Browse the corpus

ations differed substantially between different populations and, within specific populations, between different studies. For intracerebral haemorrhage, the findings have been conflicting, due partly to the relatively small number of events recorded and lack of neuroimaging for diagnosis of stroke types in many studies. Added value of this study This study represents the world's largest investigation to date in a single study of the association between adiposity and risk of incident stroke types. In this relative lean (mean BMI 23·6 kg/m2 [SD 3·2]) and apparently healthy Chinese population, there was a strong positive and log-linear association between general (eg, BMI and body fat percentage) and central (eg, waist circumference and waist-to-hip ratio) adiposity and the incidence of ischaemic stroke, with the risk estimates consistent with those predicted by equivalent differences in systolic blood pressure associated with adiposity. By contrast, the risk estimates for intracerebral haemorrhage were less extreme than for ischaemic stroke, and much weaker than predicted by the corresponding difference in systolic blood pressure, especially at a BMI of less than 25 kg/m2, which covers two-thirds of study population. The risk estimates were higher in men than in women, due mainly to stronger associations in men of systolic blood pressure with stroke. After adjustment for usual systolic blood pressure, the positive association of baseline BMI (and other adiposity measures) with ischaemic stroke largely disappeared, whereas for intracerebral haemorrhage it was completely reversed. BMI at the age of 25 years was weakly positively associated with ischaemic stroke and intracerebral haemorrhage, and the association was largely unaffected by adjustment for systolic blood pressure.

ty measures) with ischaemic stroke largely disappeared, whereas for intracerebral haemorrhage it was completely reversed. BMI at the age of 25 years was weakly positively associated with ischaemic stroke and intracerebral haemorrhage, and the association was largely unaffected by adjustment for systolic blood pressure. Implications of all the available evidence

Introduction Stroke is a leading cause of premature death and disability worldwide, with more than 80% of the global burden of stroke now occurring in low-income and middle-income countries.1 Stroke affects about 2 million people each year in China, and about 20% of cases are due to intracerebral haemorrhage.2 The prevalence of overweight and obesity is increasing globally, including in China,3, 4 and high levels of adiposity are associated with raised blood pressure, raised blood glucose, and dyslipidaemia, each of which predisposes to stroke and ischaemic heart disease.5, 6, 7, 8, 9 About a fifth of strokes in China has been estimated to be attributable to high adiposity (defined as body-mass index [BMI] >23 kg/m2), with no separate estimates available for pathological types of stroke.10 Such estimates were based mainly on risk estimates derived from studies done in high-income countries, where, by comparison with China, the mean BMI is higher, the rates of stroke, especially intracerebral haemorrhage, are lower, and individuals identified with hypertension are generally well managed. In China, although several prospective studies have previously assessed the associations between adiposity and stroke mortality, most have predated widespread use of brain imaging for reliable diagnosis of stroke types,5, 8 which might have important differences in their associations with adiposity.5, 11

well managed. In China, although several prospective studies have previously assessed the associations between adiposity and stroke mortality, most have predated widespread use of brain imaging for reliable diagnosis of stroke types,5, 8 which might have important differences in their associations with adiposity.5, 11 Prospective studies5, 6, 8, 11, 12, 13, 14, 15, 16 have consistently reported positive associations between ischaemic and total stroke and BMI. However, the strength of the associations differs substantially between different populations and, within specific populations, between different studies.17 For intracerebral haemorrhage, the previous findings have been conflicting, with some reporting linear positive associations,8, 11, 13, 18, 19, 20 and others showing J-shaped or U-shaped,5, 8, 21 null,22, 23 or linear inverse associations.17 Few studies have assessed the relevance of central (eg, waist circumference) or other (eg, body fat percentage) measures of adiposity with stroke types,6, 24, 25 or the extent to which such effects are mediated by blood pressure, diabetes, or blood lipids.9, 16 Using data from the large nationwide China Kadoorie Biobank (CKB) prospective study, we assess the associations between general and central adiposity and incidence of stroke types and effect mediation by blood pressure in Chinese men and women. Research in context Evidence before this study

Prospective studies5, 6, 8, 11, 12, 13, 14, 15, 16 have consistently reported positive associations between ischaemic and total stroke and BMI. However, the strength of the associations differs substantially between different populations and, within specific populations, between different studies.17 For intracerebral haemorrhage, the previous findings have been conflicting, with some reporting linear positive associations,8, 11, 13, 18, 19, 20 and others showing J-shaped or U-shaped,5, 8, 21 null,22, 23 or linear inverse associations.17 Few studies have assessed the relevance of central (eg, waist circumference) or other (eg, body fat percentage) measures of adiposity with stroke types,6, 24, 25 or the extent to which such effects are mediated by blood pressure, diabetes, or blood lipids.9, 16 Using data from the large nationwide China Kadoorie Biobank (CKB) prospective study, we assess the associations between general and central adiposity and incidence of stroke types and effect mediation by blood pressure in Chinese men and women. Research in context Evidence before this study Embase and MEDLINE databases were searched for articles published in English before March 1, 2017, using a combination of terms: “stroke (ischaemic stroke [IS] OR haemorrhagic stroke OR intracerebral haemorrhage [ICH] OR stroke types) AND adiposity (body mass index OR BMI OR waist circumference OR waist hip ratio OR adiposity OR obesity) AND cohort (cohort study OR prospective study OR longitudinal study)”. The reference lists of relevant articles were reviewed. About 20 prospective cohort studies examining the association between adiposity and risk of stroke types were identified, including five in China, along with a few meta-analyses of cohort studies. With a few exceptions, most of the studies only included body-mass index (BMI) and focused on mortality rather than incidence of stroke, with little information on diagnosis of stroke types. Moreover, none examined the quantitative association between BMI and risk of stroke types in the context of the associations between BMI and blood pressure, and between blood pressure and stroke types. In general, these studies have shown positive associations between ischaemic stroke and total stroke and BMI, but the strength of the associations differed substantially between different populations and, within specific populations, between different studies. For intracerebral haemorrhage, the findings have been conflicting, due partly to the relatively small number of events recorded and lack of neuroimaging for diagnosis of stroke types in many studies.

ations differed substantially between different populations and, within specific populations, between different studies. For intracerebral haemorrhage, the findings have been conflicting, due partly to the relatively small number of events recorded and lack of neuroimaging for diagnosis of stroke types in many studies. Added value of this study

ty measures) with ischaemic stroke largely disappeared, whereas for intracerebral haemorrhage it was completely reversed. BMI at the age of 25 years was weakly positively associated with ischaemic stroke and intracerebral haemorrhage, and the association was largely unaffected by adjustment for systolic blood pressure. Implications of all the available evidence Globally, the mean levels of adiposity have been increasing steadily in recent decades. In this Chinese population, high adiposity (BMI >23 kg/m2) accounted for 15% of total stroke (17% ischaemic stroke and 7% intracerebral haemorrhage). The positive association between adiposity and ischaemic stroke is chiefly mediated through its effect on blood pressure. However, there was no association between adiposity and intracerebral haemorrhage across a broad normal range (ie, BMI <25 kg/m2), suggesting that leanness, either per se or through some other factor (or factors), might increase risk, thereby offsetting the protective effects of lower blood pressure. Future Mendelian randomisation studies are needed to establish the causal relevance of adiposity for intracerebral haemorrhage. Since ischaemic stroke constitutes the majority of total stroke cases in China and elsewhere, the findings on intracerebral haemorrhage should not diminish the fundamental importance of high adiposity as a major modifiable determinant of overall stroke. Given the substantial risks associated with high blood pressure, the present study further highlights the importance of controlling blood pressure (and other intermediate risk factors) for prevention of both ischaemic and haemorrhagic stroke, irrespective of levels of adiposity.

odifiable determinant of overall stroke. Given the substantial risks associated with high blood pressure, the present study further highlights the importance of controlling blood pressure (and other intermediate risk factors) for prevention of both ischaemic and haemorrhagic stroke, irrespective of levels of adiposity. Methods Study population Details of the design, methods, and participants in the CKB have been described previously.26, 27, 28 Briefly, the baseline survey took place between June 25, 2004, and July 15, 2008, involving ten (five urban and five rural) areas across China, chosen from China's nationally representative Disease Surveillance Points to maximise geographical and social diversity. About 1·8 million registered residents aged 35–74 years in the 100–150 administrative units (rural villages or urban residential committees) within each of the ten study areas were identified through residential records and invited to participate in the study. 512 891 people, including 12 668 just outside this age range (extending the baseline age range to 30–79 years), were enrolled in the study. Prior ethics approval was obtained from the relevant international, national, and local ethics committees. All participants provided written informed consent.

icipate in the study. 512 891 people, including 12 668 just outside this age range (extending the baseline age range to 30–79 years), were enrolled in the study. Prior ethics approval was obtained from the relevant international, national, and local ethics committees. All participants provided written informed consent. Data collection Information was collected on sociodemographic factors, lifestyle (eg, smoking, alcohol consumption, diet, and physical activity), and medical history by trained health workers using interviewer-administered laptop-based questionnaires. Physical measurements (eg, blood pressure, lung function, and anthropometric measures) were recorded using standard methods. A non-fasting venous blood sample was collected for storage and on-site random plasma glucose testing using the SureStep Plus system (LifeScan; Milpitas, CA, USA). Two periodic resurveys were done from May 26, to Oct 10, 2008, and from Aug 4, 2013, to Sept 18, 2014, on approximately 5% of randomly selected participants, collecting similar information as at baseline to enable assessment of within-person variation and temporal trends in exposures.

tem (LifeScan; Milpitas, CA, USA). Two periodic resurveys were done from May 26, to Oct 10, 2008, and from Aug 4, 2013, to Sept 18, 2014, on approximately 5% of randomly selected participants, collecting similar information as at baseline to enable assessment of within-person variation and temporal trends in exposures. Anthropometric measurements were recorded once, and usually to the nearest 0·1 cm or 0·1 kg.27, 28 Standing and sitting heights were measured using a stadiometer, and weight was measured by a body composition analyser (TANITA-TBF-300 GS; Tanita Corporation, Tokyo, Japan), with subtraction of weight of clothing by 0·5–1·5 kg depending on the season. Body fat percentage was the proportion of total weight due to fat weight estimated by the body composition analyser. Waist circumference was measured, using a soft non-stretchable tape, midway between the lowest rib and the iliac crest. Hip circumference was measured at the maximum circumference around the buttocks (usually over underclothing, but subtracting 1 cm if over a skirt, or 2·5 cm if over trousers). BMI was calculated as measured weight in kilograms divided by the square of the standing height in metres (kg/m2), BMI at age 25 years used self-reported weight at age 25 years and height measured at baseline, and waist-to-hip ratio was the ratio of waist circumference to hip circumference.

or 2·5 cm if over trousers). BMI was calculated as measured weight in kilograms divided by the square of the standing height in metres (kg/m2), BMI at age 25 years used self-reported weight at age 25 years and height measured at baseline, and waist-to-hip ratio was the ratio of waist circumference to hip circumference. Blood pressure was measured twice using the UA-779 digital sphygmomanometer (A&D Instruments; Abingdon, UK) after participants had remained at rest in the seated position for at least 5 min. If the difference between the two measurements was more than 10 mm Hg for systolic blood pressure (in about 5% of participants), a third measurement was made and the mean of the last two measurements was used.

Instruments; Abingdon, UK) after participants had remained at rest in the seated position for at least 5 min. If the difference between the two measurements was more than 10 mm Hg for systolic blood pressure (in about 5% of participants), a third measurement was made and the mean of the last two measurements was used. Follow-up for morbidity and mortality The vital status of participants was obtained periodically from local death registries,29 and checked annually against residential records and health insurance data and by active confirmation through local residential administrators. Causes of death were sought chiefly from official death certificates. Additional information on hospitalised events was collected through electronic linkage, via unique personal identification numbers, to disease registries (for stroke, ischaemic heart disease, cancer, and diabetes) and to the national health insurance claim system, which has almost universal coverage and included disease descriptions, International Classification of Diseases 10th revision (ICD-10) codes, and procedure codes.27 All reported fatal or non-fatal stroke (and other major diseases) events from different sources were reviewed and integrated centrally by trained clinical staff, who were masked to baseline information.

d disease descriptions, International Classification of Diseases 10th revision (ICD-10) codes, and procedure codes.27 All reported fatal or non-fatal stroke (and other major diseases) events from different sources were reviewed and integrated centrally by trained clinical staff, who were masked to baseline information. The main stroke types examined were ischaemic stroke (ICD-10 code I63), intracerebral haemorrhage (I61), subarachnoid haemorrhage (I60), and total stroke (I60, I61, I63, and I64). All analyses were restricted to first events of any stroke type occurring between the ages of 35 years and 79 years, with censoring when participants had a stroke or reached 80 years of age (or loss to follow-up). Statistical analyses Individuals with previous history of stroke or transient ischaemic attack (n=8884) or ischaemic heart disease (n=15 472) at baseline, and those with missing, implausible, or extreme values of anthropometric measurements (n=493) were excluded, leaving 489 301 for our report. Individuals with hypertension or diabetes were retained (and not adjusted for in standard analyses), as these could be mechanisms whereby adiposity predisposes to stroke.

