Browse the corpus

h outcomes have received comparatively little attention despite the fact that injuries, noncommunicable diseases (NCDs), and acquired chronic conditions with childhood onset profoundly affect long-term health trajectories, future health care needs, intellectual development, and economic and productivity prospects.6,7,8 High return on investment is expected when evidence-based interventions are implemented to address the health and well-being of children and adolescents.9 During the past decades, the world experienced rapid economic changes along with declines in fertility and greater longevity in many countries, collectively leading to marked changes in global demographics.10,11 The identification of successes, unmet needs, and emerging challenges must therefore consider sociodemographic information to contextualize levels and trends of disease burden.5,12 This information can guide prevention and intervention efforts, tracking and allocation of resources for health and other youth-centric services (eg, education), and monitoring progress for countries at all points on the spectrum of economic development.

Introduction Reducing mortality among children younger than 5 years has been a focus of significant international attention for several decades, beginning with the Convention on the Rights of the Child, accelerating during the Millennium Development Goal era, and continuing with the Sustainable Development Goals (SDGs).1,2,3 Global progress in reducing death in children younger than 5 years has been substantial,4 but much less attention has been focused on quantifying and minimizing mortality burden among older children and adolescents.5 Likewise, nonfatal health outcomes have received comparatively little attention despite the fact that injuries, noncommunicable diseases (NCDs), and acquired chronic conditions with childhood onset profoundly affect long-term health trajectories, future health care needs, intellectual development, and economic and productivity prospects.6,7,8

rmation to contextualize levels and trends of disease burden.5,12 This information can guide prevention and intervention efforts, tracking and allocation of resources for health and other youth-centric services (eg, education), and monitoring progress for countries at all points on the spectrum of economic development. Two comprehensive reports on the burden of diseases and injuries in young persons were published following the Global Burden of Diseases, Injuries, and Risk Factors (GBD) 2013 Study.13,14 The first report covered children and adolescents 19 years or younger; the second described disease burden in young persons aged 10 to 24 years.15 In the present study—an extension of GBD 2015—we again focus on children and adolescents 19 years or younger, extending the data to 2015 and to 195 countries and territories. We present results separately by sex, describe the epidemiologic factors of several highly disabling conditions that arise from multiple GBD causes, report levels and trends in pregnancy complications among adolescents, and evaluate the association between metrics of disease burden and the Socio-demographic Index (SDI), a composite indicator of development status generated for GBD 2015. Methods Detailed methods for each analytic step in GBD 2015 are described elsewhere and are compliant with the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER).4,16,17,18,19,20,21 Data are available online at the Global Health Data Exchange (http://ghdx.healthdata.org).

Two comprehensive reports on the burden of diseases and injuries in young persons were published following the Global Burden of Diseases, Injuries, and Risk Factors (GBD) 2013 Study.13,14 The first report covered children and adolescents 19 years or younger; the second described disease burden in young persons aged 10 to 24 years.15 In the present study—an extension of GBD 2015—we again focus on children and adolescents 19 years or younger, extending the data to 2015 and to 195 countries and territories. We present results separately by sex, describe the epidemiologic factors of several highly disabling conditions that arise from multiple GBD causes, report levels and trends in pregnancy complications among adolescents, and evaluate the association between metrics of disease burden and the Socio-demographic Index (SDI), a composite indicator of development status generated for GBD 2015. Methods Detailed methods for each analytic step in GBD 2015 are described elsewhere and are compliant with the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER).4,16,17,18,19,20,21 Data are available online at the Global Health Data Exchange (http://ghdx.healthdata.org). Briefly, we quantified an extensive set of health loss metrics—with corresponding uncertainty intervals (UIs)—from 1990 to 2015 for 20 age groups and both sexes in 195 countries and territories. For the present study, we further analyzed levels and trends for children and adolescents 19 years or younger, which includes the first 7 age groups of the GBD 2015 analyses. Health loss metrics in this analysis include all-cause mortality, cause-specific mortality (deaths and years of life lost [YLLs]), nonfatal health outcomes (prevalence and years lived with disability [YLDs]), and total disease burden (disability-adjusted life years [DALYs]). Countries and territories were hierarchically organized into 21 regions and 7 super-regions, which are aggregates of the 21 regions in the GBD location hierarchy. The GBD cause list organizes all diseases and injuries into a 4-level hierarchy. The first level has 3 categories: (1) communicable, maternal, neonatal, and nutritional disorders (group I conditions); (2) NCDs; and (3) injuries. Level 2 of the hierarchy has 21 cause groups, while levels 3 (166 causes) and 4 (261 causes) contain more disaggregated causes and cause groups. The full GBD cause list with corresponding International Classification of Diseases (ICD)-9 and ICD-10 codes is available in previous publications on cause-specific mortality and nonfatal health outcomes.16,17

use groups, while levels 3 (166 causes) and 4 (261 causes) contain more disaggregated causes and cause groups. The full GBD cause list with corresponding International Classification of Diseases (ICD)-9 and ICD-10 codes is available in previous publications on cause-specific mortality and nonfatal health outcomes.16,17 Our all-cause and cause-specific mortality analyses used systematic approaches to address data challenges such as variation in both death certification practices and coding schemes, inconsistent age group reporting, and misclassification of human immunodeficiency virus (HIV) or AIDS. Each death was assigned to a single underlying cause. Cause-of-death ensemble modeling was the most widely used statistical tool for estimating cause-specific mortality across GBD 2015. Cause-of-death ensemble modeling uses a train-test-test approach to evaluate a wide range of families of statistical models, maximizing out-of-sample predictive validity of final models. Years of life lost were calculated by multiplying counts of age-specific death and normative life expectancy at the age of death.16

. Cause-of-death ensemble modeling uses a train-test-test approach to evaluate a wide range of families of statistical models, maximizing out-of-sample predictive validity of final models. Years of life lost were calculated by multiplying counts of age-specific death and normative life expectancy at the age of death.16 Analyses of nonfatal health outcomes used detailed epidemiologic data from systematic reviews of the literature, hospital and claims databases, health surveys, case notification systems, cohort studies, and disease-specific registries. DisMod-MR 2.1, a statistical modeling method developed in-house, was the most widely used statistical method in GBD 2015; it is a Bayesian meta-regression tool that synthesizes all available data, adjusting for different case definitions or sampling strategies, to generate internally consistent results for prevalence, incidence, remission, and excess mortality in each population.22 Each most-specific cause was paired with a variable number of mutually exclusive and collectively exhaustive sequelae, which quantify the main outcomes (including asymptomatic states) of diseases and injuries and are the units of analysis for nonfatal health outcomes. Years lived with disability were calculated as the product of sequela-specific prevalence and corresponding GBD disability weights derived from population surveys with more than 60 000 respondents.23,24 Disability weights were assumed to be invariant by geography, but the distribution of sequelae—and therefore cumulative disability per case—varies by geography, year, sex, and age. Finally, we adjusted for comorbid illness using a microsimulation framework within each population and proportionally adjusting YLDs for each comorbid condition. Disability-adjusted life years are the sum of YLLs and YLDs.17

sequelae—and therefore cumulative disability per case—varies by geography, year, sex, and age. Finally, we adjusted for comorbid illness using a microsimulation framework within each population and proportionally adjusting YLDs for each comorbid condition. Disability-adjusted life years are the sum of YLLs and YLDs.17 We developed the SDI for GBD 2015, as described previously, to characterize epidemiologic transitions more robustly than is possible with analyses based only on income.4,16,17,18,19 The SDI is a composite measure of developmental status as it is associated with health, calculated as the geometric mean of the following 3 indicators: total rate of fertility, log income per capita, and mean years of education among those 15 years or older. Socio-demographic Index scores were scaled from 0 (highest fertility, lowest income, and lowest education) to 1 (highest income, highest education, and lowest fertility), and each geographical unit was assigned an SDI score for each year. We analyzed the average association between SDI score and all-ages rates of YLLs, YLDs, and DALYs for all level 2 and level 3 causes. For comparisons across SDI quintiles, each geographical unit was assigned to a single quintile according to its SDI in 2015 (eFigure 1 in the Supplement).

ned an SDI score for each year. We analyzed the average association between SDI score and all-ages rates of YLLs, YLDs, and DALYs for all level 2 and level 3 causes. For comparisons across SDI quintiles, each geographical unit was assigned to a single quintile according to its SDI in 2015 (eFigure 1 in the Supplement). For all results, 95% UIs were derived from 1000 draws of the posterior distribution at each analytic step and represent the ordinal 25th and 975th draws. Unlike confidence intervals, which capture only sampling error, UIs provide a means of also capturing other sources of uncertainty owing to model specification (eg, parameter selection) and estimation (eg, data adjustments from nonreference categories and β values for covariates). Cumulative and annualized rates of change were calculated on point estimates, and corresponding UIs were derived from the same calculations performed at the draw level.

owing to model specification (eg, parameter selection) and estimation (eg, data adjustments from nonreference categories and β values for covariates). Cumulative and annualized rates of change were calculated on point estimates, and corresponding UIs were derived from the same calculations performed at the draw level. We present results as both total numbers to illustrate the absolute magnitude of burden, and all-age rates, to compare across geographical areas with differently sized populations. We completed age standardization for ages 19 years or younger for the 10 highest-ranked global causes of death and disability to help compare across populations with different age structures; all other results are presented as total number and all-ages rates only. Results for the global level, along with SDI quintile and region in order of decreasing SDI, are presented in the main article. Results for each country and territory are contained in the Supplement and are available online at http://vizhub.healthdata.org/gbd-compare by age group and sex.

umber and all-ages rates only. Results for the global level, along with SDI quintile and region in order of decreasing SDI, are presented in the main article. Results for each country and territory are contained in the Supplement and are available online at http://vizhub.healthdata.org/gbd-compare by age group and sex. Results All-Cause Mortality and Cause-Specific Mortality in Children and Adolescents Total deaths and the age-standardized mortality rate (per 100 000 population) for all causes combined, as well as the 10 largest level 3 causes of death globally, are shown for children and adolescents 19 years or younger in 1990 and 2015 in Table 1. Corresponding country-level results, with uncertainty and cumulative percent change, are in eTable 1 in the Supplement for children and adolescents 19 years or younger and eTable 2 in the Supplement for children and adolescents 5 years or younger. In 2015, there were 7.26 million (95% UI, 7.14 million to 7.39 million) deaths among children and adolescents globally, of which 5.82 million (95% UI, 5.69 million to 5.95 million) occurred among children younger than 5 years, 463 000 (95% UI, 453 000-473 000) among those aged 5 to 9 years, 391 000 (95% UI, 383 000-402 000) among children aged 10 to 14 years, and 588 000 (95% UI, 567 000-610 000) among those aged 15 to 19 years. Table 1. Top 10 Global Causes of Death in Children and Adolescents 19 Years or Younger, Both Sexes, 1990 and 2015 GBD Location No.

Results All-Cause Mortality and Cause-Specific Mortality in Children and Adolescents Total deaths and the age-standardized mortality rate (per 100 000 population) for all causes combined, as well as the 10 largest level 3 causes of death globally, are shown for children and adolescents 19 years or younger in 1990 and 2015 in Table 1. Corresponding country-level results, with uncertainty and cumulative percent change, are in eTable 1 in the Supplement for children and adolescents 19 years or younger and eTable 2 in the Supplement for children and adolescents 5 years or younger. In 2015, there were 7.26 million (95% UI, 7.14 million to 7.39 million) deaths among children and adolescents globally, of which 5.82 million (95% UI, 5.69 million to 5.95 million) occurred among children younger than 5 years, 463 000 (95% UI, 453 000-473 000) among those aged 5 to 9 years, 391 000 (95% UI, 383 000-402 000) among children aged 10 to 14 years, and 588 000 (95% UI, 567 000-610 000) among those aged 15 to 19 years. Table 1. Top 10 Global Causes of Death in Children and Adolescents 19 Years or Younger, Both Sexes, 1990 and 2015 GBD Location No. of Deaths (Death Rate) per 100 000 Population All Causes Neonatal Preterm Birth Complications Lower Respiratory Tract Infections Neonatal Encephalopathy Due to Birth Asphyxia or Trauma Diarrheal Diseases Congenital Anomalies Malaria Neonatal Sepsis and Other Neonatal Infections Meningitis Other Neonatal Disorders HIV and AIDS 2015 Global 7 263 484 (285.4) 805 778 (31.4) 792 992 (31.1) 740 424 (28.8) 569 737 (22.4) 543 314 (21.3) 534 007 (21.0) 351 667 (13.7) 220 530 (8.7) 220 247 (8.6) 202 929 (8.1) SDI High 118 122 (43.9) 13 493 (5.2) 4399 (1.7) 5509 (2.1) 832 (0.3) 23 775 (9.1) 1 (0.0) 2920 (1.1) 958 (0.4) 5802 (2.2) 699 (0.3) High-middle 536 318 (118.3) 75 776 (17.4) 43 653 (9.7) 49 568 (11.4) 11 265 (2.5) 79 782 (18.0) 428 (0.1) 18 756 (4.3) 7066 (1.5) 21 014 (4.8) 21 175 (4.6) Middle 1 191 374 (174.8) 178 438 (26.2) 108 851 (16.0) 133 759 (19.7) 49 594 (7.3) 132 103 (19.4) 6396 (0.9) 44 875 (6.6) 23 862 (3.5) 36 854 (5.4) 21 159 (3.1) Low-middle 3 418 022 (425.4) 410 824 (50.0) 382 444 (47.5) 432 718 (52.6) 322 586 (40.2) 193 453 (23.8) 252 862 (31.7) 173 049 (21.1) 99 627 (12.5) 99 004 (12.1) 91 953 (11.7) Low 1 996 606 (581.6) 126 934 (34.2) 253 251 (72.3) 118 676 (32.0) 185 251 (53.7) 113 896 (32.0) 274 248 (80.4) 111 929 (30.2) 88 928 (26.2) 57 449 (15.5) 67 860 (22.1) GBD region High-income North America 42 322 (48.5) 5914 (7.3) 826 (1.0) 1683 (2.1) 328 (0.4) 7144 (8.6) 0 (0.0) 730 (0.9) 252 (0.3) 2382 (2.9) 95 (0.1) Australasia 2582 (35.4) 205 (3.0) 55 (0.8) 150 (2.2) 16 (0.2) 513 (7.3) 0 (0.0) 36 (0.5) 18 (0.3) 141 (2.0) 2 (0.0) High-income Asia Pacific 8211 (26.0) 519 (1.8) 325 (1.1) 205 (0.7) 54 (0.2) 1718 (5.9) 0 (0.0) 172 (0.6) 39 (0.1) 303 (1.1) 14 (0.0) Western Europe 25 449 (29.6) 3090 (3.8) 594 (0.7) 1247 (1.5) 176 (0.2) 5728 (6.9) 0 (0.0) 514 (0.6) 232 (0.3) 1205 (1.5) 59 (0.1) Southern Latin America 16 800 (85.5) 2582 (13.6) 892 (4.6) 579 (3.0) 173 (0.9) 3821 (19.9) 0 (0.0) 658 (3.5) 186 (1.0) 677 (3.6) 77 (0.4) Eastern Europe 32 817 (76.0) 2390 (5.1) 2112 (4.7) 2082 (4.5) 199 (0.4) 6781 (14.8) 0 (0.0) 1318 (2.8) 429 (0.9) 1584 (3.4) 532 (1.2) Central Europe 10 849 (48.1) 1330 (6.3) 955 (4.3) 423 (2.0) 71 (0.3) 2323 (10.8) 0 (0.0) 164 (0.8)

92 (4.6) 579 (3.0) 173 (0.9) 3821 (19.9) 0 (0.0) 658 (3.5) 186 (1.0) 677 (3.6) 77 (0.4) Eastern Europe 32 817 (76.0) 2390 (5.1) 2112 (4.7) 2082 (4.5) 199 (0.4) 6781 (14.8) 0 (0.0) 1318 (2.8) 429 (0.9) 1584 (3.4) 532 (1.2) Central Europe 10 849 (48.1) 1330 (6.3) 955 (4.3) 423 (2.0) 71 (0.3) 2323 (10.8) 0 (0.0) 164 (0.8) 98 (0.4) 532 (2.5) 25 (0.1) Central Asia 61 200 (180.5) 7663 (21.8) 14 789 (42.6) 8096 (23.0) 1777 (5.1) 6966 (20.0) 3 (0.0) 1242 (3.5) 793 (2.4) 2421 (6.9) 48 (0.2) Central Latin America 114 654 (128.2) 13 254 (15.5) 11 000 (12.5) 5894 (6.9) 4339 (4.9) 20 035 (23.0) 20 (0.0) 6190 (7.2) 1039 (1.2) 3064 (3.6) 596 (0.7) Andean Latin America 30 164 (135.9) 3123 (14.2) 4242 (19.1) 2508 (11.4) 1012 (4.6) 3657 (16.5) 2 (0.0) 2472 (11.2) 413 (1.9) 642 (2.9) 22 (0.1) Caribbean 32 608 (222.5) 3476 (24.3) 3652 (25.2) 2384 (16.7) 2913 (20.2) 3669 (25.3) 36 (0.2) 1959 (13.7) 932 (6.4) 1544 (10.8) 933 (6.1) Tropical Latin America 83 965 (133.3) 10 140 (17.5) 6045 (10.0) 6083 (10.5) 1978 (3.3) 10 729 (18.1) 20 (0.0) 5271 (9.1) 1446 (2.3) 4525 (7.8) 710 (1.1) East Asia 309 899 (95.6) 39 620 (12.5) 28 066 (8.7) 27 558 (8.7) 2230 (0.7) 53 615 (16.7) 27 (0.0) 2766 (0.9) 3041 (0.9) 7121 (2.2) 1655 (0.5) Southeast Asia 365 942 (162.2) 47 066 (21.2) 46 590 (20.8) 31 717 (14.3) 17 561 (7.8) 39 415 (17.6) 3273 (1.4) 18 379 (8.3) 8862 (3.9) 10 376 (4.7) 2295 (1.0) Oceania 15 005 (290.4) 1244 (23.5) 2778 (53.1) 738 (14.0) 617 (12.0) 851 (16.2) 413 (8.3) 314 (5.9) 371 (7.2) 671 (12.7) 64 (1.3) North Africa and Middle East 529 160 (222.5) 83 998 (34.3) 55 221 (22.8) 19 930 (8.1) 24 608 (10.1) 81 812 (33.6) 4371 (1.8) 17 737 (7.2) 11 040 (4.6) 19 646 (8.0) 494 (0.2) South Asia 2 205 667 (343.6) 379 162 (59.8) 235 756 (37.0) 413 928 (65.2) 175 213 (27.4) 111 162 (17.4) 21 434 (3.3) 61 781 (9.8) 55 233 (8.6) 71 394 (11.3) 15 984 (2.4) Southern sub-Saharan Africa 126 790 (386.8) 10 049 (29.4) 9265 (27.8) 9157 (26.8) 11 466 (34.3) 4269 (12.7) 873 (2.6) 4274 (12.5) 1932 (5.9) 6571 (19.2) 40 778 (128.4) Western sub-Saharan Africa 1 680 122 (665.5) 93 613 (34.1) 170 118 (66.7) 105 859 (38.6) 197 475 (77.6) 72 544 (27.8) 353 769 (141.8) 141 738 (51.7) 71 368 (28.9) 29 955 (11.0) 47 729 (21.4) Eastern sub-Saharan Africa 1 106 529 (476.2) 70 810 (28.4) 140 010 (59.0) 74 910 (30.1) 96 769 (41.8) 81 468 (33.9) 73 950 (31.9) 57 493 (23.1) 44 598 (19.4) 39 987 (16.1) 78 604 (37.0) Central sub-Saharan Africa 462 738 (591.5) 26 521 (30.7) 59 690 (74.4) 25 282 (29.3) 30 751 (38.8) 25 084 (30.4) 75 807 (98

729 (21.4) Eastern sub-Saharan Africa 1 106 529 (476.2) 70 810 (28.4) 140 010 (59.0) 74 910 (30.1) 96 769 (41.8) 81 468 (33.9) 73 950 (31.9) 57 493 (23.1) 44 598 (19.4) 39 987 (16.1) 78 604 (37.0) Central sub-Saharan Africa 462 738 (591.5) 26 521 (30.7) 59 690 (74.4) 25 282 (29.3) 30 751 (38.8) 25 084 (30.4) 75 807 (98 .2) 26 449 (30.7) 18 196 (23.7) 15 497 (18.0) 12 202 (18.2) 1990 Global 14 182 624 (584.6) 1 795 211 (71.5) 2 241 773 (91.6) 915 323 (36.4) 1 536 806 (63.3) 696 037 (28.3) 791 867 (32.9) 329 296 (13.1) 376 652 (15.6) 351 304 (14.0) 39 363 (1.6) SDI High 295 736 (100.5) 42 760 (15.5) 17 451 (6.1) 17 881 (6.5) 3350 (1.2) 50 953 (18.1) 9 (0.0) 4644 (1.7) 4573 (1.6) 9008 (3.3) 927 (0.3) High-middle 1 666 079 (319.0) 286 715 (54.8) 248 629 (47.4) 96 928 (18.5) 107 312 (20.5) 161 837 (30.9) 918 (0.2) 25 555 (4.9) 32 409 (6.2) 54 939 (10.5) 1269 (0.2) Middle 3 608 743 (473.5) 575 921 (73.6) 622 238 (80.8) 240 833 (30.8) 317 411 (41.5) 219 456 (28.5) 12 876 (1.7) 62 606 (8.0) 84 860 (11.2) 87 112 (11.2) 1502 (0.2) Low-middle 6 148 482 (934.4) 765 273 (107.7) 1 023 454 (153.8) 459 777 (64.6) 784 970 (120.3) 186 302 (27.4) 366 926 (57.3) 149 124 (21.0) 156 226 (24.2) 137 559 (19.5) 17 592 (2.7) Low 2 457 431 (1297.4) 123 952 (56.3) 328 981 (168.1) 99 603 (45.2) 322 951 (170.3) 77 154 (38.1) 410 936 (224.0) 87 218 (39.8) 98 394 (52.6) 62 494 (28.6) 18 035 (9.6) GBD region High-income North America 74 124 (91.8) 12 644 (15.7) 1903 (2.4) 3113 (3.9) 250 (0.3) 11 613 (14.4) 0 (0.0) 963 (1.2) 891 (1.1) 3264 (4.1) 487 (0.6) Australasia 4856 (80.1) 658 (11.5) 120 (2.1) 277 (4.8) 10 (0.2) 829 (14.3) 0 (0.0) 60 (1.1) 57 (1.0) 77 (1.3) 7 (0.1) High-income Asia Pacific 30 483 (69.1) 2632 (7.1) 1550 (3.7) 886 (2.4) 213 (0.5) 5776 (15.0) 5 (0.0) 489 (1.3) 373 (0.9) 635 (1.7) 29 (0.1) Western Europe 67 742 (74.7) 10 249 (12.4) 1934 (2.2) 4157 (5.0) 261 (0.3) 12 813 (15.1) 0 (0.0) 1041 (1.3) 1073 (1.2) 1225 (1.5) 212 (0.2) Southern Latin America 33 064 (172.3) 7143 (37.0) 3231 (16.7) 1866 (9.7) 981 (5.1) 5072 (26.3) 2 (0.0) 1171 (6.1) 668 (3.5) 1078 (5.6) 94 (0.5) Eastern Europe 97 965 (164.7) 9661 (17.7) 9579 (16.5) 8884 (16.3) 1901 (3.3) 16 902 (29.6) 0 (0.0) 1805 (3.3) 2274 (3.8) 3293 (6.0) 199 (0.3) Central Europe 51 452 (150.4) 8394 (26.4) 8268 (24.8) 2840 (8.9) 856 (2.6) 7929 (24.2) 0 (0.0) 574 (1.8) 929 (2.7) 2772 (8.7) 136 (0.4) Central Asia 136 834 (390.1) 13 318 (36.7) 51 286 (143.6) 12 803 (35.3) 13 444 (37.4) 8107 (22.7) 12 (0.0) 1524 (4.2) 2341 (6.8) 3213 (8.9) 18 (0.1) Cent

274 (3.8) 3293 (6.0) 199 (0.3) Central Europe 51 452 (150.4) 8394 (26.4) 8268 (24.8) 2840 (8.9) 856 (2.6) 7929 (24.2) 0 (0.0) 574 (1.8) 929 (2.7) 2772 (8.7) 136 (0.4) Central Asia 136 834 (390.1) 13 318 (36.7) 51 286 (143.6) 12 803 (35.3) 13 444 (37.4) 8107 (22.7) 12 (0.0) 1524 (4.2) 2341 (6.8) 3213 (8.9) 18 (0.1) Cent ral Latin America 262 420 (297.0) 31 927 (35.4) 34 869 (39.1) 18 532 (20.5) 40 522 (45.4) 22 447 (25.1) 233 (0.3) 8584 (9.5) 3995 (4.5) 6459 (7.2) 364 (0.4) Andean Latin America 107 369 (508.1) 8778 (39.8) 24 661 (115.2) 4969 (22.6) 13 822 (65.0) 3676 (17.0) 50 (0.3) 4127 (18.8) 1412 (6.8) 1557 (7.1) 25 (0.1) Caribbean 73 033 (455.0) 7303 (44.7) 8756 (54.3) 4272 (26.1) 14 855 (92.1) 4551 (28.2) 160 (1.0) 2378 (14.6) 2687 (16.7) 2538 (15.5) 775 (4.8) Tropical Latin America 238 315 (343.6) 42 309 (61.3) 32 888 (47.3) 14 189 (20.5) 41 650 (60.2) 13 525 (19.5) 485 (0.7) 9689 (14.0) 6676 (9.6) 4045 (5.9) 392 (0.6) East Asia 1 864 295 (383.0) 333 663 (67.9) 409 264 (83.1) 72 246 (14.7) 64 343 (13.2) 165 662 (33.8) 76 (0.0) 7371 (1.5) 27 587 (5.7) 38 942 (7.9) 55 (0.0) Southeast Asia 1 068 595 (480.1) 122 834 (54.4) 211 599 (94.8) 62 520 (27.7) 102 556 (46.1) 48 562 (21.7) 15 537 (7.1) 25 592 (11.3) 26 666 (12.0) 29 145 (12.9) 733 (0.3) Oceania 19 733 (517.9) 1410 (34.8) 4610 (118.0) 700 (17.3) 1526 (40.3) 647 (16.5) 592 (16.1) 231 (5.7) 546 (14.4) 592 (14.7) 13 (0.4) North Africa and Middle East 1 045 563 (531.5) 153 951 (75.1) 168 843 (84.4) 28 635 (14.0) 123 274 (61.4) 105 344 (52.3) 4665 (2.4) 17 328 (8.5) 23 954 (12.1) 37 480 (18.3) 91 (0.0) South Asia 4 939 233 (808.2) 831 361 (127.9) 741 686 (120.5) 497 476 (76.5) 586 134 (97.0) 136 714 (21.8) 56 232 (9.7) 79 611 (12.3) 122 461 (20.5) 111 976 (17.4) 537 (0.1) Southern sub-Saharan Africa 144 842 (497.4) 11 693 (38.6) 20 799 (70.5) 7794 (25.7) 29 998 (101.6) 4607 (15.5) 831 (2.9) 3686 (12.2) 2585 (9.0) 12 584 (41.6) 3422 (11.7) Western sub-Saharan Africa 1 853 426 (1333.6) 84 305 (52.7) 226 611 (159.6) 81 707 (51.0) 246 187 (177.6) 53 610 (36.3) 368 547 (275.3) 96 745 (60.6) 76 499 (56.0) 32 635 (20.5) 5590 (4.1) Eastern sub-Saharan Africa 1 612 637 (1193.5) 80 419 (52.0) 222 288 (159.4) 72 023 (46.5) 214 691 (159.5) 53 005 (37.0) 249 176 (189.0) 48 757 (31.7) 58 781 (44.1) 45 654 (29.6) 23 340 (17.3) Central sub-Saharan Africa 456 634 (1160.0) 20 547 (44.8) 57 016 (139.7) 15 424 (33.7) 39 323 (98.5) 14 638 (34.4) 95 256 (248.6) 17 562 (38.4) 14 185 (36.8) 12 130 (26.6) 2833 (7.4) Abbreviations: GBD, Global Burden

) 214 691 (159.5) 53 005 (37.0) 249 176 (189.0) 48 757 (31.7) 58 781 (44.1) 45 654 (29.6) 23 340 (17.3) Central sub-Saharan Africa 456 634 (1160.0) 20 547 (44.8) 57 016 (139.7) 15 424 (33.7) 39 323 (98.5) 14 638 (34.4) 95 256 (248.6) 17 562 (38.4) 14 185 (36.8) 12 130 (26.6) 2833 (7.4) Abbreviations: GBD, Global Burden of Diseases, Injuries, and Risk Factors Study; HIV, human immunodeficiency virus; SDI, Socio-demographic Index.

) 214 691 (159.5) 53 005 (37.0) 249 176 (189.0) 48 757 (31.7) 58 781 (44.1) 45 654 (29.6) 23 340 (17.3) Central sub-Saharan Africa 456 634 (1160.0) 20 547 (44.8) 57 016 (139.7) 15 424 (33.7) 39 323 (98.5) 14 638 (34.4) 95 256 (248.6) 17 562 (38.4) 14 185 (36.8) 12 130 (26.6) 2833 (7.4) Abbreviations: GBD, Global Burden of Diseases, Injuries, and Risk Factors Study; HIV, human immunodeficiency virus; SDI, Socio-demographic Index. As can be seen in Table 1, mortality in children and adolescents 19 years or younger decreased in all SDI quintiles, but inequality increased. Nearly 75% of all deaths among children and adolescents in 2015 occurred in the 2 lowest SDI quintiles (compared with 61% in 1990), while only 1.6% occurred in the highest SDI quintile (compared with 2.1% in 2015). Age-standardized rates of death declined from 1990 to 2015 at similar rates of 55% and 56% in the 2 lowest and highest SDI quintiles, respectively, while they declined by 63% in middle and high-middle SDI quintiles. South Asia accounted for 2.21 million (95% UI, 2.15 million to 2.27 million) child and adolescent deaths, 30.4% of the global total and the most of any region. Next were Western sub-Saharan Africa (1.68 million; 95% UI, 1.61 million to 1.76 million [23.1%]), Eastern sub-Saharan Africa (1.11 million; 95% UI, 1.07 million to 1.14 million [15.3%]), North Africa and the Middle East (529 000; 95% UI, 499 000-562 000 [7.3%]), and central sub-Saharan Africa (463 000; 95% UI, 408 000-524 000 [6.4%]). Geographical patterns of mortality in children younger than 5 years were similar to those in children and adolescents 19 years or younger but with a slightly greater concentration of mortality burden in the 2 lowest SDI quintiles (77% of total). Mortality rates (per 100 000 population) varied from a low of 26.0 (95% UI, 25.1-26.8) in the high-income Asia Pacific region to a high of 666 (95% UI, 638-696) in Western sub-Saharan Africa among all children and adolescents 19 years or younger and from 58.8 (95% UI, 55.8-61.8) in the high-income Asia Pacific region to 2133 (95% UI, 2029-2245) in Western sub-Saharan Africa for children 5 years or younger in 2015.

h-income Asia Pacific region to a high of 666 (95% UI, 638-696) in Western sub-Saharan Africa among all children and adolescents 19 years or younger and from 58.8 (95% UI, 55.8-61.8) in the high-income Asia Pacific region to 2133 (95% UI, 2029-2245) in Western sub-Saharan Africa for children 5 years or younger in 2015. Cause-Specific Mortality As seen in Table 1 across the entire age range, rankings were dominated by those affecting the youngest children. Globally, the most common causes of death were neonatal preterm birth complications (mortality rate, 31.4 per 100 000 population; 95% UI, 29.1-34.2 deaths per 100 000 population), lower respiratory tract infections (LRIs) (31.1; 95% UI, 29.2-33.0), neonatal encephalopathy owing to birth asphyxia and trauma (28.8; 95% UI, 26.5-31.5), diarrheal diseases (22.4; 95% UI, 20.5-24.2), congenital anomalies (21.3; 95% UI, 19.7-23.1), malaria (21.0; 95% UI, 16.2-25.6), neonatal sepsis (13.7; 95% UI, 10.7-16.7), other neonatal disorders (8.6; 95% UI, 7.4-10.3), meningitis (8.7; 95% UI, 6.8-10.4), and HIV and AIDS (8.1; 95% UI, 7.8-8.3). With the exception of the infectious causes (malaria, diarrheal diseases, and meningitis) each cause was highly ranked in all regions.

UI, 16.2-25.6), neonatal sepsis (13.7; 95% UI, 10.7-16.7), other neonatal disorders (8.6; 95% UI, 7.4-10.3), meningitis (8.7; 95% UI, 6.8-10.4), and HIV and AIDS (8.1; 95% UI, 7.8-8.3). With the exception of the infectious causes (malaria, diarrheal diseases, and meningitis) each cause was highly ranked in all regions. Rankings of the 25 leading level 3 causes of death among children and adolescents 19 years or younger, disaggregated by sex, are shown in Figure 1. Besides the causes listed above, others ranking in the top 10 in specific regions included hemoglobinopathies and hemolytic anemias (in Western sub-Saharan Africa, where sickle cell disease is the largest level 4 cause of hemoglobinopathies), as well as selected infections (measles, HIV and AIDS, whooping cough, intestinal infectious disease, sexually transmitted infections excluding HIV [ie, congenital syphilis], and encephalitis) and injuries (drowning, road injuries, direct effects of war [ie, collective violence] and natural disasters, exposure to mechanical forces, aspiration of a foreign body, and fire).

whooping cough, intestinal infectious disease, sexually transmitted infections excluding HIV [ie, congenital syphilis], and encephalitis) and injuries (drowning, road injuries, direct effects of war [ie, collective violence] and natural disasters, exposure to mechanical forces, aspiration of a foreign body, and fire). Figure 1. Ranking of the Top 25 Global Causes of Death in 2015 by 5 Socio-demographic Index (SDI) Quintiles and 21 Regions in the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) Ranking of causes of death in females and males. Global causes, SDI quintiles, and GBD regions appear in columns, sorted in order of decreasing SDI status. The causes are sorted according to their ranking at the global level. The color coding indicates the relative ranking of each cause, with red the highest and green the lowest. The numbers appearing in each column indicate the geography-specific ranking of that cause in 2015. Blanks indicate causes that were not contracted in that geographical area. HIV indicates human immunodeficiency virus.

color coding indicates the relative ranking of each cause, with red the highest and green the lowest. The numbers appearing in each column indicate the geography-specific ranking of that cause in 2015. Blanks indicate causes that were not contracted in that geographical area. HIV indicates human immunodeficiency virus. Differences in Causes of Death by Geography, Age, and Sex We found important differences in mortality patterns for each of the 7 component age groups 19 years or younger in 2015 (eFigure 2A-G in the Supplement). During the neonatal period (ie, 6 days or less and 7-27 days), rankings across SDI quintiles and regions were broadly similar; mortality was dominated by neonatal complications, congenital anomalies, and LRIs. Divergence began to appear during the postneonatal period (ie, 28-364 days), when acquired infections such as LRIs, diarrhea, malaria, and meningitis predominated in lower-SDI geographical areas and congenital anomalies and sudden infant death syndrome predominated in higher-SDI geographical areas. Protein-energy malnutrition also emerged as an important cause of death in the postneonatal period in several regions, especially in males, a trend that continued into children aged 1 to 4 years, where it ranked fourth globally in both sexes. Malaria, LRIs, and diarrhea were the 3 highest-ranked causes of death in children aged 1 to 4 years; because protein-energy malnutrition and other forms of malnutrition raise the mortality risk for each, the effect of malnutrition is even higher than that reflected in results for protein-energy malnutrition alone. Geographic heterogeneity was also observed in other causes of death in children aged 1 to 4 years for both females and males at the global level, including measles (concentrated in the lowest 3 SDI quintiles, particularly Oceania and Southeast Asia), leukemia, road injuries, and drowning (all concentrated in the 3 highest SDI quintiles).

neity was also observed in other causes of death in children aged 1 to 4 years for both females and males at the global level, including measles (concentrated in the lowest 3 SDI quintiles, particularly Oceania and Southeast Asia), leukemia, road injuries, and drowning (all concentrated in the 3 highest SDI quintiles). Geographical differences in causes of death in 2015 were more pronounced with increasing age (ie, 5-9 years, 10-14 years, and 15-19 years). Congenital anomalies and cancers (leukemia, brain cancer, and other neoplasms [eg, sarcomas]) were highly ranked in high-SDI regions in all age groups, simultaneously reflecting continued risk of mortality beyond the time of initial diagnosis and lower overall risk of mortality in the population. Intestinal infectious disease was highly ranked globally (second in children aged 5-9 years for both males and females), driven primarily by very large mortality numbers in South Asia and Southeast Asia. Human immunodeficiency virus and AIDS rose to be ranked first globally among children aged 10 to 14 years, driven almost entirely by epidemics in the Caribbean and sub-Saharan Africa. Diarrhea, LRIs, malaria, and protein-energy malnutrition remained important causes of death throughout all age groups but were largely limited except in geographical areas with lower SDIs. Five level 3 causes of maternal mortality—hemorrhage, hypertensive disorders, indirect causes, other direct causes, and the combined category of abortion, ectopic pregnancy, and miscarriage—were in the top 25 causes of maternal mortality globally in females aged 15 to 19 years, reflecting the high burden of maternal mortality among adolescents in the 2 lowest SDI quintiles.

pertensive disorders, indirect causes, other direct causes, and the combined category of abortion, ectopic pregnancy, and miscarriage—were in the top 25 causes of maternal mortality globally in females aged 15 to 19 years, reflecting the high burden of maternal mortality among adolescents in the 2 lowest SDI quintiles. The ranking of injuries as causes of death increased consistently with age and with increasing SDI; all injuries except self-harm ranked higher in males than females. Road injuries were the leading injury-associated cause of death in all age groups, rising to first globally among all causes for both sexes in adolescents aged 15 to 19 years. Drowning was the next highest-ranked cause of injury-associated death in children aged 5 to 9 years (ninth overall among females and sixth among males) and 10 to 14 years (eighth overall in females and third in males), while self-harm (second overall in females and third in males) and interpersonal violence (14th overall in females and second in males) were the next most common injury-associated causes of death among adolescents aged 15 to 19 years. The direct mortality burden of war was extremely large in North Africa and the Middle East, where it ranked second for each sex among children aged 1 to 4 years and first in all subsequent age groups in 2015.

nd in males) were the next most common injury-associated causes of death among adolescents aged 15 to 19 years. The direct mortality burden of war was extremely large in North Africa and the Middle East, where it ranked second for each sex among children aged 1 to 4 years and first in all subsequent age groups in 2015. Leading Causes of Nonfatal Health Outcomes in Children and Adolescents Total prevalent cases and the age-standardized prevalence rate (per 100 000 population) for all causes combined, as well as the 10 leading level 3 causes with the most YLDs globally, are shown for children and adolescents 19 years or younger in Table 2. Corresponding country-level results for 1990 and 2015, with uncertainty and mean annualized rates of change, are in eTable 3 in the Supplement for children and adolescents 19 years or younger and eTable 4 in the Supplement for children 5 years or younger. In 2015, nonfatal health outcomes caused 154 million (95% UI, 117 million to 196 million) YLDs among children and adolescents, of which 33.3 million (95% UI, 23.5 million to 45.3 million) were in children 5 years or younger, 35.0 million (95% UI, 24.9 million to 47.4 million) in those aged 5 to 9 years, 40.9 million (95% UI, 29.8 million to 54.9 million) in those aged 10 to 14 years, and 44.4 million (95% UI, 32.9 million to 58.0 million) in those aged 15 to 19 years. Table 2. Top 10 Global Causes of Years Lived With Disability (YLDs) in Children and Adolescents 19 Years or Younger, Both Sexes, 1990 and 2015 GBD Location No.

Leading Causes of Nonfatal Health Outcomes in Children and Adolescents Total prevalent cases and the age-standardized prevalence rate (per 100 000 population) for all causes combined, as well as the 10 leading level 3 causes with the most YLDs globally, are shown for children and adolescents 19 years or younger in Table 2. Corresponding country-level results for 1990 and 2015, with uncertainty and mean annualized rates of change, are in eTable 3 in the Supplement for children and adolescents 19 years or younger and eTable 4 in the Supplement for children 5 years or younger. In 2015, nonfatal health outcomes caused 154 million (95% UI, 117 million to 196 million) YLDs among children and adolescents, of which 33.3 million (95% UI, 23.5 million to 45.3 million) were in children 5 years or younger, 35.0 million (95% UI, 24.9 million to 47.4 million) in those aged 5 to 9 years, 40.9 million (95% UI, 29.8 million to 54.9 million) in those aged 10 to 14 years, and 44.4 million (95% UI, 32.9 million to 58.0 million) in those aged 15 to 19 years. Table 2. Top 10 Global Causes of Years Lived With Disability (YLDs) in Children and Adolescents 19 Years or Younger, Both Sexes, 1990 and 2015 GBD Location No. (Rate) of Prevalent Cases and YLDs per 100 000 Population All Causes Iron-Deficiency Anemia Skin and Subcutaneous Diseases Asthma Hemoglobinopathies and Hemolytic Anemias Diarrheal Diseases Congenital Anomalies Protein-Energy Malnutrition Epilepsy Malaria Neonatal Preterm Birth Complications Prevalence Global 2 289 784 742 (91 528) 713 016 539 (28 435) 841 794 320 (33 722) 158 151 385 (6340) 590 315 873 (23 585) 23 261 098 (920) 33 930 983 (1355) 22 448 815 (881) 8 507 896 (340) 185 157 379 (7397) 19 663 514 (780) SDI High 226 113 174 (82 348) 61 970 977 (22 899) 81 551 525 (29 298) 16 969 001 (6194) 33 076 375 (12 086) 208 573 (78) 4 816 429 (1759) 233 687 (88) 797 827 (291) 173 (0) 2 609 151 (968) High-middle 426 126 367 (89 979) 116 899 236 (24 950) 161 438 549 (33 674) 31 001 285 (6567) 78 113 169 (16 524) 2 383 153 (519) 7 606 302 (1609) 1 996 215 (439) 1 685 760 (357) 1 903 940 (407) 4 007 704 (866) Middle 620 437 513 (91 245) 183 177 818 (26 995) 235 523 024 (34 526) 40 051 924 (5918) 147 366 794 (21 680) 5 365 803 (790) 9 523 656 (1401) 5 312 581 (780) 2 279 318 (335) 8 147 684 (1204) 6 206 477 (912) Low-middle 728 493 883 (94 109) 264 078 445 (33 931) 254 287 797 (33 195) 47 579 774 (6136) 242 252 419 (31 257) 10 271 205 (1300) 8 677 889 (1118) 10 895 631 (1359) 3 043 617 (394) 96 717 358 (12 427) 5 147 836 (654) Low 287 519 634 (95 950) 86 753 142 (27 923) 108 638 223 (37 243) 22 328 931 (7575) 89 910 958 (29 897) 5059 868 (1574) 3 262 823 (1078) 4 053 749 (1174) 692 913 (230) 78 540 101 (26 024) 1 678 626 (511) GBD region High-income North America 73 330 440 (79 912) 21 068 515 (23 475) 24 019 562 (25 692) 5 707 411 (6278) 14 931 029 (16 373) 4563 (5) 1 711 199 (1877) 57 (0) 307 601 (335) 0 (0) 1 139 482 (1295) Australasia 5 588 277 (77 136) 1 640 974 (23 319) 2 146 577 (29 085) 832 472 (11 552) 283 121 (3945) 1639 (23) 115 479 (1603) 3 (0) 15 614 (216) 0 (0) 64 327 (900) High-income Asia Pacific 26 481 561 (79 059) 6 465 007 (19 259) 10 625 742 (30 803) 1 693 356 (5094) 1 375 105 (4121) 6409 (21) 617 514 (1861) 22 (0) 86 975 (262) 170 (1) 221 380 (702) Western Europe 74 207 193 (82 398) 20 289 274 (23 000) 25 919 813 (28 333) 6 752 467 (7434) 11 491

(0) 15 614 (216) 0 (0) 64 327 (900) High-income Asia Pacific 26 481 561 (79 059) 6 465 007 (19 259) 10 625 742 (30 803) 1 693 356 (5094) 1 375 105 (4121) 6409 (21) 617 514 (1861) 22 (0) 86 975 (262) 170 (1) 221 380 (702) Western Europe 74 207 193 (82 398) 20 289 274 (23 000) 25 919 813 (28 333) 6 752 467 (7434) 11 491 219 (12 828) 38 753 (45) 1 320 347 (1474) 43 (0) 271 816 (303) 0 (0) 737 461 (848) Southern Latin America 17 481 603 (85 921) 5 289 625 (26 396) 5 828 636 (28 374) 1 743 887 (8559) 1 305 899 (6438) 14 815 (75) 317 631 (1565) 325 (2) 81 606 (401) 0 (0) 163 368 (817) Eastern Europe 35 014 905 (87 740) 10 661 588 (26 359) 13 133 232 (33 901) 1 682 897 (4390) 2 945 210 (7331) 109 724 (245) 710 175 (1762) 143 535 (304) 74 876 (189) 0 (0) 364 503 (860) Central Europe 20 577 466 (88 408) 5 438 029 (24 068) 7 978 808 (33 409) 1 063 290 (4589) 1 860 190 (8037) 24 565 (110) 426 788 (1840) 86 646 (391) 64 416 (276) 0 (0) 188 351 (833) Central Asia 27 637 487 (90 263) 8 895 722 (28 648) 9 860 451 (32 898) 1 287 972 (4347) 2 758 875 (8982) 191 465 (572) 430 890 (1398) 200 781 (567) 162 189 (541) 9 (0) 492 888 (1523) Central Latin America 83 994 289 (89 808) 20 274 661 (21 845) 30 778 797 (32 582) 9 302 241 (9967) 10 040 751 (10 760) 548 915 (616) 1 353 276 (1452) 189 979 (216) 431 697 (462) 144 994 (161) 539 148 (591) Andean Latin America 20 988 527 (93 924) 7 087 131 (31 757) 7 825 043 (34 967) 2 708 694 (12 148) 2 196 360 (9831) 242 067 (1085) 267 125 (1196) 29 637 (133) 86 487 (387) 56 971 (256) 228 882 (1028) Caribbean 14 098 610 (91 995) 4 867 798 (32 025) 5 415 326 (35 027) 2 437 520 (15 925) 1 639 505 (10 711) 161 040 (1090) 191 664 (1253) 61 637 (423) 47 701 (310) 31 611 (211) 298 919 (2000) Tropical Latin America 62 922 628 (92 101) 18 749 359 (28 291) 24 326 804 (34 888) 9 541 805 (13 939) 11 600 835 (17 051) 588 023 (943) 951 436 (1401) 113 333 (190) 201 109 (296) 155 544 (247) 781 985 (1202) East Asia 291 154 416 (88 314) 70 013 496 (21 384) 123 484 074 (37 206) 9 882 773 (3008) 60 640 901 (18 421) 599 691 (184) 5 838 748 (1773) 1 016 159 (312) 706 018 (215) 71 784 (22) 2 584 742 (791) Southeast Asia 212 766 895 (92 671) 54 989 778 (24 111) 87 545 667 (37 908) 17 619 693 (7689) 45 115 712 (19 666) 2 162 600 (953) 3 206 636 (1398) 1 908 798 (846) 833 383 (364) 6 782 640 (2958) 2 874 530 (1265) Oceania 4 801 545 (95 827) 1 171 545 (23 282) 2 038 890 (40 964) 533 158 (10 659) 964 886 (19 230) 43 411 (852) 64 763 (1290) 23 563 (453) 13 568 (270) 448 392 (8959) 86 629

67 (37 908) 17 619 693 (7689) 45 115 712 (19 666) 2 162 600 (953) 3 206 636 (1398) 1 908 798 (846) 833 383 (364) 6 782 640 (2958) 2 874 530 (1265) Oceania 4 801 545 (95 827) 1 171 545 (23 282) 2 038 890 (40 964) 533 158 (10 659) 964 886 (19 230) 43 411 (852) 64 763 (1290) 23 563 (453) 13 568 (270) 448 392 (8959) 86 629 (1689) North Africa and Middle East 203 284 245 (90 320) 54 678 673 (24 033) 63 541 624 (28 678) 15 382 502 (6882) 47 859 607 (21 205) 2 159 395 (902) 2 804 537 (1241) 2 057 860 (843) 1 161 561 (514) 2 712 071 (1201) 1 604 151 (690) South Asia 620 847 125 (93 528) 248 419 970 (37 653) 214 109 490 (32 007) 31 947 985 (4798) 199 986 527 (30 145) 8 261 636 (1274) 7 813 202 (1179) 10 067 488 (1587) 2 884 194 (434) 15 450 728 (2344) 4 572 823 (699) Southern sub-Saharan Africa 29 695 592 (94 467) 8 012 341 (25 410) 11 355 636 (36 285) 3 433 792 (11 094) 6 406 425 (20 356) 396 166 (1220) 429 423 (1362) 160 371 (478) 66 419 (212) 650 182 (2065) 192 464 (597) Western sub-Saharan Africa 206 177 716 (97 079) 71 758 741 (32 759) 71 801 710 (35 351) 12 737 815 (6197) 86 474 216 (40 608) 3 127 410 (1350) 2 318 812 (1079) 3 389 017 (1340) 435 017 (204) 95 826 414 (44 837) 996 271 (422) Eastern sub-Saharan Africa 195 800 822 (95 264) 55 706 419 (26 096) 76 085 688 (37 921) 14 829 604 (7311) 56 946 975 (27 597) 3 235 815 (1488) 2 311 022 (1113) 2 290 920 (974) 393 064 (189) 40 319 184 (19 621) 1 143 654 (513) Central sub-Saharan Africa 62 933 394 (97 393) 17 537 885 (26 368) 23 972 738 (38 407) 7 030 042 (11 074) 23 492 513 (36 235) 1 342 987 (1851) 730 306 (1114) 708 630 (896) 182 575 (284) 22 506 678 (35 097) 387 546 (525) YLDs Global 153 738 779 (6151) 28 929 775 (1154) 18 299 658 (732) 7 170 928 (288) 4 676 015 (187) 3 780 968 (150) 3 169 555 (127) 2 779 412 (109) 2 512 221 (100) 2 471 320 (99) 2 222 098 (89) SDI High 13 873 053 (5004) 2 268 673 (844) 1 903 844 (690) 773 452 (282) 333 309 (124) 34 413 (13) 525 720 (192) 29 393 (11) 142 020 (52) 5 (0) 234 807 (86) High-middle 26 454 141 (5549) 4 564 413 (981) 3 547 547 (745) 1 445 951 (306) 717 114 (154) 402 587 (88) 684 409 (145) 257 535 (57) 444 219 (94) 53 564 (12) 489 408 (104) Middle 38 514 608 (5656) 7 075 700 (1047) 4 977 816 (731) 1 785 676 (264) 1 328 362 (197) 856 900 (126) 847 432 (125) 645 602 (95) 618 917 (91) 193 942 (29) 765 470 (113) Low-middle 53 056 669 (6878) 11 300 366 (1445) 5 505 932 (715) 2 153 411 (278) 1 798 392 (230) 1 668 008 (211) 840 951 (108) 1 348 008 (168) 1 042 262 (135) 1 097 615 (140) 634 0

5 700 (1047) 4 977 816 (731) 1 785 676 (264) 1 328 362 (197) 856 900 (126) 847 432 (125) 645 602 (95) 618 917 (91) 193 942 (29) 765 470 (113) Low-middle 53 056 669 (6878) 11 300 366 (1445) 5 505 932 (715) 2 153 411 (278) 1 798 392 (230) 1 668 008 (211) 840 951 (108) 1 348 008 (168) 1 042 262 (135) 1 097 615 (140) 634 0 62 (82) Low 21 751 788 (7341) 3 706 095 (1179) 2 352 608 (795) 1 005 581 (341) 497 047 (162) 816 972 (254) 269 420 (89) 497 910 (144) 263 420 (87) 1 125 089 (368) 96 395 (31) GBD region High-income North America 4 837 450 (5182) 768 122 (858) 638 731 (694) 260 458 (287) 129 016 (142) 768 (1) 215 027 (236) 7 (0) 50 603 (55) 0 (0) 100 598 (110) Australasia 390 234 (5303) 59 268 (858) 49 823 (686) 37 895 (526) 4278 (62) 271 (4) 15 090 (210) 0 (0) 2535 (35) 0 (0) 6068 (84) High-income Asia Pacific 1 520 962 (4430) 251 785 (760) 262 974 (782) 77 277 (233) 20 599 (63) 1068 (4) 64 847 (196) 2 (0) 14 260 (43) 5 (0) 16 610 (50) Western Europe 4 510 597 (4953) 737 447 (837) 551 452 (607) 307 706 (339) 126 465 (142) 6406 (8) 141 916 (158) 5 (0) 44 328 (49) 0 (0) 65 889 (73) Southern Latin America 1 077 954 (5267) 200 602 (1002) 140 337 (687) 79 492 (390) 12 957 (64) 2432 (12) 32 320 (159) 39 (0) 16 539 (81) 0 (0) 17 326 (85) Eastern Europe 2 064 354 (5313) 396 743 (992) 276 676 (703) 76 460 (199) 35 684 (90) 18 078 (40) 58 397 (145) 18 042 (38) 17 728 (45) 0 (0) 34 663 (86) Central Europe 1 161 550 (4923) 191 996 (863) 165 603 (702) 48 427 (209) 18 791 (84) 4049 (18) 40 841 (176) 10 900 (49) 14 567 (63) 0 (0) 18 275 (79) Central Asia 1 682 951 (5565) 336 361 (1095) 204 701 (676) 58 557 (198) 37 625 (124) 31 413 (94) 32 170 (104) 25 109 (71) 41 611 (139) 0 (0) 48 759 (158) Central Latin America 4 890 942 (5182) 755 908 (814) 706 785 (753) 423 348 (454) 91 052 (97) 90 295 (101) 130 554 (140) 23 811 (27) 113 023 (121) 4853 (5) 61 239 (66) Andean Latin America 1 350 888 (6042) 273 511 (1228) 181 745 (813) 122 875 (551) 13 153 (59) 39 559 (177) 24 264 (109) 3687 (17) 21 852 (98) 1476 (7) 21 706 (97) Caribbean 1 011 612 (6566) 189 319 (1248) 134 395 (875) 110 321 (721) 10 623 (70) 26 205 (177) 17 342 (113) 7640 (53) 13 827 (90) 545 (4) 30 241 (198) Tropical Latin America 4 242 662 (6106) 718 422 (1091) 537 515 (781) 433 087 (633) 51 193 (76) 95 950 (154) 91 862 (135) 14 096 (24) 50 513 (74) 5329 (8) 92 446 (136) East Asia 15 445 604 (4670) 2 583 236 (796) 2 497 965 (755) 449 977 (137) 851 152 (262) 98 975 (30) 398 918 (121) 127 719 (39) 186 159 (57) 1426 (0) 315 729 (96) Southea

merica 4 242 662 (6106) 718 422 (1091) 537 515 (781) 433 087 (633) 51 193 (76) 95 950 (154) 91 862 (135) 14 096 (24) 50 513 (74) 5329 (8) 92 446 (136) East Asia 15 445 604 (4670) 2 583 236 (796) 2 497 965 (755) 449 977 (137) 851 152 (262) 98 975 (30) 398 918 (121) 127 719 (39) 186 159 (57) 1426 (0) 315 729 (96) Southea st Asia 13 292 217 (5774) 2 030 377 (893) 1 954 826 (849) 800 218 (349) 359 444 (158) 353 453 (156) 254 392 (111) 238 088 (106) 236 992 (103) 140 513 (62) 343 313 (150) Oceania 344 463 (6914) 49 509 (987) 57 805 (1154) 24 064 (481) 6559 (130) 7041 (138) 5006 (100) 2933 (56) 4823 (96) 13 309 (265) 7384 (147) North Africa and Middle East 13 596 683 (6087) 2 051 891 (903) 1 462 893 (653) 698 541 (313) 499 378 (221) 353 666 (148) 265 386 (117) 256 463 (105) 377 985 (167) 42 659 (19) 183 712 (81) South Asia 45 458 863 (6839) 10 764 532 (1635) 4 597 805 (690) 1 445 214 (217) 1 485 374 (225) 1 338 785 (207) 898 954 (136) 1 243 899 (196) 907 416 (136) 328 591 (50) 689 132 (104) Southern sub-Saharan Africa 2 006 303 (6418) 302 072 (967) 248 647 (793) 155 539 (503) 19 923 (64) 64 622 (199) 42 545 (135) 19 996 (60) 20 611 (66) 13 685 (44) 20 124 (64) Western sub-Saharan Africa 15 184 327 (7224) 3 262 703 (1463) 1 433 936 (694) 574 266 (279) 530 793 (242) 505 014 (218) 162 495 (75) 416 459 (165) 167 049 (78) 1 031 751 (475) 55 031 (24) Eastern sub-Saharan Africa 13 844 433 (6824) 2 250 727 (1048) 1 589 604 (783) 670 667 (331) 268 953 (128) 525 699 (242) 212 384 (102) 283 413 (121) 145 305 (70) 564 410 (271) 76 371 (36) Central sub-Saharan Africa 5 823 721 (9255) 755 235 (1114) 605 429 (944) 316 530 (498) 102 992 (154) 217 210 (299) 64 833 (98) 87 093 (110) 64 486 (100) 322 762 (494) 17 470 (25) Abbreviations: GBD, Global Burden of Diseases, Injuries, and Risk Factors Study; SDI, Socio-demographic Index.

(70) 564 410 (271) 76 371 (36) Central sub-Saharan Africa 5 823 721 (9255) 755 235 (1114) 605 429 (944) 316 530 (498) 102 992 (154) 217 210 (299) 64 833 (98) 87 093 (110) 64 486 (100) 322 762 (494) 17 470 (25) Abbreviations: GBD, Global Burden of Diseases, Injuries, and Risk Factors Study; SDI, Socio-demographic Index. Iron-deficiency anemia was the highest-ranking level 3 cause of YLDs in children and adolescents, followed by skin and subcutaneous diseases, asthma, hemoglobinopathies and hemolytic anemias, diarrheal diseases, congenital anomalies, protein-energy malnutrition, epilepsy, malaria, and neonatal complications of preterm birth. Among children 5 years or younger, there was higher relative importance of disability owing to protein-energy malnutrition (third highest-ranking cause) and diarrheal diseases (fourth highest-ranking cause), as well as neonatal encephalopathy (ninth highest-ranking cause) and other neonatal disorders (tenth highest-ranking cause). Although the age-standardized prevalence and rate of YLDs decreased for most conditions, it increased for malaria and congenital anomalies. The burden of most conditions either decreased with increasing SDI or was relatively constant across different SDI quintiles. Two exceptions were congenital anomalies, which increased with increasing SDI, and hemoglobinopathies, which were highest in low- to middle-SDI geographical areas.

aria and congenital anomalies. The burden of most conditions either decreased with increasing SDI or was relatively constant across different SDI quintiles. Two exceptions were congenital anomalies, which increased with increasing SDI, and hemoglobinopathies, which were highest in low- to middle-SDI geographical areas. Disability Burden From Conditions With Multiple Causes Many clinical conditions cause significant disease burden in children and adolescents, but because they can arise from multiple causes, their effect is not obvious when examining causes of GBD. Examples that would be in the top 10 global causes of YLDs if considered alone are anemia, developmental intellectual disability, epilepsy, hearing loss, and vision loss. For example, while iron-deficiency anemia was the leading level 3 cause of disability, it accounted for only about two-thirds of total anemia in children and adolescents 19 years or younger in 2015 (eTable 5 in the Supplement), and each case tended to be less severe than other etiologic causes of anemia. Infectious diseases, hemoglobinopathies, malaria, hookworm, gynecologic conditions, and gastritis and duodenitis were other important causes of anemia in children and adolescents. Neonatal disorders were the most common nonidiopathic cause of both developmental intellectual disability (eTable 6 in the Supplement) and epilepsy (eTable 7 in the Supplement). Autism, iodine deficiency, and congenital disorders were important causes of intellectual disability, while much of the rest of intellectual disability and much of nonidiopathic epilepsy were secondary to infectious causes, especially malaria and meningitis. Hearing and vision loss also contributed to the disease burden within children and adolescents 19 years or younger, with age-associated and other hearing loss accounting for most hearing loss burden (eTable 8 in the Supplement). For vision loss, a range of causes contributed to the burden among children and adolescents 19 years or younger, including neonatal disorders and nutritional deficiencies (eTable 9 in the Supplement).

unger, with age-associated and other hearing loss accounting for most hearing loss burden (eTable 8 in the Supplement). For vision loss, a range of causes contributed to the burden among children and adolescents 19 years or younger, including neonatal disorders and nutritional deficiencies (eTable 9 in the Supplement). Pregnancy Complications in Adolescents Mortality was the primary driver of health loss owing to maternal disorders in adolescents. The global maternal mortality ratio per 100 000 live births was 278 (95% UI, 229-339) and 142 (95% UI, 123-166) in 2015 for children and adolescents aged 10 to 14 and 15 to 19 years, respectively, causing 1343 (95% UI, 1105-1640) and 26 855 (95% UI, 23 254-31 521) maternal deaths. Both age groups had a maternal mortality ratio higher than the global aggregate of 132 (95% UI, 117-153) seen in women aged 25 to 29 years (eTable 10 in the Supplement and GBD 2015 maternal mortality publication18). The mean annualized decline in the maternal mortality ratio among adolescents aged 10 to 19 years was only 1.4% (95% UI, 0.8%-2.0%), which was slower than the global improvement rate of 2.6% for overall maternal mortality. Maternal hemorrhage was the highest-ranked level 3 cause of maternal mortality globally, driven largely by its prominence in low-SDI geographical areas where teenage pregnancy and the burden of maternal mortality are the highest (eFigures 3 and 4 in the Supplement). Other top-ranked causes of maternal mortality included maternal hypertensive disorders, other direct maternal disorders (eg, pulmonary embolism, cardiomyopathy, and surgical and anesthetic complications), and the combined category of abortion, ectopic pregnancy, and/or miscarriage. The risk of nonfatal complications during pregnancy is also higher in adolescents than in women in their 20s (eFigure 5 in the Supplement). Abortion, ectopic pregnancy, and/or miscarriage is the most common disabling outcome of pregnancy among adolescents, followed by maternal hemorrhage, maternal hypertensive disorders, maternal sepsis and other maternal infections, and obstructed labor.

dolescents than in women in their 20s (eFigure 5 in the Supplement). Abortion, ectopic pregnancy, and/or miscarriage is the most common disabling outcome of pregnancy among adolescents, followed by maternal hemorrhage, maternal hypertensive disorders, maternal sepsis and other maternal infections, and obstructed labor. Ranking and Trends of DALYs in Children and Adolescents Ranking of the 25 leading level 3 causes of DALYs in 1990, 2005, and 2015, along with the changes in total number, all-ages rate, and age-standardized rate, are shown in Figure 2 for children and adolescents 19 years or younger. Corresponding DALY rankings disaggregated by SDI quintile are in eFigure 6A-E in the Supplement. Between 1990 and 2005, more than 40% declines in DALYs in children and adolescents 19 years or younger were seen for LRIs, diarrheal diseases, measles, tetanus, drowning, and neonatal hemolytic disease and other neonatal jaundice; similar declines between 2005 and 2015 were seen for malaria, measles, tetanus, and neonatal hemolytic disease and other neonatal jaundice. The most significant increase was for HIV and AIDS, which increased by close to 600% to rank 11th globally in 2005, a ranking that stayed largely static through 2015 despite a nearly 30% drop in DALYs from 2005 to 2015. Malaria and iron-deficiency anemia were the other group I conditions with significantly increased DALYs between 1990 and 2005; DALYs for both conditions also subsequently decreased significantly by 2015. Several NCDs increased in ranking from 1990 to 2015 for children and adolescents 19 years or younger. Some diseases—including congenital anomalies, asthma, and hemoglobinopathies and hemolytic anemias—increased in ranking despite registering decreased age-standardized DALY rates for each time period. In contrast, other causes—including sense organ diseases, skin diseases, and mental and substance abuse disorders such as depression, anxiety, and conduct disorder—increased in ranking with largely unchanged or slightly increased global age-standardized DALY rates from 1990 to 2015. Road injuries and drowning were the 2 highest-ranking injuries in terms of DALYs despite significant decreases from 1990 to 2015.

ance abuse disorders such as depression, anxiety, and conduct disorder—increased in ranking with largely unchanged or slightly increased global age-standardized DALY rates from 1990 to 2015. Road injuries and drowning were the 2 highest-ranking injuries in terms of DALYs despite significant decreases from 1990 to 2015. Figure 2. Leading Level 3 Causes of Global Disability-Adjusted Life Years (DALYs) in the Global Burden of Diseases, Injuries, and Risk Factors Study This figure shows the rankings for the top 25 causes of global disability-adjusted life years among children and adolescents 19 years or younger at the global level in 1990, 2005, and 2015. Lines connecting the boxes illustrate changes in ranking. Any cause that appears in the top 25 in any year is listed, along with its ranking during each year. Group I causes (infectious, neonatal, nutritional, and maternal) are shown in gray, noncommunicable diseases in red, and injuries in green. Changes in total DALYs are in the first column next to 2005, followed by changes in all-ages rates of DALYs, and age-standardized rates of DALYs. Statistically significant differences appear in bold. HIV indicates human immunodeficiency virus, and STI, sexually transmitted infection.

ases in red, and injuries in green. Changes in total DALYs are in the first column next to 2005, followed by changes in all-ages rates of DALYs, and age-standardized rates of DALYs. Statistically significant differences appear in bold. HIV indicates human immunodeficiency virus, and STI, sexually transmitted infection. SDI and Epidemiologic Transition in Children and Adolescents Figure 3A shows the mean association between SDI and cause-specific YLLs and YLDs from 1990 to 2015 for all level 2 causes. Nonlinearity of associations at times followed the nonlinearity of the SDI itself. There is a clear and substantial downward gradient in child and adolescent health loss with increasing SDI. Years of life lost are the dominant component of DALYs in the geographical areas with the lowest SDI, a trend that continues until an SDI of roughly 0.80, after which YLDs become responsible for a larger proportion of DALYs. There is also a clear increase in the all-ages rate of YLDs in the geographical areas with the highest SDI to the point where, at the highest SDI, 67% of all DALYs are owing to nonfatal health outcomes. Figure 3B shows the corresponding information displayed as a proportion of total rates of YLL and YLD owing to each level 2 cause at each SDI level. For most level 2 causes, the proportion of all YLLs owing to group I causes decreases with increasing SDI. The exceptions are neonatal disorders and HIV and AIDS and tuberculosis, which increased in relative importance with increasing SDI. In the geographical areas with the highest SDI, self-harm and interpersonal violence, other NCDs, and neoplasms were responsible for an increasing proportion of YLLs. The proportion of YLDs owing to group I causes similarly decreased with increasing SDI, while the proportion owing to NCDs generally increased. Most level 3 causes followed this same pattern, with 2 notable exceptions among the top causes of child and adolescent DALYs: congenital anomalies (eFigure 7A in the Supplement) and neonatal disorders (eFigure 7B in the Supplement). For congenital anomalies, there was a consistent decrease in the rate of YLLs, with increasing SDI for all causes, but increases in the rate of YLDs for most level 3 congenital anomalies, especially congenital heart anomalies and other congenital anomalies.

lement) and neonatal disorders (eFigure 7B in the Supplement). For congenital anomalies, there was a consistent decrease in the rate of YLLs, with increasing SDI for all causes, but increases in the rate of YLDs for most level 3 congenital anomalies, especially congenital heart anomalies and other congenital anomalies. For neonatal disorders, increasing SDI was associated with consistent improvements in neonatal sepsis, hemolytic disease of the newborn, and other neonatal disorders but not for preterm birth complications. Rates of DALYs for preterm birth complications increased initially, and there was little suggestion of further improvement beyond an SDI of 0.85 for any neonatal disorder, especially preterm birth complications.

sis, hemolytic disease of the newborn, and other neonatal disorders but not for preterm birth complications. Rates of DALYs for preterm birth complications increased initially, and there was little suggestion of further improvement beyond an SDI of 0.85 for any neonatal disorder, especially preterm birth complications. Figure 3. Expected Association Between Rates of Years of Life Lost (YLL) and Years Lived With Disability (YLD) Rates With Socio-demographic Index (SDI) for 21 Level 2 Causes in the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) A, Expected association between rates of YLL and YLD with SDI for the 21 GBD level 2 causes in children and adolescents 19 years or younger, both sexes, 1990-2015. Each geography is assigned an SDI value for each year, and nonlinear spline regressions are used to find the average relationship between SDI and cause-specific burden rates. B, Expected association between rates of YLL and YLD and SDI for the 21 GBD level 2 causes as a proportion of total rates of YLL and YLD in children and adolescents 19 years or younger, both sexes, 1990-2015. Each geographical area is assigned an SDI value for each year, and nonlinear spline regressions are used to find the mean association between SDI and cause-specific rates of disease burden. HIV indicates human immunodeficiency virus.

of YLL and YLD in children and adolescents 19 years or younger, both sexes, 1990-2015. Each geographical area is assigned an SDI value for each year, and nonlinear spline regressions are used to find the mean association between SDI and cause-specific rates of disease burden. HIV indicates human immunodeficiency virus. The mean association between SDI and sex-specific rates of DALYs of level 3 causes also showed broad consistency in that the burden owing to most causes decreased with increasing SDI (eFigure 8A-V in the Supplement). Exceptions were the level 3 causes that consistently increased with increasing SDI—musculoskeletal and mental and substance abuse disorders—as well as those that either peaked in high-middle and middle SDI or improved only marginally until the highest SDI levels—asthma, epilepsy, migraine, road injuries, self-harm, interpersonal violence, and neoplasms such as leukemia, lymphoma, brain cancer, and other neoplasms. Sex differences were notable in that some conditions, such as neonatal disorders, neoplasms, nutritional deficiencies, hemoglobinopathies and hemolytic anemias, epilepsy, and all categories of injuries, had higher all-ages rates of DALYs in males, while others, such as migraine, gastrointestinal disorders, musculoskeletal disorders, and congenital anomalies, were higher in females.

onatal disorders, neoplasms, nutritional deficiencies, hemoglobinopathies and hemolytic anemias, epilepsy, and all categories of injuries, had higher all-ages rates of DALYs in males, while others, such as migraine, gastrointestinal disorders, musculoskeletal disorders, and congenital anomalies, were higher in females. Discussion We found widespread reductions in total disease burden among children and adolescents, but progress has been unequal. In 2015, an even greater share of global mortality burden was concentrated in the lowest-SDI countries than it was in 1990, and there has been a significant increase in the global proportion of nonfatal disease burden. Global reductions in disease burden from infectious, neonatal, maternal, and nutrition-associated causes have been accompanied by a growing importance of NCDs and injuries. Neonatal disorders and congenital anomalies remain large and, in some cases, growing problems in many countries. Nutritional deficiencies, along with infections such as HIV and AIDS, diarrhea, LRIs, malaria, intestinal infectious diseases, and vaccine-preventable diseases, also still cause enormous health loss in some countries. Some populations of children and adolescents have been struck by the massive effect of war on health, while others are struggling with the detrimental effects of road injuries, drowning, self-harm, and interpersonal violence, and the growing importance of NCDs such as mental and substance use disorders, cancer, congenital anomalies, and hemoglobinopathies. Leading causes of death and disability vary as a function of age, sex, and SDI status, so the precise challenges for each country may be very different. The SDGs set a series of absolute time-bound targets for improving health, the environment, and societal development. By design, the SDG agenda is broader than the Millennium Development Goal framework, in which there was only one child target (Millennium Development Goal 4: reduce mortality by two-thirds in children younger than 5 years). By broadening the agenda and shifting to absolute targets, the SDGs provide an excellent starting point to judge progress at the country level and should rightfully focus attention on the countries with the most progress yet to achieve.25 Tracking the entire spectrum of disease in children and adolescents facilitates monitoring of the SDGs but can also highlight non-SDG health challenges, and it should be used to inform final decision making with respect to SDG indicators.

htfully focus attention on the countries with the most progress yet to achieve.25 Tracking the entire spectrum of disease in children and adolescents facilitates monitoring of the SDGs but can also highlight non-SDG health challenges, and it should be used to inform final decision making with respect to SDG indicators. Besides neonatal mortality and mortality in children younger than 5 years,3 other health-associated SDGs are important for understanding disease burden in children and adolescents given the age pattern of these conditions. Examples include infection-associated SDG targets addressing malaria, HIV and tuberculosis, neglected tropical diseases, severe malnutrition, and access to safe water, sanitation, and hygiene. Our analysis shows that these SDG indicators are improving for children and adolescents in most geographical areas. Sustainable Development Goals focusing on reproductive health targets such as maternal mortality, adolescent fertility, universal access to modern contraception, skilled birth attendance, and neonatal support services are also an important part of the discussion about child and adolescent health loss given the continued importance of teenage pregnancy, pregnancy-associated complications, and maternal death in many settings as well as the intricate links between maternal and child health. Many of the injury-associated SDG targets also are relevant to tracking child health progress, including those on disaster preparedness, road injuries, poisoning, self-harm, interpersonal violence, and war.

plications, and maternal death in many settings as well as the intricate links between maternal and child health. Many of the injury-associated SDG targets also are relevant to tracking child health progress, including those on disaster preparedness, road injuries, poisoning, self-harm, interpersonal violence, and war. Most of the SDG targets addressing NCDs, such as those associated with mental health conditions and cardiovascular diseases, not only focus on NCDs that are primarily problems in adults but also specifically exclude children and adolescents. This exclusion is problematic, especially because of the growing importance found in our study of mental health, substance abuse, and self-harm among children and adolescents. Half of all mental illnesses begin by age 14 years and three-quarters begin by the mid-20s. If untreated, these conditions can predispose to self-harm and severely influence children’s development, educational attainment, and long-term fulfillment and economic potential.26 Other SDGs, such as those addressing education and sex equality, are not specifically associated with health but can have a significant effect on the health of children and adolescents.

o self-harm and severely influence children’s development, educational attainment, and long-term fulfillment and economic potential.26 Other SDGs, such as those addressing education and sex equality, are not specifically associated with health but can have a significant effect on the health of children and adolescents. One possible explanation for growing inequality in disease burden among children and adolescents is that many of the geographical areas with the lowest SDIs have not historically been significant recipients of development assistance for health (DAH).27,28 Although development assistance for child and newborn health has been among the fastest-growing focus areas of DAH since 1990 and is one of the few areas in which funding has continued to increase since 2010, it has been uneven. For example, most countries in Central and Western sub-Saharan Africa received less than half the DAH per all-cause DALY in 2010 to 2012 received by countries in Southern sub-Saharan Africa, Eastern sub-Saharan Africa, and Central Latin America.29 Many DAH programs have concentrated on funding widespread delivery of proven preventive measures, including vaccines; nutritional support; maternal education; improved water, sanitation, and hygiene; and treatment of diarrhea, LRIs, and HIV and AIDS. Synergistic maternal health programs have also enjoyed strong support during this period. These programs must continue and be expanded, especially in low-SDI settings in which progress in child and adolescent health has been comparatively slow, but focus must also be turned to strengthening health systems, especially in geographical areas in which such systems are underdeveloped or have been weakened by conflict.

programs must continue and be expanded, especially in low-SDI settings in which progress in child and adolescent health has been comparatively slow, but focus must also be turned to strengthening health systems, especially in geographical areas in which such systems are underdeveloped or have been weakened by conflict. This finding leads to another possible explanation for growing inequality; namely, that there has been inadequate focus on increasing local health system capacity and capabilities. High-SDI settings have seen improved prevention and better outcomes for many childhood illnesses, such as neonatal disorders, as well as many congenital birth defects, injuries, and cancers, such as leukemia. Improved treatment has led to increasing numbers of children now reaching adulthood with ongoing medical needs.30,31,32,33,34 These medical advances have not been as readily realized in geographical areas with lower SDIs or in older children and adolescents.35,36 Sustained investment is needed to improve prevention, diagnosis, and treatment for causes not traditionally targeted by DAH, including strengthening of workforces and facilities,37,38,39 cooperation between health centers in the same region,40 clinician specialization,41,42,43,44 improved surgical and anesthetic care,34 evidence-based case management for life-threatening complications,45,46 injury prevention, and active screening to identify and treat high-risk children (eg, those with congenital heart disease47,48,49 and hemoglobinopathies50,51). Integration of health care services across facilities and specialties to meet the varied needs of children and adolescents is paramount to successful intervention programs, especially in geographical areas with middle SDIs where there is a high likelihood of comorbid NCDs, injuries, and group I conditions.

thies50,51). Integration of health care services across facilities and specialties to meet the varied needs of children and adolescents is paramount to successful intervention programs, especially in geographical areas with middle SDIs where there is a high likelihood of comorbid NCDs, injuries, and group I conditions. Indeed, the World Health Organization Global Strategy for Women’s, Children’s, and Adolescents’ Health (2016-2030)9 addresses both explanations for growing inequality. In addition to advocating for programs supported by DAH, it recommends specific actions to address the interplay between environment, economics, and health through multisector action including investments and policy implementation, leadership engagement at regional and country levels, and building resilience of health systems through workforce development, human capital investment, sex equality, and youth empowerment.

e interplay between environment, economics, and health through multisector action including investments and policy implementation, leadership engagement at regional and country levels, and building resilience of health systems through workforce development, human capital investment, sex equality, and youth empowerment. One aspect of child and adolescent health that is not comprehensively addressed by the World Health Organization Global Strategy for Women’s, Children’s, and Adolescents’ Health (2016-2030)9 is injuries. Minimizing the burden of injuries in children and adolescents requires the implementation of specific prevention policies where they do not exist and support for treatment and rehabilitation of injured youth. First and foremost, this implementation necessitates strict limitations on child labor and elimination of child slavery.52 To reduce suicide, Sri Lanka and South Korea have both restricted the availability of pesticides with good effect,53,54 while Australia has taken the step of implementing comprehensive mental health screening and treatment.55 Along with suicide, interpersonal violence may be partially addressed via implementation of more stringent gun regulations,56 although many other social, demographic, and societal factors, such as poverty, education, and drug use, also must be addressed in policies aimed at reducing homicide.57,58 Road injuries, falls, and interpersonal violence are common etiologic causes of traumatic brain injury, a condition that can have significant mortality and morbidity.59,60 Policies requiring seat belts, appropriate use of car seats, and helmets for cyclists can greatly reduce the risk of traumatic brain injury following road injuries.61 Timely hospital care and ongoing rehabilitation services are necessary to minimize the long-term health burden of traumatic brain injury,62,63 although optimal approaches are still in development.64

e of car seats, and helmets for cyclists can greatly reduce the risk of traumatic brain injury following road injuries.61 Timely hospital care and ongoing rehabilitation services are necessary to minimize the long-term health burden of traumatic brain injury,62,63 although optimal approaches are still in development.64 In addition to effective medical programs, comprehensive community-based approaches are needed to maximize the development and empowerment of children and adolescents at the population level. One approach that could serve as a model—and may be possible in many high-SDI countries—is the Healthy Child, Healthy Future campaign in Northern Ireland.65 The central tenet of the program is to promote a shift from an approach to child health that relies on medical screening to identify children with treatable conditions to an approach that also emphasizes promotion of general health, primary prevention of diseases and injuries, and active intervention for children at risk. Involving parents, schools, and communities in this effort is seen as a crucial part because they are the entities who have the most influence on the physical and social environment of children, and their own personal behaviors are likely to have a direct effect on children in their care.

rvention for children at risk. Involving parents, schools, and communities in this effort is seen as a crucial part because they are the entities who have the most influence on the physical and social environment of children, and their own personal behaviors are likely to have a direct effect on children in their care. Limitations This analysis has several limitations. First, it is based on the GBD 2015 study results and is therefore subject to the same potential problems as the overall study, including variations in availability and quality of data, variations in coding practices, delays in release of data after their collection, paucity of data following conflict and natural disasters, and incomplete information or potential underreporting on nonfatal health outcomes in many geographical areas. Despite potential limitations, drawing conclusions on levels of and trends in child and adolescent health loss from this central study at least ensures that comparisons are all internally consistent with one another. Second, in this study we concentrated on only the leading global causes of fatal and nonfatal health loss in the aggregate age groups of children and adolescents 19 years or younger. This approach allows reporting on general patterns of health loss in children and adolescents, but given that childhood is a period of rapid development and change, it does not highlight less common conditions and may obscure some age-specific and geography-specific subtleties of epidemiologic factors of disease and injury. Third, data availability for children aged 5 to 14 years is not as robust as that for those 5 years or younger and individuals aged 15 to 19 years. Estimates of all-cause mortality were thus based primarily on extensive data sources in children 5 years or younger and individuals aged 15 to 19 years, with estimates for intervening age groups derived from a large collection of empirical life tables and use of survival history data from surveys and censuses.16 We found broad agreement between our all-cause mortality estimates and primary data sources from the Sample Registration System and Demographic and Health Surveys in India, which supports the GBD findings. Hill and colleagues66 have argued against this approach on the basis that it could underestimate mortality in populations in which interventions targeted at children 5 years or younger have not also led to improved survival in children aged 5 to 14 years.

ic and Health Surveys in India, which supports the GBD findings. Hill and colleagues66 have argued against this approach on the basis that it could underestimate mortality in populations in which interventions targeted at children 5 years or younger have not also led to improved survival in children aged 5 to 14 years. The implication is that such underestimation would dilute the perceived importance of causes that disproportionately affect children older than 5 years (eg, injuries, NCDs, and maternal disorders), but even if that were the case, it should have minimal effect on the interpretation of levels and trends within each age group or country. Fourth, given the geography-centric approach to GBD analyses, we have limited ability to evaluate disease burden in subpopulations that are not geographically based, such as refugees and many indigenous peoples. Fifth, while GBD 2015 has analyzed the total burden and underlying etiologic causes of several broad conditions, such as anemia and developmental intellectual disability, there are others that may be particularly important in children and adolescents that have not received the same level of scrutiny, including, for example, traumatic brain injury, hydrocephalus, and sepsis. Sixth, our quantification of disease burden focuses only on affected individuals and therefore does not capture indirect effects, including effects on education and development of child labor,52 the long-term effects of war,67 or the burden on parents, siblings, and communities of caring for ill or injured children and adolescents.68

tification of disease burden focuses only on affected individuals and therefore does not capture indirect effects, including effects on education and development of child labor,52 the long-term effects of war,67 or the burden on parents, siblings, and communities of caring for ill or injured children and adolescents.68 Conclusions Timely, robust, and comprehensive assessment of disease burden among children and adolescents provides information that is essential to health policy decision making in countries at all points along the spectrum of economic development. Understanding the burden of disease and how it is changing helps identify context-specific successes, unmet needs, and future challenges. Child and adolescent health has dramatically improved from 1990 to 2015 throughout the world. International attention and investment have led to improvements in many infectious and nutritional diseases, but progress has been uneven. The burden of disease during childhood and adolescence is now even more concentrated in lower-SDI countries and territories than it was a generation ago, and there is an ongoing epidemiologic transition with a growing relative burden of NCDs and injuries. If we are going to continue the current pace of improvement in child and adolescent health, we must invest in better data collection, continue to monitor trends in population disease burden, and adapt health systems to meet the ongoing and changing needs of children and adolescents so that all can have a chance to grow up to be healthy.

continue the current pace of improvement in child and adolescent health, we must invest in better data collection, continue to monitor trends in population disease burden, and adapt health systems to meet the ongoing and changing needs of children and adolescents so that all can have a chance to grow up to be healthy. Supplement. eFigure 1. Socio-Demographic Index quintiles by GBD subnational level 1 geography, 2015 eFigure 2a. Top 25 Global Causes of Death, by 5 SDI quintiles and 21 GBD regions, Aged 0-6 days, Females & Males, 2015 eFigure 2b. Top 25 Global Causes of Death, by 5 SDI quintiles and 21 GBD regions, Aged 7-27 days, Females & Males, 2015 eFigure 2c. Top 25 Global Causes of Death, by 5 SDI quintiles and 21 GBD regions, Aged 28-364 days, Females & Males, 2015 eFigure 2d. Top 25 Global Causes of Death, by 5 SDI quintiles and 21 GBD regions, Aged 1-4 years, Females & Males, 2015 eFigure 2e. Top 25 Global Causes of Death, by 5 SDI quintiles and 21 GBD regions, Aged 5-9 years, Females & Males, 2015 eFigure 2f. Top 25 Global Causes of Death, by 5 SDI quintiles and 21 GBD regions, Aged 10-14 years, Females & Males, 2015 eFigure 2g. Top 25 Global Causes of Death, by 5 SDI quintiles and 21 GBD regions, Aged 15-19 years, Females & Males, 2015 eFigure 3. Top Global Causes of Maternal Mortality, by 5 SDI quintiles and 21 GBD regions, Aged 10 to 19 Years, Female, 2015 eFigure 4. The expected relationship between age, cause-specific YLLs and YLDs, and population with the Socio-Demographic Index (SDI) for the level 3 maternal causes, Females, 1990 to 2015

eFigure 2g. Top 25 Global Causes of Death, by 5 SDI quintiles and 21 GBD regions, Aged 15-19 years, Females & Males, 2015 eFigure 3. Top Global Causes of Maternal Mortality, by 5 SDI quintiles and 21 GBD regions, Aged 10 to 19 Years, Female, 2015 eFigure 4. The expected relationship between age, cause-specific YLLs and YLDs, and population with the Socio-Demographic Index (SDI) for the level 3 maternal causes, Females, 1990 to 2015 eFigure 5. Global pregnancy complication ratio (events per 100 live births) by type of complication and age group in 2015 eFigure 6a. Leading 25 GBD cause hierarchy level 3 causes of low SDI DALYs for both sexes combined for 1990, 2005 and 2015, 0-19 years eFigure 6b. Leading 25 GBD cause hierarchy level 3 causes of low-middle SDI DALYs for both sexes combined for 1990, 2005 and 2015, 0-19 years eFigure 6c. Leading 25 GBD cause hierarchy level 3 causes of middle SDI DALYs for both sexes combined for 1990, 2005 and 2015, 0-19 years eFigure 6d. Leading 25 GBD cause hierarchy level 3 causes of middle-high SDI DALYs for both sexes combined for 1990, 2005 and 2015, 0-19 years eFigure 6e. Leading 25 GBD cause hierarchy level 3 causes of high SDI DALYs for both sexes combined for 1990, 2005 and 2015, 0-19 years eFigure 7a. The expected relationship between YLL and YLD rates with SDI for GBD level 3 congenital causes, aged 0 to 19 years, both sexes, 1990 to 2015 eFigure 7b. The expected relationship between YLL and YLD rates with SDI for GBD level 3 neonatal causes, aged 0 to 19 years, both sexes, 1990 to 2015

eFigure 6e. Leading 25 GBD cause hierarchy level 3 causes of high SDI DALYs for both sexes combined for 1990, 2005 and 2015, 0-19 years eFigure 7a. The expected relationship between YLL and YLD rates with SDI for GBD level 3 congenital causes, aged 0 to 19 years, both sexes, 1990 to 2015 eFigure 7b. The expected relationship between YLL and YLD rates with SDI for GBD level 3 neonatal causes, aged 0 to 19 years, both sexes, 1990 to 2015 eFigure 8. The expected relationship between cause-specific all-ages DALY rates for 0- 19 years, and Sociodemographic Index (SDI) for males (left) and females (right) eTable 1. Number of Deaths, Death Rates (per 100,000 population), and Cumulative Percent Change with 95% Uncertainty Intervals (UI) for the Top 10 Global Causes of Death in 195 Countries and Territories, Aged 0 to 19, Both Sexes, 1990 and 2015 eTable 2. Number of Deaths, Death Rates (per 100,000 population), and Cumulative Percent Change with 95% Uncertainty Intervals (UI) for the Top 10 Global Causes of Death in 195 Countries and Territories, Aged Under 5 Years, Both Sexes, 1990 and 2015 eTable 3. Prevalent Cases, Rates (per 100,000 population), Years Lived with Disability (YLDs), and Cumulative Percent Change with 95% Uncertainty Interval (UI) for the Top 10 Global Causes of YLDs in Children and Adolescents in 195 Countries and Territories, Aged 0 to 19 Years, Both Sexes, 1990 and 2015

eTable 2. Number of Deaths, Death Rates (per 100,000 population), and Cumulative Percent Change with 95% Uncertainty Intervals (UI) for the Top 10 Global Causes of Death in 195 Countries and Territories, Aged Under 5 Years, Both Sexes, 1990 and 2015 eTable 3. Prevalent Cases, Rates (per 100,000 population), Years Lived with Disability (YLDs), and Cumulative Percent Change with 95% Uncertainty Interval (UI) for the Top 10 Global Causes of YLDs in Children and Adolescents in 195 Countries and Territories, Aged 0 to 19 Years, Both Sexes, 1990 and 2015 eTable 4. Prevalent Cases, Rates (per 100,000 population), Years Lived with Disability (YLDs), and Cumulative Percent Change with 95% Uncertainty Interval (UI) for the Top 10 Global Causes of YLDs in Children and Adolescents in 195 Countries and Territories, Aged Under 5 Years, Both Sexes, 1990 and 2015 eTable 5. Prevalent Cases and Years Lived with Disability (YLDs), Percent Change, and Percent Change in Age-standardised Rates between 2005 and 2015 of Anemia by Cause, Aged 0 to 19 Years, Both Sexes, 2015 eTable 6. Prevalent Cases and Years Lived with Disability (YLDs), Percent Change, and Percent Change in Age-standardised Rates between 2005 and 2015 of Developmental Intellectual Disability by Cause, Aged 0 to 19 Years, Both Sexes, 2015 eTable 7. Prevalent Cases and Years Lived with Disability (YLDs), Percent Change, and Percent Change in Age-standardised Rates between 2005 and 2015 of Epilepsy by Cause, Aged 0 to 19, Both Sexes, 2015

eTable 6. Prevalent Cases and Years Lived with Disability (YLDs), Percent Change, and Percent Change in Age-standardised Rates between 2005 and 2015 of Developmental Intellectual Disability by Cause, Aged 0 to 19 Years, Both Sexes, 2015 eTable 7. Prevalent Cases and Years Lived with Disability (YLDs), Percent Change, and Percent Change in Age-standardised Rates between 2005 and 2015 of Epilepsy by Cause, Aged 0 to 19, Both Sexes, 2015 eTable 8. Prevalent Cases and Years Lived with Disability (YLDs), Percent Change, and Percent Change in Age-standardised Rates between 2005 and 2015 of Hearing Loss by Cause, Aged 0 to 19, Both Sexes, 2015 eTable 9. Prevalent Cases and Years Lived with Disability (YLDs), Percent Change, and Percent Change in Age-standardised Rates between 2005 and 2015 of Vision Loss by Cause, Aged 0 to 19, Both Sexes, 2015 eTable 10. Number of Maternal Deaths, Maternal Mortality Ratio (MMR, number of deaths per 100,000 live births), and Annualized Rate of Change (ARC), in percent, for 195 Countries and Territories, Aged 10 to 19 Years, Females, 1990 to 2015 Click here for additional data file.