aseline, and those with missing, implausible, or extreme values of anthropometric measurements (n=493) were excluded, leaving 489 301 for our report. Individuals with hypertension or diabetes were retained (and not adjusted for in standard analyses), as these could be mechanisms whereby adiposity predisposes to stroke. The prevalence or mean values of baseline characteristics were calculated across seven BMI categories (cut points <18·0, <20·5, <23·0, <25·0, <27·5, and <30·0 kg/m2), standardised by 5-year age groups, sex, and study area. The cut points were chosen to cover different thresholds for defining overweight and obesity in Chinese people and to ensure reasonable participant numbers in each group. Cox proportional hazards models with time since baseline as the timescale were used to estimate hazard ratios (HRs) for incident stroke by adiposity and systolic blood pressure, stratified by 5-year age-at-risk, sex, and area (ten groups), and adjusted for education (five groups), smoking (four groups), alcohol use (five groups), physical activity (metabolic equivalent of task [MET] hours per day), and self-reported health status (four groups). Floating variance estimates (reflecting independent variability within each group, including the reference group) were used to estimate group-specific 95% CIs for each log HR to enable comparisons between any two categories (rather than just pairwise comparisons with the reference category).30

lth status (four groups). Floating variance estimates (reflecting independent variability within each group, including the reference group) were used to estimate group-specific 95% CIs for each log HR to enable comparisons between any two categories (rather than just pairwise comparisons with the reference category).30 To correct for regression dilution bias,31 two serial measurements recorded in a random subset of approximately 20 000 participants at resurveys were used to calculate regression dilution ratios (ie, the ratio of the uncorrected to the corrected regression coefficient). Log HR estimates associated with baseline measures were divided by the regression dilution ratio (0·61 for systolic blood pressure and 0·93 for BMI) to obtain HRs for usual levels of exposure measures. Departure from linearity was assessed using the likelihood ratio test, with inclusion of a second-degree term in the model.32 The proportional hazards assumption was tested by examining the HRs for the first 4 years and for subsequent years of follow-up (and showed no strong evidence of departure). HRs were also compared across strata of other baseline factors (eg, age, sex, and area), and χ2 values for trend and heterogeneity were calculated. Additional sensitivity analyses, presented mainly in the appendix, were done after excluding individuals with poor self-rated health status, any self-reported previous disease at baseline, ever smokers, or stroke events occurring during the first 3-years of follow-up. All analyses were done with SAS version 9.3.

lated. Additional sensitivity analyses, presented mainly in the appendix, were done after excluding individuals with poor self-rated health status, any self-reported previous disease at baseline, ever smokers, or stroke events occurring during the first 3-years of follow-up. All analyses were done with SAS version 9.3. Role of the funding source The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. ZC, AI, and LL had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results Among the 489 301 participants included, the mean age was 51·1 years (SD 10·4), and the mean BMI was 23·6 kg/m2 (3·2), with 331 723 (67·8%) having a BMI of less than 25 kg/m2 and 18 862 (3·9%) having a BMI of 30 kg/m2 or more (ie, obese). Overall mean BMI was slightly higher in women than in men (23·7 kg/m2 [3·4] vs 23·4 kg/m2 [3·2]; p<0·0001), and higher in urban than in rural areas (24·2 kg/m2 [3·4] vs 23·1 kg/m2 [3·3]; p<0·0001). BMI was weakly positively associated with education (p=0·0008) and alcohol intake (p=0·0009), especially at a BMI of less than 25 kg/m2, and inversely associated with smoking in both men and women. Individuals with a BMI of less than 18 kg/m2 or more than 25 kg/m2 tended to have lower levels of physical activity and poor self-rated health status than did those with a normal BMI (table 1).Table 1 Baseline characteristics of participants by baseline body-mass index among people without previous stroke or ischaemic heart disease

ith a BMI of less than 18 kg/m2 or more than 25 kg/m2 tended to have lower levels of physical activity and poor self-rated health status than did those with a normal BMI (table 1).Table 1 Baseline characteristics of participants by baseline body-mass index among people without previous stroke or ischaemic heart disease Categories of baseline body-mass index (kg/m2) All <18·0 18·0 to <20·5 20·5 to <23·0 23·0 to <25·0 25·0 to <27·5 27·5 to <30·0 ≥30·0 Number of participants 13 527 72 411 135 616 110 169 95 573 43 143 18 862 489 301 Body-mass index 17·1 (0·9) 19·5 (0·8) 21·8 (0·7) 23·9 (0·6) 26·1 (0·7) 28·5 (0·8) 31·7 (2·0) 23·6 (3·2) Age and socioeconomic factors Age (years) 55·4 (15·1) 51·2 (11.9) 50·4 (10·6) 50·7 (10·1) 51·0 (10·1) 51·1 (10·5) 51·0 (12·2) 51·1 (10·4) Male 37·8% 45·3% 42·1% 40·2 40·0% 37·5% 29·1% 40·9% Female 62·2% 54·7% 57·9% 59·8% 60·0% 62·5% 70·9% 59·1% Urban residents 31·7% 32·0% 38·4% 45·5% 50·5% 53·9% 57·2% 43·2% Rural residents 68·3% 68·0% 61·6% 54·5% 49·5% 46·1% 42·8% 56·8% ≥6 years of education 48·1% 48·8% 49·4% 50·2% 49·9% 48·9% 47·6% 49·2% Lifestyle factors Male ever-regular smoker 79·7% 79·7% 75·9% 72·5% 70·9% 70·5% 70·7% 74·4% Female ever-regular smoker 5·4% 3·7% 3·1% 2·9% 2·7% 2·8% 3·2% 3·1% Male ever-regular alcohol drinker 31·8% 36·1% 37·8% 37·6% 37·0% 37·7% 36·2% 37·2% Female ever-regular alcohol drinker 2·4% 2·3% 2·5% 2·5% 2·6% 2·6% 2·2% 2·5% Physical activity (MET h/day) 20·7 (17·1) 22·1 (13·0) 22·1 (11·9) 21·6 (11·8) 21·0 (12·2) 20·4 (13·0) 19·5 (15·1) 21·5 (11·8) Medical history and health status Diabetes 3·6% 3·2% 4·2% 5·5% 6·8% 8·1% 10·4% 5·4% Hypertension* 2·8% 4·3% 6·7% 10·1% 14·0% 18·1% 23·9% 9·9% Self-rated poor health 19·1% 11·5% 9·0% 8·2% 8·4% 9·4% 11·6% 9·4% Anthropometry and blood pressure Weight (kg) 43·2 (5·3) 49·1 (4·4) 54·9 (4·3) 60·4 (4·4) 65·9 (5·0) 72·1 (5·9) 80·3 (9·0) 59·1 (9·1) Standing height (cm) 158·6 (8·4) 158·6 (6·3) 158·6 (5·6) 158·7 (5·4) 158·9 (5·6) 158·9 (6·1) 159·0 (7·2) 158·7 (5·5) Sitting height (cm) 84·4 (4·8) 84·7 (3·7) 85·0 (3·3) 85·4 (3·1) 85·7 (3·2) 86·0 (3·5) 86·3 (4·0) 85·3 (3·2) Waist circumference (cm) 64·8 (6·4) 69·9 (5·5) 75·6 (5·1) 81·1 (5·0) 86·3 (5·4) 91·8 (6·2) 98·8 (8·5) 80·0 (9·0) Waist-to-hip ratio 0·80 (0·07) 0·83 (0·06) 0·86 (0·05) 0·89 (0·05) 0·91 (0·06) 0·94 (0·06) 0·96 (0·07) 0·88 (0·06) Body fat percentage 15·8 (4·0) 20·3 (3·2) 24·8 (3·3) 28·8 (3·5) 32·4 (4·1) 36·0 (4·9) 40·0 (6·8) 27·9 (6·4) Systolic blood pressure (mm Hg) 119·8 (24·3) 123·8 (21·3) 127·7 (19·1) 131·2 (18·9) 134·8 (19·

0) Waist-to-hip ratio 0·80 (0·07) 0·83 (0·06) 0·86 (0·05) 0·89 (0·05) 0·91 (0·06) 0·94 (0·06) 0·96 (0·07) 0·88 (0·06) Body fat percentage 15·8 (4·0) 20·3 (3·2) 24·8 (3·3) 28·8 (3·5) 32·4 (4·1) 36·0 (4·9) 40·0 (6·8) 27·9 (6·4) Systolic blood pressure (mm Hg) 119·8 (24·3) 123·8 (21·3) 127·7 (19·1) 131·2 (18·9) 134·8 (19· 8) 138·4 (22·0) 142·7 (26·4) 130·6 (19·3) Diastolic blood pressure (mm Hg) 72·7 (14·4) 74·0 (11·7) 75·8 (10·6) 78·0 (10·5) 80·1 (11·1) 82·0 (12·3) 83·9 (14·8) 77·7 (10·7) Random plasma glucose concentration (mmol/L) 5·90 (3·88) 5·81 (2·56) 5·89 (2·29) 6·02 (2·27) 6·18 (2·46) 6·33 (2·71) 6·61 (3·79) 6·03 (2·23) Lipid concentration Data available 653 2929 4850 3731 3491 1712 773 18 139 LDL cholesterol (mmol/L) 2·05 (0·82) 2·20 (0·76) 2·30 (0·68) 2·44 (0·68) 2·49 (0·69) 2·54 (0·78) 2·59 (1·08) 2·37 (0·67) HDL cholesterol (mmol/L) 1·42 (0·43) 1·35 (0·37) 1·28 (0·31) 1·21 (0·27) 1·15 (0·27) 1·12 (0·27) 1·11 (0·30) 1·24 (0·29) Total cholesterol (mmol/L) 4·32 (1·15) 4·46 (1·01) 4·57 (0·95) 4·74 (0·95) 4·81 (0·97) 4·88 (1·08) 4·95 (1·53) 4·66 (0·94) Triglycerides (mmol/L) 1·19 (0·77) 1·44 (1·40) 1·72 (1·28) 2·09 (1·73) 2·45 (1·93) 2·71 (2·18) 2·84 (2·32) 2·00 (1·57) Baseline characteristics adjusted for age, sex, and area as appropriate. Data are mean (SD) and weighted percentages. MET h/day=metabolic equivalent of task hours per day. * When combined self-reported hypertension with measured high blood pressure (ie, ≥140/90 mm Hg), 159 377 (32·6%) participants had hypertension at baseline.

8) 138·4 (22·0) 142·7 (26·4) 130·6 (19·3) Diastolic blood pressure (mm Hg) 72·7 (14·4) 74·0 (11·7) 75·8 (10·6) 78·0 (10·5) 80·1 (11·1) 82·0 (12·3) 83·9 (14·8) 77·7 (10·7) Random plasma glucose concentration (mmol/L) 5·90 (3·88) 5·81 (2·56) 5·89 (2·29) 6·02 (2·27) 6·18 (2·46) 6·33 (2·71) 6·61 (3·79) 6·03 (2·23) Lipid concentration Data available 653 2929 4850 3731 3491 1712 773 18 139 LDL cholesterol (mmol/L) 2·05 (0·82) 2·20 (0·76) 2·30 (0·68) 2·44 (0·68) 2·49 (0·69) 2·54 (0·78) 2·59 (1·08) 2·37 (0·67) HDL cholesterol (mmol/L) 1·42 (0·43) 1·35 (0·37) 1·28 (0·31) 1·21 (0·27) 1·15 (0·27) 1·12 (0·27) 1·11 (0·30) 1·24 (0·29) Total cholesterol (mmol/L) 4·32 (1·15) 4·46 (1·01) 4·57 (0·95) 4·74 (0·95) 4·81 (0·97) 4·88 (1·08) 4·95 (1·53) 4·66 (0·94) Triglycerides (mmol/L) 1·19 (0·77) 1·44 (1·40) 1·72 (1·28) 2·09 (1·73) 2·45 (1·93) 2·71 (2·18) 2·84 (2·32) 2·00 (1·57) Baseline characteristics adjusted for age, sex, and area as appropriate. Data are mean (SD) and weighted percentages. MET h/day=metabolic equivalent of task hours per day. * When combined self-reported hypertension with measured high blood pressure (ie, ≥140/90 mm Hg), 159 377 (32·6%) participants had hypertension at baseline. BMI was strongly positively associated with weight (p<0·0001), and weakly positively associated with sitting and standing height (p=0·0006; table 1; appendix p 3). BMI, waist circumference, waist-to-hip ratio, and body fat percentage were highly intercorrelated with each other, and BMI at the age of 25 years was modestly correlated with adulthood BMI and less so with other adiposity measures (appendix p 3). BMI, and other measures of adiposity,28 were strongly positively associated with blood pressure and prevalence of physician-diagnosed hypertension and diabetes (table 1). BMI was also positively correlated with plasma glucose concentration. In a subset of 18 139 participants not receiving lipid-lowering treatment, BMI was positively correlated with concentrations of LDL cholesterol, total cholesterol, and triglycerides, and inversely correlated with HDL cholesterol (table 1).

(table 1). BMI was also positively correlated with plasma glucose concentration. In a subset of 18 139 participants not receiving lipid-lowering treatment, BMI was positively correlated with concentrations of LDL cholesterol, total cholesterol, and triglycerides, and inversely correlated with HDL cholesterol (table 1). Incident strokes from enrolment until Jan 1, 2016 (a median of 9 years [IQR 8–10]), are included in these analyses, by which time 37 289 (7·3%) participants had died and 4875 (1·0%) were lost to follow-up. During 3·85 million person-years of follow-up, there were 32 448 incident cases of stroke (about 90% confirmed by neuroimaging; 3493 [10·8%] fatal) at ages 35–79 years, including 25 210 (77·7%) due to ischaemic stroke (mean age at event 65·1 years [SD 9·6]; 648 [2·6%] fatal), 5380 (16·6%) due to intracerebral haemorrhage (mean age at event 64·6 years [10·4]; 2495 [46·4%] fatal), 434 (1·3%) due to subarachnoid haemorrhage (mean age at event 61·4 years [10·7]; 43 [9·9%] fatal), and 1424 (4·4%) unspecified strokes (mean age at event 65·0 years [10·1]; 307 [21·6%] fatal). The standardised total stroke incidence per 10 000 person-years was 84·3, and was higher in urban than rural areas (94·6 vs 76·8). The incidence of ischaemic stroke per 10 000 person-years was higher in urban than in rural areas (82·2 vs 53·3), but the converse was true for intracerebral haemorrhage (7·5 vs 18·7).

al). The standardised total stroke incidence per 10 000 person-years was 84·3, and was higher in urban than rural areas (94·6 vs 76·8). The incidence of ischaemic stroke per 10 000 person-years was higher in urban than in rural areas (82·2 vs 53·3), but the converse was true for intracerebral haemorrhage (7·5 vs 18·7). Throughout the distribution, each 5 kg/m2 higher BMI was associated with 8·3 mm Hg (SE 0·04) higher systolic blood pressure (and 4·3 mm Hg [0·02] higher diastolic blood pressure; figure 1). Blood pressure was strongly positively associated with risk of stroke, especially intracerebral haemorrhage, with no evidence of any threshold throughout the range studied (figure 1). After adjusting for regression dilution bias, each 10 mm Hg higher usual systolic blood pressure was associated with an adjusted HR of 1·31 (95% CI 1·30–1·32) for ischaemic stroke and 1·61 (1·58–1·63) for intracerebral haemorrhage, with greater HRs at younger than older ages and in men than women (appendix p 4). Corresponding adjusted HRs per 10 mm Hg higher systolic blood pressure were 1·21 (1·13–1·30) for subarachnoid haemorrhage, 1·38 (1·33–1·43) for unspecified stroke, and 1·36 (1·36–1·37) for total stroke. Given the linear associations in figure 1, it can be expected that a 5 kg/m2 higher BMI, corresponding to 8·3 mm Hg higher usual systolic blood pressure, would be associated with HRs of 1·25 (1·24–1·26) for ischaemic stroke, 1·48 (1·46–1·50) for intracerebral haemorrhage, and 1·29 (1·29–1·30) for total stroke.Figure 1 Associations between baseline BMI and SBP and between usual SBP and incidence of stroke types

esponding to 8·3 mm Hg higher usual systolic blood pressure, would be associated with HRs of 1·25 (1·24–1·26) for ischaemic stroke, 1·48 (1·46–1·50) for intracerebral haemorrhage, and 1·29 (1·29–1·30) for total stroke.Figure 1 Associations between baseline BMI and SBP and between usual SBP and incidence of stroke types (A) BMI versus SBP. Means are adjusted for age, sex, area, education, smoking, alcohol consumption, physical activity, and self-rated health status. The area of the squares is inversely proportional to the variance of the mean SBP. The vertical lines indicate 95% CIs. (B) SBP versus ischaemic stroke. (C) SBP versus intracerebral haemorrhage. HRs are stratified by age, sex, and area, and adjusted simultaneously for education, smoking, alcohol consumption, physical activity, and self-rated health status. Each closed square represents HR with the area inversely proportional to the variance of the log HR. The vertical lines indicate 95% CIs. BMI=body-mass index. SBP=systolic blood pressure. HR=hazard ratio.

justed simultaneously for education, smoking, alcohol consumption, physical activity, and self-rated health status. Each closed square represents HR with the area inversely proportional to the variance of the log HR. The vertical lines indicate 95% CIs. BMI=body-mass index. SBP=systolic blood pressure. HR=hazard ratio. There was a log-linear positive association between BMI and incidence of ischaemic stroke throughout the BMI range examined (mean 17·1 kg/m2 [SD 0·9] to 31·7 kg/m2 [2·0]), with each 5 kg/m2 higher baseline BMI associated with an adjusted HR of 1·30 (95% CI 1·28–1·33; table 2; figure 2), consistent with the expected associations with systolic blood pressure in figure 1. By contrast, each 5 kg/m2 higher BMI was only associated with an adjusted HR of 1·11 (1·07–1·16) for intracerebral haemorrhage, an effect size that was weaker than that for ischaemic stroke, and substantially weaker than predicted from the corresponding difference in systolic blood pressure (ie, 1·48 [1·46–1·50] per 8·3 mm Hg systolic blood pressure). Moreover, the shape of the association with intracerebral haemorrhage appeared non-linear (p<0·0001 for test of non-linearity). Among the 68% of participants with a BMI of less than 25 kg/m2, there were no associations between BMI and risk of intracerebral haemorrhage, even though systolic/diastolic blood pressure differed by 11·5/5·3 mm Hg across the four BMI groups studied. Above a BMI of 25 kg/m2, the risk of intracerebral haemorrhage increased with increasing levels of BMI, but the association was still weaker than that predicted by the difference in systolic blood pressure. BMI was unrelated to subarachnoid haemorrhage, but the number of events (n=434) was small (table 2). For unspecified stroke, the association resembled that for ischaemic stroke (HR 1·28 [1·18–1·38] per 5 kg/m2), as was the case for total stroke (1·27 [1·25–1·29] per 5 kg/m2; table 2). After adjusting for regression dilution bias, the HRs became slightly stronger (1·33 [1·30–1·36] for ischaemic stroke, 1·12 [1·07–1·17] for intracerebral haemorrhage, and 1·29 [1·27–1·31] for total stroke, per 5 kg/m2 higher usual BMI).

5 kg/m2), as was the case for total stroke (1·27 [1·25–1·29] per 5 kg/m2; table 2). After adjusting for regression dilution bias, the HRs became slightly stronger (1·33 [1·30–1·36] for ischaemic stroke, 1·12 [1·07–1·17] for intracerebral haemorrhage, and 1·29 [1·27–1·31] for total stroke, per 5 kg/m2 higher usual BMI). In this population, high adiposity (defined as BMI >23 kg/m2) accounted for 14·7% of total stroke, including 16·5% of ischaemic stroke, and 6·7% of intracerebral haemorrhage, and overweight (ie, BMI ≥25 kg/m2) accounted for 9·2% of total stroke, including 9·9% of ischaemic stroke and 6·7% of intracerebral haemorrhage (appendix p 5).Figure 2 Adjusted HRs for ischaemic stroke (A) and intracerebral haemorrhage (B) by baseline BMI, body fat percentage, and waist circumference HRs are stratified by age, sex, and area, and adjusted simultaneously for education, smoking, alcohol consumption, physical activity, and self-rated health status. Each closed square represents HR with the area inversely proportional to the variance of the log HR. The vertical lines indicate 95% CIs. HR=hazard ratio. BMI=body-mass index. Table 2 Standardised stroke incidence rates per 10 000 person-years and adjusted HRs by baseline BMI among people without previous stroke or ischaemic heart disease

HRs are stratified by age, sex, and area, and adjusted simultaneously for education, smoking, alcohol consumption, physical activity, and self-rated health status. Each closed square represents HR with the area inversely proportional to the variance of the log HR. The vertical lines indicate 95% CIs. HR=hazard ratio. BMI=body-mass index. Table 2 Standardised stroke incidence rates per 10 000 person-years and adjusted HRs by baseline BMI among people without previous stroke or ischaemic heart disease Ischaemic stroke Intracerebral haemorrhage Subarachnoid haemorrhage Unspecified stroke All stroke Events Rate* HR (95% CI)† Events Rate* HR (95% CI)† Events Rate* HR (95% CI)† Events Rate* HR (95% CI)† Events Rate* HR (95% CI)† BMI<18·0 590 49·3 1·00 (0·92–1·09) 253 15·8 1·00 (0·88–1·13) 10 1·2 1·00 (0·53–1·87) 37 2·3 1·00 (0·72–1·39) 890 68·6 1·00 (0·94–1·07) BMI 18·0–20·4 2740 50·4 1·10 (1.05–1·14) 980 13·5 0·97 (0·91–1·03) 71 1·1 1·53 (1·20–1·94) 207 3·5 1·41 (1·23–1·63) 3998 68·5 1·08 (1·04–1·11) BMI 20·5–22·9 5790 59·6 1·33 (1·29–1·36) 1494 13·7 1·01 (0·96–1·06) 127 1·3 1·57 (1·32–1·87) 349 3·4 1·47 (1·32–1·63) 7760 78·0 1·24 (1·22–1·27) BMI 23·0–24·9 5600 67·5 1·51 (1·47–1·55) 1020 12·9 0·95 (0·90–1·02) 81 1·0 1·25 (1·01–1·56) 306 3·8 1·61 (1·44–1·80) 7007 85·2 1·37 (1·33–1·40) BMI 25·0–27·4 5946 78·6 1·75 (1·70–1·79) 960 14·6 1·11 (1·04–1·18) 89 1·2 1·57 (1·28–1·94) 289 4·0 1·75 (1·56–1·97) 7284 98·4 1·58 (1·54–1·62) BMI 27·5–29·9 3038 86·6 1·90 (1·84–1·97) 464 17·7 1·24 (1·13–1·36) 43 1·3 1·66 (1·22–2·25) 150 4·5 2·03 (1·73–2·39) 3695 110·1 1·73 (1·68–1·79) BMI≥30·0 1506 96·7 2·03 (1·92–2·13) 209 19·5 1·32 (1·15–1·51) 13 1·2 1·05 (0·60–1·82) 86 6·9 2·60 (2·09–3·22) 1814 124·3 1·85 (1·76–1·94) Overall 25 210 65·5 .. 5380 14·0 .. 434 1·1 .. 1424 3·7 .. 32 448 84·3 .. ptrend .. .. <0·0001 .. .. <0·0001‡ .. .. 0·9686 .. .. <0·0001 .. .. <0·0001 HRs per 5 kg/m2 Observed .. .. 1·30 (1·28–1·33) .. .. 1·11 (1·07–1·16) .. .. 0·96 (0·83–1·11) .. .. 1·28 (1·18–1·38) .. .. 1·27 (1·25–1·29) Expected¶ .. .. 1·25 (1·24–1·26) .. .. 1·48 (1·46–1·50) .. .. 1·17 (1·11–1·24) .. .. 1·31 (1·27–1·35) .. .. 1·29 (1·29–1·30) Body-mass index (BMI) unit is kg/m2. HR=hazard ratio. ..=not applicable.

bserved .. .. 1·30 (1·28–1·33) .. .. 1·11 (1·07–1·16) .. .. 0·96 (0·83–1·11) .. .. 1·28 (1·18–1·38) .. .. 1·27 (1·25–1·29) Expected¶ .. .. 1·25 (1·24–1·26) .. .. 1·48 (1·46–1·50) .. .. 1·17 (1·11–1·24) .. .. 1·31 (1·27–1·35) .. .. 1·29 (1·29–1·30) Body-mass index (BMI) unit is kg/m2. HR=hazard ratio. ..=not applicable. * Per 10 000 person-years. Standardised to age, sex, and study area structure of the China Kadoorie Biobank population. † Stratified by age, sex, and study area and adjusted for education, smoking, alcohol consumption, physical activity, and self-rated health status. ‡ p value <0·0001 both for trend test and for test of non-linearity across seven groups. ¶ Estimated according to the corresponding difference in usual systolic blood pressure (8·3 mm Hg) and its associated stroke risk. The HRs for ischaemic stroke per 5 kg/m2 baseline BMI were greater at younger than at older ages (figure 3). The HRs were also greater in men than women (1·39 [95% CI 1·35–1·43] vs 1·26 [1·23–1·28]; pheterogeneity<0·0001), especially at age-at-risk of more than 60 years. For intracerebral haemorrhage, there was a weak inverse association with BMI at ages 70–79 years but, at younger than 70 years, BMI was weakly and positively associated with intracerebral haemorrhage.Figure 3 Adjusted HRs for ischaemic stroke (A,B) and intracerebral haemorrhage (C,D) by baseline BMI, stratified by age (A,C) and age and sex (B,D)

ge, there was a weak inverse association with BMI at ages 70–79 years but, at younger than 70 years, BMI was weakly and positively associated with intracerebral haemorrhage.Figure 3 Adjusted HRs for ischaemic stroke (A,B) and intracerebral haemorrhage (C,D) by baseline BMI, stratified by age (A,C) and age and sex (B,D) The left panel shows adjusted HRs for stroke types by baseline BMI, stratified by age-at-risk. HRs are stratified by sex and area, and adjusted simultaneously for education, smoking, alcohol consumption, physical activity, and self-rated health status. HRs are plotted on a floating absolute scale. Each closed square represents HR with the area inversely proportional to the variance of the log HR. The vertical lines indicate 95% CIs. The right panel shows HRs for stroke types per 5 kg/m2 higher BMI, stratified by age-at-risk and sex. Each square represents HR with the area inversely proportional to the variance of the log HR. The horizontal lines indicate 95% CIs. The dashed vertical line indicates the overall HR for men and women combined, and open diamonds indicate combined values and their 95% CIs. HR=hazard ratio. BMI=body-mass index.

and sex. Each square represents HR with the area inversely proportional to the variance of the log HR. The horizontal lines indicate 95% CIs. The dashed vertical line indicates the overall HR for men and women combined, and open diamonds indicate combined values and their 95% CIs. HR=hazard ratio. BMI=body-mass index. The associations between BMI and ischaemic stroke and intracerebral haemorrhage were similar in urban and rural areas (appendix p 6) and across the ten study areas (appendix p 7). The associations were unaltered by smoking status (appendix p 8), and by other factors except education (pheterogeneity=0·035 for ischaemic stroke and pheterogeneity=0·0038 for intracerebral haemorrhage) and, to a lesser extent, by physical activity (ptrend<0·0001 for ischaemic stroke and ptrend=0·088 for intracerebral haemorrhage; appendix p 9). These associations were unaltered by additional adjustment for certain dietary factors (eg, fresh fruit, preserved vegetable, dairy, and meat intake; HR 1·31 [95% CI 1·29–1·33] for ischaemic stroke and 1·12 [1·08–1·17] for intracerebral haemorrhage, per 5 kg/m2). The associations were slightly attenuated by additional exclusion of people with any previous diseases, people with self-reported poor health status at baseline, the first 3 years of follow-up, and ever-regular smokers (appendix p 10), but there was still no association with intracerebral haemorrhage at a BMI of less than 25 kg/m2.