Introduction Because of its influence on a wide variety of life-course outcomes, educational attainment is arguably the most important long-term outcome of early childhood interventions (ECIs). Low educational attainment (ie, high school credential or less) is a major risk factor for all 7 health metrics of the American Heart Association (eg, hypertension, smoking, and obesity),1,2,3 economic disparities, criminal behavior, and mental health problems.4,5,6,7,8 For these reasons, educational attainment is the leading social determinant of health in Healthy People 2020.9 Early childhood interventions are one of the most promising and consequential of all prevention programs,10,11,12,13 but few, if any, studies have examined the entire spectrum of education from high school dropout to postsecondary success, primarily because of the lack of follow-up beyond 25 years of age, when continuing education is prevalent. Findings are also inconsistent across studies,14,15,16,17,18,19 with some showing positive effects on high school graduation and/or college attendance and others showing no such effects, a mixed pattern favoring high school or college attainment but not both, and differences by sex or racial groups. Moreover, to our knowledge, no previous studies have assessed degree completion for large-scale, public programs after the mid-20s.

l graduation and/or college attendance and others showing no such effects, a mixed pattern favoring high school or college attainment but not both, and differences by sex or racial groups. Moreover, to our knowledge, no previous studies have assessed degree completion for large-scale, public programs after the mid-20s. In the Perry Preschool Study,17 for example, program participants had significantly higher rates of high school graduation than controls, but no differences were observed for postsecondary degree attainment. The reverse pattern was found in the Carolina Abecedarian Project.18 Because of small sample sizes, no credible interpretations about subgroup associations were possible. A designed replication of the Abecedarian Project, the Infant Health and Development Program, found no differences in high school dropout.16 Other long-term follow-up studies, including of Head Start,15,19 have reported positive associations with educational attainment up to 25 years of age. Differences by sex vary across studies,17,18,19,20 but high-risk groups, including the most economically disadvantaged, have had greater benefits.13,20,21

ol dropout.16 Other long-term follow-up studies, including of Head Start,15,19 have reported positive associations with educational attainment up to 25 years of age. Differences by sex vary across studies,17,18,19,20 but high-risk groups, including the most economically disadvantaged, have had greater benefits.13,20,21 Given the relatively small samples and limited number of programs analyzed in long-term follow-up studies, differences by child and family characteristics and for different dosages of intervention are underinvestigated.22,23 This limits the implications of findings for health promotion and economic mobility. Because postsecondary attainment for low-income, minority populations is not usually completed until the early 30s,24 midlife follow-up is essential for understanding impacts. Sex and family socioeconomic status moderate life-course outcomes.4,15,24 In our study, we assessed for the first time, to our knowledge, links between participation in the Child-Parent Center (CPC) program, a large-scale and multicomponent school-based ECI, and educational attainment at 35 years of age. We addressed whether (1) CPC program participation beginning at 3 or 4 years of age was linked to greater educational attainment, including postsecondary degree completion; (2) duration of participation up to 6 years was associated with greater attainment; and (3) increases were modified by sex and parental educational level. We hypothesized that earlier and continuing participation would link to attainment and that men and those whose parents had less education would show greater benefits.

) duration of participation up to 6 years was associated with greater attainment; and (3) increases were modified by sex and parental educational level. We hypothesized that earlier and continuing participation would link to attainment and that men and those whose parents had less education would show greater benefits. Methods Study Design The Chicago Longitudinal Study (CLS) is a multisite, prospective investigation of CPC and early experiences on well-being across the life course.14,25,26 The study sample includes 1539 low-income, minority children born in 1979 or 1980 who grew up in high-poverty neighborhoods in Chicago, Illinois. In this matched-group, alternative-intervention design, all 989 children who entered preschool in 1983 or 1984 and completed kindergarten in 1986 from 20 CPCs were included. The same-age comparison group of 550 children was enrolled from 5 randomly selected Chicago public schools with alternative ECIs (full-day kindergarten; 15% with Head Start preschool) or 8 CPC sites without preschool experience. Previous reports25,26,27 have described the sampling, baseline characteristics, and design elements (eTable 1 and eTable 2 in the Supplement). All data collection procedures have been approved by the University of Minnesota Institutional Review Board. All participants provided written or oral informed consent, and all data were deidentified for analysis.

bed the sampling, baseline characteristics, and design elements (eTable 1 and eTable 2 in the Supplement). All data collection procedures have been approved by the University of Minnesota Institutional Review Board. All participants provided written or oral informed consent, and all data were deidentified for analysis. Because of living in high-poverty neighborhoods, children in this cohort were eligible for and participated in government-funded early childhood programs. Like most other studies of established programs, random assignment to the intervention was not possible and would have violated the legal rules that require enrollment of the neediest children on a first-come, first-serve basis.25,27 Data were collected from administrative records, schools, and families from birth to 35 years of age.

studies of established programs, random assignment to the intervention was not possible and would have violated the legal rules that require enrollment of the neediest children on a first-come, first-serve basis.25,27 Data were collected from administrative records, schools, and families from birth to 35 years of age. Intervention Because the CPC program has been described in depth elsewhere,21,25,27 herein, we provide an overview. The CPC program is a school-based preventive intervention for preschool- and school-aged children in high-poverty neighborhoods that are not being otherwise served. The program is designed to promote children's school competence, especially school readiness, achievement, and family involvement in learning. Each center is colocated in a school or on school grounds and serves 100 to 150 children beginning in preschool who then matriculate to school-age services in the same location. Any child who visits a CPC site receives services. Participation age ranges from 1 to 6 years. In addition to a leadership team in each site, all teachers have bachelor’s degrees and are state certified (eMethods in the Supplement).

eginning in preschool who then matriculate to school-age services in the same location. Any child who visits a CPC site receives services. Participation age ranges from 1 to 6 years. In addition to a leadership team in each site, all teachers have bachelor’s degrees and are state certified (eMethods in the Supplement). Major program elements include the following: (1) high-quality educational enrichment through reduced class sizes and a balance of teacher- and child-directed learning; (2) family support services that include participation in school activities, support groups and workshops, and home visits; and (3) comprehensive services that include nutritional and health supports (ie, subsidized meals, health screening, and speech therapy). Service continuity from preschool to third grade (ages 3-9 years) is a hallmark, thus providing a stable and predictable learning environment. The CPC program is being expanded as a multicomponent school reform model.28,29

ude nutritional and health supports (ie, subsidized meals, health screening, and speech therapy). Service continuity from preschool to third grade (ages 3-9 years) is a hallmark, thus providing a stable and predictable learning environment. The CPC program is being expanded as a multicomponent school reform model.28,29 Outcomes The full spectrum of secondary and postsecondary education was assessed. All measures were dichotomous except years of education. Data were obtained from administrative records of various sources and were supplemented by interviews with youths and family members. For higher education, the National Student Clearinghouse was the primary source.30 More than 3600 public and private colleges, representing 98% of all students, report enrollment and graduation data. We used data collected from matched records from those reports and the Illinois Shared Enrollment and Graduation Consortium (now part of the National Student Clearinghouse) from January 1, 2002, to May 31, 2015 (mean age, 35.1 years) (eMethods in the Supplement). The amount and sources of information were similar among the intervention groups.

om matched records from those reports and the Illinois Shared Enrollment and Graduation Consortium (now part of the National Student Clearinghouse) from January 1, 2002, to May 31, 2015 (mean age, 35.1 years) (eMethods in the Supplement). The amount and sources of information were similar among the intervention groups. Secondary education was measured by 5 indicators: dropout by 16 years of age, 4-year graduation (earned diploma within 4 years of ninth grade entry), and 3 others, including high school completion (earned diploma or General Educational Development credential). Postsecondary education was measured by years of education (range, 7-22 years), college attendance (2 or 4 years), degree completion (associate’s and higher), and a credential (eg, occupational certificate) (eMethods in the Supplement).

h school completion (earned diploma or General Educational Development credential). Postsecondary education was measured by years of education (range, 7-22 years), college attendance (2 or 4 years), degree completion (associate’s and higher), and a credential (eg, occupational certificate) (eMethods in the Supplement). Statistical Analysis With use of methods of previous studies,14,21,25 outcomes were analyzed by probit and linear regression in Stata statistical software, version 14 (StataCorp).31 Findings are reported as means or percentages and group differences after adjusting for the influence of the covariates through inverse probability weighting (IPW).32,33 Inverse probability weighting adjusted for each individual’s propensity to enroll in the program (selection model) and to leave the study sample (attrition model) (eMethods in the Supplement). After major assumptions were corroborated, propensity scores for each model were estimated through probit regression, with the resulting values used as inverse weights to estimate associations.34 As a double adjustment, the weights were multiplied together.

e study sample (attrition model) (eMethods in the Supplement). After major assumptions were corroborated, propensity scores for each model were estimated through probit regression, with the resulting values used as inverse weights to estimate associations.34 As a double adjustment, the weights were multiplied together. Inverse probability weighting is a more flexible and comprehensive analytic approach than traditional covariate adjustment or matching and yields estimates with the lowest variances and SEs in large samples while correcting for potential biases.34,35 The covariates of the propensity scores were measured primarily between birth and 5 years of age, including family risk indicators, sex, race/ethnicity, and child welfare history. There were 17 covariates for the selection model and 26 for the attrition model (eTables 3 and 4 in the Supplement). Models were estimated separately for preschool and school-age and for extended intervention groups. Dichotomously coded CPC preschool participation at 3 or 4 years of age and school-aged participation from first to third grade were assessed simultaneously. The CPC extended intervention for 4 to 6 years was assessed against the comparison group with less extensive participation (0-3 years). Duration was also analyzed among 6 groups consistent with a dosage-response relationship.

s of age and school-aged participation from first to third grade were assessed simultaneously. The CPC extended intervention for 4 to 6 years was assessed against the comparison group with less extensive participation (0-3 years). Duration was also analyzed among 6 groups consistent with a dosage-response relationship. A dummy code for missing data was included to determine whether estimates based on multiple imputation (expectation maximization) affected outcomes. We adjusted SEs for school-level clustering (mean intraclass r value, 0.05), although this did not affect estimates. Group differences and marginal effects were emphasized, with 95% CIs. Robustness analyses included each IPW model separately (selection and attrition) and the traditional covariate approach. Sex and mother’s educational level subgroups were estimated in separate models, with interpretations emphasizing differences shown in the overall model.

arginal effects were emphasized, with 95% CIs. Robustness analyses included each IPW model separately (selection and attrition) and the traditional covariate approach. Sex and mother’s educational level subgroups were estimated in separate models, with interpretations emphasizing differences shown in the overall model. Results Cohort Follow-up at 35 Years of Age A total of 1539 participants (mean [SD] age, 35.1 [0.32] years; 1423 [92.9%] black and 108 [7.1%] Hispanic) were included in the study. In May 2015, a total of 1398 participants (90.8%) from the original sample had data on educational attainment. Retention rates were 91.4% for the intervention group and 89.8% for the comparison group. The high rate of retention was attributable to the use of multiple sources of data, including administrative, self-report, and follow-up tracking. Among those with available information, 963 (72.5%) in the total sample resided in Illinois, with most others (213 [16.0%]) remaining in the Midwest (70 participants had missing contact information). Table 1 gives the rates of participation over time and attrition. The comparison group received a variety of early childhood services. There was no evidence of selective attrition by program status.14,25 The attrition sample included those who moved and were not located or deceased (eTable 1 and eMethods in the Supplement).

1 gives the rates of participation over time and attrition. The comparison group received a variety of early childhood services. There was no evidence of selective attrition by program status.14,25 The attrition sample included those who moved and were not located or deceased (eTable 1 and eMethods in the Supplement). Table 1. Participation of Original Child-Parent Center (CPC) Program and Comparison Groups in the Chicago Longitudinal Study Study Category Total Sample Program Groupa Comparison Groupa Program participant characteristics at start of studyb No. in original sample 1539 989 550 No. with preschool participation 1073 989 84 No. with CPC preschool 989 989 0 Years in CPC preschool (range) 0.99 (0-2) 1.55 (1-2) 0 No. with kindergarten participation 1539 989 550 No. in full-day kindergarten 1141 591 550 No. with CPC school-aged participation 850 684 166 Years in school-aged program (range) 1.16 (0-3) 1.43 (1-2) 0.68 (1-2) No. with CPC extended intervention (4-6 y) 553 553 0 Total years in CPC program (range) 2.78 (0-6) 3.95 (1-2) 0.68 (1-2) No. unavailable for follow-up in postprogram years Movedc 75 41 34 Deceased 66 44 22 Follow-up study at 35 y of age, No. with data Educational attainment 1398 904 494 Completed high school (diploma or equivalent) 1157 774 383 Attended college (2- or 4-y institution) 813 555 258 Earned associates’ degree or higher 239 176 63 Earned bachelors’ degree or higher 167 124 43 a Program participation covers the 6-year period (1983-1989) that defines enrollment in the CPC intervention.

4 494 Completed high school (diploma or equivalent) 1157 774 383 Attended college (2- or 4-y institution) 813 555 258 Earned associates’ degree or higher 239 176 63 Earned bachelors’ degree or higher 167 124 43 a Program participation covers the 6-year period (1983-1989) that defines enrollment in the CPC intervention. b The CPC preschool comparison group participated in a full-day kindergarten program, and 84 participated in Head Start preschool. A total of 109 parents in the comparison group reported that their child participated in other child care or education in preschool, and 176 individuals in the comparison group were eligible to receive limited services in the CPC program for kindergarten but enrolled in different classrooms; they are not part of the original CPC intervention group. Some individuals in the comparison group participated in the school-aged program because it was open to any child enrolled in elementary school from first to third grade. Fifteen children in the CPC intervention group enrolled in the alternative full-day kindergarten.

re not part of the original CPC intervention group. Some individuals in the comparison group participated in the school-aged program because it was open to any child enrolled in elementary school from first to third grade. Fifteen children in the CPC intervention group enrolled in the alternative full-day kindergarten. c These categories account for attrition from the original study sample of 1539. Individuals were unavailable for follow-up during the postprogram years because they moved from Chicago and could not be located, were deceased, or did not have sufficient identifying information to track, refused to participate, or were incarcerated (other). At 35 years of age, the total number of deceased individuals in the study was 73. Seven individuals who died after July 1, 2009, were included in the study sample. The attrition sample of 141 had missing data on educational attainment between the ages of 28 and 35 years.

ed to participate, or were incarcerated (other). At 35 years of age, the total number of deceased individuals in the study was 73. Seven individuals who died after July 1, 2009, were included in the study sample. The attrition sample of 141 had missing data on educational attainment between the ages of 28 and 35 years. Group Comparability At follow-up, the program and comparison groups were similar on most characteristics (Table 2). These characteristics were measured from administrative records and family surveys primarily between birth and 5 years of age. The groups were similar in employment, single-parent family status, neighborhood poverty, and adverse experiences (eg, parent substance abuse). They differed in mother’s educational level and child welfare but were equivalent among male participants. The original sample had similar group equivalence (eTable 2 in the Supplement). The original and follow-up samples had similar attributes (eg, family risk, program participation) (eTable 1 in the Supplement). However, the attrition sample had a higher percentage of male participants. Inverse probability weighting accounted for differences in analyses (eTables 3 and 4 in the Supplement). Table 2.

Group Comparability At follow-up, the program and comparison groups were similar on most characteristics (Table 2). These characteristics were measured from administrative records and family surveys primarily between birth and 5 years of age. The groups were similar in employment, single-parent family status, neighborhood poverty, and adverse experiences (eg, parent substance abuse). They differed in mother’s educational level and child welfare but were equivalent among male participants. The original sample had similar group equivalence (eTable 2 in the Supplement). The original and follow-up samples had similar attributes (eg, family risk, program participation) (eTable 1 in the Supplement). However, the attrition sample had a higher percentage of male participants. Inverse probability weighting accounted for differences in analyses (eTables 3 and 4 in the Supplement). Table 2. Preprogram Attributes of Child-Parent Center Program and Comparison Groups for the Follow-up Studya Child or Family Characteristic Follow-up Sample at 35 Years of Age (n = 1398) Male Program vs Control (n = 666) Female Program vs Control (n = 732) Program Group (n = 904) Comparison Group (n = 494) Difference (95% CI) Program Group (n = 414) Comparison Group (n = 252) Difference (95% CI) Program Group (n = 490) Comparison Group (n = 242) Difference (95% CI) Sample recovery 91.4 89.8 1.6 (–1.5 to 4.7) 87.0 88.5 –1.5 (–6.2 to 3.3) 95.7 93.8 1.9 (–1.5 to 5.3) Female 54.2 49.0 5.2 (–0.3 to 10.7) Black 93.3 93.2 0.1 (–2.6 to 2.9) 93.5 91.3 2.2 (–2.0 to 6.4) 93.1 95.1 –2.0 (–5.5 to 1.6) Family risk index by child’s age of 3 yb 4.49 4.50 –0.7 (–0.19 to 0.18) 4.37 4.52 –0.15 (–0.41 to 0.12) 4.59 4.47 0.12 (–0.14 to 0.37) ≥4 Risk factors by child’s age of 3 y 72.8 70.9 1.9 (–3.0 to 6.9) 69.6 70.3 –0.7 (–7.9 to 6.5) 75.5 71.5 4.0 (–2.8 to 10.9) Mother did not complete high school by child’s age of 3 yc 50.1 58.7 –8.6 (–14.0 to –3.2)d 52.7 57.6 –4.9 (–12.7 to 2.9) 48.0 60.0 –12.0 (–19.6 to −4.4)d Mother completed some college by child’s age of 3 y 13.4 10.5 2.9 (–0.6 to 6.4) 13.0 11.5 1.5 (–3.6 to 6.6) 13.7 9.5 4.2 (–0.6 to 9.0) Single parent by child’s age of 3 yc 76.6 75.5 1.0 (–3.7 to 5.7) 74.4 78.2 –3.8 (–10.4 to 2.8) 78.4 72.8 5.6 (–1.1 to 12.3) Mother not employed by child’s age of 3 yc 66.6 62.6 4.0 (–1.2 to 9.3) 61.8 64.7 –2.9 (–10.4 to 4.7) 70.6 60.3 10.3 (2.9 to 17.7)d Ever reported receiving free lunch by child’s age of 3 yc 83.6 82.8 0.8 (–3.3 to 5.0) 50.7 52.2 –1.5 (–7.5 to 4.6) 86.1 83.4 2.7 (–2.9 to 8.2) Ever reported receiving AFDC by child’s age of 3 yc 62.5 60.3 2.2 (–3.2 to 7.5) 58.5 61.6 –3.1 (–10.7 to 4.6) 65.9 59.1 6.8 (–0.7 to 14.3) ≥4 Children at home by child’s age of 3 yc 16.0 19.0 –3.0 (–7.2 to 1.2) 16.7 15.5 1.2 (–4.5 to 6.9) 15.5 22.7 –7.2 (–13.4 to –1.0)d Children in school area in which ≥60% of children reside in low-income familiesc 77.8 73.1 4.7 (–0.1 to 9.5) 75.6 71.8 3.8 (–3.2 to 10.7) 79.6 74.4 5.2 (–1.3 to 11.8) Any child welfare history by child’s age of 3 y 2.9 5.5 –2.6 (–4.9 to –0.3)d 2.4 5.1 –2.7 (–5.9 to 0.4) 3.3 5.8 –2.5 (–5.9 to 0

2.7 –7.2 (–13.4 to –1.0)d Children in school area in which ≥60% of children reside in low-income familiesc 77.8 73.1 4.7 (–0.1 to 9.5) 75.6 71.8 3.8 (–3.2 to 10.7) 79.6 74.4 5.2 (–1.3 to 11.8) Any child welfare history by child’s age of 3 y 2.9 5.5 –2.6 (–4.9 to –0.3)d 2.4 5.1 –2.7 (–5.9 to 0.4) 3.3 5.8 –2.5 (–5.9 to 0 .8) Mother was teenager at child’s birthc 15.9 17.8 –1.9 (–6.0 to 2.3) 16.7 20.3 –3.6 (–9.7 to 2.6) 15.3 15.3 0 (–5.5 to 5.6) Missing any family risk indicators 13.7 15.6 –1.9 (–5.8 to 2.0) 15.0 19.1 –4.1 (–10.0 to 1.9) 12.7 12.0 0.7 (–4.4 to 5.7) Low birth weight (<2500 g) 11.3 14.0 –2.7 (–6.4 to 1.0) 10.6 11.5 –0.9 (–5.8 to 4.1) 11.8 16.5 –4.7 (–10.2 to 0.8) Home environment problem at ages 0-5 y 51.9 55.7 –3.8 (–9.3 to 1.7) 51.7 55.6 –3.9 (–11.7 to 3.9) 52.0 55.7 –3.7 (–11.0 to 3.9) No. of adverse childhood experiences at age 0-5 y 0.34 0.31 0.03 (–0.05 to 0.10) 0.42 0.36 0.06 (–0.05 to 0.18) 0.26 0.26 0 (–0.09 to 0.09) Abbreviation: AFDC, Aid to Families with Dependent Children.

.8) Mother was teenager at child’s birthc 15.9 17.8 –1.9 (–6.0 to 2.3) 16.7 20.3 –3.6 (–9.7 to 2.6) 15.3 15.3 0 (–5.5 to 5.6) Missing any family risk indicators 13.7 15.6 –1.9 (–5.8 to 2.0) 15.0 19.1 –4.1 (–10.0 to 1.9) 12.7 12.0 0.7 (–4.4 to 5.7) Low birth weight (<2500 g) 11.3 14.0 –2.7 (–6.4 to 1.0) 10.6 11.5 –0.9 (–5.8 to 4.1) 11.8 16.5 –4.7 (–10.2 to 0.8) Home environment problem at ages 0-5 y 51.9 55.7 –3.8 (–9.3 to 1.7) 51.7 55.6 –3.9 (–11.7 to 3.9) 52.0 55.7 –3.7 (–11.0 to 3.9) No. of adverse childhood experiences at age 0-5 y 0.34 0.31 0.03 (–0.05 to 0.10) 0.42 0.36 0.06 (–0.05 to 0.18) 0.26 0.26 0 (–0.09 to 0.09) Abbreviation: AFDC, Aid to Families with Dependent Children. a Data are presented as percentage of individuals unless otherwise indicated. The child and family background indicators were measured from administrative records (birth certificates, school records) primarily from birth to 5 years of age. Parent reports from ages 7 to 12 years were used to supplement some risk indicators (eg, number of children). These and other indicators were used in the inverse probability weighting models. Home environment problems (eg, frequent family conflict) and adverse child experiences were from retrospective reports. The preschool to third grade group showed similar patterns and is not shown. b The index ranges from 0 to 7, with higher numbers indicating greater levels of family risk. c Family risk indicators. d The 95% CI does not include zero.

a Data are presented as percentage of individuals unless otherwise indicated. The child and family background indicators were measured from administrative records (birth certificates, school records) primarily from birth to 5 years of age. Parent reports from ages 7 to 12 years were used to supplement some risk indicators (eg, number of children). These and other indicators were used in the inverse probability weighting models. Home environment problems (eg, frequent family conflict) and adverse child experiences were from retrospective reports. The preschool to third grade group showed similar patterns and is not shown. b The index ranges from 0 to 7, with higher numbers indicating greater levels of family risk. c Family risk indicators. d The 95% CI does not include zero. Outcomes for Preschool and School-aged Participation As indicated in Table 3, after controlling for IPW selection and attrition, the preschool group had significantly higher levels of educational attainment on 9 of 12 outcomes. These outcomes included a higher rate of 4-year high school graduation (51.0% vs 44.0%; difference, 7.0%; 95% CI, 1.4%-12.6%), college attendance (61.2% vs 53.1%; difference, 8.1%; 95% CI, 0.8%-15.4%), associate’s degree or higher (15.7% vs 10.7%; difference, 5.0%; 95% CI 1.0%-9.0%), and master’s degree or higher (4.2% vs 1.5%; difference, 2.7%; 95% CI, 1.3%-4.1%) (Figure). The advantage for years of education was a half year (12.81 vs 12.32 years), but the difference for an earned bachelors’ degree was smaller (11.0% vs 7.8%). The only difference for CPC school-age was in on-time high school graduation.

ster’s degree or higher (4.2% vs 1.5%; difference, 2.7%; 95% CI, 1.3%-4.1%) (Figure). The advantage for years of education was a half year (12.81 vs 12.32 years), but the difference for an earned bachelors’ degree was smaller (11.0% vs 7.8%). The only difference for CPC school-age was in on-time high school graduation. Table 3. Adjusted Rates of Educational Attainment by Preschool, School-aged, and Extended-Intervention Group Statusa Educational Outcomes by 35 Years of Age Preschool Groups School-aged Groups Extended Intervention Groupsb Intervention (n = 904) Comparison (n = 494) Difference (95% CI) Intervention (n = 776) Comparison (n = 622) Difference (95% CI) Intervention (n = 514) Comparison (n = 884) Difference (95% CI) Dropout by 16 y of age 11.2 13.9 –2.7 (–6.8 to 1.3) 12.4 11.6 0.8 (–2.2 to 3.8) 11.6 12.6 –1.0 (–5.2 to 3.1) 4-y High school graduation 51.0 44.0 7.0 (1.4 to 12.6)c 49.1 46.3 2.8 (–4.4 to 10.0) 55.3 44.7 10.6 (3.4 to 17.8)c On-time high school graduation 42.1 34.3 7.8 (2.5 to 13.1)c 42.3 35.1 7.2 (0.8 to 13.6)c 48.5 34.7 13.8 (6.0 to 21.6)c High school completion 86.9 80.7 6.2 (0.9 to 11.6)c 85.2 83.5 1.7 (–2.5 to 5.9) 87.3 82.8 4.5 (1.3 to 7.8)c High school graduation 56.0 50.5 5.5 (0.3 to 10.8)c 53.7 52.4 1.3 (–6.3 to 8.9) 59.6 50.4 9.2 (2.3 to 16.0)c Years of education 12.81 12.32 0.49 (0.20 to 0.77)c 12.65 12.55 0.1 (–0.14 to 0.33) 12.95 12.45 0.5 (0.17 to 0.84)c College attendance 61.2 53.1 8.1 (0.8 to 15.4)c 59.4 56.5 2.9 (–3.1 to 9.0) 63.2 55.8 7.4 (1.4 to 13.4)c 4-y College attendance 29.3 21.4 7.9 (1.9 to 14) 25.6 26.0 –0.4 (–7.1 to 6.4) 31.4 24.0 7.4 (0.4 to 14.4) Associates’ degree or higher 15.7 10.7 5.0 (1.0 to 9.0) 14.2 13.4 0.8 (–3.8 to 5.3) 18.5 12.5 6.0 (1.0 to 11.0) Bachelor’s degree or higher 11.0 7.8 3.2 (–0.3 to 6.7) 10.8 8.7 2.1 (–1.8 to 6.0) 14.3 8.2 6.1 (1.3 to 10.9) Master’s degree or higher 4.2 1.5 2.7 (1.3 to 4.1) 3.8 2.3 1.5 (–0.5 to 3.4) 5.9 2.3 3.6 (1.4 to 5.9) Postsecondary credential 18.3 17.2 4.1 (–1.1 to 9.3) 20.4 19.2 1.2 (–4.4 to 6.7) 25.0 18.1 6.9 (0.9 to 12.9) a Data are percentage of individuals unless otherwise indicated. Adjusted with inverse probability weighting for program selection and attrition. Comparisons for other extended intervention groups showed a similar pattern. Child welfare history by 4 years of age was not included in the models of bachelor’s degree and master’s degree or higher because it predicted the outcomes. A total of 57 individuals (4%) reported having a master’s degree or higher.

attrition. Comparisons for other extended intervention groups showed a similar pattern. Child welfare history by 4 years of age was not included in the models of bachelor’s degree and master’s degree or higher because it predicted the outcomes. A total of 57 individuals (4%) reported having a master’s degree or higher. The extended intervention model was estimated separately from the preschool and school-aged model. b Extended intervention was 4 to 6 years; comparison, less than 4 years. c The 95% CI does not include zero. Figure. Adjusted Rates for 3 Measures of Educational Attainment by Child-Parent Center (CPC) Program Participation Values are marginal rates adjusted for inverse probability weighting (IPW) for program selection and attrition. A-C, Findings according to preschool CPC participation. In A-C, CPC program school-aged participation was also included in the model (see eTables 3 and 4 in the Supplement for IPW input models and eTables 5-12 in the Supplement for other outcomes). D-F, Findings according to preschool to third grade (PK-3) CPC program participation. Error bars indicate SD.

PC participation. In A-C, CPC program school-aged participation was also included in the model (see eTables 3 and 4 in the Supplement for IPW input models and eTables 5-12 in the Supplement for other outcomes). D-F, Findings according to preschool to third grade (PK-3) CPC program participation. Error bars indicate SD. Outcomes for Extended Program Participation Compared with fewer years of participation, the CPC 4- to 6-year group (extended intervention) had higher levels of educational attainment on 11 of 12 outcomes. These outcomes included a higher rate of 4-year high school graduation (55.3% vs 44.7%; difference, 10.6%; 95% CI, 3.4%-17.8%) (Table 3 and Figure), 4-year college attendance (31.4% vs 24.0%; difference, 7.4%; 95% CI, 0.4%-14.4%), associate’s degree or higher (18.5% vs 12.5%; difference, 6.0%; 95% CI, 1.0%-11.0%), bachelor’s degree (14.3% vs 8.2%; difference, 6.1%; 95% CI, 1.3%-10.9%), and postsecondary credential (25.0% vs 18.1%; difference, 6.9%; 95% CI, 0.9%-12.9%) (eFigure 1 in the Supplement). The percentage improvements over the comparison group were 38.1% for associate’s degree or higher and 48.0% for postsecondary credentials. Other comparisons were similar (eTable 5 in the Supplement).

.9%), and postsecondary credential (25.0% vs 18.1%; difference, 6.9%; 95% CI, 0.9%-12.9%) (eFigure 1 in the Supplement). The percentage improvements over the comparison group were 38.1% for associate’s degree or higher and 48.0% for postsecondary credentials. Other comparisons were similar (eTable 5 in the Supplement). Duration (Dosage) of Participation to Third Grade Table 4 gives the levels of attainment for specific dosage groups ranging from 0 to 5 to 6 years of participation. Although, as expected, there were increasing attainment rates as years of intervention increased, 3 results were evident (see biserial correlations and marginal effects for years of intervention in Table 4). First, the largest differences between the minimum and maximum duration of services was for on-time high school graduation, 4-year high school graduation, college attendance, and years of education. For on-time graduation, the difference was 22.6 percentage points (53.2% vs 30.6%; marginal effect, years = 3.0%; 95% CI, 1.8%-4.1%).

argest differences between the minimum and maximum duration of services was for on-time high school graduation, 4-year high school graduation, college attendance, and years of education. For on-time graduation, the difference was 22.6 percentage points (53.2% vs 30.6%; marginal effect, years = 3.0%; 95% CI, 1.8%-4.1%). Table 4. Adjusted Means and Marginal Effects of Select Measures of Educational Attainment by Program Dosagea Measure Mean Valueb Difference (95% CI) PK to Third Grade (n = 160) PK to Second Grade (n = 351) PK to First Grade (n = 116) PK/PK and Kindergarten (n = 277) No PK/School-aged (n = 149) No Participation (n = 345) Marginal Effect, Duration of Intervention Biserial Correlation Dropout by 16 y of age 7.9 13.3 9.8 11.6 16.1 13.7 –0.7 (–1.7 to 0.4) –0.08 (–0.14 to –0.03) 4-y High school graduation 60.7c 51.6 42.2 45.0 41.0 43.1 2.1 (0.7 to 3.4)d 0.14 (0.09 to 0.20)d On-time high school graduation 53.2c 44.4 33.5 35.0 36.8 30.6 3.0 (1.8 to 4.1)d 0.18 (0.13 to 0.23)d High school completion 87.8 86.8 87.1 84.9 81.1 79.0 1.6 (0.3 to 2.9)d 0.15 (0.09 to 0.20)d High school graduation 65.9c 55.2 46.0 49.8 46.3 50.8 1.7 (0.3 to 3.1)d 0.11 (0.06 to 0.17)d Years of education 12.89c 12.95 12.58 12.59 12.14 12.32 0.10 (0.05 to 0.16)d 0.14 (0.09 to 0.19)d College attendance 62.5 61.7 57.7 58.1 53.6 50.7 2.0 (0.6 to 3.4)d 0.11 (0.06 to 0.17)d 4-y College attendance 27.5 30.6 22.3 28.3 20.0 22.4 1.4 (0.4 to 2.4)d 0.12 (0.06 to 0.17)d Associate’s degree or higher 15.5 18.1 11.9 13.6 6.8 9.9 0.9 (0.17 to 1.6)d 0.14 (0.09 to 0.19)d Bachelor’s degree or higher 11.0 15.4c 9.5 8.5 4.7 7.4 0.8 (0.3 to 1.3)d 0.15 (0.10 to 0.20)d Master’s degree or higher 3.1 5.8 4.0 0 1.6 2.2 0.4 (0.3 to 0.6)d 0.21 (0.16 to 0.26)d Postsecondary credential 27.3 24.0 15.8 17.7 12.2 17.9 0.9 (–0.1 to 1.8) 0.12 (0.07 to 0.18)d Abbreviation: PK, prekindergarten (preschool).

achelor’s degree or higher 11.0 15.4c 9.5 8.5 4.7 7.4 0.8 (0.3 to 1.3)d 0.15 (0.10 to 0.20)d Master’s degree or higher 3.1 5.8 4.0 0 1.6 2.2 0.4 (0.3 to 0.6)d 0.21 (0.16 to 0.26)d Postsecondary credential 27.3 24.0 15.8 17.7 12.2 17.9 0.9 (–0.1 to 1.8) 0.12 (0.07 to 0.18)d Abbreviation: PK, prekindergarten (preschool). a Data are presented as percentage of individuals unless otherwise indicated. Adjusted with inverse probability weighting (IPW) for program selection and attrition. The groups define the extent of intervention in years. A total of 57 individuals (4%) reported having a master’s degree or higher. When examined by the 6 groups, the no PK group and the any school-aged group predicted the outcome and thus the rates were not reported. Marginal effect assessed years of Child-Parent Center program participation (range, 0-6 years) after IPW adjustments. Unadjusted biserial correlation between years of intervention (defined by the groups in the table) and the outcome is adjusted to account for dichotomous outcomes. b PK to third grade represented individuals aged 5 to 6 years; PK to second grade, 4 to 5 years; PK to first grade, 3 to 4 years; PK/PK and kindergarten, 2 to 3 years; no PK/school age, 1 to 3 years; and no participation, 0 years. The reference group for marginal rates is the PK/PK and kindergarten group. c The 95% CI does not include zero compared with the reference group (PK/PK + kindergarten [2- to 3-year olds]). d The 95% CI for marginal effects and correlations do not include zero. The exceptions were dropout by 16 years of age and postsecondary credential.

b PK to third grade represented individuals aged 5 to 6 years; PK to second grade, 4 to 5 years; PK to first grade, 3 to 4 years; PK/PK and kindergarten, 2 to 3 years; no PK/school age, 1 to 3 years; and no participation, 0 years. The reference group for marginal rates is the PK/PK and kindergarten group. c The 95% CI does not include zero compared with the reference group (PK/PK + kindergarten [2- to 3-year olds]). d The 95% CI for marginal effects and correlations do not include zero. The exceptions were dropout by 16 years of age and postsecondary credential. The second result was that the high-dosage groups, which participated in the entire program from preschool to at least second grade (4-5 years or 5-6 years), had the greatest attainment difference with the preschool and kindergarten group for all 3 high school graduation measures and bachelor’s degree or higher (eg, 15.4% vs 8.5% for preschool to second grade vs preschool only). They also had increased educational attainment compared with the lower dosage groups, including the school-aged intervention group with no preschool (eTable 6 in the Supplement). Third, the attainment levels of the 2 high-dosage groups were similar. Program groups had increases in degree attainment between 28 and 35 years of age that were more than double those of the comparison groups (eFigure 2 in the Supplement).

The second result was that the high-dosage groups, which participated in the entire program from preschool to at least second grade (4-5 years or 5-6 years), had the greatest attainment difference with the preschool and kindergarten group for all 3 high school graduation measures and bachelor’s degree or higher (eg, 15.4% vs 8.5% for preschool to second grade vs preschool only). They also had increased educational attainment compared with the lower dosage groups, including the school-aged intervention group with no preschool (eTable 6 in the Supplement). Third, the attainment levels of the 2 high-dosage groups were similar. Program groups had increases in degree attainment between 28 and 35 years of age that were more than double those of the comparison groups (eFigure 2 in the Supplement). Differences by Subgroups Separate estimates were reported for sex and mothers’ educational level (eTables 7 to 10 in the Supplement). Compared with male participants, female participants had higher rates of educational attainment. Although postsecondary outcomes were increased for female participants, in the extended intervention, the difference was significant for college attendance only (female participants: 73.5% vs 62.6%; difference, 10.9%; 95% CI, 3.3%-18.5%; male participants: 50.8% vs 49.0%; difference, 1.8%; 95% CI, −7.7%-11.2%) (eTable 7 in the Supplement). For mothers’ educational level, differences in years of education, 4-year college attendance, associate’s or higher degree, and postsecondary credential were greater for preschool participants whose mothers dropped out of high school compared with those with mothers who were high school graduates (eTable 9 and eTable 10 in the Supplement). This finding and others for 3 postsecondary outcomes are shown in the Figure. The difference in high school attainment was greatest for male participants, especially as duration of intervention increased (eTable 8 in the Supplement).

ith mothers who were high school graduates (eTable 9 and eTable 10 in the Supplement). This finding and others for 3 postsecondary outcomes are shown in the Figure. The difference in high school attainment was greatest for male participants, especially as duration of intervention increased (eTable 8 in the Supplement). Robustness Analysis We compared the main IPW model with 3 others: (1) covariate-adjusted model with no IPW, (2) IPW for program selection, and (3) IPW for attrition. As reported in eFigure 4 in the Supplement, program estimates were robust across models that balanced the covariates between groups (eTables 11 and 12 in the Supplement). The covariate-adjusted model yielded slightly lower estimates of impact (eFigures 1, 3, and 4 in the Supplement). Discussion Lower educational attainment is a major risk factor for many indicators of health and well-being. To our knowledge, no previous early childhood studies have investigated degree completion for large-scale programs after the mid-20s. Given the low generalizability and small sample sizes of previous studies,17,18,22,23 differences by subgroups and dosage of intervention are underinvestigated. The benefits of continuing services beyond 5 years of age are especially limited.

ve investigated degree completion for large-scale programs after the mid-20s. Given the low generalizability and small sample sizes of previous studies,17,18,22,23 differences by subgroups and dosage of intervention are underinvestigated. The benefits of continuing services beyond 5 years of age are especially limited. As the most comprehensive longitudinal study of an established large-scale program, CPC was associated with educational attainment in midlife. We found positive outcomes for 4-year high school graduation, college attendance, and degree completion (associate’s degree or higher). Given that educational attainment is the leading social determinant of health, findings demonstrate that school-based early childhood programs, such as the CPC program, have significant potential to advance life-course health and well-being. For example, all 7 ideal health metrics of the American Heart Association1,2,3 (eg, hypertension) are associated with educational attainment in a dosage-response fashion.36,37 Higher levels of education are associated with greater economic well-being,5,38,39 reduced depression, and involvement in the justice system.5,6 Increased access to high-quality programs provides an important avenue of improved well-being.

n) are associated with educational attainment in a dosage-response fashion.36,37 Higher levels of education are associated with greater economic well-being,5,38,39 reduced depression, and involvement in the justice system.5,6 Increased access to high-quality programs provides an important avenue of improved well-being. The program had compensatory effects for those at elevated risk: black male participants and children of school dropouts. Although the results varied by duration of intervention, they indicate that prevention can be particularly effective in increasing educational attainment for those with the largest disparities in outcomes. For example, the preschool group from low-education households had a 2-fold increase in bachelor’s degree attainment compared with the comparison group. This result dovetails with previous findings14,21,25 and supports the CPC program in enhancing quality, intensity, and continuity of learning.27,28,29

ies in outcomes. For example, the preschool group from low-education households had a 2-fold increase in bachelor’s degree attainment compared with the comparison group. This result dovetails with previous findings14,21,25 and supports the CPC program in enhancing quality, intensity, and continuity of learning.27,28,29 Our focus on 4-year high school graduation is unique and reflects a key Healthy People metric.9 The association between intervention and the educational attainment continuum in midlife increases the likelihood that the CPC program can influence adult health.40 This finding is supported by increasing literature that early childhood experiences link to healthy behaviors.39,41 A major implication is that investments in the ECI, whether during the preschool or early school-aged years, are investments in public health and can affect long-term outcomes. Although access to publicly funded preschool has increased in recent years, only 42% of children aged 4 years and 15% of those aged 3 years enroll in state prekindergarten, Head Start, or other programs.42 Continuing participation in programs similar to the CPC program is not tracked but has been estimated to be less than 10%.22,23,28 Our findings indicate the importance of increasing the availability of such services.

4 years and 15% of those aged 3 years enroll in state prekindergarten, Head Start, or other programs.42 Continuing participation in programs similar to the CPC program is not tracked but has been estimated to be less than 10%.22,23,28 Our findings indicate the importance of increasing the availability of such services. Another unique feature of the study was examination of preschool to third grade services within a dosage framework. Research on dosage effects have been limited despite their direct policy implications. Studies43,44 have found that an additional year of preschool leads to improvements in educational outcomes. Whether the effects are sustained largely depends on the quality of subsequent experiences.4,23,27 Our findings extend previous work by showing that years of participation from preschool to second or third grade were linked to all high school graduation outcomes. Years were also associated with higher rates of earned degrees, including a bachelor’s degree. Long-term effects on education can be strengthened if services continue through at least second grade.

ng that years of participation from preschool to second or third grade were linked to all high school graduation outcomes. Years were also associated with higher rates of earned degrees, including a bachelor’s degree. Long-term effects on education can be strengthened if services continue through at least second grade. The processes that account for the observed links with educational attainment are complex, but several have been identified. Through educational enrichment, the CPC program and other interventions promote early cognitive scholastic skills, which lead to better school performance and adjustment and increased school commitment, thereby reducing the need for treatment.11,15,17,22 Similar cumulative processes of effects have been found for family support and parenting behaviors and school support and quality.22,23,45 Socioemotional advantages (eg, self-control) also have a key role in explaining links with crime prevention and college persistence.11,45 The linchpin of this mediational process is that initial links are sufficiently strong as a function of effective implementation. How this process carries over to physical health and health behaviors in midlife warrants further investigation, but educational attainment is a key conduit that contributes to a process of cumulative advantage.4,36,37,45

l process is that initial links are sufficiently strong as a function of effective implementation. How this process carries over to physical health and health behaviors in midlife warrants further investigation, but educational attainment is a key conduit that contributes to a process of cumulative advantage.4,36,37,45 A rare feature that further strengthens inferences is the availability of complete records of attainment from multiple sources. Education is a continuous process that extends into the 30s for the CLS cohort and in general. The CPC preschool to third graders and preschoolers had nearly doubled rates of degree attainment from 28 to 35 years of age, whereas the rates for comparison groups increased slightly (eFigure 1 in the Supplement). The program groups also had increased rates of earned degrees by nearly 50% (Table 3 and Table 4). Despite economic barriers that make it more difficult for poor and minority youths to pursue postsecondary education,46 these findings indicate that prevention early in life can help reduce economic disparities in attainment at a time when the returns to college are increasing.5,46

by nearly 50% (Table 3 and Table 4). Despite economic barriers that make it more difficult for poor and minority youths to pursue postsecondary education,46 these findings indicate that prevention early in life can help reduce economic disparities in attainment at a time when the returns to college are increasing.5,46 Nevertheless, these sizable benefits do not eliminate disparities across the education continuum with more advantaged groups, especially for degree completion.47,48 Early childhood programs, which are targeted to families at higher levels of risk, would not be expected to permanently compensate for continuing disadvantages. For example, the rate of arrest among CLS youths is double the national rate,49 and this difference accounts for the higher rates of attainment for female participants in the study and substantially contributes to the lower rates of degree completion compared with the general population.

e for continuing disadvantages. For example, the rate of arrest among CLS youths is double the national rate,49 and this difference accounts for the higher rates of attainment for female participants in the study and substantially contributes to the lower rates of degree completion compared with the general population. Limitations This study has 3 limitations. First, the findings are based on a quasi-experimental design, and thus the strength of inferences may not be as strong as in well-executed randomized experiments. Nevertheless, the groups were largely equivalent and findings were consistent across analytic methods. Second, measures of postsecondary education were collected from administrative records and supplemented with self-reports. It is possible that some participants were missed from the records search even though we obtained a 91% rate of retention. Third, generalizability of findings may be restricted. Although other programs are supportive of the long-term benefits reported here,10,11,13 the population served in our study had relatively high levels of economic disadvantage. In addition, the CPC program has a long history of effectiveness because of its quality and comprehensiveness, and thus results may not apply to programs of modest quality.

supportive of the long-term benefits reported here,10,11,13 the population served in our study had relatively high levels of economic disadvantage. In addition, the CPC program has a long history of effectiveness because of its quality and comprehensiveness, and thus results may not apply to programs of modest quality. Conclusions This study suggests that an established and large-scale ECI contributes positively to educational attainment in midlife, a key social determinant of health and well-being. Replication and extension of findings to other contemporary programs, contexts, and samples should further strengthen confidence in the benefits of multiyear and multicomponent preventive interventions. Supplement. eTable 1. Study Sample Characteristics by Attrition Status eTable 2. Original Study Sample Characteristics by Program Groups and Gender (N = 1539) eTable 3. Propensity Score Predictors of Participation in the CPC Program eTable 4. Propensity Score Predictors of Being in the Study Sample eTable 5. Adjusted Rates and Means for Educational Attainment by Program Groups eTable 6. Adjusted Rates, Means, and Marginal Effects for Educational Attainment by Program Dosage eTable 7. Educational Attainment by Program Group Status for Females and Males eTable 8. Educational Attainment by Gender and Program Dosage eTable 9. Educational Attainment by Mothers’ Education and Program Groups eTable 10. Educational Attainment by Mothers’ Education and Program Dosage eTable 11. Robustness by Analytic Technique for the Program Groups eTable 12. Robustness by Analytic Technique and Program Dosage

eTable 8. Educational Attainment by Gender and Program Dosage eTable 9. Educational Attainment by Mothers’ Education and Program Groups eTable 10. Educational Attainment by Mothers’ Education and Program Dosage eTable 11. Robustness by Analytic Technique for the Program Groups eTable 12. Robustness by Analytic Technique and Program Dosage eMethods. Study Background, Measures, and Analysis eFigure 1. Adjusted Rates for Two Measures of Attainment by Preschool to Third Grade Participation eFigure 2. Rate Differences in Educational Attainment Between Ages 28 and 35 by CPC Groups eFigure 3. Adjusted Rates for Two Measures of Attainment by Preschool Participation eFigure 4. Propensity-Score Weighted Differences by Program Groups Click here for additional data file.

Introduction Approximately 12% to 15% of US children experience developmental delays or disabilities, which range in severity and scope from isolated delays in achieving certain developmental milestones to functional impairments in hearing or vision, as well as diagnosable learning, emotional, and behavioral disorders.1,2 Early identification and intervention are critical to optimize language, cognitive, motor, and socioemotional development as well as educational success,3,4,5 yet only an estimated 10% of children with delays are identified and receive intervention.2 The American Academy of Pediatrics (AAP) first recommended developmental screening in 2001 and later expanded guidelines in 2006 with an algorithm that included standardized developmental screening at 9-, 18-, and 30-month well-child visits or whenever a parent or clinician expresses concern through ongoing surveillance at every preventive visit, as well as autism-specific screening at 18 and 24 months of age.6,7,8 Subsequent randomized clinical trials of standardized developmental screening relative to surveillance alone and interventions to increase screening have bolstered these recommendations by demonstrating improvements in timely diagnosis, treatment referrals, and early intervention.9,10

g at 18 and 24 months of age.6,7,8 Subsequent randomized clinical trials of standardized developmental screening relative to surveillance alone and interventions to increase screening have bolstered these recommendations by demonstrating improvements in timely diagnosis, treatment referrals, and early intervention.9,10 During the past decade, there have been several initiatives to improve developmental screening, including the AAP’s Developmental Surveillance and Screening Policy Implementation Project in 17 pediatric practices,11 Pediatric Improvement Partnerships in 12 states and Washington, DC,12,13 and multiple iterations of the Commonwealth Fund’s Assuring Better Child Development initiative, which supported 26 states, Washington, DC, and Puerto Rico to implement policy and practice changes.14 Federal campaigns have included the Agency for Children and Family’s “Birth to Five, Watch me Thrive” program and the Centers for Disease Control and Prevention’s “Learn the Signs. Act Early” program. Developmental screening was included as a Healthy People 2020 objective,15 as part of the Medicaid and Children’s Health Insurance Program child core set of quality measures with demonstration grants for 10 states,16 and, most recently, as a national performance measure for the Title V Maternal and Child Health Services Block Grant,17 with 42 states electing to work toward improvement during the 5-year planning cycle beginning in 2015.

h Insurance Program child core set of quality measures with demonstration grants for 10 states,16 and, most recently, as a national performance measure for the Title V Maternal and Child Health Services Block Grant,17 with 42 states electing to work toward improvement during the 5-year planning cycle beginning in 2015. Data from the National Survey of Children’s Health (NSCH), the only nationally representative data source for standardized developmental screening, indicate that, in 2007, fewer than 1 in 5 children had received a standardized parent-completed developmental screening from a health care professional in the past year,18,19 which increased to about 1 in 3 children by 2011-2012.20,21 Substantial state-level variation, both within and across survey years, far exceeded differences by child or household characteristics. Using the 2016 NSCH, we sought to provide a current examination of the prevalence and variation of recommended developmental screening and surveillance, which serves to establish a baseline for new initiatives and identify opportunities for improvement.

s, far exceeded differences by child or household characteristics. Using the 2016 NSCH, we sought to provide a current examination of the prevalence and variation of recommended developmental screening and surveillance, which serves to establish a baseline for new initiatives and identify opportunities for improvement. Methods Data Source and Study Population Data for this analysis come from the 2016 NSCH—a nationally representative, parent-completed survey of US children younger than 18 years funded and directed by the Health Resources and Services Administration and conducted by the US Census Bureau. Relative to prior survey iterations in 2003, 2007, and 2011-2012, the 2016 NSCH was significantly redesigned to merge content with the former National Survey of Children With Special Health Care Needs and, owing to declining response rates, to change administration from telephone-based interviews to a mailed, self-administered survey with paper and web-based response options. Therefore, data from the 2016 NSCH are not directly comparable to data from prior years of the NSCH. After a parent or caregiver completed a household-based screening instrument to determine the presence of children by special health care needs status, 1 child per household was selected for the survey, with oversampling for those with special health care needs. Surveys were available in both English and Spanish and were completed between June 2016 and February 2017 (N = 50 212). The survey completion rate among households with a confirmed child was 69.7%, and the overall response rate including households without children was 40.7%. All analyses were weighted to account for the unequal probability of selection and differential nonresponse by various sociodemographic factors, and represent all noninstitutionalized US children residing in housing units. Additional details of the NSCH are available elsewhere.22,23 Participation was voluntary and confidential under Title 13, United States Code, Section 9. Prior to public release, all data products are reviewed for adherence to privacy protection and disclosure avoidance guidelines by the US Census Bureau’s Disclosure Review Board.

details of the NSCH are available elsewhere.22,23 Participation was voluntary and confidential under Title 13, United States Code, Section 9. Prior to public release, all data products are reviewed for adherence to privacy protection and disclosure avoidance guidelines by the US Census Bureau’s Disclosure Review Board. We restricted the study population to children between the ages of 9 and 35 months (N = 5668) for consistency with AAP guidelines for developmental screening7; Healthy People 2020 Maternal, Infant, and Child Health Objective 29.115; the Medicaid and Children’s Health Insurance Program core quality measure24; and other initiatives aimed at children from birth to 3 years of age.

ages of 9 and 35 months (N = 5668) for consistency with AAP guidelines for developmental screening7; Healthy People 2020 Maternal, Infant, and Child Health Objective 29.115; the Medicaid and Children’s Health Insurance Program core quality measure24; and other initiatives aimed at children from birth to 3 years of age. Outcomes Developmental Screening The NSCH captures the receipt of standardized, parent-completed developmental screening, hereafter referred to as developmental screening, through 3 survey items that were iteratively developed and validated.18 Children are considered to have received developmental screening if a parent or caregiver responded affirmatively to whether a doctor or health care professional had them or another caregiver complete a questionnaire about specific concerns or observations they may have had about the child’s development, communication, or social behavior in the past year and whether this questionnaire included 2 additional age-specific content components capturing language development and social behavior. Although not all standardized screening tools involve a parent-completed component, they are more commonly used by pediatricians and are favored for both their efficiency and their ability to engage parents.11,25 A total of 5492 children 9 through 35 months of age (96.9%) had complete parental or caregiver responses to the developmental screening survey items.

ols involve a parent-completed component, they are more commonly used by pediatricians and are favored for both their efficiency and their ability to engage parents.11,25 A total of 5492 children 9 through 35 months of age (96.9%) had complete parental or caregiver responses to the developmental screening survey items. Developmental Surveillance Children were considered to have received developmental monitoring or surveillance by a health care professional if a parent or caregiver responded affirmatively to 1 item assessing whether physicians or other health care professionals asked if they had concerns about their child’s learning, development, or behavior. This measure captures the first of 5 recommended surveillance steps: eliciting parental concerns about their child’s development, documenting and maintaining a developmental history, making accurate observations of the child, identifying risk and protective factors, and maintaining an accurate record of documenting the process and findings.7 Although elicitation may occur through a previsit questionnaire similar to a parent-completed screening tool, this surveillance measure captures complementary verbal communication regarding developmental concerns. A total of 5652 children 9 through 35 months of age (99.7%) had a valid parental or caregiver response to the developmental surveillance survey item.

evisit questionnaire similar to a parent-completed screening tool, this surveillance measure captures complementary verbal communication regarding developmental concerns. A total of 5652 children 9 through 35 months of age (99.7%) had a valid parental or caregiver response to the developmental surveillance survey item. Covariates Following the behavioral model of health care use,26,27,28,29 various child-, family-, and health care–related factors were examined that may influence developmental screening and surveillance through parent and clinician behavior. Predisposing sociodemographic characteristics that may influence the use or receipt of services included the child’s age, sex, race/ethnicity, primary language, family structure, and highest household educational level. Enabling factors that may influence the ability to access developmental screening included household income to poverty ratio, insurance coverage and type, the presence of a medical home, having a preventive visit in the past year, and state of residence. Health status characteristics that may influence a perceived need for screening by parents or health care professionals included the child’s health status and the presence of a special health care need. The NSCH measures the AAP-defined medical home through a multi-item composite of 5 components: having a usual source of care, having a personal physician or nurse, receiving family-centered care, receiving referrals for specialty care if needed, and receiving help coordinating health and health-related care if needed.30 Special health care need status is determined using a validated screener that captures chronic physical, developmental, behavioral, or emotional conditions requiring increased use of health or related services.31,32 Owing to a large percentage of missing data (18.6%) and use in the weighting process, the household income to poverty ratio was multiply imputed by the US Census Bureau using regression imputation methods.22

, developmental, behavioral, or emotional conditions requiring increased use of health or related services.31,32 Owing to a large percentage of missing data (18.6%) and use in the weighting process, the household income to poverty ratio was multiply imputed by the US Census Bureau using regression imputation methods.22 Statistical Analysis Bivariate associations between the covariates and the receipt of screening and surveillance were examined using χ2 tests for significance. Multivariable logistic regression models were used to estimate adjusted associations with screening and surveillance using all covariates in a single model. State-level estimates were achieved with fixed effects and complex variance estimation to account for clustering within states. To improve interpretation and translation, estimated odds were converted to marginal probabilities for presentation of adjusted prevalence estimates and rate ratios.33 Unadjusted and model-adjusted state-level prevalence estimates were compared to assess the contribution of child-, family-, and health care–related characteristics in explaining state variation. State-level estimates were also compared with national estimates using t tests for overlapping groups.34 Given the potential mediating role of preventive services and the possibility that certain special health care needs are identified through developmental screening, sensitivity analyses were conducted without these covariates in multivariable models. Approximately 5% of the study sample had missing data on nonimputed covariates and were excluded from regression analyses. All analyses adjusted the variance estimates to account for the complex survey design and multiple imputation of poverty using SAS-callable SUDAAN, version 11.0.1 (Research Triangle Institute).

ls. Approximately 5% of the study sample had missing data on nonimputed covariates and were excluded from regression analyses. All analyses adjusted the variance estimates to account for the complex survey design and multiple imputation of poverty using SAS-callable SUDAAN, version 11.0.1 (Research Triangle Institute). Results In 2016, an estimated 30.4% (95% CI, 28.0%-33.0%) of children 9 through 35 months of age were reported by a parent or caregiver to have received a parent-completed developmental screening from a health care professional in the past year that included assessment of communication and behavior (Table 1). A higher percentage of children (37.1%; 95% CI, 34.4%-39.8%) had received developmental surveillance in which a health care professional had asked if the parent or caregiver had any concerns about their development. Slightly less than 1 in 5 children (19.2%) had received both screening and surveillance, while more than half (51.6%) had received neither screening nor surveillance (Table 2). Table 1. Developmental Screening and Surveillance by Predisposing, Enabling, and Need Characteristics Characteristic Overall Distribution (N = 5668), Weighted % Developmental Screening (n = 5492) Developmental Surveillance (n = 5652) Weighted % (95% CI) P Valuea Weighted % (95% CI) P Valuea Total 100.0 30.4 (28.0-33.0) 37.1 (34.4-39.8) Weighted population size, No.

ce by Predisposing, Enabling, and Need Characteristics Characteristic Overall Distribution (N = 5668), Weighted % Developmental Screening (n = 5492) Developmental Surveillance (n = 5652) Weighted % (95% CI) P Valuea Weighted % (95% CI) P Valuea Total 100.0 30.4 (28.0-33.0) 37.1 (34.4-39.8) Weighted population size, No. 9.0 million 2.7 million 3.3 million Predisposing Characteristics Age, mo 9-23 55.5 29.4 (26.0-32.9) .33 36.1 (32.4-40.1) .66 24-35 44.5 31.8 (28.4-35.3) 38.2 (34.7-41.9) Sex Male 51.5 32.0 (28.4-35.7) .21 38.7 (35.0-42.6) .22 Female 48.5 28.8 (25.7-32.1) 35.3 (31.7-39.2) Race/ethnicity Non-Hispanic white 53.1 34.4 (31.7-37.1) <.001 40.1 (37.3-43.0) .03 Non-Hispanic black 11.9 24.8 (16.7-35.2) 30.4 (22.7-39.5) Hispanic 22.5 24.3 (18.5-31.3) 32.6 (25.5-40.8) Non-Hispanic other single race 6.2 20.2 (14.1-28.1) 31.2 (23.1-40.7) Non-Hispanic multiple race 6.3 38.8 (30.3-48.2) 45.3 (36.5-54.5) Primary household language English 84.7 33.3 (30.8-35.9) <.001 38.5 (35.9-41.2) .02 Non-English 15.3 14.2 (9.6-20.4) 26.4 (18.4-36.3) Family structure 2 Parents, married 67.8 34.8 (31.9-37.8) <.001 39.9 (37.0-42.9) .008 2 Parents, unmarried 12.6 25.7 (18.9-33.9) 35.3 (27.2-44.4) Single mother or other 19.6 20.9 (16.1-26.6) 28.6 (22.8-35.2) Highest household educational level <High school 7.3 16.0 (7.6-30.6) <.001 20.3 (10.7-35.1) .03 High school 17.4 22.4 (16.5-29.7) 34.1 (26.9-42.1) Some college 22.2 25.8 (21.4-30.7) 37.6 (32.1-43.5) ≥College degree 53.1 37.7 (34.7-40.9) 40.5 (37.6-43.5) Enabling Characteristics Household income-to-poverty ratio, % federal poverty level <100 21.3 22.7 (16.6-30.2) .009 32.1 (25.0-40.1) .24 100-199 20.7 26.1 (20.6-32.4) 40.9 (33.6-48.7) 200-399 28.8 33.9 (29.2-39.0) 35.5 (30.6-40.9) ≥400 29.3 36.1 (32.3-40.0) 39.4 (35.9-43.1) Insurance coverage and type Any public 36.8 23.7 (19.8-28.1) <.001 36.5 (31.5-41.7) .04 Private only 58.2 36.0 (32.9-39.1) 39.0 (36.0-42.2) Uninsured 5.0 16.9 (8.7-30.3) 20.1 (10.1-36.0) Has medical home Yes 53.0 37.1 (33.9-40.5) <.001 41.8 (38.4-45.1) <.001 No 47.0 22.7 (19.3-26.5) 31.7 (27.7-36.0) Preventive visit in past year Yes 91.1 32.3 (29.8-35.0) <.001 39.5 (36.8-42.4) <.001 No 8.9 9.2 (5.7-14.7) 13.1 (7.6-21.7) Need Characteristics Child health status Excellent or very good 93.8 31.4 (28.9-34.1) .006 37.0 (34.3-39.8) .60 Good, fair, or poor 6.2 17.7 (11.6-26.1) 40.2 (28.9-52.7) Special health care needs status Yes 8.0 39.1 (31.4-47.4) .02 53.3 (44.8-61.6) <.001 No 92.0 29.7 (27.2-32.3) 35.6 (32.9-38.5) a Determined

7.6-21.7) Need Characteristics Child health status Excellent or very good 93.8 31.4 (28.9-34.1) .006 37.0 (34.3-39.8) .60 Good, fair, or poor 6.2 17.7 (11.6-26.1) 40.2 (28.9-52.7) Special health care needs status Yes 8.0 39.1 (31.4-47.4) .02 53.3 (44.8-61.6) <.001 No 92.0 29.7 (27.2-32.3) 35.6 (32.9-38.5) a Determined by the χ2 test.

7.6-21.7) Need Characteristics Child health status Excellent or very good 93.8 31.4 (28.9-34.1) .006 37.0 (34.3-39.8) .60 Good, fair, or poor 6.2 17.7 (11.6-26.1) 40.2 (28.9-52.7) Special health care needs status Yes 8.0 39.1 (31.4-47.4) .02 53.3 (44.8-61.6) <.001 No 92.0 29.7 (27.2-32.3) 35.6 (32.9-38.5) a Determined by the χ2 test. Table 2. Developmental Screening by Surveillance Developmental Screening (n = 5481) Developmental Surveillance, % Yes No Yes 19.2 11.3 No 17.9 51.6 Several sociodemographic characteristics were associated with both screening and surveillance (Table 1). For example, non-Hispanic white children were about 10 percentage points more likely than non-Hispanic black or Hispanic children to have received screening (non-Hispanic white, 34.4%; 95% CI, 31.7%-37.1% vs non-Hispanic black 24.8%; 95% CI, 16.7%-35.2% and Hispanic, 24.3%; 95% CI, 18.5%-31.3%) and surveillance (non-Hispanic white, 40.1%; 95% CI, 37.3%-43.0% vs non-Hispanic black, 30.4%; 95% CI, 22.7%-39.5% and Hispanic, 32.6%; 95% CI, 25.5%-40.8%). There was also a strong educational gradient, with children of college-educated parents at least 20 percentage points more likely than those with less than a high school degree to have received screening (college educated, 37.7%; 95% CI, 34.7%-40.9%; less than high school degree, 16.0%; 95% CI, 7.6%-30.6%) and surveillance (college educated, 40.5%; 95% CI, 37.6%-43.5%; less than high school degree, 20.3%; 95% CI, 10.7%-35.1%). Other enabling and health factors associated with both screening and surveillance included having a medical home (screening, 37.1%; 95% CI, 33.9%-40.5% vs 22.7%; 95% CI, 19.3%-26.5%; surveillance, 41.8%; 95% CI, 38.4%-45.1% vs 31.7%; 95% CI, 27.7%-36.0%), a past-year preventive visit (screening, 32.3%; 95% CI, 29.8%-35.0% vs 9.2%; 95% CI, 5.7%-14.7%; surveillance, 39.5%; 95% CI, 36.8%-42.4% vs 13.1%; 95% CI, 7.6%-21.7%), and a special health care need (screening, 39.1%; 95% CI, 31.4%-47.4% vs 29.7%; 95% CI, 27.2%-32.3%; surveillance, 53.3%; 95% CI, 44.8%-61.6% vs 35.6%; 95% CI, 32.9%-38.5%). However, several other factors were associated with screening only, such as income (≥400% poverty, 36.1%; 95% CI, 32.3%-40.0% vs <100% poverty, 22.7%; 95% CI, 16.6%-30.2%) and insurance type (private only, 36.0%; 95% CI, 32.9%-39.1% vs any public, 23.7%; 95% CI, 19.8%-28.1%).

44.8%-61.6% vs 35.6%; 95% CI, 32.9%-38.5%). However, several other factors were associated with screening only, such as income (≥400% poverty, 36.1%; 95% CI, 32.3%-40.0% vs <100% poverty, 22.7%; 95% CI, 16.6%-30.2%) and insurance type (private only, 36.0%; 95% CI, 32.9%-39.1% vs any public, 23.7%; 95% CI, 19.8%-28.1%). After adjustment with logistic regression models, most factors except for race/ethnicity and insurance remained significantly associated with screening or surveillance (Table 3). Similar to bivariate associations, there were more factors associated with standardized screening than with more general surveillance. Specifically, primary household language, highest household educational level, and child health status were only associated with screening. Compared with children living in an English primary language household, those in non-English primary language households were 40% less likely to have received screening in the past year (adjusted rate ratio, 0.60; 95% CI, 0.39-0.92). Having a medical home was associated with a 34% increase in screening (adjusted rate ratio, 1.34; 95% CI, 1.13-1.57) and a 24% increase in surveillance (adjusted rate ratio, 1.24; 95% CI, 1.08-1.42), corresponding to an 8 to 9 absolute percentage point increase. In additional models, medical home components—usual source of care, personal physician or nurse, and family-centered care—were associated with both screening and surveillance (eTable 1 in the Supplement). The small proportion of children without a past-year preventive visit were approximately 60% less likely to have received screening and surveillance, while those with special health care needs were more than 50% more likely to have received screening and surveillance. Additional models without a past-year preventive visit and special health care need status did not substantively alter associations (eTable 2 in the Supplement).

ve received screening and surveillance, while those with special health care needs were more than 50% more likely to have received screening and surveillance. Additional models without a past-year preventive visit and special health care need status did not substantively alter associations (eTable 2 in the Supplement). Table 3.

ve received screening and surveillance, while those with special health care needs were more than 50% more likely to have received screening and surveillance. Additional models without a past-year preventive visit and special health care need status did not substantively alter associations (eTable 2 in the Supplement). Table 3. Adjusted Associations With Developmental Screening and Surveillance Characteristic Developmental Screening (n = 5229) Developmental Surveillance (n = 5373) Adjusted Prevalence Rate Ratio (95% CI) Adjusted Prevalence Rate Ratio (95% CI) Predisposing Characteristics Age, mo 9-23 30.2 1 [Reference] 36.4 1 [Reference] 24-35 32.7 1.08 (0.94-1.25) 37.7 1.03 (0.91-1.17) Sex Male 32.7 1 [Reference] 38.3 1 [Reference] Female 29.9 0.92 (0.80-1.05) 35.7 0.93 (0.82-1.06) Race/ethnicity Non-Hispanic white 31.0 1 [Reference] 36.7 1 [Reference] Non-Hispanic black 31.3 1.01 (0.72-1.42) 33.5 0.91 (0.69-1.21) Hispanic 32.7 1.06 (0.84-1.34) 37.9 1.03 (0.83-1.28) Non-Hispanic other single race 25.4 0.82 (0.58-1.16) 37.4 1.02 (0.76-1.37) Non-Hispanic multiple race 36.0 1.16 (0.91-1.49) 43.2 1.18 (0.96-1.45) Primary household language English 32.9 1 [Reference] 38.2 1 [Reference] Non-English 19.8 0.60 (0.39-0.92) 28.7 0.75 (0.52-1.08) Family structure 2 Parents, married 33.4 1 [Reference] 38.9 1 [Reference] 2 Parents, unmarried 30.2 0.90 (0.68-1.21) 36.1 0.93 (0.72-1.19) Single mother or other 24.0 0.72 (0.54-0.96) 30.4 0.78 (0.62-0.99) Highest household educational level <High school 29.5 0.85 (0.47-1.56) 28.7 0.73 (0.41-1.29) High school 26.5 0.77 (0.58-1.02) 32.5 0.83 (0.65-1.05) Some college 26.6 0.77 (0.62-0.95) 36.4 0.92 (0.78-1.10) ≥College degree 34.5 1 [Reference] 39.3 1 [Reference] Enabling Characteristics Household income-to-poverty ratio, % federal poverty level <100 34.4 1.20 (0.82-1.76) 37.9 1.09 (0.84-1.41) 100-199 31.2 1.09 (0.82-1.46) 43.6 1.26 (1.02-1.55) 200-399 33.0 1.15 (0.95-1.39) 34.5 0.99 (0.84-1.17) ≥400 28.7 1 [Reference] 34.7 1 [Reference] Insurance coverage and type Any public 30.1 0.93 (0.74-1.18) 39.6 1.11 (0.90-1.36) Private only 32.2 1 [Reference] 35.8 1 [Reference] Uninsured 26.6 0.83 (0.49-1.39) 33.9 0.95 (0.55-1.63) Has medical home Yes 35.1 1.34 (1.13-1.57) 40.4 1.24 (1.08-1.42) No 26.3 1 [Reference] 32.6 1 [Reference] Preventive visit in past year Yes 32.6 1 [Reference] 38.6 1 [Reference] No 12.3 0.38 (0.23-0.62) 16.8 0.43 (0.27-0.70) Need Characteristics Child health status Excellent or very good 32.1 1 [Reference] 36.8 1 [Reference] Good, fair,

s 35.1 1.34 (1.13-1.57) 40.4 1.24 (1.08-1.42) No 26.3 1 [Reference] 32.6 1 [Reference] Preventive visit in past year Yes 32.6 1 [Reference] 38.6 1 [Reference] No 12.3 0.38 (0.23-0.62) 16.8 0.43 (0.27-0.70) Need Characteristics Child health status Excellent or very good 32.1 1 [Reference] 36.8 1 [Reference] Good, fair, or poor 18.8 0.58 (0.41-0.84) 39.3 1.07 (0.83-1.37) Special health care needs status Yes 46.7 1.55 (1.29-1.85) 54.5 1.54 (1.31-1.80) No 30.2 1 [Reference] 35.5 1 [Reference] Both screening and surveillance varied substantially across states by more than 40 percentage points (Figure 1 and Figure 2).

s 35.1 1.34 (1.13-1.57) 40.4 1.24 (1.08-1.42) No 26.3 1 [Reference] 32.6 1 [Reference] Preventive visit in past year Yes 32.6 1 [Reference] 38.6 1 [Reference] No 12.3 0.38 (0.23-0.62) 16.8 0.43 (0.27-0.70) Need Characteristics Child health status Excellent or very good 32.1 1 [Reference] 36.8 1 [Reference] Good, fair, or poor 18.8 0.58 (0.41-0.84) 39.3 1.07 (0.83-1.37) Special health care needs status Yes 46.7 1.55 (1.29-1.85) 54.5 1.54 (1.31-1.80) No 30.2 1 [Reference] 35.5 1 [Reference] Both screening and surveillance varied substantially across states by more than 40 percentage points (Figure 1 and Figure 2). The prevalence of screening ranged from 17.2% in Mississippi to 58.8% in Oregon, corresponding to a rate ratio of 3.4. Similarly, developmental surveillance ranged from 19.1% in Mississippi to 60.8% in Oregon, corresponding to a rate ratio of 3.2. States with significantly lower rates of developmental screening than the nation overall included Kentucky (17.5%), New York (17.5%), and Florida (20.4%), while states with rates significantly exceeding the national rate included Oregon (58.8%), Colorado (50.2%), Minnesota (50.1%), North Carolina (47.6%), Alaska (46.8%), Montana (46.3%), Massachusetts (46.3%), and Maryland (43.0%). Mississippi was the only state with a significantly lower surveillance rate (19.1%) than the nation overall, while Oregon (60.8%), Iowa (59.4%), Massachusetts (54.9%), Alaska (53.9%), and Idaho (49.8%) had significantly higher rates. The correlation coefficient between state-level screening and surveillance rates indicated only a moderate association (r = 0.55), with 4 states having a 20-percentage point difference between surveillance and screening. Adjustment for sociodemographic, health care, and health status characteristics explained only 4% and 13% of the state-level variance in screening and surveillance, respectively, with a mean absolute state change of 1 percentage point (eTables 3 and 4 in the Supplement).

t difference between surveillance and screening. Adjustment for sociodemographic, health care, and health status characteristics explained only 4% and 13% of the state-level variance in screening and surveillance, respectively, with a mean absolute state change of 1 percentage point (eTables 3 and 4 in the Supplement). Figure 1. Developmental Screening Rates by State—National Survey of Children’s Health, 2016 This map illustrates state rates of developmental screening in a continuous blue color, ranging from 17.2% in Mississippi (lightest blue) to 58.8% in Oregon (darkest blue). Figure 2. Developmental Surveillance Rates by State—National Survey of Children’s Health, 2016 This map illustrates state rates of developmental surveillance in a continuous blue color, ranging from 19.1% in Mississippi (lightest blue) to 60.8% in Oregon (darkest blue).