ons were slightly attenuated by additional exclusion of people with any previous diseases, people with self-reported poor health status at baseline, the first 3 years of follow-up, and ever-regular smokers (appendix p 10), but there was still no association with intracerebral haemorrhage at a BMI of less than 25 kg/m2. The associations were, however, substantially attenuated by additional adjustment for blood pressure. For ischaemic stroke, simultaneous adjustment for usual systolic blood pressure (and diabetes status) reduced the HR from 1·30 (95% CI 1·28–1·33) to 1·05 (1·03–1·07; appendix p 11), whereas for intracerebral haemorrhage, the association was reversed (from 1·11 [1·07–1·16] to 0·73 [0·70–0·77]), including among individuals at age 35–49 years (0·66 [0·58–0·75]). There was no longer any sex difference in the strength of the associations between BMI and risk of ischaemic stroke and intracerebral haemorrhage after adjustment for usual systolic blood pressure. Likewise, the results were unaltered in the subset of participants after additional adjustment for LDL cholesterol, HDL cholesterol, and triglycerides (1·03 [0·99–1·08] for ischaemic stroke and 0·83 [0·78–0·87] for intracerebral haemorrhage, per 5 kg/m2).

haemorrhage after adjustment for usual systolic blood pressure. Likewise, the results were unaltered in the subset of participants after additional adjustment for LDL cholesterol, HDL cholesterol, and triglycerides (1·03 [0·99–1·08] for ischaemic stroke and 0·83 [0·78–0·87] for intracerebral haemorrhage, per 5 kg/m2). The shape and strength of the associations with other measures of adiposity (eg, body fat percentage and waist circumference) were generally similar to those for BMI (figure 2), as were the attenuating effects of adjustment for mediators (appendix pp 11–12) and augmenting effects of adjustment for regression dilution bias (appendix p 13). BMI at the age of 25 years was weakly positively associated with ischaemic stroke and intracerebral haemorrhage, and the association was largely unaffected by adjustment for systolic blood pressure (appendix p 14). BMI change between the age of 25 years and the baseline survey was also positively associated with ischaemic stroke, but the association disappeared almost completely after adjustment for systolic blood pressure, whereas for intracerebral haemorrhage, there was no association without adjustment and a strongly inverse association after adjustment for systolic blood pressure (appendix p 14).

positively associated with ischaemic stroke, but the association disappeared almost completely after adjustment for systolic blood pressure, whereas for intracerebral haemorrhage, there was no association without adjustment and a strongly inverse association after adjustment for systolic blood pressure (appendix p 14). Discussion This large prospective study showed that, among relatively lean but healthy Chinese adults, adiposity was strongly positively associated with systolic blood pressure, and that systolic blood pressure was strongly positively related to stroke incidence, particularly intracerebral haemorrhage, without evidence of any threshold throughout the range studied. However, despite these two continuous and apparently causal relationships, adiposity showed an expected strong positive association only with ischaemic stroke, but not with intracerebral haemorrhage.

troke incidence, particularly intracerebral haemorrhage, without evidence of any threshold throughout the range studied. However, despite these two continuous and apparently causal relationships, adiposity showed an expected strong positive association only with ischaemic stroke, but not with intracerebral haemorrhage. Several large prospective studies, and meta-analyses of such studies, have consistently reported positive associations between BMI and risks of ischaemic stroke and total stroke.5, 6, 8, 11, 12, 13, 15, 16, 17 In a meta-analysis of 16 large prospective studies, including eight from east Asia, the pooled relative risk (RR) for ischaemic stroke was greater in east Asian than in high-income populations (1·35 vs 1·22, per 5 kg/m2 higher BMI),17 but varied greatly across different studies both within and between east Asian and high-income populations. Our RR estimate for ischaemic stroke (1·30 per 5 kg/m2 BMI) was similar to that observed in previous east Asian studies. Likewise, a meta-analysis of 97 prospective studies, including 33 from Asia, which examined the associations of BMI with total stroke only, reported pooled RR in Asian cohorts that were similar to those reported in the present study (1·29 vs 1·27 per 5 kg/m2 higher BMI), but stronger than those in high-income populations (1·14).16 For total stroke, our study estimated that high BMI (defined as >23 kg/m2) accounted for 15% of total stroke, which was less than the 20% previously estimated for China in a recent Global Burden of Disease report10 that used the same BMI cut points. Moreover, our study also demonstrated that the RRs for ischaemic stroke were substantially greater at younger than older ages, although the absolute excess risks were greater at older ages. Previous studies showed that, given BMI levels, men tend to have greater insulin resistance, ectopic fat levels in the liver and elsewhere, and higher risks of type 2 diabetes and cardiovascular disease.9, 33 In our study, the RRs were also higher in men than in women, which was due mainly to the stronger associations in men between systolic blood pressure and stroke. Consistent with previous studies,6, 16 the observed association of BMI with ischaemic stroke was mediated mainly through blood pressure in this Chinese population.

r study, the RRs were also higher in men than in women, which was due mainly to the stronger associations in men between systolic blood pressure and stroke. Consistent with previous studies,6, 16 the observed association of BMI with ischaemic stroke was mediated mainly through blood pressure in this Chinese population. Most previous studies have focused mainly on haemorrhagic stroke rather than intracerebral haemorrhage specifically, and the results have been inconsistent, partly because of the relatively small number of cases studied and the strong possibility of misclassification of stroke types in studies without widespread use of brain imaging. In the most recent meta-analysis, which included more than 17 000 haemorrhagic stroke cases,17 the overall association between BMI and risk of haemorrhagic stroke differed substantially between high-income and east Asian cohorts (RR 0·91 vs 1·16, per 5 kg/m2 higher BMI). In the high-income cohorts, more than 80% of the haemorrhagic stroke cases were from one UK study, which used self-reported BMI and had a disproportionally large number of subarachnoid haemorrhage cases compared with intracerebral haemorrhage cases (3062 vs 2790). Moreover, this UK study reported no increased risk of intracerebral haemorrhage even among those who were obese, nor a positive association with risk of ischaemic stroke at BMI levels of less than 27·5 kg/m2, in contrast with the findings of this and many other previous studies.5, 6, 11, 12, 13, 16

ases (3062 vs 2790). Moreover, this UK study reported no increased risk of intracerebral haemorrhage even among those who were obese, nor a positive association with risk of ischaemic stroke at BMI levels of less than 27·5 kg/m2, in contrast with the findings of this and many other previous studies.5, 6, 11, 12, 13, 16 Consistent with our study, several large prospective studies in east Asian populations have reported little or no clear positive associations between BMI and risk of haemorrhagic stroke at a BMI of less than 25 kg/m2.5, 8, 11, 12 However, none of the previous studies included sufficient numbers of well characterised intracerebral haemorrhage cases as in this study, or attempted to quantify the associations in the context of the association between adiposity and systolic blood pressure and between systolic blood pressure and stroke risk. Among the two-thirds of CKB participants with a BMI of less than 25 kg/m2, there was no association between adiposity and intracerebral haemorrhage. Given the strong positive associations observed between adiposity and systolic blood pressure and between systolic blood pressure and intracerebral haemorrhage, the lack of any apparent association with intracerebral haemorrhage below BMI 25 kg/m2 was unexpected and unexplained. There was a mean difference of at least 11/5 mm Hg in systolic/diastolic blood pressure across the BMI range below 25 kg/m2, and this difference should, other things being equal, correspond to an HR of 1·6 for intracerebral haemorrhage, but it was totally flat. The disparities are too extreme to be accounted for by chance, by known confounding factors, or by reverse causality, although one cannot exclude the possibility that some other unmeasured or unknown risk factors associated with low BMI might offset the protective effects of lower blood pressure. Future Mendelian randomisation studies are needed to establish the causal relevance of adiposity for intracerebral haemorrhage.34

causality, although one cannot exclude the possibility that some other unmeasured or unknown risk factors associated with low BMI might offset the protective effects of lower blood pressure. Future Mendelian randomisation studies are needed to establish the causal relevance of adiposity for intracerebral haemorrhage.34 Few prospective cohort studies have examined the associations between stroke types and measures of adiposity other than BMI, such as waist circumference, waist-to-hip ratio, and body fat percentage.6, 24, 25 Compared with raised BMI, measures of central adiposity, such as waist circumference and waist-to-hip ratio, might be better indicators of accumulation of visceral fat and an adverse metabolic profile.35, 36, 37 However, in the Emerging Risk Factors Collaboration meta-analysis of 21 studies involving about 2400 ischaemic stroke cases,6 BMI, waist circumference, and waist-to-hip ratio each showed a similar strength of association with ischaemic stroke risk. This study included a ten times greater number of ischaemic stroke cases than did the latter meta-analysis and confirmed the similar associations between ischaemic stroke and different measures of adiposity. These associations are consistent with the fact that most of the adiposity measures (eg, BMI and waist circumference) were highly correlated with each other. Moreover, this study also provided a new assessment of the relevance of central adiposity and other measures of adiposity (eg, BMI at age 25 years and body fat percentage) for intracerebral haemorrhage.

he fact that most of the adiposity measures (eg, BMI and waist circumference) were highly correlated with each other. Moreover, this study also provided a new assessment of the relevance of central adiposity and other measures of adiposity (eg, BMI at age 25 years and body fat percentage) for intracerebral haemorrhage. Our study had several strengths, including a very large study population, availability of different measures of adiposity, completeness of follow-up, and a high proportion of stroke types reliably diagnosed by neuroimaging. Moreover, this study explored the effects of mediators (eg, systolic blood pressure) on the associations of adiposity with stroke types both directly and indirectly. However, the study also had several limitations. First, despite the widespread use of neuroimaging, the study was not able to examine the associations between adiposity and different subtypes of ischaemic stroke (eg, lacunar vs non-lacunar ischaemic stroke) or of intracerebral haemorrhage (eg, lobar vs non-lobar intracerebral haemorrhage). Second, information collected on diet was limited and did not include salt intake, so residual confounding might still persist, even though adjustment for fresh fruit and meat consumption did not alter the associations. Third, it was not possible to fully explore the mediating effects of lipids and other blood-related factors, although examination of a subset of data indicated that further adjustment for lipids had little additional impact over and above blood pressure.

fresh fruit and meat consumption did not alter the associations. Third, it was not possible to fully explore the mediating effects of lipids and other blood-related factors, although examination of a subset of data indicated that further adjustment for lipids had little additional impact over and above blood pressure. In summary, our study shows that, among relatively lean but healthy Chinese adults, adiposity was strongly and positively associated with ischaemic stroke, mainly through its effect on blood pressure. However, there was no association between adiposity and intracerebral haemorrhage across the normal range (ie, BMI <25 kg/m2), suggesting that leanness, either per se or through some other factor (or factors), might increase risk, thereby offsetting the protective effects associated with lower blood pressure. Since ischaemic stroke constitutes the majority of total stroke cases, the findings on intracerebral haemorrhage should not diminish the fundamental importance of high adiposity as a major modifiable determinant of overall stroke. In view of the substantial risks associated with high blood pressure, this study highlights the importance of controlling blood pressure (and other intermediate risk factors) for prevention of both ischaemic stroke and intracerebral haemorrhage, irrespective of levels of adiposity. Supplementary Material Supplementary appendix

In summary, our study shows that, among relatively lean but healthy Chinese adults, adiposity was strongly and positively associated with ischaemic stroke, mainly through its effect on blood pressure. However, there was no association between adiposity and intracerebral haemorrhage across the normal range (ie, BMI <25 kg/m2), suggesting that leanness, either per se or through some other factor (or factors), might increase risk, thereby offsetting the protective effects associated with lower blood pressure. Since ischaemic stroke constitutes the majority of total stroke cases, the findings on intracerebral haemorrhage should not diminish the fundamental importance of high adiposity as a major modifiable determinant of overall stroke. In view of the substantial risks associated with high blood pressure, this study highlights the importance of controlling blood pressure (and other intermediate risk factors) for prevention of both ischaemic stroke and intracerebral haemorrhage, irrespective of levels of adiposity. Supplementary Material Supplementary appendix Acknowledgments The chief acknowledgment is to the participants, the project staff, and the China National Centre for Disease Control and Prevention and its regional offices for access to death and disease registries. The Chinese National Health Insurance scheme provides electronic linkage to all hospital admission data. The baseline survey and the first re-survey were supported by a research grant from the Kadoorie Charitable Foundation in Hong Kong. The long-term continuation of the project is supported by programme grants from the UK Wellcome Trust (088158/Z/09/Z, 104085/Z/14/Z, 088158/Z/09/Z), the Chinese Ministry of Science and Technology (2011BAI09B01, 2012-14), the Chinese National Natural Science Foundation (81390540/81390541/81390544, 2013-18), and the National Key Research and Development Program of China (2016YFC0900500/2016YFC0900501/2016YFC0900504/2016YFC1303904, 2016-2021). The British Heart Foundation (BHF), UK Medical Research Council, and Cancer Research UK provide core funding to the Oxford Clinical Trial Service Unit and Epidemiological Studies Unit. FB acknowledges support from the BHF Centre of Research Excellence, Oxford.

0/2016YFC0900501/2016YFC0900504/2016YFC1303904, 2016-2021). The British Heart Foundation (BHF), UK Medical Research Council, and Cancer Research UK provide core funding to the Oxford Clinical Trial Service Unit and Epidemiological Studies Unit. FB acknowledges support from the BHF Centre of Research Excellence, Oxford. Contributors All authors were involved in study design, conduct, long-term follow-up, analysis of data, interpretation, or writing of the report. Declaration of interests We declare no competing interests.

Introduction In 2013, WHO set global targets for the control of non-communicable diseases, including a 25% relative reduction in the prevalence of elevated blood pressure by 2025.1 About 80% of premature deaths from non-communicable diseases occur in low-income and middle-income countries.2 However, estimates of the effects of elevated blood pressure on the worldwide burden of non-communicable diseases, such as ischaemic heart disease and stroke, are based largely on findings of prospective studies in high-income countries.3 Studies assessing the hazards of elevated blood pressure in low-income and middle-income countries, including China, are needed to inform local and regional disease prevention strategies. Results of such studies might also substantially improve the global estimates of the burden of disease attributable to elevated blood pressure.

essing the hazards of elevated blood pressure in low-income and middle-income countries, including China, are needed to inform local and regional disease prevention strategies. Results of such studies might also substantially improve the global estimates of the burden of disease attributable to elevated blood pressure. Meta-analyses of prospective studies have demonstrated the importance of moderate differences in blood pressure for risk of ischaemic heart disease and stroke,4, 5 and randomised trials of blood-pressure-lowering medication have confirmed the reversibility of most of the excess risks within 4–5 years of initiating treatment.6 In China, however, despite high rates of vascular mortality (particularly from stroke),7 the strengths of the associations between blood pressure and the main types of vascular disease have not been well established; large-scale prospective studies, with measurement of long-term average (ie, usual) levels of blood pressure and reliable phenotyping of the main stroke types (ischaemic stroke and intracerebral haemorrhage), have only become feasible in China over the past decade or so. Research in context Evidence before this study

Meta-analyses of prospective studies have demonstrated the importance of moderate differences in blood pressure for risk of ischaemic heart disease and stroke,4, 5 and randomised trials of blood-pressure-lowering medication have confirmed the reversibility of most of the excess risks within 4–5 years of initiating treatment.6 In China, however, despite high rates of vascular mortality (particularly from stroke),7 the strengths of the associations between blood pressure and the main types of vascular disease have not been well established; large-scale prospective studies, with measurement of long-term average (ie, usual) levels of blood pressure and reliable phenotyping of the main stroke types (ischaemic stroke and intracerebral haemorrhage), have only become feasible in China over the past decade or so. Research in context Evidence before this study We did a literature search to identify articles from prospective studies in China that reported on the association between blood pressure and chronic disease. We searched PubMed for articles published between Jan 1, 1960, and March 1, 2017, using the terms “blood pressure” (or “hypertension”) and “prospective study” (or “cohort study” or “longitudinal study”) and “China” (or “Chinese”). We used search terms in English but did not apply any language restrictions. The reference lists of relevant articles were also reviewed. We identified numerous small prospective studies (<50 000 participants) but few large studies, and the largest of these had less than half the number of vascular events than this study. All identified studies reported positive associations between blood pressure and vascular disease (and several reported positive associations with chronic kidney disease and cancer), but most examined effects on mortality only and did not correct for regression dilution bias. Furthermore, the larger prospective studies tended to pre-date the routine use of brain imaging required to reliably differentiate the main stroke types (ischaemic stroke and intracerebral haemorrhage). As such, although blood pressure is an established cause of vascular disease, uncertainty remained about the strength of its association with the major types of vascular disease in China.

tine use of brain imaging required to reliably differentiate the main stroke types (ischaemic stroke and intracerebral haemorrhage). As such, although blood pressure is an established cause of vascular disease, uncertainty remained about the strength of its association with the major types of vascular disease in China. Added value of this study This large prospective study reliably quantifies the age-specific associations between blood pressure and major vascular and non-vascular chronic diseases in men and women in China. For each decade of age at risk (40–79 years), there were positive, log-linear associations between systolic blood pressure and risk of ischaemic heart disease, ischaemic stroke, intracerebral haemorrhage, and the aggregate of all major vascular diseases, with no evidence of a threshold down to at least 120 mm Hg. Overall, each 10 mm Hg higher usual systolic blood pressure was associated with an approximately 30% higher risk of ischaemic heart disease and of ischaemic stroke, but the association with intracerebral haemorrhage was about twice as steep. Usual systolic blood pressure was also positively associated with chronic kidney disease and diabetes, but not with cancer (overall or with several common types), liver cirrhosis, or chronic obstructive pulmonary disease; the causal relevance of these positive associations with non-vascular chronic disease remains unclear. Implications of all the available evidence

This large prospective study reliably quantifies the age-specific associations between blood pressure and major vascular and non-vascular chronic diseases in men and women in China. For each decade of age at risk (40–79 years), there were positive, log-linear associations between systolic blood pressure and risk of ischaemic heart disease, ischaemic stroke, intracerebral haemorrhage, and the aggregate of all major vascular diseases, with no evidence of a threshold down to at least 120 mm Hg. Overall, each 10 mm Hg higher usual systolic blood pressure was associated with an approximately 30% higher risk of ischaemic heart disease and of ischaemic stroke, but the association with intracerebral haemorrhage was about twice as steep. Usual systolic blood pressure was also positively associated with chronic kidney disease and diabetes, but not with cancer (overall or with several common types), liver cirrhosis, or chronic obstructive pulmonary disease; the causal relevance of these positive associations with non-vascular chronic disease remains unclear. Implications of all the available evidence Our study showed stronger associations between blood pressure and vascular disease than have previous prospective studies in China. The magnitude of the age-specific associations between blood pressure and ischaemic heart disease was similar to those reported in previous large meta-analyses of prospective studies. In contrast to these meta-analyses, however, the present study found strong evidence that, throughout middle age and into old age, there were steeper associations for intracerebral haemorrhage than ischaemic stroke. It was estimated that about half of all vascular deaths in China were attributable to elevated blood pressure (ie, systolic blood pressure >120 mm Hg), accounting for around 1 million deaths (age <80 years) annually; given the strength of the associations between blood pressure and vascular disease, even small reductions in mean blood pressure at a population level would be expected to have a major impact on vascular morbidity and mortality.

lood pressure >120 mm Hg), accounting for around 1 million deaths (age <80 years) annually; given the strength of the associations between blood pressure and vascular disease, even small reductions in mean blood pressure at a population level would be expected to have a major impact on vascular morbidity and mortality. Reliable classification of the main stroke types requires brain imaging,8 but such imaging is sometimes not performed or is not readily available for epidemiological research. Meta-analyses of prospective studies in western (mainly European and North American) countries, with cohorts that largely pre-date the routine use of brain imaging, report similar associations between systolic blood pressure and risk of both stroke types.4 By contrast, several studies done in east Asia (where haemorrhagic stroke is more common than in western populations) have reported more extreme associations with intracerebral haemorrhage than ischaemic stroke.5, 9, 10, 11 Few large-scale prospective studies, however, have achieved the high rates of brain imaging required to avoid substantial misclassification between stroke types.

c stroke is more common than in western populations) have reported more extreme associations with intracerebral haemorrhage than ischaemic stroke.5, 9, 10, 11 Few large-scale prospective studies, however, have achieved the high rates of brain imaging required to avoid substantial misclassification between stroke types. Hence, although blood pressure is an established cause of vascular disease, uncertainty remains about the strength of its association with the main stroke types and with ischaemic heart disease in China. Furthermore, studies in western populations have reported strong positive associations between blood pressure and risk of several non-vascular diseases, including chronic kidney disease, diabetes, and certain cancers, but such associations have not been reliably assessed in China.12, 13, 14 Using data from the large nationwide China Kadoorie Biobank (CKB) prospective study, we aimed to assess the shape and strength of the associations between usual systolic blood pressure and major vascular diseases, including ischaemic heart disease, ischaemic stroke, and intracerebral haemorrhage; to compare these associations with those of usual systolic blood pressure with incident non-vascular chronic diseases; and to estimate the burden of vascular deaths attributable to elevated blood pressure in China.

or vascular diseases, including ischaemic heart disease, ischaemic stroke, and intracerebral haemorrhage; to compare these associations with those of usual systolic blood pressure with incident non-vascular chronic diseases; and to estimate the burden of vascular deaths attributable to elevated blood pressure in China. Methods Study population Details of the study design and survey methods have been reported previously.15, 16 Briefly, 512 891 adults (aged 30–79 years) were recruited from the general population in ten areas of China between June 25, 2004, and July 15, 2008. Approximately 30% of the resident population in each area is estimated to have participated in the study. Participants attended temporary assessment centres where interviewers recorded their age, sex, socioeconomic status, aspects of lifestyle (including alcohol intake, smoking, diet, and physical activity), and medical history. Measurements of blood pressure, height, and weight were recorded, and a blood sample was collected. Ethics approval was obtained from the relevant local, national, and international ethics committees, and all participants provided written informed consent.

ke, smoking, diet, and physical activity), and medical history. Measurements of blood pressure, height, and weight were recorded, and a blood sample was collected. Ethics approval was obtained from the relevant local, national, and international ethics committees, and all participants provided written informed consent. Measurement of blood pressure After participants had sat for at least 5 min, blood pressure was measured twice using a UA-779 digital sphygmomanometer (A&D Instruments; Abingdon, UK). If the difference between the two measurements was more than 10 mm Hg for systolic blood pressure, a third measurement was taken. Only the last two readings were recorded, and the mean of these readings was used in all analyses. Sphygmomanometers were supplied centrally, calibrated daily, and only used by trained fieldworkers. Repeat measurements of blood pressure were obtained from a random sample of 19 788 participants (approximately 5%) about 3 years after baseline (between May 26, and Oct 10, 2008) using procedures identical to those used in the baseline survey. Repeated blood pressure measures were used to correct prospective analyses for the regression dilution bias that results from the inaccuracy with which a single measurement of an exposure at baseline characterises an individual's usual level.17 Rosner's regression method was used to calculate the proportional reduction in the strength of the associations that resulted from regression dilution (the regression dilution ratio) for systolic blood pressure, whereby the ratio is equal to the slope of the regression line between baseline and resurvey systolic blood pressure values.18

n method was used to calculate the proportional reduction in the strength of the associations that resulted from regression dilution (the regression dilution ratio) for systolic blood pressure, whereby the ratio is equal to the slope of the regression line between baseline and resurvey systolic blood pressure values.18 Morbidity and mortality follow-up Deaths were identified by electronic linkage to local mortality records, supplemented by annual reviews of local residential registers to check for completeness. Non-fatal events were identified from local chronic disease registries (for stroke, ischaemic heart disease, cancer, and diabetes) and from claims to the national health insurance system, which has almost universal coverage in all study areas. The underlying causes of death and hospital diagnoses were coded in accordance with the International Classification of Diseases 10th revision (ICD-10; see appendix p 3 for ICD-10 codes used to define disease endpoints). Residential records were used to identify participants who moved out of the study areas and were lost to follow-up.

auses of death and hospital diagnoses were coded in accordance with the International Classification of Diseases 10th revision (ICD-10; see appendix p 3 for ICD-10 codes used to define disease endpoints). Residential records were used to identify participants who moved out of the study areas and were lost to follow-up. Statistical analysis We excluded participants with missing information on blood pressure or any covariates, those with extreme values of systolic blood pressure (<80 mm Hg or ≥250 mm Hg) or diastolic blood pressure (<40 mm Hg or ≥150 mm Hg), those with no follow-up in the age range 40–79 years, and those with a (self-reported) history of vascular disease at baseline (myocardial infarction, stroke, or transient ischaemic attack). For prospective analyses of each non-vascular disease, participants with a history of the relevant disease at baseline were excluded (diabetes at baseline includes both self-reported and screen-detected cases; appendix p 4). The main analyses are of disease associations with systolic blood pressure, but the corresponding analyses for diastolic blood pressure are provided in the appendix pp 17–20.

ory of the relevant disease at baseline were excluded (diabetes at baseline includes both self-reported and screen-detected cases; appendix p 4). The main analyses are of disease associations with systolic blood pressure, but the corresponding analyses for diastolic blood pressure are provided in the appendix pp 17–20. We used Cox regression analysis to produce adjusted hazard ratios (HRs) for the associations between systolic blood pressure and first occurrence (non-fatal or fatal) of major vascular disease (defined as first myocardial infarction, first stroke, or vascular death), chronic kidney disease, type 2 diabetes, cancer, liver cirrhosis, or chronic obstructive pulmonary disease. Major vascular disease was further classified into ischaemic heart disease (first myocardial infarction or other ischaemic heart disease death), stroke types (ischaemic stroke, intracerebral haemorrhage, and other stroke), and other vascular diseases.

, cancer, liver cirrhosis, or chronic obstructive pulmonary disease. Major vascular disease was further classified into ischaemic heart disease (first myocardial infarction or other ischaemic heart disease death), stroke types (ischaemic stroke, intracerebral haemorrhage, and other stroke), and other vascular diseases. HRs were adjusted for sex, education, area, smoking, alcohol intake, and body-mass index (BMI). An additional term was fitted that allowed the HR in each decade of age to be estimated as the geometric mean of the HRs in the first and second half of that decade, which avoids any assumptions as to whether the proportional associations of risk to usual systolic blood pressure are similar in different decades of age. The log HRs for a 10 mm Hg higher systolic blood pressure were obtained by fitting systolic blood pressure as a continuous variable in men and women separately for each decade of age at risk (40–49, 50–59, 60–69, and 70–79 years). Age-specific and sex-specific HRs were then corrected for age-specific and sex-specific regression dilution ratios to give the HR for a 10 mm Hg higher usual systolic blood pressure (appendix pp 5–7), and combined by calculating their inverse-variance weighted average.

ge at risk (40–49, 50–59, 60–69, and 70–79 years). Age-specific and sex-specific HRs were then corrected for age-specific and sex-specific regression dilution ratios to give the HR for a 10 mm Hg higher usual systolic blood pressure (appendix pp 5–7), and combined by calculating their inverse-variance weighted average. In categorical analyses, participants were divided into five groups according to baseline systolic blood pressure (80–124 mm Hg, 125–144 mm Hg, 145–164 mm Hg, 165–184 mm Hg, and 185–249 mm Hg). HRs were calculated relative to the lowest systolic blood pressure group, and plotted against mean usual systolic blood pressure in each of these groups (appendix p 7); the 95% CIs for these HRs were estimated with the variance of the log risk.19 Fatal events were defined as death from any cause within 28 days of the event.

ere calculated relative to the lowest systolic blood pressure group, and plotted against mean usual systolic blood pressure in each of these groups (appendix p 7); the 95% CIs for these HRs were estimated with the variance of the log risk.19 Fatal events were defined as death from any cause within 28 days of the event. Assuming causality of the association between blood pressure and vascular mortality, we estimated the fraction of deaths attributable to elevated systolic blood pressure by Pd × (RR – 1)/RR, where Pd represents the proportion of all vascular deaths in a given systolic blood pressure group (with the same systolic blood pressure groups already described) and RR, the relative risk, is approximated by the group-specific HR (this approximation is valid early in follow-up, when the event rate is low) relative to the lowest systolic blood pressure group.20 The overall fraction of vascular deaths attributable to elevated systolic blood pressure was calculated as the sum of the group-specific fractions. All analyses were done with SAS software (version 9.3), and figures were plotted using R software (version 3.3). Role of the funding source The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. BL, SL, RCl, ZC, and LL had full access to the data and analyses and had final responsibility for the decision to submit for publication.