Figure 1. Developmental Screening Rates by State—National Survey of Children’s Health, 2016 This map illustrates state rates of developmental screening in a continuous blue color, ranging from 17.2% in Mississippi (lightest blue) to 58.8% in Oregon (darkest blue). Figure 2. Developmental Surveillance Rates by State—National Survey of Children’s Health, 2016 This map illustrates state rates of developmental surveillance in a continuous blue color, ranging from 19.1% in Mississippi (lightest blue) to 60.8% in Oregon (darkest blue). Discussion The results of this analysis indicate that less than one-third of all children 9 through 35 months of age have received a standardized parent-completed developmental screening from a health care professional in the last year. At 37.1%, developmental surveillance or elicitation regarding concerns during a health care visit is not substantially higher, and only 1 in 5 children received both screening and surveillance. Although ongoing surveillance is recommended to initiate formal screening for those at risk for delay, universal screening is shown to be more effective than surveillance alone at identifying delays and receiving referrals and intervention eligibility on a timelier basis.10 Thus, the low rates of screening are concerning and were no higher than 39.1% by various child, family, and health care characteristics.

for delay, universal screening is shown to be more effective than surveillance alone at identifying delays and receiving referrals and intervention eligibility on a timelier basis.10 Thus, the low rates of screening are concerning and were no higher than 39.1% by various child, family, and health care characteristics. In bivariate analyses, children of minority race/ethnicity, single mothers, and less-educated or lower-income parents, who were either uninsured or publicly insured, tended to have lower levels of screening and/or surveillance than their corresponding counterparts. This finding represents a notable shift from 2007, when these less-advantaged groups were more likely to be screened,18,19,35 perhaps as a result of a perceived risk of delay and/or efforts that focused on poor children, such as the Early and Periodic Screening Diagnostic and Treatment and Assuring Better Child Development programs.14 As rates of screening and general awareness of its importance have increased, familiar patterns of sociodemographic advantage in accessing quality care may have emerged.

rts that focused on poor children, such as the Early and Periodic Screening Diagnostic and Treatment and Assuring Better Child Development programs.14 As rates of screening and general awareness of its importance have increased, familiar patterns of sociodemographic advantage in accessing quality care may have emerged. Even after adjustment for other characteristics, children living in non–English-speaking primary language households were considerably less likely to receive screening, which demonstrates the continued relevance of previously identified language barriers,18,19,36 despite the availability of parent-completed screening tools in at least 14 languages.7 This disparity may also reflect issues of cultural competence and health care quality, given that only the rates of screening and not surveillance were significantly lower for children living in non–English-speaking primary language households. That fewer factors were significantly associated with surveillance in general may reflect the greater ease with which simple monitoring can occur owing to frequently cited clinician screening barriers of time and staffing limitations and inadequate reimbursement.36,37

n non–English-speaking primary language households. That fewer factors were significantly associated with surveillance in general may reflect the greater ease with which simple monitoring can occur owing to frequently cited clinician screening barriers of time and staffing limitations and inadequate reimbursement.36,37 Early identification of developmental disorders is an explicit function of the medical home,38 which is among the more modifiable enabling characteristics to increase rates of screening and surveillance. Previous studies have shown an unadjusted association between the medical home and developmental screening19 and an adjusted association with developmental surveillance.29 To our knowledge, this is the first study to have examined and reported an adjusted association between the medical home and rates of developmental screening, with 3 of 5 components driving this association: having a usual source of care, having a personal physician or nurse, and receiving family-centered care. Thus, efforts to promote continuous and comprehensive primary care within a medical home may result in improved quality and use of preventive services,39 including developmental screening and surveillance. The potential for significant improvement exists, with nearly half of children 9 to 35 months of age lacking a medical home, as estimated in this study.

d comprehensive primary care within a medical home may result in improved quality and use of preventive services,39 including developmental screening and surveillance. The potential for significant improvement exists, with nearly half of children 9 to 35 months of age lacking a medical home, as estimated in this study. Financial barriers to preventive services, including developmental screening, have been nearly eliminated through expansions of coverage and benefit requirements for private plans to cover all Bright Futures recommendations. Early and Periodic Screening Diagnostic and Treatment requires the same for state Medicaid programs, although details of coverage, codes, and rates of reimbursement may need to be updated in guidelines.40 More than 90% of young children were reported to have had a preventive visit in the past year, consistent with data from other surveys.41,42 However, other data suggest that there are still gaps in children receiving the recommended number of well-child visits,42,43 which could be addressed through reminder-recall and home visiting programs.44,45 Nonetheless, clinician-focused efforts including training, automated prompts in electronic medical records, and learning collaboratives may be necessary and are shown to be effective in promoting adherence to screening recommendations.9,46

hich could be addressed through reminder-recall and home visiting programs.44,45 Nonetheless, clinician-focused efforts including training, automated prompts in electronic medical records, and learning collaboratives may be necessary and are shown to be effective in promoting adherence to screening recommendations.9,46 Similar to previous analyses,18,20 state variation far exceeded that observed by child and family characteristics, with a range of more than 40 percentage points on an absolute scale and a tripling on a relative scale. Moreover, most of this variation could not be explained by available sociodemographic, enabling, or health characteristics, which suggests a role for unmeasured policies, practices, and quality improvement efforts. That many top performers, including Alaska, Colorado, Maryland, Massachusetts, Montana, and Oregon, have likely more than doubled their rates of screening in the past decade demonstrates that improvement is possible across the country. In particular, Oregon had one of the lowest rates of developmental screening in 200718 and now has the highest rate in the nation—nearly twice the national rate, at 58.5%. This success may be attributable to tracking and incentivizing quality improvement through pay-for-performance metrics in coordinated care organizations, established as part of a Medicaid demonstration waiver. The medical home and developmental screening are among 5 emphasized incentive metrics, the latter of which tripled from 20.9% in 2011 to 62.2% in 2016.47

g and incentivizing quality improvement through pay-for-performance metrics in coordinated care organizations, established as part of a Medicaid demonstration waiver. The medical home and developmental screening are among 5 emphasized incentive metrics, the latter of which tripled from 20.9% in 2011 to 62.2% in 2016.47 Limitations The major limitation of this analysis involves potential underestimation of developmental screening by capturing only screenings with a parent-completed component. However, physicians favor parent-completed screening tools for their efficiency and ability to engage parents in the process of developmental promotion.11,25 Our estimate of developmental screening is roughly comparable to the median of the 2016 Medicaid claims-based reporting from 26 states (36%) that includes tools without a parent-completed component.48 It is lower than the latest available AAP member survey report of always or almost always using any formal screening tool (47.7% in 2009)25; however, pediatricians may overreport screening and not all children receive primary care from AAP member pediatricians. Conversely, surveillance may be overestimated given that only the first of 5 recommended surveillance steps is captured. Regardless of counterbalancing assessment issues, these data represent the only current national estimates and indicate that only half of children younger than 3 years have received either screening or surveillance.

eillance may be overestimated given that only the first of 5 recommended surveillance steps is captured. Regardless of counterbalancing assessment issues, these data represent the only current national estimates and indicate that only half of children younger than 3 years have received either screening or surveillance. Conclusions Despite more than a decade of initiatives, the rates of developmental screening and surveillance remain low. However, state-level variation indicates continued potential for improvement. Systems-level quality improvement efforts that integrate the medical home and build on lessons learned from state-based initiatives will be necessary to achieve universal screening and surveillance that optimizes early identification, intervention, and developmental trajectories for children with delays. Supplement. eTable 1. Adjusted Associations Between Medical Home Components and Developmental Screening and Surveillance eTable 2. Adjusted Associations With Developmental Screening and Surveillance (Without Preventive Medical Visit and Special Health Care Needs Status) eTable 3. Unadjusted and Adjusted Developmental Screening Rates by State eTable 4. Unadjusted and Adjusted Developmental Surveillance Rates by State Click here for additional data file.

Introduction Kawasaki disease (KD) is an acute inflammatory disorder predominantly seen in young children. Since it was first described in Japan,1 KD has emerged as the most common cause of acquired heart disease, with an incidence in children younger than 5 years ranging from 265 cases per 100 000 in Japan,2 to 51 to 194 cases per 100 000 in other Asian countries,3,4,5 to 8 to 20 cases per 100 000 in Europe6 and the United States,7 respectively. What makes KD of such concern is its association with vasculitis, affecting predominantly the coronary arteries, which results in coronary artery aneurysms (CAAs) in up to 25% of untreated children.8 Death from myocardial infarction may occur due to thrombotic occlusion of the aneurysms or from the later development of stenotic lesions due to vascular remodeling in the damaged artery. Long-term outcome studies9,10 of children with giant CAAs indicate a worrisome prognosis, with more than 50% needing revascularization or experiencing myocardial infarction within a 30-year period.

usion of the aneurysms or from the later development of stenotic lesions due to vascular remodeling in the damaged artery. Long-term outcome studies9,10 of children with giant CAAs indicate a worrisome prognosis, with more than 50% needing revascularization or experiencing myocardial infarction within a 30-year period. Treatment with intravenous immunoglobulin (IVIG) and, for those who do not respond, additional IVIG11 or other anti-inflammatory agents, such as corticosteroids and infliximab, is effective in abrogating the inflammatory process and reduces the risk of CAAs to 5% to 10%.12 Because KD is difficult to distinguish from other common febrile conditions, many children with KD are not diagnosed and treated early enough to prevent development of CAAs.13 Furthermore, patients who do not fulfill the clinical criteria for diagnosing KD (so-called incomplete KD) may experience CAAs. Delayed diagnosis is a consistent risk factor for development of CAAs, and treatment is often commenced only when coronary dilatation is already demonstrated on echocardiography.

nt of CAAs.13 Furthermore, patients who do not fulfill the clinical criteria for diagnosing KD (so-called incomplete KD) may experience CAAs. Delayed diagnosis is a consistent risk factor for development of CAAs, and treatment is often commenced only when coronary dilatation is already demonstrated on echocardiography. The symptoms of KD are similar to those of several other childhood febrile illnesses, including staphylococcal and streptococcal toxic shock syndromes, measles and other viral illnesses (eg, adenovirus infection, Rocky Mountain spotted fever), and childhood inflammatory diseases, leading to diagnostic difficulty and thus delay in diagnosis and treatment. Guidelines have been developed to facilitate diagnosis based on clinical signs and symptoms, echocardiography, and laboratory variables,14 but there remains an urgent need for an accurate test to distinguish KD from other conditions causing prolonged fever in children.

ficulty and thus delay in diagnosis and treatment. Guidelines have been developed to facilitate diagnosis based on clinical signs and symptoms, echocardiography, and laboratory variables,14 but there remains an urgent need for an accurate test to distinguish KD from other conditions causing prolonged fever in children. In the era of precision medicine, diagnosis of many conditions previously based on clinical features alone is being replaced by diagnosis based on molecular pathology. Host blood gene expression signatures have been shown to identify several specific infectious and inflammatory diseases, including tuberculosis,15 bacterial and viral infections,16,17 and systemic lupus erythematosus.18 Support for a diagnostic approach to KD based on gene expression signatures comes from identification of microRNA biomarkers in KD,19,20 although existing studies are limited by the range of comparator patients or a need to extract RNA from exosomes. We explored use of whole-blood gene expression patterns to distinguish KD from other childhood infectious and inflammatory conditions. We present a gene expression signature, discovered and validated in independent patient groups, that distinguishes KD from a range of bacterial, viral, and inflammatory illnesses.

omes. We explored use of whole-blood gene expression patterns to distinguish KD from other childhood infectious and inflammatory conditions. We present a gene expression signature, discovered and validated in independent patient groups, that distinguishes KD from a range of bacterial, viral, and inflammatory illnesses. Methods Ethical Approval and Informed Consent Patients were recruited, with written parental informed consent, under approvals by the research ethics committees of the United Kingdom (St Mary’s Hospital 09/H0712/58, 13/LO/0026); Spain (Ethical Committee of Clinical Investigation of Galicia [CEIC] 2010/015); Amsterdam, the Netherlands (NL41023.018.12 and NL34230.018.10); and the University of California San Diego (Human Research Protection Program 140220).

rch ethics committees of the United Kingdom (St Mary’s Hospital 09/H0712/58, 13/LO/0026); Spain (Ethical Committee of Clinical Investigation of Galicia [CEIC] 2010/015); Amsterdam, the Netherlands (NL41023.018.12 and NL34230.018.10); and the University of California San Diego (Human Research Protection Program 140220). Patient Study Groups The differential diagnosis for KD includes multiple infectious and inflammatory conditions. Therefore, in this case-control study, we established a discovery group of children with KD and a range of other infectious and inflammatory diseases with clinical signs, inflammatory markers, and duration of fever overlapping KD. Patients were prospectively recruited at pediatric centers in the United Kingdom, Spain, the Netherlands, and the United States if they had febrile illness and required blood testing for clinical investigation as part of the UK-based Immunopathology of Respiratory, Inflammatory and Infectious Disease Study (IRIS)17; the Spanish GENDRES (Genetic, Vitamin D, and Respiratory Infections Research Network) study (http://www.gendres.org); the Dutch Kawasaki Study; or the US-based Kawasaki Disease Research Center Program (https://medschool.ucsd.edu/som/pediatrics/research/centers/kawasaki-disease). The training and test discovery group comprised 404 children with infectious and inflammatory conditions (78 KD, 84 other inflammatory diseases, and 242 bacterial or viral infections) and 55 healthy controls. The independent validation group comprised 102 patients with KD, including 72 in the first 7 days of illness, and 130 febrile controls. The study dates were March 1, 2009, to November 14, 2013, and data analysis took place from January 1, 2015, to December 31, 2017.

cterial or viral infections) and 55 healthy controls. The independent validation group comprised 102 patients with KD, including 72 in the first 7 days of illness, and 130 febrile controls. The study dates were March 1, 2009, to November 14, 2013, and data analysis took place from January 1, 2015, to December 31, 2017. Children with KD represented a combination of those seen directly in emergency departments and patients referred from regional centers. Our study included only patients recruited before initiation of IVIG for treatment. For discovery of a diagnostic signature, we included patients with KD in the first 7 days of illness because we aimed to develop a test for use early in the illness before coronary artery damage occurs. However, to explore the performance of the signature at all stages of illness, we included patients up to day 10 of their illness in the validation study.

re, we included patients with KD in the first 7 days of illness because we aimed to develop a test for use early in the illness before coronary artery damage occurs. However, to explore the performance of the signature at all stages of illness, we included patients up to day 10 of their illness in the validation study. Febrile controls whose duration of illness before hospital presentation varied were recruited, with blood samples collected as soon as possible after presentation and before clinical diagnosis was confirmed. Febrile controls were assigned to diagnostic groups using predefined criteria once the results of all investigations were available (Figure 1 and eMethods in the Supplement). Children with comorbidities likely to influence gene expression, such as immunosuppressive treatments, were excluded. We included comparator groups of children seen with inflammatory illness, including juvenile idiopathic arthritis and Henoch-Schönlein purpura. Comparison of the duration of illness, inflammatory markers. and demographics between patients with KD and febrile controls is summarized in Table 1. Figure 1. Assignment of Patients to Diagnostic Groups The diagnostic algorithm demonstrates the method of assigning patients to diagnostic groups. AHA indicates American Heart Association; CAA, coronary artery aneurysm; CRP, C-reactive protein; HSP, Henoch-Schönlein purpura; JIA, juvenile idiopathic arthritis; and KD, Kawasaki disease. To convert C-reactive protein level to nanomoles per liter, multiply by 9.524; to convert neutrophil count to ×109/L, multiply by 0.001.

icates American Heart Association; CAA, coronary artery aneurysm; CRP, C-reactive protein; HSP, Henoch-Schönlein purpura; JIA, juvenile idiopathic arthritis; and KD, Kawasaki disease. To convert C-reactive protein level to nanomoles per liter, multiply by 9.524; to convert neutrophil count to ×109/L, multiply by 0.001. Table 1. Clinical Characteristics and Initial Laboratory Values for Patients With Kawasaki Disease and Febrile Controls in Discovery and Validation Study Groupsa Variable Discovery Set Validation Set Kawasaki Disease Febrile Controlsb Kawasaki Diseasec Febrile Controlsb No. of patients 78 326 72 130 Age, median (IQR), mo 27 (16 to 45) 37 (9 to 116) 34 (17 to 51) 17 (5 to 47) Male sex, No. (%) 43 (55.1) 184 (56.4) 45 (62.5) 74 (56.9) Illness day at sample collection, median (IQR)d 5 (4 to 6) 6 (4 to 9) 5 (5 to 6) 5 (3 to 7) Laboratory values, median (IQR) Hemoglobin z scoree −1.3 (−2.0 to −0.3) NA −1.2 (−2.0 to −0.4) NA C-reactive protein, mg/L 119 (48 to 192) 66 (23 to 174) 87 (59 to 173) 62 (16 to 162) Platelet count, ×103/μL 352 (303 to 448) 254 (167 to 351) 408 (324 to 474) 277 (176 to 352) White blood cell count, /μL 14 200 (10 400 to 18 300) 8000 (6000 to 12 900) 13 900 (11 000 to 19 000) 11 000 (7700 to 16 000) Neutrophil count, /μL 9000 (6600 to 12400) 5000 (3100 to 9400) 10000 (7300 to 12600) 7000 (3600 to 13400) Ethnicity, No. (%) No. not statedf 0 23 (7.1) 0 10 (8.3) African 3 (3.8) 28 (8.6) 2 (2.8) 23 (19.2) Asian, including Indian subcontinent and Far East 12 (15.4) 29 (8.9) 12 (16.7) 12 (10.0) European 20 (25.6) 186 (57.1) 20 (27.8) 68 (56.7) Hispanic 25 (32.1) 20 (6.1) 14 (19.4) 0 Mixed 15 (19.2) 28 (8.6) 23 (31.9) 7 (5.8) Other 3 (3.8) 12 (3.7) 1 (1.4) 10 (8.3) Coronary artery status, No. (%) Normal 45 (57.7) NA 52 (72.2) NA Dilated 25 (32.1) NA 15 (20.8) NA Aneurysm 8 (10.3) NA 5 (6.9) NA IVIG resistant, No. (%) 18 (23.1) NA 15 (20.8) NA Abbreviations: IQR, interquartile range; IVIG, intravenous immunoglobulin; NA, not applicable.

5.8) Other 3 (3.8) 12 (3.7) 1 (1.4) 10 (8.3) Coronary artery status, No. (%) Normal 45 (57.7) NA 52 (72.2) NA Dilated 25 (32.1) NA 15 (20.8) NA Aneurysm 8 (10.3) NA 5 (6.9) NA IVIG resistant, No. (%) 18 (23.1) NA 15 (20.8) NA Abbreviations: IQR, interquartile range; IVIG, intravenous immunoglobulin; NA, not applicable. SI conversation factors: To convert C-reactive protein level to nanomoles per liter, multiply by 9.524; neutrophil count to ×109/L, multiply by 0.001; platelet count to ×109/L, multiply by 1.0; and white blood cell count to ×109/L, multiply by 0.001. a There were no significant differences between patients with Kawasaki disease in the discovery and validation sets. b Healthy controls were not included. c Data refer to the 72 patients in the first week of Kawasaki disease. d Illness day 1 is the first day of fever (in Kawasaki disease) or symptoms (in febrile controls). e Hemoglobin was normalized by age (data unavailable for febrile controls). f Ethnicity percentages were calculated for the denominator with recorded data. Patients in the validation study group were similarly recruited as part of biomarker studies of febrile children seen in the hospital and requiring blood tests, as described previously.21,22 Healthy control children with no recent history of fever or immunization were recruited alongside patients with KD and febrile control patients in the discovery and validation studies. Data from healthy controls were used to standardize data obtained in different microarray experiments but were not used to evaluate the performance of the signature.

no recent history of fever or immunization were recruited alongside patients with KD and febrile control patients in the discovery and validation studies. Data from healthy controls were used to standardize data obtained in different microarray experiments but were not used to evaluate the performance of the signature. KD Case Definition Kawasaki disease was diagnosed on the basis of the American Heart Association criteria,14 with 2-dimensional echocardiography performed soon after presentation (2 and 6 weeks after onset). Patients with fewer than 4 of the 5 classic criteria (bilateral nonpurulent conjunctivitis, oral mucosal changes, peripheral extremity changes, rash, and cervical lymphadenopathy >1.5 cm) were included as having incomplete KD if the maximum coronary artery z score (Zmax) (standard deviation units from the mean internal diameter normalized for body surface area) at any time during the illness for the left anterior descending or right coronary arteries was 2.5 or higher or if the patients satisfied the algorithm for incomplete KD in the American Heart Association guidelines. Patients were classified as having normal (Zmax <2.5), small (Zmax 2.5 to <5.0), or large (Zmax ≥5.0) CAAs. Because of interoperator variability in coronary artery dimensions, we set a high (Zmax ≥5.0) threshold to define patients with confirmed aneurysms and thus definite diagnosis of KD.

Heart Association guidelines. Patients were classified as having normal (Zmax <2.5), small (Zmax 2.5 to <5.0), or large (Zmax ≥5.0) CAAs. Because of interoperator variability in coronary artery dimensions, we set a high (Zmax ≥5.0) threshold to define patients with confirmed aneurysms and thus definite diagnosis of KD. Further Classification of KD by Diagnostic Certainty Because there is no gold standard for diagnosis, some patients may meet the criteria for KD but have other conditions. Therefore, we further categorized patients with KD in the validation study group based on certainty of clinical diagnosis. All clinical records, laboratory results, echocardiogram reports, response to treatment, and follow-up were reviewed by an independent pediatric infectious disease specialist and expert on KD (M.P.G.) masked to the analysis. Patients with documented CAAs (Zmax ≥5.0) persisting 6 weeks after onset were considered to have definite KD because there is no other known self-resolving inflammatory illness in childhood that leads to CAAs. The remaining patients (all of whom were treated with IVIG for suspected KD) were classified as having highly probable, possible, or unlikely KD by the expert reviewer. This review identified no unlikely KD cases.

te KD because there is no other known self-resolving inflammatory illness in childhood that leads to CAAs. The remaining patients (all of whom were treated with IVIG for suspected KD) were classified as having highly probable, possible, or unlikely KD by the expert reviewer. This review identified no unlikely KD cases. Febrile Control Children With Infection or Other Inflammatory Syndromes Children seen with febrile illnesses were recorded as having definite bacterial infection, definite viral infection, suspected bacterial or viral infection, juvenile idiopathic arthritis, or Henoch-Schönlein purpura. The criteria shown in Figure 1 and described in eMethods in the Supplement were used to make this determination. Oversight and Conduct of the Study Patients were categorized into disease groups (Figure 1) after evaluation of all results by 2 independent clinicians not involved in the patients’ care (J.A.H., A.M.B., J.T.K., M.P.G., and J.C.B.). All blood samples were anonymized, and transcriptomic data sets were analyzed only after clinical assignments were finalized and dispatched for independent verification (eMethods in the Supplement).

all results by 2 independent clinicians not involved in the patients’ care (J.A.H., A.M.B., J.T.K., M.P.G., and J.C.B.). All blood samples were anonymized, and transcriptomic data sets were analyzed only after clinical assignments were finalized and dispatched for independent verification (eMethods in the Supplement). Discovery and Validation of the Gene Expression Signature The overall study design and signature discovery pipeline are shown in Figure 2. Whole blood was collected at the time of recruitment into blood RNA tubes (PAXgene; PreAnalytiX), frozen, extracted, and analyzed on arrays (HumanHT-12 version 4.0 BeadChip; Illumina). An earlier array (HumanHT-12 version 3.0 BeadChip; Illumina) with largely overlapping probes was used in a subset of the validation study group. Details of laboratory methods are provided in eMethods in the Supplement. Figure 2. Study Design The overall study pipeline shows sample handling, derivation of test and training data sets, data processing, and analysis pipeline. Version 3 arrays indicate HumanHT-12, version 3.0 BeadChip (Illumina); version 4 arrays indicate HumanHT-12, version 4.0 BeadChip (Illumina); and ComBat indicates the ComBat algorithm.23 DB indicates definite bacterial; DV, definite viral; FC, fold change; HC, healthy controls; HSP, Henoch-Schönlein purpura; JIA, juvenile idiopathic arthritis; KD, Kawasaki disease; PReMS, parallel regularized regression model search; SDE, significantly differentially expressed; and U, infections of uncertain bacterial or viral etiology.

ite bacterial; DV, definite viral; FC, fold change; HC, healthy controls; HSP, Henoch-Schönlein purpura; JIA, juvenile idiopathic arthritis; KD, Kawasaki disease; PReMS, parallel regularized regression model search; SDE, significantly differentially expressed; and U, infections of uncertain bacterial or viral etiology. aSee Supplemental Methods (RNA sample extraction and processing), as well as Statistical Methods in eMethods in the Supplement. bHealthy controls were used in model building but were excluded from estimates of model accuracy. cSee Statistical Methods in eMethods in the Supplement; 146 acute KD samples (HumanHT-12, version 4.0) were used in Combat, of which 101 were taken forward. dDiagnostic performance was assessed on 72 patients (within the first 7 days of illness). eIncludes convalescent KD and healthy controls. Statistical Analysis Transcript Signature Discovery Analysis of the transcriptomic data was conducted with statistical software (R, version 3.2.2; R Foundation for Statistical Computing). As shown in Figure 2, the discovery study group was randomly divided into an 80% training set and a 20% test set. The signature was identified in the training set and validated in the test set and in the validation study group, established using previously reported acute and convalescent patients with KD21 and acute bacterial and viral patients22 (eMethods in the Supplement). After quality control and filtering (eMethods in the Supplement), significantly differentially expressed transcripts in patients with KD compared with all other diseases were identified in the training set.

acute and convalescent patients with KD21 and acute bacterial and viral patients22 (eMethods in the Supplement). After quality control and filtering (eMethods in the Supplement), significantly differentially expressed transcripts in patients with KD compared with all other diseases were identified in the training set. Small Signature Discovery Using Parallel Regularized Regression Model Search A range of statistical methods are available to identify signatures from significantly differentially expressed transcripts, including least absolute shrinkage and selection operator (LASSO)24 and elastic net.25 However, these approaches produce large signatures that may not be easy to translate into a bedside diagnostic test. Therefore, we developed a novel variable selection method, parallel regularized regression model search, that identifies and ranks transcript signatures on the basis of their least number of transcripts and highest accuracy in discrimination26 (eMethods in the Supplement). The method first evaluates all possible 1- and 2-transcript models distinguishing KD from comparator diseases based on all significantly differentially expressed transcripts and takes the 100 best-fitting 2-transcript models to the next round, when a further transcript is added to the model and all combinations are again evaluated. The process continues with the incremental addition of 1 further transcript at a time to the best 100 models. The optimal signature for a given number of transcripts (model size) was selected after ranking each model by its Watanabe-Akaike information criterion, which is a Bayesian estimate of the out-of-sample error.27 The optimal model size was determined by cross-validation.

ther transcript at a time to the best 100 models. The optimal signature for a given number of transcripts (model size) was selected after ranking each model by its Watanabe-Akaike information criterion, which is a Bayesian estimate of the out-of-sample error.27 The optimal model size was determined by cross-validation. Disease Risk Score and Assessment of Model Accuracy We applied the previously reported disease risk score (DRS) method that assigns individual disease risk based on the transcripts in the diagnostic signature.15 The DRS combines the fluorescence intensity of upregulated transcripts and subtracts the combined fluorescence intensity of down-regulated transcripts15 and might facilitate development of tests from complex signatures. Healthy controls were used in model building but were excluded from estimates of model accuracy, assessed by area under the curve (AUC), sensitivity, and specificity at the optimal cut point according to the Youden index.

of down-regulated transcripts15 and might facilitate development of tests from complex signatures. Healthy controls were used in model building but were excluded from estimates of model accuracy, assessed by area under the curve (AUC), sensitivity, and specificity at the optimal cut point according to the Youden index. Results The numbers of patients in each diagnostic category are shown in Figure 2. Clinical and demographic features of patients with KD and febrile controls are summarized in Table 1, with further details of control patients listed in eTable 1 in the Supplement. Principal component analysis of the normalized transcript expression profiles was performed separately on the discovery (training and test) and validation groups (eFigure 1 and eFigure 2 in the Supplement). Study groups clustered together in the discovery group and in the validation group after combining KD and case-control data using the ComBat algorithm23 (Statistical Methods in eMethods in the Supplement).

separately on the discovery (training and test) and validation groups (eFigure 1 and eFigure 2 in the Supplement). Study groups clustered together in the discovery group and in the validation group after combining KD and case-control data using the ComBat algorithm23 (Statistical Methods in eMethods in the Supplement). Identification of Minimal Transcript Signatures In total, 1600 transcripts passed quality control and were significantly differentially expressed between KD and all other diseases and healthy controls (defined as log2 fold change >1 in KD vs at least 1 of the comparator groups). To identify small signatures suitable for developing as a diagnostic test, we next undertook variable selection using parallel regularized regression model search. This approach identified a 13-transcript signature (Table 2) that, when implemented as a DRS, had a diagnostic performance in the discovery test set distinguishing KD from other infectious and inflammatory conditions, with an AUC of 96.2% (95% CI, 92.5%-99.9%), sensitivity of 81.7% (95% CI, 60.0%-94.8%), and specificity of 92.1% (95% CI, 84.0%-97.0%) (Figure 3A and B).

(Table 2) that, when implemented as a DRS, had a diagnostic performance in the discovery test set distinguishing KD from other infectious and inflammatory conditions, with an AUC of 96.2% (95% CI, 92.5%-99.9%), sensitivity of 81.7% (95% CI, 60.0%-94.8%), and specificity of 92.1% (95% CI, 84.0%-97.0%) (Figure 3A and B). Table 2. Genes Included in the Diagnostic Signature Gene Symbol Gene Name HGNC Identification No. Probe Identification No. Location Logistic Regression Coefficienta CACNA1E Calcium voltage-gated channel subunit alpha1 E 1392 7510647 1q25.3 0.955 DDIAS DNA damage–induced apoptosis suppressor 26351 2570019 11q14.1 0.844 KLHL2 Kelch-like family member 2 6353 1070593 4q32.3 0.789 PYROXD2 Pyridine nucleotide-disulphide oxidoreductase domain 2 23517 1684497 10q24.2 0.727 SMOX Spermine oxidase 15862 270068 20p13 0.675 ZNF185 Zinc finger protein 185 with domain 12976 6840674 Xq28 0.646 LINC02035 Long intergenic non–protein coding RNA 2035 52875 3236239 3q21.1 0.561 CLIC3 Chloride intracellular channel 3 2064 5870136 9q34.3 0.464 S100P S100 calcium-binding protein P 10504 1510424 4p16.1 −0.405 IFI27 Interferon alpha–inducible protein 27 5397 3990170 14q32.12 −0.426 HS.553068 BX103476 NCI_CGAP_Lu5 Homo sapiens cDNA clone NA 1470450 NA −0.599 CD163 CD163 molecule 1631 2680092 12p13.31 −0.638 RTN1 Reticulon 1 10467 6860193 14q23.1 −0.690 Abbreviations: cDNA, complementary DNA; HGNC, Hugo Gene Nomenclature Committee; NA, not applicable.

ducible protein 27 5397 3990170 14q32.12 −0.426 HS.553068 BX103476 NCI_CGAP_Lu5 Homo sapiens cDNA clone NA 1470450 NA −0.599 CD163 CD163 molecule 1631 2680092 12p13.31 −0.638 RTN1 Reticulon 1 10467 6860193 14q23.1 −0.690 Abbreviations: cDNA, complementary DNA; HGNC, Hugo Gene Nomenclature Committee; NA, not applicable. a The logistic regression coefficient indicates the power of the gene to discriminate Kawasaki disease in the parallel regularized regression model search. Genes with positive values show increased expression in Kawasaki disease relative to other diseases, and genes with negative values show decreased expression in Kawasaki disease.

regression coefficient indicates the power of the gene to discriminate Kawasaki disease in the parallel regularized regression model search. Genes with positive values show increased expression in Kawasaki disease relative to other diseases, and genes with negative values show decreased expression in Kawasaki disease. Figure 3. Performance of the 13-Transcript Signature on the Discovery Test Set and the Validation Set Shown is classification (A) and ROC curve (B) of the 13-transcript signature in the discovery test set, comprising patients with KD and patients with other diseases, using the disease risk score. Shown is classification (C) and ROC curves (D) of the 13-transcript signature in the validation set, comprising 3 KD clinical subgroups of differing diagnostic certainty and patients with other diseases. In box plots, horizontal lines represent the median; lower and upper edges represent interquartile ranges; and whiskers represent the range or 1.5 times the interquartile range, whichever is smaller. The horizontal blue line indicates the disease risk score threshold that separates patients predicted as having KD (above the line) or not having KD (below the line) as determined by the point in the ROC curve that maximized sensitivity and specificity in the discovery training group. DB indicates definite bacterial; DV, definite viral; HSP, Henoch-Schönlein purpura; JIA, juvenile idiopathic arthritis; KD, Kawasaki disease; KD-Def, definite KD; KD-HP, highly probable KD; KD-P, possible KD; ROC, receiver operating characteristic; and U, infections of uncertain bacterial or viral etiology.

ning group. DB indicates definite bacterial; DV, definite viral; HSP, Henoch-Schönlein purpura; JIA, juvenile idiopathic arthritis; KD, Kawasaki disease; KD-Def, definite KD; KD-HP, highly probable KD; KD-P, possible KD; ROC, receiver operating characteristic; and U, infections of uncertain bacterial or viral etiology. Signature Performance in Validation Set When the signature was applied to all of the 72 KD cases in the validation set, who were in the first 7 days of illness, the AUC was 94.6% (95% CI, 91.3%-98.0%), with sensitivity of 85.9% (95% CI, 76.8%-92.6%) and specificity of 89.1% (95% CI, 83.0%-93.7%). The performance was slightly reduced in the 30 patients diagnosed later (days 8-10) (eTable 2 and eFigure 3 in the Supplement). Because clinical features of KD overlap those of other conditions and because any KD study group is likely to include patients misclassified as KD, we assessed whether certainty of clinical diagnosis corresponded to the predictive performance of the KD DRS. The performance of the 13-transcript signature in the patients with definite, highly probable, or possible KD in the validation set mirrored certainty of clinical diagnosis, with AUCs of 98.1% (95% CI, 94.5%-100%), 96.3% (95% CI, 93.3%-99.4%), and 70.0% (95% CI, 53.4%-86.6%), respectively (Figure 3C and D and eTable 2 in the Supplement).

the 13-transcript signature in the patients with definite, highly probable, or possible KD in the validation set mirrored certainty of clinical diagnosis, with AUCs of 98.1% (95% CI, 94.5%-100%), 96.3% (95% CI, 93.3%-99.4%), and 70.0% (95% CI, 53.4%-86.6%), respectively (Figure 3C and D and eTable 2 in the Supplement). Discussion We identified a 13-transcript signature that distinguishes patients with KD from patients with bacterial, viral, and inflammatory illnesses. The high sensitivity and specificity of this signature for early diagnosis of KD suggests that it might form the basis of a diagnostic test. Our findings herein extend previous gene expression studies21,28,29,30,31,32 in KD that focused on immunopathogenesis.

patients with bacterial, viral, and inflammatory illnesses. The high sensitivity and specificity of this signature for early diagnosis of KD suggests that it might form the basis of a diagnostic test. Our findings herein extend previous gene expression studies21,28,29,30,31,32 in KD that focused on immunopathogenesis. The diagnosis of KD now relies on the presence of 4 of the 5 characteristic clinical criteria. Fewer criteria are accepted if coronary artery abnormalities (dilatation or aneurysms) are detected on echocardiography. Children with incomplete KD who do not fulfil the classic diagnostic criteria but have prolonged fever and inflammation are at an increased risk of developing CAAs.33 One reason for the greater risk of CAAs in incomplete KD is the delayed diagnosis that often occurs in patients lacking all clinical features. Because the clinical features of KD overlap those of many other common childhood conditions,34 treatment with IVIG may be delayed while awaiting exclusion of other conditions. Conversely, because the diagnosis of KD is considered in the differential diagnosis of many childhood febrile illnesses and because the consequences of delayed treatment may be severe, overtreatment with IVIG or immunosuppressant second-line treatments may occur. A diagnostic test that accurately distinguishes KD from other infectious and inflammatory processes would be a significant advance in management of the disorder, reduce unnecessary investigations and inappropriate treatments, and enable earlier treatment with IVIG and other anti-inflammatory agents.

eatments may occur. A diagnostic test that accurately distinguishes KD from other infectious and inflammatory processes would be a significant advance in management of the disorder, reduce unnecessary investigations and inappropriate treatments, and enable earlier treatment with IVIG and other anti-inflammatory agents. In establishing our discovery and validation study groups, we aimed to include a wide range of disorders with features overlapping those of KD, including both infectious and inflammatory diseases. The signature that we have identified distinguished KD from a wide range of other conditions with similar duration of fever and overlapping levels of inflammation. Because KD is diagnosed based on a constellation of clinical features and because there is no gold standard for diagnosis, evaluation of biomarker test results is difficult. Any cohort of children treated with IVIG for presumed KD is likely to include some patients with non-KD illness but with similar features. To evaluate the correspondence of the KD DRS with levels of diagnostic certainty, we categorized patients in the validation set as having definite, highly probable, or possible KD based on independent review of the clinical data. We observed a higher sensitivity and specificity of our signature in the definite and highly probable KD groups than in the possible KD group.

vels of diagnostic certainty, we categorized patients in the validation set as having definite, highly probable, or possible KD based on independent review of the clinical data. We observed a higher sensitivity and specificity of our signature in the definite and highly probable KD groups than in the possible KD group. Regarding the transcripts in the signature (Table 2), expression was lower in patients with KD compared with the non-KD group for 5 of the 13 transcripts. Of these, S100P, previously reported to have increased expression in acute KD relative to convalescence35 or viral infections,32,35 was most abundant in patients with bacterial infection. The IFI27 gene has been reported to be upregulated in children with viral compared with bacterial infections36 and autoimmune diseases,37,38 consistent with reduced expression of genes induced by type 1 interferons reported in acute KD vs adenovirus infection.32 CD163 is a transmembrane receptor expressed in macrophages and monocytes involved in bacterial clearance during acute infection.39 A network analysis of the signature using pathway analysis (Ingenuity Pathways Analysis; Ingenuity Systems) revealed that 7 of the 13 transcripts in the signature were connected in a network around a central hub of tumor necrosis factor and interleukin 6 (eFigure 4 in the Supplement).

learance during acute infection.39 A network analysis of the signature using pathway analysis (Ingenuity Pathways Analysis; Ingenuity Systems) revealed that 7 of the 13 transcripts in the signature were connected in a network around a central hub of tumor necrosis factor and interleukin 6 (eFigure 4 in the Supplement). Strengths and Limitations We recognize both strengths and limitations in our study. First, the epidemiology of KD varies globally by ethnicity. Although we included patients with a mix of ethnicities in both discovery and validation cohorts, further studies are required to establish the performance of the signature in other geographic populations. Second, in the validation experiment, data from different Illumina microarray versions and studies were combined using the ComBat algorithm to achieve normalization. This normalization may reduce both experimental and biological sources of variability between data sets; consequently, the accuracy (AUC) of the signature in the validation set may be an underestimate. Conversely, although we showed that the ComBat algorithm successfully normalized the data sets, residual batch associations may have falsely increased the performance of the signature. Third, to develop a signature applicable in a wide range of febrile conditions, we discovered the signature through comparison of KD cases with a wide range of febrile controls with a spectrum of KD diagnosis likelihoods. This potentially biased the test toward better discriminatory value than might be applicable in the clinical setting. Fourth, we discovered the signature using patients with KD in the first 7 days of illness, with the aim of identifying a test for use early in the disease to enable treatment before coronary injury has occurred. Because the performance of the signature was lower in patients with KD seen after the seventh day, further work is required to establish the optimal signature for diagnosis in patients with KD with late presentation.

dentifying a test for use early in the disease to enable treatment before coronary injury has occurred. Because the performance of the signature was lower in patients with KD seen after the seventh day, further work is required to establish the optimal signature for diagnosis in patients with KD with late presentation. Conclusions The results of our study suggest that KD can be distinguished from the range of infectious and inflammatory conditions with which it is often clinically confused using 13 transcripts in blood. Development of a test based on this gene expression signature is made more achievable because of the small number of transcripts in our signature and the rapidly evolving technologies for detecting nucleic acids. A diagnostic test would be a major advance allowing earlier treatment and thus prevention of cardiac complications of this serious childhood disease. Our findings represent a step toward better diagnosis of diseases based on molecular signatures rather than the clinical criteria and are relevant to many other clinical syndromes. Supplement. eMethods. Supplemental Methods eTable 1A. Clinical Features of Children in the Juvenile Idiopathic Arthritis Cohort (Discovery) eTable 1B. Clinical Features of Children in the Henoch-Schönlein Purpura Group (Discovery) eTable 1C. Clinical Features of Children With Bacterial and Viral Infection, Infections of Uncertain Bacterial or Viral Etiology, and Healthy Controls (Discovery and Validation) eTable 1D. Viral and Bacterial Causative Pathogens in Patients in the Definite Bacterial and Viral Groups

eTable 1B. Clinical Features of Children in the Henoch-Schönlein Purpura Group (Discovery) eTable 1C. Clinical Features of Children With Bacterial and Viral Infection, Infections of Uncertain Bacterial or Viral Etiology, and Healthy Controls (Discovery and Validation) eTable 1D. Viral and Bacterial Causative Pathogens in Patients in the Definite Bacterial and Viral Groups eTable 2. Summary of Performance of Models eFigure 1. Principal Component Analysis on the Discovery Cohort eFigure 2. Principal Component Analysis on Validation Sets Before and After Merging Using ComBat eFigure 3. Performance of the 13-Transcript Signature by Illness Day at Sample Collection in Validation Set eFigure 4. Gene Network Derived From 13-Transcript Signature Click here for additional data file.

Introduction Human body mass index (BMI) (calculated as weight in kilograms divided by height in meters squared) is highly heritable, as indicated in recent reviews of twin studies.1,2 However, there is substantial variation in BMI heritability estimates, which range from 31% to 90%.2 This variation has been attributed to both population and socioenvironmental characteristics. The heritability of BMI is higher in populations with higher average BMIs,2 in countries with higher gross domestic product,2 in populations born later,3 and in families of lower socioeconomic status.4,5 These findings are in line with the hypothesis that obesity-related genes are more strongly associated with BMI in more obesogenic home environments. Molecular genetic studies have corroborated findings from twin studies, showing that the environment modifies the association between measured genetic risk of obesity and BMI. In a large European sample of children (n = 4406), the effect of the FTO genotype on BMI was stronger among children with parents of low socioeconomic status.6 In another study, the association between a composite indicator of genetic risk of obesity and BMI was stronger for more recent birth cohorts, who by implication had had greater exposure to the obesogenic environment.7

ect of the FTO genotype on BMI was stronger among children with parents of low socioeconomic status.6 In another study, the association between a composite indicator of genetic risk of obesity and BMI was stronger for more recent birth cohorts, who by implication had had greater exposure to the obesogenic environment.7 Differences in economic growth and socioeconomic status are macro-level influences of the environment. The food, physical activity, and entertainment environments are proximal or micro-level influences on energy intake and physical activity; these include the home, school, and neighborhood settings.8 Some research has found that living in more walkable neighborhood environments suppresses genetic variance in adult BMI.9 However, no studies have examined whether the heritability of BMI varies by the home environment in childhood. This is an important research endeavor because the home environment is within an individual’s control and has been identified as a key influence on early weight trajectories.10,11 Understanding the role of the home environment from a gene-environment perspective can further inform home-based childhood obesity prevention and treatment efforts, which have been ineffective.12

environment is within an individual’s control and has been identified as a key influence on early weight trajectories.10,11 Understanding the role of the home environment from a gene-environment perspective can further inform home-based childhood obesity prevention and treatment efforts, which have been ineffective.12 The obesogenic home environment incorporates food, physical activity, and media-related influences, such as the availability of healthy and unhealthy foods, opportunities for physical activity, and parental rules around media use.13,14 Any single aspect of the home environment probably has limited influence on weight-related outcomes; therefore, composite measures should capture overall obesogenic risk most effectively. Recent findings have shown that preschool children who lived in higher-risk home environments, as measured by the Home Environment Interview (HEI) (the sum of 21 food-related, 6 physical activity-related, and 5 media-related factors), had poorer diets, engaged in less physical activity, and watched more television than did children who lived in lower-risk home environments.15 This study expands previous research by examining whether the heritability of child BMI varies by the early obesogenic home environment. It is hypothesized that the heritability of BMI will be higher among children living in higher-risk home environments compared with those living in lower-risk home environments.

The obesogenic home environment incorporates food, physical activity, and media-related influences, such as the availability of healthy and unhealthy foods, opportunities for physical activity, and parental rules around media use.13,14 Any single aspect of the home environment probably has limited influence on weight-related outcomes; therefore, composite measures should capture overall obesogenic risk most effectively. Recent findings have shown that preschool children who lived in higher-risk home environments, as measured by the Home Environment Interview (HEI) (the sum of 21 food-related, 6 physical activity-related, and 5 media-related factors), had poorer diets, engaged in less physical activity, and watched more television than did children who lived in lower-risk home environments.15 This study expands previous research by examining whether the heritability of child BMI varies by the early obesogenic home environment. It is hypothesized that the heritability of BMI will be higher among children living in higher-risk home environments compared with those living in lower-risk home environments. Methods Sample Gemini cohort data (a nationally representative twin study of early growth16) were used in this study. In total, 2402 of 6754 families (36% of those with live twin births in England and Wales during March-December 2007) gave written consent to participate and completed a baseline questionnaire when their children were a mean (SD) of 8.2 (2.2) months of age (range, 4–20 months). The HEI was completed by 1113 of 2402 families (46% of the total sample) when the children were a mean (SD) of 4.2 (0.4) years of age (range, 3–5 years). This study sample comprised 925 twin pairs (1850 twins) with data on all study variables. Data were analyzed from July to October 2013 and in June 2018. Ethical approval was granted by the University College London Committee for the Ethics of non–National Health Service Human Research. Data were deidentified.

5 years). This study sample comprised 925 twin pairs (1850 twins) with data on all study variables. Data were analyzed from July to October 2013 and in June 2018. Ethical approval was granted by the University College London Committee for the Ethics of non–National Health Service Human Research. Data were deidentified. Measures Zygosity Opposite-sex twins were classified as dizygotic (DZ). Parents of same-sex twins were asked to complete a previously validated 20-item zygosity questionnaire,17 which assesses the twins’ physical likeness, blood type, how easily friends and family members can tell the twins apart, and parents and health professionals’ opinions about the twins’ zygosity. The questionnaire showed 100% agreement with DNA samples of 81 randomly selected Gemini twin pairs (43 monozygotic [MZ] twins and 38 DZ twins) at 29 months of age.18 Body Mass Index Electronic weighing scales and height charts were sent to all families when the twins were 2 years of age to collect parent-reported measurements every 3 months. Parents also provided their twins’ heights and weights at the time of the HEI. The BMI SD scores, adjusted for age and sex, were calculated using British 1990 growth reference data19 and the LMS growth macro for Excel (Microsoft Corporation). Home Environment Primary caregivers (1102 of 1113 caregivers [99%] were mothers) completed the HEI by telephone when their twins were 4 years of age. The HEI is a comprehensive home environment measure assessing food, physical activity, and media-related influences.15

Body Mass Index Electronic weighing scales and height charts were sent to all families when the twins were 2 years of age to collect parent-reported measurements every 3 months. Parents also provided their twins’ heights and weights at the time of the HEI. The BMI SD scores, adjusted for age and sex, were calculated using British 1990 growth reference data19 and the LMS growth macro for Excel (Microsoft Corporation). Home Environment Primary caregivers (1102 of 1113 caregivers [99%] were mothers) completed the HEI by telephone when their twins were 4 years of age. The HEI is a comprehensive home environment measure assessing food, physical activity, and media-related influences.15 As described elsewhere,15 the level of obesogenic risk was determined by creating composite scores, guided by feedback from an international panel of 30 experts in pediatric obesity. A total of 32 constructs were included in the composites (eTable 1 in the Supplement). Constructs associated with lower risk of excessive weight gain were reverse-scored so that higher total scores would reflect higher obesogenic risk. Each variable was standardized using z scores and summed to create composite scores for the home food environment (21 variables), the home activity environment (6 variables), and the home media environment (5 variables). There were few cases with missing data on home environment variables; these were recoded to 0 (the mean value for each standardized variable). The 3 composites were summed to create an overall home environment composite, dividing by the number of variables per composite so that each domain contributed equally to the overall score (food composite/21 + activity composite/6 + media composite/5).

were recoded to 0 (the mean value for each standardized variable). The 3 composites were summed to create an overall home environment composite, dividing by the number of variables per composite so that each domain contributed equally to the overall score (food composite/21 + activity composite/6 + media composite/5). Test-retest reliability of the home environment composites from 7 to 19 days (mean [SD], 9.6 [3.4] days) was acceptable to high. The intraclass correlation coefficients were 0.71 (95% CI, 0.52–0.83) for food, 0.83 (95% CI, 0.72–0.91) for activity, 0.92 (95% CI, 0.85–0.95) for media, and 0.92 (95% CI, 0.86–0.96) overall. An overview of the measurement points is given in eTable 2 in the Supplement. Statistical Analyses Heritability Analyses Genetic and environmental contributions to variation in a trait can be estimated by comparing similarity between MZ twins (who share 100% of their genes) with that between DZ twins (who share approximately 50% of their genes). Comparing MZ and DZ correlations enables variation in a trait to be decomposed into 3 latent factors (the ACE model): additive genetic effects (ie, heritability) (A); shared environmental influence (shared experiences that make twins within a pair similar) (C); and nonshared environmental influence (experiences unique to an individual that make twins within a pair different) (E), which also includes random measurement error.20

model): additive genetic effects (ie, heritability) (A); shared environmental influence (shared experiences that make twins within a pair similar) (C); and nonshared environmental influence (experiences unique to an individual that make twins within a pair different) (E), which also includes random measurement error.20 Two methods were used to estimate the heritability of BMI at 4 years of age: twin correlations and maximum likelihood structural equation modeling (MLSEM).21 For each method, 4-year BMI SD score was residualized for age at BMI measurement and sex effects using linear regression.22 The analyses were repeated using BMI SD scores additionally residualized for gestational age, which is also exactly correlated within twin pairs. Heritability estimates for 4-year BMI SD scores were calculated for the total sample and for home environment groups dichotomized on the mean (0): lower (≤0) and higher (>0) overall risk, food, activity, and media home environments. Twin Correlations Intraclass correlations were calculated for each zygosity (MZ and DZ) and for each zygosity by each home environment group (eg, MZs living in a home environment with higher overall risk) in R23 using the structural equation modeling software OpenMx, version 2.2.6.24

Heritability estimates for 4-year BMI SD scores were calculated for the total sample and for home environment groups dichotomized on the mean (0): lower (≤0) and higher (>0) overall risk, food, activity, and media home environments. Twin Correlations Intraclass correlations were calculated for each zygosity (MZ and DZ) and for each zygosity by each home environment group (eg, MZs living in a home environment with higher overall risk) in R23 using the structural equation modeling software OpenMx, version 2.2.6.24 Model Fitting Univariate twin models were created in R23 using the structural equation modeling software OpenMx, version 2.2.624 to produce reliable parameter estimates for the whole sample with 95% CIs and goodness-of-fit statistics. A heterogeneity model was used to test for differences in the magnitude of A, C, and E between the lower-risk and higher-risk home environment groups (eFigure in the Supplement). A, C, and E were estimated using the covariance between twins. Because MZs share 100% of their genes and DZs share approximately 50% of their genes, the genetic correlations within MZ and DZ pairs were fixed at 1.0 and 0.5, respectively. Because it is assumed that shared environmental influences are equal for MZ and DZ twins, the shared environmental correlation was fixed at 1.0 for both zygosities.

00% of their genes and DZs share approximately 50% of their genes, the genetic correlations within MZ and DZ pairs were fixed at 1.0 and 0.5, respectively. Because it is assumed that shared environmental influences are equal for MZ and DZ twins, the shared environmental correlation was fixed at 1.0 for both zygosities. A common effects model was fitted to compare parameter estimates in lower-risk and higher-risk home environment groups. This model allows the magnitude of variance explained by A, C, and E to differ between groups. The fit of more constrained nested models was then compared with the original model using likelihood ratio tests. A significant difference between the negative log-likelihood of the nested model and that of the original model indicates a deterioration in model fit.25,26 The 2 nested models in this study were the scalar model, which allows variance differences but not quantitative differences between groups, and the null model, which constrains all parameters to be the same across the 2 groups. If the scalar or null models show a better fit than the common effects model, there are no quantitative differences in parameter estimates between groups.25,26 Statistical significance was set at .05, and P values were 1-sided.

ups, and the null model, which constrains all parameters to be the same across the 2 groups. If the scalar or null models show a better fit than the common effects model, there are no quantitative differences in parameter estimates between groups.25,26 Statistical significance was set at .05, and P values were 1-sided. Results Sample Characteristics Of the total HEI sample (1113 families; 2226 twins), 12 twin-pairs had unknown zygosity, and 174 first-born twins and 177 second-born twins had missing data for 4-year BMI. This left a sample of 925 twin pairs (1850 twins; 915 [49.5%] male and 935 [50.5%] female; mean [SD] age, 4.1 [0.4] years). There were no significant differences between the study sample and the total HEI sample with respect to the study variables (eTable 3 in the Supplement).

orn twins had missing data for 4-year BMI. This left a sample of 925 twin pairs (1850 twins; 915 [49.5%] male and 935 [50.5%] female; mean [SD] age, 4.1 [0.4] years). There were no significant differences between the study sample and the total HEI sample with respect to the study variables (eTable 3 in the Supplement). Three hundred fourteen of 925 twin pairs (34%) were MZ. There were slightly more twin pairs living in lower-risk home environments than higher-risk homes (508 [56%] vs 417 [46%]). Mean (SD) 4-year BMI SD score was below that of the reference population (first-born twins: −0.01 [1.03]; second-born twins: −0.10 [1.03]). The ranges for the home environment composites (standardized scores) showed that there was substantial variation (overall, −2.44 to 4.02; food, −19.24 to 25.24; activity, −4.93 to 16.15; media, −7.00 to 18.12). Sample characteristics by higher-risk and lower-risk home environments (overall) are shown in Table 1. Families living in higher-risk home environments had significantly higher risk scores for each of the food (t838 = −19.35; P < .001), physical activity (t683.44 = −18.85; P < .001), and media (t628.05 = −18.73; P < .001) environment composites compared with those living in lower-risk home environments. The proportion of university-educated mothers (χ22 = 31.57) and families with professional occupations (χ22 = 26.70) was significantly smaller among those living in higher-risk home environments (P < .001).

a (t628.05 = −18.73; P < .001) environment composites compared with those living in lower-risk home environments. The proportion of university-educated mothers (χ22 = 31.57) and families with professional occupations (χ22 = 26.70) was significantly smaller among those living in higher-risk home environments (P < .001). Table 1. Characteristics of the Study Sample by Overall Home Environment Risk Characteristics Overall Higher-Risk Home Environment (n = 417) Overall Lower-Risk Home Environment (n = 508) P Value Differencea Age at HEI, mean (SD), y 4.13 (0.44) 4.16 (0.37) .19 Sex of twin pair, No. (%) Male 147 (35.3) 167 (32.9) .74 Female 144 (34.5) 180 (35.4) Opposite sex 126 (30.2) 161 (31.7) Zygosity, No. (%) Monozygotic 151 (36.2) 163 (32.1) .19 Dizygotic 266 (63.8) 345 (67.9) Maternal educational level, No. (%)b Low 80 (19.2) 56 (11.0) <.001 Medium 170 (40.8) 157 (30.9) High 167 (40.0) 295 (58.1) NSSEC, No. (%)c Low 75 (18.0) 46 (9.1) <.001 Medium 76 (18.3) 62 (12.2) High 265 (63.7) 399 (78.7) Composite score, mean (range) 0.81 (−0.03 to 4.02) −0.70 (−2.44 to −0.03) <.001 Food score, mean (range) 3.84 (−11.35 to 25.24) −3.09 (−19.24 to 9.46) <.001 Activity score, mean (range) 1.85 (−4.93 to 16.15) −1.49 (−4.93 to 5.79) <.001 Media score, mean (range) 1.86 (−6.45 to 18.12) −1.81 (−7.00 to 4.37) <.001 4-y BMI SD score, mean (SD) −0.06 (1.05) −0.02 (0.99) .57 Abbreviations: BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); HEI, Home Environment Interview; NSSEC, National Statistics Socio-economic Classification.

mean (range) 1.86 (−6.45 to 18.12) −1.81 (−7.00 to 4.37) <.001 4-y BMI SD score, mean (SD) −0.06 (1.05) −0.02 (0.99) .57 Abbreviations: BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); HEI, Home Environment Interview; NSSEC, National Statistics Socio-economic Classification. a Characteristics of those living in higher-risk vs lower-risk home environments were compared using χ2 for categorical variables and t tests for continuously distributed variables. One twin was selected at random to avoid clustering effects. b Educational level categorized as low (no qualifications or basic high school education), medium (vocational or advanced high school education), and high (university-level education). c NSSEC level categorized as low (lower supervisory and technical occupations, routine or semiroutine occupations, never worked, and long-term unemployed), medium (intermediate occupations, small employers, and own-account workers), and high (higher and lower managerial and professional occupations).

b Educational level categorized as low (no qualifications or basic high school education), medium (vocational or advanced high school education), and high (university-level education). c NSSEC level categorized as low (lower supervisory and technical occupations, routine or semiroutine occupations, never worked, and long-term unemployed), medium (intermediate occupations, small employers, and own-account workers), and high (higher and lower managerial and professional occupations). Twin Correlations The intraclass correlation coefficients for 4-year BMI SD score (adjusted for age and sex) by zygosity and home environment groups are shown in Table 2. Correlations were higher between MZ than DZ twins (ranges, 0.78-0.87 vs 0.37-0.54), indicating additive genetic variation in BMI. The size of the difference between MZ and DZ twins varied by the level of home environment risk, with greater differences in higher-risk than lower-risk home environments (overall, 0.46 vs 0.27; food, 0.43 vs 0.28; activity, 0.46 vs 0.27), although the difference was smaller between higher-risk and lower-risk media environments (0.39 vs 0.32). The results were the same when additionally adjusting 4-year BMI SD score for gestational age.

s in higher-risk than lower-risk home environments (overall, 0.46 vs 0.27; food, 0.43 vs 0.28; activity, 0.46 vs 0.27), although the difference was smaller between higher-risk and lower-risk media environments (0.39 vs 0.32). The results were the same when additionally adjusting 4-year BMI SD score for gestational age. Table 2. Intraclass Correlations of BMI SD Score at 4 Years by Zygosity and Home Environment Risk Home Environment Risk Group No. (%) of Twin Pairs Intraclass Correlation Coefficient (95% CI) MZ (n = 314) DZ (n = 611) MZ DZ Overall home environment Lower risk 166 (52.9) 351 (57.4) 0.78 (0.71-0.83) 0.51 (0.43-0.58) Higher risk 148 (47.1) 260 (42.6) 0.87 (0.83-0.91) 0.41 (0.31-0.51) Home food environment Lower risk 146 (46.5) 333 (54.5) 0.80 (0.73-0.85) 0.52 (0.44-0.59) Higher risk 168 (53.5) 278 (45.5) 0.84 (0.79-0.88) 0.41 (0.31-0.50) Home activity environment Lower risk 179 (57.0) 350 (57.3) 0.81 (0.76-0.86) 0.54 (0.46-0.61) Higher risk 135 (53.0) 261 (42.7) 0.83 (0.77-0.88) 0.37 (0.26-0.47) Home media environment Lower risk 174 (55.4) 375 (61.4) 0.80 (0.74-0.85) 0.48 (0.40-0.55) Higher risk 140 (44.6) 236 (38.6) 0.84 (0.78-0.88) 0.45 (0.35-0.55) Abbreviations: BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); DZ, dizygotic; MZ, monozygotic.

.88) 0.37 (0.26-0.47) Home media environment Lower risk 174 (55.4) 375 (61.4) 0.80 (0.74-0.85) 0.48 (0.40-0.55) Higher risk 140 (44.6) 236 (38.6) 0.84 (0.78-0.88) 0.45 (0.35-0.55) Abbreviations: BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); DZ, dizygotic; MZ, monozygotic. Maximum Likelihood Structural Equation Modeling For the total sample, variance in BMI was largely attributable to additive genetic factors (62%; 95% CI, 49%-75%), moderately attributable to shared environmental factors (18%; 95% CI, 5%-29%), and moderately attributable to nonshared environmental factors (20%; 95% CI, 17%-24%). Parameter estimates for higher-risk and lower-risk home environments are summarized in Table 3. For the overall home environment, the common effects model gave the best fit to the data, indicating that the heritability of BMI SD score was significantly and substantially higher (86% vs 39%) in higher-risk home environments. There was also a difference in the proportion of variance in 4-year BMI SD score attributable to shared environmental factors across the 2 groups; 34% for lower-risk home environments and 0% for higher-risk home environments. For the home food and media environments, the common effects model also provided the best fit to the data. For the home physical activity environment, there were observable differences in the parameter estimates for the higher-risk and lower-risk groups. However, the scalar model was not a significantly worse fit to the data than the common effects model, and a null model did not fit the data well. This indicated that there were significant differences in variances across the higher-risk and lower-risk groups. These results were replicated when additionally adjusting 4-year BMI SD score for gestational age.

nificantly worse fit to the data than the common effects model, and a null model did not fit the data well. This indicated that there were significant differences in variances across the higher-risk and lower-risk groups. These results were replicated when additionally adjusting 4-year BMI SD score for gestational age. Table 3. Parameter Estimates and Goodness-of-Fit Statistics for Home Environment Interaction Models That Examined the Heritability of BMI SD Score at 4 Years of Agea Home Environment, Modelb Estimate Change in AIC P Valued Additive Genetic Environment Shared Nonsharedc Overall Common effects Lower risk 0.39 (0.21-0.57) 0.34 (0.18-0.49) 0.27 (0.21-0.33) NA NA Higher risk 0.86 (0.68-0.89) 0.00 (0.00-0.17) 0.14 (0.11-0.18) NA NA Scalar 0.62 (0.49-0.75) 0.18 (0.05-0.29) 0.20 (0.17-0.24) 15.183 <.001 Null 0.62 (0.49-0.75) 0.18 (0.05-0.29) 0.20 (0.17-0.24) −1.524 .49 Food Common effects Lower risk 0.40 (0.23-0.58) 0.35 (0.18-0.49) 0.25 (0.20-0.31) NA NA Higher risk 0.83 (0.65-0.87) 0.00 (0.00-0.18) 0.17 (0.13-0.21) NA NA Scalar 0.62 (0.49-0.76) 0.18 (0.05-0.29) 0.20 (0.17-0.24) 6.693 .005 Null 0.62 (0.49-0.75) 0.18 (0.05-0.29) 0.20 (0.17-0.24) −1.446 .46 Activity Common effects Lower risk 0.49 (0.33-0.65) 0.31 (0.15-0.44) 0.21 (0.17-0.26) NA NA Higher risk 0.80 (0.60-0.84) 0.00 (0.00-0.00) 0.20 (0.16-0.26) NA NA Scalar 0.62 (0.49-0.75) 0.18 (0.05-0.29) 0.20 (0.17-0.24) 0.288 .10 Null 0.62 (0.49-0.75) 0.18 (0.05-0.29) 0.20 (0.17-0.24) −1.987 .91 Media Common effects Lower risk 0.60 (0.42-0.78) 0.18 (0.01-0.33) 0.23 (0.18-0.29) NA NA Higher risk 0.65 (0.46-0.84) 0.17 (0.00-0.34) 0.18 (0.14-0.23) NA NA Scalar 0.62 (0.49-0.76) 0.18 (0.05-0.29) 0.20 (0.17-0.24) 9.123 .002 Null 0.62 (0.49-0.75) 0.18 (0.05-0.29) 0.20 (0.17-0.24) −1.002 .32 Abbreviations: AIC, Akaike information criterion; BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); NA, not applicable.

18 (0.14-0.23) NA NA Scalar 0.62 (0.49-0.76) 0.18 (0.05-0.29) 0.20 (0.17-0.24) 9.123 .002 Null 0.62 (0.49-0.75) 0.18 (0.05-0.29) 0.20 (0.17-0.24) −1.002 .32 Abbreviations: AIC, Akaike information criterion; BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); NA, not applicable. a The BMI SD scores modeled were residuals adjusted for age at BMI measurement and sex. Presented models include all children with valid data for age, sex, Home Environment Interview score, and 4-year BMI SD score. An additional 7 cases in which just 1 twin within the pair had available BMI data were included in the maximum-likelihood structural equation modeling, performed with OpenMx software, version 2.2.6. b Statistical analyses: standard ACE model-fitting analyses for continuous data were used to model BMI SD score at 4 years of age. c Includes measurement error. d P values were based on the likelihood ratio test and AIC. A better-fitting submodel showed a change in χ2 that did not represent a significant worsening of fit designated by the P value.

b Statistical analyses: standard ACE model-fitting analyses for continuous data were used to model BMI SD score at 4 years of age. c Includes measurement error. d P values were based on the likelihood ratio test and AIC. A better-fitting submodel showed a change in χ2 that did not represent a significant worsening of fit designated by the P value. Discussion This is the first study, to our knowledge, to test behavioral susceptibility theory’s hypothesis that the heritability of BMI will be higher among children who live in more obesogenic home environments. As hypothesized, heritability of BMI was higher among children living in overall higher-risk home environments compared with those living in lower-risk home environments. The modeling indicated that none of the variance in BMI was attributable to shared environmental factors in the higher-risk group. In contrast, a similar proportion of the variance in BMI was attributable to shared environmental factors and additive genetic factors in the lower-risk group. The findings were similar when examining the heritability of BMI in the separate food and physical activity environment domains.

ctors in the higher-risk group. In contrast, a similar proportion of the variance in BMI was attributable to shared environmental factors and additive genetic factors in the lower-risk group. The findings were similar when examining the heritability of BMI in the separate food and physical activity environment domains. For the total sample, 62% (95% CI, 49%-75%) of the variance in 4-year BMI SD score was attributable to additive genetic factors, 18% (95% CI, 5%-29%) to shared environmental factors, and 20% (95% CI, 17%-24%) to nonshared environmental factors. These estimates largely concur with previous studies of 4-year-old children.27 The heritability of BMI increases throughout childhood,27,28,29 perhaps as individuals seek out environments in line with their genotype and allow it to be expressed freely (active gene-environment correlation)30 or because gene expression changes developmentally.31 This study builds on earlier findings that the heritability of BMI is higher in populations with higher average BMIs, with higher levels of gross domestic product, and with lower socioeconomic status.2 Examining the role of proximal environmental exposures is important because these factors are within an individual’s control, and it is easier to hypothesize about their potential association with neurobiological pathways that mediate the development of overweight and obesity.32

t, and with lower socioeconomic status.2 Examining the role of proximal environmental exposures is important because these factors are within an individual’s control, and it is easier to hypothesize about their potential association with neurobiological pathways that mediate the development of overweight and obesity.32 According to behavioral susceptibility theory,33,34,35 an individual’s appetitive traits confer differential susceptibility to the obesogenic environment. Individuals who have high food responsiveness and low sensitivity to satiety are more likely to overeat when there is increased opportunity to do so.33,34,35 Appetitive traits play a causal role in the development of weight,36,37 they are highly heritable,38,39 and they explain part of the association between obesity-related genes and weight.40 Many weight-related genes are highly expressed in the hypothalamus, a key regulator of appetite and food intake.41 Evidence also indicates that food intake is influenced by brain regions related to reward sensitivity and incentive motivation.42,43 It is feasible that a home environment with multiple food cues triggers appetitive and reward-related pathways, which prompt increased food intake and, subsequently, weight gain. In line with this idea, children with the FTO polymorphism associated with obesity risk had stronger responses to food commercials in the nucleus accumbens, a reward-related brain region,44 and they were more likely to consume excess calories.45 Physical activity suppresses the effect of obesity-related genes on BMI, perhaps also via appetitive and reward-related pathways.46,47 Future research should directly examine whether the home environment moderates genetic influence on BMI using a genetic risk score, because BMI is a highly polygenic trait.48,49

ries.45 Physical activity suppresses the effect of obesity-related genes on BMI, perhaps also via appetitive and reward-related pathways.46,47 Future research should directly examine whether the home environment moderates genetic influence on BMI using a genetic risk score, because BMI is a highly polygenic trait.48,49 Although there were large observable differences in parameter estimates when comparing higher-risk and lower-risk home physical activity environments (80% vs 49% for variance attributable to additive genetic factors), the model-fitting indicated that the 2 groups could be combined, with no significant worsening of fit. Significant differences may emerge in larger, higher powered samples and in more extreme home physical activity environments, because there was a skew toward lower risk in this sample.50,51 Of note, although the common effects model provided the best fit for the home media environment data, the differences in parameter estimates when comparing higher-risk and lower-risk groups were substantially smaller than those observed for the overall environment and food domain (65% vs 60% for variance attributable to additive genetic factors). There was no difference in the proportion of variance in BMI attributable to shared environmental factors across the higher-risk and lower-risk groups (17% vs 18%). It is therefore questionable that the differences observed for the home media environment are meaningful. It is possible that gene-environment effects of the home media environment are stronger in more extreme environments50,51 and later in development, when media influences are more prominent.52 Research should further examine gene-environment effects of the separate food, physical activity, and media domains in larger and more diverse samples to clarify their relative contributions.

environment are stronger in more extreme environments50,51 and later in development, when media influences are more prominent.52 Research should further examine gene-environment effects of the separate food, physical activity, and media domains in larger and more diverse samples to clarify their relative contributions. Limitations Although the findings suggest gene-environment interaction, they may be partly explained by gene-environment correlation.30,53 For example, a child may be born into a home environment that is correlated with their genotype (passive gene-environment correlation), and some aspects of the home environment, such as parental feeding practices, may be responsive to the child’s genotype (reactive gene-environment correlation). Models have been developed to take into account gene-environment correlation effects,54 but larger sample sizes are needed than that available in this study.

, and some aspects of the home environment, such as parental feeding practices, may be responsive to the child’s genotype (reactive gene-environment correlation). Models have been developed to take into account gene-environment correlation effects,54 but larger sample sizes are needed than that available in this study. There are also some limitations of the twin method, which may lead to overestimation of heritability estimates. The assumption of equal shared environments among DZ and MZ twins has been challenged by individuals who believe that MZ twins experience environments that are more similar than those experienced by DZ twins.55,56 There is also evidence that the prenatal environment may make MZ twins less similar than the twin method assumes.57 However, studying twins reared apart overcomes the equal environments assumption, and principal findings match those reported in twin modeling studies.58 Twins are less representative of the general population than singletons in several ways, including their growth59; however, there is no evidence that growth patterns differ between MZ and DZ twins, which would compromise findings from twin studies.

and principal findings match those reported in twin modeling studies.58 Twins are less representative of the general population than singletons in several ways, including their growth59; however, there is no evidence that growth patterns differ between MZ and DZ twins, which would compromise findings from twin studies. Although it is not clear whether or how gene-environment interaction would vary by race/ethnicity, some research suggests that heritability of BMI is higher among white adolescents than East Asian adolescents.60 It would therefore be informative to replicate our findings in an ethnically diverse sample. Finally, as in other cohort studies, heritability estimates were derived from parent reports of height and weight. However, research supports the validity of parent-reported BMI, especially when the measures are taken at home, as in this study.61 Conclusions This is the first study, to our knowledge, to examine whether the heritability of child BMI varies by the extent to which the early home environment is obesogenic. Heritability of BMI was higher in higher-risk home environments, which supports the theory that obesity-related genes are more strongly associated with BMI in more obesogenic environments and suggests pathways through which macro-level factors, such as socioeconomic status, are associated with obesity. These findings provide further insight into the mechanisms underlying overweight and obesity and how they may be prevented. Supplement. eTable 1. Constructs included in the home environment composite scores

Conclusions This is the first study, to our knowledge, to examine whether the heritability of child BMI varies by the extent to which the early home environment is obesogenic. Heritability of BMI was higher in higher-risk home environments, which supports the theory that obesity-related genes are more strongly associated with BMI in more obesogenic environments and suggests pathways through which macro-level factors, such as socioeconomic status, are associated with obesity. These findings provide further insight into the mechanisms underlying overweight and obesity and how they may be prevented. Supplement. eTable 1. Constructs included in the home environment composite scores eTable 2. Overview of the study measures and assessment points in Gemini eTable 3. Comparison of the study sample and the total HEI sample on the study variables eFigure 1. Heterogeneity model for higher-risk and lower-risk obesogenic home environments Click here for additional data file.