Assuming causality of the association between blood pressure and vascular mortality, we estimated the fraction of deaths attributable to elevated systolic blood pressure by Pd × (RR – 1)/RR, where Pd represents the proportion of all vascular deaths in a given systolic blood pressure group (with the same systolic blood pressure groups already described) and RR, the relative risk, is approximated by the group-specific HR (this approximation is valid early in follow-up, when the event rate is low) relative to the lowest systolic blood pressure group.20 The overall fraction of vascular deaths attributable to elevated systolic blood pressure was calculated as the sum of the group-specific fractions. All analyses were done with SAS software (version 9.3), and figures were plotted using R software (version 3.3). Role of the funding source The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. BL, SL, RCl, ZC, and LL had full access to the data and analyses and had final responsibility for the decision to submit for publication. Results Of 512 891 participants, we excluded two with missing information, 110 with extreme systolic blood pressure, 62 with extreme diastolic blood pressure, 488 without follow-up in the age range 40–79 years, and 23 104 with a baseline history of vascular disease. By Jan 1, 2016, 37 289 (7·3%) participants had died and 4875 (1·0%) participants were lost to follow-up.

th missing information, 110 with extreme systolic blood pressure, 62 with extreme diastolic blood pressure, 488 without follow-up in the age range 40–79 years, and 23 104 with a baseline history of vascular disease. By Jan 1, 2016, 37 289 (7·3%) participants had died and 4875 (1·0%) participants were lost to follow-up. Among 489 125 participants included in the main analyses, the mean age at baseline was 51 years (SD 10) and 289 224 (59·1%) were women (table). Overall, 211 085 (43·2%) participants resided in urban areas and 240 380 (49·1%) had been educated to middle school or higher. Mean BMI was 23·6 kg/m2 (SD 3·4), and 26 393 (5·4%) participants had diabetes (13 300 [2·7%] were clinically diagnosed and a further 13 093 [2·7%] were screen detected from blood glucose tests at baseline). Among men, 123 902 (62·0%) were current smokers and 67 693 (33·9%) consumed alcohol at least weekly, but only 6616 (2·3%) women were smokers and 6017 (2·1%) consumed alcohol at least weekly. Overall, mean systolic blood pressure was 130·6 mm Hg (SD 21·0), and 17 623 (3·6%) participants reported current use of blood-pressure-lowering medication.Table Baseline characteristics

93 (33·9%) consumed alcohol at least weekly, but only 6616 (2·3%) women were smokers and 6017 (2·1%) consumed alcohol at least weekly. Overall, mean systolic blood pressure was 130·6 mm Hg (SD 21·0), and 17 623 (3·6%) participants reported current use of blood-pressure-lowering medication.Table Baseline characteristics Baseline systolic blood pressure All (n=489 125) 80–124 mm Hg (n=158 958) 125–144 mm Hg (n=224 885) 145–164 mm Hg (n=71 081) 165–184 mm Hg (n=24 952) 185–249 mm Hg (n=9249) Age (years) 47 (10) 51 (10) 56 (10) 58 (10) 59 (9) 51 (10) Female 105 630 (66·5%) 123 152 (54·8%) 39 748 (56·7%) 14 905 (59·7%) 5789 (62·6%) 289 224 (59·1%) Systolic blood pressure (mm Hg) 109·7 (7·3) 130·8 (6·9) 153·3 (5·7) 172·9 (5·6) 197·8 (11·9) 130·6 (21·0) Diastolic blood pressure (mm Hg) 69·1 (7·0) 78·3 (7·9) 86·4 (9·5) 92·4 (10·7) 100·1 (12·2) 77·7 (11·1) Body-mass index (kg/m2) 22·6 (3·0) 23·8 (3·3) 24·6 (3·5) 24·7 (3·7) 24·8 (3·7) 23·6 (3·4) MET-h/day of physical activity 22·3 (13·6) 22·0 (14·1) 19·7 (13·9) 18·6 (13·4) 17·8 (12·8) 21·5 (13·9) Urban area 77 932 (49·0%) 91 615 (40·7%) 29 194 (40·9%) 9276 (37·2%) 3068 (33·2%) 211 085 (43·2%) Educated to middle school or higher* 93 691 (58·9%) 109 894 (48·9%) 26 815 (37·7%) 7590 (30·4%) 2390 (25·8%) 240 380 (49·1%) Current smokers (male only†) 35 090 (65·8%) 62 845 (61·8%) 18 053 (57·6%) 5849 (58·2%) 2065 (59·7%) 123 902 (62·0%) Alcohol consumption at least weekly (male only†) 15 507 (29·1%) 35 539 (34·9%) 11 712 (37·4%) 3637 (36·2%) 1298 (37·5%) 67 693 (33·9%) Fresh fruit consumed daily 35 367 (22·2%) 38 588 (17·2%) 11 077 (15·6) 3314 (13·3%) 1044 (11·3%) 89 390 (18·3%) Fresh vegetable consumed daily 150 054 (94·4%) 213 669 (95·0%) 67 361 (94·8%) 23 474 (94·1%) 8609 (93·1%) 463 167 (94·7%) Diabetes‡ 4197 (2·6%) 11 813 (5·3%) 6615 (9·3%) 2695 (10·8%) 1073 (11·6%) 26 393 (5·4%) Taking blood-pressure-lowering medication 929 (0·6%) 6399 (2·8%) 5890 (8·3%) 3068 (12·3%) 1337 (14·5%) 17 623 (3·6%) Data are mean (SD) and n (%). The main analysis excludes people with no follow-up at ages 40–79 years; those with a previous diagnosis of vascular disease at baseline (myocardial infarction, stroke, or transient ischaemic attack); those with out-of-range baseline systolic or diastolic blood pressure; and those with missing information on blood pressure or other important factors. MET=metabolic equivalent.

s 40–79 years; those with a previous diagnosis of vascular disease at baseline (myocardial infarction, stroke, or transient ischaemic attack); those with out-of-range baseline systolic or diastolic blood pressure; and those with missing information on blood pressure or other important factors. MET=metabolic equivalent. * Middle school is typically attended between ages of 12 years and 15 years. † 6616 (2·3%) women smoked and 6017 (2·1%) reported drinking alcohol at least weekly. ‡ Includes those with a (self-reported) previous medical diagnosis of diabetes and those detected by blood glucose tests at baseline. Mean systolic blood pressure increased linearly with age in both sexes (figure 1). Mean diastolic blood pressure also increased with age until about 50 years, but decreased thereafter. Systolic blood pressure was positively associated with BMI (about 8 mm Hg higher systolic blood pressure per 5 kg/m2 higher BMI), and was inversely associated with level of education and residing in an urban area. At resurvey, about 3 years after the baseline survey, the overall regression dilution ratio for systolic blood pressure was 0·6, but it was slightly lower (ie, reflecting greater within-person variability) in men than women, and in older than in younger individuals (appendix p 6).Figure 1 Mean blood pressure at baseline, by age and sex Means standardised for region. Analyses in 489 125 participants; exclusions as in the table. SBP=systolic blood pressure. DBP=diastolic blood pressure.

Mean systolic blood pressure increased linearly with age in both sexes (figure 1). Mean diastolic blood pressure also increased with age until about 50 years, but decreased thereafter. Systolic blood pressure was positively associated with BMI (about 8 mm Hg higher systolic blood pressure per 5 kg/m2 higher BMI), and was inversely associated with level of education and residing in an urban area. At resurvey, about 3 years after the baseline survey, the overall regression dilution ratio for systolic blood pressure was 0·6, but it was slightly lower (ie, reflecting greater within-person variability) in men than women, and in older than in younger individuals (appendix p 6).Figure 1 Mean blood pressure at baseline, by age and sex Means standardised for region. Analyses in 489 125 participants; exclusions as in the table. SBP=systolic blood pressure. DBP=diastolic blood pressure. During 3·9 million person-years of follow-up (median 9 years [IQR 8–10]), 42 662 incident major vascular disease events occurred, including 4450 cases of ischaemic heart disease, 29 343 of ischaemic stroke, 6011 of intracerebral haemorrhage, 1572 of other (or unspecified) stroke, and 1286 of other vascular events. About 90% of stroke cases were diagnosed for stroke type with brain imaging (CT scan or MRI), as estimated from a detailed review of the medical records of a representative subset of cases. Additionally, there were 21 044 incident cases of cancer, 12 241 of diabetes, 8478 of chronic obstructive pulmonary disease, 1940 of chronic kidney disease, and 1740 of chronic liver disease.

brain imaging (CT scan or MRI), as estimated from a detailed review of the medical records of a representative subset of cases. Additionally, there were 21 044 incident cases of cancer, 12 241 of diabetes, 8478 of chronic obstructive pulmonary disease, 1940 of chronic kidney disease, and 1740 of chronic liver disease. For each decade of age at risk, usual systolic blood pressure was positively and log-linearly associated with risk of major vascular disease throughout the systolic blood pressure range examined, with no evidence of a threshold down to at least 120 mm Hg (figure 2). The HRs were about twice as steep at ages 40–49 years than 70–79 years and, within each decade, were somewhat greater in men than women. Overall (mean age at event 64 years [SD 9]), each 10 mm Hg higher usual systolic blood pressure was associated with 36% higher risk (HR 1·36 [95% CI 1·35–1·37]) of major vascular disease.Figure 2 Age-specific incidence of major vascular disease versus usual SBP HRs are adjusted for age at risk (5-year groups), sex, area, education, smoking, alcohol consumption, and body-mass index. (A) The area of each square is inversely proportional to the variance of the category-specific log risk. (B) The area of each square is inversely proportional to the variance of the log HR. Corresponding 95% CIs are plotted as lines. Analyses were done in 489 125 participants at risk and reasons for exclusion are shown in the table. SBP=systolic blood pressure. HR=hazard ratio.

variance of the category-specific log risk. (B) The area of each square is inversely proportional to the variance of the log HR. Corresponding 95% CIs are plotted as lines. Analyses were done in 489 125 participants at risk and reasons for exclusion are shown in the table. SBP=systolic blood pressure. HR=hazard ratio. The associations between usual systolic blood pressure and ischaemic heart disease, ischaemic stroke, and intracerebral haemorrhage were also approximately log-linear throughout the systolic blood pressure range examined (figure 3). Overall, each 10 mm Hg higher usual systolic blood pressure was associated with about 30% higher risk of ischaemic heart disease (HR 1·31 [95% CI 1·28–1·34]) and ischaemic stroke (1·30 [1·29–1·31]), but the associations with intracerebral haemorrhage were about twice as steep (1·68 [1·65–1·71]). The HRs for each of these vascular diseases were substantially greater at ages 40–49 years than 70–79 years (figure 4; appendix p 8). The associations were also somewhat greater in men than women for both stroke types, but not for ischaemic heart disease. In addition, associations were greater for fatal than non-fatal ischaemic heart disease (1·35 [1·32–1·39] vs 1·23 [1·19–1·28]) and for fatal than non-fatal ischaemic stroke (1·47 [1·40–1·55] vs 1·30 [1·29–1·31]), but there were no such differences for intracerebral haemorrhage.Figure 3 Incidence of ischaemic heart disease, ischaemic stroke, and intracerebral haemorrhage versus usual SBP

emic heart disease (1·35 [1·32–1·39] vs 1·23 [1·19–1·28]) and for fatal than non-fatal ischaemic stroke (1·47 [1·40–1·55] vs 1·30 [1·29–1·31]), but there were no such differences for intracerebral haemorrhage.Figure 3 Incidence of ischaemic heart disease, ischaemic stroke, and intracerebral haemorrhage versus usual SBP HRs at ages 40–79 years, adjusted for age at risk (5-year groups), sex, area, education, smoking, alcohol consumption, and body-mass index. For each category, the area of each square is inversely proportional to the variance of the category-specific log risk, which also determines the 95% CI. The HR is shown above each square and numbers of events below. Analyses were done in 489 125 participants at risk and reasons for exclusion are shown in the table. SBP=systolic blood pressure. HR=hazard ratio. Figure 4 Effect of 10 mm Hg higher SBP on incidence of ischaemic heart disease, ischaemic stroke, and intracerebral haemorrhage, by age and sex HRs for 10 mm Hg higher usual SBP at ages 40–79 years, adjusted for age at risk (5-year groups), sex, area, education, smoking, alcohol consumption, and body-mass index. For each category, area of square is inversely proportional to the variance log HR, which also determines the 95% CI. Analyses were done in 489 125 participants at risk and reasons for exclusion are shown in the table. SBP=systolic blood pressure. HR=hazard ratio.