Introduction Atopic dermatitis ranks among the largest components of the nonfatal disease burden worldwide.1 Sleep disturbances have been identified as central to quality-of-life decrements in atopic dermatitis,2,3 but little is known about their association with sleep in the general population. Pruritus, a hallmark of atopic dermatitis, is often worst at night, resulting in scratching that may interfere with the process of falling asleep and cause disruptions in ongoing sleep.4,5 Small polysomnography and actigraphy studies among clinic-based populations have found that children with atopic dermatitis are more restless in their sleep, awaken more often, and spend more time awake after the onset of sleep.6,7,8,9,10 Adequate sleep is critical to well-being and health; in children, acute and chronic sleep disturbances have been associated with a wide range of cognitive, mood, and behavioral impairments and have been linked to poor educational performance.11,12,13 Atopic dermatitis has a chronic relapsing and remitting course, and it is unknown how variations in disease activity and severity affect sleep at different periods throughout childhood. Longitudinal studies can help characterize and quantify the burden of atopic dermatitis–associated sleep loss during these critical developmental periods. We aimed to determine whether children with active atopic dermatitis have impaired sleep duration and quality throughout childhood and whether the severity of atopic dermatitis affects sleep outcomes in a population-based birth cohort.

atopic dermatitis–associated sleep loss during these critical developmental periods. We aimed to determine whether children with active atopic dermatitis have impaired sleep duration and quality throughout childhood and whether the severity of atopic dermatitis affects sleep outcomes in a population-based birth cohort. Methods We performed a longitudinal cohort study using data collected from 1990 to 2008 from the Avon Longitudinal Study of Parents and Children (ALSPAC). Participants in ALSPAC provided written informed consent, and ethical approval was obtained from the ALSPAC Ethics and Law Committee and the Local Research Ethics Committees. The present study was considered exempt from review by the University of California, San Francisco Institutional Review Board because all of the data obtained by investigators were fully deidentified. Data analysis was performed from September 2017 to September 2018. Participants Pregnant women residing in Avon, United Kingdom, were recruited between 1990 and 1992 and followed up in the ensuing 2 decades with standardized questionnaires and clinical assessment visits, as described in detail elsewhere.14,15 The ALSPAC study enrolled a total of 14 541 pregnancies, which resulted in 14 062 live births, of which 13 988 were alive at 1 year of age. The current study sample is limited to children alive at 1 year of age and includes assessments through age 16 (n = 11 620; 83% of those alive at 1 year of age). The ALSPAC website contains details of all of the data available through a fully searchable data dictionary and variable search tool.16

live at 1 year of age. The current study sample is limited to children alive at 1 year of age and includes assessments through age 16 (n = 11 620; 83% of those alive at 1 year of age). The ALSPAC website contains details of all of the data available through a fully searchable data dictionary and variable search tool.16 Exposure The primary exposure was atopic dermatitis annual period prevalence, measured by a standardized question about flexural (joints and creases) dermatitis answered by the mother or the child (latest time point only) at 12 time points between age 6 months and 16 years: Has your child had an itchy, dry skin rash in the joints and creases of his body (eg, behind the knees, elbows, under the arms) in the past year? This question is comparable to that used in the large International Study of Asthma and Allergies in Childhood (ISAAC).17 Individuals were considered to have active atopic dermatitis if they had at least 2 reports of flexural dermatitis, up to and including the time point being considered.18,19,20 On the first report of flexural dermatitis, individuals were categorized as being indeterminate for atopic dermatitis and not included in the control group for that time point. Disease severity was assessed at each time point by a question that asked mothers to categorize their child’s disease over the past year as no problem, mild, quite bad, or very bad. Finally, children were classified as having inactive atopic dermatitis if they met the definition of active atopic dermatitis previously but responded negatively at the time point being considered.

ion that asked mothers to categorize their child’s disease over the past year as no problem, mild, quite bad, or very bad. Finally, children were classified as having inactive atopic dermatitis if they met the definition of active atopic dermatitis previously but responded negatively at the time point being considered. Sleep Outcomes Sleep duration was assessed by standardized questionnaires at 8 time points (30, 42, 57, 69, 81, 115, 140, and 186 months) between ages 2 and 16 years. Nighttime sleep duration was calculated on the basis of maternal or self-report (16 years only) of the time the child usually went to sleep and usually woke up in the morning. Mothers were also asked about daytime sleep duration at 5 time points between ages 2 and 7 years. Total sleep duration was calculated by adding nighttime and daytime duration up until age 7 years, and it was equal to nighttime sleep duration alone after age 7 years. Both nighttime and total sleep duration were found to be approximately normally distributed. Sleep quality was assessed using 4 standardized questions at 6 time points (30, 42, 57, 69, 81, and 115 months) between ages 2 and 10 years. Mothers were asked about nighttime awakenings (≥1 per night) and whether the child regularly experienced early morning awakenings, difficulty falling asleep, and nightmares over the past year. Responses were analyzed individually and combined into a composite sleep-quality outcome scored from 0 to 4, assigning 1 point for each item.

sked about nighttime awakenings (≥1 per night) and whether the child regularly experienced early morning awakenings, difficulty falling asleep, and nightmares over the past year. Responses were analyzed individually and combined into a composite sleep-quality outcome scored from 0 to 4, assigning 1 point for each item. Additional Covariates Potential confounders and effect modifiers were identified from the literature and incorporated into a directed acyclic graph that was used to guide the modeling strategy (Figure 1).22,23,24,25,26,27 These covariates included child and mother demographic characteristics (child sex, age, and race/ethnicity as well as maternal age at delivery), indicators of socioeconomic status, household smoking exposure, and comorbid asthma or allergic rhinitis. Socioeconomic status was measured using prenatal questionnaires collected from parents at study enrollment, including the highest educational qualification of the mother; social class based on occupation (highest of either parent); household crowding index (number of people living in the household divided by the number of rooms in the house); and a financial difficulties score assessing the mother’s self-reported difficulty to afford food, clothing, heating, rent or mortgage, and items necessary to care for her baby.

n (highest of either parent); household crowding index (number of people living in the household divided by the number of rooms in the house); and a financial difficulties score assessing the mother’s self-reported difficulty to afford food, clothing, heating, rent or mortgage, and items necessary to care for her baby. Figure 1. Directed Acyclic Graph A directed acyclic graph represents associations between covariates and primary exposure and outcome. Gray circles represent ancestors of the exposure and outcome (ie, confounders), blue circles represent ancestors of the outcome (ie, causal determinants of the outcome), and light blue circles represent unobserved (ie, latent) variables. Green lines represent causal paths, and gray lines represent biasing paths. The minimally sufficient adjustment set represents covariates such that the adjustment for this set of variables will minimize confounding bias when estimating the association between the exposure and the outcome. The minimally sufficient adjustment set was determined using the DAGitty software.21 Child comorbid asthma or allergic rhinitis was considered to be a collider, which was appropriately accounted for by adjusting for additional variables contained on the backdoor paths shared by this collider. The final minimally sufficient adjustment set comprised child sex, age, race/ethnicity, and comorbid asthma or allergic rhinitis; maternal age at delivery; socioeconomic status (SES); and household smoking exposure.

unted for by adjusting for additional variables contained on the backdoor paths shared by this collider. The final minimally sufficient adjustment set comprised child sex, age, race/ethnicity, and comorbid asthma or allergic rhinitis; maternal age at delivery; socioeconomic status (SES); and household smoking exposure. Time-varying covariates included comorbid asthma or allergic rhinitis and household smoking exposure. A child was determined to have comorbid asthma or allergic rhinitis at each time point examined if the mother reported asthma and/or allergic rhinitis symptoms at that time point, based on standardized questions similar to those used in the ISAAC study.17 Finally, models also included a measure of household smoking exposure, assessed by a maternal questionnaire about the number of smokers living in the household at multiple time points throughout childhood. Missing Data As has been described in detail elsewhere, the ALSPAC cohort has both intermittent missing data and attrition (ie, loss to follow-up).14 For example, the mean response rate to 12 surveys during the adolescence phase was 48%, but 75% of individuals responded at least once during adolescence. Multiple imputation was performed to impute missing exposure, outcome, and covariate data. Iterative chained equations were used, as most variables in the models did not follow a normal distribution. Fifty imputed data sets were generated and used to repeat primary analyses.

responded at least once during adolescence. Multiple imputation was performed to impute missing exposure, outcome, and covariate data. Iterative chained equations were used, as most variables in the models did not follow a normal distribution. Fifty imputed data sets were generated and used to repeat primary analyses. Statistical Analysis Cross-sectional regression analyses were performed to compare sleep outcomes between children with and children without atopic dermatitis at each time point (linear models for sleep duration, logistic models for binary sleep-quality outcomes, and ordered logistic regression models for the composite sleep-quality measure). Longitudinal analyses with repeated measures of the exposure, outcome, and time-varying covariates (asthma or allergic rhinitis and household smoking exposure) were then conducted using mixed-effects models with random slopes and intercepts for each individual. For all longitudinal analyses, table legends include the total number of individuals included in each model and the mean number of observations (ie, number of time points used) per individual. The minimally sufficient adjustment set was determined using a directed acyclic graph (Figure 1). We tested for interactions between atopic dermatitis and comorbid asthma or allergic rhinitis, child’s age, child sex, and maternal educational level. Of these variables, only the interaction with comorbid asthma or allergic rhinitis was found to be statistically significant for sleep-quality outcomes (P < .05); thus, these results are presented stratified by asthma or allergic rhinitis status. All statistical analyses were performed using Stata, version 14.2 (StataCorp Inc).

variables, only the interaction with comorbid asthma or allergic rhinitis was found to be statistically significant for sleep-quality outcomes (P < .05); thus, these results are presented stratified by asthma or allergic rhinitis status. All statistical analyses were performed using Stata, version 14.2 (StataCorp Inc). Results The study sample was composed of 13 988 children (7220 male [51.7%]) followed up for a median (interquartile range [IQR]) duration of 11 (5-14) years. Overall, 4938 children (35.3%) met the definition of atopic dermatitis between 2 and 16 years of age. Children with atopic dermatitis were more likely to be female, have comorbid asthma or allergic rhinitis, have a family history of atopic conditions, be from a family of a high social class, and have a mother with a high level of education (Table 1). The annual period prevalence of active atopic dermatitis ranged from 13% to 21% from age 2 to 16 years, and 22% to 40% of individuals with atopic dermatitis reported quite bad to very bad disease at any given time point (eTable 1 in the Supplement).

ial class, and have a mother with a high level of education (Table 1). The annual period prevalence of active atopic dermatitis ranged from 13% to 21% from age 2 to 16 years, and 22% to 40% of individuals with atopic dermatitis reported quite bad to very bad disease at any given time point (eTable 1 in the Supplement). Table 1. Cohort Characteristics Variable Total, No. No. (%) Without Atopic Dermatitis (n = 5555) With Atopic Dermatitis (n = 4938)a P Valueb Child, No. (%) Male sex 13 978 7220 (51.7) 3008 (54.2) 2377 (48.1) <.001 White race/ethnicity 12 077 11 468 (95.0) 4755 (95.5) 4490 (95.7) .65 Asthma everc 12 612 3237 (25.7) 1058 (19.1) 1789 (36.2) <.001 Allergic rhinitis everc 10 156 1375 (13.5) 339 (7.8) 913 (20.3) <.001 Asthma or allergic rhinitis everc 12 620 3919 (31.1) 1272 (22.9) 2188 (44.3) <.001 Asthma and allergic rhinitis everc 12 620 693 (5.5) 125 (2.3) 514 (10.4) <.001 Maternal age at delivery, No. (%), y <.001 ≤20 13 972 1004 (7.2) 375 (6.8) 212 (4.3) 21-34 13 972 11 585 (82.9) 4589 (82.6) 4196 (85.0) ≥35 13 972 1383 (9.9) 588 (10.6) 530 (10.7) History of atopic condition, No. (%)d Maternal 12 454 5659 (45.4) 2154 (42.0) 2402 (50.3) <.001 Paternal 8545 3418 (40.0) 1307 (36.8) 1543 (45.2) <.001 Household smoking exposure, No. (%)e 10 188 3683 (36.2) 1643 (37.7) 1476 (33.5) <.001 Maternal highest educational level, No. (%)f <.001 CSE/none 12 412 2504 (20.2) 1101 (21.5) 726 (15.2) Vocational 12 412 1224 (9.9) 510 (9.9) 446 (9.3) O level 12 412 4294 (34.6) 1833 (35.7) 1656 (34.7) A level 12 412 2791 (22.5) 1130 (22.0) 1182 (24.8) Degree 12 412 1599 (12.9) 559 (10.9) 762 (16.0) Social class, No. (%)g <.001 Unskilled 12 254 227 (1.9) 86 (1.7) 54 (1.2) Partly skilled 12 254 920 (7.5) 375 (7.5) 285 (6.2) Skilled manual 12 254 1666 (13.6) 712 (14.2) 516 (11.2) Skilled nonmanual 12 254 3780 (30.8) 1582 (31.6) 1390 (30.2) Managerial and technical 12 254 4566 (37.3) 1862 (37.2) 1846 (40.1) Professional 12 254 1095 (8.9) 391 (7.8) 508 (11.1) Financial difficulties quartile, No. (%) .08 Lowest 12 083 4337 (35.9) 1798 (35.9) 1781 (38.4) Mild 12 083 3006 (24.9) 1254 (25.1) 1142 (24.6) Moderate 12 083 2324 (19.2) 970 (19.4) 862 (18.6) Highest 12 083 2416 (20.0) 979 (19.6) 851 (18.4) Crowding index, No. (%)h <.001 <0.5 12 799 5329 (41.6) 2058 (39.9) 2231 (46.9) >0.5-0.75 12 799 4013 (31.3) 1623 (31.5) 1509 (31.7) >0.75-1 12 799 2579 (20.2) 1106 (21.5) 810 (17.0) >1 12 799 878 (6.9) 368 (7.1) 207 (4.4) Abbreviation: CSE, Certificate of Secondary Education.

st 12 083 2416 (20.0) 979 (19.6) 851 (18.4) Crowding index, No. (%)h <.001 <0.5 12 799 5329 (41.6) 2058 (39.9) 2231 (46.9) >0.5-0.75 12 799 4013 (31.3) 1623 (31.5) 1509 (31.7) >0.75-1 12 799 2579 (20.2) 1106 (21.5) 810 (17.0) >1 12 799 878 (6.9) 368 (7.1) 207 (4.4) Abbreviation: CSE, Certificate of Secondary Education. a Children who met the definition of atopic dermatitis by age 16 years (ie, had at least 2 reports of flexural dermatitis). There were 3505 individuals with only 1 report of flexural dermatitis. b χ2 Test comparing children who never reported atopic dermatitis with children who ever reported atopic dermatitis through age 16 years. c At least 2 reports of symptoms throughout childhood. d Including atopic dermatitis, asthma, or allergic rhinitis. e At 1.75 years of age. f UK educational levels: CSE, certificate after passing national school examinations at 16 years of age; vocational; O level, qualification after passing national school examinations at 16 years of age; higher than CSE; A level, qualification after passing national school examinations at 18 years of age; degree, university degree, or higher. g Highest of either parent, based on occupation. h The number of persons living in the household divided by the number of rooms in that household.

f UK educational levels: CSE, certificate after passing national school examinations at 16 years of age; vocational; O level, qualification after passing national school examinations at 16 years of age; higher than CSE; A level, qualification after passing national school examinations at 18 years of age; degree, university degree, or higher. g Highest of either parent, based on occupation. h The number of persons living in the household divided by the number of rooms in that household. Sleep Duration The mean (SD) nighttime sleep duration ranged from 11.2 (1.0) hours at 2 years of age to 8.7 (0.9) hours at 16 years of age. The median (IQR) daytime sleep duration was 30 (0-90) minutes at 2 years of age and 0 minutes thereafter. Overall, throughout childhood, nighttime sleep duration was similar between children with active atopic dermatitis and those without atopic dermatitis (Table 2). In adjusted models, the estimated difference was 0 minutes (95% CI, −2 to 2), and we did not find any statistically significant differences or gradient by atopic dermatitis severity levels. For total sleep duration (including daytime naps though age 7 years), we found a statistically significant but clinically negligible difference: individuals with active atopic dermatitis were estimated to sleep a mean 2 minutes less per day throughout childhood (95% CI, −4 to 0), and this association was similar across all disease severity levels (Table 2).

time naps though age 7 years), we found a statistically significant but clinically negligible difference: individuals with active atopic dermatitis were estimated to sleep a mean 2 minutes less per day throughout childhood (95% CI, −4 to 0), and this association was similar across all disease severity levels (Table 2). Table 2. Estimated Differences in Sleep Duration According to Atopic Dermatitis Disease Activity and Severitya Disease Activity and Severity Estimated Difference in Sleep Duration (95% CI), min Unadjusted Adjusted Nighttime sleep durationb,c Never reported atopic dermatitis 1 [Reference] 1 [Reference] Overall active atopic dermatitis −8 (−10 to −6) 0 (−2 to 2) No problem −9 (−13 to −6) 1 (−2 to 4) Mild −11 (−13 to −9) 1 (−1 to 3) Quite bad −7 (−10 to −5) 0 (−2 to 2) Very bad −5 (−10 to 0) −1 (−6 to 3) Inactive atopic dermatitis −33 (−35 to −32) 2 (1 to 4) Total sleep durationd,e Never reported atopic dermatitis 1 [Reference] 1 [Reference] Overall active atopic dermatitis −10 (−13 to −8) −2 (−4 to 0) No problem −15 (−19 to −10) −2 (−5 to 1) Mild −17 (−19 to −14) −2 (−4 to −1) Quite bad −8 (−11 to −5) −1 (−4 to 1) Very bad −3 (−8 to 2) −3 (−7 to 1) Inactive atopic dermatitis −43 (−45 to −42) −1 (−3 to 0) a Results from unadjusted and adjusted multivariable mixed-effects linear regression models examining the association between atopic dermatitis and nighttime and total sleep duration at 8 time points (30, 42, 57, 69, 81, 115, 140, and 186 months) between ages 2 and 16 years. The multivariable model adjusted for potential confounders, including child’s sex, age, race/ethnicity, and comorbid asthma or allergic rhinitis; household smoking exposure; maternal educational level, social class, and age at delivery; crowding index; and financial difficulties score.

ths) between ages 2 and 16 years. The multivariable model adjusted for potential confounders, including child’s sex, age, race/ethnicity, and comorbid asthma or allergic rhinitis; household smoking exposure; maternal educational level, social class, and age at delivery; crowding index; and financial difficulties score. b Unadjusted model with 11 549 individuals; mean (range) of 5.3 (1-8) observations per individual. c Adjusted model with 9109 individuals; mean (range) of 5.0 (1-7) observations per individual. d Unadjusted model with 11 531 individuals; mean (range) of 5.3 (1-8) observations per individual. e Adjusted model with 9101 individuals; mean (range) of 5.0 (1-7) observations per individual. Sleep Quality At any point between 2 and 10 years of age, the mean number of sleep disturbances ranged from 1.3 to 1.8, and 72% to 87% of the population experienced 1 or more sleep-quality disturbances. Overall, 5075 (50.0%) of 10 159 children reported regularly waking at least once in the night at age 2 years, which decreased to 1001 (13.5%) of 7435 children by age 10 years. A large proportion of all children reported regularly waking early in the morning (36.3% [2806 of 7739] to 58.2% [5930 of 10 195]), regularly having difficulty falling asleep (37.1% [3729 of 10 047] to 63.0% [5312 of 8434]), and regularly experiencing nightmares (26.2% [2667 of 10 166] to 49.5% [4138 of 8364]) at any given time point. The proportion of children with active atopic dermatitis experiencing all 4 sleep-quality disturbances according to child’s age is shown in Figure 2.

g asleep (37.1% [3729 of 10 047] to 63.0% [5312 of 8434]), and regularly experiencing nightmares (26.2% [2667 of 10 166] to 49.5% [4138 of 8364]) at any given time point. The proportion of children with active atopic dermatitis experiencing all 4 sleep-quality disturbances according to child’s age is shown in Figure 2. Figure 2. Proportion of Children With Active Atopic Dermatitis Experiencing Sleep-Quality Disturbances by Child Age Proportion of children with active atopic dermatitis reporting each of the 4 sleep-quality disturbances based on cross-sectional data at different child ages.

g asleep (37.1% [3729 of 10 047] to 63.0% [5312 of 8434]), and regularly experiencing nightmares (26.2% [2667 of 10 166] to 49.5% [4138 of 8364]) at any given time point. The proportion of children with active atopic dermatitis experiencing all 4 sleep-quality disturbances according to child’s age is shown in Figure 2. Figure 2. Proportion of Children With Active Atopic Dermatitis Experiencing Sleep-Quality Disturbances by Child Age Proportion of children with active atopic dermatitis reporting each of the 4 sleep-quality disturbances based on cross-sectional data at different child ages. In cross-sectional analyses, children with active atopic dermatitis were more likely to report worse sleep-quality outcomes at all ages (eTable 2 in the Supplement). We found evidence for statistically significant interaction between atopic dermatitis and comorbid asthma or allergic rhinitis for the occurrence of impaired sleep-quality outcomes (overall test for interaction P = .04; Table 3). Children with only active atopic dermatitis had nearly 50% higher odds of reporting more sleep-quality disturbances throughout childhood (adjusted odds ratio [aOR], 1.48; 95% CI, 1.33-1.66), compared with those who never reported atopic dermatitis. Children with both active atopic dermatitis and either asthma or allergic rhinitis had nearly 80% increased odds of reporting more sleep-quality disturbances throughout childhood (aOR, 1.79; 95% CI, 1.54-2.09). In comparison, those with only asthma or allergic rhinitis and no atopic dermatitis had about 40% greater odds of reporting more sleep-quality disturbances throughout childhood (aOR, 1.42; 95% CI, 1.26-1.60).

ncreased odds of reporting more sleep-quality disturbances throughout childhood (aOR, 1.79; 95% CI, 1.54-2.09). In comparison, those with only asthma or allergic rhinitis and no atopic dermatitis had about 40% greater odds of reporting more sleep-quality disturbances throughout childhood (aOR, 1.42; 95% CI, 1.26-1.60). Table 3. Adjusted Participant-Specific Odds of Reporting More Sleep-Quality Disturbances According to Atopic Dermatitis Disease Activity and Severitya,b Disease Activity and Severity Odds Ratio (95% CI) Without Asthma or Allergic Rhinitis With Asthma or Allergic Rhinitis Never reported atopic dermatitis 1 [Reference] 1 [Reference] Overall active atopic dermatitis 1.48 (1.33-1.66) 1.79 (1.54-2.09) No problem 1.24 (1.04-1.48) 1.52 (1.07-2.16) Mild 1.40 (1.27-1.54) 1.47 (1.26-1.71) Quite bad 1.51 (1.32-1.73) 2.03 (1.66-2.48) Very bad 1.85 (1.41-2.45) 2.28 (1.65-3.15) Inactive atopic dermatitis 1.41 (1.28-1.55) 1.52 (1.31-1.77) a Results from an adjusted multivariable mixed-effects ordinal logistic regression model examining the association between atopic dermatitis and a composite measure of sleep quality (including nighttime awakenings, early morning awakenings, difficulty going to sleep, and nightmares) at 6 time points (30, 42, 57, 69, 81, and 115 months) between ages 2 and 10 years. The model adjusted for potential confounders, including child’s sex, age, race/ethnicity, and comorbid asthma or allergic rhinitis; household smoking exposure; maternal educational level, social class, and age at delivery; crowding index; and financial difficulties score as well as an interaction term between atopic dermatitis and comorbid asthma or allergic rhinitis. Model with 9112 individuals; mean (range) of 4.5 (1-6) observations per individual.

usehold smoking exposure; maternal educational level, social class, and age at delivery; crowding index; and financial difficulties score as well as an interaction term between atopic dermatitis and comorbid asthma or allergic rhinitis. Model with 9112 individuals; mean (range) of 4.5 (1-6) observations per individual. b Overall test for interaction: P = .04. More severe disease was associated with worse sleep-quality outcomes, among children with and without comorbid asthma or allergic rhinitis (Table 3). For those with quite bad or very bad active disease, children with only active atopic dermatitis had nearly 1.7 times the odds of reporting more sleep-quality disturbances throughout childhood (aOR, 1.68; 95% CI, 1.42-1.98), and those with both active atopic dermatitis and comorbid asthma or allergic rhinitis had 2.15 times the odds of reporting more sleep-quality disturbances throughout childhood (aOR, 2.15; 95% CI, 1.75-2.64), compared with those who never reported atopic dermatitis. Children with inactive disease had similar odds of reporting more sleep-quality disturbances (OR, 1.41; 95% CI, 1.28-1.55) as those with active mild disease (OR, 1.40; 95% CI, 1.27-1.54), and both were statistically significantly higher than the reference group (Table 3). These results were largely consistent across individual sleep-quality outcomes (eTables 3-6 in the Supplement).

e sleep-quality disturbances (OR, 1.41; 95% CI, 1.28-1.55) as those with active mild disease (OR, 1.40; 95% CI, 1.27-1.54), and both were statistically significantly higher than the reference group (Table 3). These results were largely consistent across individual sleep-quality outcomes (eTables 3-6 in the Supplement). Multiple Imputation Results Primary analyses yielded results that were largely consistent with those estimated from the imputed data (eTables 3-8 in the Supplement). For sleep quality, estimates using the imputed data were slightly attenuated toward the null; however, results remained qualitatively similar and statistically significant. For sleep duration, the results were nearly identical. Discussion Among the 13 988 children from the ALSPAC cohort followed up from birth through adolescence, we found similar sleep duration between children with active atopic dermatitis and those without. In contrast, children with active atopic dermatitis experienced worse sleep quality throughout childhood. This association was largest among children with more severe disease and among children with comorbid asthma or allergic rhinitis, but it remained statistically significant even for those with inactive and mild disease.

ntrast, children with active atopic dermatitis experienced worse sleep quality throughout childhood. This association was largest among children with more severe disease and among children with comorbid asthma or allergic rhinitis, but it remained statistically significant even for those with inactive and mild disease. These findings are consistent with those of small cross-sectional studies of clinic populations that used objective measures of sleep, including actigraphy and polysomnography. In those studies, despite increases in sleep fragmentation and reductions in sleep efficiency, overall sleep duration was similar between children with and without atopic dermatitis.6,7,8,9,10 In contrast, time spent awake after sleep onset is consistently greater among children with atopic dermatitis, ranging from approximately 45 to 100 minutes.6,7,8,9 In addition to increased nighttime awakenings and difficulty falling asleep, we found that children with active atopic dermatitis were more likely to report nightmares and early morning awakenings, which has not been previously studied.10

dren with atopic dermatitis, ranging from approximately 45 to 100 minutes.6,7,8,9 In addition to increased nighttime awakenings and difficulty falling asleep, we found that children with active atopic dermatitis were more likely to report nightmares and early morning awakenings, which has not been previously studied.10 A strength of this longitudinal study was that it enabled us to examine the association between sleep and atopic dermatitis activity and severity at multiple time points throughout childhood, allowing a look at sleep outcomes of individuals whose disease was no longer active at any given time point. Children with inactive disease still reported increased odds of impaired sleep quality, at a level similar to those with active but mild disease. Our findings are consistent with previous data from a polysomnographic study: Despite being in a period of clinical remission, children with a history of previously active atopic dermatitis experienced a substantially higher number of arousals and awakenings compared with controls.28 Moreover, scratching episodes only accounted for 15% of the arousals and awakenings, suggesting that scratching alone does not explain the sleep fragmentation experienced by these patients.28 Fishbein and colleagues5 have proposed that this phenomenon may be associated with a heightened sensitivity to sensory stimulation at night secondary to skin damage, which may represent an underlying mechanism of hyperarousability despite good disease control.6,29 Other factors that may be implicated in atopic dermatitis–associated sleep disturbances include cytokine and melatonin dysregulation and disrupted circadian rhythms of the skin.30

lation at night secondary to skin damage, which may represent an underlying mechanism of hyperarousability despite good disease control.6,29 Other factors that may be implicated in atopic dermatitis–associated sleep disturbances include cytokine and melatonin dysregulation and disrupted circadian rhythms of the skin.30 From a clinical perspective, our findings suggest that pediatricians should consider screening all children with atopic dermatitis, even if their disease is mild or no longer active. Clinicians may offer anticipatory guidance, education, behavioral interventions, and referrals if appropriate.31,32,33 Early detection and management of sleep problems in children with atopic dermatitis is critical to prevent quality-of-life impairments, daytime fatigue, as well as behavioral and mood disorders reported in children with atopic dermatitis.2,9,10,34,35 Additional research is needed to understand whether more aggressive treatment of atopic dermatitis symptoms will lead to improvements in children’s sleep quality.

al to prevent quality-of-life impairments, daytime fatigue, as well as behavioral and mood disorders reported in children with atopic dermatitis.2,9,10,34,35 Additional research is needed to understand whether more aggressive treatment of atopic dermatitis symptoms will lead to improvements in children’s sleep quality. Sleep disturbances have been studied separately in children with asthma and allergic rhinitis, but few studies have examined differences in sleep quality in children with more than 1 atopic condition.10,36,37 One of the strengths of the current study is the inclusion of several time-varying covariates, notably, asthma and allergic rhinitis, which were measured at all of the same time points as the primary exposure. We found that children with both atopic dermatitis and comorbid asthma or allergic rhinitis reported substantially more sleep-quality impairments, which has important clinical and therapeutic implications. This result suggests that children with several atopic diseases may represent a group at higher risk of experiencing disrupted sleep and its consequences, including impaired quality of life, daytime fatigue, poor school performance, and behavioral problems. Clinicians caring for children with several atopic conditions should inquire about nocturnal symptoms and sleep disturbances during routine clinic visits and should consider treating these conditions more aggressively.

ncluding impaired quality of life, daytime fatigue, poor school performance, and behavioral problems. Clinicians caring for children with several atopic conditions should inquire about nocturnal symptoms and sleep disturbances during routine clinic visits and should consider treating these conditions more aggressively. Limitations This study has several limitations that warrant discussion. As in all large-scale longitudinal studies, the study was missing data and attrition occurred over time. For this reason, we repeated the analyses after conducting multiple imputation and found that the results were similar, which helped address concerns of potential selection bias.

limitations that warrant discussion. As in all large-scale longitudinal studies, the study was missing data and attrition occurred over time. For this reason, we repeated the analyses after conducting multiple imputation and found that the results were similar, which helped address concerns of potential selection bias. Another important limitation was the possibility for misclassification bias, because both exposure and outcomes were parent- or self-reported. Previous studies have found that parental report closely approximates physician assessment of atopic dermatitis,38 and the estimates of the annual period prevalence of atopic dermatitis were consistent with UK estimates from the population-based ISAAC study, which included physical assessment in childhood.39 Our findings for both sleep duration and quality were highly consistent with smaller studies that used objective measures of sleep. Comparisons between parental assessment of children’s sleep and objective measures, including polysomnography and actigraphy, have found that parents tend to overestimate sleep duration and underestimate nighttime awakenings,40,41,42 both of which would tend to bias our results toward the null. Although our composite measure for sleep quality has not been validated, it included items that are similar to those in the Children’s Sleep Habits Questionnaire, a validated and reliable screening instrument to identify sleep problems in school-aged children.43 In addition, although our cohort is fairly representative of the UK population,14 the extent to which our results are generalizable to other settings is unclear.

lar to those in the Children’s Sleep Habits Questionnaire, a validated and reliable screening instrument to identify sleep problems in school-aged children.43 In addition, although our cohort is fairly representative of the UK population,14 the extent to which our results are generalizable to other settings is unclear. Notwithstanding these limitations, this study has several implications for future research and clinical care. Currently, only a few atopic dermatitis clinical outcome measures address sleep and may not adequately capture the extent of sleep-quality disturbances.44,45 Our findings support the development of standardized and validated clinical outcome measures of sleep disturbance that explicitly address several aspects of sleep quality.45,46 This refinement would enable future trials to assess the effectiveness of atopic dermatitis interventions in reducing poor sleep. Conclusions Atopic dermatitis appeared to negatively affect sleep quality throughout childhood, even among patients with mild and inactive disease. Increasing disease severity and comorbid asthma or allergic rhinitis appeared to be associated with worse sleep-quality outcomes. Clinical outcome measures for atopic dermatitis should explicitly address sleep quality, and additional work should investigate interventions to improve sleep quality and examine the association between atopic dermatitis treatment and children’s sleep. Supplement. eTable 1. Atopic Dermatitis Disease Activity and Severity by Child Age

Conclusions Atopic dermatitis appeared to negatively affect sleep quality throughout childhood, even among patients with mild and inactive disease. Increasing disease severity and comorbid asthma or allergic rhinitis appeared to be associated with worse sleep-quality outcomes. Clinical outcome measures for atopic dermatitis should explicitly address sleep quality, and additional work should investigate interventions to improve sleep quality and examine the association between atopic dermatitis treatment and children’s sleep. Supplement. eTable 1. Atopic Dermatitis Disease Activity and Severity by Child Age eTable 2. Unadjusted Odds of Sleep Disturbances for Children with Active Atopic Dermatitis by Child Age Compared to Children Who Never Reported Atopic Dermatitis eTable 3. Subject-specific Odds of Waking at Least Once per Night According to Atopic Dermatitis Disease Activity and Severity Stratified by Comorbid Asthma or Allergic Rhinitis Compared to Children Who Never Reported Atopic Dermatitis eTable 4. Subject-specific Odds of Having Difficulty Falling Asleep According to Atopic Dermatitis Disease Activity and Severity Stratified by Comorbid Asthma or Allergic Rhinitis Compared to Children Who Never Reported Atopic Dermatitis eTable 5. Subject-specific Odds of Early Morning Awakening According to Atopic Dermatitis Disease Activity and Severity Stratified by Comorbid Asthma or Allergic Rhinitis Compared to Children Who Never Reported Atopic Dermatitis

eTable 4. Subject-specific Odds of Having Difficulty Falling Asleep According to Atopic Dermatitis Disease Activity and Severity Stratified by Comorbid Asthma or Allergic Rhinitis Compared to Children Who Never Reported Atopic Dermatitis eTable 5. Subject-specific Odds of Early Morning Awakening According to Atopic Dermatitis Disease Activity and Severity Stratified by Comorbid Asthma or Allergic Rhinitis Compared to Children Who Never Reported Atopic Dermatitis eTable 6. Subject-specific Odds of Nightmares According to Atopic Dermatitis Disease Activity and Severity Stratified by Comorbid Asthma or Allergic Rhinitis Compared to Children Who Never Reported Atopic Dermatitis eTable 7. Estimated Differences in Nighttime and Total Sleep Duration According to Atopic Dermatitis Disease Activity and Severity Compared to Children Who Never Reported Atopic Dermatitis Using Imputed Data eTable 8. Subject-specific Odds of Individual Sleep Quality Disturbances According to Atopic Dermatitis Disease Activity and Severity Compared to Children Who Never Reported Atopic Dermatitis Using Imputed Data Click here for additional data file.

Introduction First sexual intercourse marks an important transition in an individual’s life,1 and early adolescence is a critical developmental period when experimentation with sexual feelings and behaviors often begins.2,3 During these formative years, expectations to adhere to gender roles and norms intensify. The Centers for Disease Control and Prevention’s Youth Risk Behavior Surveillance System (YRBSS) tracks sexual intercourse before age 13 years as a core surveillance metric and finds that males are more than twice as likely as females to experience first sexual intercourse before age 13 years.4 Sex education guidelines recommend providing children with comprehensive sex education starting at least by kindergarten, and clinical care guidelines recommend clinicians set time alone with young patients to address confidential care inclusive of sexual health starting during early adolescence.5,6 However, most males start having sex before receiving sex education, and the quality of sexual health care delivery to male adolescents is poor.7,8,9

ical care guidelines recommend clinicians set time alone with young patients to address confidential care inclusive of sexual health starting during early adolescence.5,6 However, most males start having sex before receiving sex education, and the quality of sexual health care delivery to male adolescents is poor.7,8,9 Estimates of young males’ transition to first sexual intercourse do not examine the prevalence of sexual activity in early adolescence across the intersecting demographics of sex, race/ethnicity, and location, likely missing important variations.10,11,12 A nationally representative study of sexual behavior reports it as “rare” among those 12 years and younger.13 Yet this conclusion may miss subgroups of males for whom sexual initiation before age 13 years is more common. Given the higher prevalence of first sexual intercourse before age 13 years among males compared with females, understanding the variation in the timing of sexual onset among adolescent males in the United States is critical to supporting their healthy sexual development.

initiation before age 13 years is more common. Given the higher prevalence of first sexual intercourse before age 13 years among males compared with females, understanding the variation in the timing of sexual onset among adolescent males in the United States is critical to supporting their healthy sexual development. Males’ experiences with regard to emergent manhood and sexuality are shaped by dimensions of masculinity, race/ethnicity, socioeconomic status, and location. Broad cultural scripts about masculinity and sex hold that men should start having sex early and have sex often.14 For young men of color, particularly black males, racist stereotypes of hypermasculinity may also contribute to expectations of early sexual initiation.15,16,17 Yet research highlights that males in early and middle adolescence do not necessarily follow such scripts and a later transition to first sexual intercourse may be valued.18,19,20,21,22 Understanding males’ wantedness of the sexual experience may be particularly important for interpreting early sexual activity.23,24 Aspects of adolescents’ communities may also be factors in the transition to first sexual intercourse.25,26,27,28 Cultural norms and values associated with masculinity may differ across communities.29 Some studies have found differences in the timing of first sexual intercourse between urban and rural settings.30,31 Even across specific urban areas, young men’s experiences may vary.

ansition to first sexual intercourse.25,26,27,28 Cultural norms and values associated with masculinity may differ across communities.29 Some studies have found differences in the timing of first sexual intercourse between urban and rural settings.30,31 Even across specific urban areas, young men’s experiences may vary. The current study examined the prevalence of sexual initiation before age 13 years among adolescent males in the United States and the variation in the timing of their sexual initiation by race/ethnicity, location, and maternal educational level and by their characterization of the wantedness of this first sexual experience. Race/ethnicity, socioeconomic status, and location are not the only factors in the timing of first sexual intercourse, but they inform the context in which these experiences and other correlates occur. We used 2 complementary large-scale representative survey systems to assess the timing of sexual onset among adolescent males in the United States, examine key sociodemographic correlates, and consider how reporting issues may affect estimates. Methods This cross-sectional analysis received approval from the Guttmacher Institute Institutional Review Board, which exempted secondary data analyses that used the YRBSS and National Survey of Family Growth (NSFG) data. The study was conducted from September 2017 to June 2018.

The current study examined the prevalence of sexual initiation before age 13 years among adolescent males in the United States and the variation in the timing of their sexual initiation by race/ethnicity, location, and maternal educational level and by their characterization of the wantedness of this first sexual experience. Race/ethnicity, socioeconomic status, and location are not the only factors in the timing of first sexual intercourse, but they inform the context in which these experiences and other correlates occur. We used 2 complementary large-scale representative survey systems to assess the timing of sexual onset among adolescent males in the United States, examine key sociodemographic correlates, and consider how reporting issues may affect estimates. Methods This cross-sectional analysis received approval from the Guttmacher Institute Institutional Review Board, which exempted secondary data analyses that used the YRBSS and National Survey of Family Growth (NSFG) data. The study was conducted from September 2017 to June 2018. Data Sources We used data from 2 cross-sectional US surveys to investigate the timing of age at first sexual intercourse among males: the YRBSS and the NSFG. The YRBSS, conducted by the Centers for Disease Control and Prevention, surveys middle school and high school students in classrooms using a paper-and-pencil, self-administered questionnaire.32 The national YRBSS high school sample is representative of public and private school students in grades 9 to 12. We pooled the 2011, 2013, and 2015 surveys for a total sample size of 19 916 male high school students, after excluding students with missing information on sex (<1%), age at first sexual intercourse (9%), or race/ethnicity (2%). Sample characteristics of the YRBSS and NSFG surveys are shown in Table 1.

grades 9 to 12. We pooled the 2011, 2013, and 2015 surveys for a total sample size of 19 916 male high school students, after excluding students with missing information on sex (<1%), age at first sexual intercourse (9%), or race/ethnicity (2%). Sample characteristics of the YRBSS and NSFG surveys are shown in Table 1. Table 1. Sample Characteristics of All Male Respondentsa Variable No. (Weighted %) YRBSS NSFG All respondents 19 916 (100) 7739 (100) Age <15 1950 (10.0) NA 15-17 14 647 (74.1) 2685 (30.5) 18-19b 3304 (15.9) 1780 (19.3) 20-24 NA 3274 (50.2) Race/ethnicity Non-Hispanic black 3148 (13.1) 1523 (15.0) Non-Hispanic white 8789 (57.1) 3737 (58.0) Hispanic 5914 (20.7) 1972 (20.1) Non-Hispanic other 2065 (9.2) 507 (6.8) Maternal educational level <College UA 5793 (72.2) ≥College degree UA 1897 (27.8) Community type Urban UA 3051 (34.8) Suburban UA 3529 (47.1) Rural UA 1159 (18.1) Survey year: NSFG 2006-2010 NA 4111 (50.7) 2011-2015 NA 3628 (49.3) Survey year: YRBSS 2011 6910 (35.4) NA 2013 6326 (30.5) NA 2015 6680 (34.1) NA Abbreviations: NA, not applicable; NSFG, National Survey of Family Growth; UA, unavailable; YRBSS, Youth Risk Behavior Surveillance System. a Data were pooled from the 2011, 2013, and 2015 YRBSS surveys and from the male respondents aged 15 to 24 years; data were pooled from the 2006 to 2010 and 2011 to 2015 continuous survey rounds of the NSFG. b The national YRBSS has only 1 top age category (18 years or older). Thus, this category potentially includes a small number of male high school students aged 20 years or older.

a Data were pooled from the 2011, 2013, and 2015 YRBSS surveys and from the male respondents aged 15 to 24 years; data were pooled from the 2006 to 2010 and 2011 to 2015 continuous survey rounds of the NSFG. b The national YRBSS has only 1 top age category (18 years or older). Thus, this category potentially includes a small number of male high school students aged 20 years or older. To examine the variation across geographic settings, we used representative YRBSS data from 8 middle school and 17 high school metropolitan areas. We used all publicly available YRBSS metropolitan area surveys and obtained permission to use some nonpublic YRBSS data from additional areas that had weighted data for at least 2 of the 3 survey years. In these high school samples, missingness on age at first sexual intercourse ranged from 8% to 26%. No national middle school YRBSS is conducted, but we included available metropolitan-area middle school data; these data were collected from younger students, potentially limiting recall bias in reporting age at first sexual intercourse.

ool samples, missingness on age at first sexual intercourse ranged from 8% to 26%. No national middle school YRBSS is conducted, but we included available metropolitan-area middle school data; these data were collected from younger students, potentially limiting recall bias in reporting age at first sexual intercourse. The NSFG is a periodic national probability survey of the noninstitutionalized population of females and males aged 15 to 44 years in the United States.33,34,35 To increase our sample size while minimizing the length of retrospective recall, we examined pooled data from the 2006 to 2010 and 2011 to 2015 continuous survey rounds and limited our analyses to 7739 males aged 15 to 24 years at the time of interview. The NSFG is administered through face-to-face interviews, and more sensitive questions are asked through the ACASI (audio computer-assisted self-interview) method. Measures Age at First Sexual Intercourse The YRBSS and NSFG surveys asked the age at which respondents first had sexual intercourse. The YRBSS asked age at first sexual intercourse without specifying the sex of the respondent’s partner, whereas the NSFG asked the timing of first heterosexual sexual intercourse. For both surveys, the answer to had sexual intercourse before age 13 years was dichotomized, with 0 indicating no and 1 indicating yes; respondents who had not had sex by the time of interview were coded as 0’s.

sex of the respondent’s partner, whereas the NSFG asked the timing of first heterosexual sexual intercourse. For both surveys, the answer to had sexual intercourse before age 13 years was dichotomized, with 0 indicating no and 1 indicating yes; respondents who had not had sex by the time of interview were coded as 0’s. Wantedness of First Sexual Intercourse The NSFG asked this question of all sexually experienced male respondents aged 18 years or older: Think back to the very first time you had vaginal intercourse with a female. Which would you say comes closest to describing how much you wanted that first vaginal intercourse to happen? Response choices were as follows: “I really didn’t want it to happen at the time,” “I had mixed feelings—part of me wanted it to happen at the time and part of me didn’t,” and “I really wanted it to happen at the time.”

ld you say comes closest to describing how much you wanted that first vaginal intercourse to happen? Response choices were as follows: “I really didn’t want it to happen at the time,” “I had mixed feelings—part of me wanted it to happen at the time and part of me didn’t,” and “I really wanted it to happen at the time.” Demographic and Survey Characteristics The surveys included comparable measures of race/ethnicity (non-Hispanic white, non-Hispanic black, non-Hispanic other, and Hispanic). Non-Hispanic other comprised Asian, American Indian, Alaskan Native, and Hawaiian and Pacific Islander males, but small sample sizes did not allow for separate analyses of each group. The national YRBSS did not include other sociodemographic covariates or details on location. From the NSFG data, we examined maternal educational level (less than college, or completed college or more) as a proxy for socioeconomic status, and community type (urban, suburban, or rural). We also constructed indicators for survey wave. To identify possible retrospective reporting bias, we used respondents’ integer age at interview (NSFG) and grade in school (YRBSS).

evel (less than college, or completed college or more) as a proxy for socioeconomic status, and community type (urban, suburban, or rural). We also constructed indicators for survey wave. To identify possible retrospective reporting bias, we used respondents’ integer age at interview (NSFG) and grade in school (YRBSS). Statistical Analysis We estimated the percentage of male high school students (national YRBSS) and males aged 15 to 24 years (NSFG) reporting sexual onset before age 13 years, testing for within-survey differences by race/ethnicity using unadjusted logistic regression. We examined differentials by metropolitan area using the YRBSS metropolitan samples, and we tested for differences by race/ethnicity within each site, controlling for grade and survey year to adjust for differences in sample composition. To account for censoring in the middle school data, given that some respondents were not yet 13 years of age, we estimated proportions from a Cox proportional hazards regression model (adjusted for grade and survey year); significance tests are for differences in site-specific relative hazards of early sexual onset between race/ethnicity groups.

the middle school data, given that some respondents were not yet 13 years of age, we estimated proportions from a Cox proportional hazards regression model (adjusted for grade and survey year); significance tests are for differences in site-specific relative hazards of early sexual onset between race/ethnicity groups. Using NSFG data, we estimated multivariable logistic regression models of sexual onset before age 13 years as a function of race/ethnicity, maternal educational level, community type, survey year, and age at interview. Then, we estimated probabilities of sexual onset before age 13 years by race/ethnicity and maternal educational level, after adding an interaction term between those 2 variables to the model. We also examined differences by age at first sexual intercourse in wantedness of first sexual intercourse (among respondents aged 18-24 years in the NSFG). To further compare differences in estimates between the YRBSS and NSFG, we conducted sensitivity analyses exploring alternate specifications in the NSFG, including using ACASI measures of age at first sexual intercourse, limiting the sample to those currently aged 15 to 19 years and in school and dropping all imputed data on age at first sexual intercourse. To test if differences persisted across age, we produced Kaplan-Meier life table estimates of age at first sexual intercourse for both data sets.

of age at first sexual intercourse, limiting the sample to those currently aged 15 to 19 years and in school and dropping all imputed data on age at first sexual intercourse. To test if differences persisted across age, we produced Kaplan-Meier life table estimates of age at first sexual intercourse for both data sets. All analyses were weighted. All P values were calculated using 2-tailed Wald tests, with SEs adjusted for the complex survey design of each data source using Stata, version 15.1 (StataCorp).36 Statistical significance was determined using α = .05. Results In total, this study analyzed 19 916 male high school students from the YRBSS data set and 7739 males aged 15 to 24 years from the NSFG data set. The sample was largely composed of non-Hispanic white males: 8789 (57.1%) from the YRBSS and 3737 (58.0%) from the NSFG. Prevalence of Early Sexual Onset by Race/Ethnicity The YRBSS and NSFG national data show different levels of sexual activity before age 13 years, but similar differentials by race/ethnicity. Sexual onset before age 13 years was reported by 7.6% (95% CI, 6.8%-8.4%) of male students in grades 9 to 12 in the YRBSS and 3.6% (95% CI, 3.0%-4.2%) of males aged 15 to 24 years in the NSFG (Table 2; results by survey year are available in eTable 1 in the Supplement). In both surveys, reports of early first sexual intercourse were statistically significantly higher among non-Hispanic black males than other racial/ethnic groups.

he YRBSS and 3.6% (95% CI, 3.0%-4.2%) of males aged 15 to 24 years in the NSFG (Table 2; results by survey year are available in eTable 1 in the Supplement). In both surveys, reports of early first sexual intercourse were statistically significantly higher among non-Hispanic black males than other racial/ethnic groups. Table 2. Weighted Percentage of Male Students Reporting Their First Sexual Intercourse Before Age 13 Years, by Race/Ethnicitya Race/Ethnicity YRBSS P Value NSFG P Value Total Unweighted Sample Size % (95% CI) Total Unweighted Sample Size % (95% CI) Total 19 916 7.6 (6.8-8.4) NA 7739 3.6 (3.0-4.2) NA Non-Hispanic black [reference] 3148 19.0 (16.7-21.3) NA 1523 10.5 (8.4-12.5) NA Non-Hispanic white 8789 4.4 (3.8-5.0) <.001 3737 2.2 (1.5-2.9) <.001 Hispanic 5914 9.0 (7.8-10.1) <.001 1972 3.4 (2.3-4.5) <.001 Non-Hispanic other 2065 7.8 (5.8-9.7) <.001 507 1.6 (0.3-2.9) <.001 Abbreviations: NA, not applicable; NSFG, National Survey of Family Growth; YRBSS, Youth Risk Behavior Surveillance System. a Data were pooled from the 2011, 2013, and 2015 YRBSS surveys and from the male respondents aged 15 to 24 years; data were pooled from the 2006 to 2010 and 2011 to 2015 continuous survey rounds of the NSFG.

Table 2. Weighted Percentage of Male Students Reporting Their First Sexual Intercourse Before Age 13 Years, by Race/Ethnicitya Race/Ethnicity YRBSS P Value NSFG P Value Total Unweighted Sample Size % (95% CI) Total Unweighted Sample Size % (95% CI) Total 19 916 7.6 (6.8-8.4) NA 7739 3.6 (3.0-4.2) NA Non-Hispanic black [reference] 3148 19.0 (16.7-21.3) NA 1523 10.5 (8.4-12.5) NA Non-Hispanic white 8789 4.4 (3.8-5.0) <.001 3737 2.2 (1.5-2.9) <.001 Hispanic 5914 9.0 (7.8-10.1) <.001 1972 3.4 (2.3-4.5) <.001 Non-Hispanic other 2065 7.8 (5.8-9.7) <.001 507 1.6 (0.3-2.9) <.001 Abbreviations: NA, not applicable; NSFG, National Survey of Family Growth; YRBSS, Youth Risk Behavior Surveillance System. a Data were pooled from the 2011, 2013, and 2015 YRBSS surveys and from the male respondents aged 15 to 24 years; data were pooled from the 2006 to 2010 and 2011 to 2015 continuous survey rounds of the NSFG. Prevalence of Early Sexual Onset by Metropolitan Area and Race/Ethnicity Across the 15 metropolitan sites with available YRBSS high school data, the proportion of male students who reported having their first sexual intercourse before age 13 years varied widely, from 5% (95% CI, 4%-7%) in San Francisco, California, to 25% (95% CI, 23%-28%) in Memphis, Tennessee, after controlling for grade or year of survey (Table 3). In most areas, higher proportions of non-Hispanic black male students reported having sexual intercourse before age 13 years compared with those in other racial/ethnic groups. Prevalence by race/ethnicity varied across metropolitan areas, ranging from 12% (95% CI, 7%-17%) in Seattle, Washington, to 28% (95% CI, 25%-31%) in Memphis among non-Hispanic black males; from 6% (95% CI, 4%-7%) in Los Angeles, California, to 17% in Seattle (95% CI, 10%-24%) and in Memphis (95% CI, 8%-27%) among Hispanic males; from 2% (95% CI, 0%-3%) in Charlotte-Mecklenburg, North Carolina, to 10% (95% CI, 3%-16%) in Chicago, Illinois, among non-Hispanic white males; and from 2% (95% CI, 1%-3%) in San Francisco to 17% (95% CI, 8%-27%) in Chicago among male students in the non-Hispanic other race/ethnicity category.

7%) among Hispanic males; from 2% (95% CI, 0%-3%) in Charlotte-Mecklenburg, North Carolina, to 10% (95% CI, 3%-16%) in Chicago, Illinois, among non-Hispanic white males; and from 2% (95% CI, 1%-3%) in San Francisco to 17% (95% CI, 8%-27%) in Chicago among male students in the non-Hispanic other race/ethnicity category. Table 3. Estimated Proportion of Male High School Students Reporting Their First Sexual Intercourse Before Age 13 Years, by Race/Ethnicity and Location, Adjusted for Grade and Survey Yeara Location Total Unweighted Sample Size % (95% CI) P Value Hispanic, % (95% CI) P Value Non-Hispanic Other, % (95% CI) P Value Total Non-Hispanic Black Non-Hispanic White Memphis, TN 1089 25 (23-28) 28 (25-31) 3 (0-6) <.001 17 (8-27) .08 16 (7-26) .06 Milwaukee, WI 1076 19 (16-22) 25 (20-29) 5 (0-11) .002 15 (10-20) .006 11 (6-17) .003 Chicago, IL 1243 18 (15-21) 29 (23-35) 10 (3-16) .002 11 (8-14) <.001 17 (8-27) .01 Boston, MA 1507 14 (12-17) 17 (12-22) 9 (4-13) .03 16 (12-19) .67 8 (4-12) .002 Duval County, FL 3770 13 (11-14) 20 (18-23) 6 (4-7) <.001 15 (11-18) .03 10 (7-12) <.001 Houston, TX 2507 12 (11-14) 22 (18-27) 6 (2-10) <.001 10 (8-11) <.001 5 (2-9) <.001 Miami-Dade County, FL 3146 11 (10-13) 20 (17-23) 6 (2-11) <.001 9 (8-11) <.001 15 (7-23) .35 Charlotte-Mecklenburg, NC 1293 11 (8-13) 20 (15-24) 2 (0-3) <.001 10 (7-14) <.001 11 (6-16) .01 Palm Beach County, FL 1650 10 (9-12) 18 (14-22) 5 (3-7) <.001 12 (9-16) .03 12 (7-17) .08 Broward County, FL 1840 10 (9-12) 19 (15-22) 3 (1-4) <.001 9 (7-11) <.001 8 (3-13) .02 Orange County, FL 1986 10 (9-12) 20 (15-25) 4 (2-6) <.001 10 (8-13) <.001 8 (4-12) <.001 New York City, NY 10 228 9 (8-10) 15 (12-17) 3 (2-4) <.001 11 (10-12) .005 3 (2-3) <.001 San Bernardino, CA 1204 9 (7-11) 14 (9-20) 9 (3-14) .14 9 (7-11) .03 4 (0-8) .01 San Diego, CA 2381 7 (6-9) 13 (8-18) 5 (3-7) .002 9 (7-11) .08 4 (2-6) <.001 Los Angeles, CA 1393 7 (5-8) 14 (3-25) 5 (1-9) .06 6 (4-7) .03 4 (2-7) .02 Seattle, WA 1441 6 (5-8) 12 (7-17) 3 (2-5) <.001 17 (10-24) .23 3 (1-4) <.001 San Francisco, CA 2637 5 (4-7) 27 (17-37) 3 (0-6) <.001 10 (8-13) <.001 2 (1-3) <.001 a Data were pooled from the 2011, 2013, and 2015 Youth Risk Behavior Surveillance System surveys. To be included in the sample, sites needed to have data available for at least 2 of the 3 survey years.

001 17 (10-24) .23 3 (1-4) <.001 San Francisco, CA 2637 5 (4-7) 27 (17-37) 3 (0-6) <.001 10 (8-13) <.001 2 (1-3) <.001 a Data were pooled from the 2011, 2013, and 2015 Youth Risk Behavior Surveillance System surveys. To be included in the sample, sites needed to have data available for at least 2 of the 3 survey years. Proportions were estimated from a logistic regression model controlling for grade and survey year; race/ethnicity-specific proportions were estimated from models with race/ethnicity by site interaction terms included as covariates. All significance tests used non-Hispanic black as a reference group. In the YRBSS middle school data (eTable 2 in the Supplement), reporting of having sexual intercourse before age 13 years had similar differentials by race/ethnicity as the high school reports. However, in almost all metropolitan areas, the proportion of students who reported having intercourse before age 13 years was higher in the middle school sample compared with the corresponding high school sample.

ual intercourse before age 13 years had similar differentials by race/ethnicity as the high school reports. However, in almost all metropolitan areas, the proportion of students who reported having intercourse before age 13 years was higher in the middle school sample compared with the corresponding high school sample. Multivariable Models of Early Sexual Onset Analyses of the NSFG data revealed that non-Hispanic white males (odds ratio [OR], 0.21; 95% CI, 0.13-0.32), Hispanic males (OR, 0.27; 95% CI, 0.18-0.4), and non-Hispanic other males (OR, 0.16; 95% CI, 0.06-0.38) were statistically significantly less likely to report having their first sexual intercourse before age 13 years than non-Hispanic black males, after adjusting for sociodemographic variables (Table 4). Respondents whose mothers had a college degree or higher educational level were statistically significantly less likely (OR, 0.31; 95% CI, 0.19-0.49) to report having sexual intercourse before age 13 years compared with those whose mothers did not have a college degree. No statistically significant differences were found by community type or by survey wave. Respondents’ age at interview was positively associated with reporting having sexual intercourse before age 13 years. The absolute difference was modest, however; we estimated that 3% of 16-year-old and 4% of 20-year-old respondents would report their sexual activity before 13 years of age.

nity type or by survey wave. Respondents’ age at interview was positively associated with reporting having sexual intercourse before age 13 years. The absolute difference was modest, however; we estimated that 3% of 16-year-old and 4% of 20-year-old respondents would report their sexual activity before 13 years of age. Table 4. Multivariable Logistic Regression of the Association Between Sexual Intercourse Before Age 13 Years and Various Demographic Characteristics Variable OR (95% CI) P Value Race/ethnicity Non-Hispanic black 1 [Reference] Non-Hispanic white 0.21 (0.13-0.32) <.001 Hispanic 0.27 (0.18-0.4) <.001 Non-Hispanic other 0.16 (0.06-0.38) <.001 Maternal educational level <College 1 [Reference] ≥College degree 0.31 (0.19-0.49) <.001 Community type Urban 0.87 (0.52-1.45) .58 Suburban 0.76 (0.45-1.26) .29 Rural 1 [Reference] Survey year 2006-2010 1 [Reference] 2011-2015 1.16 (0.81-1.67) .40 Age of respondent at interview 1.09 (1.03-1.14) .001 Estimated probabilities for race/ethnicity by maternal educational level, % (95% CI)a Non-Hispanic black: <college [reference] 12 (9-14) Non-Hispanic black: ≥college degree 6 (2-10) .03 Non-Hispanic white: <college 3 (2-4) <.001 Non-Hispanic white: ≥college degree 0 (0-1) <.001 Hispanic: <college 4 (2-5) <.001 Hispanic: ≥college degree 2 (0-3) <.001 Non-Hispanic other: <college 2 (0-4) <.001 Non-Hispanic other: ≥college degree 1 (0-2) <.001 Abbreviations: NSFG, National Survey of Family Growth; OR, odds ratio.

nic white: <college 3 (2-4) <.001 Non-Hispanic white: ≥college degree 0 (0-1) <.001 Hispanic: <college 4 (2-5) <.001 Hispanic: ≥college degree 2 (0-3) <.001 Non-Hispanic other: <college 2 (0-4) <.001 Non-Hispanic other: ≥college degree 1 (0-2) <.001 Abbreviations: NSFG, National Survey of Family Growth; OR, odds ratio. a Estimated from additional model, including interaction term between race/ethnicity and maternal educational level (eTable 3 in the Supplement). Data from the male respondents aged 15 to 24 years pooled from the 2006 to 2010 and 2011 to 2015 continuous survey rounds of the NSFG. After adding to the model interaction terms between race/ethnicity and maternal educational level (eTable 3 in the Supplement), we estimated that 12% (95% CI, 9%-14%) of non-Hispanic black males whose mothers did not have a college degree would report having their first intercourse before age 13 years (Table 4); this probability is statistically significantly higher than for males in any other combination of race/ethnicity and maternal educational level, which ranged from 0% (95% CI, 0%-1%) for non-Hispanic white males whose mothers had a college degree or more to 6% (95% CI, 2%-10%) for non-Hispanic black males whose mothers had a college degree or more.

ally significantly higher than for males in any other combination of race/ethnicity and maternal educational level, which ranged from 0% (95% CI, 0%-1%) for non-Hispanic white males whose mothers had a college degree or more to 6% (95% CI, 2%-10%) for non-Hispanic black males whose mothers had a college degree or more. Wantedness of First Sexual Intercourse by Age of Sexual Onset Among respondents aged 18 to 24 years who reported having their first sexual intercourse before age 13 years, 8.5% (95% CI, 3.8%-17.8%) characterized it as unwanted, 37.0% (95% CI, 28.0%-47.0%) had mixed feelings about it, and 54.6% (95% CI, 44.7%-64.1%) described it as wanted (Figure). These distributions were not statistically significantly different from those among males who had sexual intercourse at age 13 years or later: 5.4% reported it as unwanted, 31.3% had mixed feelings, and 63.3% described it as wanted. Most of the respondents who reported having sex before age 13 years described their first sexual partner as a friend (eTable 4 in the Supplement). Figure. Percent Distribution of Wantedness of First Sexual Intercourse Among Sexually Experienced Male Respondents Aged 18 to 24 Years, by Age at First Sexual Intercourse Data were pooled from the 2006 to 2010 and 2011 to 2015 continuous survey rounds of the National Survey of Family Growth.

Wantedness of First Sexual Intercourse by Age of Sexual Onset Among respondents aged 18 to 24 years who reported having their first sexual intercourse before age 13 years, 8.5% (95% CI, 3.8%-17.8%) characterized it as unwanted, 37.0% (95% CI, 28.0%-47.0%) had mixed feelings about it, and 54.6% (95% CI, 44.7%-64.1%) described it as wanted (Figure). These distributions were not statistically significantly different from those among males who had sexual intercourse at age 13 years or later: 5.4% reported it as unwanted, 31.3% had mixed feelings, and 63.3% described it as wanted. Most of the respondents who reported having sex before age 13 years described their first sexual partner as a friend (eTable 4 in the Supplement). Figure. Percent Distribution of Wantedness of First Sexual Intercourse Among Sexually Experienced Male Respondents Aged 18 to 24 Years, by Age at First Sexual Intercourse Data were pooled from the 2006 to 2010 and 2011 to 2015 continuous survey rounds of the National Survey of Family Growth. Sensitivity Analysis None of the sensitivity adjustments to the NSFG data (using ACASI measures of age at first sexual intercourse, limiting the sample to those currently aged 15 to 19 years and in school, or dropping all imputed data on age at first sexual intercourse) narrowed the gap between the NSFG and YRBSS prevalence estimates (eTable 5 in the Supplement). Kaplan-Meier life table estimates of age at first sexual intercourse between the NSFG and the YRBSS data sets showed persistent differences across age (eFigure in the Supplement).

uted data on age at first sexual intercourse) narrowed the gap between the NSFG and YRBSS prevalence estimates (eTable 5 in the Supplement). Kaplan-Meier life table estimates of age at first sexual intercourse between the NSFG and the YRBSS data sets showed persistent differences across age (eFigure in the Supplement). Discussion Drawing on representative surveys, we found that reported rates of sexual onset before age 13 years among adolescent males in the United States varied substantially by race/ethnicity, location, and maternal educational level. Although males in the YRBSS reported a higher prevalence of sex before 13 years of age than males in the NSFG, the differentials by race/ethnicity in each survey were extremely similar. The NSFG data also showed substantial differentials by maternal educational level, even after controlling for race/ethnicity. These findings underscore the need for providing comprehensive sex education that is culturally informed and inclusive before an individual’s first sexual encounter and ensuring that health care practitioners discuss sex with their male patients starting during middle school years or earlier.

olling for race/ethnicity. These findings underscore the need for providing comprehensive sex education that is culturally informed and inclusive before an individual’s first sexual encounter and ensuring that health care practitioners discuss sex with their male patients starting during middle school years or earlier. This study extends previous findings that showed a high prevalence of sexual activity before age 13 years, particularly among males of color, and highlights substantial variation across metropolitan areas.4,13,37 These findings reinforce that males’ identities and community contexts are associated with their experiences. Other studies have found that age at first sexual intercourse is associated with identifiable systemic barriers in communities, such as racial segregation25 and neighborhood disadvantage.28,30 Adolescent males’ experiences of emerging sexuality are informed by their social context, which is often defined by location. To this end, investments at the local level will be critical to support and promote youth development generally and healthy sexual development specifically.

5 and neighborhood disadvantage.28,30 Adolescent males’ experiences of emerging sexuality are informed by their social context, which is often defined by location. To this end, investments at the local level will be critical to support and promote youth development generally and healthy sexual development specifically. Some of the reported first sexual experiences were characterized as unwanted and mixed feelings were common, but more than half reported their experience as wanted. This finding underscores the need to include young men’s views when identifying and interpreting their sexual and developmental trajectories. However, wantedness of early sexual intercourse among male adolescents may also represent efforts to conform to traditional cultural expectations of masculinity and may not on its own represent healthy sexual choices. Health education and counseling can create opportunities for young males, their families, and communities to discuss healthy sexuality, including topics of consent, coercion, and development of sexual expression.3,38 Systems of care are also needed to support and treat young males who have experienced unwanted sexual encounter.

ucation and counseling can create opportunities for young males, their families, and communities to discuss healthy sexuality, including topics of consent, coercion, and development of sexual expression.3,38 Systems of care are also needed to support and treat young males who have experienced unwanted sexual encounter. Normative masculinity values may be a factor in reporting of age at first sexual experience, explaining some of the observed race/ethnicity differentials. Studies have found inconsistent reporting of first sexual intercourse among black males, which is suggestive of overreporting of younger ages at first sexual intercourse.39 Social pressures for such reporting may be a particular factor in responses to school-based surveys, which may partially explain the higher rates in the YRBSS data compared with rates in the NSFG data.40 Other survey differences, including sampling frames, survey modes, and survey measures, may also explain some of these differences.41 Although race/ethnicity and gender-specific norms may be a factor in reporting of age at first sexual intercourse, such norms in and of themselves are relevant to adolescent males’ actual experiences and development. For example, black children, particularly males, often are viewed as more adult than white children, without the need for protection afforded to others.42 Health care practitioners should work together with parents or guardians and their sons during late middle childhood and early adolescence to identify and respond to the normative pressures around masculinity.43

s, often are viewed as more adult than white children, without the need for protection afforded to others.42 Health care practitioners should work together with parents or guardians and their sons during late middle childhood and early adolescence to identify and respond to the normative pressures around masculinity.43 Early sexual onset has been associated with increased prevalence of negative physical and mental health outcomes.44 The causal role of age at first sexual intercourse is unclear, however, and associations may be driven by unmeasured confounding factors such as childhood sexual abuse, pubertal timing, or parental engagement. Emerging research suggests that the context of sexual initiation may have greater implications for sexual health than age alone.45 Our focus on sexual onset before age 13 years recognizes an important developmental and social marker. Variation exists in when cognitive, emotional, and physical development milestones are reached, and other studies have used different chronological cutoffs in their investigation of early first sexual intercourse,44,46 but there is no particular biological age at which individuals are uniformly ready for this transition. Policy approaches that aim to delay sexual onset may focus on the wrong drivers of behavior and may risk causing harm through stigmatization of and barriers to sex education and sexual health services.47

course,44,46 but there is no particular biological age at which individuals are uniformly ready for this transition. Policy approaches that aim to delay sexual onset may focus on the wrong drivers of behavior and may risk causing harm through stigmatization of and barriers to sex education and sexual health services.47 Healthy People 2020 calls for providing sex education to adolescents before age 18 years.48 Ideally, this education should begin before an adolescent first has sexual intercourse, but fewer than half of all teenaged males receive formal instruction on birth control before their first sexual encounter.7 In addition, parents are less likely to talk with their sons than daughters about many sexual health topics.39 Health care practitioners should expand how they fill these gaps during early adolescence and late middle childhood. Although the Bright Futures guidelines for early adolescence discuss screening and anticipatory guidance for sexual activity, as well as the importance of individual time with a clinician, its guideline for middle childhood is focused on sexual abuse and offers less guidance on the importance of visit time alone.49 Our finding that substantial shares of male adolescents have their first sexual intercourse before age 13 years underscores the need to provide comprehensive sex education and sexual and reproductive health care that start early and are developmentally appropriate for children’s age.6 Greater attention also needs to be given to providing sex education that is culturally informed and inclusive.50

l intercourse before age 13 years underscores the need to provide comprehensive sex education and sexual and reproductive health care that start early and are developmentally appropriate for children’s age.6 Greater attention also needs to be given to providing sex education that is culturally informed and inclusive.50 Limitations This study has several limitations. Age at first sexual intercourse is only 1 indicator of sexuality; efforts to understand the trajectories of sexual development should consider behaviors beyond vaginal-penile intercourse.51 Furthermore, the available measures are fairly blunt constructs of the intersection of sex, race/ethnicity, socioeconomic status, and location and likely miss further nuance and differentiation. In addition, the increased reporting of earlier occurrence of first sexual intercourse in the middle school data compared with the high school YRBSS and by age in the NSFG is suggestive of reporting bias associated with age or length of recall. The study’s use of multiple high-quality surveys to triangulate reports offsets this limitation.

eased reporting of earlier occurrence of first sexual intercourse in the middle school data compared with the high school YRBSS and by age in the NSFG is suggestive of reporting bias associated with age or length of recall. The study’s use of multiple high-quality surveys to triangulate reports offsets this limitation. Conclusions Rates of sexual onset before age 13 years among young males in the United States varied widely by race/ethnicity, location, and maternal educational level, with higher rates among non-Hispanic black males in most metropolitan areas. These findings may have major implications for the timing of sex education and sexual and reproductive health care. Helping parents or guardians, schools, and communities support male adolescents’ healthy sexual development should be a priority. Health care practitioners must recognize and address all of the developmental needs and pathways to healthy trajectories for young males. Supplement. eTable 1. Percent of Male Students Reporting Their First Sexual Intercourse Before Age 13 Years, 2011, 2013 and 2015 YRBS; Percent of Young Men Aged 15-24 Reporting Their First Sexual Intercourse Before Age 13 Years, 2006-2010 and 2011-2015 NSFG eTable 2. Predicted Proportion of Male Middle School Students Reporting Their First Sexual Intercourse Before Age 13 Years, by Race/Ethnicity and Location eTable 3. Full Multivariable Logistic Regression Model of the Association Between First Sexual Intercourse Before Age 13 Years and Various Demographic Characteristics, Used to Produce Predicted Probabilities in eTable 2

eTable 2. Predicted Proportion of Male Middle School Students Reporting Their First Sexual Intercourse Before Age 13 Years, by Race/Ethnicity and Location eTable 3. Full Multivariable Logistic Regression Model of the Association Between First Sexual Intercourse Before Age 13 Years and Various Demographic Characteristics, Used to Produce Predicted Probabilities in eTable 2 eTable 4. Relationship with First Sexual Partner, Among Men Aged 15-24 Reporting Their First Sexual Intercourse Before Age 13 Years, Pooled 2006-2010 and 2011-2015 NSFG eTable 5. Sensitivity Analyses of Percent of Males Aged 15-24 Reporting Their First Sexual Intercourse Before Age 13 Years, Pooled 2006-10 and 2011-15 NSFG; According to Race/Ethnicity and Sensitivity Specification eFigure. Kaplan-Meier Failure Curve of Age at First Sex Among Male 15-24 Year Olds, Pooled NSFG 2006-10 and 2011-15, and Male Students, Pooled YRBS 2011-2015 Click here for additional data file.

Introduction Autism spectrum disorder (ASD) is a common disorder of childhood, affecting 1 in 59 children.1 It is also becoming clear that ASD has its beginnings during prenatal life.2 Because many children with ASD have clinical signs within the first year, such as failure to respond to their name3 and reduced positive affect,4 there is a considerable demand for early detection, intervention, and services.5 Although several studies have shown that early signs of ASD can sometimes be detected using parent report screens as early as 126,7,8 or 18 months of age,9,10 the mean patient age at ASD detection is several years later, generally between 3 and 4 years of age.1 This late age of detection is a missed opportunity given the accelerated pace of brain development that occurs between birth and 3 to 4 years of age.11 Despite the appeal of the concept of early detection and treatment in ASD, there are many unknowns. Foundational questions regarding early-age diagnostic stability, age of clinical symptom onset, and overlap of early-age clinical symptoms between ASD and other disorders, such as language delay or global developmental delay, remain unanswered. A previous report12 by the US Preventive Services Task Force did not endorse early universal screening for ASD given the lack of clarity regarding the balance of benefits and harms of early screening and detection.

between ASD and other disorders, such as language delay or global developmental delay, remain unanswered. A previous report12 by the US Preventive Services Task Force did not endorse early universal screening for ASD given the lack of clarity regarding the balance of benefits and harms of early screening and detection. The months surrounding the first birthday are a remarkable time for a toddler’s development. At this age, toddlers learn to walk,13 speak their first word,14 and engage in a range of joint social attention behaviors, such as pointing and showing objects to others to share social attentional focus.15 The toddler stage is also the earliest age that ASD can be detected and treatment started,6,16 yet the stability of an ASD diagnosis at this pivotal age is unknown.

irst word,14 and engage in a range of joint social attention behaviors, such as pointing and showing objects to others to share social attentional focus.15 The toddler stage is also the earliest age that ASD can be detected and treatment started,6,16 yet the stability of an ASD diagnosis at this pivotal age is unknown. A previous report17 stated that most studies examining the diagnostic stability of ASD before 3 years of age have involved slightly older, clinic-referred cohorts, usually at approximately 2 years of age. Stability coefficients within these studies have been high (mean, 88%, range, 63%-100%).17 Two studies examined stability at an even younger age (18 months) but examined this question from within multiplex families using the infant sibling design. One of these studies reported that 93% of siblings first diagnosed as having ASD at 18 months retained that diagnosis at a final diagnostic age of 36 months,17 but only 69% of siblings first diagnosed as having ASD at 24 months did so (ie, 27 of 39 retained diagnosis).17 Although studies collectively suggest that an ASD diagnosis is moderately stable at young ages,17 there are several key questions remaining. First, it is unclear whether stability estimates from infant sibling designs would be found within a general population cohort. Second, none of the previous clinic-referred cohort studies included large groups of toddlers without ASD ascertained in the same manner as the toddlers with ASD. Such contrast groups are essential to understand how the ASD phenotype emerges from and overlaps with clinical expressions from other diagnostic groups, such as language and developmental delay, commonly found in clinical settings. Third, clinic-referred studies are small, usually containing 50 to 100 participants, and may generate less stable results. Moreover, children referred to a clinic because of already suspected ASD may generate artificially high stability rates relative to a community-ascertained sample.17 Fourth, despite the potential of the infant sibling design to study ASD from birth, stability estimates have only been reported starting at 18 months of age, leaving questions surrounding younger ages unanswered.

suspected ASD may generate artificially high stability rates relative to a community-ascertained sample.17 Fourth, despite the potential of the infant sibling design to study ASD from birth, stability estimates have only been reported starting at 18 months of age, leaving questions surrounding younger ages unanswered. Interleaved with these gaps in knowledge is the recent finding from infant sibling studies17,18 that 50% to 80% of toddlers eventually diagnosed as having ASD at 3 years of age were not identified as having ASD by expert clinicians at 18 months of age. In short, despite extensive clinical testing that included the gold standard tool the Autism Diagnostic Observation Schedule (ADOS),19 these diagnoses were missed. A newer study,20 however, suggests that such so-called late-onset cases may be attributable to weaknesses inherent in standardized diagnostic tools at early ages, rather than a lack of observable ASD symptoms per se. Determining the degree to which such late-onset cases may be present in a general population cohort is essential, because if rates are as high as in infant sibling cohorts, it would strongly underscore the American Academy of Pediatrics recommendation for repeat screening at multiple ages. It would also add further urgency to the search for early behavioral or biological tests for ASD to more readily detect ASD during the earliest ages when detection is the most challenging. In this study, we sought to examine the diagnostic stability of ASD in a large cohort of toddlers starting at 12 months of age and to compare this stability with that of toddlers with other disorders, such as developmental delay.

o more readily detect ASD during the earliest ages when detection is the most challenging. In this study, we sought to examine the diagnostic stability of ASD in a large cohort of toddlers starting at 12 months of age and to compare this stability with that of toddlers with other disorders, such as developmental delay. Methods Participants A total of 2241 toddlers 12 to 36 months of age were referred for a diagnostic evaluation to an autism expert evaluation center created at University of California, San Diego. Referrals were given through our early detection program, Get SET Early,6,21 which systematically screens for ASD and other disorders in the general population at 12-, 18-, and 24-month well-child checkups or through the general community. Typically developing (TD) toddlers were also recruited from the same pediatric offices participating in the Get SET Early program (eMethods in the Supplement). A total of 1269 of the 2241 toddlers were longitudinally evaluated 2 or more times and were the focus of this study. In this sample, approximately 75% came from the Get SET Early program and approximately 25% from community referral. Additional eligibility requirements included an interval of 6 months or longer between the first and last evaluations. Figure 1 and eFigure 1 and eFigure 2 in the Supplement show the cohort characteristics. This study was overseen by the institutional review board at the University of California, San Diego, and written informed consent was obtained from caregivers before study enrollment. At the data analysis phase of the study, the patient names were removed from our spreadsheets to protect their identity.

aracteristics. This study was overseen by the institutional review board at the University of California, San Diego, and written informed consent was obtained from caregivers before study enrollment. At the data analysis phase of the study, the patient names were removed from our spreadsheets to protect their identity. Figure 1. Sample Characteristics Distribution of key features associated with the study cohort, including the age (in months) that toddlers received their first comprehensive diagnostic evaluation (A), the number of toddlers who received 2, 3, or 4 or more diagnostic evaluations (B), the age (in months) that toddlers received their last diagnostic evaluation, and the interval (in months) between a toddler’s first and last diagnostic evaluation (D).

eliably diagnosed with ASD several years earlier, as young as 14 months. The implications of this finding extend beyond information that relates to diagnostic stability and may open new opportunities to consider if and how treatments started at this early age are associated with long-term outcomes of affected children. An initial ASD diagnosis was more stable than any other diagnosis, including TD. In our cohort, 84% of toddlers initially diagnosed with ASD at their first visit retained this diagnosis at 3 to 4 years of age. Most toddlers within the remaining 16% did not lose their delays entirely but instead presented with milder delays at their final diagnostic visit. The most common transition was ASD to ASD features, a diagnostic category used for toddlers with signs of ASD but not enough to meet DSM criteria. The least common transition was ASD to TD (ie, only 1.8% of toddlers initially designated as having ASD transitioned to TD). Because all toddlers were immediately referred for treatment once any delay was detected, improvements in symptom severity could have been associated with a positive impact of very early treatment, which research suggests may be more beneficial than treatment started at older ages.26,27,28 From a public policy perspective, this finding suggests that it is important to initiate treatment immediately after an initial designation of ASD, even at the youngest ages. The human brain has an enormous capacity to resculpt and remodel, particularly during the first postnatal years. The few studies that have examined treatment during this transformative time window have found that toddlers with ASD,26,27,29,30 cerebral palsy,31 premature birth,32,33,34 and severe hearing loss35 experience significant positive changes, such as an increase in 15 IQ points29 or improvements in speech perception and language ability.36 The caveat, however, is that early-age diagnostic evaluations should be conducted by practitioners with considerable experience in early ASD development. In many places in the United States, such experience is severely lacking.37

oddlers received their first comprehensive diagnostic evaluation (A), the number of toddlers who received 2, 3, or 4 or more diagnostic evaluations (B), the age (in months) that toddlers received their last diagnostic evaluation, and the interval (in months) between a toddler’s first and last diagnostic evaluation (D). Diagnostic and Psychometric Testing Highly experienced, licensed psychologists with PhD degrees performed diagnostic and psychometric tests, including the ADOS-2 (module T, 1, or 2 as appropriate),19 Mullen Scales of Early Learning,22 and Vineland Adaptive Behavior Scales.23 Toddlers who received their first diagnostic evaluation at younger than 36 months were diagnostically tested approximately every 12 months until 3 years of age. After each visit, psychologists filled out a diagnostic judgment form and entered it into a database. Psychologists were not masked to previous diagnoses during longitudinal test visits. A toddler was designated as having 1 of the following: ASD, ASD features, developmental delay, language delay (LD), other issue, TD, or typical sibling of an ASD proband. Parents were given feedback regarding their child’s performance after completion of testing and referred for treatment as appropriate. A description of psychologist training, diagnostic criteria used, data quality control process, and estimated Mullen T scores generated for 9% of toddlers who scored below a standard T score of 20 are given in the eMethods in the Supplement. The Table and eFigure 2 in the Supplement give information regarding the Diagnostic and Statistical Manual of Mental Disorders (DSM) version used.

ta quality control process, and estimated Mullen T scores generated for 9% of toddlers who scored below a standard T score of 20 are given in the eMethods in the Supplement. The Table and eFigure 2 in the Supplement give information regarding the Diagnostic and Statistical Manual of Mental Disorders (DSM) version used. Table.

ta quality control process, and estimated Mullen T scores generated for 9% of toddlers who scored below a standard T score of 20 are given in the eMethods in the Supplement. The Table and eFigure 2 in the Supplement give information regarding the Diagnostic and Statistical Manual of Mental Disorders (DSM) version used. Table. Demographic Information and Clinical Test Scores for Each Diagnostic Groupa Characteristic at Last Diagnostic Visit ASD (n = 441) ASD Features (n = 78) DD (n = 89) LD (n = 80) Other (n = 91) Typical Sibling (n = 51) TD (n = 439) Sex Male 361 (81.9) 68 (87.2) 66 (74.2) 58 (72.5) 61 (67.0) 26 (51.0) 278 (65.6) Female 80 (18.1) 10 (12.8) 23 (25.8) 22 (27.5) 30 (33.0) 25 (49.0) 161 (36.7) Age, mean (SD), mo 42.84 (20.28) 40.77 (17.61) 35.91 (10.15) 35.44 (11.42) 42.92 (13.08) 38.44 (13.76) 37.10 (9.84) Final DSM diagnosisb DSM-IV 135 19 24 23 34 26 202 DSM-5 306 59 65 57 57 25 237 Ethnicity Hispanic/Latino 128 (29.0) 17 (21.8) 36 (40.4) 38 (47.5) 20 (22.0) 15 (29.4) 87 (19.8) Non-Hispanic/Latino 263 (59.6) 53 (67.9) 47 (52.8) 39 (48.8) 64 (70.3) 31 (60.8) 325 (74.0) Not reported 50 (11.3) 8 (10.3) 6 (6.7) 3 (3.8) 7 (7.7) 5 (9.8) 27 (6.2) Race White 237 (53.7) 51 (65.4) 48 (53.9) 47 (58.8) 60 (65.9) 31 (60.8) 299 (68.1) Black/African American 9 (2.0) 1 (1.3) 2 (2.2) 2 (2.5) 5 (5.5) 2 (3.9) 12 (2.7) Asian 48 (10.9) 7 (9.0) 7 (7.9) 1 (1.3) 2 (2.2) 1 (2.0) 41 (9.3) Pacific Islander 4 (0.90) 3 (3.8) 4 (4.5) 2 (2.5) 2 (2.2) 1 (2.0) 5 (1.1) Native American/Alaska 2 (0.50) 0 1 (1.1) 3 (3.8) 1 (1.1) 0 0 Mixed race 57 (12.9) 7 (9.0) 9 (10.1) 1 (1.3) 11 (12.1) 7 (13.7) 46 (10.5) Not reported 84 (19.0) 9 (11.5) 18 (20.2) 24 (30.0) 10 (11.0) 9 (17.6) 36 (8.2) Mullen T score, mean (SD)c Visual reception 38.0 (14.9) 51.4 (13.5) 35.3 (13.4) 49.6 (11.7) 54.1 (13.4) 61.0 (9.7) 59.2 (10.6) Fine motor 34.0 (12.6) 43.8 (11.6) 31.2 (11.0) 46.4 (10.4) 45.3 (13.2) 53.1 (10.3) 52.4 (10.4) Receptive language 32.1 (15.0) 46.1 (12.0) 33.4 (12.0) 40.8 (10.6) 48.7 (11.0) 52.7 (10.3) 53.8 (9.0) Expressive language 30.6 (16.9) 48.6 (12.0) 30.8 (13.9) 33.9 (9.4) 49.2 (12.4) 54.2 (8.8) 53.2 (8.7) ELC 71.5 (22.1) 94.7 (22.6) 68.6 (17.7) 86.0 (15.1) 99.4 (18.7) 110.5 (14.7) 109.2 (14.4) Vineland standard score, mean (SD) Communication 72.1 (25.0) 96.3 (21.2) 78.3 (21.2) 84.6 (19.5) 98.2 (17.2) 101.2 (18.0) 102.0 (19.4) Daily living 75.2 (22.5) 95.1 (18.2) 83.8 (19.0) 94.9 (18.6) 96.3 (15.7) 98.7 (16.8) 100.0 (17.7) Socialization 72.6 (21.5) 95.2 (18.6) 85.9 (18.5) 92.3 (18.4) 97.0 (15.4) 103 (16.9) 10

Vineland standard score, mean (SD) Communication 72.1 (25.0) 96.3 (21.2) 78.3 (21.2) 84.6 (19.5) 98.2 (17.2) 101.2 (18.0) 102.0 (19.4) Daily living 75.2 (22.5) 95.1 (18.2) 83.8 (19.0) 94.9 (18.6) 96.3 (15.7) 98.7 (16.8) 100.0 (17.7) Socialization 72.6 (21.5) 95.2 (18.6) 85.9 (18.5) 92.3 (18.4) 97.0 (15.4) 103 (16.9) 10 2.0 (18.0) Motor skills 76.2 (27.0) 92.5 (20.4) 80.4 (20.3) 91.9 (23.9) 91.3 (17.9) 95.8 (15.9) 96.3 (19.9) Adaptive behavior composite 73.3 (21.8) 95.5 (16.2) 80.5 (13.1) 91.23 (11.4) 96.0 (13.4) 100.7 (10.5) 101.6 (12.3) ADOS (module T, 1, or 2) score, mean (SD)d ADOS SA/CoSo score 12.9 (4.1) 4.4 (2.7) 3.8 (3.3) 2.4 (2.1) 3.1 (2.4) 2.0 (1.8) 2.2 (1.7) ADOS RRB score 4.6 (1.9) 2.6 (1.5) 1.4 (1.5) 0.6 (0.9) 0.7 (0.8) 0.3 (0.7) 0.6 (1.0) ADOS total score 17.6 (4.8) 7.0 (3.1) 5.2 (3.1) 3.0 (2.3) 3.8 (2.6) 2.4 (1.8) 2.8 (2.0) Abbreviations: ADOS, Autism Diagnostic Observation Schedule; ASD, autism spectrum disorder; CoSo, Communication Social Score; DD, developmental delay; DSM, Diagnostic and Statistical Manual of Mental Disorders; ELC, early learning composite; LD, language delay; RRB, restricted and repetitive behavior; SA, social affect; TD, typical development. a Data are presented as number (percentage) of toddlers unless otherwise indicated. b Version of the DSM used at the final diagnostic evaluation (eMethods and eFigure 2 in the Supplement).

2.0 (18.0) Motor skills 76.2 (27.0) 92.5 (20.4) 80.4 (20.3) 91.9 (23.9) 91.3 (17.9) 95.8 (15.9) 96.3 (19.9) Adaptive behavior composite 73.3 (21.8) 95.5 (16.2) 80.5 (13.1) 91.23 (11.4) 96.0 (13.4) 100.7 (10.5) 101.6 (12.3) ADOS (module T, 1, or 2) score, mean (SD)d ADOS SA/CoSo score 12.9 (4.1) 4.4 (2.7) 3.8 (3.3) 2.4 (2.1) 3.1 (2.4) 2.0 (1.8) 2.2 (1.7) ADOS RRB score 4.6 (1.9) 2.6 (1.5) 1.4 (1.5) 0.6 (0.9) 0.7 (0.8) 0.3 (0.7) 0.6 (1.0) ADOS total score 17.6 (4.8) 7.0 (3.1) 5.2 (3.1) 3.0 (2.3) 3.8 (2.6) 2.4 (1.8) 2.8 (2.0) Abbreviations: ADOS, Autism Diagnostic Observation Schedule; ASD, autism spectrum disorder; CoSo, Communication Social Score; DD, developmental delay; DSM, Diagnostic and Statistical Manual of Mental Disorders; ELC, early learning composite; LD, language delay; RRB, restricted and repetitive behavior; SA, social affect; TD, typical development. a Data are presented as number (percentage) of toddlers unless otherwise indicated. b Version of the DSM used at the final diagnostic evaluation (eMethods and eFigure 2 in the Supplement). c A total of 9% percent of the overall sample had a chronologic or mental age that exceeded the validated age range for use with the Mullen scales at their last diagnostic evaluation visit and received a Wechsler Preschool and Primary Scale of Intelligence instead. d Administered ADOS module depended on the age and language ability of the toddler at the time of testing. For these individuals, their most recent available Mullen scores were used.

c A total of 9% percent of the overall sample had a chronologic or mental age that exceeded the validated age range for use with the Mullen scales at their last diagnostic evaluation visit and received a Wechsler Preschool and Primary Scale of Intelligence instead. d Administered ADOS module depended on the age and language ability of the toddler at the time of testing. For these individuals, their most recent available Mullen scores were used. Statistics and Data Visualization Diagnostic Stability Stability coefficients were first calculated within 2-month age bands by determining the proportion of toddlers with a particular diagnosis at their first diagnostic visit who retained that same diagnosis at their last visit. Diagnostic transition tables were created for overall and 2-month–interval age-binned data. Diagnostic stability was modeled using logistic regression, with sex, age at first diagnosis, interval between first and last diagnosis, and diagnostic group at first visit as variables and results reported as odds ratios (ORs) (eTable 1 in the Supplement). To examine the association of age at first diagnosis with stability coefficients while optimizing statistical power, we binned age to 4 roughly equally populated groups: younger than 14 months, 14 to 17.99 months, 18 to 23.99 months, 24 months or older. No significant association between sex or interval with stability coefficients was found. Follow-up models with the 4 age bins as the only covariates were used to examine the association of age at first diagnosis with stability coefficients within each diagnostic group. A B-spline method with 3 df was also used to test the nonlinear, continuous association of age at first diagnosis with stability coefficients within each diagnosis group and to visualize the data. All analyses were performed in the R programming environment (R Foundation for Statistical Computing) (eMethods in the Supplement). Given caution from the ADOS developers regarding use of the ADOS with toddlers with nonverbal mental ages younger than 12 months and those who are not ambulatory,24 stability coefficients were also recalculated after removing such cases.

environment (R Foundation for Statistical Computing) (eMethods in the Supplement). Given caution from the ADOS developers regarding use of the ADOS with toddlers with nonverbal mental ages younger than 12 months and those who are not ambulatory,24 stability coefficients were also recalculated after removing such cases. Transition Patterns Visualized Using Diagnostic Heat Maps To visualize how phenotypic expression of ASD and other disorders changed across visits, diagnostic heat maps were generated using the ggplot library in R software. With the use of this approach, diagnostic judgments were illustrated inward out from first diagnosis to the final diagnosis, and each diagnostic judgment was assigned a unique color. In this way, transition patterns across diagnosis visits could be visually deciphered.

generated using the ggplot library in R software. With the use of this approach, diagnostic judgments were illustrated inward out from first diagnosis to the final diagnosis, and each diagnostic judgment was assigned a unique color. In this way, transition patterns across diagnosis visits could be visually deciphered. ASD Identification Designation and Clinical Differences For comparison with infant sibling diagnostic stability studies, the ASD cohort was categorized as having an early-age persistent ASD diagnosis, which was defined as an ASD diagnosis at or before 18 months of age that was retained at final diagnosis; middle-age persistent ASD diagnosis, which was defined as an ASD diagnosis after 18 months of age that was retained at final diagnosis; and late-identified ASD, which was defined as ASD not diagnosed at first diagnostic visit regardless of intake age. The 270 toddlers who were initially identified as having TD and retained this diagnosis at final diagnostic age were used as a contrast cohort. One-way analyses of variance with follow-up planned contrasts and Cohen d were used to examine clinical differences. Expanded comparisons that included all diagnostic groups were also conducted (eFigure 3 in the Supplement).

s having TD and retained this diagnosis at final diagnostic age were used as a contrast cohort. One-way analyses of variance with follow-up planned contrasts and Cohen d were used to examine clinical differences. Expanded comparisons that included all diagnostic groups were also conducted (eFigure 3 in the Supplement). Results Participant Characteristics Among the 1269 toddlers, 918 (72.3%) were male, median age at first evaluation was 17.6 months (interquartile range, 14.0-24.4 months), mean number of diagnostic visits was 2.7 (interquartile range, 2-3), and median age at final evaluation was 36.2 months (interquartile range, 33.4-40.9 months). The Table gives the demographic information and clinical test scores for each diagnostic group.

uation was 17.6 months (interquartile range, 14.0-24.4 months), mean number of diagnostic visits was 2.7 (interquartile range, 2-3), and median age at final evaluation was 36.2 months (interquartile range, 33.4-40.9 months). The Table gives the demographic information and clinical test scores for each diagnostic group. Diagnostic Stability Overall stability was 0.84 (95% CI, 0.80-0.87) for an ASD diagnosis and 0.79 (95% CI, 0.74-0.83) for a TD diagnosis (Figure 2A). Results from the overall logistic regression model showed a significant association of age and diagnosis at first visit with stability (eTable 1 in the Supplement). No significant differences were found in stability based on sex (OR, 0.76; 95% CI, 0.56-1.04) or interval between first and last diagnostic evaluations (OR, 0.99; 95% CI, 0.98-1.00). Logistic regression analyses showed a nonsignificant difference in stability coefficients between ASD and TD (OR, 0.86; 95% CI, 0.57-1.29). In contrast, significant differences were found between ASD and the remaining diagnostic groups (OR, 0.11 [95% CI, 0.03-0.32] vs ASD features; OR, 0.15 [95% CI, 0.09-0.25] vs DD; OR, 0.04 [95% CI, 0.03-0.06] vs LD; and OR, 0.16 [95% CI, 0.09-0.28] vs other) (eTable 1 in the Supplement). For ASD, stability was weakest at 12 to 13 months of age (stability coefficient, 0.50; 95% CI, 0.32-0.69). Stability of an ASD diagnosis increased to 0.79 by 14 months of age and 0.83 by 16 months of age (age bands of 12 vs 14 and 16 months; OR, 4.25; 95% CI, 1.59-11.74) (Figure 3 and eFigure 4, eFigure 5, eTable 2, and eTable 3 in the Supplement). When toddlers with ASD features were considered to have ASD, the stability coefficients increased to 0.70 (95% CI, 0.52-0.85) at 12 months of age, 0.85 (95% CI, 0.71-0.94) at 14 months of age, and 0.94 (95% CI, 0.81-0.99) at 16 months of age. Given the transient nature of many early delays,25 overall stability was low for the remaining delay groups (Figure 2 and Figure 3 and eFigure 4 and eTable 4 in the Supplement). Exclusion of 73 toddlers (34 with ASD, 1 with ASD features, 24 with DD, 7 with other disorders, 1 with a typical sibling, and 6 with TD) whose nonverbal mental age based on the visual reception component of the Mullen scale was younger than 12 months (mean nonverbal mental age, 9.6 months) did not improve the stability coefficient of ASD at 12 to 13 months (eTable 5 and eFigure 6 in the Supplement).

er disorders, 1 with a typical sibling, and 6 with TD) whose nonverbal mental age based on the visual reception component of the Mullen scale was younger than 12 months (mean nonverbal mental age, 9.6 months) did not improve the stability coefficient of ASD at 12 to 13 months (eTable 5 and eFigure 6 in the Supplement). Figure 2. Transition Table and Diagnostic Heat Maps A, Summary of the proportion of toddlers from within the entire sample (N = 1269) who retained or moved to a different diagnostic group between their first and last diagnostic visits. Stability coefficients are denoted within each cell (coefficients are not adjusted for the age at first diagnosis; a high concordance with age-adjusted coefficients was observed) (eTable 4 in the Supplement gives coefficients adjusted for age at first diagnosis). To read the table, compare values across each row or vertically within each column. For example, of the 400 toddlers initially designated as having autism spectrum disorder (ASD), 336 retained this diagnosis at their last (final) diagnostic visit, yielding a diagnostic stability coefficient of 0.84, whereas 35 toddlers had ASD features but no longer met ASD criteria, 6 tested as developmentally delayed, 6 as language delayed, 10 had other developmental issues, and 7 were designated as typically developing with no residual symptoms. For transition tables with stability coefficients within 2-month age bands, see eFigure 4 in the Supplement. B, Changes in diagnosis across visits. Colors represent each diagnostic group and rings represent each diagnostic visit, with the smallest center ring representing the first visit. The left-most panel summarizes diagnostic changes for toddlers who received 2 diagnostic evaluations a mean of 15 months apart; the center represents toddlers who received 3 diagnostic evaluations a mean of 11 months apart; and the right-most panel represents toddlers who received 4 or more diagnostic evaluations a mean of 8 months apart. The heat map indicates that ASD was the most stable diagnostic category, and that toddlers initially suspected as having developmental delay (DD) or language delay (LD) frequently received a final diagnosis of ASD. TD indicates typical development.

who received 4 or more diagnostic evaluations a mean of 8 months apart. The heat map indicates that ASD was the most stable diagnostic category, and that toddlers initially suspected as having developmental delay (DD) or language delay (LD) frequently received a final diagnosis of ASD. TD indicates typical development. Figure 3. Diagnostic Stability Plots by Age at First Diagnosis A, Plots show diagnostic stability per group across 2-month age intervals based on the age at first diagnostic evaluation. Age intervals with missing data points reflect an absence of toddlers who received their first diagnostic evaluation at that age. B-spline logistic regression line is shown; bands represent 95% CIs for the fit line. Overall stability was highest in toddlers initially designated as having autism spectrum disorder (ASD) or typical development as illustrated by the relatively tight CI bands, and the largely consistent stability coefficients within each age band. B, Diagnostic stability coefficients in the 4 age bins used in the logistic regression model across diagnostic groups. The lines represent 95% CIs. Coefficients were estimated based on within group logistic regression models. eFigure 5 and eFigure 6 in the Supplement give complementary analyses.

s within each age band. B, Diagnostic stability coefficients in the 4 age bins used in the logistic regression model across diagnostic groups. The lines represent 95% CIs. Coefficients were estimated based on within group logistic regression models. eFigure 5 and eFigure 6 in the Supplement give complementary analyses. Transition Patterns Diagnostic heat maps (Figure 2B) illustrate diagnostic transition patterns for toddlers who were evaluated 2, 3, or 4 or more times. The transition from an initial diagnosis of LD or developmental delay to ASD was the most common transition type. Transitioning from an initial designation of ASD to a final diagnosis of TD was rare and occurred in only 1.8% of overall cases (ie, 7 toddlers of 400 toddlers initially designated as ASD). However, 5 of these 7 toddlers with false-positive results were initially evaluated at the youngest ages (12-13 months of age) (eFigure 4 in the Supplement).

al designation of ASD to a final diagnosis of TD was rare and occurred in only 1.8% of overall cases (ie, 7 toddlers of 400 toddlers initially designated as ASD). However, 5 of these 7 toddlers with false-positive results were initially evaluated at the youngest ages (12-13 months of age) (eFigure 4 in the Supplement). ASD Identification Patterns and Clinical Differences For ASD, 66 toddlers (15.9%) were classified into the early-age diagnosis group, 270 (61.2%) into the middle-age diagnosis group, and 105 (23.8%) into the late-identified group. Overall, F tests revealed a significant between-group difference for all clinical domains (eTable 6 in the Supplement). Follow-up pairwise analyses revealed that these differences were driven by the late-identified group who had consistently better test scores than the other 2 ASD diagnosis groups for all measures at the first evaluation visit (range of Cohen d effect sizes, 0.44-2.35). However, toddlers with ASD in this group had significantly worse test scores than toddlers with TD (range of Cohen d effect sizes, 1.43-1.84), suggesting that symptoms were present. No clinical differences were found between the early and middle diagnosis groups (Figure 4).

t (range of Cohen d effect sizes, 0.44-2.35). However, toddlers with ASD in this group had significantly worse test scores than toddlers with TD (range of Cohen d effect sizes, 1.43-1.84), suggesting that symptoms were present. No clinical differences were found between the early and middle diagnosis groups (Figure 4). Figure 4. Comparison of Clinical Features in Toddlers With Autism Spectrum Disorder (ASD) Stratified by Identification Age Violin plots show differences in Autism Diagnostic Observation Schedule (ADOS) total scores (A), Mullen Expressive Language T scores (B), Vineland Adaptive Behavior Composite scores (C), and Mullen Receptive Language T scores (D) at the first diagnostic evaluation between toddlers with ASD identified at 12 to 18 months of age (early-age persistent ASD diagnosis), toddlers with ASD identified after 18 months (middle-age persistent ASD diagnosis), or toddlers not identified as having ASD at their first diagnostic visit (late-identified ASD). Black dots represent the mean. The width of the shape represents patient density, and the length illustrates the range of the scores. Data from 270 toddlers with typical development (TD) identified at their first diagnosis visit and retaining that diagnosis at their last visit are shown for comparison. Note that scores from the late-identified group were significantly different from toddlers with TD across all clinical domains, suggesting that symptoms were already present at the first diagnostic visit in a large fraction of late-identified ASD cases. Also note that 39 toddlers in the late-identified group (37%) did fall within the range of concern on the Autism Diagnostic Observation Schedule toddler module (cutoff score for concern using the few to no words algorithm = 10), however, were designated as non-ASD based on practitioner judgment, underscoring the challenges in differential diagnoses particularly at the youngest ages. Effect sizes are reported as Cohen d (95% CI). eFigure 3 in the Supplement gives an expanded figure that includes all diagnostic groups.

no words algorithm = 10), however, were designated as non-ASD based on practitioner judgment, underscoring the challenges in differential diagnoses particularly at the youngest ages. Effect sizes are reported as Cohen d (95% CI). eFigure 3 in the Supplement gives an expanded figure that includes all diagnostic groups. Discussion Children with ASD are detected and treated nationally at approximately 4 years of age.1 However, we found that within the context of an early detection program,6 children can be reliably diagnosed with ASD several years earlier, as young as 14 months. The implications of this finding extend beyond information that relates to diagnostic stability and may open new opportunities to consider if and how treatments started at this early age are associated with long-term outcomes of affected children.

an increase in 15 IQ points29 or improvements in speech perception and language ability.36 The caveat, however, is that early-age diagnostic evaluations should be conducted by practitioners with considerable experience in early ASD development. In many places in the United States, such experience is severely lacking.37 Although the overall stability of an ASD diagnosis was high, examination of the data within 2-month age bands revealed that stability was selectively low at 12 months, with a stability coefficient of 0.50. The lower stability coefficient for ASD specifically at 12 months is likely reflective of some limitations in the diagnostic tools used at that age, which included the ADOS-2 and DSM. In the first diagnosis ASD sample, which contained 400 toddlers, only 7 transitioned to a final diagnosis of TD, and 5 of these were within the 12-month age band. Research from the ADOS-2 developers cautions that it is not valid for use with toddlers who have a nonverbal mental age younger than 12 months,24 yet even when such toddlers were removed from analyses, the stability coefficient did not improve. Twelve months is an age when toddlers learn to talk, walk, merge, and shift attention with objects and others, and it is not surprising that this age would be the most diagnostically challenging. When toddlers with ASD features were included in the calculation at 12, 14, and 16 months of age, stability coefficients increased to 0.70 at 12 months, 0.85 at 14 months, and 0.94 at 16 months. This finding is likely related to the possibility that ASD is a dimensional rather than categorical disorder38,39 and the strict cutoff boundaries defined by the DSM may artificially affect results.

4, and 16 months of age, stability coefficients increased to 0.70 at 12 months, 0.85 at 14 months, and 0.94 at 16 months. This finding is likely related to the possibility that ASD is a dimensional rather than categorical disorder38,39 and the strict cutoff boundaries defined by the DSM may artificially affect results. Our study also found that toddlers diagnosed as having an LD at their first visit overwhelmingly transitioned into testing within the typical range by 3 to 4 years of age. Such transient LD cases have been commonly noted in the literature.25 Toddlers who exhibited an LD were referred for immediate treatment and generally received 1 to 2 hours per week of speech therapy. Our study was not designed to determine whether such early treatments affected the speed with which toddlers caught up by the time they reached final diagnosis age. Another possibility is that the psychometric test that we used (Mullen Scales of Early Learning) may be less reliable at very young ages.

f speech therapy. Our study was not designed to determine whether such early treatments affected the speed with which toddlers caught up by the time they reached final diagnosis age. Another possibility is that the psychometric test that we used (Mullen Scales of Early Learning) may be less reliable at very young ages. The importance of understanding ASD diagnostic stability in a general population, community-ascertained cohort should not be underestimated, particularly when early screening starting at 18 months is recommended by the American Academy of Pediatrics,40 yet most screening studies9,10,41,42 validate diagnoses only once, usually at approximately the age of screening, leaving unclear the degree to which an initial diagnosis persists at later ages. Although this was not a screening study and the cohort contained approximately 25% community-referred cases, most of the ASD cases were detected using a broadband screening tool, the Communication and Symbolic Behavior Scales Infant-Toddler Checklist,43 administered at well-child visits.

agnosis persists at later ages. Although this was not a screening study and the cohort contained approximately 25% community-referred cases, most of the ASD cases were detected using a broadband screening tool, the Communication and Symbolic Behavior Scales Infant-Toddler Checklist,43 administered at well-child visits. In this study, with a sample size of 1269 toddlers from the general population, each with multiple evaluation visits, generating a total of more than 3000 evaluation visits, we found that 105 (23.8%) who ultimately received a diagnosis of ASD at 3 to 4 years initially had ASD missed at their first evaluation visit. This percentage is substantially lower than the 50% to 80% late-identified ASD cases reported in infant sibling studies.17,18 Among the patients with late-identified ASD, 45 (42.8%) were initially suspected of having only an LD. That is, practitioners focused on a child’s delays in language rather than subclinical social delays. This finding is consistent with an infant sibling study17 that reported lower-than-expected language scores on the Mullen Scales of Early Learning test within their late-identified group.

uspected of having only an LD. That is, practitioners focused on a child’s delays in language rather than subclinical social delays. This finding is consistent with an infant sibling study17 that reported lower-than-expected language scores on the Mullen Scales of Early Learning test within their late-identified group. Limitations One limitation to our study was that the practitioners who made the final diagnoses were not masked to previous diagnoses. This lack of masking was because DSM-5 procedures and criteria require consideration of historical information regarding ASD symptoms. Although unlikely, it is possible that review of this information could have biased psychologists in favor of increased diagnostic stability. Evidence counter to this point is the relatively weak diagnostic stability found in other delay groups. Conclusions Autism spectrum disorder is a common disorder that begins in prenatal life.2 Because of this, there is a demand for early detection, intervention, and services,5 and,in response, marked effort and funding have gone into discovery of methods for early-age universal screening, detection, and diagnosis. Therefore, when ASD is not detected in an infant or toddler, it is likely because it was missed.20 Our findings suggest that ASD detection and diagnosis can reliably start as young as 14 months. Our next challenge is to determine best treatments and the degree to which such early engagement benefits toddlers and their families in the long term. Supplement. eMethods. Supplementary methods

Conclusions Autism spectrum disorder is a common disorder that begins in prenatal life.2 Because of this, there is a demand for early detection, intervention, and services,5 and,in response, marked effort and funding have gone into discovery of methods for early-age universal screening, detection, and diagnosis. Therefore, when ASD is not detected in an infant or toddler, it is likely because it was missed.20 Our findings suggest that ASD detection and diagnosis can reliably start as young as 14 months. Our next challenge is to determine best treatments and the degree to which such early engagement benefits toddlers and their families in the long term. Supplement. eMethods. Supplementary methods eFigure 1. Examination of clinical characteristics between excluded (1 visit) and included (≥2 visits) toddlers eFigure 2. Distribution of diagnostic judgements based on the DSM version eFigure 3. Clinical characteristics of ASD group stratified by identification age and other DX groups eFigure 4. Diagnostic transition tables within each 2-month age band eFigure 5. Diagnostic stability plots by age at first diagnosis eFigure 6. Diagnostic stability after removing toddlers with non-verbal mental age <12 mo eTable 1. Overall logistic regression model eTable 2. Effect of age at first DX on stability coefficients eTable 3. Stability coefficients within 2 months age bands eTable 4. Overall age adjusted and unadjusted stability coefficients eTable 5. Distribution of toddlers with non-verbal age below 12 mo. Across diagnosis groups eTable 6. Between-group difference for all clinical domains eReferences

eTable 2. Effect of age at first DX on stability coefficients eTable 3. Stability coefficients within 2 months age bands eTable 4. Overall age adjusted and unadjusted stability coefficients eTable 5. Distribution of toddlers with non-verbal age below 12 mo. Across diagnosis groups eTable 6. Between-group difference for all clinical domains eReferences Click here for additional data file.

Introduction Childhood and adolescence are vulnerable periods and a crucial window for adult health determination. While improvements in the mortality rate of children younger than 5 years (the population often called under-5) have been undeniably dramatic and positive,1 the full story of child and adolescent health is more nuanced and heterogeneous, with a notably broader range of characteristics than can be told with a single summary statistic.2 The effects of acute and chronic infectious diseases, nutrition, physical functioning, mental health, and intellectual development set the stage for both individual prosperity and the future human capital of all societies.3 Eleven of the 18 Sustainable Development Goals (SDGs) and 19 of the 53 health-associated SDG indicators are about child and adolescent health.4,5 These include ending all forms of malnutrition (SDG 2.2), reducing maternal mortality ratio to fewer than 70 per 100 000 live births (SDG 3.1), decreasing neonatal and under-5 mortality rate to fewer than 12 and 25 per 1000 live births, respectively (SDG 3.2), ensuring universal access to reproductive health care (SDG 3.7), and multiple objectives aimed at combating specific causes of health loss, such as malaria, tuberculosis, HIV, road traffic crashes, air pollution, substance abuse, and noncommunicable diseases (NCDs). However, many of the leading drivers of health loss among children and adolescents are notably absent from the SDG agenda.6

ultiple objectives aimed at combating specific causes of health loss, such as malaria, tuberculosis, HIV, road traffic crashes, air pollution, substance abuse, and noncommunicable diseases (NCDs). However, many of the leading drivers of health loss among children and adolescents are notably absent from the SDG agenda.6 We have compiled this third annual global report to detail the levels, trends, causes, and correlates of health loss from birth through age 19 years. It reflects several notable improvements from Global Burden of Disease (GBD) 2017. First, we have generated a complete set of internally consistent demographics estimates, with uncertainty intervals (UIs), for age-specific fertility, population, and all-cause mortality.7 Second, 5 additional countries (Ethiopia, Iran, New Zealand, Norway, and Russia) were estimated at the subnational level. Third, in addition to adding many new sources of data, we have improved data-processing algorithms. Methods for redistributing deaths coded to nonspecific, implausible, or intermediate causes of death were updated to incorporate statistical uncertainty of cause reassignment. Clinical administrative data (hospital and claims) processing methods were updated to better account for hospital readmissions, multiple clinical visits, and ordering of discharge codes by age, sex, location, and time. Fourth, we have improved the epidemiological transition analysis through improved estimation of the SDI.

nical administrative data (hospital and claims) processing methods were updated to better account for hospital readmissions, multiple clinical visits, and ordering of discharge codes by age, sex, location, and time. Fourth, we have improved the epidemiological transition analysis through improved estimation of the SDI. Methods Comprehensive descriptions of each analytic component of GBD 2017 are detailed elsewhere1,7,8,9,10,11,12 and compliant with the Guidelines for Accurate and Transparent Health Estimates Reporting.13 The GBD 2017 included 11 467 unique sources for cause of death estimation and 26 007 for estimation of nonfatal health loss. Data sources for each cause-level analysis are available online at the Global Health Data Exchange.14

and compliant with the Guidelines for Accurate and Transparent Health Estimates Reporting.13 The GBD 2017 included 11 467 unique sources for cause of death estimation and 26 007 for estimation of nonfatal health loss. Data sources for each cause-level analysis are available online at the Global Health Data Exchange.14 The GBD 2017 used a geographic hierarchy of 7 superregions (high-income countries; Latin America and the Caribbean; North Africa and the Middle East; South Asia; sub-Saharan Africa [SSA]; Central Asia, Central Europe, and Eastern Europe; and Southeast Asia, East Asia, and Oceania) containing 21 regions and 195 countries and territories. Fifteen countries were estimated at the subnational level: Brazil, China, England, Ethiopia, India, Indonesia, Iran, Kenya, Mexico, New Zealand, Norway, the United States, Russia, Sweden, and South Africa. Estimates were produced for male individuals and female individuals separately in each of 23 standard age groups. We cover the first 7 of these age groups in this report: early neonatal (0-6 days’ postbirth age), late neonatal (7-27 days’ postbirth age), postneonatal (28-364 days’ postbirth age), 1 to 4 years, 5 to 9 years, 10 to 14 years, and 15 to 19 years.

als and female individuals separately in each of 23 standard age groups. We cover the first 7 of these age groups in this report: early neonatal (0-6 days’ postbirth age), late neonatal (7-27 days’ postbirth age), postneonatal (28-364 days’ postbirth age), 1 to 4 years, 5 to 9 years, 10 to 14 years, and 15 to 19 years. Each of 359 diseases and injuries were arranged in a 4-level mutually exclusive and collectively exhaustive cause hierarchy; most were analyzed as causing both death and disability. The first level (level 1) of the cause list has 3 categories: communicable, maternal, neonatal, and nutritional conditions (CMNN); NCDs; and injuries. At level 2, there are 22 cause groups, and level 3 includes more disaggregated causes of burden (169 causes), as does level 4 (293 causes). The full GBD cause list, including corresponding International Classification of Diseases, Ninth Revision (ICD-9) and Tenth Revision (ICD-10) codes, is detailed in appendices to the GBD 2017 summary publications.8,9

ps, and level 3 includes more disaggregated causes of burden (169 causes), as does level 4 (293 causes). The full GBD cause list, including corresponding International Classification of Diseases, Ninth Revision (ICD-9) and Tenth Revision (ICD-10) codes, is detailed in appendices to the GBD 2017 summary publications.8,9 All-cause mortality, cause-specific mortality, and years of life lost (YLLs) were estimated using standardized approaches of data identification, extraction, and processing to address data challenges such as incompleteness, variation in classification systems and coding practices, and inconsistent age group and sex reporting. Nonfatal estimates were generated using data from literature, hospital discharge and claims data systems, cross-sectional surveys, cohort studies, case notification systems, and disease-specific registries. Cause-specific years lived with disability (YLDs) were calculated by multiplying sequela-level prevalence with corresponding disability weights that were derived from population and internet surveys of more than 60 000 persons and adjusted for comorbidity through microsimulation.15,16 Disability-adjusted life years (DALYs) are the sum of YLDs and YLLs and used to measure the comprehensive health status of a population for a given location, sex, year, and age combination.

erived from population and internet surveys of more than 60 000 persons and adjusted for comorbidity through microsimulation.15,16 Disability-adjusted life years (DALYs) are the sum of YLDs and YLLs and used to measure the comprehensive health status of a population for a given location, sex, year, and age combination. We sampled 1000 draws of the posterior distribution of quantity at the most granular level of each analysis, and 95% UIs represent the range of values between the ordinal 25th and 975th draws. Unlike confidence intervals, which only capture sampling error in a single statistical test, UIs also incorporate uncertainty from other associated steps. Aggregate estimates (eg, DALYs, combined age groups, geographical groups) were calculated by summing draw-level results assuming independence of each quantity. All draw-level results were summarized as mean values and 95% UIs.

r in a single statistical test, UIs also incorporate uncertainty from other associated steps. Aggregate estimates (eg, DALYs, combined age groups, geographical groups) were calculated by summing draw-level results assuming independence of each quantity. All draw-level results were summarized as mean values and 95% UIs. We performed 3 secondary analyses for this report. First, we decomposed probability of death from birth to 19 years to illustrate how cause-specific trends are associated with overall survival improvements. Second, we explored the historical association between burden metrics and the SDI, a composite indicator of development based on per capita income, adult education, and total fertility rate for individuals younger than 25 years.1 Each GBD location’s SDI can vary by year, but for reporting purposes, each was assigned to an quintile based on its SDI in 2017. A map of SDI quintile assignments is shown in eFigure 1 in the Supplement and SDI values for each country by year are in eTable 1 in the Supplement. Observed values are the actual disease burden rates in each location-year, while expected values were determined by Gaussian process regression on the range of rates observed for each level of SDI. Third, given the intricate association between the health of women and their children, we examined the historical association between maternal mortality and DALY rates of children and adolescents.

while expected values were determined by Gaussian process regression on the range of rates observed for each level of SDI. Third, given the intricate association between the health of women and their children, we examined the historical association between maternal mortality and DALY rates of children and adolescents. We present a number of different formulations of results in the GBD 2017. Total number illustrates the cumulative size of burden, rates best compare between differently sized populations, and cause fraction (%) allows the comparison of relative importance of specific causes. We refer to those younger than 28 days as neonates, those younger than 1 year as infants, those younger than 10 years collectively as children, and those aged 10 to 19 years as adolescents. We focus on presenting aggregate results for the global level, SDI quintiles, and the GBD regions, either for birth to 19 years en bloc or for infants, children, and adolescents separately. Except when noted, results are for both sexes combined. More granular results are publicly available in an interactive online visualization tool called GBD Compare (https://vizhub.healthdata.org/gbd-compare/) and for download from the GBD Results Tool (http://ghdx.healthdata.org/gbd-results-tool).

olescents separately. Except when noted, results are for both sexes combined. More granular results are publicly available in an interactive online visualization tool called GBD Compare (https://vizhub.healthdata.org/gbd-compare/) and for download from the GBD Results Tool (http://ghdx.healthdata.org/gbd-results-tool). Results All-Cause Mortality and Decomposition of Causes of Death Premature mortality is the dominant component of health loss in children and adolescents. The Table shows deaths by age group globally and by SDI quintile. eTable 2 in the Supplement shows the same for superregions, regions, countries, and territories. All-cause child and adolescent deaths decreased 51.7% from 13.77 million (95% UI, 13.60–13.93 million) in 1990 to 6.64 million (95% UI, 6.44-6.87 million) in 2017. More than half (60.1% [95% UI, 59.6%-60.5%]) occurred in infants younger than 1 year and, of those, 46.6% (95% UI, 46.0%-47.3%) occurred in the first week of life. The fastest decline was among children aged 1 to 4 years, in whom global deaths decreased by 61% from 3.62 million (95% UI, 3.52-3.72 million) in 1990 to 1.40 million (95% UI, 1.34-1.48 million) in 2017. Over the same period, mortality decreased 51% to 3.99 million (95% UI, 3.85-4.14 million) in infants younger than 1 year, by 52% to 0.41 million (95% UI, 0.40-0.42 million) in children aged 5 to 9 years, and by 27% to 0.84 million (95% UI, 0.82-0.85 million) in children aged 10 to 19 years. Improvements by age were similar across SDI quintiles.

ality decreased 51% to 3.99 million (95% UI, 3.85-4.14 million) in infants younger than 1 year, by 52% to 0.41 million (95% UI, 0.40-0.42 million) in children aged 5 to 9 years, and by 27% to 0.84 million (95% UI, 0.82-0.85 million) in children aged 10 to 19 years. Improvements by age were similar across SDI quintiles. Table. All-Cause Mortality in 1990, 2000, and 2017, With Mean Percentage Changes for Combined Sexes by Age Metric Individuals, No. (95% Uncertainty Intervals) Early Neonatala Late Neonatalb Postneonatalc Aged 1 to 4 y Aged 5 to 9 y Aged 10 to 14 y Aged 15 to 19 y <1 y <5 y <20 y Aged 10 to 19 y Global Deaths, 1990, No. 3 037 044 (2 966 409- 3 104 943) 1 263 098 (1 220 453- 1 308 864) 3 854 632 (3 771 297- 3 942 560) 3 615 809 (3 519 278- 3 715 515) 853 941 (844 956- 862 845) 466 727 (462 844- 470 900) 674 033 (666 858- 681 883) 8 154 774 (8 024 357- 8 285 234) 11 770 583 (11 618 335- 11 926 603) 13 765 285 (13 601 506- 13 934 158) 1 140 760 (1 130 272- 1 151 953) Deaths, 2000, No. 2 743 663 (2 683 053- 2 807 354) 918 145 (887 698- 954 872) 3 092 086 (3 030 247- 3 163 815) 2 926 972 (2 853 818- 3 007 898) 703 302 (696 132- 711 149) 483 735 (479 703- 487 837) 678 423 (670 973- 686 282) 6 753 895 (6 641 564- 6 883 034) 9 680 867 (9 545 127- 9 822 519) 11 546 328 (11 402 470- 11 696 143) 1 162 159 (1 151 054- 1 173 659) Deaths, 2017, No. 1 859 529 (1 793 656- 1 933 838) 507 698 (489 055- 527 088) 1 621 135 (1 558 657- 1 693 323) 1 403 200 (1 340 221- 1 475 816) 412 113 (404 482- 420 010) 319 630 (314 713- 325 138) 516 509 (506 942- 527 177) 3 988 362 (3 847 837- 4 140 124) 5 391 562 (5 195 363- 5 612 906) 6 639 815 (6 437 215- 6 870 058) 836 139 (822 746- 851 197) % Change, 1990-2017 −38.8 (−41.5 to −35.9) −59.8 (−61.8 to −57.6) −57.9 (−59.9 to −55.8) −61.2 (−63.3 to −59.0) −51.7 (−52.8 to −50.7) −31.5 (−32.6 to −30.3) −23.4 (−24.9 to −21.7) −51.1 (−53.1 to −49.1) −54.2 (−56.0 to −52.3) −51.8 (−53.4 to −50.0) −26.7 (−28.0 to −25.3) % Change, 2000-2017 −32.2 (−35.1 to −29.1) −44.7 (−47.5 to −41.7) −47.6 (−49.9 to −45.0) −52.1 (−54.5 to −49.4) −41.4 (−42.7 to −40.1) −33.9 (−35.0 to −32.9) −23.9 (−25.2 to −22.5) −41.0 (−43.2 to −38.3) −44.3 (−46.4 to −41.9) −42.5 (−44.4 to −40.3) −28.1 (−29.2 to −26.9) Low SDI Deaths, 1990, No.

26.7 (−28.0 to −25.3) % Change, 2000-2017 −32.2 (−35.1 to −29.1) −44.7 (−47.5 to −41.7) −47.6 (−49.9 to −45.0) −52.1 (−54.5 to −49.4) −41.4 (−42.7 to −40.1) −33.9 (−35.0 to −32.9) −23.9 (−25.2 to −22.5) −41.0 (−43.2 to −38.3) −44.3 (−46.4 to −41.9) −42.5 (−44.4 to −40.3) −28.1 (−29.2 to −26.9) Low SDI Deaths, 1990, No. 1 048 756 (1 000 660- 1 099 967) 499 125 (462 152- 541 079) 1 407 561 (1 351 942- 1 473 057) 1 649 664 (1 571 877- 1 744 345) 294 379 (288 394- 300 299) 134 914 (132 405- 137 132) 173 480 (168 538- 178 248) 2 955 442 (2 859 649- 3 044 743) 4 605 107 (4 498 615- 4 697 610) 5 207 880 (5 094 452- 5 309 147) 308 395 (301 354- 315 308) % of Total, 1990 34.5 (33.4-35.8) 39.5 (37.6-41.4) 36.5 (35.5-37.7) 45.6 (44.3-47.2) 34.5 (34.0-35.0) 28.9 (28.5-29.3) 25.7 (25.1-26.3) 36.2 (35.4-37.0) 39.1 (38.5-39.7) 37.8 (37.2-38.4) 27.0 (26.5-27.5) Deaths, 2000, No. 1 047 243 (1 006 123- 1 091 383) 377 909 (352 083- 410 913) 1 312 429 (1 273 115- 1 357 139) 1 433 329 (1 378 745- 1 493 988) 269 317 (263 349- 274 995) 152 533 (150 081- 154 909) 192 073 (188 328- 195 751) 2 737 581 (2 658 104- 2 824 273) 4 170 910 (4 078 681- 4 267 250) 4 784 833 (4 684 232- 4 887 154) 344 605 (338 629- 350 377) % of Total, 2000 38.2 (37.0-39.3) 41.2 (39.3-43.4) 42.4 (41.4-43.6) 49.0 (47.7-50.2) 38.3 (37.7-38.9) 31.5 (31.1-31.9) 28.3 (27.9-28.8) 40.5 (39.6-41.4) 43.1 (42.4-43.8) 41.4 (40.8-42.1) 29.6 (29.2-30.1) Deaths, 2017, No.

273) 4 170 910 (4 078 681- 4 267 250) 4 784 833 (4 684 232- 4 887 154) 344 605 (338 629- 350 377) % of Total, 2000 38.2 (37.0-39.3) 41.2 (39.3-43.4) 42.4 (41.4-43.6) 49.0 (47.7-50.2) 38.3 (37.7-38.9) 31.5 (31.1-31.9) 28.3 (27.9-28.8) 40.5 (39.6-41.4) 43.1 (42.4-43.8) 41.4 (40.8-42.1) 29.6 (29.2-30.1) Deaths, 2017, No. 799 558 (756 228- 844 647) 216 363 (204 706- 228 297) 764 529 (731 080- 800 527) 697 767 (666 056- 730 782) 170 474 (165 388- 175 721) 116 484 (113 419- 119 513) 164 477 (160 277- 168 964) 1 780 450 (1 700 258- 1 869 233) 2 478 217 (2 374 041- 2 592 363) 2 929 653 (2 817 698- 3 050 894) 280 962 (274 764- 287 360) % of Total, 2017 43.0 (41.2-44.9) 42.6 (40.9-44.5) 47.2 (45.1-49.2) 49.7 (47.4-52.1) 41.4 (40.5-42.4) 36.4 (35.7-37.2) 31.8 (31.1-32.6) 44.6 (42.8-46.4) 46.0 (44.1-47.7) 44.1 (42.5-45.6) 33.6 (32.9-34.3) % Change, 1990-2017 −23.8 (−29.0 to −17.6) −56.6 (−60.4 to −52.0) −45.7 (−49.1 to −42.2) −57.7 (−60.8 to −54.4) −42.1 (−44.1 to −39.9) −13.7 (−16.3 to −10.7) −5.2 (−8.6 to −1.6) −39.8 (−43.0 to −36.1) −46.2 (−48.6 to −43.4) −43.8 (−46.1 to −41.2) −8.9 (−11.8 to −6.0) % Change, 2000-2017 −23.6 (−28.8 to −17.7) −42.8 (−48.2 to −37.3) −41.8 (−44.8 to −38.6) −51.3 (−54.2 to −48.2) −36.7 (−39.2 to −34.3) −23.6 (−26.0 to −21.3) −14.4 (−16.9 to −12.0) −35.0 (−38.5 to −31.2) −40.6 (−43.4 to −37.4) −38.8 (−41.5 to −35.7) −18.5 (−20.6 to −16.4) Low-middle SDI Deaths, 1990, No. 1 050 591 (1 005 227- 1 101 086) 420 295 (402 095- 440 764) 1 231 685 (1 187 111- 1 281 266) 1 241 777 (1 188 516- 1 291 581) 252 941 (248 961- 256 718) 135 510 (133 125- 137 988) 191 154 (186 315- 196 539) 2 702 571 (2 635 701- 2 772 458) 3 944 348 (3 854 061- 4 037 247) 4 523 953 (4 433 296- 4 620 989) 326 664 (319 584- 334 446) % of Total, 1990 34.6 (33.4-35.9) 33.3 (31.9-34.8) 31.9 (31.0-33.0) 34.3 (33.0-35.6) 29.6 (29.2-30.0) 29.0 (28.6-29.5) 28.4 (27.8-29.0) 33.1 (32.5-33.9) 33.5 (33.0-34.1) 32.9 (32.4-33.5) 28.6 (28.1-29.2) Deaths, 2000, No. 1 027 894 (985 060- 1 072 304) 337 892 (322 334- 353 887) 1 103 094 (1 057 458- 1 152 018) 1 111 181 (1 061 360- 1 162 163) 226 250 (222 741- 229 883) 145 940 (143 053- 148 953) 207 612 (201 623- 214 285) 2 468 880 (2 386 134- 2 550 868) 3 580 061 (3 482 534-

.0-34.1) 32.9 (32.4-33.5) 28.6 (28.1-29.2) Deaths, 2000, No. 1 027 894 (985 060- 1 072 304) 337 892 (322 334- 353 887) 1 103 094 (1 057 458- 1 152 018) 1 111 181 (1 061 360- 1 162 163) 226 250 (222 741- 229 883) 145 940 (143 053- 148 953) 207 612 (201 623- 214 285) 2 468 880 (2 386 134- 2 550 868) 3 580 061 (3 482 534- 3 681 176) 4 159 863 (4 056 664- 4 267 343) 353 552 (344 879- 363 321) % of total, 2000 37.5 (36.2-38.7) 36.8 (35.0-38.4) 35.7 (34.6-36.8) 38.0 (36.6-39.3) 32.2 (31.7-32.6) 30.2 (29.7-30.6) 30.6 (30.0-31.3) 36.5 (35.6-37.5) 37.0 (36.2-37.7) 36.0 (35.4-36.7) 30.4 (29.9-31.0) Deaths, 2017, No. 741 928 (687 548- 799 669) 201 579 (187 087- 217 116) 598 744 (546 451- 659 900) 556 133 (501 195- 617 348) 143 064 (137 522- 148 843) 109 895 (106 336- 114 277) 171 571 (163 916- 180 817) 1 542 251 (1 428 045- 1 667 161) 2 098 384 (1 931 802- 2 278 025) 2 522 915 (2 349 624- 2 716 084) 281 467 (270 507- 294 587) % of Total, 2017 39.9 (37.9-42.0) 39.7 (37.7-41.8) 36.9 (34.6-39.4) 39.6 (36.9-42.2) 34.7 (33.7-35.7) 34.4 (33.6-35.3) 33.2 (32.1-34.4) 38.7 (36.7-40.7) 38.9 (36.9-41.0) 38.0 (36.2-39.8) 33.7 (32.7-34.7) % Change, 1990-2017 −29.4 (−35.7 to −22.9) −52.0 (−56.3 to −47.8) −51.4 (−56.1 to −45.6) −55.2 (−59.9 to −50.0) −43.4 (−45.8 to −41.0) −18.9 (−21.6 to −15.7) −10.2 (−14.5 to −5.3) −42.9 (−47.3 to −38.1) −46.8 (−50.9 to −42.1) −44.2 (−48.0 to −39.8) −13.8 (−17.3 to −9.6) % Change, 2000-2017 −27.8 (−34.1 to −21.1 −40.3 (−45.4 to −34.9) −45.7 (−51.3 to −39.6) −50.0 (−55.1 to −44.0) −36.8 (−39.5 to −34.0) −24.7 (−27.0 to −22.1) −17.4 (−20.8 to −13.5) −37.5 (−42.4 to −32.3) −41.4 (−46.1 to −36.1) −39.4 (−43.7 to −34.5) −20.4 (−23.3 to −17.1) Middle SDI Deaths, 1990, No. 628 643 (610 187- 646 947) 238 799 (230 555- 247 253) 868 238 (837 395- 901 662) 537 299 (517 549- 556 637) 202 488 (198 125- 206 891) 121 109 (119 519- 122 793) 173 289 (170 735- 175 772) 1 735 679 (1 683 546- 1 790 834) 2 272 978 (2 208 184- 2 340 940) 2 769 865 (2 698 798- 2 842 310) 294 398 (290 461- 298 249) % of Total, 1990 20.7 (20.0-21.4) 18.9 (18.1-19.8) 22.5 (21.8-23.3) 14.9 (14.2-15.4) 23.7 (23.3-24.1) 25.9 (25.6-26.3) 25.7 (25.3-26.1) 21.3 (20.7-21.9) 19.3 (18.8-19.8) 20.1 (19.6-20.6) 25.8 (25.4-26.2) Deaths, 2000, No.

834) 2 272 978 (2 208 184- 2 340 940) 2 769 865 (2 698 798- 2 842 310) 294 398 (290 461- 298 249) % of Total, 1990 20.7 (20.0-21.4) 18.9 (18.1-19.8) 22.5 (21.8-23.3) 14.9 (14.2-15.4) 23.7 (23.3-24.1) 25.9 (25.6-26.3) 25.7 (25.3-26.1) 21.3 (20.7-21.9) 19.3 (18.8-19.8) 20.1 (19.6-20.6) 25.8 (25.4-26.2) Deaths, 2000, No. 462 225 (451 844- 472 926) 139 884 (135 990- 143 999) 493 464 (480 843- 506 778) 278 288 (269 324- 287 279) 141 272 (138 903- 143 669) 117 620 (116 463- 118 791) 157 626 (155 839- 159 372) 1 095 573 (1 073 483- 1 120 170) 1 373 862 (1 345 819- 1 401 556) 1 790 380 (1 759 629- 1 820 258) 275 246 (272 638- 277 738) % of Total, 2000 16.9 (16.4-17.3) 15.2 (14.6-15.9) 16.0 (15.5-16.4) 9.5 (9.1-9.9) 20.1 (19.7-20.4) 24.3 (24.0-24.6) 23.2 (22.9-23.6) 16.2 (15.8-16.6) 14.2 (13.9-14.5) 15.5 (15.2-15.8) 23.7 (23.4-23.9) Deaths, 2017, No. 222 226 (214 317- 230 498) 60 514 (58 427- 62 842) 183 100 (174 148- 193 196) 104 767 (100 420- 109 319) 68 791 (67 224- 70 555) 64 108 (63 168- 65 105) 116 098 (113 774- 118 242) 465 841 (449 615- 483 118) 570 608 (551 051- 591 996) 819 604 (797 304- 842 529) 180 206 (177 373- 182 959) % of Total, 2017 11.9 (11.4-12.6) 11.9 (11.3-12.5) 11.3 (10.7-12.1) 7.5 (7.0-7.9) 16.7 (16.2-17.1) 20.1 (19.6-20.5) 22.5 (21.9-23.0) 11.7 (11.2-12.2) 10.6 (10.1-11.1) 12.3 (11.9-12.8) 21.6 (21.1-22.0) % Change, 1990-2017 −64.7 (−66.2 to −62.9) −74.7 (−75.8 to −73.3) −78.9 (−80.3 to −77.4) −80.5 (−81.5 to −79.3) −66.0 (−67.2 to −64.7) −47.1 (−48.2 to −45.9) −33.0 (−34.6 to −31.4) −73.2 (−74.4 to −71.8) −74.9 (−76.0 to −73.7) −70.4 (−71.5 to −69.2) −38.8 (−40.1 to −37.5) % Change, 2000-2017 −51.9 (−53.9 to −49.8) −56.7 (−58.6 to −54.6) −62.9 (−64.9 to −60.5) −62.4 (−64.2 to −60.4) −51.3 (−52.7 to −49.8) −45.5 (−46.5 to −44.5 −26.4 (−27.9 to −24.7) −57.5 (−59.0 to −55.6) −58.5 (−60.0 to −56.7) −54.2 (−55.6 to −52.7) −34.5 (−35.7 to −33.3) High-middle SDI Deaths, 1990, No. 251 380 (243 625- 259 229) 88 124 (85 426- 91 122) 294 385 (285 064- 304 510) 156 447 (149 526- 163 923) 85 156 (83 615- 86 820) 57 768 (57 072- 58 454) 86 847 (85 614- 88 149) 633 890 (616 360- 653 955) 790 337 (768 924- 813 197) 1 020 107 (996 550- 1 045 005) 144 615 (142 872- 146 484) % of Total, 1990 8.3 (8.0-8.6) 7.0 (6.7-7.3) 7.6 (7.3-7.9) 4.3 (4.1-4.5) 10.0 (9.8-10.2) 12.4 (12.2-12.5) 12.9 (12.7-13.1) 7.8 (7.5-8.1) 6.7 (6.5-6.9) 7.4 (7.2-7.6) 12.7 (12.5-12.9) Deaths, 2000, No.

88 149) 633 890 (616 360- 653 955) 790 337 (768 924- 813 197) 1 020 107 (996 550- 1 045 005) 144 615 (142 872- 146 484) % of Total, 1990 8.3 (8.0-8.6) 7.0 (6.7-7.3) 7.6 (7.3-7.9) 4.3 (4.1-4.5) 10.0 (9.8-10.2) 12.4 (12.2-12.5) 12.9 (12.7-13.1) 7.8 (7.5-8.1) 6.7 (6.5-6.9) 7.4 (7.2-7.6) 12.7 (12.5-12.9) Deaths, 2000, No. 168 748 (163 067- 174 696) 50 722 (48 938- 52 476) 152 970 (147 699- 158 189) 85 695 (81 187- 90 292) 54 832 (53 762- 55 920) 54 631 (53 990- 55 338) 84 218 (83 489- 85 016) 372 439 (361 974- 383 119) 458 135 (445 459- 471 978) 651 816 (637 814- 666 613) 138 849 (137 574- 140 224) % of Total, 2000 6.2 (5.9-6.4) 5.5 (5.2-5.8) 5.0 (4.8-5.1) 2.9 (2.8-3.1) 7.8 (7.6-8.0) 11.3 (11.1-11.4) 12.4 (12.2-12.6) 5.5 (5.3-5.7) 4.7 (4.6-4.9) 5.7 (5.5-5.8) 11.9 (11.8-12.1) Deaths, 2017, No. 69 837 (66 696- 73 005) 21 430 (20 560- 22 308) 56 477 (53 589- 59 480) 33 866 (32 037- 35 979) 23 303 (22 679- 23 920) 21 805 (21 438- 22 160) 42 409 (41 619- 43 208) 147 743 (141 663- 153 908) 181 609 (174 033- 188 764) 269 126 (260 627- 277 283) 64 214 (63 108- 65 265) % of Total, 2017 3.8 (3.5-4.0) 4.2 (4.0-4.5) 3.5 (3.3-3.7) 2.4 (2.2-2.6) 5.7 (5.5-5.8) 6.8 (6.7-7.0) 8.2 (8.0-8.4) 3.7 (3.5-3.9) 3.4 (3.2-3.5) 4.0 (3.9-4.2) 7.7 (7.5-7.8) % Change, 1990-2017 −72.2 (−73.7 to −70.7) −75.7 (−76.9 to −74.4) −80.8 (−82.0 to −79.6) −78.3 (−79.8 to −76.8) −72.6 (−73.4 to −71.9) −62.2 (−63.0 to −61.5) −51.2 (−52.3 to −50.0) −76.7 (−77.8 to −75.5) −77.0 (−78.1 to −76.0) −73.6 (−74.6 to −72.7) −55.6 (−56.5 to −54.7) % Change, 2000-2017 −58.6 (−60.7 to −56.2) −57.8 (−59.9 to −55.5) −63.1 (−65.3 to −60.6 −60.5 (−63.4 to −57.4) −57.5 (−58.8 to −56.2) −60.1 (−60.9 to −59.3) −49.6 (−50.8 to −48.6) −60.3 (−62.2 to −58.4) −60.4 (−62.2 to −58.5) −58.7 (−60.2 to −57.2) −53.8 (−54.7 to −52.8) High SDI Deaths, 1990, No. 51 138 (50 076-52 156) 14 066 (13 636-14 520) 40 672 (39 568-41 787) 23 862 (23 087-24 712) 16 274 (15 946-16 603) 15 870 (15 745-15 997) 47 148 (47 037-47 267) 105 875 (104 730- 107 028) 129 737 (128 508- 131 047) 209 029 (207 361- 210 885) 63 018 (62 791- 63 243) % of Total, 1990 1.7 (1.6-1.7) 1.1 (1.1-1.2) 1.1 (1.0-1.1) 0.7 (0.6-0.7) 1.9 (1.9-1.9) 3.4 (3.4-3.4) 7.0 (6.9-7.1) 1.3 (1.3-1.3) 1.1 (1.1-1.1) 1.5 (1.5-1.5) 5.5 (5.5-5.6) Deaths, 2000, No.

97) 47 148 (47 037-47 267) 105 875 (104 730- 107 028) 129 737 (128 508- 131 047) 209 029 (207 361- 210 885) 63 018 (62 791- 63 243) % of Total, 1990 1.7 (1.6-1.7) 1.1 (1.1-1.2) 1.1 (1.0-1.1) 0.7 (0.6-0.7) 1.9 (1.9-1.9) 3.4 (3.4-3.4) 7.0 (6.9-7.1) 1.3 (1.3-1.3) 1.1 (1.1-1.1) 1.5 (1.5-1.5) 5.5 (5.5-5.6) Deaths, 2000, No. 33 111 (32 535-33 688) 10 224 (9938-10 536) 23 583 (22 957-24 177) 14 817 (14 389-15 295) 9693 (9641-9746) 11 387 (11 347-11 430) 35 359 (35 294-35 431) 66 918 (66 183-67 627) 81 735 (81 099-82 476) 138 175 (137 405-138 996) 46 747 (46 647-46 849) % of Total, 2000 1.2 (1.2-1.2) 1.1 (1.1-1.2) 0.8 (0.7-0.8) 0.5 (0.5-0.5) 1.4 (1.4-1.4) 2.4 (2.3-2.4) 5.2 (5.2-5.3) 1.0 (1.0-1.0) 0.8 (0.8-0.9) 1.2 (1.2-1.2) 4.0 (4.0-4.1) Deaths, 2017, No. 24 050 (23 335-24 802) 7181 (6937-7450) 16 108 (15 624-16 662) 9390 (9153-9642) 5757 (5681-5836) 6702 (6640-6765) 21 083 (20 783-21 392) 47 340 (45 953-48 790) 56 729 (55 145-58 397) 90 272 (88 442-92 159) 27 786 (27 447-28 137) % of Total, 2017 1.3 (1.2-1.4) 1.4 (1.3-1.5) 1.0 (0.9-1.1) 0.7 (0.6-0.7) 1.4 (1.4-1.4) 2.1 (2.1-2.1) 4.1 (4.0-4.2) 1.2 (1.1-1.2) 1.1 (1.0-1.1) 1.4 (1.3-1.4) 3.3 (3.2-3.4) % Change, 1990-2017 −53.0 (−54.6 to −51.1) −48.9 (−51.3 to −46.3) −60.4 (−62.1 to −58.8) −60.6 (−62.1 to −58.9) −64.6 (−65.4 to −63.8) −57.8 (−58.3 to −57.3) −55.3 (−55.9 to −54.6) −55.3 (−56.6 to −53.9) −56.3 (−57.5 to −54.9) −56.8 (−57.7 to −55.9) −55.9 (−56.5 to −55.3) % Change, 2000-2017 −27.4 (−29.9 to −24.9 −29.8 (−33.1 to −26.3) −31.7 (−34.3 to −28.8) −36.6 (−39.1 to −34.0) −40.6 (−41.5 to −39.8) −41.1 (−41.8 to −40.5) −40.4 (−41.2 to −39.5) −29.3 (−31.5 to −27.0) −30.6 (−32.6 to −28.5) −34.7 (−36.0 to −33.3) −40.6 (−41.3 to −39.8) Abbreviation: SDI, Socio-Demographic Index.

5.3) % Change, 2000-2017 −27.4 (−29.9 to −24.9 −29.8 (−33.1 to −26.3) −31.7 (−34.3 to −28.8) −36.6 (−39.1 to −34.0) −40.6 (−41.5 to −39.8) −41.1 (−41.8 to −40.5) −40.4 (−41.2 to −39.5) −29.3 (−31.5 to −27.0) −30.6 (−32.6 to −28.5) −34.7 (−36.0 to −33.3) −40.6 (−41.3 to −39.8) Abbreviation: SDI, Socio-Demographic Index. a Early neonatal includes individuals aged 0 to 6 days. b Late neonatal includes individuals aged 7 to 27 days. c Postneonatal includes individuals aged 28 to 364 days.

5.3) % Change, 2000-2017 −27.4 (−29.9 to −24.9 −29.8 (−33.1 to −26.3) −31.7 (−34.3 to −28.8) −36.6 (−39.1 to −34.0) −40.6 (−41.5 to −39.8) −41.1 (−41.8 to −40.5) −40.4 (−41.2 to −39.5) −29.3 (−31.5 to −27.0) −30.6 (−32.6 to −28.5) −34.7 (−36.0 to −33.3) −40.6 (−41.3 to −39.8) Abbreviation: SDI, Socio-Demographic Index. a Early neonatal includes individuals aged 0 to 6 days. b Late neonatal includes individuals aged 7 to 27 days. c Postneonatal includes individuals aged 28 to 364 days. Decomposition of changes in probability of death between birth and age 20 years from 1990 to 2017 revealed different level 2 cause-level drivers across GBD regions (Figure 1; country results are in eFigure 2 in the Supplement). Decreases in deaths owing to infectious diseases, neonatal disorders, and unintentional injuries drove improvements at the global level and for many less-developed regions (eg, CMNN deaths were virtually absent in high-SDI regions). In western, central, and eastern SSA, the probability of surviving to adulthood increased from 1990 to 2017 (western SSA: 1990, 78.8%; 2017, 89.1%; central SSA: 1990, 80.5%; 2017, 90.7%; eastern SSA: 1990, 80.0%; 2017, 92.1%), primarily as a result of decreased mortality owing to respiratory infections (percentage of decreased mortality owing to this change: western SSA, 20.4%; central SSA, 19.1%; eastern SSA, 17.9%), enteric infections (western SSA, 24.6%; central SSA, 13.3%; eastern SSA, 13.7%), neglected tropical diseases and malaria (western SSA, 11.9%; central SSA, 22.6%; eastern SSA, 22.6%), other infectious diseases (western SSA, 25.0%; central SSA, 16.7%; eastern SSA, 19.6%), and nutritional deficiencies (western SSA, 6.4%; central SSA, 8.5%; eastern SSA, 7.1%). The total decrease in mortality from these causes was 88.4% in western SSA, 80.3% in central SSA, and 81.0% in eastern SSA.

eastern SSA, 22.6%), other infectious diseases (western SSA, 25.0%; central SSA, 16.7%; eastern SSA, 19.6%), and nutritional deficiencies (western SSA, 6.4%; central SSA, 8.5%; eastern SSA, 7.1%). The total decrease in mortality from these causes was 88.4% in western SSA, 80.3% in central SSA, and 81.0% in eastern SSA. Figure 1. Decomposition of the Probability of Death Globally and in the Socio-Demographic Index (SDI) Quintiles and Global Burden of Disease Regions for Individuals Younger Than 20 Years of Both Sexes, From 1990 to 2017 Probability of death is plotted globally by SDI quintile and Global Disease Burden study region for 1990 (dashed vertical line) and 2017 (solid vertical line). The relative contribution of that change owing to different causes is indicated by different color bars. The SDI quintiles are sorted from lowest probability of death in 2017 to highest, and GBD regions are sorted from highest probability of death in 2017 to lowest.

dashed vertical line) and 2017 (solid vertical line). The relative contribution of that change owing to different causes is indicated by different color bars. The SDI quintiles are sorted from lowest probability of death in 2017 to highest, and GBD regions are sorted from highest probability of death in 2017 to lowest. Decreased mortality from other NCDs (primarily congenital birth defects and hemoglobinopathies) and neonatal disorders contributed the most to survival improvements in most of high-middle–SDI and high-SDI regions (decrease in death rate, 1990-2017, for congenital birth defects: high-SDI countries, 52.6% [95% UI, 41.6%-55.8%]; high-middle–SDI countries, 59.4% [95% UI, 53.1%-64.8%]; hemoglobinopathies: high-SDI countries, 54.1% [95% UI, 38.8%-60.0%]; high-middle–SDI countries, 61.3% [95% UI, 49.1%-68.7%]; neonatal disorders: high-SDI countries, 45.5% [95% UI, 42.3%-50.4%]; high-middle–SDI countries, 65.0% [95% UI, 61.3%-68.4%]). Exceptions to broad survival improvements included a 0.6% increased probability of death owing to HIV/AIDS and sexually transmitted infections (1990, 0.6%; 2017, 1.2%; death rate per 100 000 population) in individuals younger than 20 years from HIV/AIDS and STI: 1990, 33.6 [95% UI, 22.4-46.7]; 2017, 57.2 [95% UI, 49.5-65.9]) in Southern sub-Saharan Africa.

nts included a 0.6% increased probability of death owing to HIV/AIDS and sexually transmitted infections (1990, 0.6%; 2017, 1.2%; death rate per 100 000 population) in individuals younger than 20 years from HIV/AIDS and STI: 1990, 33.6 [95% UI, 22.4-46.7]; 2017, 57.2 [95% UI, 49.5-65.9]) in Southern sub-Saharan Africa. There were a total of 50 countries where the probability of death by self-harm and interpersonal violence increased between 1990 and 2017. Nine of them had increase of more than 0.1% in the overall probability of death owing to self-harm and interpersonal violence between birth and age 20 years: Syria (1990, 0.04%; 2017, 3.45%), Iraq (1990, 0.26%; 2017, 1.22%), Yemen (1990, 0.07%; 2017, 0.89%), Central African Republic (1990, 0.1%; 2017, 0.64%), South Sudan (1990, 0.35%; 2017, 0.73%), Libya (1990, 0.05%; 2017, 0.41%), Venezuela (1990, 0.19%; 2017, 0.54%), Mexico (1990, 0.17%; 2017, 0.29%), and Lesotho (1990, 0.25%; 2017, 0.35%).

2017, 3.45%), Iraq (1990, 0.26%; 2017, 1.22%), Yemen (1990, 0.07%; 2017, 0.89%), Central African Republic (1990, 0.1%; 2017, 0.64%), South Sudan (1990, 0.35%; 2017, 0.73%), Libya (1990, 0.05%; 2017, 0.41%), Venezuela (1990, 0.19%; 2017, 0.54%), Mexico (1990, 0.17%; 2017, 0.29%), and Lesotho (1990, 0.25%; 2017, 0.35%). Temporal and Sociodemographic Trends in DALYs Total DALYs in children and adolescents decreased by 46% from 1.31 billion (95% UI, 1.27-1.36 billion) in 1990 to 709 million (95% UI, 665-757 million) in 2017 (Figure 2). Absolute DALYs decreases were greatest in the low-SDI quintile (1990, 476 million [95% UI, 464-490 million]; 2017, 292 million [95% UI, 277-309 million]), but slower relative gains led to increased inequity as the proportion of global DALYs rose from 36% in 1990 to 41% in 2017. The CMNN causes were by far the largest contributor to DALYs in the low-SDI quintile (75.3% of total DALYs [95% UI, 73.5%-77.0%]), low-middle–SDI quintile (71.6% of total DALYs [95% UI, 69.6%-73.6%]), and middle-SDI quintile (49.8% of total DALYs [95% UI, 47.6%-52.0%]) but also posted the biggest improvements, decreasing globally by 52.3% from 992 million (95% UI, 962 million-1.025 billion) to 473 million (95% UI, 452-498 million) in 2017, including a 72.3% reduction in the middle-SDI quintile from 179 million (95% UI, 172-187 million) to 49.5 million (95% UI, 46.5-52.9 million) in 2017. The NCD DALY rates were relatively similar across SDI quintiles and had the slowest global decrease (17.1%), from 210 million (95% UI, 186-238 million) to 174 million (95% UI, 151-201 million) in 2017. The NCDs were the bulk of DALYs in high-SDI locations (63.8% [95% UI, 60.1%-67.3%]), while the high-middle–SDI quintile transitioned in 2000 to also having the largest proportion of child and adolescent DALYs owing to NCDs (1990, 32.0% [95% UI, 29.9%-34.3%]; 2000, 39.2% [95% UI, 36.7%-41.8%]; 2017, 47.7% [95% UI, 44.7%-50.8%]). Global DALYs owing to injuries decreased by 46.0% from 113 million (95% UI, 101-122 million) to 61.0 million (95% UI, 58.0-64.1 million) in 2017 but also fluctuated widely by country and region, mainly because of war and (notably) the 1994 Rwandan genocide, 2004 tsunami in Southeast Asia, the 2010 earthquake in Haiti, and other recent conflicts and disasters.

by 46.0% from 113 million (95% UI, 101-122 million) to 61.0 million (95% UI, 58.0-64.1 million) in 2017 but also fluctuated widely by country and region, mainly because of war and (notably) the 1994 Rwandan genocide, 2004 tsunami in Southeast Asia, the 2010 earthquake in Haiti, and other recent conflicts and disasters. Figure 2. Trends of Disability-Adjusted Life Years From 1990 to 2017 for Global and Socio-Demographic Index (SDI) Quintiles for Children and Adolescents Younger Than 20 Years Temporal trends in disability adjusted life-years (DALYs) are plotted for children and adolescents younger than 20 years. Global trends are plotted in the top left subpanel, and the corresponding trends for each SDI quintile are plotted in the next 5 subpanels. Shaded areas show 95% uncertainty intervals. Communicable, maternal, neonatal, and nutritional disorders are shown in orange, noncommunicable disease causes in blue, and injuries in gray.

trends are plotted in the top left subpanel, and the corresponding trends for each SDI quintile are plotted in the next 5 subpanels. Shaded areas show 95% uncertainty intervals. Communicable, maternal, neonatal, and nutritional disorders are shown in orange, noncommunicable disease causes in blue, and injuries in gray. Absolute difference in DALY rates between the lowest-SDI and highest-SDI locations were more apparent in children younger than 5 years than other age groups (eFigure 3 and eFigure 4 in the Supplement). The CMNN category was associated with the most DALYs in all SDI quintiles for children younger than 1 year (DALYs per 100 000 population, by SDI: low, 374 554 [95% UI, 357 5343-392 461]; low-middle, 301 824 [95% UI, 278 689-326 511]; middle, 96 368 [95% UI, 92 615-100 368]; high-middle, 48 272 [95% UI, 46 183-50 491]; high, 21 319 [95% UI, 19 936-22 237]). The NCDs increased in importance with age, causing the most DALYs in the middle-SDI quintile (3355 DALYs [95% UI, 2657-4169] per 100 000), high-middle–SDI quintile (3266 [95% UI, 2570-4102] per 100 000), and high-SDI quintile (2938 [95% UI, 2195-3800] per 100 000) in children aged 5 to 9 years, and in all but the low-SDI quintile in children aged 10 to 14 years (per SDI: low-middle quintile, 5116.1 [95% UI, 4061.4-6371.4]; middle quintile, 4503.3 [95% UI, 3519.9-5643.3]; high-middle quintile, 4446.0 [95% UI, 3453.1-5604.5]; high quintile, 4662.8 [95% UI, 3478.0-6089.5]) and 15 to 19 years (per SDI: low-middle quintile, 7266.1 [95% UI, 5839.7-8903.7]; middle quintile, 6370.9 [95% UI, 5039.9-7939.6]; high-middle quintile, 6420.5 [95% UI, 5018.1-8095.4]; high quintile, 7774.3 [95% UI, 5919.3-9924.6]). In the low-SDI quintile, NCDs and CMNN DALYs were close in value (NCDs, children aged 10-14 years, 4990.0 [95% UI, 3908.5-6217.1] DALYs per 100 000; those aged 15-19 years, 7051.07 [95% UI, 5632.0-8659.6] DALYs per 100 000; CMNNs: children aged 10-14 years, 5843.5 [95% UI, 5016.5-6828.1] DALYs per 100 000; those aged 15-19 years, 5943.6 [95% UI, 5284.8-6789.5] DALYs per 100 000).

ue (NCDs, children aged 10-14 years, 4990.0 [95% UI, 3908.5-6217.1] DALYs per 100 000; those aged 15-19 years, 7051.07 [95% UI, 5632.0-8659.6] DALYs per 100 000; CMNNs: children aged 10-14 years, 5843.5 [95% UI, 5016.5-6828.1] DALYs per 100 000; those aged 15-19 years, 5943.6 [95% UI, 5284.8-6789.5] DALYs per 100 000). Despite decreasing rates in 179 of 195 countries between 1990 and 2017, injuries caused an increasing proportion of overall DALYs with advancing age of children and adolescents (global rates of injury DALYs per age group: 10-14 years, 1466.7 [95% UI, 1371.0-1570.3] DALYs per 100 000; 15-19 years, 2979.0 [95% UI, 2842.3-3127.0] DALYs per 100 000), to the point that total DALYs owing to injuries were only marginally lower than CMNN DALYs in those aged 10 to 19 years. The sociodemographic gradient for injuries was not as stark as for CMNN until the high-SDI quintile was reached, however. In this quintile, there was a dramatic decline in injury DALYs compared with other SDI settings from 1990 to 2017 (low, −58% [95% UI, −64% to −47%]; low-middle, −48%, [95% UI, −53% to −39%]; middle, −57% [95% UI, −59% to −57%]; high-middle, −58% [95% UI, −60% to −56%]; high, −52% [95% UI, −54% to −50%]), but this category rose in relative importance to cause nearly 20% of total DALYs by 2017 in all SDI settings (low, 6.5% [95% UI, 6.0%-7.1%]; low-middle, 7.5% [95% UI, 7.0%-7.9%]; middle, 14.0% [95% UI, 13.0%-15.0%]; high-middle, 15.9% [95% UI, 14.7%-17.2%]; high, 13.5% [95% UI, 12.2%-14.9%]).

% [95% UI, −54% to −50%]), but this category rose in relative importance to cause nearly 20% of total DALYs by 2017 in all SDI settings (low, 6.5% [95% UI, 6.0%-7.1%]; low-middle, 7.5% [95% UI, 7.0%-7.9%]; middle, 14.0% [95% UI, 13.0%-15.0%]; high-middle, 15.9% [95% UI, 14.7%-17.2%]; high, 13.5% [95% UI, 12.2%-14.9%]). Historical patterns in SDI and DALY rates illustrate the epidemiologic transition by age (eFigure 5 in the Supplement). The slope of the SDI gradient decreased with increasing age for all causes. For CMNN causes, DALY rates tracked closely with SDI differences in all regions, except Southern sub-Saharan Africa and the Caribbean. The association between SDI groups and NCDs was similar in all children younger than 10 years (range across all SDI levels: children younger than 1 year, 30 413.3-63 232.9; aged 1-4 years, 3504.3-12 070.2; aged 5-9 years, 3273.1-5017.1), but flattened somewhat in adolescents (range across all SDI levels: aged 10-14 years, 4757.6-5685.7; aged 15-19 years, 6835.6-8537.9). In this group, the DALY rate was also higher. In several regions, as evidenced by steep temporal slope of regional plots, especially in Andean Latin America and East Asia, improvements in NCD DALYs outpaced what would have been expected on the basis of SDI improvements alone (O:E DALY rates: Andean Latin America, 1990-2000, −24.1; 2000-2017, −17.0; East Asia, 1990-2000, 12.2; 2000-2017, −9.1). A trend toward increasing DALY rates owing to injury in children aged 5 to 9 years and 10 to 19 years was seen at the lower end of the development spectrum. Eastern Europe and Southern sub-Saharan Africa had consistently higher injury DALY rates than expected by SDI grouping.

st Asia, 1990-2000, 12.2; 2000-2017, −9.1). A trend toward increasing DALY rates owing to injury in children aged 5 to 9 years and 10 to 19 years was seen at the lower end of the development spectrum. Eastern Europe and Southern sub-Saharan Africa had consistently higher injury DALY rates than expected by SDI grouping. Identifying Exemplars Changes in the ratio of observed-to-expected (O:E) DALY rates from 1990 to 2000 and 2000 to 2017 are mapped in Figure 3 (and by age group in eFigure 6 in the Supplement). Before 2000, 117 countries improved more than expected on the basis of SDI changes, while 116 countries did so after 2000. Seventy-six countries had faster than expected improvement in both periods; the O:E ratio of all-cause DALY rates were most notable in Liberia (1990-2000, −35.7; 2000-2017, −30.0), Niger (1990-2000, −14.7; 2000-2017, −43.3), Kyrgyzstan (1990-2000, −36.1; 2000-2017, −21.0), Peru (1990-2000, −32.8; 2000-2017, −19.0), and Georgia (1990-2000, −37.6; 2000-2017, −12.0). On the other end of the performance spectrum, 38 countries performed worse than expected in both periods. From 1990 to 2000, North Korea had O:E DALY rates increase 225% more than expected. However, many countries that underperformed expectations before 2000 reversed course post-2000. Notable exceptions include Dominica (all-cause O:E DALY rates: 1990-2000, 31.0; 2000-2017, 99.2), Equatorial Guinea (all-cause O:E DALY rates: 1990-2000, 41.8; 2000-2017, 23.1), Bosnia and Herzegovina (all-cause O:E DALY rates: 1990-2000, 29.8; 2000-2017, 34.2), and Lesotho (all-cause O:E DALY rates: 1990-2000, 46.9; 2000-2017, 14.9).

ns include Dominica (all-cause O:E DALY rates: 1990-2000, 31.0; 2000-2017, 99.2), Equatorial Guinea (all-cause O:E DALY rates: 1990-2000, 41.8; 2000-2017, 23.1), Bosnia and Herzegovina (all-cause O:E DALY rates: 1990-2000, 29.8; 2000-2017, 34.2), and Lesotho (all-cause O:E DALY rates: 1990-2000, 46.9; 2000-2017, 14.9). Figure 3. Percentage Change From 1990 to 2017 in Observed-to-Expected (O:E) Disability-Adjusted Life-Years Ratio in Children and Adolescents of Both Sexes, Aged 0 to 19 Years Percentage changes in observed-to-expected disability-adjusted life-years (DALYs) are plotted for 1990 to 2017 for all children and adolescents younger than 20 years. Both sexes are combined. Subnational differentiation occurs within each country’s Global Burden of Disease models at the subnational level. Inset plots provided for detailed inspection of small or clustered regions. ATG indicates Antigua and Barbuda; BRB, Barbados; COM, Comoros; DMA, Dominica; E Med, Eastern Mediterranean; FJI, Fiji; FSM, Federated States of Micronesia; GRD, Grenada; KIR, Kiribati; LCA, St Lucia; MDV, Maldives; MHL, Marshall Islands; MLT, Malta; MUS, Mauritius; SGP, Singapore; SLB, Solomon Islands; SYC, Seychelles; TLS, Timor-Leste; TON, Tonga; TTO, Trinidad and Tobago; W Africa, West Africa; WSM, Samoa; VCT, St Vincent and the Grenadines; and VUT, Vanuatu.

icronesia; GRD, Grenada; KIR, Kiribati; LCA, St Lucia; MDV, Maldives; MHL, Marshall Islands; MLT, Malta; MUS, Mauritius; SGP, Singapore; SLB, Solomon Islands; SYC, Seychelles; TLS, Timor-Leste; TON, Tonga; TTO, Trinidad and Tobago; W Africa, West Africa; WSM, Samoa; VCT, St Vincent and the Grenadines; and VUT, Vanuatu. Syria was also an outlier in how much worse than expected observed DALY rates were in children and adolescents (O:E ratios in 1990: all-cause, 0.48; injuries, 0.49; O:E ratios in 2017: all-cause, 1.09; injuries, 4.73). This was primarily owing to increased injury rates (for all individuals younger than 20 years: 1990, 189 305 [95 UI, 154 987-224 769]; 2017, 1 253 214 [95% UI, 1 225 825-1 288 824]; percentage change, 562.0% [95% UI, 462.5%-705.8%]). Corresponding maps depicting data for children younger than 1 year, 1 to 4 years, 5 to 9 years, and 10 to 19 years for CMNN, NCDs, and injuries separately are shown in eFigure 7 and eTable 3 in the Supplement. For most countries in sub-Saharan Africa, improvements were much faster than expected between 2000 and 2017 for children aged 1 to 4 years in particular, with several countries also having more rapid DALY improvement than expected in children younger than 1 year and aged 5 to 9 years. Among adolescents, on the other hand, there was little evidence of accelerated improvement after the turn of the century, with almost half of the countries in sub-Saharan Africa lagging behind expected improvements in DALY rates.

apid DALY improvement than expected in children younger than 1 year and aged 5 to 9 years. Among adolescents, on the other hand, there was little evidence of accelerated improvement after the turn of the century, with almost half of the countries in sub-Saharan Africa lagging behind expected improvements in DALY rates. Leading Causes of DALYs The top 10 level 3 GBD causes of DALYs globally in 2017 for each region and country, along with their O:E DALY rates on the basis of SDI, are shown in eFigure 8 in the Supplement. Globally, for all children and adolescents, only 1 primarily nonfatal disease ranked in the top 10 of global DALYs: iron-deficient anemia (eighth; O:E ratio, 2.08). The rest of the top 10 are also leading causes of death, including neonatal disorders (O:E ratio, 1.38), lower respiratory infection (O:E ratio, 1.83), diarrhea (O:E ratio, 4.96), congenital birth defects (O:E ratio, 0.78), malaria (O:E ratio, 4596.33), meningitis (O:E ratio, 1.54), road injuries (O:E ratio, 0.69), protein-energy malnutrition O:E ratio, (9.73), and HIV/AIDS (O:E ratio, 10.49). Every country in sub-Saharan Africa had either neonatal disorders, malaria, or HIV/AIDS as the leading cause of DALYs, with either diarrhea or lower respiratory infection often ranked second. Neonatal disorders or congenital birth defects were ranked either first or second in most other countries. Important country-specific exceptions included natural disasters ranked first in Puerto Rico (O:E ratio, 5993.81), interpersonal violence ranked second or third in Brazil (O:E ratio, 5.36) and most of Central Latin America (example O:E ratios: Mexico, 3.59; Honduras, 3.11; Guatemala, 2.74; El Salvador, 5.89; Colombia, 3.36; Panama, 2.84; Venezuela, 7.18), and conflict and terror ranked first in Syria (O:E ratio, 11 497.8) and second in Iraq (3204.1) and Libya (O:E ratio, 3442.5).

Brazil (O:E ratio, 5.36) and most of Central Latin America (example O:E ratios: Mexico, 3.59; Honduras, 3.11; Guatemala, 2.74; El Salvador, 5.89; Colombia, 3.36; Panama, 2.84; Venezuela, 7.18), and conflict and terror ranked first in Syria (O:E ratio, 11 497.8) and second in Iraq (3204.1) and Libya (O:E ratio, 3442.5). Also notable was the burden of sudden infant death syndrome (SIDS) in infants; SIDS was ranked third cause of DALYs in children younger than 1 year in the high-SDI quintile and in the top 10 for all high-income countries, plus all of Eastern Europe and Central Europe, accounting for 0.71% of deaths in the late neonatal period (age range, 7-27 days) and 2.24% in the postneonatal period (age range, 28-364 days) in 2017 globally. Sudden infant death syndrome also accounted for 3.4% of postneonatal deaths and 17.0% of deaths in the late neonatal period in high-SDI locations, but only 0.67% and 2.15% of deaths, respectively, in low-SDI settings.

(age range, 7-27 days) and 2.24% in the postneonatal period (age range, 28-364 days) in 2017 globally. Sudden infant death syndrome also accounted for 3.4% of postneonatal deaths and 17.0% of deaths in the late neonatal period in high-SDI locations, but only 0.67% and 2.15% of deaths, respectively, in low-SDI settings. The Growing Burden of Nonfatal Health Loss Rates of YLDs decreased only slightly and nonsignificantly between 1990 and 2017. Amidst a backdrop of decreasing premature childhood death and population growth, global YLDs increased 4.7% to a total of 145 million (95% UI, 107-190 million) among children and adolescents. The YLD rates increased with age, from 4366 (3168-5797) per 100 000 population in children younger than 1 year to 4486 (3242-5956) per 100 000 population in children aged 1 to 4 years, 4981 (3560-6619) per 100 000 population in children aged 5 to 9 years, to 6542 (4845-8493) per 100 000 population in those aged 10 to 19 years. Temporal trends by region in YLL-to-YLD ratio and SDI (eFigure 9 in the Supplement) demonstrate an epidemiologic transition to nonfatal health loss even more pronounced than the transition in DALYs. The association of the YLL-to-YLD ratio and SDI categories was consistent for CMNN in all age groups (children younger than 1 year, 67.0 [95% UI, 48.8-90.4]; 1-4 years, 6.9 [95% UI, 4.9-9.0]; 5-9 years, 1.5 [95% UI, 0.8-1.6]; 10-19 years, 1.5 [95% UI, 0.8-1.4]) but was stark for NCDs in children younger than 1 year (children younger than 1 year, 34.0 [95% UI, 24.7-46.4]; 1-4 years, 2.1 [95% UI, 1.2-2.4]; 5-9 years, 0.9 [95% UI, 0.3-0.6]; 10-19 years, 0.8 [95% UI, 0.2-0.4]).

1-4 years, 6.9 [95% UI, 4.9-9.0]; 5-9 years, 1.5 [95% UI, 0.8-1.6]; 10-19 years, 1.5 [95% UI, 0.8-1.4]) but was stark for NCDs in children younger than 1 year (children younger than 1 year, 34.0 [95% UI, 24.7-46.4]; 1-4 years, 2.1 [95% UI, 1.2-2.4]; 5-9 years, 0.9 [95% UI, 0.3-0.6]; 10-19 years, 0.8 [95% UI, 0.2-0.4]). In this case, there was barely any association between SDI level and the YLL-to-YLD ratio until the highest SDI strata, which may have reflected a poor penetration of prevention and treatment services for congenital birth defects and neoplasms outside of high-income countries. In the case of NCDs, increasing SDI was associated with an increased YLL-to-YLD ratio in children aged 5 to 9 years and 10 to 19 years. A similar association was found in the case of injuries (children younger than 1 year, 169.2 [95% UI, 117.2-245.8]; 1-4 years, 38.7 [95% UI, 27.2-52.7]; 5-9 years, 9.4 [95% UI, 6.4-11.7]; 10-19 years, 7.1 [95% UI, 4.8-8.3]). This could possibly reflect poorer prevention and treatment access for these causes and the effect of wars and natural disasters on disease burden.

ies (children younger than 1 year, 169.2 [95% UI, 117.2-245.8]; 1-4 years, 38.7 [95% UI, 27.2-52.7]; 5-9 years, 9.4 [95% UI, 6.4-11.7]; 10-19 years, 7.1 [95% UI, 4.8-8.3]). This could possibly reflect poorer prevention and treatment access for these causes and the effect of wars and natural disasters on disease burden. In 2017, the top 10 level-3 causes of YLDs globally were iron-deficient anemia (O:E ratio, 2.08), vitamin A deficiency (O:E ratio, 1.88), headache (O:E ratio, 0.82), conduct disorder (O:E ratio, 0.87), neonatal disorders (O:E ratio, 1.13), anxiety disorder (O:E ratio, 0.86), skin diseases (O:E ratio, 0.79), lower back pain (O:E ratio, 0.79), congenital disorders (O:E ratio, 0.99), and depression (O:E ratio, 0.87) (eFigure 10 in the Supplement for results by age). Iron-deficient anemia or asthma sometimes lead most low and low-middle SDI countries, with neonatal disorders leading in most middle, high-middle, and high SDI countries.

k pain (O:E ratio, 0.79), congenital disorders (O:E ratio, 0.99), and depression (O:E ratio, 0.87) (eFigure 10 in the Supplement for results by age). Iron-deficient anemia or asthma sometimes lead most low and low-middle SDI countries, with neonatal disorders leading in most middle, high-middle, and high SDI countries. Neonatal disorders was the only level 3 cause that ranked in the top 10 of both mortality and disability globally, ranking among the top 10 causes of YLDs in many countries in North Africa and the Middle East and sub-Saharan Africa. Musculoskeletal and mental health disorders (including anxiety disorders, conduct disorder, depression, autism spectrum disorders, and drug use disorders) were all highly ranked in high-income countries, in central and eastern Europe, and throughout Asia, Latin America, and the Caribbean. Hemoglobinopathies, such as sickle cell disorders and thalassemias, were also in the top 10 by O:E ratio in a number of countries, including Yemen (O:E ratio, 2.7), Burkina Faso (1.74), Côte d’Ivoire (2.03), Guinea (1.74), Liberia (1.85), Nigeria (3.21), and Sierra Leone (2.32). Among CMNN causes, HIV/AIDS was among the top 10 causes of YLDs in Malawi (O:E ratio, 55.32), Mozambique (O:E ratio, 54.35), Lesotho (O:E ratio, 150.1), Namibia (O:E ratio, 188.47), South Africa (O:E ratio, 339.23), Eswatini (also known as Swaziland; O:E ratio, 246.47), and Zimbabwe (O:E ratio, 58.79). Protein-energy malnutrition was in the top 10 causes in India (O:E ratio, 18.45), Mauritania (O:E ratio, 27.03), Djibouti (O:E ratio, 13.81), and South Sudan (O:E ratio, 1.82). Malaria was ranked in the top 5 causes in most countries of western and central sub-Saharan Africa, as well as Mozambique (O:E ratio, 2.48). Diarrhea, onchocerciasis, and intestinal nematode infections were the other CMNN causes among the top 10 causes of YLDs in certain countries. Injuries did not rank high in most countries, with notable exceptions of Iraq (where conflict ranked ninth; O:E ratio, 672.19) and Syria (where conflict ranked first; O:E ratio, 2962.5).