ion, smoking, alcohol consumption, and body-mass index. For each category, area of square is inversely proportional to the variance log HR, which also determines the 95% CI. Analyses were done in 489 125 participants at risk and reasons for exclusion are shown in the table. SBP=systolic blood pressure. HR=hazard ratio. Only 4% of strokes were classified as other or unspecified type, and the strength of these associations with usual systolic blood pressure was similar to that for ischaemic stroke (figure 5; appendix p 9). For the aggregate of all other vascular events, the strength of the association was consistent with all major vascular disease. Usual systolic blood pressure was also strongly and positively associated with risk of chronic kidney disease (HR 1·40 [95% CI 1·35–1·44]) and less strongly associated with diabetes (1·14 [1·12–1·15]), but was not associated with cancer, liver cirrhosis, or chronic obstructive pulmonary disease.Figure 5 Effect of 10 mm Hg higher SBP on incidence of vascular and non-vascular chronic disease HRs for 10 mm Hg higher usual SBP at ages 40–79 years, adjusted for age at risk (5-year groups), sex, area, education, smoking, alcohol consumption, and body-mass index. For each category, the area of the square is inversely proportional to the variance of the log HR, which also determines the 95% CI. Reasons for exclusion are shown in the table; analyses of each non-vascular disease further exclude participants with a previous diagnosis of the relevant disease at baseline (appendix p 4). SBP=systolic blood pressure. HR=hazard ratio.

rsely proportional to the variance of the log HR, which also determines the 95% CI. Reasons for exclusion are shown in the table; analyses of each non-vascular disease further exclude participants with a previous diagnosis of the relevant disease at baseline (appendix p 4). SBP=systolic blood pressure. HR=hazard ratio. For vascular mortality at ages 40–79 years, each 10 mm Hg higher usual systolic blood pressure was associated with about a 50% higher risk of vascular death (HR 1·49 [95% CI 1·47–1·51]; mean age at death 68 years [SD 9]; appendix p 10). Non-vascular mortality was also positively associated with usual systolic blood pressure, although the risks were much weaker than those for vascular mortality 1·07 [1·06–1·08]). Assuming causality, the excess vascular mortality associated with elevated blood pressure (ie, systolic blood pressure >120 mm Hg) accounted for about half (45% [41–48%]) of all vascular deaths (ie, roughly 1·1 million deaths) between ages 40 years and 79 years in China in 2015 (appendix p 11).

ality 1·07 [1·06–1·08]). Assuming causality, the excess vascular mortality associated with elevated blood pressure (ie, systolic blood pressure >120 mm Hg) accounted for about half (45% [41–48%]) of all vascular deaths (ie, roughly 1·1 million deaths) between ages 40 years and 79 years in China in 2015 (appendix p 11). In sensitivity analyses (appendix pp 12–16), the associations were similar in urban and rural areas (except those for ischaemic stroke, which were somewhat shallower in urban than rural areas), and were not materially altered by further adjustment for other potential confounders (including physical activity, frequency of fruit and vegetable consumption, and, where appropriate, diabetes); excluding the first 3 years of follow-up (to assess for reverse causality); excluding participants who were using blood-pressure-lowering medication at baseline; or censoring at the first occurrence of chronic disease of any type (to assess for competing risks). Furthermore, there was no evidence that the associations varied by type of chronic kidney disease or by type of common cancer. The associations for diastolic blood pressure were broadly consistent with those for systolic blood pressure: there were log-linear associations of usual diastolic blood pressure with each of the major types of vascular disease throughout the blood pressure range examined (75–100 mm Hg usual diastolic blood pressure), and the strengths of these associations for 5 mm Hg higher usual diastolic blood pressure were equivalent to about 10 mm Hg higher usual systolic blood pressure (appendix pp 17–20).

th each of the major types of vascular disease throughout the blood pressure range examined (75–100 mm Hg usual diastolic blood pressure), and the strengths of these associations for 5 mm Hg higher usual diastolic blood pressure were equivalent to about 10 mm Hg higher usual systolic blood pressure (appendix pp 17–20). Discussion In this Chinese population, usual blood pressure was positively and log-linearly associated with risk of ischaemic heart disease, ischaemic stroke, intracerebral haemorrhage, and the aggregate of all major vascular diseases. These relationships continued down to at least 120 mm Hg systolic blood pressure and 75 mm Hg diastolic blood pressure, below which the evidence was scarce. The associations attenuated with increasing age, but even at ages 70–79 years there were strong associations with each of the main types of vascular disease. At all ages, there were steeper associations between blood pressure and intracerebral haemorrhage than ischaemic stroke. We estimated that about half of all premature vascular deaths were attributable to elevated blood pressure (ie, systolic blood pressure >120 mm Hg).

s with each of the main types of vascular disease. At all ages, there were steeper associations between blood pressure and intracerebral haemorrhage than ischaemic stroke. We estimated that about half of all premature vascular deaths were attributable to elevated blood pressure (ie, systolic blood pressure >120 mm Hg). Our study showed stronger associations between blood pressure and vascular disease than have previous prospective studies in China, largely because those studies did not correct for within-person variability in blood pressure (regression dilution bias).21, 22, 23 Compared with previous large meta-analyses of prospective studies that corrected for regression dilution (the Prospective Studies Collaboration4 and the Asia Pacific Cohort Studies Collaboration5), the findings are consistent for ischaemic heart disease but differ somewhat for stroke types. By contrast with these meta-analyses, our study found strong evidence that, throughout middle age and into old age, there were steeper associations for intracerebral haemorrhage than ischaemic stroke.

Studies Collaboration5), the findings are consistent for ischaemic heart disease but differ somewhat for stroke types. By contrast with these meta-analyses, our study found strong evidence that, throughout middle age and into old age, there were steeper associations for intracerebral haemorrhage than ischaemic stroke. These differences might, in part, reflect the high rate of brain imaging in this study. The use of brain imaging for diagnosis of stroke types was less common before the mid-1990s and, hence, it is possible that there was substantial misclassification of stroke types in some of the older studies included in these meta-analyses. By contrast, brain imaging was used to diagnose stroke types in about 90% of stroke cases in the CKB study. The greater use of brain imaging might also have allowed more non-fatal strokes to be identified than previously. There were shallower associations for non-fatal than fatal ischaemic stroke (unlike for intracerebral haemorrhage), and there is some evidence that both case fatality and the strength of associations with blood pressure might differ between the different aetiological subtypes of ischaemic stroke (large-artery atherosclerosis, small-vessel disease, cardioembolism, or other pathology).8, 24 Information on subtypes of ischaemic stroke is not yet available in the present study, but small-vessel (lacunar) infarcts, with low case fatality, might well be more common in Chinese than white populations.25

chaemic stroke (large-artery atherosclerosis, small-vessel disease, cardioembolism, or other pathology).8, 24 Information on subtypes of ischaemic stroke is not yet available in the present study, but small-vessel (lacunar) infarcts, with low case fatality, might well be more common in Chinese than white populations.25 Randomised controlled trials6, 26 of blood-pressure-lowering medication have shown the reversibility of some of the excess vascular risks associated with elevated blood pressure. A comparison of the findings from blood-pressure-lowering trials with the predicted effects of blood pressure from prospective cohort studies suggests that about 60% of the predicted risks of ischaemic heart disease and 80% of the predicted risks of stroke were reversed within 4–5 years of initiating treatment.26 Furthermore, the trial results indicate that the proportional effects of blood pressure on vascular risk were similar among people with and without cardiovascular disease, so the absolute benefits are very much greater among those with existing disease.6 Similarly, the absolute benefits of blood pressure lowering are likely to be greater in older than younger adults, given the greater absolute risk of vascular disease at older age.

ar among people with and without cardiovascular disease, so the absolute benefits are very much greater among those with existing disease.6 Similarly, the absolute benefits of blood pressure lowering are likely to be greater in older than younger adults, given the greater absolute risk of vascular disease at older age. Guidelines on the management of hypertension generally recommend initiating treatment in adults with an average systolic blood pressure of 140 mm Hg or more, or diastolic blood pressure of 90 mm Hg or more, and some recent randomised trials, in selected populations, have found benefit in initiating treatment at lower levels.27, 28 However, the use of blood-pressure-lowering treatment in China is much lower than in western populations.29 Our previous analyses of the CKB study indicate that, consistent with nationally representative surveys in China,30, 31 about a third of adults had hypertension at baseline (defined as systolic blood pressure ≥140 mm Hg, or diastolic blood pressure ≥90 mm Hg, or receiving treatment for hypertension).32 Of those with hypertension, about a third were diagnosed; of those diagnosed, about half were treated; and, of those treated, about a third had their hypertension controlled. The result was that less than 5% of participants with hypertension achieved properly controlled blood pressure.

ng treatment for hypertension).32 Of those with hypertension, about a third were diagnosed; of those diagnosed, about half were treated; and, of those treated, about a third had their hypertension controlled. The result was that less than 5% of participants with hypertension achieved properly controlled blood pressure. Previous studies reported associations between blood pressure variability and risk of vascular disease independently of baseline systolic blood pressure,33, 34 but there were an insufficient number of repeat blood pressure measures to assess any such effects in this study. However, other studies have suggested that, after correction for regression dilution bias in blood pressure variability, the effects of any such variability in systolic blood pressure on risk of vascular diseases, independent of usual levels of systolic blood pressure, are likely to be modest.35

effects in this study. However, other studies have suggested that, after correction for regression dilution bias in blood pressure variability, the effects of any such variability in systolic blood pressure on risk of vascular diseases, independent of usual levels of systolic blood pressure, are likely to be modest.35 We observed significant positive associations between systolic blood pressure and chronic kidney disease and diabetes, but the causal relevance of these associations remains uncertain. Neither association is strongly supported by evidence from trials, and might well be accounted for by reverse causality or residual confounding. A large meta-analysis6 of randomised controlled trials of blood-pressure-lowering medications reported that lowering blood pressure had no significant effect on the incidence of renal failure (10 mm Hg lower systolic blood pressure was associated with RR 0·95 [95% CI 0·84–1·07]). However, renal failure might not be entirely consistent with chronic kidney disease as defined in this study (the level of renal impairment among incident chronic kidney disease cases was not available). For diabetes, there is some evidence from randomised trials that lower blood pressure is associated with a reduced incidence of diabetes, but the effects are limited to certain medications only (angiotensin-converting-enzyme inhibitors and angiotensin receptor blockers), indicating that the renin–angiotensin system might be causally related to diabetes rather than actual levels of blood pressure.36

sure is associated with a reduced incidence of diabetes, but the effects are limited to certain medications only (angiotensin-converting-enzyme inhibitors and angiotensin receptor blockers), indicating that the renin–angiotensin system might be causally related to diabetes rather than actual levels of blood pressure.36 The chief strengths of this study include the large number of disease events, high-quality measurements of blood pressure (including the training of technicians), and long-term follow-up of a wide range of clinically validated disease outcomes. The main findings support efforts to address the substantial burden of blood pressure on vascular disease in China, and indicate that even small reductions in mean blood pressure at a population level can have a substantial effect on disease risk: for example, 5 mm Hg lower usual blood pressure would prevent around 350 000 deaths per year among individuals younger than 80 years of age (and around 200 000 deaths at <70 years). The use of blood-pressure-lowering medication is low in China, and strategies to improve rates of awareness, detection, and treatment of individuals with hypertension are likely to be highly cost-effective, especially among those with existing vascular disease.37 In the absence of well developed primary care systems, population-based approaches are also required to address the major determinants of elevated blood pressure in China, including high sodium intake (eg, through salt reduction), high alcohol consumption, excess adiposity, lack of regular physical activity, and poor home heating in winter.38, 39, 40

primary care systems, population-based approaches are also required to address the major determinants of elevated blood pressure in China, including high sodium intake (eg, through salt reduction), high alcohol consumption, excess adiposity, lack of regular physical activity, and poor home heating in winter.38, 39, 40 In conclusion, among men and women in China, higher levels of blood pressure were continuously and positively associated with higher risks of major vascular disease, with no evidence of a threshold down to at least 120 mm Hg systolic blood pressure and 75 mm Hg diastolic blood pressure. Unlike studies in western populations, blood pressure was more strongly associated with intracerebral haemorrhage than ischaemic stroke. There was also evidence of positive associations with some non-vascular chronic diseases, but the causality of these associations remains unclear. It was estimated that about half of all vascular deaths in China were attributable to elevated blood pressure, accounting for roughly 1 million deaths (age <80 years) annually; given the strength of the associations between blood pressure and vascular disease, even small reductions in mean blood pressure at a population level would be expected to have a major impact on vascular morbidity and mortality. Supplementary Material Supplementary appendix

In conclusion, among men and women in China, higher levels of blood pressure were continuously and positively associated with higher risks of major vascular disease, with no evidence of a threshold down to at least 120 mm Hg systolic blood pressure and 75 mm Hg diastolic blood pressure. Unlike studies in western populations, blood pressure was more strongly associated with intracerebral haemorrhage than ischaemic stroke. There was also evidence of positive associations with some non-vascular chronic diseases, but the causality of these associations remains unclear. It was estimated that about half of all vascular deaths in China were attributable to elevated blood pressure, accounting for roughly 1 million deaths (age <80 years) annually; given the strength of the associations between blood pressure and vascular disease, even small reductions in mean blood pressure at a population level would be expected to have a major impact on vascular morbidity and mortality. Supplementary Material Supplementary appendix Acknowledgments The baseline survey and the first resurvey were supported by a research grant from the Kadoorie Charitable Foundation in Hong Kong. The long-term continuation of the project is supported by programme grants from the UK Wellcome Trust (088158/Z/09/Z, 104085/Z/14/Z), the Chinese Ministry of Science and Technology (2011BAI09B01, 2012–14), and the Chinese National Natural Science Foundation (81390541). The British Heart Foundation (BHF), UK Medical Research Council, and Cancer Research UK provide core funding to the Clinical Trial Service Unit, University of Oxford, UK. FB acknowledges support from the BHF Centre of Research Excellence, Oxford. The chief acknowledgment is to the participants, the project staff, and the China National Centre for Disease Control and Prevention (CDC) and its regional offices for access to death and disease registries. The Chinese National Health Insurance scheme provided electronic linkage to all hospital admission data.