, onchocerciasis, and intestinal nematode infections were the other CMNN causes among the top 10 causes of YLDs in certain countries. Injuries did not rank high in most countries, with notable exceptions of Iraq (where conflict ranked ninth; O:E ratio, 672.19) and Syria (where conflict ranked first; O:E ratio, 2962.5). Associating Maternal Health Outcomes With Those of Children and Adolescents To evaluate the association between population-level trends in child and adolescent DALYs and those of their mothers, we compared percentage change from 1990 to 2017 in all-cause DALY rates for children younger than 1 year, 1 to 4 years, 5 to 9 years, and 10 to 19 years with percentage change in death rates owing to maternal disorders for women aged 10 to 54 years (eFigure 11 in the Supplement). There were strong correlations between trends in maternal death and all-cause DALY rates in all childhood age groups (<1 year, r = 0.589; 1-4 years, r = 0.452; 5-9 years, r = 0.507; 10-19 years, r = 0.379); those countries with the most improvement in maternal mortality also tended to have higher performance in reducing child and adolescent deaths. Statistical correlation was strongest for children younger than 1 year of age (r = 0.59), but continued even to health outcomes of older children and adolescents (r range = 0.38-0.45). The overall association between trends in maternal mortality and all-cause child and adolescent DALY rates became stronger after 2000 in all SDI quintiles other than high-middle SDI quintile (low: r = 0.539 in 1990-2000 vs r = 0.672 in 2000-2017; low-middle: r = 0.540 in 1990-2000 vs r = 0.576 in 2000-2017; middle: r = 0.419 in 1990-2000 vs r = 0.442 in 2000-2017; high-middle: r = 0.333 in 1990-2000 vs r = 0.296 in 2000-2017; high: r = 0.182 in 1990-2000 vs r = 0.379 in 2000-2017).

high-middle SDI quintile (low: r = 0.539 in 1990-2000 vs r = 0.672 in 2000-2017; low-middle: r = 0.540 in 1990-2000 vs r = 0.576 in 2000-2017; middle: r = 0.419 in 1990-2000 vs r = 0.442 in 2000-2017; high-middle: r = 0.333 in 1990-2000 vs r = 0.296 in 2000-2017; high: r = 0.182 in 1990-2000 vs r = 0.379 in 2000-2017). A total of 25 countries (Afghanistan, American Samoa, Antigua, Burundi, Botswana, Canada, Switzerland, Costa Rica, Djibouti, Guam, Jamaica, Kuwait, Lesotho, Madagascar, Montenegro, Netherlands, Rwanda, Singapore, Sierra Leone, São Tomé and Principe, Eswatini, Chad, Thailand, United States, and St Vincent and the Grenadines) had divergent trends in maternal and child health (ie, 1 mortality rate increased while the other decreased) between 1990 and 2000, while that was only true for 21 countries (American Samoa, Antigua, Bahamas, Barbados, Brunei, Canada, Dominican Republic, Georgia, Greece, Grenada, Hungary, Ireland, Libya, Saint Lucia, Panama, Puerto Rico, Suriname, Syria, United States, St Vincent and the Grenadines, and US Virgin Islands) from 2000 to 2017. Only 8 countries had divergent trends throughout the entire time period, with all examples of divergence having increases in maternal mortality and decreases in all-cause child and adolescent DALY rates: American Samoa (21.5% and −33.1%, respectively), Canada (36.0% and −20.4%, respectively), Greece (33.8% and −28.5%, respectively), Guam (49.0% and −2.88%, respectively), Jamaica (4.14% and −28.0%, respectively), St Vincent and the Grenadines (7.29% and −24.2%, respectively), the United States (67.5% and −25.7%, respectively), and Zimbabwe (15.5% and −10.6%, respectively).

a (36.0% and −20.4%, respectively), Greece (33.8% and −28.5%, respectively), Guam (49.0% and −2.88%, respectively), Jamaica (4.14% and −28.0%, respectively), St Vincent and the Grenadines (7.29% and −24.2%, respectively), the United States (67.5% and −25.7%, respectively), and Zimbabwe (15.5% and −10.6%, respectively). Discussion Children and adolescents in every country in the world were more likely to reach their 20th birthday in 2017 than ever before, but progress in improving health outcomes has been uneven. Mortality reductions were most rapid in children between the ages of 1 and 4 years, driven by global declines in deaths owing to diarrhea, lower respiratory infection, and other common infectious diseases. Improvements accelerated after 2000. The largest absolute declines were seen in Western, Eastern, and Central sub-Saharan Africa, while the fastest rates of decline were seen in East Asia, Andean Latin America, and South Asia. The pattern of change was closely associated with gains in sociodemographic development and temporally aligned with increased development assistance for health, which led to broad improvements in vaccination, early childhood nutrition, sanitation, clean water, and targeted interventions for HIV/AIDS and malaria.3,17,18,19,20

ttern of change was closely associated with gains in sociodemographic development and temporally aligned with increased development assistance for health, which led to broad improvements in vaccination, early childhood nutrition, sanitation, clean water, and targeted interventions for HIV/AIDS and malaria.3,17,18,19,20 A vast unfinished agenda in child and adolescent health remains. While malaria has decreased dramatically across the African continent, there are many countries, especially in western sub-Saharan Africa, where parasite transmission, acute illness, and mortality from malaria remain high. Lower respiratory infection, diarrhea, and acute malnutrition also remain among the top killers of children and adolescents in the world in 2017. Investment in programs targeting prevention and effective syndromic treatment of CMNN disorders clearly pays dividends, and these investments must continue. In locations with higher SDIs, a continuing shift toward nonfatal health loss from NCDs, such as congenital birth defects, mental and behavioral disorders, injuries, and asthma are challenging health systems to adapt.21 The consistent burden of NCD-attributable DALYs in adolescents over the past 28 years illustrates a need for continued research and action on NCDs as communicable disease burden declines across the development spectrum. The burden of injuries in adolescents surpasses that of CMNN causes throughout the study period for middle-SDI through high-SDI countries, and with the relative faster decline of CMNN causes in low and low-middle countries, the relative ranking of injuries may switch in those locations in the near future.

ent spectrum. The burden of injuries in adolescents surpasses that of CMNN causes throughout the study period for middle-SDI through high-SDI countries, and with the relative faster decline of CMNN causes in low and low-middle countries, the relative ranking of injuries may switch in those locations in the near future. Overall health improvements were slowest in adolescents. Few locations showed any evidence of improvements in health among adolescents that exceeded the trends expected with general societal development gains. Adolescence is a key phase of the life course and human development, including a phase of growth and maturation of the reproductive, musculoskeletal, neurodevelopmental, endocrine, metabolic, immune, and cardiometabolic systems into adulthood.22 Gains or lack thereof in adolescent health thus have the potential to influence individual and societal outcomes for periods substantially longer than the teenage years. In terms of family and home life, key issues include the improvement of sanitary and living conditions, stable food systems, quality education, and gainful employment.23 Also, HIV/AIDS remains an imminent threat to the health and well-being of older children and adolescents in many countries in sub-Saharan Africa, such as South Africa, Zimbabwe, Lesotho, Eswatini, Botswana, and Zambia. The large and growing burden of mental health and substance use disorders among older children and adolescents also is an emerging threat to the thrive component of the SDG survive and thrive agenda. While the psychological needs of children and adolescents show similarities across geographical settings,24,25,26,27 comparatively little is understood about modifiable risk factors or effective prevention programs for childhood mental illness, outside of ensuring that caregivers are attuned to the link between mental health disorders and self-harm.28,29 Injuries in general continue to be a major cause of early mortality and long-term disability among older children and adolescents in all countries.

effective prevention programs for childhood mental illness, outside of ensuring that caregivers are attuned to the link between mental health disorders and self-harm.28,29 Injuries in general continue to be a major cause of early mortality and long-term disability among older children and adolescents in all countries. While many types of injuries, such as those arising from war and natural disasters, may not be preventable with health sector–based approaches, diligent preparedness planning can help mitigate the immediate health aftermath of them.30,31,32 Others are much more amenable to policies and programs that focus on prevention using what have come to be regarded as common-sense safety measures, such as speed limits, seat belts, and cycle helmets for road traffic accidents,33,34 fencing around water hazards and swimming-skills training for drowning,35 and policies to prevent self-harm via improving safety and limiting access to firearms and chemicals.36,37

e come to be regarded as common-sense safety measures, such as speed limits, seat belts, and cycle helmets for road traffic accidents,33,34 fencing around water hazards and swimming-skills training for drowning,35 and policies to prevent self-harm via improving safety and limiting access to firearms and chemicals.36,37 At the other end of the age spectrum, neonatal disorders remain a major prevention and treatment challenge, especially for countries outside the high-SDI quintile that lack the same level of financial and human resources to dedicate to the intensive care needs of sick neonate. Investment is needed to develop and implement cost-effective interventions for neonatal disorders that take into account the dynamics of maternal health, risk-factor exposures during pregnancy, clinical care systems, supportive equipment needs, and the cultural differences around how families and communities care for newborns. It is important also to invest in the ongoing care of children who survive perinatal emergencies only to develop long-term complications, such as cerebral palsy. Congenital birth defects and hemoglobinopathies are 2 other groups of causes for which there is little evidence of improved outcomes outside the high-SDI quintile, perhaps reflecting the resource-intensive nature of averting deaths owing to such conditions and societal barriers to care38 but also likely because of a failure of recent clinical advances to be adopted in lower-resource settings.39

for which there is little evidence of improved outcomes outside the high-SDI quintile, perhaps reflecting the resource-intensive nature of averting deaths owing to such conditions and societal barriers to care38 but also likely because of a failure of recent clinical advances to be adopted in lower-resource settings.39 The close linkage between trends in maternal and child health reinforces the notion that the health of different population segments are closely interconnected.40 The simultaneous focus of the Millennium Development Goals on maternal and child mortality appears to have led to closer association between them since 2000 via alignment of funding streams, targeting of common risk factors between mothers and their children, an increased focus on delaying the age of parenthood by increasing education, contraception, and increased birth spacing, and catalyzing improved gender equity.41,42,43,44,45,46,47 There are strong ties between the physical health of women (eg, high body mass index, NCDs, nutrition) and neonatal outcomes (such as pregnancy complications, short gestational age, and low birth weight), which are in turn linked with poorer health outcomes and delayed development.3,6 This is to say nothing of the potential epigenetic connections between mothers and the health of their children that have the potential to extend beyond the neonatal period into childhood, adolescence, adulthood, and the next generation.48 The subset of countries that are outliers to this trend of concomitant improvement in maternal and child health warrant close examination to determine the underlying causes. Challenges are likely to arise whenever funding streams are decoupled, education or family planning programs are disrupted, or the health of young women is not prioritized.

at are outliers to this trend of concomitant improvement in maternal and child health warrant close examination to determine the underlying causes. Challenges are likely to arise whenever funding streams are decoupled, education or family planning programs are disrupted, or the health of young women is not prioritized. The epidemiological transition has unique implications for the health of children and adolescents and the potential trajectory of socioeconomic development. In particular, as more children survive, the human capital potential societies will expand, but as more children with health problems are also surviving, there is potential for increased burden on health and education systems. The cost of sustaining progress on child and adolescent health and well-being is not insignificant. To achieve the goal of surviving and thriving and realized the human capital potential of children and adolescents, all countries must make strategic investments in education and health systems, including human resources for health, supply chains, infrastructure, governance, and increased support for children with developmental disabilities. Alignment of funding around interconnected drivers of human development and health loss is also required to achieve the SDGs.49

nvestments in education and health systems, including human resources for health, supply chains, infrastructure, governance, and increased support for children with developmental disabilities. Alignment of funding around interconnected drivers of human development and health loss is also required to achieve the SDGs.49 The SDGs are expansive, but they should not be considered a comprehensive rubric for achieving improved child and adolescent health. For example, outside of women’s reproductive health and experiences of sexual violence during adolescence, the SDG goals, targets, and indicators remain largely silent on the unique social, environmental, and biological determinants of health occurring in adolescence across the socioeconomic development spectrum. This blind spot in international health targets, planning, and prevention fails to capture the complex transitions occurring during adolescence in particular. Many additional nonhealth SDG indicators also focus on reducing poverty, expanding education, stabilizing environments, strengthening economies, and reducing overall socioeconomic inequality within each country and throughout the world, all of which are relevant to the health and well-being of young persons.

articular. Many additional nonhealth SDG indicators also focus on reducing poverty, expanding education, stabilizing environments, strengthening economies, and reducing overall socioeconomic inequality within each country and throughout the world, all of which are relevant to the health and well-being of young persons. Limitations The GBD study is an iterative process and, despite continued methodological advancements and improvements in data, this study has a number of limitations. First, all limitations documented in the elements of the GBD estimation process that allow for YLL, YLD, and DALY estimation will contribute to uncertainty in these summary measures. Second, these summary measures of population health are influenced by data availability. Time lags in the reporting of health information by national authorities and thus subsequent incorporation into the GBD estimation mean that these estimates are based on data that are already out of date. Relatedly, data deficiencies from populations in conflict zones (eg, Syria, Iraq, Yemen, South Sudan, Afghanistan), autonomous subnational regions, and certain nongeographical subpopulations (ie, migrants, refugees, and some indigenous peoples) limit the precision of some of the estimated levels and trends of disease burden. Third, the association between YLLs, YLDs, DALYs, and SDIs, although explanatory, cannot be viewed as causal. Fourth, a nontrivial assumption of the analyses is the independence of the uncertainty calculated for YLLs and YLDs. Because of the link between death and prevalence, a positive correlation probably exists between these uncertainties that are not captured in this analysis. Study limitations specific to child and adolescent health include the comparatively poor quality of cause-of-death certification in neonates and infants vs older persons, the relatively broad age categorization of all 1-to-4-year-old children in 1 group, and the limited ability to quantify the magnitude of specific intergenerational, societal, and environmental factors that are ecologically suggested by this study.

f cause-of-death certification in neonates and infants vs older persons, the relatively broad age categorization of all 1-to-4-year-old children in 1 group, and the limited ability to quantify the magnitude of specific intergenerational, societal, and environmental factors that are ecologically suggested by this study. Conclusions Globally, the aggregate health status of children and adolescents improved dramatically between 1990 and 2017, particularly owing to declines in death owing to infectious diseases, but nonfatal health loss has increased in both absolute and relative terms, and the gap between best and worst performers has widened. Continued monitoring of the drivers of child and adolescent health loss is crucial to sustain the progress of the past 26 years in the SDG era. The global community must commit to creating systematic accounting of drivers and consequences of long-lasting negative health outcomes beginning in childhood and the effects of long-term morbidity on health systems and human capital and ensuring that no populations are left behind. Only then will we be able to accelerate progress to 2030 and beyond. Supplement. eFigure 1. GBD 2017 Socio-demographic index (SDI) quintiles by GBD administrative level 1 geography eFigure 2. Decomposition of the probability of death in 195 countries for <20 years from 1990 to 2017, both sexes combined eFigure 3. Trends of DALYs from 1990 to 2017 for global and SDI quintiles for <1, 1-4, 5-9, 10-19 years eFigure 4. Trends of DALYs from 1990 to 2017 by age group (<1, 1-4, 5-9, 10-19 years) for global and SDI quintiles

eFigure 2. Decomposition of the probability of death in 195 countries for <20 years from 1990 to 2017, both sexes combined eFigure 3. Trends of DALYs from 1990 to 2017 for global and SDI quintiles for <1, 1-4, 5-9, 10-19 years eFigure 4. Trends of DALYs from 1990 to 2017 by age group (<1, 1-4, 5-9, 10-19 years) for global and SDI quintiles eFigure 5. Coevolution of disability-adjusted life years (DALYs) and socio-demographic index (SDI) for global disease burden (GDB) study regions and level 1 causes eFigure 6. Map of percent change for observed to expected (O:E) all-cause DALY rates from 1990 to 2000, 2000 to 2017 for <1, 1-4, 5-9, 10-19 years eFigure 7. Annual percent change of observed to expected (O:E) all-cause DALY rates in <20 years from 1990 to 2000 versus 2000 to 2017 eFigure 8. Leading ten causes of DALYs with the ratio of observed to expected on the basis of SDI alone in 2017 for a) <20 years, b) <1 years, c) 1-4 years, d) 5-9 years, and e) 10-19 years for both sexes combined eFigure 9. Co-evolution of YLL to YLD ratios and SDI for GBD regions and Level 1 causes, 1990 o 2017, both sexes combined for a) all causes, b) CMNN, c) NCDs, d) Injuries eFigure 10. Leading ten causes of YLDs with the ratio of observed to expected on the basis of SDI alone in 2017 for a) <20 years, b) <1 years, c) 1-4 years, d) 5-9 years, and e) 10-19 years for both sexes combined eFigure 11. Percent change in MMR vs. percent change in all-cause child DALYs rate for <1, 1-4, 5-9, and 10-19 years, 1990-2017 eTable 1. SDI groupings by location, based on GBD 2017 values

eFigure 10. Leading ten causes of YLDs with the ratio of observed to expected on the basis of SDI alone in 2017 for a) <20 years, b) <1 years, c) 1-4 years, d) 5-9 years, and e) 10-19 years for both sexes combined eFigure 11. Percent change in MMR vs. percent change in all-cause child DALYs rate for <1, 1-4, 5-9, and 10-19 years, 1990-2017 eTable 1. SDI groupings by location, based on GBD 2017 values eTable 2. ENN, LNN, PNN, 1-4, 5-9, 10-14, 15-19, <1, <5, 10-19, and <20 years mortality in 1990, 2000 and 2017 with mean percentage change for all GBD causes eTable 3. Percent change of observed to expected DALYs rate from 1990 to 2000 and 2000 to 2017 for each GBD location, for all causes combined and separately for communicable, maternal, neonatal, and nutritional (CMNN), non-communicable diseases (NCDs), and injuries or a) <20 years, b) <1 years, c) 1-4 years, d) 5-9 years, and e)10-19 years, both sexes combined Click here for additional data file.

Introduction Puberty, the transition from childhood to adulthood that ends with attaining reproductive capability, is a milestone in human development. The pubertal process consists of a series of events that can be used to determine the timing of puberty. A physical examination of adolescent patients, and estimation of either breast development or testicular volume according to Tanner stages, is often used in routine pediatric care. Self-reported pubertal demarcations can be used to estimate pubertal timing retrospectively. Age at menarche in girls has substantial accuracy, whereas for boys there is no reliable corresponding pubertal event. Studies on male pubertal timing are therefore scarce. If several measurements of height during pubertal growth are available, the timing of the pubertal growth spurt (age at the peak height velocity [PHV]) can be used as an objective assessment of pubertal timing. Age at PHV is the age at the maximum growth velocity during puberty and occurs approximately 2 years and 1.5 years after pubertal onset in boys and girls, respectively.

e available, the timing of the pubertal growth spurt (age at the peak height velocity [PHV]) can be used as an objective assessment of pubertal timing. Age at PHV is the age at the maximum growth velocity during puberty and occurs approximately 2 years and 1.5 years after pubertal onset in boys and girls, respectively. A secular trend (ie, not a cyclical or seasonal trend, but a trend over a long period) of earlier menarcheal age has been observed in Europe and the United States since the mid-19th century, partly due to increased childhood body mass index (BMI; calculated as weight in kilograms divided by height in meters squared) during the last 60 years. Early menarche has been associated with an increased risk of diseases, such as type 2 diabetes and breast cancer. These findings indicate that changes in menarcheal age might have implications for health and disease later in life. To our knowledge, for boys, no large population-based study has been performed that covers a long time span with an objective assessment of pubertal timing that captures the entire pubertal period and with adjustment for childhood BMI.

nges in menarcheal age might have implications for health and disease later in life. To our knowledge, for boys, no large population-based study has been performed that covers a long time span with an objective assessment of pubertal timing that captures the entire pubertal period and with adjustment for childhood BMI. Whether there is a secular trend for earlier pubertal timing in boys is not established and thus represents a knowledge gap regarding male puberty. We have previously developed a method for objectively assessing pubertal timing based on the height growth curve. Using this method, we recently demonstrated that childhood BMI is inversely associated with age at PHV in boys with normal weight but not in boys with overweight in a large population-based cohort of 31 971 boys. Moreover, we have reported a pronounced increase in mean childhood BMI among 8-year-old boys from the 1940s until now. Here, we address the earlier mentioned knowledge gap regarding male pubertal timing and hypothesize that there is a secular trend for earlier pubertal timing in boys that can be partially explained by childhood BMI. This study’s aim was to investigate the association between birth year and changes in male pubertal timing independent of changes in childhood BMI. To this end, we used the population-based BMI Epidemiology Study (BEST) cohort covering 50 years of growth data in men born from 1947 to 1996 with age at PHV available.

MI. This study’s aim was to investigate the association between birth year and changes in male pubertal timing independent of changes in childhood BMI. To this end, we used the population-based BMI Epidemiology Study (BEST) cohort covering 50 years of growth data in men born from 1947 to 1996 with age at PHV available. Participants and Methods Data Collection and Study Population The population-based BEST cohort was initiated with the overall aim to determine the role of childhood obesity and pubertal timing for various diseases in adult life. The BEST cohort included individuals who completed school in Gothenburg municipality, Sweden, and afterwards had their school health record stored in the Archives of City of Gothenburg and Region Västra Götaland. The school health records include data on height and weight as measured by specially trained school nurses. For height measurements, a wall-mounted stadiometer was used and the pupils were weighed wearing lightweight clothing. These health examinations were performed according to a prespecified program throughout childhood and until the children finished secondary school and include all children in Sweden (98.5% for school health care from calendar year 1952). The data from regular health visits at school health care were noted in school health care records and have been retrieved from these records for use in research. In addition to these childhood data, height and weight have also been retrieved from the conscription register. At military conscription, which was mandatory in Sweden from 1901 to 2010, all recruits were examined and had their height measured by specially trained staff using a wall-mounted stadiometer. To determine age at PHV for individuals who lacked final height measurements from school health records or the conscription register, we retrieved heights from the passport register, which includes self-reported heights for all individuals holding a passport in Sweden (eMethods in the Supplement). The ethics committee at the University of Gothenburg approved the study and waived the requirement for written informed consent.

the conscription register, we retrieved heights from the passport register, which includes self-reported heights for all individuals holding a passport in Sweden (eMethods in the Supplement). The ethics committee at the University of Gothenburg approved the study and waived the requirement for written informed consent. Linkage With Registers From Statistics Sweden Using the individuals’ personal identity numbers (PINs), the BEST cohort was linked with the Longitudinal Integration Database for Health Insurance and Labor Market Studies at Statistics Sweden. The country of birth for each individual and his parents was also retrieved.

the conscription register, we retrieved heights from the passport register, which includes self-reported heights for all individuals holding a passport in Sweden (eMethods in the Supplement). The ethics committee at the University of Gothenburg approved the study and waived the requirement for written informed consent. Linkage With Registers From Statistics Sweden Using the individuals’ personal identity numbers (PINs), the BEST cohort was linked with the Longitudinal Integration Database for Health Insurance and Labor Market Studies at Statistics Sweden. The country of birth for each individual and his parents was also retrieved. Curve Fitting and Statistical Analyses To adequately calculate age at PHV in an unbiased manner, height measurements before, during, and after the pubertal period are required. We calculated age at PHV according to a modified infancy-childhood-puberty model as previously described. For each growth curve with sufficient information, we used the nls.lm function in the R package minpack.lm (R Foundation). The model was fitted by minimizing the sum of squares using a modification of the Levenberg-Marquardt algorithm and it also had tests controlling for convergence. A good model fitting was confirmed through visually inspecting all curves (M.B. and J.M.K.). Age at PHV was defined as the age at the maximum growth velocity during puberty and was estimated by the model. Childhood BMI at age 8 years was calculated for every included individual using all paired height and weight measurements between age 6.5 and 9.5 years. This age interval was selected to represent the childhood period after infancy but before the confounding association of puberty with body composition and BMI in boys. Because BMI is also age dependent within the selected childhood period, age adjustment of BMI within the interval was performed using a linear model with BMI as a dependent variable and age as an independent variable. For every BMI measurement within the interval, this linear assumption was used to estimate a BMI at exactly age 8 years. R statistical software, version 3.4.2 (R Foundation), was used to calculate childhood BMI and age at PHV.

med using a linear model with BMI as a dependent variable and age as an independent variable. For every BMI measurement within the interval, this linear assumption was used to estimate a BMI at exactly age 8 years. R statistical software, version 3.4.2 (R Foundation), was used to calculate childhood BMI and age at PHV. Descriptive statistics for age at PHV and childhood BMI were calculated for each birth cohort. The overall trend was calculated using a linear regression and the comparison between mean age at PHV for the different birth cohorts was tested using 1-way analysis of variance followed by the Tukey post hoc test. The distribution of age at PHV for the different birth cohorts was described using the 5th, 25th, 50th, 75th, and 95th percentiles. The association between childhood BMI and age at PHV has previously been shown to be nonlinear. The nonlinear association between childhood BMI and age at PHV in this cohort was evaluated using a piecewise linear regression model as previously described. Furthermore, given the nonlinear association between childhood BMI and age at PHV, childhood BMI and quadratic childhood BMI were included when the linear regression model was adjusted for childhood BMI. In all analyses when suitable, log-transformed and standardized childhood BMI was used. A possible nonlinear association between birth year and age at PHV was tested by including a quadratic or cubic term for the birth year. Moreover, we performed a goodness-of-fit test (F test) to evaluate if the overall model fit was improved when including the quadratic or cubic terms. For all statistical analyses, SPSS (version 24; IBM) was used, and statistical significance was set at P < .05.

by including a quadratic or cubic term for the birth year. Moreover, we performed a goodness-of-fit test (F test) to evaluate if the overall model fit was improved when including the quadratic or cubic terms. For all statistical analyses, SPSS (version 24; IBM) was used, and statistical significance was set at P < .05. Results We included 11 male birth cohorts born in 1947 (reference cohort) and every 5 years from 1951 to 1996 (n = 375 for each birth cohort from 1947-1991, n = 340 for birth cohort in 1996, and a total n = 4090 [69% of eligible individuals]) in the present BEST subcohort (eFigure 1 in the Supplement). The mean (SD) age at PHV was 13.9 (1.1) years for the total cohort (eTable 1 in the Supplement).

ery 5 years from 1951 to 1996 (n = 375 for each birth cohort from 1947-1991, n = 340 for birth cohort in 1996, and a total n = 4090 [69% of eligible individuals]) in the present BEST subcohort (eFigure 1 in the Supplement). The mean (SD) age at PHV was 13.9 (1.1) years for the total cohort (eTable 1 in the Supplement). A linear regression analysis between birth year and age at PHV revealed that age at PHV was 1.5 months earlier per decade increase during the study period (−0.12 years per decade increase in birth year; 95% CI, −0.14 to −0.10). The decrease in age at PHV was statistically significant from the birth cohort in 1976 and onwards compared with the birth cohort in 1947 (Figure 1). We did not detect a significant nonlinearity as assessed by including a quadratic or a cubic term for birth year in the regression analysis (P values were nonsignificant for the quadratic and cubic terms). Furthermore, goodness-of-fit tests demonstrated that neither adding a quadratic nor a cubic term to the model significantly improved the overall model fit compared with the linear model. The secular trend for age at pubertal timing is shown across the distribution of age at PHV in Figure 2. In subanalyses, linear regression analyses were performed for the highest and lowest percentiles of age at PHV, demonstrating a secular trend for late and early percentiles of age at PHV that was similar for the entire population (eTable 2 in the Supplement). Compared with the entire population, there was a slightly less pronounced secular trend for those with an early age at PHV (eTable 2 in the Supplement). In addition, when age at PHV was displayed according to the participants’ BMI status at age 8 years, the secular trend was observed for participants with normal weight and overweight (eFigure 2 in the Supplement).

a slightly less pronounced secular trend for those with an early age at PHV (eTable 2 in the Supplement). In addition, when age at PHV was displayed according to the participants’ BMI status at age 8 years, the secular trend was observed for participants with normal weight and overweight (eFigure 2 in the Supplement). Figure 1. Mean Age at Peak Height Velocity (PHV) for Boys Included in the BMI Epidemiology Study Cohort Born From 1947 to 1996 Values are presented as mean (95% CI), and statistically significant differences vs the birth cohort in 1947 were observed from the birth cohort in 1976 and after (P < .01). The P for trend is <.001. Figure 2. Distribution of Age at Peak Height Velocity (PHV) for Boys Included in the BMI Epidemiology Study Cohort Born From 1947 to 1996 There was an inverse association between childhood BMI and age at PHV in the present study population born 1947 to 1996 below, but not above, a threshold at 17.71 (eFigure 3 in the Supplement). Given that childhood BMI increased during the same period, we next wanted to determine the secular trend of earlier age at PHV independent of childhood BMI. When we included age at PHV and childhood BMI (median, 15.8; range 12.3-29.2) in the same linear regression model, we found that the association between birth year and age at PHV was slightly attenuated but remained significant; the age at PHV was 1.2 months earlier per decade increase during the study period (−0.10 years per decade increase in birth year; 95% CI, −0.12 to −0.07; Figure 3).

9.2) in the same linear regression model, we found that the association between birth year and age at PHV was slightly attenuated but remained significant; the age at PHV was 1.2 months earlier per decade increase during the study period (−0.10 years per decade increase in birth year; 95% CI, −0.12 to −0.07; Figure 3). Figure 3. Mean Age at Peak Height Velocity (PHV) Adjusted for Childhood Body Mass Index (BMI) for Boys Included in the BMI Epidemiology Study Cohort Born From 1947 to 1996 Values are presented as mean (95% CI). The P for trend is <.001. To investigate a possible confounding effect by demographic changes in the population from 1947 to 1996, we performed a subanalysis including only boys born in Sweden and with parents born in Sweden (3087 [75%] of the study cohort). The subanalysis demonstrated similar results as the main analysis: age at PHV was 1.2 months earlier for every decade increase in birth year (−0.10 years per decade increase in birth year, 95% CI, −0.13 to −0.08; Figure 4). Furthermore, age at PHV was 0.9 months earlier for every decade increase in birth year independent of childhood BMI (−0.07 years per decade increase in birth year; 95% CI, −0.10 to −0.05). Similar to the total cohort, there was no indication of a nonlinear association in the subanalysis that included only boys born in Sweden and with parents born in Sweden. Figure 4. Mean Age at Peak Height Velocity (PHV) Among a Subgroup of Boys Born From 1947 to 1996 in Sweden and With Parents Born in Sweden Values are presented as mean (95% CI). The P for trend is <.001.

To investigate a possible confounding effect by demographic changes in the population from 1947 to 1996, we performed a subanalysis including only boys born in Sweden and with parents born in Sweden (3087 [75%] of the study cohort). The subanalysis demonstrated similar results as the main analysis: age at PHV was 1.2 months earlier for every decade increase in birth year (−0.10 years per decade increase in birth year, 95% CI, −0.13 to −0.08; Figure 4). Furthermore, age at PHV was 0.9 months earlier for every decade increase in birth year independent of childhood BMI (−0.07 years per decade increase in birth year; 95% CI, −0.10 to −0.05). Similar to the total cohort, there was no indication of a nonlinear association in the subanalysis that included only boys born in Sweden and with parents born in Sweden. Figure 4. Mean Age at Peak Height Velocity (PHV) Among a Subgroup of Boys Born From 1947 to 1996 in Sweden and With Parents Born in Sweden Values are presented as mean (95% CI). The P for trend is <.001. Thus, there was a clear secular trend for earlier pubertal timing in Swedish boys born from 1947 to 1996 that was independent of prepubertal BMI. Similar results were seen in subanalyses that included only boys born in Sweden and with parents born in Sweden.

Figure 4. Mean Age at Peak Height Velocity (PHV) Among a Subgroup of Boys Born From 1947 to 1996 in Sweden and With Parents Born in Sweden Values are presented as mean (95% CI). The P for trend is <.001. Thus, there was a clear secular trend for earlier pubertal timing in Swedish boys born from 1947 to 1996 that was independent of prepubertal BMI. Similar results were seen in subanalyses that included only boys born in Sweden and with parents born in Sweden. Discussion In this article, we present evidence of a secular trend for earlier pubertal timing in boys using age at PHV as objective assessment of pubertal timing in 4090 boys in 11 birth cohorts born from 1947 to 1996. Age at PHV was 1.5 months earlier for every decade increase in birth year and 1.2 months earlier per decade after adjusting for childhood BMI at age 8 years. The concept of a secular trend for earlier pubertal timing and the contribution of increasing childhood BMI to the secular trend in pubertal timing is well established for girls. However, for boys, studies are scarce and, to our knowledge, it is not established whether age at pubertal timing has changed over time. Whereas menarcheal age for girls is often easily available, a corresponding self-reported, valid marker for pubertal timing in boys is lacking.

l timing is well established for girls. However, for boys, studies are scarce and, to our knowledge, it is not established whether age at pubertal timing has changed over time. Whereas menarcheal age for girls is often easily available, a corresponding self-reported, valid marker for pubertal timing in boys is lacking. A Danish study used growth data to investigate the change in pubertal timing for boys born from 1935 to 1969. They found that age at PHV was 0.3 years earlier in boys born from 1965 to 1970 compared with boys born from 1935 to 1939. However, the study had several weaknesses. Only 10% of the study population born from 1935 to 1970 was included and for the birth years 1935 to 1939, only 0.2% were included. As the height growth data in that study (from age 7 to 15 years) did not cover the entire pubertal period, boys with late puberty, not completed at age 15 years, were not included. The study did not assess the extent of the secular trend independent of childhood BMI. Moreover, information on country of birth was not available and therefore it is unclear to what degree the secular trend was explained by increasing childhood BMI or confounded by demographic shifts in the population. Another article comparing the average pubertal timing, determined using Tanner staging of secondary sex characteristics, between several different studies that were mostly cross-sectional and collected at different points, indicated that there might be a possible trend for earlier pubertal timing in boys. However, the methods for determining pubertal timing and the composition of the cohorts with regard to recruitment and ethnicity differed substantially. An independent, well-powered cross-sectional study (n = 826) using secondary sex characteristics failed to detect any evidence of a secular trend. Thus, a secular trend toward earlier pubertal timing in boys is not as well documented as in girls.

cohorts with regard to recruitment and ethnicity differed substantially. An independent, well-powered cross-sectional study (n = 826) using secondary sex characteristics failed to detect any evidence of a secular trend. Thus, a secular trend toward earlier pubertal timing in boys is not as well documented as in girls. We have previously demonstrated an increase in mean BMI and in the prevalence of overweight among 8-year-old boys born between 1946 and 2006 and an inverse association between BMI at age 8 years and pubertal timing in boys with normal weight but not overweight in 2 cohorts, one of which was born before the obesity epidemic and one that was born late during the obesity epidemic. Here we demonstrate an inverse association between childhood BMI and pubertal timing in the present study population, covering the period from the 1940s to the 1990s. Given our findings in boys regarding a strong inverse association between childhood BMI and pubertal timing and similar previous findings in girls, it is plausible that the increasing childhood BMI has contributed to the secular trend. It is therefore relevant to determine the secular trend in pubertal timing independent of prepubertal childhood BMI. When we adjusted our analysis on the association between birth year and pubertal timing for childhood BMI, we found that prepubertal childhood BMI contributes to this association but that a substantial part of the association between birth year and age at PHV was independent of childhood BMI. Moreover, a recent study actually demonstrated earlier pubertal timing in boys with overweight than in boys with obesity. In addition, we were also able to show a significant association between birth year and pubertal timing in a subgroup of men born in Sweden and with parents born in Sweden and thereby confirm that the change in pubertal timing was not explained by demographic changes in the study population between 1947 and 1996. Although the general appearance of the curve indicates that the secular trend mainly happened from birth year 1956 to birth year 1986, we did not detect any statistically significant nonlinearity.

that the change in pubertal timing was not explained by demographic changes in the study population between 1947 and 1996. Although the general appearance of the curve indicates that the secular trend mainly happened from birth year 1956 to birth year 1986, we did not detect any statistically significant nonlinearity. Some studies have suggested an increased use of endocrine disrupting chemicals (EDCs) to account for the secular trend in earlier pubertal timing in girls. The finding that female mice exposed to the estrogenic acting endocrine disruptor bisphenol A (BPA) prenatally displayed earlier sexual maturation lends some support to this hypothesis in girls. Boys with moderate exposure to the BPA had earlier pubertal timing compared with boys with the least exposure to BPA, and urine BPA levels were also inversely associated with pubertal height gain. Conversely, phthalates have antiandrogenic and obesogenic properties and high urinary phthalates were associated with late pubarche in girls but were not associated with pubertal onset or levels of serum testosterone in boys. Thus, EDCs with estrogenic, antiandrogenic, or obesogenic actions may have differing associations with puberty in boys and girls. To what extent EDCs contribute to the secular trend in male pubertal timing presented in this study is not clear. Our findings indicate that there is a robust secular trend for earlier pubertal timing in boys that is explained by other unknown factors than the obesity epidemic and demographic changes. Possible other factors of importance might include EDCs, overall better psychosocial environment, and improved nutrition as well as improved health care.

that there is a robust secular trend for earlier pubertal timing in boys that is explained by other unknown factors than the obesity epidemic and demographic changes. Possible other factors of importance might include EDCs, overall better psychosocial environment, and improved nutrition as well as improved health care. Pubertal timing can be interpreted as a measure of exposure to sex steroids. In girls, early pubertal timing is associated with an increased risk for type 2 diabetes and breast cancer. Whether the secular trend for pubertal timing has implications for future health and disease in men represents a knowledge gap that will be important to address.

erpreted as a measure of exposure to sex steroids. In girls, early pubertal timing is associated with an increased risk for type 2 diabetes and breast cancer. Whether the secular trend for pubertal timing has implications for future health and disease in men represents a knowledge gap that will be important to address. Strengths and Limitations The strengths of this population-based study include the long study period with data on pubertal timing from 11 birth cohorts during 50 years that allowed us to distinguish between temporary changes and persistent trends. With information on childhood BMI available for the boys in the cohort, we were able to determine that the secular trend was robust and only partially explained by increasing childhood BMI. In addition, we demonstrated that the results are maintained when only participants born in Sweden and with parents born in Sweden were included. Furthermore, we used an objective measurement of pubertal timing in boys. This study’s limitations were that psychosocial and socioeconomic factors were not available. Moreover, we cannot completely exclude the possibility that there is a secular trend in the interval between the onset of puberty and the growth acceleration so that the onset of puberty is unaltered despite earlier growth acceleration. Because of the retrospective design of this study, information on secondary sex characteristics is not available. However, age at PHV, although a surrogate marker for pubertal timing, is derived from direct measurements of height, producing an objective estimate of pubertal timing. Age at PHV shows a strong association with pubertal timing retrieved from detailed longitudinal physical examinations of secondary sex characteristics. In contrast, self-reported pubertal timing in men displays a modest correlation with secondary sex characteristics as determined by a physician. In addition, if repeated height measurements are available, age at PHV can, in contrast to detailed longitudinal physical examinations of secondary sex characteristics, be easily estimated for many individuals. Another limitation might be that because our population has a narrow range of prepubertal BMI, we cannot eliminate the possibility that our study underestimated the importance of prepubertal BMI on pubertal timing in a contemporary population of boys exposed to the obesity epidemic.

sily estimated for many individuals. Another limitation might be that because our population has a narrow range of prepubertal BMI, we cannot eliminate the possibility that our study underestimated the importance of prepubertal BMI on pubertal timing in a contemporary population of boys exposed to the obesity epidemic. Conclusions We provide evidence of a secular trend for the earlier timing of pubertal PHV, a marker of pubertal timing, in Swedish boys. The secular trend of earlier pubertal timing is partially explained by increasing childhood BMI, but other unknown factors also contribute. Supplement. eMethods. eFigure 1. Flow chart of included participants eFigure 2. Age at PHV according to prepubertal BMI status at 8 years of age eFigure 3. Association between childhood BMI and age at PHV eTable 1. Birth cohort descriptives eTable 2. Associations for early and late age at PHV percentiles Click here for additional data file.

Introduction For decades, community water fluoridation has been used to prevent tooth decay. Water fluoridation is supplied to about 66% of US residents, 38% of Canadian residents, and 3% of European residents. In fluoridated communities, fluoride from water and beverages made with tap water makes up 60% to 80% of daily fluoride intake in adolescents and adults.

uoridation has been used to prevent tooth decay. Water fluoridation is supplied to about 66% of US residents, 38% of Canadian residents, and 3% of European residents. In fluoridated communities, fluoride from water and beverages made with tap water makes up 60% to 80% of daily fluoride intake in adolescents and adults. Fluoride crosses the placenta, and laboratory studies show that it accumulates in brain regions involved in learning and memory and alters proteins and neurotransmitters in the central nervous system. Higher fluoride exposure from drinking water has been associated with lower children’s intelligence in a meta-analysis of 27 epidemiologic studies and in studies including biomarkers of fluoride exposure. However, most prior studies were cross-sectional and conducted in regions with higher water fluoride concentrations (0.88-31.6 mg/L; to convert to millimoles per liter, multiply by 0.05263) than levels considered optimal (ie, 0.7 mg/L) in North America. Further, most studies did not measure exposure during fetal brain development. In a longitudinal birth cohort study involving 299 mother-child pairs in Mexico City, Mexico, a 1-mg/L increase in maternal urinary fluoride (MUF) concentration was associated with a 6-point (95% CI, −10.84 to −1.74) lower IQ score among school-aged children. In this same cohort, MUF was also associated with more attention-deficit/hyperactivity disorder–like symptoms. Urinary fluoride concentrations among pregnant women living in fluoridated communities in Canada are similar to concentrations among pregnant women living in Mexico City. However, it is unclear whether fluoride exposure during pregnancy is associated with cognitive deficits in a population receiving optimally fluoridated water.

luoride concentrations among pregnant women living in fluoridated communities in Canada are similar to concentrations among pregnant women living in Mexico City. However, it is unclear whether fluoride exposure during pregnancy is associated with cognitive deficits in a population receiving optimally fluoridated water. This study examined whether exposure to fluoride during pregnancy was associated with IQ scores in children in a Canadian birth cohort in which 40% of the sample was supplied with fluoridated municipal water. Methods Study Cohort Between 2008 and 2011, the Maternal-Infant Research on Environmental Chemicals (MIREC) program recruited 2001 pregnant women from 10 cities across Canada. Women who could communicate in English or French, were older than 18 years, and were within the first 14 weeks of pregnancy were recruited from prenatal clinics. Participants were not recruited if there was a known fetal abnormality, if they had any medical complications, or if there was illicit drug use during pregnancy. Additional details are in the cohort profile description.

than 18 years, and were within the first 14 weeks of pregnancy were recruited from prenatal clinics. Participants were not recruited if there was a known fetal abnormality, if they had any medical complications, or if there was illicit drug use during pregnancy. Additional details are in the cohort profile description. A subset of 610 children in the MIREC Study was evaluated for the developmental phase of the study at ages 3 to 4 years; these children were recruited from 6 of 10 cities included in the original cohort: Vancouver, Montreal, Kingston, Toronto, Hamilton, and Halifax. Owing to budgetary restraints, recruitment was restricted to the 6 cities with the most participants who fell into the age range required for the testing during the data collection period. Of the 610 children, 601 (98.5%) completed neurodevelopmental testing; 254 (42.3%) of these children lived in nonfluoridated regions and 180 (30%) lived in fluoridated regions; for 167 (27.7%) fluoridation status was unknown owing to missing water data or reported not drinking tap water (Figure 1). Figure 1. Flowchart of Inclusion Criteria MUF indicates maternal urinary fluoride. This study was approved by the research ethics boards at Health Canada, York University, and Indiana University. All women signed informed consent forms for both mothers and children.

A subset of 610 children in the MIREC Study was evaluated for the developmental phase of the study at ages 3 to 4 years; these children were recruited from 6 of 10 cities included in the original cohort: Vancouver, Montreal, Kingston, Toronto, Hamilton, and Halifax. Owing to budgetary restraints, recruitment was restricted to the 6 cities with the most participants who fell into the age range required for the testing during the data collection period. Of the 610 children, 601 (98.5%) completed neurodevelopmental testing; 254 (42.3%) of these children lived in nonfluoridated regions and 180 (30%) lived in fluoridated regions; for 167 (27.7%) fluoridation status was unknown owing to missing water data or reported not drinking tap water (Figure 1). Figure 1. Flowchart of Inclusion Criteria MUF indicates maternal urinary fluoride. This study was approved by the research ethics boards at Health Canada, York University, and Indiana University. All women signed informed consent forms for both mothers and children. Maternal Urinary Fluoride Concentration We used the mean concentrations of MUF measured in urine spot samples collected across each trimester of pregnancy at a mean (SD) of 11.57 (1.57), 19.11 (2.39), and 33.11 (1.50) weeks of gestation. Owing to the variability of urinary fluoride measurement and fluoride absorption during pregnancy, we only included women who had all 3 urine samples. In our previous work, these samples were moderately correlated; intraclass correlation coefficient (ICC) ranged from 0.37 to 0.40.

, and 33.11 (1.50) weeks of gestation. Owing to the variability of urinary fluoride measurement and fluoride absorption during pregnancy, we only included women who had all 3 urine samples. In our previous work, these samples were moderately correlated; intraclass correlation coefficient (ICC) ranged from 0.37 to 0.40. Urinary fluoride concentration was analyzed at the Indiana University School of Dentistry using a modification of the hexamethyldisiloxane (Sigma Chemical Co) microdiffusion procedure and described in our previous work. Fluoride concentration could be measured to 0.02 mg/L. We excluded 2 samples (0.002%) because the readings exceeded the highest concentration standard (5 mg/L) and there was less certainty of these being representative exposure values. To account for variations in urine dilution at the time of measurement, we adjusted MUF concentrations for specific gravity (SG) using the following equation: MUFSG = MUFi × (SGM-1)/(SGi-1), where MUFSG is the SG-adjusted fluoride concentration (in milligrams of fluoride per liter), MUFi is the observed fluoride concentration, SGi is the SG of the individual urine sample, and SGM is the median SG for the cohort. For comparison, we also adjusted MUF using the same creatinine adjustment method that was used in the 2017 Mexican cohort.

ted fluoride concentration (in milligrams of fluoride per liter), MUFi is the observed fluoride concentration, SGi is the SG of the individual urine sample, and SGM is the median SG for the cohort. For comparison, we also adjusted MUF using the same creatinine adjustment method that was used in the 2017 Mexican cohort. Water Fluoride Concentration Water treatment plants measured fluoride levels daily if fluoride was added to municipal drinking water and weekly or monthly if fluoride was not added to water. We matched participants’ postal codes with water treatment plant zones, allowing an estimation of water fluoride concentration for each woman by averaging water fluoride concentrations (in milligrams per liter) during the duration of pregnancy. We only included women who reported drinking tap water during pregnancy.

r. We matched participants’ postal codes with water treatment plant zones, allowing an estimation of water fluoride concentration for each woman by averaging water fluoride concentrations (in milligrams per liter) during the duration of pregnancy. We only included women who reported drinking tap water during pregnancy. Daily Fluoride Intake in Mothers We obtained information on consumption of tap water and other water-based beverages (tea and coffee) from a self-report questionnaire completed by mothers during the first and third trimesters. This questionnaire was used in the original MREC cohort and has not been validated. Also, for this study, we developed methods to estimate and calculate fluoride intake that have not yet been validated. To estimate fluoride intake from tap water consumed per day (milligrams per day), we multiplied each woman’s consumption of water and beverages by her water fluoride concentration (averaged across pregnancy) and multiplied by 0.2 (fluoride content for a 200-mL cup). Because black tea contains a high fluoride content (2.6 mg/L), we also estimated the amount of fluoride consumed from black tea by multiplying each cup of black tea by 0.52 mg (mean fluoride content in a 200-mL cup of black tea made with deionized water) and added this to the fluoride intake variable. Green tea also contains varying levels of fluoride; therefore, we used the mean for the green teas listed by the US Department of Agriculture (1.935 mg/L). We multiplied each cup of green tea by 0.387 mg (fluoride content in a 200-mL cup of green tea made with deionized water) and added this to the fluoride intake variable.

tea also contains varying levels of fluoride; therefore, we used the mean for the green teas listed by the US Department of Agriculture (1.935 mg/L). We multiplied each cup of green tea by 0.387 mg (fluoride content in a 200-mL cup of green tea made with deionized water) and added this to the fluoride intake variable. Primary Outcomes We assessed children’s intellectual abilities with the Wechsler Preschool and Primary Scale of Intelligence, Third Edition. Full Scale IQ (FSIQ), a measure of global intellectual functioning, was the primary outcome. We also assessed verbal IQ (VIQ), representing verbal reasoning and comprehension, and performance IQ (PIQ), representing nonverbal reasoning, spatial processing, and visual-motor skills. Covariates We selected covariates from a set of established factors associated with fluoride metabolism (eg, time of void and time since last void) and children’s intellectual abilities (eg, child sex, maternal age, gestational age, and parity) (Table 1). Mother’s race/ethnicity was coded as white or other, and maternal education was coded as either bachelor’s degree or higher or trade school diploma or lower. The quality of a child’s home environment was measured by the Home Observation for Measurement of the Environment (HOME)–Revised Edition on a continuous scale. We also controlled for city and, in some models, included self-reported exposure to secondhand smoke (yes/no) as a covariate.

or trade school diploma or lower. The quality of a child’s home environment was measured by the Home Observation for Measurement of the Environment (HOME)–Revised Edition on a continuous scale. We also controlled for city and, in some models, included self-reported exposure to secondhand smoke (yes/no) as a covariate. Table 1. Demographic Characteristics and Exposure Outcomes for Mother-Child Pairs With MUFSG (n = 512) and Fluoride Intake Data (n = 400) by Fluoridated and Nonfluoridated Statusa Variableb No. (%) MUFSG Sample (n = 512)c Maternal-Child Pairs With Fluoride Intake, IQ, and Complete Covariate Data (n = 400) Nonfluoridated (n = 238) Fluoridated (n = 162) Mothers Age of mother at enrollment, mean (SD), y 32.33 (5.07) 32.61 (4.90) 32.52 (4.03) Prepregnancy BMI, mean (SD) 25.19 (6.02) 25.19 (6.35) 24.33 (5.10) Married or common law 497 (97) 225 (95) 159 (98) Born in Canada 426 (83) 187 (79) 131 (81) White 463 (90) 209 (88) 146 (90) Maternal education Trade school diploma/high school 162 (32) 80 (34) 38 (24) Bachelor’s degree or higher 350 (68) 158 (66) 124 (76) Employed at time of pregnancy 452 (88) 205 (86) 149 (92) Net income household >$70 000 CAD 364 (71) 162 (68) 115 (71) HOME total score, mean (SD) 47.32 (4.32) 47.28 (4.48) 48.14 (3.90) Smoked in trimester 1 12 (2) 7 (3) 2 (1) Secondhand smoke in the home 18 (4) 9 (4) 2 (1) Alcohol consumption, alcoholic drink/mo None 425 (83) 192 (81) 136 (84) <1 41 (8) 23 (10) 11 (7) ≥1 46 (9) 23 (10) 15 (9) Parity (first birth) 233 (46) 119 (50) 71 (44) Children Female 264 (52) 118 (50) 83 (51) Age at testing, mean (SD), y 3.42 (0.32) 3.36 (0.31) 3.49 (0.29) Gestation, mean (SD), wk 39.12 (1.57) 39.19 (1.47) 39.17 (1.81) Birth weight, mean (SD), kg 3.47 (0.49) 3.48 (0.48) 3.47 (0.53) FSIQ 107.16 (13.26) 108.07 (13.31) 108.21 (13.72) Boysd 104.61 (14.09) 106.31 (13.60) 104.78 (14.71) Girlsd 109.56 (11.96) 109.86 (12.83) 111.47 (11.89) Exposure variables MUFSG concentration, mg/Le No. 512 228 141 Mean (SD) 0.51 (0.36) 0.40 (0.27) 0.69 (0.42) Fluoride intake level per day, mg No. 369a 238 162 Mean (SD) 0.54 (0.44) 0.30 (0.26) 0.93 (0.43) Water fluoride concentration, mg/L No. 369a 238 162 Mean (SD) 0.31 (0.23) 0.13 (0.06) 0.59 (0.08) Abbreviations: BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); CAD, Canadian dollars; FSIQ, Full Scale IQ; HOME, Home Observation for Measurement of the Environment; MUFSG, maternal urinary fluoride adjusted for specific gravity.

ean (SD) 0.31 (0.23) 0.13 (0.06) 0.59 (0.08) Abbreviations: BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); CAD, Canadian dollars; FSIQ, Full Scale IQ; HOME, Home Observation for Measurement of the Environment; MUFSG, maternal urinary fluoride adjusted for specific gravity. SI conversion factor: To convert fluoride to millimoles per liter, multiply by 0.05263. a Owing to missing water treatment plant data and/or MUF data, the samples are distinct with some overlapping participants in both groups (n = 369). b All of the listed variables were tested as potential covariates, as well as the following: paternal variables (age, education, employment status, smoking status, and race/ethnicity); maternal chronic condition during pregnancy and birth country; breastfeeding duration; and time of void and time since last void. c Maternal urinary fluoride (averaged across all 3 trimesters) and corrected for specific gravity. d The FSIQ score has a mean (SD) of 100 (15); US population norms used. e Owing to missing water treatment plant data, the samples in the fluoridated and nonfluoridated regions do not add up to the MUF sample size.

b All of the listed variables were tested as potential covariates, as well as the following: paternal variables (age, education, employment status, smoking status, and race/ethnicity); maternal chronic condition during pregnancy and birth country; breastfeeding duration; and time of void and time since last void. c Maternal urinary fluoride (averaged across all 3 trimesters) and corrected for specific gravity. d The FSIQ score has a mean (SD) of 100 (15); US population norms used. e Owing to missing water treatment plant data, the samples in the fluoridated and nonfluoridated regions do not add up to the MUF sample size. Statistical Analyses In our primary analysis, we used linear regression analyses to estimate the associations between our 2 measures of fluoride exposure (MUFSG and fluoride intake) and children’s FSIQ scores. In addition to providing the coefficient corresponding to a 1-mg difference in fluoride exposure, we also estimated coefficients corresponding to a fluoride exposure difference spanning the 25th to 75th percentile range (which corresponds to a 0.33 mg/L and 0.62 mg F/d difference in MUFSG and fluoride intake, respectively) as well as the 10th to 90th percentile range (which corresponds to a 0.70 mg/L and 1.04 mg F/d difference in MUFSG and fluoride intake, respectively).

e exposure difference spanning the 25th to 75th percentile range (which corresponds to a 0.33 mg/L and 0.62 mg F/d difference in MUFSG and fluoride intake, respectively) as well as the 10th to 90th percentile range (which corresponds to a 0.70 mg/L and 1.04 mg F/d difference in MUFSG and fluoride intake, respectively). We retained a covariate in the model if its P value was less than .20 or its inclusion changed the regression coefficient of the variable associated factor by more than 10% in any of the IQ models. Regression diagnostics confirmed that there were no collinearity issues in any of the IQ models with MUFSG or fluoride intake (variance inflation factor <2 for all covariates). Residuals from each model had approximately normal distributions, and their Q-Q plots revealed no extreme outliers. Plots of residuals against fitted values did not suggest any assumption violations and there were no substantial influential observations as measured by Cook distance. Including quadratic or natural-log effects of MUFSG or fluoride intake did not significantly improve the regression models. Thus, we present the more easily interpreted estimates from linear regression models. Additionally, we examined separate models with 2 linear splines to test whether the MUFSG association significantly differed between lower and higher levels of MUFSG based on 3 knots, which were set at 0.5 mg/L (mean MUFSG), 0.8 mg/L (threshold seen in the Mexican birth cohort), and 1 mg/L (optimal concentration in the United States until 2015). For fluoride intake, knots were set at 0.4 mg (mean fluoride intake), 0.8 mg, and 1 mg (in accordance with MUFSG). We also examined sex-specific associations in all models by testing the interactions between child sex and each fluoride measure.

ohort), and 1 mg/L (optimal concentration in the United States until 2015). For fluoride intake, knots were set at 0.4 mg (mean fluoride intake), 0.8 mg, and 1 mg (in accordance with MUFSG). We also examined sex-specific associations in all models by testing the interactions between child sex and each fluoride measure. In sensitivity analyses, we tested whether the associations between MUFSG and IQ were confounded by maternal blood concentrations of lead, mercury, manganese, perfluoro-octanoic acid, or urinary arsenic. We also conducted sensitivity analyses by removing IQ scores that were greater than or less than 2.5 standard deviations from the sample mean. Additionally, we examined whether using MUF adjusted for creatinine instead of SG affected the results. In additional analyses, we examined the association between our 2 measures of fluoride exposure (MUFSG and fluoride intake) with VIQ and PIQ. Additionally, we examined whether water fluoride concentration was associated with FSIQ, VIQ, and PIQ scores. For all analyses, statistical significance tests with a type I error rate of 5% were used to test sex interactions, while 95% confidence intervals were used to estimate uncertainty. Analyses were conducted using R software (the R Foundation). The P value level of significance was .05, and all tests were 2-sided.

In additional analyses, we examined the association between our 2 measures of fluoride exposure (MUFSG and fluoride intake) with VIQ and PIQ. Additionally, we examined whether water fluoride concentration was associated with FSIQ, VIQ, and PIQ scores. For all analyses, statistical significance tests with a type I error rate of 5% were used to test sex interactions, while 95% confidence intervals were used to estimate uncertainty. Analyses were conducted using R software (the R Foundation). The P value level of significance was .05, and all tests were 2-sided. Results For the first measure of fluoride exposure, MUFSG, 512 of 601 mother-child pairs (85.2%) who completed the neurodevelopmental visit had urinary fluoride levels measured at each trimester of the mother’s pregnancy and complete covariate data (Figure 1); 89 (14.8%) were excluded for missing MUFSG at 1 or more trimesters (n = 75) or missing 1 or more covariates included in the regression (n = 14) (Figure 1). Of the 512 mother-child pairs with MUFSG data (and all covariates), 264 children were female (52%). For the second measure of fluoride exposure, fluoride intake from maternal questionnaire, data were available for 400 of the original 601 mother-child pairs (66.6%): 201 women (33.4%) were excluded for reporting not drinking tap water (n = 59), living outside of the predefined water treatment plant zone (n = 108), missing beverage consumption data (n = 20), or missing covariate data (n = 14) (Figure 1).

re, data were available for 400 of the original 601 mother-child pairs (66.6%): 201 women (33.4%) were excluded for reporting not drinking tap water (n = 59), living outside of the predefined water treatment plant zone (n = 108), missing beverage consumption data (n = 20), or missing covariate data (n = 14) (Figure 1). Children had mean FSIQ scores in the average range (population normed) (mean [SD], 107.16 [13.26], range = 52-143), with girls (109.56 [11.96]) showing significantly higher scores than boys (104.61 [14.09]; P < .001) (Table 1). The demographic characteristics of the 512 mother-child pairs included in the primary analysis were not substantially different from the original MIREC cohort or subset of mother-child pairs without 3 urine samples (eTable 1 in the Supplement). Of the 400 mother-child pairs with fluoride intake data (and all covariates), 118 of 238 (50%) in the group living in a nonfluoridated region were female and 83 of 162 (51%) in the group living in a fluoridated region were female. Fluoride Measurements The median MUFSG concentration was 0.41 mg/L (range, 0.06-2.44 mg/L). Mean MUFSG concentration was significantly higher among women (n = 141) who lived in communities with fluoridated drinking water (0.69 [0.42] mg/L) compared with women (n=228) who lived in communities without fluoridated drinking water (0.40 [0.27] mg/L; P < .001) (Table 1; Figure 2). Figure 2. Distribution of Fluoride Levels in Maternal Urine and for Estimated Fluoride Intake by Fluoridation Status To convert fluoride to millimoles per liter, multiply by 0.05263.

Fluoride Measurements The median MUFSG concentration was 0.41 mg/L (range, 0.06-2.44 mg/L). Mean MUFSG concentration was significantly higher among women (n = 141) who lived in communities with fluoridated drinking water (0.69 [0.42] mg/L) compared with women (n=228) who lived in communities without fluoridated drinking water (0.40 [0.27] mg/L; P < .001) (Table 1; Figure 2). Figure 2. Distribution of Fluoride Levels in Maternal Urine and for Estimated Fluoride Intake by Fluoridation Status To convert fluoride to millimoles per liter, multiply by 0.05263. The median estimated fluoride intake was 0.39 mg per day (range, 0.01-2.65 mg). As expected, the mean (SD) fluoride intake was significantly higher for women (162 [40.5%]) who lived in communities with fluoridated drinking water (mean [SD], 0.93 [0.43] mg) than women (238 [59.5%]) who lived in communities without fluoridated drinking water (0.30 [0.26] mg; P < .001) (Table 1; Figure 2). The MUFSG was moderately correlated with fluoride intake (r = 0.49; P < .001) and water fluoride concentration (r = 0.37; P < .001). Maternal Urinary Fluoride Concentrations and IQ Before covariate adjustment, a significant interaction (P for interaction = .03) between MUFSG and child sex (B = 7.24; 95% CI, 0.81- 13.67) indicated that MUFSG was associated with FSIQ in boys; an increase of 1 mg/L MUFSG was associated with a 5.01 (95% CI, −9.06 to −0.97; P = .02) lower FSIQ score in boys. In contrast, MUFSG was not significantly associated with FSIQ score in girls (B = 2.23; 95% CI, −2.77 to 7.23; P = .38) (Table 2).

5% CI, 0.81- 13.67) indicated that MUFSG was associated with FSIQ in boys; an increase of 1 mg/L MUFSG was associated with a 5.01 (95% CI, −9.06 to −0.97; P = .02) lower FSIQ score in boys. In contrast, MUFSG was not significantly associated with FSIQ score in girls (B = 2.23; 95% CI, −2.77 to 7.23; P = .38) (Table 2). Table 2. Unadjusted and Adjusted Associations Estimated From Linear Regression Models of Fluoride Exposure Variables and FSIQ Scores Variable Difference (95% CI) Unadjusted Adjusted Estimates, Regression Coefficients Indicate Change in Outcome pera 1 mg 25th to 75th Percentiles 10th to 90th Percentiles MUFSGb,c −2.60 (−5.80 to 0.60) −1.95 (−5.19 to 1.28) −0.64 (−1.69 to 0.42) −1.36 (−3.58 to 0.90) Boys −5.01 (−9.06 to −0.97) −4.49 (−8.38 to −0.60) −1.48 (−2.76 to −0.19) −3.14 (−5.86 to −0.42) Girls 2.23 (−2.77 to 7.23) 2.40 (−2.53 to 7.33) 0.79 (−0.83 to 2.42) 1.68 (−1.77 to 5.13) Fluoride intaked,e −3.19 (−5.94 to −0.44) −3.66 (−7.16 to −0.15) −2.26 (−4.45 to −0.09) −3.80 (−7.46 to −0.16) Abbreviations: FSIQ, Full Scale IQ; HOME, Home Observation for Measurement of the Environment; MUFSG, maternal urinary fluoride adjusted for specific gravity. a Adjusted estimates pertain to predicted FSIQ difference for a value spanning the interquartile range (25th to 75th percentiles) and 80th central range (10th to 90th percentiles): (1) MUFSG: 0.33 mg/L, 0.70 mg/L, respectively; (2) fluoride intake: 0.62 mg, 1.04 mg, respectively. b n = 512. c Adjusted for city, HOME score, maternal education, race/ethnicity, and including child sex interaction. d n = 400.

a Adjusted estimates pertain to predicted FSIQ difference for a value spanning the interquartile range (25th to 75th percentiles) and 80th central range (10th to 90th percentiles): (1) MUFSG: 0.33 mg/L, 0.70 mg/L, respectively; (2) fluoride intake: 0.62 mg, 1.04 mg, respectively. b n = 512. c Adjusted for city, HOME score, maternal education, race/ethnicity, and including child sex interaction. d n = 400. e Adjusted for city, HOME score, maternal education, race/ethnicity, child sex, and prenatal secondhand smoke exposure. Adjusting for covariates, a significant interaction (P for interaction = .02) between child sex and MUFSG (B = 6.89; 95% CI, 0.96-12.82) indicated that an increase of 1 mg/L of MUFSG was associated with a 4.49 (95% CI, −8.38 to −0.60; P = .02) lower FSIQ score for boys. An increase from the 10th to 90th percentile of MUFSG was associated with a 3.14 IQ decrement among boys (Table 2; Figure 3). In contrast, MUFSG was not significantly associated with FSIQ score in girls (B = 2.43; 95% CI, −2.51 to 7.36; P = .33). Figure 3. Covariate Results of Multiple Linear Regression Models of Full Scale IQ (FSIQ) from Maternal Urinary Fluoride Concentration by Child Sex (n = 512) and Total Fluoride Intake Estimated from Daily Maternal Beverage Consumption (n = 400) B, Community fluoridation status (CWF) is shown for each woman; black dots represent women living in nonfluoridated (non-Fl) communities and blue dots represent women living in fluoridated (Fl) communities.

ion by Child Sex (n = 512) and Total Fluoride Intake Estimated from Daily Maternal Beverage Consumption (n = 400) B, Community fluoridation status (CWF) is shown for each woman; black dots represent women living in nonfluoridated (non-Fl) communities and blue dots represent women living in fluoridated (Fl) communities. Estimated Fluoride Intake and IQ A 1-mg increase in fluoride intake was associated with a 3.66 (95% CI, −7.16 to −0.15; P = .04) lower FSIQ score among boys and girls (Table 2; Figure 3). The interaction between child sex and fluoride intake was not statistically significant (B = 1.17; 95% CI, −4.08 to 6.41; P for interaction = .66). Sensitivity Analyses Adjusting for lead, mercury, manganese, perfluorooctanoic acid, or arsenic concentrations did not substantially change the overall estimates of MUFSG for boys or girls (eTable 2 in the Supplement). Use of MUF adjusted for creatinine did not substantially alter the associations with FSIQ (eTable 2 in the Supplement). Including time of void and time since last void did not substantially change the regression coefficient of MUFSG among boys or girls.

estimates of MUFSG for boys or girls (eTable 2 in the Supplement). Use of MUF adjusted for creatinine did not substantially alter the associations with FSIQ (eTable 2 in the Supplement). Including time of void and time since last void did not substantially change the regression coefficient of MUFSG among boys or girls. Estimates for determining the association between MUFSG and PIQ showed a similar pattern with a statistically significant interaction between MUFSG and child sex (P for interaction = .007). An increase of 1 mg/L MUFSG was associated with a 4.63 (95% CI, −9.01 to −0.25; P = .04) lower PIQ score in boys, but the association was not statistically significant in girls (B = 4.51; 95% CI, −1.02 to 10.05; P = .11). An increase of 1 mg/L MUFSG was not significantly associated with VIQ in boys (B = −2.85; 95% CI, −6.65 to 0.95; P = .14) or girls (B = 0.55; 95% CI, −4.28 to 5.37; P = .82); the interaction between MUFSG and child sex was not statistically significant (P for interaction = .25) (eTable 3 in the Supplement). Consistent with the findings on estimated maternal fluoride intake, increased water fluoride concentration (per 1 mg/L) was associated with a 5.29 (95% CI, −10.39 to −0.19) lower FSIQ score among boys and girls and a 13.79 (95% CI, −18.82 to −7.28) lower PIQ score (eTable 4 in the Supplement).

Estimates for determining the association between MUFSG and PIQ showed a similar pattern with a statistically significant interaction between MUFSG and child sex (P for interaction = .007). An increase of 1 mg/L MUFSG was associated with a 4.63 (95% CI, −9.01 to −0.25; P = .04) lower PIQ score in boys, but the association was not statistically significant in girls (B = 4.51; 95% CI, −1.02 to 10.05; P = .11). An increase of 1 mg/L MUFSG was not significantly associated with VIQ in boys (B = −2.85; 95% CI, −6.65 to 0.95; P = .14) or girls (B = 0.55; 95% CI, −4.28 to 5.37; P = .82); the interaction between MUFSG and child sex was not statistically significant (P for interaction = .25) (eTable 3 in the Supplement). Consistent with the findings on estimated maternal fluoride intake, increased water fluoride concentration (per 1 mg/L) was associated with a 5.29 (95% CI, −10.39 to −0.19) lower FSIQ score among boys and girls and a 13.79 (95% CI, −18.82 to −7.28) lower PIQ score (eTable 4 in the Supplement). Discussion Using a prospective Canadian birth cohort, we found that estimated maternal exposure to higher fluoride levels during pregnancy was associated with lower IQ scores in children. This association was supported by converging findings from 2 measures of fluoride exposure during pregnancy. A difference in MUFSG spanning the interquartile range for the entire sample (ie, 0.33 mg/L), which is roughly the difference in MUFSG concentration for pregnant women living in a fluoridated vs a nonfluoridated community, was associated with a 1.5-point IQ decrement among boys. An increment of 0.70 mg/L in MUFSG concentration was associated with a 3-point IQ decrement in boys; about half of the women living in a fluoridated community have a MUFSG equal to or greater than 0.70 mg/L. These results did not change appreciably after controlling for other key exposures such as lead, arsenic, and mercury.

rement of 0.70 mg/L in MUFSG concentration was associated with a 3-point IQ decrement in boys; about half of the women living in a fluoridated community have a MUFSG equal to or greater than 0.70 mg/L. These results did not change appreciably after controlling for other key exposures such as lead, arsenic, and mercury. To our knowledge, this study is the first to estimate fluoride exposure in a large birth cohort receiving optimally fluoridated water. These findings are consistent with that of a Mexican birth cohort study that reported a 6.3 decrement in IQ in preschool-aged children compared with a 4.5 decrement for boys in our study for every 1 mg/L of MUF. The findings of the current study are also concordant with ecologic studies that have shown an association between higher levels of fluoride exposure and lower intellectual abilities in children. Collectively, these findings support that fluoride exposure during pregnancy may be associated with neurocognitive deficits.

The findings of the current study are also concordant with ecologic studies that have shown an association between higher levels of fluoride exposure and lower intellectual abilities in children. Collectively, these findings support that fluoride exposure during pregnancy may be associated with neurocognitive deficits. In contrast with the Mexican study, the association between higher MUFSG concentrations and lower IQ scores was observed only in boys but not in girls. Studies of fetal and early childhood fluoride exposure and IQ have rarely examined differences by sex; of those that did, some reported no differences by sex. Most rat studies have focused on fluoride exposure in male rats, although 1 study showed that male rats were more sensitive to neurocognitive effects of fetal exposure to fluoride. Testing whether boys are potentially more vulnerable to neurocognitive effects associated with fluoride exposure requires further investigation, especially considering that boys have a higher prevalence of neurodevelopmental disorders such as ADHD, learning disabilities, and intellectual disabilities. Adverse effects of early exposure to fluoride may manifest differently for girls and boys, as shown with other neurotoxicants.

ure requires further investigation, especially considering that boys have a higher prevalence of neurodevelopmental disorders such as ADHD, learning disabilities, and intellectual disabilities. Adverse effects of early exposure to fluoride may manifest differently for girls and boys, as shown with other neurotoxicants. The estimate of maternal fluoride intake during pregnancy in this study showed that an increase of 1 mg of fluoride was associated with a decrease of 3.7 IQ points across boys and girls. The finding observed for fluoride intake in both boys and girls may reflect postnatal exposure to fluoride, whereas MUF primarily captures prenatal exposure. Importantly, we excluded women who reported that they did not drink tap water and matched water fluoride measurements to time of pregnancy when estimating maternal fluoride intake. None of the fluoride concentrations measured in municipal drinking water were greater than the maximum acceptable concentration of 1.5 mg/L set by Health Canada; most (94.3%) were lower than the 0.7 mg/L level considered optimal.

water fluoride measurements to time of pregnancy when estimating maternal fluoride intake. None of the fluoride concentrations measured in municipal drinking water were greater than the maximum acceptable concentration of 1.5 mg/L set by Health Canada; most (94.3%) were lower than the 0.7 mg/L level considered optimal. Water fluoridation was introduced in the 1950s to prevent dental caries before the widespread use of fluoridated dental products. Originally, the US Public Health Service set the optimal fluoride concentrations in water from 0.7 to 1.2 mg/L to achieve the maximum reduction in tooth decay and minimize the risk of enamel fluorosis. Fluorosis, or mottling, is a symptom of excess fluoride intake from any source occurring during the period of tooth development. In 2012, 68% of adolescents had very mild to severe enamel fluorosis. The higher prevalence of enamel fluorosis, especially in fluoridated areas, triggered renewed concern about excessive ingestion of fluoride. In 2015, in response to fluoride overexposure and rising rates of enamel fluorosis, the US Public Health Service recommended an optimal fluoride concentration of 0.7 mg/L, in line with the recommended level of fluoride added to drinking water in Canada to prevent caries. However, the beneficial effects of fluoride predominantly occur at the tooth surface after the teeth have erupted. Therefore, there is no benefit of systemic exposure to fluoride during pregnancy for the prevention of caries in offspring. The evidence showing an association between fluoride exposure and lower IQ scores raises a possible new concern about cumulative exposures to fluoride during pregnancy, even among pregnant women exposed to optimally fluoridated water.

emic exposure to fluoride during pregnancy for the prevention of caries in offspring. The evidence showing an association between fluoride exposure and lower IQ scores raises a possible new concern about cumulative exposures to fluoride during pregnancy, even among pregnant women exposed to optimally fluoridated water. Strengths and Limitations Our study has several strengths and limitations. First, urinary fluoride has a short half-life (approximately 5 hours) and depends on behaviors that were not controlled in our study, such as consumption of fluoride-free bottled water or swallowing toothpaste prior to urine sampling. We minimized this limitation by using 3 serial urine samples and tested for time of urine sample collection and time since last void, but these variables did not alter our results. Second, although higher maternal ingestion of fluoride corresponds to higher fetal plasma fluoride levels, even serial maternal urinary spot samples may not precisely represent fetal exposure throughout pregnancy. Third, while our analyses controlled for a comprehensive set of covariates, we did not have maternal IQ data. However, there is no evidence suggesting that fluoride exposure differs as a function of maternal IQ; our prior study did not observe a significant association between MUF levels and maternal education level. Moreover, a greater proportion of women living in fluoridated communities (124 [76%]) had a university-level degree compared with women living in nonfluoridated communities (158 [66%]). Nonetheless, despite our comprehensive array of covariates included, this observational study design could not address the possibility of other unmeasured residual confounding. Fourth, fluoride intake did not measure actual fluoride concentration in tap water in the participant’s home; Toronto, for example, has overlapping water treatment plants servicing the same household. Similarly, our fluoride intake estimate only considered fluoride from beverages; it did not include fluoride from other sources such as dental products or food. Furthermore, fluoride intake data were limited by self-report of mothers’ recall of beverage consumption per day, which was sampled at 2 points of pregnancy, and we lacked information regarding specific tea brand. In addition, our methods of estimating maternal fluoride intake have not been validated; however, we show construct validity with MUF. Fifth, this study did not include assessment of postnatal fluoride exposure or consumption.

was sampled at 2 points of pregnancy, and we lacked information regarding specific tea brand. In addition, our methods of estimating maternal fluoride intake have not been validated; however, we show construct validity with MUF. Fifth, this study did not include assessment of postnatal fluoride exposure or consumption. However, our future analyses will assess exposure to fluoride in the MIREC cohort in infancy and early childhood. Conclusions In this prospective birth cohort study from 6 cities in Canada, higher levels of fluoride exposure during pregnancy were associated with lower IQ scores in children measured at age 3 to 4 years. These findings were observed at fluoride levels typically found in white North American women. This indicates the possible need to reduce fluoride intake during pregnancy. Supplement. eTable 1. Comparison of Current Sample to Other MIREC Samples eTable 2. Sensitivity Analyses Predicting Full Scale IQ (FSIQ) eTable 3. Unadjusted and Adjusted Effect Estimates From Linear Regression Models Of Fluoride Exposure Variables Predicting Verbal IQ and Performance IQ Scores eTable 4. Unadjusted and Adjusted Effect Estimates From Linear Regression Models of Water Fluoride Concentration (mg/L) Predicting FSIQ Scores Click here for additional data file.

Introduction Research demonstrates that both positive and adverse experiences shape brain development and health across the life span.. Understanding human development requires a model that incorporates both risks (factors that decrease the likelihood of successful development) and opportunities (factors that increase the likelihood of successful development). On the positive side, successful child development depends on secure attachment during the first years of life. As the child grows, exposure to spoken language and having the presence of safe, stable, nurturing relationships and environments are important factors for optimal development. On the other hand, children with adverse childhood experiences (ACEs) are at risk for observable changes in brain anatomy, gene expression, and delays in social, emotional, physical, and cognitive development lasting into adulthood.

nurturing relationships and environments are important factors for optimal development. On the other hand, children with adverse childhood experiences (ACEs) are at risk for observable changes in brain anatomy, gene expression, and delays in social, emotional, physical, and cognitive development lasting into adulthood. According to standardized measures, an estimated 61.5% of adults and 48% of children in the United States have been exposed to ACEs, with more than one-third of these having multiple exposures. The wide-ranging negative associations between exposure to multiple ACEs and diminished adult and child health are well documented. Most notable is the especially strong evidence linking ACEs with adult mental health problems including depression. A robust literature also exists regarding the effect of ACEs on adult relational health (often assessed by whether adults report that they get the social and emotional support they need) and how diminished adult social and emotional support contributes to poorer adult physical and mental health.

s including depression. A robust literature also exists regarding the effect of ACEs on adult relational health (often assessed by whether adults report that they get the social and emotional support they need) and how diminished adult social and emotional support contributes to poorer adult physical and mental health. Beyond the extensive and growing body of research dealing with lifelong correlates of adversity, many prior studies identify resiliency factors and adaptive skills and interventions associated with improved child development and child and adult health outcomes. For example, the Search Institute developed a list of “40 Developmental Assets” and demonstrated associations between the number of assets and both positive and negative outcomes. A national population-based study on child flourishing and resilience shows strong associations with levels of family resilience and parent-child connection for children with exposures to greater ACEs, poverty, and chronic conditions. Similar studies, such as those assessing the US Centers for Disease Control and Prevention (CDC)’s “safe, stable, and nurturing relationships” model, show similar findings.

ciations with levels of family resilience and parent-child connection for children with exposures to greater ACEs, poverty, and chronic conditions. Similar studies, such as those assessing the US Centers for Disease Control and Prevention (CDC)’s “safe, stable, and nurturing relationships” model, show similar findings. Despite these advances, standardized measures and the prevalence of positive childhood experiences (PCEs) at the population level for adults or children are still unknown. Yet prior studies, using data from small or nonrepresentative samples, have explored interactions between PCEs and ACEs. For example, 1 study, conducted by Kaiser Permanente and CDC investigators, analyzed a cohort of 4648 women. They found that adult reports of specific positive family experiences in childhood (including closeness, support, loyalty, protection, love, importance, and responsiveness to health needs) were associated with lower rates of adolescent pregnancy across all ACEs exposure levels. The protective effects of reported interpersonal PCEs against mental health problems in adulthood have also been found among pregnant women and young adults exposed to ACEs. Despite these findings, few subsequent studies on ACEs have simultaneously evaluated PCEs.

f adolescent pregnancy across all ACEs exposure levels. The protective effects of reported interpersonal PCEs against mental health problems in adulthood have also been found among pregnant women and young adults exposed to ACEs. Despite these findings, few subsequent studies on ACEs have simultaneously evaluated PCEs. Collectively, prior studies on child development point to the importance of research focusing on PCEs, especially those associated with parent-child attachment, positive parenting (eg, parental warmth, responsiveness, and support), family health, and positive relationships with friends, in school, and in the community. Knowledge of whether retrospectively reported PCEs co-occur with ACEs and how PCEs interact with ACEs to effect adult mental and relational health is needed to inform the nation’s growing focus on addressing early life and social determinants of healthy development and lifelong health.

in school, and in the community. Knowledge of whether retrospectively reported PCEs co-occur with ACEs and how PCEs interact with ACEs to effect adult mental and relational health is needed to inform the nation’s growing focus on addressing early life and social determinants of healthy development and lifelong health. This study used data from the 2015 Wisconsin Behavioral Risk Factor Survey (WI BRFS), a representative, population-based survey, to assess the prevalence of PCEs in an adult sample and evaluate hypothesized associations with adult mental and relational health across 4 ACEs exposure levels. This study builds on a 2017 Health Outcomes of Positive Experiences report featuring bivariate findings from the 2015 WI BRFS associating individual PCEs with negative adult health outcomes. Here, we construct a PCEs cumulative score measure and use multivariable regression methods to assess the magnitude and significance of associations between this PCEs score and (1) adult depression and/or poor mental health (D/PMH) and (2) adults’ reported social and emotional support (ARSES). Separate assessment of associations was conducted for each of 4 ACEs exposure levels.

ultivariable regression methods to assess the magnitude and significance of associations between this PCEs score and (1) adult depression and/or poor mental health (D/PMH) and (2) adults’ reported social and emotional support (ARSES). Separate assessment of associations was conducted for each of 4 ACEs exposure levels. Methods Population and Data Data were from the cross-sectional 2015 WI BRFS, a representative, telephone survey of noninstitutionalized Wisconsin adults 18 years and older who speak English or Spanish (n = 6188). The WI BRFS response rate was 45.0% (weighted American Association of Public Opinion Research median, 47.2%). The cooperation rate was 64.9% (weighted American Association of Public Opinion Research median, 68.0%). The 2015 WI BRFS core and state-added items data sets were linked. Institutional review board (IRB) approval was not required because data are based on a survey conducted by a public agency and do not include personal health information. Respondent oral consent methods and construction of race/ethnicity variables used standard CDC BRFS approved methods.

ed items data sets were linked. Institutional review board (IRB) approval was not required because data are based on a survey conducted by a public agency and do not include personal health information. Respondent oral consent methods and construction of race/ethnicity variables used standard CDC BRFS approved methods. There were 18.1% to 21.1% missing cases for state-added ARSES, ACEs, and PCEs items. “Don’t know/refused” responses to these questions were 0.2% to 0.9%. A 10% missing value rate for the WI BRFS state-added items is expected and is attributed to the administration of the core WI BRFS survey by another state to Wisconsin residents who have out-of-state cellular phones. In these cases, the WI BRFS state-added items were not available to be administered. The remainder of missing cases were nearly all owing to respondent dropoffs prior to administering the ARSES, ACEs, and PCEs questions after administration of the core WI BRFS. Differences in D/PMH prevalence rates between respondents and missing cases were not notable. See eTable 1 in the Supplement for additional details.

emainder of missing cases were nearly all owing to respondent dropoffs prior to administering the ARSES, ACEs, and PCEs questions after administration of the core WI BRFS. Differences in D/PMH prevalence rates between respondents and missing cases were not notable. See eTable 1 in the Supplement for additional details. Key Measures Positive Childhood Experiences Score The PCEs score included 7 items asking respondents to report how often or how much as a child they: (1) felt able to talk to their family about feelings; (2) felt their family stood by them during difficult times; (3) enjoyed participating in community traditions; (4) felt a sense of belonging in high school (not including those who did not attend school or were home schooled); (5) felt supported by friends; (6) had at least 2 nonparent adults who took genuine interest in them; and (7) felt safe and protected by an adult in their home. The PCEs score items were adapted from 4 subscales included in the Child and Youth Resilience Measure–28 : (1) 4 items from the Psychological, Caregiving subscale (see PCEs items 1, 2, 7, and 6 listed previously); (2) 1 from the Education subscale (PCEs item 4); (3) 1 from the Culture subscale (PCEs item 3), and (4) 1 from the Peer Support subscale (PCEs item 5). Items were designed in the Child and Youth Resilience Measure–28 for cultural sensitivity, and their validity was supported by associations with improved resilience. Psychometric analyses confirmed use of a PCEs cumulative score. See eTable 2 in the Supplement for details.

from the Peer Support subscale (PCEs item 5). Items were designed in the Child and Youth Resilience Measure–28 for cultural sensitivity, and their validity was supported by associations with improved resilience. Psychometric analyses confirmed use of a PCEs cumulative score. See eTable 2 in the Supplement for details. Adverse Childhood Experiences We used data from the standardized ACEs survey items defined by the CDC. The ACEs measure included 11 ACEs items assessing recollections of childhood experiences of physical or emotional abuse or neglect, sexual abuse, and household dysfunctions such as substance abuse, parental incarceration, and divorce. As recommended by the CDC, items were coded using cumulative score groupings of 0, 1, 2 to 3, or 4 to 8 ACEs. Subjective reports of experiences in childhood are the intended construct for assessment of both PCEs and ACEs and not whether what is reported would be validated using objective assessments. Adult-Reported Social and Emotional Support Adult-reported social and emotional support is assessed using a standardized single item, “How often do you get the social and emotional support you need?” Response choices were “always,” “usually,” “sometimes,” “rarely,” or “never.” Based on previous research and analysis of this ARSES variable, this study separately evaluated “always” and “usually” responses and created a combined “sometimes/rarely/never” response category.

et the social and emotional support you need?” Response choices were “always,” “usually,” “sometimes,” “rarely,” or “never.” Based on previous research and analysis of this ARSES variable, this study separately evaluated “always” and “usually” responses and created a combined “sometimes/rarely/never” response category. Depression/Poor Mental Health The D/PMH category was constructed using (1) the single item on depression asking whether a physician or other health professional “ever told you that you have a depressive disorder, including depression, major depression, dysthymia, or minor depression?”; and (2) a score of 14 or higher on the single item validated as an indicator of current poor mental health that asked, “Now thinking about your mental health, which includes stress, depression, and problems with emotions, for how many days during the past 30 days was your mental health not good?” Adults reporting either or both of these outcomes were included in the D/PMH variable. Other Covariates Demographic covariates included age (18-34 years, 35-54 years, 55-64 years, and 65 years or older), race/ethnicity (nonwhite or white/non-Hispanic), and annual income (less than $25 000, $25 000-$49 999, $50 000-$74 999, and $75 000 or more). Sample size and statistical power analysis findings required combining race/ethnicity subgroups into 2 categories for purposes of statistical analysis.

, and 65 years or older), race/ethnicity (nonwhite or white/non-Hispanic), and annual income (less than $25 000, $25 000-$49 999, $50 000-$74 999, and $75 000 or more). Sample size and statistical power analysis findings required combining race/ethnicity subgroups into 2 categories for purposes of statistical analysis. Analytic Methods Prevalence rates for all variables were computed, and bivariate associations between individual PCE items and PCEs cumulative score groups and all other variables were evaluated using χ2 tests. Iterative and recursive analyses confirmed independent variable construction and focused on confirmation of assumptions on the linearity and comparability of associations with study outcomes when ordinal (count) or cumulative score groupings of PCEs and ACEs were used. Cumulative score groups of 0 to 2, 3 to 5, and 6 to 7 PCEs and 0, 1, 2 to 3, and 4 to 8 ACEs were also selected to ensure adequate statistical power to detect meaningful associations. Such score groups also simplify reporting of results by narrowing the number of comparative groups requiring reporting. Interaction variables crossing PCEs by ACEs and PCEs by ARSES were also analyzed for each study outcome and supported decisions to assess PCEs, ACEs, and ARSES as independent (vs interacting) variables in regression models.

simplify reporting of results by narrowing the number of comparative groups requiring reporting. Interaction variables crossing PCEs by ACEs and PCEs by ARSES were also analyzed for each study outcome and supported decisions to assess PCEs, ACEs, and ARSES as independent (vs interacting) variables in regression models. As noted, multivariable logistic regression analyses evaluated 2 association pathways between PCEs items and cumulative score groups and 2 outcome variables: (1) meeting criteria for D/PMH and (2) reports of “always” on ARSES. Regression models were adjusted for age, sex, race/ethnicity, income, and ACEs. Separate models were evaluated for each ACEs exposure level (0, 1, 2-3, and 4-8) to examine stability of associations across ACEs exposure levels. We further assessed the stability of associations between D/PMH and PCEs when ARSES were or were not controlled for in regression models. This was done to further understand more nuanced association pathways between PCEs and ARSES and their individual or interacting association with D/MPH. Additional sensitivity analyses of PCEs associations when ACEs were or were not included in models were also conducted. The survey data were weighted to be representative of the Wisconsin population. We used SPSS Complex Samples, version 24 (IBM Corporation) for data analysis. A P value of .05 or less was used to determine statistical significance.

s of PCEs associations when ACEs were or were not included in models were also conducted. The survey data were weighted to be representative of the Wisconsin population. We used SPSS Complex Samples, version 24 (IBM Corporation) for data analysis. A P value of .05 or less was used to determine statistical significance. Results Population Characteristics and Prevalence of Study Outcomes by PCEs Demographic characteristics for the 2015 WI BRFS mirrored the state population: 50.7% women and 84.9% white. About half (52.3%) reported 6 to 7 PCEs, more than half (56.7%) reported ACEs, 21.2% met D/PMH criteria, and more than half (55.1%) reported “always” to getting the social and emotional support they needed (ARSES). Nonwhite, younger, and lower-income adults reported fewer levels of PCEs (Table 1). Compared with those reporting 6 to 7 PCEs, adults reporting 0 to 2 PCEs had nearly 4 times higher prevalence of D/PMH (48.2% vs 12.6%) and were half as likely to report “always” to getting the social and emotional support they needed (33.0% vs 67.9%) (Table 2). Similar variations in prevalence were observed when each of the 7 PCEs items were separately evaluated for each study outcome (Figure 1 and Figure 2). As hypothesized and shown in these Figures, stronger associations emerged for cumulative PCEs scores.

ocial and emotional support they needed (33.0% vs 67.9%) (Table 2). Similar variations in prevalence were observed when each of the 7 PCEs items were separately evaluated for each study outcome (Figure 1 and Figure 2). As hypothesized and shown in these Figures, stronger associations emerged for cumulative PCEs scores. Table 1. Study Population Characteristics and Prevalence of PCEs by D/PMH, ACEs, ARSES, and Demographic Characteristics Population Characteristics (n = Unweighted Sample Size) Statewide Population Prevalence Estimates Prevalence of PCEs (n = 4926)a P Value (Test of Independence) 0-2 PCEs 3-5 PCEs 6-7 PCEs Unweighted No. Weighted % Unweighted No. Weighted % Unweighted No. Weighted % Unweighted No. Weighted % All respondents 6188 100 635 13.2 1606 34.5 2685 52.3 NA D/PMH (n = 6187) Yes 1289 21.2 294 29.4 402 40.1 347 30.5 <.001 No 4898 78.8 341 8.7 1204 33.0 2338 58.3 ACEs exposure levels (n = 4974)a,b 0 ACEs 2275 43.3 106 4.9 567 27.3 1568 67.8 <.001 1 ACE 1142 23.0 100 8.3 406 38.6 625 53.1 2-3 ACEs 967 19.9 174 18.5 400 42.1 390 39.5 4-8 ACEs 590 13.7 255 39.4 232 39.4 100 21.2 ARSES (n = 5021)a Always 2707 55.1 195 7.9 687 27.3 1743 64.8 <.001 Usually 1337 25.8 171 12.9 507 41.9 635 45.2 Sometimes, rarely, or never 977 19.1 263 28.7 393 44.7 284 26.6 Age (n = 6127), y 18-34 977 28.8 98 13.0 267 37.9 350 49.2 .03 35-54 1737 33.0 201 15.6 407 31.9 748 52.5 55-64 1426 17.6 169 12.6 389 36.0 613 51.4 65 or older 1987 20.5 163 10.4 532 33.1 954 56.5 Sex (n = 6188) Male 2720 49.3 248 11.9 763 36.3 1133 51.7 .09 Female 3468 50.7 387 14.3 843 32.8 1552 52.9 Race/ethnicity (n = 6129) Nonwhite 757 15.1 107 17.0 208 44.7 233 38.3 <.001 White, non-Hispanic 5372 84.9 521 12.6 1385 33.1 2433 54.3 Income level (n = 5461),c $ <24 999 1331 22.5 219 22.0 387 38.3 437 39.6 <.001 25 000-49 999 1511 27.8 168 14.9 431 36.9 631 48.3 50 000-74 999 1010 18.9 83 9.7 288 39.1 465 51.3 75 000 or more 1609 30.7 105 8.2 334 25.9 888 66.0 Abbreviations: ACEs, adverse childhood experiences; ARSES, adult-reported social and emotional support; D/PMH, depression and/or poor mental health; NA, not applicable; PCEs, positive childhood experiences; WI BRFS, Wisconsin Behavioral Risk Factor Survey.

9.1 465 51.3 75 000 or more 1609 30.7 105 8.2 334 25.9 888 66.0 Abbreviations: ACEs, adverse childhood experiences; ARSES, adult-reported social and emotional support; D/PMH, depression and/or poor mental health; NA, not applicable; PCEs, positive childhood experiences; WI BRFS, Wisconsin Behavioral Risk Factor Survey. a A 10% missing value rate is expected and attributed to core WI BRFS survey administration to out-of-state cellular phone holders who never received the WI BRFS state added items. The remainder were nearly all owing to respondent dropoffs prior to administering the ARSES, ACEs, and PCEs questions, which were administered after the end of the core WI BRFS. No notable differences in prevalence of D/PMH were found between respondents and cases missing ARSES, ACEs, or PCEs data. See eTable 1 in the Supplement. b The ACEs cumulative scores were created placing adults into categories of 0, 1, 2 to 3, or 4 to 8 ACEs based on their responses to the 11 ACEs items. Three sexual abuse items were combined into a single item, and alcohol and substance abuse items were presented as a single ACEs item. c Income missing values rate was 11.7%.

b The ACEs cumulative scores were created placing adults into categories of 0, 1, 2 to 3, or 4 to 8 ACEs based on their responses to the 11 ACEs items. Three sexual abuse items were combined into a single item, and alcohol and substance abuse items were presented as a single ACEs item. c Income missing values rate was 11.7%. Table 2. Prevalence and Adjusted Odds Ratios of Adult D/PMH and Reports of “Always” on the ARSES Item by PCEs and Other Regression Model Variables Population Characteristics (Raw Sample Size) Prevalence of D/PMH P Value Adjusted Odds Ratio (95% CI) for Meeting D/PMH Criteria Prevalence of “Always” on ARSES Item P Value Adjusted Odds Ratio (95% CI) for Reports of “Always” on ARSES Itema Unweighted No. Weighted % Unweighted No. Weighted % All Respondents 1289 21.2 NA NA 2707 55.1 NA NA Positive childhood experiences (PCEs) (n = 4926)a,b,c 0-2 PCEs reported 294 48.2 <.001 1 [Reference] 195 33.0 <.001 1 [Reference] 3-5 PCEs reported 402 25.1 0.50 (0.36-0.69) 687 43.6 1.31 (0.97-1.78) 6-7 PCEs reported 347 12.6 0.28 (0.21-0.39) 1743 67.9 3.53 (2.60-4.80) Adverse childhood experiences (ACEs) (n = 4974)a No ACEs reported 252 11.9 <.001 1 [Reference] 1394 62.4 <.001 1.22 (0.88-1.69) 1 ACE reported 215 20.2 1.62 (1.18-2.21) 596 53.9 0.93 (0.67-1.30) 2-3 ACEs reported 294 9.2 2.40 (1.77-3.24) 439 47.6 0.90 (0.64-1.27) 4-8 ACEs reported 285 42.4 3.10 (2.20-4.37) 226 44.2 1 [Reference] Age (n = 6127), y 18-34 215 21.0 .01 1.09 (0.78-1.53) 408 56.8 .44 1.09 (0.84-1.42) 35-54 406 22.6 1.51 (1.10-2.06) 766 54.9 0.97 (0.76-1.23) 55-64 331 24.2 1.64 (1.20-2.24) 600 52.1 0.88 (0.69-1.13) 65 or older 332 16.9 1 [Reference] 911 55.8 1 [Reference] Sex (n = 6188) Male 444 16.9 <.001 0.59 (0.47-0.74) 1189 55.3 .80 0.97 (0.81-1.17) Female 845 25.5 1 [Reference] 1518 54.8 1 [Reference] Race/ethnicity (n = 6129) Nonwhite 203 23.8 <.25 0.98 (0.67-1.42) 294 53.5 .64 1.19 (0.84-1.70) White, non-Hispanic 1078 20.9 1 [Reference] 2391 55.2 1 [Reference] Income level (n = 5461),d $ <24 999 454 33.3 <.001 2.91 (2.11-4.02) 465 47.8 <.001 0.67 (0.51-0.88) 25 000-49 999 340 22.6 1.76 (1.29-2.41) 667 53.4 0.81 (0.64-1.03) 50 000-74 999 172 18.4 1.43 (1.02-2.01) 458 54.3 0.81 (0.62-1.05) 75 000 or more 205 13.1 1 [Reference] 857 62.3 1 [Reference] Abbreviations: ACEs, adverse childhood experiences; ARSES, adult-reported social and emotional support; D/PMH, depression and/or poor mental health; NA, not applicable; PCEs, positive childhood experiences; WI BRFS, Wisconsin Behavioral Risk Factor Survey.

5) 75 000 or more 205 13.1 1 [Reference] 857 62.3 1 [Reference] Abbreviations: ACEs, adverse childhood experiences; ARSES, adult-reported social and emotional support; D/PMH, depression and/or poor mental health; NA, not applicable; PCEs, positive childhood experiences; WI BRFS, Wisconsin Behavioral Risk Factor Survey. a A 10% missing value rate is expected and attributed to core WI BRFS 5 survey administration to out-of-state cellular phone holders who never received the WI BRFS state added items. The remainder were nearly all owing to respondent dropoffs prior to administering the ARSES, ACEs, and PCEs questions, which were administered after the end of the core WI BRFS. No notable differences in prevalence of D/PMH were found between respondents and cases missing ARSES, ACEs, or PCEs data. See eTable 1 in the Supplement. b Without adjustment for ACEs, PCEs associations with D/PMH were 0.19 (95% CI, 0.14-0.25) and 0.40 (95% CI, 0.30-0.54) for adults reporting 6 to 7 and 3 to 5 PCEs vs 0 to 2 PCEs, respectively. c Without adjustment for ACEs, PCEs associations with “always” on the ARSES variable were 3.83 (95% CI, 2.89-5.06) and 1.35 (95% CI, 1.01-1.81) for adults reporting 6 to 7 and 3 to 5 PCEs vs 0 to 2 PCEs, respectively. d Income missing values rate is 11.7%. Income was not imputed for the WI BRFS by the Wisconsin Department of Health Services so federal poverty level could not be calculated.

c Without adjustment for ACEs, PCEs associations with “always” on the ARSES variable were 3.83 (95% CI, 2.89-5.06) and 1.35 (95% CI, 1.01-1.81) for adults reporting 6 to 7 and 3 to 5 PCEs vs 0 to 2 PCEs, respectively. d Income missing values rate is 11.7%. Income was not imputed for the WI BRFS by the Wisconsin Department of Health Services so federal poverty level could not be calculated. Figure 1. Prevalence of Depression and/or Poor Mental Health Among Adults by Positive Childhood Experiences (PCEs) Single Items and Cumulative Scores aSource: authors’ analysis of the 2015 Wisconsin Behavioral Risk Factor Survey. bAdjusted odds ratios (AORs) shown are adjusted for age, sex, race/ethnicity, income, and adverse childhood experiences. cNever, rarely, or sometimes is the reference category. dNever, a little, or some of the time is the reference category. Figure 2. Prevalence of Adult Reporting Always Receiving Needed Social Emotional Support by Positive Childhood Experiences (PCEs) Single Items and Cumulative Scores aSource: authors’ analysis of the 2015 Wisconsin Behavioral Risk Factor Survey. bAdjusted odds ratios (AORs) shown are adjusted for age, sex, race/ethnicity, income, and adverse childhood experiences. cNever, rarely, or sometimes is the reference category. dNever, a little, or some of the time is the reference category.

Figure 2. Prevalence of Adult Reporting Always Receiving Needed Social Emotional Support by Positive Childhood Experiences (PCEs) Single Items and Cumulative Scores aSource: authors’ analysis of the 2015 Wisconsin Behavioral Risk Factor Survey. bAdjusted odds ratios (AORs) shown are adjusted for age, sex, race/ethnicity, income, and adverse childhood experiences. cNever, rarely, or sometimes is the reference category. dNever, a little, or some of the time is the reference category. The lowest adult D/PMH prevalences were observed for respondents reporting both 6 to 7 PCEs and either no ACEs (10.5%) or “always” on the ARSES variable (8.5%). Highest D/PMH prevalences were for those reporting 0 to 2 PCEs and either 4 to 8 ACEs (59.7%) or “sometimes/ rarely/never” on the ARSES variable (61.7%). Yet, even among those reporting always getting needed social and emotional support, a subset reported 0 to 2 PCEs, and this group had 4 times greater prevalence of D/PMH compared with those reporting 6 to 7 PCEs (33.8% vs 8.5%). Likewise, 21.2% of those with 4 to 8 ACEs and 26.6% of those reporting “sometime/rarely/never” to the ARSES item nonetheless also reported 6 to 7 PCEs. (Table 1, Table 3, and eTable 3 in the Supplement).

s, and this group had 4 times greater prevalence of D/PMH compared with those reporting 6 to 7 PCEs (33.8% vs 8.5%). Likewise, 21.2% of those with 4 to 8 ACEs and 26.6% of those reporting “sometime/rarely/never” to the ARSES item nonetheless also reported 6 to 7 PCEs. (Table 1, Table 3, and eTable 3 in the Supplement). Table 3. Prevalence of D/PMH and Reports of “Always” on the ARSES Item by PCEs Scores for Each of 4 Adverse Childhood Experiences ACEs Exposure Levels (0, 1, 2-3, or 4-8) Categories by ACEs and PCEs Meets D/PMH Criteriaa Reports of “Always” to Getting Needed Social and Emotional Support (ARSES) Unweighted No. Weighted % Adjusted Odds Ratiob (95% CI) Unweighted No. Weighted % Adjusted Odds Ratiob (95% CI) No ACEs reported 0-2 PCEs 17 12.1 1 [Reference] 35 34.6 1 [Reference] 3-5 PCEs 86 15.8 1.15 (0.51-2.62) 266 47.3 1.58 (0.84-2.95) 6-7 PCEs 148 10.5 0.88 (0.42-1.87) 1072 70.5 4.18 (2.31-7.55) 1 ACE reported 0-2 PCEs 35 45.7 1 [Reference] 38 30.9 1 [Reference] 3-5 PCEs 85 24.2 0.38 (0.17-0.83) 161 39.5 1.33 (0.68-2.62) 6-7 PCEs 94 13.4 0.21 (0.10-0.46) 390 67.6 4.93 (2.54-9.58) 2-3 ACEs reported 0-2 PCEs 87 53.3 1 [Reference] 47 30.3 1 [Reference] 3-5 PCEs 131 31.4 0.47 (0.26-0.84) 167 43.9 1.65 (0.90-3.02) 6-7 PCEs 76 16.0 0.18 (0.10-0.34) 223 59.2 2.80 (1.53-5.13) 4-8 ACEs reported 0-2 PCEs 155 59.7 1 [Reference] 75 35.1 1 [Reference] 3-5 PCEs 100 36.9 0.49 (0.28-0.84) 93 41.7 1.19 (0.69-2.03) 6-7 PCEs 29 20.7 0.23 (0.11-0.46) 56 65.6 3.37 (1.66-6.84) Abbreviations: ACEs, adverse childhood experiences; ARSES, adult-reported social and emotional support; D/PMH, depression and/or poor mental health; PCEs, positive childhood experiences.

1 [Reference] 3-5 PCEs 100 36.9 0.49 (0.28-0.84) 93 41.7 1.19 (0.69-2.03) 6-7 PCEs 29 20.7 0.23 (0.11-0.46) 56 65.6 3.37 (1.66-6.84) Abbreviations: ACEs, adverse childhood experiences; ARSES, adult-reported social and emotional support; D/PMH, depression and/or poor mental health; PCEs, positive childhood experiences. a Prevalence of D/PMH varied across levels of ACEs within each PCEs cumulative score category (0-2, 3-5, and 6-7) at P < .01. b Adjusted odds ratios adjusted for age, sex, race/ethnicity, and income. Association Pathway 1: PCEs and D/PMH After controlling for ACEs, the adjusted odds of D/PMH were 72% lower (odds ratio [OR], 0.28; 95% CI, 0.21-0.39) for adults with the highest vs lowest PCEs scores (12.6% vs 48.2%). Odds were 50% lower (OR, 0.50; 95% CI, 0.36-0.69) for those reporting intermediate PCEs scores of 3 to 5 (25.1% vs 48.2%) (Table 2). Associations were similar in magnitude for adults reporting 1, 2 to 3, or 4 to 8 ACEs (Table 3).

28; 95% CI, 0.21-0.39) for adults with the highest vs lowest PCEs scores (12.6% vs 48.2%). Odds were 50% lower (OR, 0.50; 95% CI, 0.36-0.69) for those reporting intermediate PCEs scores of 3 to 5 (25.1% vs 48.2%) (Table 2). Associations were similar in magnitude for adults reporting 1, 2 to 3, or 4 to 8 ACEs (Table 3). Association Pathway 2: PCEs and ARSES The adjusted odds of “always” reports on the ARSES item were 3.53 times (95% CI, 2.60-4.80) greater for adults with the highest vs lowest PCEs scores. Adjusted odds of reports of “always” on the ARSES variable were not significant for adults with intermediate PCEs of 3 to 5 (adjusted OR, 1.31; 95% CI, 0.97-1.78) (Table 2). Findings were similar across all ACEs exposure level subgroups (Table 3). Because PCEs and ARSES were strongly associated as hypothesized, we further examined whether each variable demonstrated an independent association with D/PMH and whether associations of PCEs with D/PMH remained stable when ARSES was included in regression models. Results showed that PCEs associations with D/PMH remained significant and changed only modestly when ARSES was included. Associations between ARSES and D/PMH also remained stable when PCEs were or were not included. See eTable 4 in the Supplement for details.

emained stable when ARSES was included in regression models. Results showed that PCEs associations with D/PMH remained significant and changed only modestly when ARSES was included. Associations between ARSES and D/PMH also remained stable when PCEs were or were not included. See eTable 4 in the Supplement for details. Discussion This study examined the prevalence of adult reports of both PCEs and ACEs in a statewide sample and found that PCEs both co-occur with and operate independently from ACEs in their associations with the adult health outcomes evaluated here. Findings also confirm the hypotheses that PCEs may exert their association with D/PMH through their association with ARSES. However, PCEs maintained an association with D/PMH independent from ARSES. Findings are both consistent with prior research showing that relational experiences in childhood are associated with adult social and relational skills and health and also point to enduring effects of PCEs on D/PMH separate from their influence on adult ARSES.

intained an association with D/PMH independent from ARSES. Findings are both consistent with prior research showing that relational experiences in childhood are associated with adult social and relational skills and health and also point to enduring effects of PCEs on D/PMH separate from their influence on adult ARSES. While PCEs associations with D/PMH were substantial and similar for adults reporting ACEs, associations were not statistically significant for those reporting no ACEs. Insignificant findings may be owing to low sample sizes for respondents with no ACEs and fewer PCEs. Results still raise questions for further exploration. We hypothesize that PCEs may have a greater influence in promoting positive health, such as getting needed social and emotional support or flourishing as an adult. In turn, these positive health attributes may reduce the burden of illness even if the illness is not eliminated. This is consistent with prior research demonstrating a dual continuum of health whereby flourishing is found to be present for many adults despite concurrent mental health conditions.

ourishing as an adult. In turn, these positive health attributes may reduce the burden of illness even if the illness is not eliminated. This is consistent with prior research demonstrating a dual continuum of health whereby flourishing is found to be present for many adults despite concurrent mental health conditions. Limitations First, this study is cross-sectional and cannot confirm causal effects. Second, the 2015 Wisconsin adult population is less diverse than the United States as a whole. Third, PCEs focused on the domain of positive emotional experiences in interpersonal relationships. Other types of positive experiences, (eg, safe and supportive environments, nature or spiritual experiences, participation in activities, or accomplishment) require further study, highlighting the need to develop and test additional measures of PCEs. Fourth, we were not able to directly examine bias in reporting of PCEs among adults with depression, although studies show an absence of such biases for reports of ACEs. Finally, the WI BRFS did not assess overall well-being or flourishing. As such, we were not able to assess whether PCEs affect positive adult health outcomes as hypothesized. Sample size limitations may have resulted in false-negative findings in some cases.

h studies show an absence of such biases for reports of ACEs. Finally, the WI BRFS did not assess overall well-being or flourishing. As such, we were not able to assess whether PCEs affect positive adult health outcomes as hypothesized. Sample size limitations may have resulted in false-negative findings in some cases. Conclusions Overall, study results demonstrate that PCEs show a dose-response association with adult mental and relational health, analogous to the cumulative effects of multiple ACEs. Findings suggest that PCEs may have lifelong consequences for mental and relational health despite co-occurring adversities such as ACEs. In this way, they support application of the World Health Organization’s definition of health emphasizing that health is more than the absence of disease or adversity. The World Health Organization’s positive construct of health is aligned with the proactive promotion of positive experiences in childhood because they are foundational to optimal childhood development and adult flourishing. Including PCEs as well as positive health outcomes measures in routinely collected public health surveillance systems, such as the National Survey of Children’s Health and state Behavioral Risk Factor Surveillance Surveys, may advance knowledge and allow the nation to track progress in promoting flourishing despite adversity or illness among children and adults in the United States.

utinely collected public health surveillance systems, such as the National Survey of Children’s Health and state Behavioral Risk Factor Surveillance Surveys, may advance knowledge and allow the nation to track progress in promoting flourishing despite adversity or illness among children and adults in the United States. Even as society continues to address remediable causes of childhood adversities such as ACEs, attention should be given to the creation of those positive experiences that both reflect and generate resilience within children, families, and communities. Success will depend on full engagement of families and communities and changes in the health care, education, and social services systems serving children and families. A joint inventory of ACEs and PCEs, such as the positive experiences assessed here, may improve efforts to assess needs, target interventions, and engage individuals in addressing the adversities they face by leveraging existing assets and strengths. Initiatives to conduct broad ACEs screening, such as those ensuing in California’s Medicaid program, may benefit from integrated assessments including PCEs.

improve efforts to assess needs, target interventions, and engage individuals in addressing the adversities they face by leveraging existing assets and strengths. Initiatives to conduct broad ACEs screening, such as those ensuing in California’s Medicaid program, may benefit from integrated assessments including PCEs. Recommendations and practice guidelines included in the National Bright Futures Guidelines for Health Supervision of Infants, Children, and Adolescents and the CDC’s Essentials for Childhood initiative encourage policies and initiatives to help child-serving professionals and programs to adopt effective approaches to promote the type of PCEs evaluated in this study. The Health Outcomes of Positive Experiences framework and the Prioritizing Possibilities national agenda for promoting child health and addressing ACEs each seek to advance existing and emerging evidence-based approaches that promote a positive construct of health in clinical, public health, and human services settings. This study adds to the growing evidence that childhood experiences have profound and lifelong effects. Results hold promise for national, state, and community efforts to achieve positive child and adult health and well-being by promoting the largely untapped potential to promote positive experiences and flourishing despite adversity.

to the growing evidence that childhood experiences have profound and lifelong effects. Results hold promise for national, state, and community efforts to achieve positive child and adult health and well-being by promoting the largely untapped potential to promote positive experiences and flourishing despite adversity. Supplement. eTable 1. Prevalence of Depression/Poor Mental Health (D/PMH) and Demographic Characteristics Among Adults With or Without Missing Values for Positive Childhood Experiences (PCEs), Adult Reported Social and Emotional Support (ARSES) and Adverse Childhood Experiences (ACEs) WI BRFS State Added Items eTable 2. Summary of Findings From Psychometric Analysesa Conducted on Positive Childhood Experiences (PCEs) Seven Item Measure eTable 3. Prevalence of Adult Depression and/or Poor Mental Health (D/PMH) by Adult Reported Social and Emotional Support (ARSES) and Positive Childhood Experiences (PCEs) eTable 4. Adjusted Odds Ratios of Adult Depression and/or Poor Mental Health (D/PMH) by Adult Reported Social and Emotional Support (ARSES) and Positive Childhood Experiences (PCEs) Under Alternative Model Configurations Click here for additional data file.

Introduction Typically, in any given school year children may be nearly a year apart in age, based on their birthdate relative to the cutoff date for school entry. Younger relative age within the school year has been associated consistently with poorer academic and sporting performance. Younger relative age has also been linked to attention-deficit/hyperactivity disorder (ADHD) diagnoses and to increased risk of intellectual disability (termed nonspecific learning disability in some countries). Few studies to date have examined the association between relative age and mental health. One study of 379 524 adolescents from 32 countries found that relatively young children reported reduced life satisfaction. Younger relative age has also been associated with lower self-esteem, reduced confidence in abilities, peer problems, and increased internalizing symptoms. One potential consequence of these differences may be an increased risk of depression. Although studies from Canada and Japan have identified an association between young relative age and increased risk of suicide, these investigations did not directly investigate the association between relative age and diagnosis of depression. In this study we aimed to investigate the association between relative age and a broader set of outcomes, including intellectual disability and ADHD, but also to examine for the first time, to our knowledge, the association between relative age and childhood depression.

Few studies to date have examined the association between relative age and mental health. One study of 379 524 adolescents from 32 countries found that relatively young children reported reduced life satisfaction. Younger relative age has also been associated with lower self-esteem, reduced confidence in abilities, peer problems, and increased internalizing symptoms. One potential consequence of these differences may be an increased risk of depression. Although studies from Canada and Japan have identified an association between young relative age and increased risk of suicide, these investigations did not directly investigate the association between relative age and diagnosis of depression. In this study we aimed to investigate the association between relative age and a broader set of outcomes, including intellectual disability and ADHD, but also to examine for the first time, to our knowledge, the association between relative age and childhood depression. Methods Clinical Practice Research Datalink The Clinical Practice Research Datalink (CPRD) is a large UK electronic primary care records database broadly representative of the UK population. A subset of the CPRD database is linked to the Hospital Episodes Statistics database, which contains data on admissions and attendances at English National Health Service hospitals and treatment centers. We used the full CPRD cohort for our main analyses and the linked subset for our secondary analyses exploring the association with ethnicity because more complete ethnicity information can be obtained using Hospital Episodes Statistics data. This study was approved by the London School of Hygiene and Tropical Medicine Research Ethics Committee and by the CPRD independent scientific advisory committee, which has approval under Section 251 of the National Health Service Act 2006 to process deidentified patient records without individual patient consent and provide anonymized data for public health research.

giene and Tropical Medicine Research Ethics Committee and by the CPRD independent scientific advisory committee, which has approval under Section 251 of the National Health Service Act 2006 to process deidentified patient records without individual patient consent and provide anonymized data for public health research. Study Population The study population included all children who were registered before January 3, 2017, at a general practice contributing high-quality data to the CPRD, and younger than 16 years at the last data collection at that general practice. Exact date of birth is redacted in the CPRD as part of the deidentification of the data for research use. However, year of birth is recorded and, for individuals younger than 16 years at last data collection, month of birth is also available. Date of birth was imputed as the 15th of the month of birth. Children were included from their imputed fourth birthday or from 12 months after registering at a practice contributing research quality data to CPRD, if later. Follow-up was censored at the earliest of date of death, date the child left the practice, date of last practice data collection, or imputed 16th birthday. Children receiving an outcome diagnosis before study entry or with missing sex were excluded. Exposure The exposure of interest was relative age within the school year. This age was calculated relative to the appropriate cutoff date for acceptance into a school year: August 31 in England and Wales, February 28 in Scotland, and July 1 in Northern Ireland (eFigure in the Supplement).

Study Population The study population included all children who were registered before January 3, 2017, at a general practice contributing high-quality data to the CPRD, and younger than 16 years at the last data collection at that general practice. Exact date of birth is redacted in the CPRD as part of the deidentification of the data for research use. However, year of birth is recorded and, for individuals younger than 16 years at last data collection, month of birth is also available. Date of birth was imputed as the 15th of the month of birth. Children were included from their imputed fourth birthday or from 12 months after registering at a practice contributing research quality data to CPRD, if later. Follow-up was censored at the earliest of date of death, date the child left the practice, date of last practice data collection, or imputed 16th birthday. Children receiving an outcome diagnosis before study entry or with missing sex were excluded. Exposure The exposure of interest was relative age within the school year. This age was calculated relative to the appropriate cutoff date for acceptance into a school year: August 31 in England and Wales, February 28 in Scotland, and July 1 in Northern Ireland (eFigure in the Supplement). To preserve analytical power and allow for nonlinear patterns to be identified, for the main analysis, relative age was analyzed using 4 categories, each spanning 3 months. The reference category was the oldest birth quarter (eg, birthdays in September to November in England).

Exposure The exposure of interest was relative age within the school year. This age was calculated relative to the appropriate cutoff date for acceptance into a school year: August 31 in England and Wales, February 28 in Scotland, and July 1 in Northern Ireland (eFigure in the Supplement). To preserve analytical power and allow for nonlinear patterns to be identified, for the main analysis, relative age was analyzed using 4 categories, each spanning 3 months. The reference category was the oldest birth quarter (eg, birthdays in September to November in England). Outcomes The primary outcomes were intellectual disability, ADHD, and depression. We included 3 negative control outcomes: appendectomy, Osgood-Schlatter disease, and incident glioma, which, a priori, we did not expect to be associated with relative age. All outcomes were primarily defined by first recorded relevant Read code (eAppendix in the Supplement). Attention-deficit/hyperactivity disorder was secondarily defined by first prescription of ADHD medication without consideration of Read code. In this secondary analysis, children with a prior Read code for narcolepsy were excluded because this diagnosis is the only other indication for these drugs. Statistical Analysis Multivariable Cox proportional hazards regression models with an underlying age timescale were used because absolute age was assumed to be the strongest a priori risk factor. Adjusted hazard ratios (aHRs) and 95% CIs were estimated, and P values were derived using likelihood ratio tests. We stratified our results by country in secondary analyses.

al hazards regression models with an underlying age timescale were used because absolute age was assumed to be the strongest a priori risk factor. Adjusted hazard ratios (aHRs) and 95% CIs were estimated, and P values were derived using likelihood ratio tests. We stratified our results by country in secondary analyses. Socioeconomic status, calendar year, and sex were potential confounders and were included as covariates in the model. Socioeconomic status was estimated by the 2015 Index of Multiple Deprivation score, measured at the general practice level, and categorized into deciles. The Index of Multiple Deprivation is an index of relative deprivation based on domains including income, employment, education, and health. We estimated the cumulative incidence and attributable risk percentage of the main outcomes due to birth quarter. We calculated this variable based on the survival function generated after estimation from the Cox proportional hazards regression model adjusted for sex, calendar year, and socioeconomic status. We estimated the cumulative incidence from age 4 to 16 years per 100 000 children. Effect modification by sex, absolute age (categorized around the midpoint of follow-up into <10 or ≥10 years to maximize statistical power), and ethnicity were explored in secondary analyses. Ethnicity was derived, among children eligible for linkage to the Hospital Episodes Statistics database with recorded ethnicity, using a validated method.

, absolute age (categorized around the midpoint of follow-up into <10 or ≥10 years to maximize statistical power), and ethnicity were explored in secondary analyses. Ethnicity was derived, among children eligible for linkage to the Hospital Episodes Statistics database with recorded ethnicity, using a validated method. In a further secondary analysis, we used Cox proportional hazards regression models to compare children born the month before the cutoff relative to those born the month after. Children born in the month on either side of the cutoff date are a year apart in school years and at opposite ends of the relative age spectrum, despite being less than 2 months apart in actual age. We anticipated that unaccounted for confounding by season would be minimized in this analysis. As a sensitivity analysis we examined censoring follow-up at earliest date of the other primary outcomes on the outcome of Cox regression for each of the 3 primary outcomes. With this analysis, we aimed to avoid any diagnostic overshadowing, where a major clinical diagnosis (ie, intellectual disability) frames and may inhibit subsequent diagnoses, such as depression. All analyses were performed using Stata MP, version 14/15 (StataCorp LLC). Data were analyzed between July 2017 and January 2019.

this analysis, we aimed to avoid any diagnostic overshadowing, where a major clinical diagnosis (ie, intellectual disability) frames and may inhibit subsequent diagnoses, such as depression. All analyses were performed using Stata MP, version 14/15 (StataCorp LLC). Data were analyzed between July 2017 and January 2019. Results There were 1 042 106 eligible children between ages 4 and 15 years within the CPRD after exclusion of 20 with missing data on sex (Figure, Table 1). There was an equal proportion of children across the 4 categories of relative age, and 532 876 were male (51.1%). The median age at study entry was 4.0 years (interquartile range [IQR], 4.0-5.0). Follow-up (ignoring outcome occurrence) was a median of 3.4 years (IQR, 1.4-6.4) and differed little between birth quartiles: fourth (youngest), 3.4 (IQR, 1.4-6.4); third, 3.4 (IQR, 1.4-6.5); second, 3.4 (IQR, 1.4-6.5); and first (oldest), 3.3 (IQR, 1.3-6.3). Median age at outcome diagnosis was 7.7 years (IQR, 6.2-9.6) for intellectual disability, 8.0 years (IQR, 6.7-9.7) for ADHD, and 13.3 years (IQR, 11.7-14.4) for depression. Figure. Study Flow Diagram ADHD indicates attention-deficit/hyperactivity disorder; CPRD, Clinical Practice Research Datalink. aRegistered at a general practice contributing research quality data to the CPRD and younger than 16 years at last practice data collection.

Results There were 1 042 106 eligible children between ages 4 and 15 years within the CPRD after exclusion of 20 with missing data on sex (Figure, Table 1). There was an equal proportion of children across the 4 categories of relative age, and 532 876 were male (51.1%). The median age at study entry was 4.0 years (interquartile range [IQR], 4.0-5.0). Follow-up (ignoring outcome occurrence) was a median of 3.4 years (IQR, 1.4-6.4) and differed little between birth quartiles: fourth (youngest), 3.4 (IQR, 1.4-6.4); third, 3.4 (IQR, 1.4-6.5); second, 3.4 (IQR, 1.4-6.5); and first (oldest), 3.3 (IQR, 1.3-6.3). Median age at outcome diagnosis was 7.7 years (IQR, 6.2-9.6) for intellectual disability, 8.0 years (IQR, 6.7-9.7) for ADHD, and 13.3 years (IQR, 11.7-14.4) for depression. Figure. Study Flow Diagram ADHD indicates attention-deficit/hyperactivity disorder; CPRD, Clinical Practice Research Datalink. aRegistered at a general practice contributing research quality data to the CPRD and younger than 16 years at last practice data collection. Table 1. Baseline Characteristics Characteristic No. (%) Birth Quarter Overall (n = 1 042 106) 4 (Youngest) (n = 265 173) 3 (n = 258 993) 2 (n = 253 017) 1 (Oldest) (n = 264 923) Sex Male 135 265 (51.0) 132 419 (51.1) 129 818 (51.3) 135 374 (51.1) 532 876 (51.1) Female 129 908 (49.0) 126 574 (48.9) 123 199 (48.7) 129 549 (48.9) 509 230 (48.9) Socioeconomic statusa 1 (Least deprived) 23 140 (8.7) 23 333 (9.0) 22 378 (8.8) 23 615 (8.9) 92 466 (8.9) 2 21 774 (8.2) 21 245 (8.2) 20 403 (8.1) 21 615 (8.2) 85 037 (8.2) 3 25 922 (9.8) 25 617 (9.9) 24 268 (9.6) 26 080 (9.8) 101 887 (9.8) 4 20 850 (7.9) 20 231 (7.8) 19 698 (7.8) 20 748 (7.8) 81 527 (7.8) 5 25 899 (9.8) 25 839 (10.0) 25 006 (9.9) 26 177 (9.9) 102 921 (9.9) 6 25 654 (9.7)) 25 313 (9.8) 24 480 (9.7) 25 685 (9.7) 101 132 (9.7) 7 30 334 (11.4) 28 859 (11.1) 28 883 (11.4) 30 088 (11.4) 118 164 (11.3) 8 26 567 (10.0) 25 784 (10.0) 25 161 (9.9) 26 195 (9.9) 103 707 (10.0) 9 36 567 (13.8) 35 321 (13.6) 35 155 (13.9) 36 079 (13.6) 143 122 (13.7) 10 (Most deprived) 28 466 (10.7) 27 451 (10.6) 27 585 (10.9) 28 641 (10.8) 112 143 (10.8) Ethnicityb White 111 352 (70.0) 107 744 (69.8) 102 925 (69.0) 110 240 (69.8) 432 261 (69.7) South Asian 9557 (6.0) 9323 (6.0) 9357 (6.4) 9684 (6.1) 37 921 (6.1) Black 6878 (4.3) 6557 (4.3) 6425 (4.3) 6761 (4.3) 26 621 (4.3) Other 3487 (2.2) 3413 (2.2) 3376 (2.3) 3501 (2.2) 13 777 (2.2) Mixed 4619 (2.9) 4330 (2.8) 4164 (2.8) 4419 (2.8) 17 532 (2.8) Missing 23 166 (14.6) 22 990 (14.9) 23 005 (15.4) 23 293 (14.8) 92 454 (14.9) Calendar period over which children contribute follow-up 1988-1995 873 (0.3) 780 (0.3) 682 (0.3) 915 (0.3) 3250 (0.3) 1995-2002 13 478 (5.1) 12 423 (4.8) 12 338 (4.9) 13 861 (5.2) 52 100 (5.0) 2002-2009 100 676 (38.0) 95 949 (37.0) 94 747 (37.4) 102 060 (38.5) 393 432 (37.8) 2009-2016 221 655 (83.6) 215 881 (83.4) 211 785 (83.7) 222 258 (83.9) 871 579 (83.6) 2016-2017 110 256 (41.6) 108 509 (41.9) 106 239 (42.0) 111 529 (42.1) 436 533 (41.9) a Measured by practice level decile of Index of Multiple Deprivation 2015 score where 1 is the least deprived decile.

4) 102 060 (38.5) 393 432 (37.8) 2009-2016 221 655 (83.6) 215 881 (83.4) 211 785 (83.7) 222 258 (83.9) 871 579 (83.6) 2016-2017 110 256 (41.6) 108 509 (41.9) 106 239 (42.0) 111 529 (42.1) 436 533 (41.9) a Measured by practice level decile of Index of Multiple Deprivation 2015 score where 1 is the least deprived decile. b Analyses including ethnicity were performed within the hospital episodes statistics linked subcohort of the Clinical Practice Research Datalink containing 620 566 patients. Association With Intellectual Disability A total of 2034 children were excluded because they received a diagnosis of intellectual disability before study entry, leaving 1 040 072 children with a median follow-up of 3.3 (IQR, 1.4-6.4) years. Cox proportional hazards regression modeling suggested evidence of an association between relative age in the school year and incidence of intellectual disability. The HR for intellectual disability increased with younger age within the school year, with adjusted HRs (aHRs) of 1.06 (95% CI, 0.96-1.17) for those born in the second quarter, 1.20 (95% CI, 1.09-1.32) for those born in the third quarter, and 1.30 (95% CI, 1.18-1.42) for those born in the fourth quarter of the school year, compared with the first quarter (oldest), (P < .01 for trend) (Table 2).

ol year, with adjusted HRs (aHRs) of 1.06 (95% CI, 0.96-1.17) for those born in the second quarter, 1.20 (95% CI, 1.09-1.32) for those born in the third quarter, and 1.30 (95% CI, 1.18-1.42) for those born in the fourth quarter of the school year, compared with the first quarter (oldest), (P < .01 for trend) (Table 2). Table 2. Incidence of Each Outcome by Birth Quarter in the School Year Adjusted for Sex, Calendar Year, and Socioeconomic Status Birth Quarter in School Year No. of Children Follow-up (Person-Years) No. of Outcomes Incidence Rate (per 1000 Person-Years) Adjusted Hazard Ratio (95% CI) P Valuea P Value for Trend Intellectual Disability 4 (Youngest) 264 583 1 090 885 1044 0.96 1.30 (1.18-1.42) <.01 <.01 3 258 513 1 071 799 945 0.88 1.20 (1.09-1.32) 2 252 522 1 044 470 821 0.79 1.06 (0.96-1.17) 1 (Oldest) 264 454 1 079 919 797 0.74 1 [Reference] ADHD 4 (Youngest) 264 432 1 087 557 2267 2.08 1.36 (1.28-1.45) <.01 <.01 3 258 329 1 068 457 2140 2.00 1.31 (1.23-1.40) 2 252 387 1 041 790 1843 1.77 1.15 (1.08-1.23) 1 (Oldest) 264 282 1 077 483 1655 1.54 1 [Reference] Depression 4 (Youngest) 265 125 1 095 746 240 0.22 1.31 (1.08-1.59) .03 <.01 3 258 950 1 076 236 206 0.19 1.13 (0.92-1.38) 2 252 981 1 048 488 187 0.18 1.05 (0.85-1.29) 1 (Oldest) 264 873 1 083 359 179 0.17 1 [Reference] Osgood-Schlatter Disease 4 (Youngest) 265 100 1 094 478 939 0.86 0.88 (0.81-0.96) .04 <.01 3 258 929 1 074 708 975 0.91 0.93 (0.85-1.01) 2 252 952 1 046 979 969 0.93 0.95 (0.87-1.03) 1 (Oldest) 264 857 1 081 814 1042 0.96 1 [Reference] Appendectomy 4 (Youngest) 264 908 1 093 131 815 0.75 1.05 (0.95-1.16) .26 .72 3 258 745 1 073 776 745 0.69 0.97 (0.88-1.08) 2 252 724 1 045 746 791 0.76 1.06 (0.96-1.18) 1 (Oldest) 264 642 1 080 720 766 0.71 1 [Reference] Glioma 4 (Youngest) 265 139 1 096 045 24 0.02 1.39 (0.75-2.59) .28 .38 3 258 967 1 076 413 28 0.03 1.66 (0.91-3.03) 2 252 995 1 048 752 28 0.03 1.70 (0.93-3.10) 1 (Oldest) 264 901 1 083 690 17 0.02 1 [Reference] Abbreviation: ADHD, attention-deficit/hyperactivity disorder.

dest) 264 642 1 080 720 766 0.71 1 [Reference] Glioma 4 (Youngest) 265 139 1 096 045 24 0.02 1.39 (0.75-2.59) .28 .38 3 258 967 1 076 413 28 0.03 1.66 (0.91-3.03) 2 252 995 1 048 752 28 0.03 1.70 (0.93-3.10) 1 (Oldest) 264 901 1 083 690 17 0.02 1 [Reference] Abbreviation: ADHD, attention-deficit/hyperactivity disorder. a Likelihood ratio test. Association With ADHD A total of 2676 children were excluded because they received a diagnosis of ADHD before study entry, leaving 1 039 430 children with a median follow-up of 3.3 (IQR, 1.4-6.4) years. Cox proportional hazards regression modeling suggested an association between relative age in the school year and incidence of ADHD. The HR for ADHD increased with younger age within the school year, with aHRs of 1.15 (95% CI, 1.08-1.23) for those born in the second quarter, 1.31 (95% CI, 1.23-1.40) for those born in the third quarter, and 1.36 (95% CI, 1.28-1.45) for those born in fourth quarter of the school year, compared with the first quarter (oldest), (P < .01) (Table 2). Analyses were repeated with ADHD defined as first prescription of a central nervous system stimulant drug. The results were similar, with aHRs of 1.15 (95% CI, 1.07-1.23) for those born in the second quarter, 1.26 (95% CI, 1.18-1.35) for those born in the third quarter, and 1.35 (95% CI, 1.27-1.45) for those born in the fourth quarter compared with the first quarter (oldest) (P < .01 for trend) (eTable 1 in the Supplement).

The results were similar, with aHRs of 1.15 (95% CI, 1.07-1.23) for those born in the second quarter, 1.26 (95% CI, 1.18-1.35) for those born in the third quarter, and 1.35 (95% CI, 1.27-1.45) for those born in the fourth quarter compared with the first quarter (oldest) (P < .01 for trend) (eTable 1 in the Supplement). Association With Depression A total of 177 children were excluded because they received a diagnosis of depression before study entry, leaving 1 041 929 children with a median follow-up of 3.4 years (IQR, 1.4-6.4). Cox proportional hazards regression modeling suggested a possible association between relative age in the school year and incidence of depression. The aHR for depression increased with younger age within the school year, with aHRs of 1.05 (95% CI, 0.85-1.29) for those born in the second quarter, 1.13 (95% CI, 0.92-1.38) for those born in the third quarter, and 1.31 (95% CI, 1.08-1.59) for those born in the fourth quarter of the school year compared with the first quarter (oldest) (P < .01 for trend) (Table 2).

e within the school year, with aHRs of 1.05 (95% CI, 0.85-1.29) for those born in the second quarter, 1.13 (95% CI, 0.92-1.38) for those born in the third quarter, and 1.31 (95% CI, 1.08-1.59) for those born in the fourth quarter of the school year compared with the first quarter (oldest) (P < .01 for trend) (Table 2). Interactions In secondary models, there was no evidence for an interaction between relative age and sex for intellectual disability (P for interaction = .35), ADHD (P for interaction = .68), and depression (P for interaction = .77). There was evidence for an interaction between relative age and ethnicity for intellectual disability (P for interaction < .01), but not for ADHD (P for interaction = .38). The association between relative age and intellectual disability was seen only in children of white or mixed ethnicity; HRs for intellectual disability comparing those born in the fourth quarter (youngest) to those born in the first quarter (oldest) were 1.44 (95% CI 1.26-1.64) for white children and 2.59 (95% CI 1.15-5.82) for those of mixed ethnicity, although the number of events in nonwhite ethnic groups was small, with only 120 events among individuals of South Asian ethnicity and 51 events among individuals of mixed ethnicity (eTable 2 in the Supplement). There were too few outcomes in nonwhite ethnic groups to assess the interaction between relative age and ethnicity for depression.

events in nonwhite ethnic groups was small, with only 120 events among individuals of South Asian ethnicity and 51 events among individuals of mixed ethnicity (eTable 2 in the Supplement). There were too few outcomes in nonwhite ethnic groups to assess the interaction between relative age and ethnicity for depression. There was suggestive evidence that the association between relative age and ADHD was greater in children younger than 10 years compared with those aged 10 years or older (P for interaction = .06) (eTable 3 in the Supplement). There was no evidence for an interaction between relative age and absolute age for intellectual disability (P for interaction = .40) or depression (P for interaction = .51). Sensitivity Analyses A restricted model comparing children born in the month on either side of the entry cutoff (those at the extremes of the year age band) supported the association between relative age and incidence of intellectual disability (aHR, 1.46; 95% CI, 1.23-1.72; P < .01) and ADHD (aHR, 1.54; 95% CI, 1.38-1.72; P < .01) (eTable 4 and eTable 5 in the Supplement). A restricted model for depression was consistent with the main analysis but with wide 95% CIs (aHR, 1.19; 95% CI, 0.84-1.68; P = .32).

relative age and incidence of intellectual disability (aHR, 1.46; 95% CI, 1.23-1.72; P < .01) and ADHD (aHR, 1.54; 95% CI, 1.38-1.72; P < .01) (eTable 4 and eTable 5 in the Supplement). A restricted model for depression was consistent with the main analysis but with wide 95% CIs (aHR, 1.19; 95% CI, 0.84-1.68; P = .32). Cox proportional hazards regression without adjustment produced similar results to adjusted Cox proportional hazards regression (eTable 6 in the Supplement). For all 3 main outcomes, aHRs changed little when censoring follow-up at the earliest diagnosis of any of the other 2 outcomes (eTable 7 in the Supplement). For instance, aHRs for intellectual disability were 1.06 (95% CI, 0.96-1.18) for those born in the second quarter, 1.19 (95% CI, 1.08-1.32) for those born in the third quarter, and 1.30 (95% CI, 1.18-1.43) for those born in the fourth quarter relative to those born in the first quarter (oldest).

e 7 in the Supplement). For instance, aHRs for intellectual disability were 1.06 (95% CI, 0.96-1.18) for those born in the second quarter, 1.19 (95% CI, 1.08-1.32) for those born in the third quarter, and 1.30 (95% CI, 1.18-1.43) for those born in the fourth quarter relative to those born in the first quarter (oldest). Incidence of Main Outcomes by Country Most children in the study were from England and Wales (n = 921 025), with fewer from Northern Ireland (n = 29 063) or Scotland (n = 92 018). Similar results as obtained in the overall cohort were seen among children from England and Wales. For example, aHRs for intellectual disability were 1.04 (95% CI, 0.93-1.15) for those born in the second quarter, 1.20 (95% CI, 1.08-1.33) for those born in the third quarter, and 1.32 (95% CI, 1.20-1.45) for those born in the fourth quarter, relative to those born in the first quarter (oldest) (eTable 8 in the Supplement). Results for Northern Ireland and Scotland were largely inconclusive owing to the low numbers of outcomes and resultant wide 95% CIs. For example, the HR for intellectual disability in Northern Ireland for the youngest birth quarter relative to oldest birth quarter was 1.19, with a wide 95% CI (0.64-2.23) (eTable 9 and eTable 10 in the Supplement).

and Scotland were largely inconclusive owing to the low numbers of outcomes and resultant wide 95% CIs. For example, the HR for intellectual disability in Northern Ireland for the youngest birth quarter relative to oldest birth quarter was 1.19, with a wide 95% CI (0.64-2.23) (eTable 9 and eTable 10 in the Supplement). Incidence of Outcomes Among children born in the fourth quarter (youngest), the estimated cumulative incidence per 100 000 was 1045.2 diagnoses of intellectual disability, 2360.6 diagnoses of ADHD, and 1086.5 diagnoses of depression by age 16 years (Table 3). In comparison, estimated cumulative incidence per 100 000 among children born in the first quarter (oldest) was 806.8 diagnoses of intellectual disability, 1741.3 diagnoses of ADHD, and 833.1 diagnoses of depression by age 16 years. The attributable risk percentage among children born in the youngest birth quarter was 23% for intellectual disability, 26% for ADHD, and 23% for depression. Table 3. Estimated Cumulative Incidence and Attributable Risk Percent of Intellectual Disability, ADHD, and Depression by Birth Quarter by Age 16 Years Birth Quarter in Year Estimated Cumulative Incidence by Age 16 y Per 100 000 Children Attributable Risk, % Intellectual Disability 4 (Youngest) 1045.2 22.8 3 966.3 16.5 2 857.9 6.0 1 (Oldest) 806.8 [Reference] ADHD 4 (Youngest) 2360.6 26.2 3 2270.3 23.3 2 1997.6 12.8 1 (Oldest) 1741.3 [Reference] Depression 4 (Youngest) 1086.5 23.3 3 937.5 11.1 2 873.3 4.6 1 (Oldest) 833.1 [Reference] Abbreviation: ADHD, attention-deficit/hyperactivity disorder.

y 4 (Youngest) 1045.2 22.8 3 966.3 16.5 2 857.9 6.0 1 (Oldest) 806.8 [Reference] ADHD 4 (Youngest) 2360.6 26.2 3 2270.3 23.3 2 1997.6 12.8 1 (Oldest) 1741.3 [Reference] Depression 4 (Youngest) 1086.5 23.3 3 937.5 11.1 2 873.3 4.6 1 (Oldest) 833.1 [Reference] Abbreviation: ADHD, attention-deficit/hyperactivity disorder. Association Incidence of Negative Control Outcomes The main analyses were replicated using 3 outcomes not previously thought to be associated with relative age: appendectomy, glioma, and Osgood-Schlatter disease. No evidence of an association was found between relative age and the first 2 negative control outcomes (Table 2). However, there was weak evidence of an association with Osgood-Schlatter disease. Relatively young children were less likely to be diagnosed with Osgood-Schlatter disease (aHR, 0.88; 95% CI, 0.81-0.96). Discussion This study provides, to our knowledge, the first evidence for what may be an association between younger relative age and increases in the diagnosis of depression. We also found, in concordance with previous studies, a possible association between younger relative age and increases in the diagnosis and prescription for treatment of ADHD and diagnosis of intellectual disability. Children born in the last quarter of the school year and therefore the youngest within their year were 1.3 times more likely to be diagnosed with intellectual disability, 1.4 times more likely to be diagnosed with ADHD, and 1.3 times more likely to be diagnosed with depression compared with children born in the first quarter of the school year.

quarter of the school year and therefore the youngest within their year were 1.3 times more likely to be diagnosed with intellectual disability, 1.4 times more likely to be diagnosed with ADHD, and 1.3 times more likely to be diagnosed with depression compared with children born in the first quarter of the school year. Two recent systematic reviews reported a consistent association between relative age and ADHD diagnoses and prescriptions in geographically diverse countries with differing school cutoffs and various school starting ages (from 4 to 7 years). There is limited evidence of an association with intellectual disability or depression: 2 US studies reported an association between relative age and intellectual disability, although the investigators of one of these studies found that this association was driven by ADHD diagnoses, whereas we found what may be an association independent of ADHD. Although, to our knowledge, there have been no previous studies on relative age and diagnosis of depression, investigations have reported lower life satisfaction, reduced self-esteem, and increased suicide rate in the relatively younger category.

agnoses, whereas we found what may be an association independent of ADHD. Although, to our knowledge, there have been no previous studies on relative age and diagnosis of depression, investigations have reported lower life satisfaction, reduced self-esteem, and increased suicide rate in the relatively younger category. Negative Controls To detect hidden confounding or other biases in our study, we repeated all analyses on 3 outcomes that were not a priori thought to be associated with relative age within the school year. For 2 of the control outcomes (appendectomy and glioma), no association was found, although statistical power was limited for glioma owing to few outcomes. For Osgood-Schlatter disease, there was a small association in the opposite direction from the main outcomes. A possible explanation for this unexpected finding could be that children who are relatively young in the school year are less likely to participate in physical education and sports at school, which are known risk factors for Osgood-Schlatter disease.

small association in the opposite direction from the main outcomes. A possible explanation for this unexpected finding could be that children who are relatively young in the school year are less likely to participate in physical education and sports at school, which are known risk factors for Osgood-Schlatter disease. Clinical and Policy Implications We have reported associations between birth month and diagnosis of intellectual disability, ADHD, and depression. Relatively young children have, in prior studies, been rated higher for hyperactivity/inattention symptoms, particularly by teachers. This elevated rate may be the result of relative immaturity in comparison with peers, overdiagnosis in relatively young children, or underdiagnosis in relatively old children with intellectual disability and/or ADHD. There have been concerns about overdiagnosis of ADHD given economic costs of treatment and uncertain long-term safety of ADHD medication. Academic performance and depression have been previously linked. This link may be because the association between relative youth in the school year and poorer academic performance is a factor in the association with depression. In addition, relative youth has been associated with poorer peer relationships, which might result in an increased incidence of depression.

reviously linked. This link may be because the association between relative youth in the school year and poorer academic performance is a factor in the association with depression. In addition, relative youth has been associated with poorer peer relationships, which might result in an increased incidence of depression. Other potential explanations for the association between relative age and depression include the possibility that length of time in school (greater at any given age for relatively young children) may be associated with depression. A seasonal association between month of birth and depression cannot be ruled out. Neither of these factors, however, explains the association between suicide and relative age reported in a Japanese study, which was observed among young adults who have finished school, when comparing 2 adjacent months, and despite a spring rather than autumn school cutoff. Historically, a seasonal association with depression existed due to poor nutrition in the winter, but this association disappeared as food availability no longer became seasonal.

bserved among young adults who have finished school, when comparing 2 adjacent months, and despite a spring rather than autumn school cutoff. Historically, a seasonal association with depression existed due to poor nutrition in the winter, but this association disappeared as food availability no longer became seasonal. There are a number of potential interventions to reduce the adverse effects of relative youth. However, existing evidence on these alternative approaches is limited. In some countries, parents of relatively young children can defer entry for a year. Deferment of children who are both relatively young and developmentally immature could reduce differences in abilities, but deferment of entry for all relatively young children would only change who is relatively young. Furthermore, deferment based on parent choice might have differential uptake according to socioeconomic status, which could lead to increased socioeconomic inequality. We believe an alternative to parental choice is entry based on ability testing. Other potential interventions include measures to increase awareness of relative age among parents, teachers, and clinicians; greater care in diagnosis, such as through the use of ADHD rating scales; and increased support for relatively young children.

There are a number of potential interventions to reduce the adverse effects of relative youth. However, existing evidence on these alternative approaches is limited. In some countries, parents of relatively young children can defer entry for a year. Deferment of children who are both relatively young and developmentally immature could reduce differences in abilities, but deferment of entry for all relatively young children would only change who is relatively young. Furthermore, deferment based on parent choice might have differential uptake according to socioeconomic status, which could lead to increased socioeconomic inequality. We believe an alternative to parental choice is entry based on ability testing. Other potential interventions include measures to increase awareness of relative age among parents, teachers, and clinicians; greater care in diagnosis, such as through the use of ADHD rating scales; and increased support for relatively young children. Limitations This study has limitations. Owing to the nature of the data source, month of birth rather than exact date was available. In Northern Ireland, school entry cutoff is on the first rather than thirty-first of the month. Children born in Northern Ireland on July 1 will, as a result, be misclassified. We expect the effect of this to be minor, given that children born in Northern Ireland in July represented 0.2% (n = 2344) of the full cohort.

In Northern Ireland, school entry cutoff is on the first rather than thirty-first of the month. Children born in Northern Ireland on July 1 will, as a result, be misclassified. We expect the effect of this to be minor, given that children born in Northern Ireland in July represented 0.2% (n = 2344) of the full cohort. We assume that all schools and local authorities adhere to the national cutoff for school year entry. In England and Wales (countries of residence for 88.4% of the study cohort), requests for delayed entry are uncommon, occurring in 1 estimate among less than 0.5% of children younger than 5 years. We believe misclassification would have led us to underestimate the true association between our exposure and outcomes. Lack of month of birth data for individuals aged 16 years or older limited our sample size relative to the CPRD population and meant that we were unable to examine the association with month of birth in individuals aged 16 years or older (ie, older adolescents and young adults). Further research into the association of relative age with depression in older adolescents appears to be warranted to see if effects persist into young adulthood. Matsubayashi and Ueda found a higher suicide rate in adolescents and young adults (age, 15-25 years) who were relatively young within their school year.

Lack of month of birth data for individuals aged 16 years or older limited our sample size relative to the CPRD population and meant that we were unable to examine the association with month of birth in individuals aged 16 years or older (ie, older adolescents and young adults). Further research into the association of relative age with depression in older adolescents appears to be warranted to see if effects persist into young adulthood. Matsubayashi and Ueda found a higher suicide rate in adolescents and young adults (age, 15-25 years) who were relatively young within their school year. Conclusions Relatively young children in their class during the school year may be at increased risk of diagnosis of intellectual disability, ADHD, and depression. Our study findings suggest that further research into interventions to reduce the negative associations of relative age with academic achievement and health is needed. Supplement. eFigure. Relative Age and Month of Birth in England, Wales, Scotland, and Northern Ireland eTable 1. Incidence of ADHD, Defined Based on ADHD Medication Prescription, by Birth Quarter in School Year Adjusted for Sex, Calendar Year, and Socioeconomic Status eTable 2. Incidence of Intellectual Disabilities by Birth Quarter in the School Year Stratified by Ethnicity and Adjusted for Sex, Calendar Year and Socioeconomic Status eTable 3. Incidence of ADHD by Birth Quarter Stratified by Absolute Age and Adjusted for Sex, Calendar Year and Socioeconomic Status

eTable 1. Incidence of ADHD, Defined Based on ADHD Medication Prescription, by Birth Quarter in School Year Adjusted for Sex, Calendar Year, and Socioeconomic Status eTable 2. Incidence of Intellectual Disabilities by Birth Quarter in the School Year Stratified by Ethnicity and Adjusted for Sex, Calendar Year and Socioeconomic Status eTable 3. Incidence of ADHD by Birth Quarter Stratified by Absolute Age and Adjusted for Sex, Calendar Year and Socioeconomic Status eTable 4. Cox Regression Adjusted for Sex, Calendar Year and Socioeconomic Status Comparing Children Born in the Month Before to the Month After the School Entry Cut-off eTable 5. Cox Regression Adjusted for Sex, Calendar Year and Socioeconomic Status Comparing Children by Month of Birth eTable 6. Unadjusted and Adjusted Hazard Ratios for Each Outcome eTable 7. Cox Regression Adjusted for Sex, Calendar Year and Socioeconomic Status Censored at the Occurrence of the Other Two Main Outcomes eTable 8. Incidence of Outcomes by Birth Quarter in England and Wales Adjusted for Sex, Calendar Year and Socioeconomic Status eTable 9. Incidence of Outcomes by Birth Quarter in Northern Ireland Adjusted for Sex, Calendar Year and Socioeconomic Status eTable 10. Incidence of Outcomes by Birth Quarter in Scotland Adjusted for Sex, Calendar Year and Socioeconomic Status eAppendix. Code Lists Click here for additional data file.

Introduction Exposure to adverse experiences, stress, and violence during childhood and adolescence is associated with elevated risk of physical and mental health problems in adulthood, many of which have inflammatory origins. Inflammation could constitute one of the underlying mechanisms responsible for the biological embedding of childhood stress, indicating an association between exposure to early-life adversity and adverse health outcomes in later life. C-reactive protein (CRP) and interleukin 6 (IL-6) are among the most commonly measured biomarkers of inflammation. Different types of adverse experiences have been associated with increased CRP and IL-6 levels, including maltreatment, bullying, and sexual abuse. However, findings are not consistent; several studies report nonsignificant associations, and not all associations were found after controlling for the confounding effects of smoking or obesity. The Dunedin Longitudinal Study recently found that exposure to childhood risk factors, including adverse childhood experiences (ACEs), was associated with higher levels of a novel biomarker of inflammation, soluble urokinase plasminogen activator receptor (suPAR), in adults independently of smoking and body mass index (BMI).

nedin Longitudinal Study recently found that exposure to childhood risk factors, including adverse childhood experiences (ACEs), was associated with higher levels of a novel biomarker of inflammation, soluble urokinase plasminogen activator receptor (suPAR), in adults independently of smoking and body mass index (BMI). suPAR is released to the bloodstream during proinflammatory conditions when the membrane-bound receptor uPAR is cleaved from the surface of immunologically active cells. Plasma levels of suPAR are thought to reflect a person’s overall level of immune activity. Elevated suPAR levels are observed in many diseases and pathologic conditions and are associated with development and progression of disease, adverse clinical outcomes, and mortality. High suPAR levels are also positively correlated with CRP and IL-6 in general and patient populations. Whereas CRP is an acute-phase reactant and a marker of acute inflammation and infections, suPAR appears to be less affected by acute conditions. Of interest, suPAR is associated with disease and mortality independent of CRP, the criterion standard marker of inflammation, suggesting that combined use of CRP and suPAR may provide a more accurate estimate of inflammatory burden by combining information about acute and chronic inflammation. Less is known about how suPAR compares with IL-6. Interleukin 6 is an important cytokine with proinflammatory or anti-inflammatory properties depending on the specific immunologic context. In addition to the immunologic functions of IL-6 in the acute-phase response, infections, inflammation, and cancer, IL-6 exerts multiple pleiotropic effects on other cell types, thereby regulating metabolism, hematopoiesis, and the neuroendocrine system. Although suPAR has been found to be associated with clinical outcomes independent of IL-6, whether the combined use of IL-6 and suPAR also provides additive information about inflammatory burden is unknown.

otropic effects on other cell types, thereby regulating metabolism, hematopoiesis, and the neuroendocrine system. Although suPAR has been found to be associated with clinical outcomes independent of IL-6, whether the combined use of IL-6 and suPAR also provides additive information about inflammatory burden is unknown. This report extends a previous study by investigating the association between suPAR and exposure to adverse experiences, stress, and violence during childhood and adolescence in the population-representative Environmental Risk (E-Risk) Longitudinal Twin Study followed up to 18 years of age. We assessed multiple types of adverse experiences during childhood and adolescence and cumulatively across the first 2 decades of life inside and outside the family, including emotional, physical, and sexual abuse; emotional and physical neglect; peer bullying; cyber bullying; and crime violence. In addition to evaluating these experiences in relation to suPAR, we compared the association of stress and violence exposure with CRP and IL-6 levels and tested whether adding suPAR to the measurement of CRP or IL-6 levels provides additional information about inflammation associated with stress and violence exposure.

In addition to evaluating these experiences in relation to suPAR, we compared the association of stress and violence exposure with CRP and IL-6 levels and tested whether adding suPAR to the measurement of CRP or IL-6 levels provides additional information about inflammation associated with stress and violence exposure. Methods Sample This cohort study included members of the E-Risk Longitudinal Twin Study, which tracks the development of a 1994-1995 birth cohort of 2232 British children. In brief, the E-Risk Longitudinal Twin Study sample was constructed from 1999 to 2000, when 1116 of 1203 eligible families (92.8%) with same-sex 5-year-old twins participated in home-visit assessments. This sample comprises 1242 (55.7%) monozygotic (MZ) and 990 (44.4%) dizygotic (DZ) twins; sex was evenly distributed within zygosity (1140 [51.1%] female). Home visits were conducted when participants were aged 5, 7, 10, 12, and, most recently, 18 years of age (n = 2066 [92.6%]). At 18 years of age, each twin was interviewed by a different interviewer. The Joint South London and Maudsley and the Institute of Psychiatry research ethics committee approved each phase of the study. Parents gave written informed consent, and twins gave oral assent between 5 and 12 years of age and then written informed consent at 18 years of age. Plasma samples were analyzed in July 2018, and statistical analysis was performed from October 1, 2018, to May 31, 2019.

thics committee approved each phase of the study. Parents gave written informed consent, and twins gave oral assent between 5 and 12 years of age and then written informed consent at 18 years of age. Plasma samples were analyzed in July 2018, and statistical analysis was performed from October 1, 2018, to May 31, 2019. The sample represents socioeconomic conditions in the United Kingdom, as reflected in the families’ distribution on a neighborhood-level socioeconomic index (ACORN [A Classification of Residential Neighborhoods], developed by CACI Inc for commercial use): 25.6% of E-Risk Longitudinal Twin Study families live in wealthy achiever neighborhoods compared with 25.3% nationwide, 5.3% vs 11.6% live in urban prosperity neighborhoods, 29.6% vs 26.9% live in comfortably off neighborhoods, 13.4% vs 13.9% live in moderate means neighborhoods, and 26.1% vs 20.7% live in hard-pressed neighborhoods. The E-Risk Longitudinal Twin Study underrepresents urban prosperity neighborhoods because such households are often childless.

urban prosperity neighborhoods, 29.6% vs 26.9% live in comfortably off neighborhoods, 13.4% vs 13.9% live in moderate means neighborhoods, and 26.1% vs 20.7% live in hard-pressed neighborhoods. The E-Risk Longitudinal Twin Study underrepresents urban prosperity neighborhoods because such households are often childless. Exposure to Adverse Experiences in Childhood and Adolescence We assessed 4 forms of stressful experiences in childhood: (1) ACEs, as introduced by the US Centers for Disease Control and Prevention ACEs Study and expanded by the Philadelphia Urban ACE Survey; (2) exposure to 6 types of severe childhood experiences of stress or violence between birth to 12 years of age; (3) exposure to 7 types of severe adolescent experiences of stress or violence at 12 to 18 years of age; and (4) exposure to cumulative stress and violence experiences throughout the life, as determined by applying latent class analysis (LCA) to stress and violence experiences data in childhood and adolescence. These measures have been reported previously and are described in the Box and detailed in eMethods 1 to 4 in the Supplement.

sure to cumulative stress and violence experiences throughout the life, as determined by applying latent class analysis (LCA) to stress and violence experiences data in childhood and adolescence. These measures have been reported previously and are described in the Box and detailed in eMethods 1 to 4 in the Supplement. Box. Description of Study Measures ACEs Twenty ACEs were measured during childhood and adolescence up to 17 years of age: 10 conventional ACEs, corresponding to the 10 subcategories of childhood adversities introduced by the CDC Adverse Childhood Experiences Study, and 10 expanded ACEs, identified from the results from the Philadelphia Urban ACE Survey and routine activity theory, as previously described. The 10 conventional ACEs included physical abuse, sexual abuse, emotional abuse, physical neglect, emotional neglect, domestic violence exposure, household substance abuse, family history of mental illness, loss of a parent (parental death, separation, or divorce), and parental antisocial behavior. The 10 expanded ACEs included experiencing bullying, living in foster care, low childhood socioeconomic status, peer substance use, low parental monitoring (as evaluated by parents), low parental monitoring (as evaluated by children), participant-perceived unsafe neighborhood, high neighbor crime violence measured via neighbor survey, neighborhood rated as unsafe through systematic social observation, and high-crime neighborhood measured through official police records. Measurement details are provided in eMethods 1 in the Supplement. Among children in this study, 197 (14.2%) had no ACEs, 238 (17.1%) had 1 ACE, 229 (16.5%) had 2 ACES, and 727 (52.3%) had 3 or more ACES.

unsafe through systematic social observation, and high-crime neighborhood measured through official police records. Measurement details are provided in eMethods 1 in the Supplement. Among children in this study, 197 (14.2%) had no ACEs, 238 (17.1%) had 1 ACE, 229 (16.5%) had 2 ACES, and 727 (52.3%) had 3 or more ACES. Severe Childhood Experiences of Stress or Violence Exposure to 6 types of severe childhood experiences of stress or violence was assessed repeatedly when the children were 5, 7, 10, and 12 years of age, including exposure to domestic intimate partner violence between the mother and her partner, frequent bullying by peers, physical maltreatment by an adult, sexual abuse, emotional abuse and neglect, and physical neglect. Exposures were coded from 12-year dossiers for each child that comprised information from home visit staff, mothers, children, family physicians, and child protection interventions. Each exposure during childhood was coded on a 3-point scale (0 indicating no exposure, 1 indicating probable or less severe exposure, and 2 indicating definite or severe exposure). Following the guidelines by Finkelhor et al, we operationalized severe childhood experiences of stress or violence as the total number of adverse event types experienced by a child. All severe childhood experiences or stress or violence were summed. Measurement details are provided in eMethods 2 in the Supplement. Among children in this study, 1004 (72.2%) had no severe childhood experiences of stress or violence, 298 (21.4%) had 1 experience, 59 (4.2%) had 2 experiences, and 30 (2.2%) had 3 or more experiences.

dhood experiences or stress or violence were summed. Measurement details are provided in eMethods 2 in the Supplement. Among children in this study, 1004 (72.2%) had no severe childhood experiences of stress or violence, 298 (21.4%) had 1 experience, 59 (4.2%) had 2 experiences, and 30 (2.2%) had 3 or more experiences. Severe Adolescent Experiences of Stress or Violence Severe adolescent experiences of stress or violence between the ages of 12 and 18 years were assessed at age 18 years of age when the twins were interviewed using the JVQ-R2, adapted as a clinical interview. The adapted JVQ-R2 comprised 45 questions covering 7 different forms of adverse experience: crime violence, peer or sibling violence, cyber bullying, sexual abuse, maltreatment, family violence, and neglect. Like severe childhood experiences of stress or violence experiences, each exposure during adolescence was coded on a 3-point scale (0 indicating no exposure, 1 indicating probable or less severe exposure, and 2 indicating definite or severe exposure). Severe adolescent experiences of stress or violence were derived by summing the number of severe adolescent experiences of stress or violence. Measurement details are provided in eMethods 3 in the Supplement. In this study, 887 adolescents (63.8%) had no severe adverse experiences, 275 (19.8%) had 1 experience, 131 (9.4%) had 2 experiences, and 98 (7.0%) had 3 or more experiences.

d by summing the number of severe adolescent experiences of stress or violence. Measurement details are provided in eMethods 3 in the Supplement. In this study, 887 adolescents (63.8%) had no severe adverse experiences, 275 (19.8%) had 1 experience, 131 (9.4%) had 2 experiences, and 98 (7.0%) had 3 or more experiences. Cumulative Stress and Violence Experiences Three groups of stress or violence experiences were identified with latent class analysis combining the 6 childhood and the 7 adolescent measures of severe experiences of stress or violence, as previously described. The latent class analysis classified participants into groups based on the degree of each participant’s exposure (none, moderate, or severe), and the analysis was performed using only participants who experienced at least 1 form of stress or violence experience. The 3 groups identified were (1) individuals who were exposed primarily to parental intimate partner violence during childhood (n = 213 [15.3%]), (2) individuals who were mainly bullied by peers and experienced street crime during childhood and adolescence (n = 354 [25.4%]), and (3) individuals who experienced multiple types of violence during childhood and adolescence (n = 129 [9.3%]). Measurement details are provided in eMethods 4 in the Supplement. Abbreviations: ACEs, adverse childhood experiences; CDC, Centers for Disease Control and Prevention; JVQ-R2, Juvenile Victimization Questionnaire, second revision.

Cumulative Stress and Violence Experiences Three groups of stress or violence experiences were identified with latent class analysis combining the 6 childhood and the 7 adolescent measures of severe experiences of stress or violence, as previously described. The latent class analysis classified participants into groups based on the degree of each participant’s exposure (none, moderate, or severe), and the analysis was performed using only participants who experienced at least 1 form of stress or violence experience. The 3 groups identified were (1) individuals who were exposed primarily to parental intimate partner violence during childhood (n = 213 [15.3%]), (2) individuals who were mainly bullied by peers and experienced street crime during childhood and adolescence (n = 354 [25.4%]), and (3) individuals who experienced multiple types of violence during childhood and adolescence (n = 129 [9.3%]). Measurement details are provided in eMethods 4 in the Supplement. Abbreviations: ACEs, adverse childhood experiences; CDC, Centers for Disease Control and Prevention; JVQ-R2, Juvenile Victimization Questionnaire, second revision. Inflammatory Biomarkers Venous blood was collected from 1700 of the 2066 participants (82.3%) with EDTA tubes. Tubes were spun at 2500g for 10 minutes and plasma samples obtained. Samples were stored at −80°C. Plasma samples were available for 1448 participants. Plasma CRP (high-sensitivity CRP) was measured using enzyme-linked immunosorbent assay (ELISA) (Quantikine ELISA Kit DCRP00, R&D Systems) following the manufacturer’s protocol. The coefficient of variation was 5.6%. Plasma IL-6 levels were measured using ELISA (Quantikine HS ELISA Kit HS600C, R&D Systems) following the manufacturer’s protocol. The coefficient of variation was 12.6%. Plasma suPAR levels were analyzed using ELISA (suPARnostic AUTO Flex ELISA, ViroGates A/S) following the manufacturer’s protocol. The coefficient of variation was 6%.

ls were measured using ELISA (Quantikine HS ELISA Kit HS600C, R&D Systems) following the manufacturer’s protocol. The coefficient of variation was 12.6%. Plasma suPAR levels were analyzed using ELISA (suPARnostic AUTO Flex ELISA, ViroGates A/S) following the manufacturer’s protocol. The coefficient of variation was 6%. Other Variables Associated With Inflammation When participants were 18 years of age, we recorded BMI ( calculated as weight in kilograms divided by height in meters squared), body temperature, current daily smoking, current illness and injury (eTable 1 in the Supplement), and use of anti-inflammatory medication (corticosteroids) within the past 2 weeks. Children exposed to ACEs or stress and violence may also be exposed to unsanitary homes, which could be associated with elevated suPAR levels and potentially confound associations between ACEs or stress and violence exposure and suPAR level. The cleanliness of the homes was assessed when children were 12 years by home visitors answering the question, “Are visible rooms of the house clean?” (no, somewhat, or yes). Childhood socioeconomic status (SES) was defined through a standardized composite of parental income, educational level, and occupation. The 3 SES indicators were highly correlated (r = 0.57-0.67) and loaded onto 1 latent factor. The population-wide distribution of the resulting factor was divided in tertiles for analyses.

dhood socioeconomic status (SES) was defined through a standardized composite of parental income, educational level, and occupation. The 3 SES indicators were highly correlated (r = 0.57-0.67) and loaded onto 1 latent factor. The population-wide distribution of the resulting factor was divided in tertiles for analyses. Statistical Analysis Both CRP and IL-6 levels were log-transformed to improve normality of their distributions, as commonly done. Distributions of CRP, IL-6, and suPAR levels are shown in the eFigure in the Supplement. Sex-adjusted regression coefficients were calculated to test associations between the inflammatory biomarkers with clinical characteristics of the sample. The association between ACEs or stress and violence exposure and inflammation was tested using ordinary least squares regression, with continuous measures of CRP, IL-6, and suPAR levels. Models were adjusted for sex, BMI, and smoking. Analyses of suPAR levels were adjusted for cleanliness of the home, childhood SES, CRP level, or IL-6 level. We report unstandardized B and standardized β coefficients, with 95% CIs adjusted to control for the nonindependence of observations of twins within families. Estimates and 95% CIs for log-transformed variables were back-transformed by exponentiating.

or cleanliness of the home, childhood SES, CRP level, or IL-6 level. We report unstandardized B and standardized β coefficients, with 95% CIs adjusted to control for the nonindependence of observations of twins within families. Estimates and 95% CIs for log-transformed variables were back-transformed by exponentiating. To analyze associations of ACEs or stress and violence exposure with combined CRP (untransformed values) and suPAR levels or combined IL-6 (untransformed values) and suPAR levels, we created groups characterized by high or low levels of CRP and suPAR or IL-6 and suPAR, as previously described. For CRP, we used the established clinical cutoff (3 mg/L [to convert to nanomoles per liter, multiply by 9.524]) to identify participants with high CRP levels; thus, high CRP level indicates a CRP level greater than 3 mg/L (n = 287 [20.6%]). Clinical cutoffs for suPAR and IL-6 have not yet been established. To identify participants with high suPAR or high IL-6 levels, we chose a cutoff for each that corresponded to a similar percentage as high CRP level; thus, high suPAR level indicates a suPAR level greater than 3.81 ng/mL (n = 286 [20.6%]), and high IL-6 level indicates an IL-6 level greater than 1.48 pg/mL (n = 286 [20.6%]). We created 4 groups of individuals characterized by (1) low CRP and low suPAR levels (n = 920 [66.1%]), (2) high CRP and low suPAR levels (n = 185 [13.3%]), (3) low CRP and high suPAR levels (n = 184 [13.2%]), and (4) high CRP and high suPAR levels (n = 102 [7.3%]). Similarly, we created 4 groups of individuals characterized by (1) low IL-6 and low suPAR levels (n = 916 [65.9%]), (2) high IL-6 and low suPAR levels (n = 189 [13.6%]), (3) low IL-6 and high suPAR levels (n = 189 [13.6%]), and (4) high IL-6 and high suPAR levels (n = 97 [7.0%]).

and high suPAR levels (n = 102 [7.3%]). Similarly, we created 4 groups of individuals characterized by (1) low IL-6 and low suPAR levels (n = 916 [65.9%]), (2) high IL-6 and low suPAR levels (n = 189 [13.6%]), (3) low IL-6 and high suPAR levels (n = 189 [13.6%]), and (4) high IL-6 and high suPAR levels (n = 97 [7.0%]). In addition, we performed an LCA that combined ln(CRP), ln(IL-6), and suPAR to classify participants into groups based on each participant’s levels of the 3 biomarkers, accounting for clustering of twins within families. The LCA identified 3 inflammation groups of individuals (eTable 2 and eTable 3 in the Supplement): low levels of all 3 biomarkers, elevated CRP and IL-6 levels, and elevated CRP, IL-6, and suPAR levels. The association between stress and violence exposure and the inflammation groups was tested using multinomial logistic regression, reporting odds ratios (ORs) with 95% CIs. Two-sided P < .05 was a priori designated statistically significant.

rkers, elevated CRP and IL-6 levels, and elevated CRP, IL-6, and suPAR levels. The association between stress and violence exposure and the inflammation groups was tested using multinomial logistic regression, reporting odds ratios (ORs) with 95% CIs. Two-sided P < .05 was a priori designated statistically significant. Results Of the 2066 children participating at 18 years of age, 1419 (68.7%) had complete data for childhood and adolescent adverse experiences; CRP, IL-6, and suPAR measurements at 18 years of age; and the covariates BMI and smoking. Participants with levels greater than 4 SDs above the means of CRP (n = 18), IL-6 (n = 7), or suPAR (n = 3) levels were excluded, leaving a final sample of 1391 (mean [SD] age, 18.4 [0.36] years; 733 [52.7%] female). No significant differences were found for those with complete data included in this article vs those without in terms of mean SES (mean, 1.99 [95% CI, 1.95-2.04] vs 2.02 [95% CI, 1.97-2.08]; P = .47), ACEs (mean, 3.22 [95% CI, 3.08-3.36] vs 3.04 [95% CI, 2.86-3.21]; P = .17), severe childhood experiences of stress or violence (mean, 0.36 [95% CI, 0.33-0.40] vs 0.34 [95% CI, 0.29-0.39]; P = .49), or severe adolescent experiences of stress or violence (mean, 0.60 [95% CI, 0.55-0.65] vs 0.55 [95% CI, 0.49-0.62]; P = .34).

s (mean, 3.22 [95% CI, 3.08-3.36] vs 3.04 [95% CI, 2.86-3.21]; P = .17), severe childhood experiences of stress or violence (mean, 0.36 [95% CI, 0.33-0.40] vs 0.34 [95% CI, 0.29-0.39]; P = .49), or severe adolescent experiences of stress or violence (mean, 0.60 [95% CI, 0.55-0.65] vs 0.55 [95% CI, 0.49-0.62]; P = .34). Correlates of CRP, IL-6, and suPAR are given in Table 1. Compared with male participants, female participants had higher levels of CRP (r = 0.14; 95% CI, 0.08-0.20; P < .001) and suPAR (r = 0.23; 95% CI, 0.17-0.29; P < .001) but not IL-6 (r = 0.002; 95% CI, −0.06 to 0.06; P = .95). Participants with high BMIs had higher levels of all 3 inflammatory biomarkers (CRP: r = 0.35; 95% CI, 0.30-0.40; P < .001; IL-6: r = 0.19; 95% CI, 0.13-0.24; P < .001; and suPAR: r = 0.29; 95% CI, 0.22-0.35; P < .001). Tobacco smoking was associated with elevated levels of IL-6 (r = 0.06; 95% CI, 0.005-0.12; P = .03) and suPAR (r = 0.22; 95% CI, 0.16-0.28; P < .001) but not CRP (r = 0.04; 95% CI, −0.02 to 0.09; P = .20). In contrast, acute conditions, such as body temperature and current illness or injury (Table 1 and eTable 1 in the Supplement), were associated with elevated levels of CRP (body temperature: r = 0.06; 95% CI, 0.004-0.12; P = .04; current illness or injury: r = 0.18; 95% CI, 0.12-0.23; P < .001) and IL-6 (current illness or injury: r = 0.14; 95% CI, 0.08-0.20; P < .001) but not suPAR (body temperature: r = 0.03; 95% CI, −0.03 to 0.09; P = .31; current illness or injury: r = 0.02; 95% CI, −0.03 to 0.08; P = .36). Use of anti-inflammatory medication (corticosteroids) in the past 2 weeks was not associated with any of the inflammatory biomarkers (CRP: r = 0.01; 95% CI, −0.04 to 0.06; P = .65: IL-6: r = −0.04; 95% CI, −0.15 to 0.07; P = .46; suPAR: r = 0.008; 95% CI, −0.07 to 0.08; P = .83). The 3 inflammatory biomarkers (CRP, IL-6, and suPAR) were positively intercorrelated (r = 0.25–0.39) (Table 1).

s was not associated with any of the inflammatory biomarkers (CRP: r = 0.01; 95% CI, −0.04 to 0.06; P = .65: IL-6: r = −0.04; 95% CI, −0.15 to 0.07; P = .46; suPAR: r = 0.008; 95% CI, −0.07 to 0.08; P = .83). The 3 inflammatory biomarkers (CRP, IL-6, and suPAR) were positively intercorrelated (r = 0.25–0.39) (Table 1). Table 1. Sex-Adjusted Correlates of Plasma CRP, IL-6, and suPAR at 18 Years of Age in the E-Risk Longitudinal Twin Study Variable Total CRPa IL-6a suPAR r (95% CI)b P Value r (95% CI)b P Value r (95% CI)b P Value Correlates of inflammation Female, No./total No. (%) 733/1391 (52.7) 0.14 (0.08 to 0.20) <.001 0.002 (−0.06 to 0.06) .95 0.23 (0.17 to 0.29) <.001 BMI, mean (SE) 22.9 (0.15) 0.35 (0.30 to 0.40) <.001 0.19 (0.13 to 0.24) <.001 0.29 (0.22 to 0.35) <.001 Daily smoking, No./total No. (%) 315/1391 (22.6) 0.04 (−0.02 to 0.09) .20 0.06 (0.005 to 0.12) .03 0.22 (0.16 to 0.28) <.001 Body temperature, mean (SE), °C 36.3 (0.02) 0.06 (0.004 to 0.12) .04 0.06 (−0.01 to 0.12) .07 0.03 (−0.03 to 0.09) .31 Current illness or injury, No./total No. (%)c 323/1390 (23.2) 0.18 (0.12 to 0.23) <.001 0.14 (0.08 to 0.20) <.001 0.02 (−0.03 to 0.08) .36 Anti-inflammatory medication use, No./total No. (%)d 17/1391 (1.2) 0.01 (−0.04 to 0.06) .65 −0.04 (−0.15 to 0.07) .46 0.008 (−0.07 to 0.08) .83 Cleanliness of home, No./total No. (%) 107/1342 (8.0)e −0.04 (−0.10 to 0.02) .22 −0.07 (−0.12 to −0.005) .03 −0.11 (−0.17 to −0.05) <.001 Socioeconomic status, No./total No. (%) 466/1391 (33.5)f −0.05 (−0.11 to 0.01) .09 −0.10 (−0.16 to −0.03) .002 −0.16 (−0.22 to −0.10) <.001 Inflammatory biomarkers, mean (SE) CRP level, mg/La 2.34 (0.11) NA NA 0.39 (0.33 to 0.44) <.001 0.25 (0.20 to 0.31) <.001 IL-6 level, pg/mLa 1.19 (0.03) 0.39 (0.33 to 0.44) <.001 NA NA 0.27 (0.21 to 0.33) <.001 suPAR level, ng/mL 3.23 (0.03) 0.25 (0.20 to 0.31) <.001 0.27 (0.21 to 0.33) <.001 NA NA Abbreviations: BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); CRP, C-reactive protein; E-Risk, Environmental Risk; IL-6, interleukin 6; NA, not applicable; suPAR, soluble urokinase plasminogen activator receptor.

3) 0.25 (0.20 to 0.31) <.001 0.27 (0.21 to 0.33) <.001 NA NA Abbreviations: BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); CRP, C-reactive protein; E-Risk, Environmental Risk; IL-6, interleukin 6; NA, not applicable; suPAR, soluble urokinase plasminogen activator receptor. SI conversion factor: to convert CRP to nanomoles per liter, multiply by 9.524. a Log-transformed (natural logarithm). b Standardized estimated regression coefficients; all correlations were adjusted for sex, and P values were adjusted for clustering within families. c The current illness or injury index is a count of 16 conditions present on the day of blood sample obtainment. Correlations with individual illnesses and injuries are provided in eTable 1 in the Supplement. d Any use of anti-inflammatory medication (corticosteroids) in the past 2 weeks. e Number of homes reported to be unclean by home visitors when children were 12 years of age. A total of 205 (15.3%) were reported as somewhat clean, and 1030 (76.8%) were reported as clean. f Number of families scored as low on the social class composite. A total of 469 (33.7%) were scored as middle social class, and 456 (32.8%) were scored as high social class.

e Number of homes reported to be unclean by home visitors when children were 12 years of age. A total of 205 (15.3%) were reported as somewhat clean, and 1030 (76.8%) were reported as clean. f Number of families scored as low on the social class composite. A total of 469 (33.7%) were scored as middle social class, and 456 (32.8%) were scored as high social class. Association of Adverse Experiences With Inflammatory Biomarker Levels Children exposed to more ACEs had higher levels of suPAR and IL-6 but not CRP at 18 years of age. Only the association with suPAR remained after adjustment for sex, BMI, and smoking (CRP: B = 1.00; 95% CI, 0.97-1.03; IL-6: B = 1.01; 95% CI, 1.00-1.03; suPAR: B = 0.03; 95% CI, 0.01-0.05) (Table 2). In general, adverse experiences were most strongly associated with suPAR at 18 years of age. Participants who experienced stress and violence exposure as children had higher levels of suPAR and IL-6 but not CRP even after controlling for sex, BMI, and smoking (suPAR: B = 0.09, 95% CI, 0.01-0.17; IL-6: B = 1.06; 95% CI, 1.01-1.12; CRP: B = 1.04; 95% CI, 0.92-1.17) (Table 2). Participants who experienced stress and violence exposure as adolescents had higher levels of suPAR and CRP but not IL-6 (suPAR: B = 0.11; 95% CI, 0.05-0.18; CRP: B = 1.09, 95% CI, 1.00-1.18; IL-6: B = 1.02; 95% CI, 0.99-1.06), but none of these associations remained after adjustment for covariates (Table 2). Participants exposed to cumulative adverse experiences across childhood and adolescence had elevated suPAR levels. In particular, levels of suPAR (but not CRP or IL-6) were elevated among those exposed to domestic violence (suPAR: B = 0.25, 95% CI, 0.10-0.40; CRP: B = 1.04; 95% CI, 0.83-1.29; IL-6: B = 1.01; 95% CI, 0.91-1.13) and those who experienced multiple types of violence in childhood and adolescence (suPAR: B = 0.26; 95% CI, 0.07-0.45; CRP: B = 1.06; 95% CI, 0.81-1.38; IL-6: B = 1.02; 95% CI, 0.88-1.18) even after controlling for sex, BMI, and smoking (Table 2).

40; CRP: B = 1.04; 95% CI, 0.83-1.29; IL-6: B = 1.01; 95% CI, 0.91-1.13) and those who experienced multiple types of violence in childhood and adolescence (suPAR: B = 0.26; 95% CI, 0.07-0.45; CRP: B = 1.06; 95% CI, 0.81-1.38; IL-6: B = 1.02; 95% CI, 0.88-1.18) even after controlling for sex, BMI, and smoking (Table 2). Table 2.

40; CRP: B = 1.04; 95% CI, 0.83-1.29; IL-6: B = 1.01; 95% CI, 0.91-1.13) and those who experienced multiple types of violence in childhood and adolescence (suPAR: B = 0.26; 95% CI, 0.07-0.45; CRP: B = 1.06; 95% CI, 0.81-1.38; IL-6: B = 1.02; 95% CI, 0.88-1.18) even after controlling for sex, BMI, and smoking (Table 2). Table 2. Associations of Childhood Adversities With Plasma CRP, IL-6, and suPAR Levels at 18 Years of Age in 1391 Participants in the E-Risk Longitudinal Twin Study Measure CRPa IL-6a suPAR Unadjusted Adjustedb Unadjusted Adjustedb Unadjusted Adjustedb B (95% CI)c,d β (95% CI)e B (95% CI)c,d β (95% CI)e B (95% CI)c,d β (95% CI)e B (95% CI)c,d β (95% CI)e B (95% CI)c β (95% CI)e B (95% CI)c β (95% CI)e Adverse childhood experiences 1.01 (0.98 to 1.04) 0.02 (−0.04 to 0.08) 1.00 (0.97 to 1.03) 0.002 (−0.06 to 0.06) 1.02 (1.01 to 1.03) 0.08 (0.02 to 0.13) 1.01 (1.00 to 1.03) 0.06 (−0.01 to 0.12) 0.05 (0.02 to 0.07) 0.13 (0.06 to 0.19) 0.03 (0.01 to 0.05) 0.08 (0.02 to 0.15) Childhood experiences of stress or violence 1.07 (0.95 to 1.22) 0.03 (−0.03 to 0.09) 1.04 (0.92 to 1.17) 0.02 (−0.04 to 0.08) 1.08 (1.03 to 1.14) 0.09 (0.03 to 0.14) 1.06 (1.01 to 1.12) 0.07 (0.01 to 0.13) 0.15 (0.06 to 0.24) 0.11 (0.05 to 0.17) 0.09 (0.01 to 0.17) 0.07 (0.01 to 0.12) Adolescent experiences of stress or violence 1.09 (1.00 to 1.18) 0.05 (−0.002 to 0.11) 1.05 (0.97 to 1.14) 0.03 (−0.02 to 0.09) 1.02 (0.99 to 1.06) 0.03 (−0.02 to 0.09) 1.01 (0.97 to 1.05) 0.01 (−0.05 to 0.07) 0.11 (0.05 to 0.18) 0.11 (0.05 to 0.18) 0.04 (−0.02 to 0.10) 0.04 (−0.02 to 0.10) Groups of Cumulative Experiences of Stress or Violence Exposure to parental intimate partner violence in childhoodf 1.10 (0.87 to 1.40) 0.07 (−0.10 to 0.24) 1.04 (0.83 to 1.29) 0.03 (−0.13 to 0.18) 1.04 (0.93 to 1.16) 0.07 (−0.11 to 0.24) 1.01 (0.91 to 1.13) 0.02 (−0.15 to 0.20) 0.33 (0.17 to 0.49) 0.36 (0.19 to 0.53) 0.25 (0.10 to 0.40) 0.27 (0.11 to 0.43) Exposure to peer and street crime stress and violence during childhood and adolescencef 1.20 (1.00 to 1.44) 0.13 (−0.003 to 0.26) 1.06 (0.89 to 1.27) 0.04 (−0.08 to 0.17) 1.02 (0.94 to 1.11) 0.04 (−0.09 to 0.17) 0.99 (0.91 to 1.07) −0.02 (−0.15 to 0.11) 0.16 (0.03 to 0.29) 0.17 (0.03 to 0.31) 0.02 (−0.10 to 0.13) 0.02 (−0.11 to 0.14) Exposure to multiple types of violence during childhood and adolescencef 1.14 (0.86 to 1.52) 0.10 (−0.11 to 0.30) 1.06 (0.81 to 1.38) 0.04 (−0.16 to 0.23) 1.07 (0.93 to 1.23) 0.11 (−0.12 to 0.33) 1.02 (0.88 to 1.18) 0.03 (−0.20 to 0.26) 0.43 (0.22 to 0.64)

) 0.17 (0.03 to 0.31) 0.02 (−0.10 to 0.13) 0.02 (−0.11 to 0.14) Exposure to multiple types of violence during childhood and adolescencef 1.14 (0.86 to 1.52) 0.10 (−0.11 to 0.30) 1.06 (0.81 to 1.38) 0.04 (−0.16 to 0.23) 1.07 (0.93 to 1.23) 0.11 (−0.12 to 0.33) 1.02 (0.88 to 1.18) 0.03 (−0.20 to 0.26) 0.43 (0.22 to 0.64) 0.47 (0.24 to 0.69) 0.26 (0.07 to 0.45) 0.28 (0.08 to 0.48) Abbreviations: CRP, C-reactive protein; E-Risk, Environmental Risk; IL-6, interleukin 6; suPAR, soluble urokinase plasminogen activator receptor. a Log-transformed (natural logarithm). b Adjusted for sex, body mass index, and smoking. c Unstandardized B coefficient for ordinary least squares regression model, in which a 1-unit change in the variable (eg, adverse childhood experiences) is associated with a corresponding change in B, holding all other variables constant. d Log-transformed estimates and 95% CIs were back-transformed by exponentiating. e Standardized β coefficients. f Estimates represent mean differences from the no adverse experience group. Correlations among different types of adverse experiences (eg, physical abuse, street crime, and cyber bullying) and CRP, IL-6, and suPAR levels are presented in eTable 4 in the Supplement. suPAR was associated with various forms of adverse experiences, suggesting that the elevation in suPAR levels among young people who underwent adverse experiences was not specific to any particular type of adverse experience but was rather a function of cumulative exposure.

are presented in eTable 4 in the Supplement. suPAR was associated with various forms of adverse experiences, suggesting that the elevation in suPAR levels among young people who underwent adverse experiences was not specific to any particular type of adverse experience but was rather a function of cumulative exposure. Home visitor ratings allowed us to investigate whether living in a nonhygienic home during childhood and adolescence explained the association between adverse experiences and inflammation. Cleanliness of the home was negatively correlated with levels of suPAR (r = −0.11; 95% CI, −0.17 to −0.05; P < .001) and IL-6 (r = −0.07; 95% CI, −0.12 to −0.005; P = .03) but not CRP (r = −0.04; 95% CI, −0.10 to 0.02; P = .22) (Table 1). When controlling for cleanliness of the home, all adverse experience exposures remained associated with elevated suPAR level at 18 years of age (eTable 5 in the Supplement). Childhood SES was negatively correlated with levels of suPAR (r = −0.16; 95% CI, −0.22 to −0.10; P < .001) and IL-6 (r = −0.10; 95% CI, −0.16 to −0.03; P = .002) but not CRP (r = −0.05; 95% CI, −0.11 to 0.01; P = .09) (Table 1). When controlling for SES, all adverse experience exposures remained associated with elevated suPAR level at 18 years of age (eTable 5 in the Supplement).

r = −0.16; 95% CI, −0.22 to −0.10; P < .001) and IL-6 (r = −0.10; 95% CI, −0.16 to −0.03; P = .002) but not CRP (r = −0.05; 95% CI, −0.11 to 0.01; P = .09) (Table 1). When controlling for SES, all adverse experience exposures remained associated with elevated suPAR level at 18 years of age (eTable 5 in the Supplement). Importance of suPAR in Adverse Experience–Associated Inflammation Adverse experiences were associated with suPAR apart from any association with CRP and IL-6 (eTable 5 in the Supplement). Measuring suPAR in addition to CRP or IL-6 increased the association between stress exposure and inflammatory burden. For example, after adjusting for CRP and IL-6 levels, each additional adverse childhood experience was associated with a 0.05-mL (95% CI, 0.03-0.07 ng/mL) increase in suPAR, each additional severe childhood experience of stress or violence was associated with a 0.14-ng/mL (95% CI, 0.06-0.22 ng/mL) increase in suPAR, and each additional severe adolescent experience of stress or violence was associated with a 0.10-ng/mL (95% CI, 0.04-0.16 ng/mL) increase in suPAR. Young people who were exposed to more ACEs, childhood stress and violence exposure, adolescent stress and violence exposure, and cumulative stress and violence exposure were significantly more likely to have elevated levels of both CRP and suPAR at 18 years of age (ACEs: OR, 1.41; 95% CI, 1.16-1.71; P < .001; childhood stress and violence exposure: OR, 1.33; 95% CI, 1.12-1.57; P = .001; adolescent stress and violence exposure: OR, 1.28; 95% CI, 1.05-1.56; P = .02; cumulative stress and violence exposure: OR, 2.51; 95% CI, 1.27-4.97; P = .008) (Figure, A, and eTable 6 in the Supplement) as well as elevated levels of IL-6 and suPAR at 18 years of age (ACEs: OR, 1.39; 95% CI, 1.13-1.71; P = .002; childhood stress and violence exposure: OR, 1.34; 95% CI, 1.11-1.63; P = .003; adolescent stress and violence exposure: OR, 1.35; 95% CI, 1.12-1.64; P = .002; cumulative stress and violence exposure: OR, 3.04; 95% CI, 1.61-5.73; P < .001) (Figure, B, and eTable 6 in the Supplement).

, 1.39; 95% CI, 1.13-1.71; P = .002; childhood stress and violence exposure: OR, 1.34; 95% CI, 1.11-1.63; P = .003; adolescent stress and violence exposure: OR, 1.35; 95% CI, 1.12-1.64; P = .002; cumulative stress and violence exposure: OR, 3.04; 95% CI, 1.61-5.73; P < .001) (Figure, B, and eTable 6 in the Supplement). Figure. Frequency of Adverse Childhood Experiences (ACEs) and Severe Experiences of Stress or Violence in Childhood or Adolescence Stratified by Inflammatory Biomarker Levels Mean number of ACEs, number of types of adverse experiences during childhood, number of types of adverse experiences during adolescence, or percentage of young people exposed to multiple types of violence during childhood or adolescence for individuals of the Environmental Risk (E-Risk) Longitudinal Twin Study stratified in groups of high C-reactive protein (CRP) level (>3 mg/L) and/or high soluble urokinase plasminogen activator receptor (suPAR) level (>3.81 ng/mL) (A), high interleukin 6 (IL-6) level (>1.48 pg/mL) and/or high suPAR level (>3.81 ng/mL) (B); or inflammation groups identified by latent class analysis (ie, low inflammation, elevated CRP and IL-6 levels, and elevated CRP, IL-6, and suPAR levels) (C). Error bars indicate 95% CIs.

r receptor (suPAR) level (>3.81 ng/mL) (A), high interleukin 6 (IL-6) level (>1.48 pg/mL) and/or high suPAR level (>3.81 ng/mL) (B); or inflammation groups identified by latent class analysis (ie, low inflammation, elevated CRP and IL-6 levels, and elevated CRP, IL-6, and suPAR levels) (C). Error bars indicate 95% CIs. However, adverse experiences were also prominent in the group of participants with low CRP and high suPAR (or low IL-6 and high suPAR) levels, who would inadvertently have been misclassified as having low inflammation if suPAR levels had not been measured. Similarly, the LCA revealed that young people who had elevated suPAR levels at 18 years of age in addition to elevated CRP and IL-6 levels had been exposed to more ACEs and childhood, adolescent, and cumulative stress and violence experiences compared with those with low levels of all 3 inflammatory biomarkers or with those with mainly high levels of CRP and IL-6 but not suPAR (Figure, C, and eTable 7 in the Supplement).

on to elevated CRP and IL-6 levels had been exposed to more ACEs and childhood, adolescent, and cumulative stress and violence experiences compared with those with low levels of all 3 inflammatory biomarkers or with those with mainly high levels of CRP and IL-6 but not suPAR (Figure, C, and eTable 7 in the Supplement). Discussion In this 2-decade prospective cohort study, we tested the usefulness of a new biomarker, suPAR, to understand the biological association with stress in the first part of the life course. First, children exposed to adversities and to multiple forms of stress and violence during childhood and adolescence had elevated suPAR levels by the time they reached 18 years of age. The association could not be attributed to BMI, smoking, or living in unhygienic homes during that period. Second, suPAR levels appear to add information about the health implications of stressful experiences in childhood beyond the more established biomarkers CRP and IL-6. We observed the strongest associations between stress exposure and inflammation when combining biomarkers, and we also found that adverse experiences were prominent in the group of participants with low CRP or low IL-6 level, who would have inadvertently been assigned to the low inflammation group if suPAR levels had not been assayed. These results replicate findings from the Dunedin Longitudinal Study, which found that exposure to ACEs was associated with elevated levels of suPAR at midlife. Thus, suPAR may be a valuable adjunct to estimating inflammatory burden that may be associated with childhood stress exposure. This conclusion was supported by an LCA that revealed that young people with elevated inflammation could be separated into 2 categories: those with elevated CRP and IL-6 levels and those with elevated CRP, IL-6, and suPAR levels, the latter of which had stronger associations with adverse experiences.

d stress exposure. This conclusion was supported by an LCA that revealed that young people with elevated inflammation could be separated into 2 categories: those with elevated CRP and IL-6 levels and those with elevated CRP, IL-6, and suPAR levels, the latter of which had stronger associations with adverse experiences. Inflammation has been the target of investigation for researchers seeking to understand the biological variables associated with stress. Much of this work has relied on measuring CRP and IL-6 levels. A meta-analysis revealed significant associations between ACEs and levels of these 2 inflammatory biomarkers. However, effects were small, and more than half of the studies, many of which were well powered, did not reveal positive associations.

with stress. Much of this work has relied on measuring CRP and IL-6 levels. A meta-analysis revealed significant associations between ACEs and levels of these 2 inflammatory biomarkers. However, effects were small, and more than half of the studies, many of which were well powered, did not reveal positive associations. Part of the difficulty may be that traditional markers of inflammation mix historical and acute effects; for example, CRP and IL-6 are involved in the acute-phase response. suPAR, the soluble form of uPAR, has been put forward as a marker of chronic inflammation. The expression and shedding of uPAR are upregulated under inflammatory conditions and in response to immunologic stimuli. Thus, the suPAR level is elevated in many diseases with an inflammatory component, whereas it is generally low, although still detectable, in healthy individuals. Common risk factors for chronic disease, such as smoking and morbid obesity, are linked to elevated suPAR levels, and elevated suPAR levels are associated with development and progression of disease and adverse outcomes, including mortality. In contrast to many markers of inflammation, which are labile and rapidly upregulated and downregulated, suPAR appears to be more stable and less sensitive to acute influences and does not fluctuate with circadian rhythm. Of the 2 categories of elevated inflammation identified by the LCA in this study, the one with elevated suPAR in addition to CRP and IL-6 was more strongly associated with stress and violence exposure during childhood and adolescence than the one with mainly elevated CRP and IL-6 levels. This finding supports the conclusion that adding suPAR to CRP and IL-6 measurement may improve the assessment of chronic inflammation associated with early-life stress.

was more strongly associated with stress and violence exposure during childhood and adolescence than the one with mainly elevated CRP and IL-6 levels. This finding supports the conclusion that adding suPAR to CRP and IL-6 measurement may improve the assessment of chronic inflammation associated with early-life stress. Limitations This study has limitations. First, plasma samples were only available for 1448 participants in the longitudinal study. However, no significant stress-exposure differences were found between participants who did and did not provide blood samples. Second, we studied twins, who may not represent singletons. However, the same pattern of associations between adverse experiences and suPAR levels was found in the Dunedin Longitudinal Study cohort of singletons. Third, although the distributional properties of suPAR are appealing for research purposes, the optimal threshold for its use as a diagnostic biomarker has not yet been determined. Fourth, the detected effect sizes were modest for suPAR, although this is to be expected in a sample of generally healthy young adults. Moreover, after we examined cumulative adverse experiences, effect sizes increased, a finding that underscores the importance of evaluating multiple risk factor exposures rather than any single exposure. Fifth, we collected inflammation data for the first time only in participants at 18 years of age, preventing us from analyzing the association of stress and violence exposure with inflammation trajectories over time. Longitudinal studies of suPAR are needed. In addition, randomized clinical trials of interventions intended to reduce effects of violence exposure should include inflammation biomarkers as outcome measures. One study found that favorable lifestyle changes were associated with reduced suPAR levels, suggesting suPAR as an outcome measure for studies of potential reversibility of chronic inflammation associated with early-life risk exposures. Sixth, we were able to identify risk factors associated with elevated suPAR levels, but because of the observational study design, we cannot rule out noncausal, alternative explanations of the associations.

measure for studies of potential reversibility of chronic inflammation associated with early-life risk exposures. Sixth, we were able to identify risk factors associated with elevated suPAR levels, but because of the observational study design, we cannot rule out noncausal, alternative explanations of the associations. Conclusions Inflammation has been suggested to fill the black box that connects childhood stress exposure to poor adult health, and it is under vigorous investigation. The results of the present study suggest that adult inflammation is associated with childhood stress exposure, with inflammation beginning by the time young people exposed to stress reach 18 years of age. Along with a previous report, the present study suggests that adding information about suPAR to traditional biomarkers of inflammation may improve the measurement of stress-related inflammatory burden. Supplement. eMethods 1. Assessment of Adverse Childhood Experiences (ACEs) eMethods 2. Assessment of Severe Childhood Experiences of Stress or Violence eMethods 3. Assessment of Severe Adolescent Experiences of Stress or Violence eMethods 4. Assessment of Cumulative Stress and Violence Experiences eFigure. Distributions of CRP, IL-6, and suPAR in the E-Risk Longitudinal Twin Study eTable 1. Correlations (Sex-Adjusted) of Plasma CRP, Plasma IL-6, and Plasma suPAR With Individual Illnesses or Injuries on the Day of Blood Sampling at Age 18 Years in the E-Risk Study (n = 1390) eTable 2. Results of a Latent Class Analysis Using Data About Inflammation Measured With CRP, IL-6, and suPAR (n = 1390)

eFigure. Distributions of CRP, IL-6, and suPAR in the E-Risk Longitudinal Twin Study eTable 1. Correlations (Sex-Adjusted) of Plasma CRP, Plasma IL-6, and Plasma suPAR With Individual Illnesses or Injuries on the Day of Blood Sampling at Age 18 Years in the E-Risk Study (n = 1390) eTable 2. Results of a Latent Class Analysis Using Data About Inflammation Measured With CRP, IL-6, and suPAR (n = 1390) eTable 3. Levels of CRP, IL-6, and suPAR in the 3 Inflammation Groups Identified by Latent Class Analysis (n = 1390) eTable 4. Correlations Between Different Types of Adverse Experiences and CRP, IL-6, and suPAR at Age 18 Years in the E-Risk Study eTable 5. Associations of Childhood Adversities With Plasma suPAR at Age 18 Years in the E-Risk Study After Adjustment for Sex and Indicated Correlates eTable 6. Associations Between Adverse Experiences and Inflammation Groups at Age 18 Years eTable 7. Associations Between Adverse Experiences and Latent Class Inflammation Groups at Age 18 Years eReferences Click here for additional data file.

Introduction Global coverage with the first dose of a measles vaccine has plateaued at approximately 85% since 2010, increases in measles incidence have been noted in 5 of the 6 World Health Organization Regions since 2016, and at least 1 country in the Americas, Venezuela, has reestablished endemic measles virus transmission. The decelerating progress in global elimination efforts implies that measles will remain endemic in many parts of the world and that the virus will continue to test immunity levels in elimination settings for the foreseeable future. Sizeable outbreaks have recently occurred in several US states (eg, New York, Washington, and New Jersey) and in other countries (eg, Canada, Vietnam, and the Philippines), pointing to heterogeneity in vaccination coverage.

virus will continue to test immunity levels in elimination settings for the foreseeable future. Sizeable outbreaks have recently occurred in several US states (eg, New York, Washington, and New Jersey) and in other countries (eg, Canada, Vietnam, and the Philippines), pointing to heterogeneity in vaccination coverage. Factors other than lack of vaccination might contribute to measles virus transmission in settings with mature control programs. First, as with other respiratory illnesses, measles transmission is affected by contact patterns, particularly mixing within and between age groups. Second, intense contact and high population density (eg, in schools and metropolitan areas) have been associated with an increased risk for measles outbreaks. Third, studies have shown reduced antibody responses and a higher risk for measles when the first dose of the measles vaccine is administered at 12 to 14 months of age compared with when the vaccine is given at 15 months of age or older. Fourth, in the absence of boosting from wild-type disease, vaccine-induced antibody titers are known to decline over time, and vaccinated persons are potentially susceptible to infection and disease as a result of waning immunity. The ability of vaccine nonresponders and of individuals with waning immunity to transmit measles is poorly understood.

sting from wild-type disease, vaccine-induced antibody titers are known to decline over time, and vaccinated persons are potentially susceptible to infection and disease as a result of waning immunity. The ability of vaccine nonresponders and of individuals with waning immunity to transmit measles is poorly understood. A better understanding of the factors affecting measles virus transmission could help improve the allocation of public health resources for measles prevention and control in elimination and near-elimination settings. We aimed to discern factors associated with measles virus transmission in the United States after elimination. Methods Measles is nationally notifiable in the United States. Cases are reported by health care professionals and clinical laboratories, investigated by local and state health departments, classified according to standard case definitions, linked into clusters epidemiologically, and reported to the Centers for Disease Control and Prevention. We analyzed available information on all confirmed cases of measles in the United States from January 1, 2001, to December 31, 2017. Data were collected as part of standardized public health surveillance and determined by the Centers for Disease Control and Prevention not to be research involving human participants.

n. We analyzed available information on all confirmed cases of measles in the United States from January 1, 2001, to December 31, 2017. Data were collected as part of standardized public health surveillance and determined by the Centers for Disease Control and Prevention not to be research involving human participants. In this cross-sectional study, we measured the transmissibility of measles by estimation of the effective reproduction number (R), or mean number of secondary cases of measles generated per single infectious individual in a population with some level of immunity (the basic reproduction number, R0, describes transmissibility in a fully susceptible population). Sustaining measles elimination requires maintenance of R below the threshold value of 1. If R is greater than 1, on average, each person spreads measles to more than 1 other person, and a self-sustaining outbreak can occur; by contrast, if R is less than 1, on average, each person spreads measles to less than 1 other person, and transmission cannot be sustained. Building on previous analyses, we adapted an existing algorithm that uses a maximum likelihood procedure to infer R for each case, or cohort of cases, given the time in days between cases in an outbreak and the probability density function of the serial interval (time between the onset of symptoms in primary cases of measles and the secondary cases they generate). We used a serial interval for measles derived from household transmission studies with a γ probability distribution and a mean (SD) of 11.1 (2.5) days. In brief, in any given measles case series, the weight that patient i infected patient j, Wij, is the serial interval distribution applied to the number of days between the rash onsets of patients i and j, and the probability that patient j was infected by patient i, Pij, is given by Pij = Wij/(∑kWkj), where the sum in the denominator is over all potential infectors k of patient j. The estimate of the R for patient i is Ri = ∑jPij (eMethods in the Supplement).

e number of days between the rash onsets of patients i and j, and the probability that patient j was infected by patient i, Pij, is given by Pij = Wij/(∑kWkj), where the sum in the denominator is over all potential infectors k of patient j. The estimate of the R for patient i is Ri = ∑jPij (eMethods in the Supplement). We applied the method to measles surveillance data by performing the procedure for all cases of measles after the index case (first identified case in a transmission chain) in each reported cluster of cases (2-case chains and outbreaks of ≥3 cases). The algorithm assigns singleton cases (single cases with no other cases epidemiologically linked to them) an R of 0. Chains of transmission in which 2 consecutive cases of measles are too close or too far away in time based on the distribution of the serial interval and that are unexplained by other cases in the outbreak are likely to be an artifact of surveillance (eg, an unidentified common source or a missing case in a chain) and may erroneously be considered a transmission pair by the model. To account for this possibility, if a secondary case could not be ascribed to a case of measles presenting 6 through 18 days prior (ie, the observed range of serial interval values and equivalent to the central 95% CI profile of the serial interval), the supposed connection was excluded and the secondary case was reassigned as an index case and the procedure continued. This method allowed for inclusion in the analyses of all transmissions before and after the supposed transmission between consecutive cases.

ent to the central 95% CI profile of the serial interval), the supposed connection was excluded and the secondary case was reassigned as an index case and the procedure continued. This method allowed for inclusion in the analyses of all transmissions before and after the supposed transmission between consecutive cases. When there is more than 1 measles case in any particular day of an outbreak, the method averages the number of forward transmissions originating from the cohort of patients with measles presenting that day, thereby increasing or decreasing the estimated contribution of any given case to transmission. However, the resolution of cases driving the transmission in each cohort can be improved by weighting the transmissibility of each case in a given day by characteristics associated to measles transmissibility. We weighted the transmissibility of each case in a given day by the number of doses of a measles-containing vaccine received (0, 1, ≥2, or unknown) and whether the patient was born before 1957 or on or after 1957. We adjusted transmissibility specifically by these 2 factors because receipt of measles vaccine and birth in the prevaccine era (ie, before 1957) are considered presumptive evidence of measles immunity, and levels of immunity are thought to be linked to the capacity to transmit the virus (eMethods and eTable 1 in the Supplement).

transmissibility specifically by these 2 factors because receipt of measles vaccine and birth in the prevaccine era (ie, before 1957) are considered presumptive evidence of measles immunity, and levels of immunity are thought to be linked to the capacity to transmit the virus (eMethods and eTable 1 in the Supplement). We assessed R based on the following characteristics of patients with measles: vaccination status (0, 1, or ≥2 doses or unknown), birth prior to 1957 (presumed immune from natural exposure), sex, importation status (imported or US-acquired), residency status (US resident or foreign visitor), age (in months) at first dose, time (years) since vaccination, hospitalization, presence of complications, reporting US state, and genotype. We dichotomized age at first dose (<15 or ≥15 months), as previous studies indicate reduced antibody responses and increased susceptibility to measles when the first dose is given before 15 months of age. To evaluate for changes in transmissibility due to waning immunity, we dichotomized time since vaccination (<12 or ≥12 years), with 12 years being the median number of years since vaccination for available data. In addition, we evaluated R based on the vaccination status (0 doses, ≥1 dose, or unknown) and age group (<1, 1-4, 5-17, 18-29, 30-49, and ≥50 years) of both primary cases of measles and the secondary cases of measles that they generated.

h 12 years being the median number of years since vaccination for available data. In addition, we evaluated R based on the vaccination status (0 doses, ≥1 dose, or unknown) and age group (<1, 1-4, 5-17, 18-29, 30-49, and ≥50 years) of both primary cases of measles and the secondary cases of measles that they generated. We describe the demographic and epidemiologic characteristics of potential superspreading events, defined as a case with an estimated R greater than or equal to 5 (≥99th percentile of all estimates in this data set). Sensitivity analyses were performed to examine the choice of the measles serial interval, the width of the time window for allowable connections to be made between consecutive cases, and the characteristics included in the weighting procedure to determine the factors associated with transmission (eResults in the Supplement).

ivity analyses were performed to examine the choice of the measles serial interval, the width of the time window for allowable connections to be made between consecutive cases, and the characteristics included in the weighting procedure to determine the factors associated with transmission (eResults in the Supplement). Results From 2001 to 2017, a total of 2218 confirmed measles cases were reported in the United States. Of these, 490 were single cases, 90 were 2 case-chains, and 116 were outbreaks of 3 or more cases. The median size of outbreaks was 5 cases (range, 3-383 cases) and median duration of outbreaks was 22 days (range, 3-121 days). Among the 2218 measles cases, 573 (25.8%) were internationally imported and 1645 (74.2%) were acquired in the United States. Most patients with measles were unvaccinated (1508 [68.0%]) or had an unknown vaccination status (435 [19.6%]). The date of vaccine receipt was poorly populated in our data set (available for 100 of 275 vaccinated individuals [36.4%]). Additional key characteristics of measles cases are shown in eTable 2 in the Supplement. A graphical representation of the transmission matrix for one outbreak is shown in the eFigure in the Supplement.

te of vaccine receipt was poorly populated in our data set (available for 100 of 275 vaccinated individuals [36.4%]). Additional key characteristics of measles cases are shown in eTable 2 in the Supplement. A graphical representation of the transmission matrix for one outbreak is shown in the eFigure in the Supplement. Estimates of R for measles in the United States were 0.76 (95% CI, 0.71-0.81) among patients who had received no doses of a measles-containing vaccine, 0.17 (95% CI, 0.11-0.26) among patients who had received 1 dose, 0.27 (95% CI, 0.17-0.39) among patients who had received 2 or more doses, and 0.52 (95% CI, 0.44-0.60) among those who had an unknown vaccination status. Among patients born before 1957, R was 0.35 (95% CI, 0.20-0.58), and among those born on or after 1957, R was 0.64 (95% CI, 0.61-0.68) (Figure 1). Figure 1. Estimates of the Measles Case Reproduction Number, R, by Vaccination Status and Birth Before 1957 The bars represent the 95% CIs, and the horizontal dashed line indicates the threshold value of R = 1. Results are self-consistently adjusted by the number of doses of a measles-containing vaccine received and birth before 1957.

tes of the Measles Case Reproduction Number, R, by Vaccination Status and Birth Before 1957 The bars represent the 95% CIs, and the horizontal dashed line indicates the threshold value of R = 1. Results are self-consistently adjusted by the number of doses of a measles-containing vaccine received and birth before 1957. Among unvaccinated primary cases of measles in patients who infected unvaccinated and vaccinated (≥1 doses) secondary cases of measles, R estimates were 0.61 (95% CI, 0.57-0.65) among unvaccinated individuals and 0.06 (95% CI, 0.05-0.08) among vaccinated individuals. Among vaccinated primary cases of measles in patients who infected unvaccinated and vaccinated secondary cases of measles, R estimates were 0.10 (95% CI, 0.06-0.15) among unvaccinated individuals and 0.07 (95% CI, 0.04-0.11) among vaccinated individuals (Table 1). Table 1. Estimates of the Measles Reproduction Number, R, Among Primary and Secondary Cases of Measles, by Vaccination Statusa Vaccination Status of Primary Cases Vaccination Status of Secondary Cases, R (95% CI) Unknown Dose(s)b 0 Dosesb ≥1 Dose(s)b Unknown dose(s)b 0.16 (0.12-0.20) 0.27 (0.22-0.34) 0.09 (0.06-0.12) 0 Dosesb 0.09 (0.08-0.11) 0.61 (0.57-0.65) 0.06 (0.05-0.08) ≥1 Dose(s)b 0.05 (0.03-0.09) 0.10 (0.06-0.15) 0.07 (0.04-0.11) a Results are self-consistently adjusted by the number of doses of a measles-containing vaccine received and birth before 1957. b Doses of a measles-containing vaccine; doses were counted if given at least 1 maximum incubation period (21 days) prior to the onset of rash.

Table 1. Estimates of the Measles Reproduction Number, R, Among Primary and Secondary Cases of Measles, by Vaccination Statusa Vaccination Status of Primary Cases Vaccination Status of Secondary Cases, R (95% CI) Unknown Dose(s)b 0 Dosesb ≥1 Dose(s)b Unknown dose(s)b 0.16 (0.12-0.20) 0.27 (0.22-0.34) 0.09 (0.06-0.12) 0 Dosesb 0.09 (0.08-0.11) 0.61 (0.57-0.65) 0.06 (0.05-0.08) ≥1 Dose(s)b 0.05 (0.03-0.09) 0.10 (0.06-0.15) 0.07 (0.04-0.11) a Results are self-consistently adjusted by the number of doses of a measles-containing vaccine received and birth before 1957. b Doses of a measles-containing vaccine; doses were counted if given at least 1 maximum incubation period (21 days) prior to the onset of rash. Transmission was generally assortative by age groups (ie, transmission tended to be higher between individuals of a similar age group). R estimates were higher when primary and secondary cases of measles were patients aged 5 to 17 years (0.36 [95% CI, 0.31-0.42]) compared with assortative transmission in other age groups (<1 year, 0.14 [95% CI, 0.10-0.20]; 1-4 years, 0.25 [95% CI, 0.20-0.30]; 18-29 years, 0.19 [95% CI, 0.15-0.24]; 30-49 years, 0.15 [95% CI, 0.11-0.20]; ≥50 years, 0.04 [95% CI, 0.01-0.10]) (Table 2).

es were patients aged 5 to 17 years (0.36 [95% CI, 0.31-0.42]) compared with assortative transmission in other age groups (<1 year, 0.14 [95% CI, 0.10-0.20]; 1-4 years, 0.25 [95% CI, 0.20-0.30]; 18-29 years, 0.19 [95% CI, 0.15-0.24]; 30-49 years, 0.15 [95% CI, 0.11-0.20]; ≥50 years, 0.04 [95% CI, 0.01-0.10]) (Table 2). Table 2. Estimates of the Measles Reproduction Number, R, Among Primary and Secondary Cases of Measles, by Age Groupa Age Group of Primary Cases Age Group of Secondary Cases, R (95% CI) <1 y 1-4 y 5-17 y 18-29 y 30-49 y ≥50 y <1 y 0.14 (0.10-0.20) 0.12 (0.08-0.19) 0.09 (0.05-0.15) 0.07 (0.04-0.12) 0.05 (0.03-0.10) 0.009 (0.002-0.03) 1-4 y 0.08 (0.05-0.12) 0.25 (0.20-0.30) 0.14 (0.11-0.19) 0.09 (0.06-0.12) 0.10 (0.07-0.14) 0.02 (0.008-0.04) 5-17 y 0.04 (0.03-0.07) 0.10 (0.08-0.14) 0.36 (0.31-0.42) 0.12 (0.09-0.16) 0.11 (0.08-0.14) 0.01 (0.005-0.03) 18-29 y 0.06 (0.04-0.09) 0.10 (0.07-0.13) 0.19 (0.15-0.25) 0.19 (0.15-0.24) 0.13 (0.09-0.17) 0.02 (0.01-0.05) 30-49 y 0.05 (0.03-0.08) 0.07 (0.04-0.10) 0.11 (0.08-0.16) 0.11 (0.08-0.16) 0.15 (0.11-0.20) 0.02 (0.01-0.04) ≥50 y 0.07 (0.03-0.15) 0.07 (0.03-0.16) 0.12 (0.06-0.24) 0.13 (0.06-0.25) 0.14 (0.07-0.25) 0.04 (0.01-0.10) a Results are self-consistently adjusted by the number of doses of a measles-containing vaccine received and birth before 1957.

0.07 (0.04-0.10) 0.11 (0.08-0.16) 0.11 (0.08-0.16) 0.15 (0.11-0.20) 0.02 (0.01-0.04) ≥50 y 0.07 (0.03-0.15) 0.07 (0.03-0.16) 0.12 (0.06-0.24) 0.13 (0.06-0.25) 0.14 (0.07-0.25) 0.04 (0.01-0.10) a Results are self-consistently adjusted by the number of doses of a measles-containing vaccine received and birth before 1957. Estimates of R were not substantially different based on sex, residence status, hospitalization, age at first dose, or time since vaccination (Figure 2). Estimates of R among patients who acquired measles abroad was estimated to be 0.56 (95% CI, 0.50-0.62) and among patients who acquired measles in the United States to be 0.67 (95% CI, 0.63-0.71). Estimates of R among patients reporting complications was 0.76 (95% CI, 0.66-0.88) and among those not reporting complications was 0.62 (95% CI, 0.59-0.66). Some differences in R estimates were seen based on the genotype and reporting state (Figure 3); some of these estimates were based on few cases, and most 95% CIs overlapped. Figure 2. Estimates of the Measles Case Reproduction Number, R, by Various Case Characteristics The bars represent the 95% CIs, and the horizontal dashed line indicates the threshold value of R = 1. Results are self-consistently adjusted by the number of doses of a measles-containing vaccine received and birth before 1957.

re 2. Estimates of the Measles Case Reproduction Number, R, by Various Case Characteristics The bars represent the 95% CIs, and the horizontal dashed line indicates the threshold value of R = 1. Results are self-consistently adjusted by the number of doses of a measles-containing vaccine received and birth before 1957. Figure 3. Estimates of the Measles Case Reproduction Number, R, by Reporting State and Genotype A, R estimates for selected states. Forty-seven US states, Washington, DC, and New York City reported measles cases during the study period. New York City Department of Health and Mental Hygiene and New York State Department of Health report separately. State data shown are for localities reporting 20 or more cases of measles, ordered by R point estimates: Utah (UT), 22; Michigan (MI), 28; Oregon (OR), 28; Kansas (KS), 29; North Carolina (NC), 32; New York (NY), 41; New Jersey (NJ), 42; Florida (FL), 48; Missouri (MO), 48; Hawaii (HI), 58; Texas (TX), 59; Pennsylvania (PA), 63; Arizona (AZ), 64; Massachusetts (MA), 68; Illinois (IL), 70; Indiana (IN), 73; Washington (WA), 105; Minnesota (MN), 121; New York City (NYC), 213; Ohio (OH), 397; and California (CA), 419. Measles cases are reported by state of residence, which may not necessarily be where the infection was acquired. B, R estimates for selected genotypes. Genotype data shown are for 8 genotypes identified in 15 or more cases, ordered by R point estimates: D9, 439; D5, 68; B3, 530; D3, 19; D8, 339; D4, 264; H1, 86; and D7, 15. The bars represent the 95% CIs, and the horizontal dashed line indicates the threshold value of R = 1. Results are self-consistently adjusted by the number of doses of a measles-containing vaccine received and birth before 1957.

R point estimates: D9, 439; D5, 68; B3, 530; D3, 19; D8, 339; D4, 264; H1, 86; and D7, 15. The bars represent the 95% CIs, and the horizontal dashed line indicates the threshold value of R = 1. Results are self-consistently adjusted by the number of doses of a measles-containing vaccine received and birth before 1957. We identified 23 possible superspreading events during the study period (median R = 6.1 [range, 5.0-18.1]) (eTable 3 in the Supplement). The median age of superspreaders was 17 years (range, 9 months-63 years). Nineteen of the 23 individuals (82.6%) were unvaccinated (the remaining 4 had an unknown vaccination status), and 22 (95.7%) were born on or after 1957. Superspreading events occurred during 16 outbreaks (typically early in the outbreak), which had a median size of 21 cases (range, 6-383 cases) and median duration of 44 days (range, 18-121 days). Superspreading events occurred mostly in close-contact settings (eg, hospitals, households, and schools), and most individuals with measles reported in these outbreaks were unvaccinated. Sensitivity analyses showed that varying several of the assumptions in this evaluation resulted in only small changes in general patterns of transmission (eTables 4-12 in the Supplement).

We identified 23 possible superspreading events during the study period (median R = 6.1 [range, 5.0-18.1]) (eTable 3 in the Supplement). The median age of superspreaders was 17 years (range, 9 months-63 years). Nineteen of the 23 individuals (82.6%) were unvaccinated (the remaining 4 had an unknown vaccination status), and 22 (95.7%) were born on or after 1957. Superspreading events occurred during 16 outbreaks (typically early in the outbreak), which had a median size of 21 cases (range, 6-383 cases) and median duration of 44 days (range, 18-121 days). Superspreading events occurred mostly in close-contact settings (eg, hospitals, households, and schools), and most individuals with measles reported in these outbreaks were unvaccinated. Sensitivity analyses showed that varying several of the assumptions in this evaluation resulted in only small changes in general patterns of transmission (eTables 4-12 in the Supplement). Discussion By pooling means of R based on various case characteristics, we were able to discern the factors associated with measles transmission in this postelimination setting. Principally, we found a gradient of transmission in which unvaccinated patients with measles are approximately 3 to 4 times more infectious compared with patients with measles who have been vaccinated once or twice, and that transmission is concentrated among unvaccinated primary and secondary cases of measles. Furthermore, our description of superspreading events highlight lack of vaccination as the initial spark for large outbreaks of measles. Although the measles vaccine is known to be highly effective in decreasing measles susceptibility—1 dose is 93% effective against measles and 2 doses are 97% effective—our findings also suggest an association of vaccination with limiting measles communicability and underscore the fact that measles transmission in the United States is driven by failure to vaccinate rather than a failure of vaccine performance. In addition, the low transmissibility observed from adults born when measles was still endemic (assumed to be naturally infected) supports the use of birth before vaccine introduction as acceptable presumptive evidence of measles immunity in elimination settings.

vaccinate rather than a failure of vaccine performance. In addition, the low transmissibility observed from adults born when measles was still endemic (assumed to be naturally infected) supports the use of birth before vaccine introduction as acceptable presumptive evidence of measles immunity in elimination settings. Measles transmission was assortative with age (among persons aged <50 years, approximately 30%-50% of transmission events occurred within the same age group), consistent with age-specific mixing reported in studies that quantify social encounters that are potentially infectious. A key feature associated with the preferential interaction within age groups is the finding of more pronounced contacts among school-aged children (relative to contacts between adults). Our evaluation similarly shows school-aged children as a primary conduit of measles transmission in the United States and emphasizes the importance of policies aimed at ensuring high 2-dose vaccine coverage of these children (eg, school entry immunization requirements) or presumptive communication (informing parents that vaccines are scheduled during the visit) instead of participatory communication (asking parents if they would like their children to be vaccinated) during parent-clinician encounters. Age-specific R estimates derived from the probabilistic model could help clarify the extent by which social contact patterns explain disease transmission.

uled during the visit) instead of participatory communication (asking parents if they would like their children to be vaccinated) during parent-clinician encounters. Age-specific R estimates derived from the probabilistic model could help clarify the extent by which social contact patterns explain disease transmission. R estimates among vaccinated patients with measles were generally very low, including toward unvaccinated cases (R = 0.10). These estimates might be biased because we did not differentiate between primary vaccine failure (failure to seroconvert after vaccination) and secondary vaccine failure (waning of immunity after seroconversion), and cases of measles owing to primary vaccine failure might be as transmissible as cases of measles in unvaccinated individuals. Individuals with secondary vaccine failure have a vigorous amnestic response to measles and thus might have milder symptoms and shed less virus. The presence of complications (a marker of disease severity) was independently associated with measles contagiousness. The largely restricted transmission of measles from vaccinated persons is in agreement with previous observations of no transmission from twice-vaccinated individuals with measles who develop robust antibody responses (despite exposing numerous persons). Because almost all persons who do not respond to the first dose of measles vaccine are expected to develop protective immunity after the second dose, our study provides further evidence for use of a 2-dose schedule in elimination efforts. Because measles antibody titers are known to decline slowly after measles vaccination, continued monitoring of measles among vaccinated persons is warranted in low-incidence settings.

elop protective immunity after the second dose, our study provides further evidence for use of a 2-dose schedule in elimination efforts. Because measles antibody titers are known to decline slowly after measles vaccination, continued monitoring of measles among vaccinated persons is warranted in low-incidence settings. There were subtle differences in transmissibility based on other factors. Compared with measles among foreign visitors and imported cases of measles, R values for cases of measles among US residents and US-acquired cases of measles tended to be higher. This finding might reflect the transitory nature of stays by foreign visitors and that they are less likely to contact local at-risk communities. Although some differences were also noted in R point estimates based on genotype and reporting state, 95% CIs overlapped for many of these estimates. The results presented here do not indicate that genotype B3 has increased transmissibility compared with other genotypes. Because the chance of measles spreading is dependent on the setting in which measles is introduced, differences in the observed transmissibility of a given genotype should be interpreted cautiously. For example, the higher R for Ohio and D9 is associated with an outbreak in an Amish community in 2014, likely owing to this community being highly underimmunized rather than to any characteristic of the virus (excluding this outbreak, the R estimates for Ohio was 0.16 and for D9 was 0.71). Other genotypes have been associated with large outbreaks in other settings (eg, H1 in Mongolia and D4 in France), and importations of these genotypes might have led to a similar outbreak in other underimmunized populations and would not have changed public health response efforts. Estimation of R associated with specific outbreaks can nonetheless serve as a marker of the extent of a particular immunity gap, and careful characterization of these susceptible communities can help pinpoint areas in which preventive interventions might be needed.

would not have changed public health response efforts. Estimation of R associated with specific outbreaks can nonetheless serve as a marker of the extent of a particular immunity gap, and careful characterization of these susceptible communities can help pinpoint areas in which preventive interventions might be needed. Limitations Our study has some limitations. The algorithm does not conclusively establish who infected whom and cannot replace careful epidemiologic investigation, but it is useful in identifying the overall direction of transmission. Because the likelihood of transmission depends on several factors, including the status (eg, vaccination) of both the infector and infectee, the setting in which the exposure occurred, and outbreak containment interventions, it is challenging to account for the effect of each potential confounder. For example, we did not directly evaluate the association of population density with transmission of measles, although our analysis of superspreading events indicates that close-contact settings provide opportunities for rapid dissemination of measles. Similarly, we did not evaluate the association of clustering with transmission of measles, and geographic clustering of unvaccinated persons has been linked to measles outbreaks. Unvaccinated primary cases of measles were more likely to infect unvaccinated rather than vaccinated individuals, whereas vaccinated primary cases of measles infected a similar number of unvaccinated and vaccinated individuals. Furthermore, the range of R values during superspreading events was similar to the commonly cited range of values for R0. Both observations imply that there are pockets of underimmunization in the United States. The date of vaccine receipt was poorly populated in our data set (available for approximately 36% of vaccinated cases), and our results of no difference in transmissibility by age at first dose and time since vaccination were based on few cases. These findings were also confounded by lack of differentiation between primary and secondary vaccine failure, which requires specialized testing (avidity and neutralizing antibody titers). However, our analyses suggest that vaccinated persons are inefficient transmitters of measles, and we found no notable differences in transmissibility between vaccinated individuals with measles with and without reported vaccination dates (eTable 13 in the Supplement).

esting (avidity and neutralizing antibody titers). However, our analyses suggest that vaccinated persons are inefficient transmitters of measles, and we found no notable differences in transmissibility between vaccinated individuals with measles with and without reported vaccination dates (eTable 13 in the Supplement). The outbreaks we evaluated occurred in diverse populations and were affected by several individual- and context-specific factors; thus, the relative importance of the different factors associated with transmission might not be generalizable. Finally, our comparisons of R values were qualitative and not statistical, although clear differences in transmissibility were noted and explained by underlying covariates. Conclusions The method we used allowed us to identify leading factors associated with the spread of measles in an elimination setting from high-quality surveillance data. Our findings show predominantly subcritical (R < 1) transmission of measles and maintenance of elimination in the United States for the past 17 years, establish the public health value of the measles vaccine in limiting measles infectiousness, and underscore the importance of having high targets for 2-dose measles vaccine coverage, especially among school-aged children. Supplement. eMethods. Additional Details of the Data Used for Analyses eResults. Comparison of Unadjusted and Adjusted Estimates of R eTable 1. Results of a Multiple Linear Regression Model to Assess the Relationship Between R and Several Case Characteristics

Conclusions The method we used allowed us to identify leading factors associated with the spread of measles in an elimination setting from high-quality surveillance data. Our findings show predominantly subcritical (R < 1) transmission of measles and maintenance of elimination in the United States for the past 17 years, establish the public health value of the measles vaccine in limiting measles infectiousness, and underscore the importance of having high targets for 2-dose measles vaccine coverage, especially among school-aged children. Supplement. eMethods. Additional Details of the Data Used for Analyses eResults. Comparison of Unadjusted and Adjusted Estimates of R eTable 1. Results of a Multiple Linear Regression Model to Assess the Relationship Between R and Several Case Characteristics eTable 2. Key Characteristics of 2218 Measles Case-Patients Reported in the United States, 2001-2017 eTable 3. Summary Characteristics of 23 Potential Measles Superspreading Events During 16 Measles Outbreaks eTable 4. Estimates of the Measles Case Reproduction Number, R, According to Several Characteristics, Adjusting Transmissibility by Various Covariates, United States, 2001-2017 eTable 5. Estimates of the Measles Case Reproduction Number, R, Among Primary and Secondary Cases According to Vaccination Status, Adjusting Transmissibility by Various Covariates, United States, 2001-2017 eTable 6. Estimates of the Measles Case Reproduction Number, R, According to Age Group, Adjusting Transmissibility by Various Covariates, United States, 2001-2017

eTable 5. Estimates of the Measles Case Reproduction Number, R, Among Primary and Secondary Cases According to Vaccination Status, Adjusting Transmissibility by Various Covariates, United States, 2001-2017 eTable 6. Estimates of the Measles Case Reproduction Number, R, According to Age Group, Adjusting Transmissibility by Various Covariates, United States, 2001-2017 eTable 7. Estimates of the Measles Case Reproduction Number, R, According to Several Characteristics, Estimated Using Three Different Serial Intervals, United States, 2001-2017 eTable 8. Estimates of the Measles Case Reproduction Number, R, Among Primary and Secondary Cases, by Vaccination Status, Estimated Using Three Different Serial Intervals, United States, 2001-2017 eTable 9. Measles Case Reproduction Numbers, R, Among Primary and Secondary Cases, by Age Groups, Estimated Using Three Different Serial Intervals, United States, 2001-2017 eTable 10. Estimates of the Measles Case Reproduction Number, R, According to Several Characteristics, Estimated Using Different Minimum and Maximum Serial Intervals, United States, 2001-2017 eTable 11. Estimates of the Measles Case Reproduction Number, R, Among Primary and Secondary Cases, by Vaccination Status, Estimated Using Different Minimum and Maximum Serial Intervals, United States, 2001-2017 eTable 12. Measles Case Reproduction Numbers, R, Among Primary and Secondary Cases, by Age Groups, Estimated Using Different Minimum and Maximum Serial Intervals, United States, 2001-2017

eTable 11. Estimates of the Measles Case Reproduction Number, R, Among Primary and Secondary Cases, by Vaccination Status, Estimated Using Different Minimum and Maximum Serial Intervals, United States, 2001-2017 eTable 12. Measles Case Reproduction Numbers, R, Among Primary and Secondary Cases, by Age Groups, Estimated Using Different Minimum and Maximum Serial Intervals, United States, 2001-2017 eTable 13. Estimates of the Measles Case Reproduction Number, R, Among Vaccinated Cases With and Without Dates of Vaccination Reported, United States, 2001-2017 eFigure. Outbreak Transmission Matrix Click here for additional data file.

More than 200 million cases of malaria occur yearly, with most in Africa, where infants younger than 5 years account for two-thirds of all malaria deaths. This highlights the need for successful prevention of malaria infection, especially in early life. Breastfeeding is the most efficient way to prevent child morbidity and mortality attributable to respiratory and gastrointestinal tract infectious diseases. In contrast, there is conflicting evidence on malaria prevention by breastfeeding. Mouse and human data have shown that the presence of foreign antigens in breast milk, such as allergens or viral antigens, could elicit strong immune responses in offspring who are breastfed. Therefore, we propose what is to our knowledge an original hypothesis: the presence of malaria antigen in breast milk stimulates antimalarial immune defenses and reduces malaria risk in infants who are breastfed. Here, as a critical first step to address this hypothesis, we investigated whether Plasmodium falciparum histidine-rich protein 2 (pHRP-2) and lactate dehydrogenase (pLDH) are detectable in the breast milk of mothers from Uganda, a country with endemic malaria. Methods This study included mothers who were lactating and who visited our malaria clinic at St Anne Health Center III, Katakwi District, northeastern Uganda, during the high or low malaria-transmission seasons. Five-milliliter samples of breast milk and fingerprick blood samples were collected after the mothers provided informed consent. Ethical approval for the study was provided by the Uganda National Council for Science and Technology.

akwi District, northeastern Uganda, during the high or low malaria-transmission seasons. Five-milliliter samples of breast milk and fingerprick blood samples were collected after the mothers provided informed consent. Ethical approval for the study was provided by the Uganda National Council for Science and Technology. The blood samples were used immediately to detect asymptomatic malaria by an ultrasensitive P falciparum HRP-2–based rapid diagnostic test (uRDT) (Alere Malaria Ag P.f [Standard Diagnostics Inc]). The presence of malaria antigens in breast milk samples was investigated by P falciparum–specific pHRP-2 and pLDH enzyme-linked immunosorbent assays (Quantimal CELISA [Cellabs]), with protocol adaptation (detection levels were 1.2 pg/mL and 4.8 units/mL, respectively). Data analyses were performed with Prism version 6 (GraphPad Software). We used 2-sided Fisher exact tests to address differences between groups, and P values less than .05 were considered significant. Data collection and analysis occurred from March 2018 to December 2018. Results A total of 123 mothers who were lactating visited the malaria clinic during the low malaria-transmission season; an additional 201 visited during the high transmission season. The overall mean [SD] age, body mass index (calculated as weight in kilograms divided by height in meters squared), and lactation duration of the mothers analyzed in this study were 26.2 [6.8] years, 23.6 [2.8], and 12.3 [5.5] months, respectively.

season; an additional 201 visited during the high transmission season. The overall mean [SD] age, body mass index (calculated as weight in kilograms divided by height in meters squared), and lactation duration of the mothers analyzed in this study were 26.2 [6.8] years, 23.6 [2.8], and 12.3 [5.5] months, respectively. None of the mothers had clinical malaria. When malaria transmission was low and high, 14 of 123 women (11.4%) and 74 of 201 women (36.8%), respectively, harbored asymptomatic malaria (P < .001). Among the 88 breast milk samples from mothers with asymptomatic malaria, 7 had detectable pHRP-2 (7.9%) with a median (interquartile range) level of 45.0 (2.0-180.2) pg/mL, and 10 had detectable pLDH (11.3%) with median (interquartile range) values of 6.6 (5.6-9.9) arbitrary units/mL (Figure 1). Overall, 14 breast milk samples (15.9%) were positive for either pLDH or pHRP-2, and 3 (3.4%) were positive for both pLDH and pHRP-2. Forty-four milk samples from mothers without malaria were used as control samples, and none of these showed detectable pHRP-2 or pLDH antigens (Figure 1).

arbitrary units/mL (Figure 1). Overall, 14 breast milk samples (15.9%) were positive for either pLDH or pHRP-2, and 3 (3.4%) were positive for both pLDH and pHRP-2. Forty-four milk samples from mothers without malaria were used as control samples, and none of these showed detectable pHRP-2 or pLDH antigens (Figure 1). Figure 1. Plasmodium falciparum Histidine-Rich Protein 2 and Lactate Dehydrogenase From Plasmodium Falciparum Are Present in Breast Milk From Mothers With Asymptomatic Malaria Data show the concentrations of Plasmodium falciparum histidine-rich protein 2 (pHRP-2) and lactate dehydrogenase (pLDH) in breast milk samples from mothers positive vs negative for asymptomatic malaria, as gauged by detection of pLDH in their blood by an ultrasensitive rapid diagnostic test in the absence of malaria clinical symptoms. Dotted lines indicate the limits of detection of pHRP-2 and pLDH antigens in breast milk, as determined by enzyme-linked immunoabsorbent assays. Solid lines indicate the median values among samples with detectable values.

heir blood by an ultrasensitive rapid diagnostic test in the absence of malaria clinical symptoms. Dotted lines indicate the limits of detection of pHRP-2 and pLDH antigens in breast milk, as determined by enzyme-linked immunoabsorbent assays. Solid lines indicate the median values among samples with detectable values. To address whether the detection of malaria antigens in breast milk depended on the density of P falciparum parasites in mothers’ blood circulation, we categorized the intensity of the test bands of the uRDT readout for 74 malaria-positive blood samples as faint, moderate, or intense, as a proxy measure of parasite density. In the faint category, 1 of 28 samples was positive, with a value of 1.52 pg/mL; in the moderate category, 1 of 18 samples was positive, with a value of 5.4 pg/mL; and in the intense category, 4 of 28 samples were positive, with a median (interquartile range) value of 112.0 (12.6-212.3) pg/mL (Figure 2). Further statistical analysis could not be performed because of the limited size sample. These preliminary data suggest that percentage of breast milk samples positive for pHRP-2 and the concentration of pHRP-2 in breast milk increased with the intensity of test bands.

ange) value of 112.0 (12.6-212.3) pg/mL (Figure 2). Further statistical analysis could not be performed because of the limited size sample. These preliminary data suggest that percentage of breast milk samples positive for pHRP-2 and the concentration of pHRP-2 in breast milk increased with the intensity of test bands. Figure 2. Plasmodium falciparum Histidine-Rich Protein 2 Levels in Breast Milk and Association With Levels in Maternal Blood Data show the concentrations of Plasmodium falciparum histidine-rich protein 2 (pHRP-2) in breast milk samples from mothers negative for malaria or with various levels of P falciparum parasites in their blood, as gauged by the intensity of test bands of the ultrasensitive rapid diagnostic test (uRDT). Discussion This study shows (to our knowledge for the first time) that 15% of breast milk samples from mothers with asymptomatic malaria contain malaria antigens. Our preliminary data indicate that blood levels of malaria antigens determine their levels in breast milk. These findings may have important implications for child susceptibility to malaria, since the levels and the nature of malaria antigens in breast milk may strongly influence immune responses to malaria infections in children who are breastfed. Future studies will need to address the immunological outcomes and malaria risk in infants exposed to 1 or multiple malaria antigens through breast milk. This should pave the way for novel and efficient strategies for malaria prevention that are adapted to early childhood.

Evidence to support current guidelines recommending routine antibiotic use in the outpatient management of uncomplicated severe acute malnutrition (SAM) is limited and based largely on data from historical inpatient settings. The evidence from 2 clinical trials on the effect of routine antibiotic use on nutritional recovery differs. In Malawi, where HIV and kwashiorkor prevalence are high, routine antibiotics increased nutritional recovery and decreased mortality. In Niger, where HIV and kwashiorkor prevalence are low, we found no benefit of routine amoxicillin on nutritional recovery or mortality, although children receiving amoxicillin had a reduced risk of transfer to inpatient care. Both reports only considered short-term risks and benefits during nutritional treatment (mean [SD] time to recovery, 29 [19] days in Malawi and 29 [13] days in Niger), although immunodeficiencies and risk of relapse or morbidity associated with SAM may persist beyond nutritional recovery. To broaden the available evidence, we present the first analysis (to our knowledge) to assess the outcome of routine antibiotic use in outpatient SAM management, including follow-up during and after nutritional treatment.

cies and risk of relapse or morbidity associated with SAM may persist beyond nutritional recovery. To broaden the available evidence, we present the first analysis (to our knowledge) to assess the outcome of routine antibiotic use in outpatient SAM management, including follow-up during and after nutritional treatment. Methods A complete description of the trial design, procedures, and outcomes has been published previously. Briefly, children aged 6 to 59 months who presented with SAM in Madarounfa, Niger, between October 2012 and November 2013 were randomly assigned 1:1 to receive amoxicillin (80 mg/kg/d) or a placebo for 7 days. All children received standard care for uncomplicated SAM for a minimum of 3 weeks and a maximum of 8 weeks. Children were followed up weekly for anthropometric, clinical, and vital status during treatment and (as per trial protocol) at 4, 8, and 12 weeks after admission. The Comité Consultatif National d’Ethique, Niger, and the Comité de Protection des Personnes, Île-de-France, France, provided ethical approval. An independent data safety monitoring board reviewed study progress and safety events, and all participants provided written informed consent.

8, and 12 weeks after admission. The Comité Consultatif National d’Ethique, Niger, and the Comité de Protection des Personnes, Île-de-France, France, provided ethical approval. An independent data safety monitoring board reviewed study progress and safety events, and all participants provided written informed consent. We used the weighted Kaplan-Meier method and Cox proportional hazard models to assess the effect of routine amoxicillin vs placebo on sustained nutritional recovery, transfer to inpatient care, and death from admission to 12 weeks. Inverse probability weights were used to account for censoring at the time of death or transfer to inpatient care. The intervention outcomes on total anthropometric gains among children who had recovered from admission to 12 weeks were assessed using t tests (of weight) and linear regression adjusted for baseline measurements (of mid–upper arm circumference and height). Anthropometric gains over time were estimated by intervention group using hierarchical generalized linear models with a cubic spline. Data analysis took place from December 2017 to June 2018. Analyses were performed using R, version 3.3 (R Foundation for Statistical Computing). Two-sided P values were considered significant at less than .05.

We used the weighted Kaplan-Meier method and Cox proportional hazard models to assess the effect of routine amoxicillin vs placebo on sustained nutritional recovery, transfer to inpatient care, and death from admission to 12 weeks. Inverse probability weights were used to account for censoring at the time of death or transfer to inpatient care. The intervention outcomes on total anthropometric gains among children who had recovered from admission to 12 weeks were assessed using t tests (of weight) and linear regression adjusted for baseline measurements (of mid–upper arm circumference and height). Anthropometric gains over time were estimated by intervention group using hierarchical generalized linear models with a cubic spline. Data analysis took place from December 2017 to June 2018. Analyses were performed using R, version 3.3 (R Foundation for Statistical Computing). Two-sided P values were considered significant at less than .05. Results All 2399 children (mean [SD] age, 16.7 [8.6] months; 1196 female children [49.9%]) of the primary analysis were eligible for inclusion in this extended analysis. Analysis found no association of routine amoxicillin administration with the risk of nutritional recovery, transfer to inpatient care or death, total weight, mid–upper arm circumference, or height gain from admission to 12 weeks (Table). The nutritional and anthropometric benefits of amoxicillin reported previously, may have been limited to the first 2 to 4 weeks after admission to the nutritional program and not maintained thereafter, including a decreased risk of transfer to inpatient care (cumulative incidence 0-<2 weeks postadmission: amoxicillin group, 8%; placebo group, 9%; 2-12 weeks postadmission: amoxicillin group, 36%; placebo group, 27%) and improved mean (SD) weight gain (during treatment: amoxicillin group, 6.47 [2.65] g/kg/day; placebo group, 5.85 [2.85] g/kg/ day; after program discharge: amoxicillin group, 1.01 [1.12] g/kg/day; placebo group, 1.06 [1.14] g/kg/day) (Figure).

9%; 2-12 weeks postadmission: amoxicillin group, 36%; placebo group, 27%) and improved mean (SD) weight gain (during treatment: amoxicillin group, 6.47 [2.65] g/kg/day; placebo group, 5.85 [2.85] g/kg/ day; after program discharge: amoxicillin group, 1.01 [1.12] g/kg/day; placebo group, 1.06 [1.14] g/kg/day) (Figure). Table. Clinical Outcomes and Anthropometric Gains From Admission to 12 Weeks by Intervention Group Clinical Outcome No. of Events per Person-Year Hazard Ratio or Mean Difference (95% CI) P Value Amoxicillin Placebo Sustained nutritional recovery 3.26 3.20 0.95 (0.86-1.05)a .36 Transfer to inpatient care 1.72 2.05 0.97 (0.84-1.13)a .70 Death 0.10 0.08 1.11 (0.58-2.13)a .75 Anthropometric gains, mean (SD) Weight, g/kg/d 2.82 (1.05) 2.75 (1.02) 0.07 (−0.04 to 0.18)b .21 Mid–upper arm circumference, mm/d 0.18 (0.08) 0.17 (0.07) 0.00 (0.00-0.01)b .22 Height, mm/d 0.19 (0.13) 0.19 (0.13) 0.01 (−0.01 to 0.02)b .41 a Hazard ratios and 95% CIs for amoxicillin relative to placebo are based on the univariate weighted Cox proportional hazard model. b The mean difference and 95% CIs of anthropometric gains were calculated from unweighted linear regression among children who remained recovered at their final study visit.

Table. Clinical Outcomes and Anthropometric Gains From Admission to 12 Weeks by Intervention Group Clinical Outcome No. of Events per Person-Year Hazard Ratio or Mean Difference (95% CI) P Value Amoxicillin Placebo Sustained nutritional recovery 3.26 3.20 0.95 (0.86-1.05)a .36 Transfer to inpatient care 1.72 2.05 0.97 (0.84-1.13)a .70 Death 0.10 0.08 1.11 (0.58-2.13)a .75 Anthropometric gains, mean (SD) Weight, g/kg/d 2.82 (1.05) 2.75 (1.02) 0.07 (−0.04 to 0.18)b .21 Mid–upper arm circumference, mm/d 0.18 (0.08) 0.17 (0.07) 0.00 (0.00-0.01)b .22 Height, mm/d 0.19 (0.13) 0.19 (0.13) 0.01 (−0.01 to 0.02)b .41 a Hazard ratios and 95% CIs for amoxicillin relative to placebo are based on the univariate weighted Cox proportional hazard model. b The mean difference and 95% CIs of anthropometric gains were calculated from unweighted linear regression among children who remained recovered at their final study visit. Figure. Risk of Transfer to Inpatient Care and Weight Gain During and After Program Discharge A, Weighted Kaplan-Meier estimates of survival (1 − risk) of transfer to inpatient care by intervention group during and after program discharge. Shaded areas are log-transformed 95% CIs. B, Weight gains during and after program discharge by intervention group, estimated using a hierarchical generalized linear model with a cubic spline, including a knot at the time of program discharge. Shaded areas are 95% CIs from 10 000 bootstraps.

nd after program discharge. Shaded areas are log-transformed 95% CIs. B, Weight gains during and after program discharge by intervention group, estimated using a hierarchical generalized linear model with a cubic spline, including a knot at the time of program discharge. Shaded areas are 95% CIs from 10 000 bootstraps. Discussion Results from this clinical trial with extended follow-up from admission to 12 weeks suggest no longer-term benefit of routine antibiotic use in the treatment of uncomplicated SAM. Current guidelines rightly acknowledge that it would be inappropriate to withhold an intervention that may substantially reduce mortality in a high-risk population, and in settings where it can save lives, routine antibiotic use should remain part of clinical protocols for uncomplicated SAM. This use, however, should be weighed against the risk of the emergence of antibiotic resistance, implications for program costs and coverage, and likely short-lived individual benefits. Guidance that allows treatment protocols to be adapted and simplified in specific contexts while maintaining individual effectiveness, protecting public health safety, and assuring access to care should be prioritized.