Oxford. The chief acknowledgment is to the participants, the project staff, and the China National Centre for Disease Control and Prevention (CDC) and its regional offices for access to death and disease registries. The Chinese National Health Insurance scheme provided electronic linkage to all hospital admission data. Contributors BL, SL, RCl, ZC, and LL designed and planned the paper; BL, XLK, SL, RCl, ZC, and LL were involved in data analysis, interpretation, and reporting; BL, SL, RCl, ZC, and LL were involved in manuscript drafting; RCl, ZC, LL, RP, RCo, RGW, YG, and JC, as members of China Kadoorie Biobank steering committee, designed and supervised the overall conduct of the study and obtained funding. YG, YC, ZB, and ZC coordinated the data acquisition (for baseline surveys, resurveys, and long-term follow-up). All authors provided critical comments on the manuscript. Declaration of interests We declare no competing interests.

Introduction Stroke is a leading cause of death and permanent disability worldwide. Around three-quarters of the global burden of stroke deaths (approximately 6·5 million per year) and associated disability-adjusted life years (113 million) now occur in low-income and middle-income countries, including China.1 In recent decades, the incidence and mortality rates of first stroke cases have declined progressively in high-income countries, but have not changed or have increased in many low-income and middle-income countries.1 Stroke is the leading cause of death and disability in China, accounting for more than 3 million new cases and around 2 million deaths in 2013,2, 3 and a higher proportion of strokes are intracerebral haemorrhages in the Chinese population than in western populations.2, 3

n many low-income and middle-income countries.1 Stroke is the leading cause of death and disability in China, accounting for more than 3 million new cases and around 2 million deaths in 2013,2, 3 and a higher proportion of strokes are intracerebral haemorrhages in the Chinese population than in western populations.2, 3 Previous large, nationwide studies in China have provided reliable estimates for stroke prevalence,4 incidence,3 and mortality,1 and have assessed the role of major risk factors for stroke.5, 6 However, there is little reliable contemporary evidence in China (and many other low-income and middle-income countries) about the case-fatality rate or recurrent risk after first stroke of different types. Previous such studies have been constrained by small sample sizes,7, 8 a focus on particular stroke types or minor cases,7, 8, 9 short duration of follow-up,3 or restriction to hospital-based rather than community-based settings.10 As in most high-income countries, considerable improvements have been made in the diagnosis of stroke (ie, widespread use of brain imaging), the acute treatment of stroke (eg, thrombolysis and antiplatelet therapy), and long-term secondary prevention (eg, blood-pressure lowering, antiplatelet and lipid-lowering therapy) in the past few decades in China.11 However, considerable uncertainty persists about whether such advances have resulted in any substantial improvement in both the short-term and long-term prognosis after first stroke in the general population. Reliable contemporary evidence on the natural history of stroke types in low-income and middle-income countries is needed to guide the delivery of appropriate services for the treatment and prevention of stroke.

ial improvement in both the short-term and long-term prognosis after first stroke in the general population. Reliable contemporary evidence on the natural history of stroke types in low-income and middle-income countries is needed to guide the delivery of appropriate services for the treatment and prevention of stroke. Research in context Evidence before this study We searched PubMed from inception to Sept 1, 2019, for studies that had investigated the recurrence or mortality after stroke and major stroke types, using the search terms (“stroke” or “Ischemic stroke” or “haemorrhagic stroke” or “Chinese” and “recurrence” or “mortality or fatality”) in articles published in English. In a national representative survey on stroke incidence and mortality, recurrence was not reported and patients were only followed up for 1 year. Although considerable improvements have been made in the diagnosis and treatment of stroke, substantial uncertainty persists about whether such advances have resulted in any appreciable improvements in prognosis after a first stroke in China. Added value of this study

We searched PubMed from inception to Sept 1, 2019, for studies that had investigated the recurrence or mortality after stroke and major stroke types, using the search terms (“stroke” or “Ischemic stroke” or “haemorrhagic stroke” or “Chinese” and “recurrence” or “mortality or fatality”) in articles published in English. In a national representative survey on stroke incidence and mortality, recurrence was not reported and patients were only followed up for 1 year. Although considerable improvements have been made in the diagnosis and treatment of stroke, substantial uncertainty persists about whether such advances have resulted in any appreciable improvements in prognosis after a first stroke in China. Added value of this study In this prospective study of adults recruited into the China Kadoorie Biobank during 2004–08, 489 586 participants without previous cardiovascular disease events were followed up for 9 years to record cases of first incident and recurrent stroke and cause-specific mortality. Risks of death and recurrent stroke were estimated separately by stroke type. 45 732 participants had a first stroke, including 36 588 (80%) individuals who had an ischaemic stroke, 7440 (16%) who had an intracerebral haemorrhage, 702 (2%) who had a subarachnoid haemorrhage, and 1002 (2%) who had an unspecified stroke type. The prognosis after a first stroke was poor, with 28-day mortality of 3% after ischaemic stroke, 47% after cerebral haemorrhage, 19% after subarachnoid haemorrhage, and 24% after unspecified stroke, with higher 28-day mortality in rural than urban areas, particularly for intracerebral haemorrhage (52% vs 32%). Among stroke survivors at 28 days, 41% had recurrent stroke and 17% died by 5 years after first stroke.

, 47% after cerebral haemorrhage, 19% after subarachnoid haemorrhage, and 24% after unspecified stroke, with higher 28-day mortality in rural than urban areas, particularly for intracerebral haemorrhage (52% vs 32%). Among stroke survivors at 28 days, 41% had recurrent stroke and 17% died by 5 years after first stroke. Implications of all the available evidence Despite progress in the treatments for stroke worldwide, the short-term and long-term prognosis following stroke is still poor in middle-income countries such as China, highlighting the urgent need for further improvements in secondary prevention of stroke in these countries. We present relevant findings from a 9-year follow-up of a population-based study of 0·5 million adults from ten diverse geographical areas in China. The aims of this study were to assess the 28-day case-fatality rates after first stroke for the major pathological stroke types; to estimate the short-term and long-term risks of recurrent stroke and mortality after a first stroke event by stroke type; and to investigate the pathological types of recurrent stroke after a first stroke of different types.

ss the 28-day case-fatality rates after first stroke for the major pathological stroke types; to estimate the short-term and long-term risks of recurrent stroke and mortality after a first stroke event by stroke type; and to investigate the pathological types of recurrent stroke after a first stroke of different types. Methods Study design and participants Details of the design and methods used in the China Kadoorie Biobank (CKB) have been previously reported.12, 13 In brief, the 2004–08 baseline survey was conducted in ten geographical areas (five urban, five rural) of China, chosen from China's nationally representative Disease Surveillance Points. These ten areas were chosen to maximise regional and social diversity, differences in risk exposures, and differences in burden of major diseases, including stroke. In each area, all permanent residents aged 35–74 years without disability were invited to participate. 512 715 individuals (including a small number of individuals outside the target age range) agreed to participate in the study.

ferences in risk exposures, and differences in burden of major diseases, including stroke. In each area, all permanent residents aged 35–74 years without disability were invited to participate. 512 715 individuals (including a small number of individuals outside the target age range) agreed to participate in the study. Trained health workers administered laptop-based questionnaires at local study clinics, which collected detailed information on sociodemographic status, smoking, alcohol consumption, diet, physical activity, and medical history. All participants also had several clinical measurements recorded, including height, weight, waist and hip circumference, lung function, blood pressure, and heart rate, and had a blood sample collected for long-term storage. Previous international, national and local ethics approval was obtained and all participants provided written informed consent to participate in the study and for their health status to be monitored by linkage with their electronic health records.

and heart rate, and had a blood sample collected for long-term storage. Previous international, national and local ethics approval was obtained and all participants provided written informed consent to participate in the study and for their health status to be monitored by linkage with their electronic health records. Procedures and outcomes After the baseline survey, the vital status of participants was monitored regularly through Disease Surveillance Point death registries, supplemented by annual checks with local residential records and active confirmation by contacting local street committees or village administrators.13 The causes of death, usually available from official death certificates, were supplemented by available medical records. In the small proportion of deaths (<5%) occurring without recent medical attention, standard procedures (verbal autopsy) were used to determine the probable causes of death based on symptoms and signs provided by informants (usually family members).14

death certificates, were supplemented by available medical records. In the small proportion of deaths (<5%) occurring without recent medical attention, standard procedures (verbal autopsy) were used to determine the probable causes of death based on symptoms and signs provided by informants (usually family members).14 All hospitalised cases of stroke (and other diseases) were identified by electronic linkage, via a unique personal identification number, to established registries of major diseases (ie, stroke, ischaemic heart disease, cancer, and diabetes) and to local health insurance claims databases for any hospital admissions, which covered more than 97% of study participants (these procedures also yielded a few additional deaths that had not been identified through death registries). To minimise bias due to loss to follow-up, any uninsured participants were also actively followed up by annual home visits in addition to annual checking of death and disease registries for any hospital admissions with stroke and other diseases and deaths in such participants.

been identified through death registries). To minimise bias due to loss to follow-up, any uninsured participants were also actively followed up by annual home visits in addition to annual checking of death and disease registries for any hospital admissions with stroke and other diseases and deaths in such participants. All fatal and non-fatal stroke cases reported by different sources were coded by trained medical staff, who were blinded to other personal information, using the International Classification of Diseases 10th revision (ICD-10), with further checking and review done centrally by a clinical research fellow. Any hospital-reported cases of first stroke also underwent separate clinical adjudication, involving retrieval and review of original medical records and brain imaging reports by clinical specialists in China using a bespoke web-based system.

ith further checking and review done centrally by a clinical research fellow. Any hospital-reported cases of first stroke also underwent separate clinical adjudication, involving retrieval and review of original medical records and brain imaging reports by clinical specialists in China using a bespoke web-based system. First incident and recurrent strokes For first incident stroke, the major pathological types examined were ischaemic stroke (ICD-10 code I63), including lacunar infarction and non-lacunar infarction; intracerebral haemorrhage (I61); subarachnoid haemorrhage (I60); and unspecified stroke (I64). Ischaemic stroke was defined as a focal neurological dysfunction lasting for more than 24 h with or without neuroimaging evidence of a cerebral infarct. Lacunar infarction was defined as stroke with neuroimaging evidence of an infarct in the brain of less than 1·5 cm in diameter on CT or MRI reports, with or without focal neurological deficits, including lacunar syndrome. Non-lacunar infarction was any other type of ischaemic stroke, excluding lacunar infarction. Intracerebral haemorrhage was defined as neurological dysfunction caused by haemorrhage into the brain parenchyma or the ventricular system, excluding those induced by injury, with or without neuroimaging evidence of brain haemorrhage. Subarachnoid haemorrhage was defined as neurological dysfunction considered to be caused by haemorrhage into the subarachnoid space, excluding those induced by injury, with or without relevant neuroimaging evidence. 92% of the reported first stroke cases had their diagnosis confirmed by brain imaging (CT or MRI). Radiological reports (but not the brain imaging) of individuals with non-fatal stroke were reviewed by Chinese neurologists online.

space, excluding those induced by injury, with or without relevant neuroimaging evidence. 92% of the reported first stroke cases had their diagnosis confirmed by brain imaging (CT or MRI). Radiological reports (but not the brain imaging) of individuals with non-fatal stroke were reviewed by Chinese neurologists online. The primary outcome measures were 28-day mortality, recurrent stroke, major vascular events, vascular mortality, and all-cause mortality. Recurrent stroke was defined as any hospitalised stroke events (ICD-10 I60, I61, I63, I64) occurring more than 24 h after the onset of a known first stroke. Major vascular events were defined as recurrent stroke, myocardial infarction (I21), or vascular death (I00–99), whichever was reported first. Statistical analysis The present analysis included all participants who had a stroke between enrolment and Jan 1, 2017, without a previous history of stroke or transient ischaemic attack, or ischaemic heart disease at baseline. The incidences of first stroke were standardised by sex, area, and age group to the overall CKB population for each decade of age at the time of diagnosis, excluding those with previous ischaemic heart disease or a previous stroke or transient ischaemic attack and are presented as incidence per 100 000 person-years. 28-day mortality was calculated as the proportion of participants who died within 28 days after a first incident stroke.

cade of age at the time of diagnosis, excluding those with previous ischaemic heart disease or a previous stroke or transient ischaemic attack and are presented as incidence per 100 000 person-years. 28-day mortality was calculated as the proportion of participants who died within 28 days after a first incident stroke. Cumulative mortality after first incident stroke was calculated as one minus Kaplan-Meier survival probability. The cumulative event rates (95% CIs) for recurrent stroke and other cardiovascular disease outcomes were calculated by the cumulative incidence function, treating death from any cause as a competing risk. The cumulative incidence of competing risks was estimated using the cmprsk package in R and the corresponding 95% CIs were based on the log-minus-log transformation.15, 16 The event rates were estimated for all participants who had a first incident stroke, and separately by stroke type. Event rates were also calculated separately by sex and area (urban or rural). In analyses of multiple events (vascular or non-vascular deaths; recurrent strokes classified by type; myocardial infarction, angina, or other cardiac diseases), different event types were treated as competing risks in the analyses. Unless otherwise specified, recurrent cases of stroke, major vascular events, or deaths were restricted to those occurring more than 28 days after a first stroke event. All analyses were done using R 3.5.0.

ction, angina, or other cardiac diseases), different event types were treated as competing risks in the analyses. Unless otherwise specified, recurrent cases of stroke, major vascular events, or deaths were restricted to those occurring more than 28 days after a first stroke event. All analyses were done using R 3.5.0. Role of the funding source The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. YC, RCl, and ZC had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results By Jan 1, 2017 (the censoring date for the present analyses), 44 066 (8·6%) participants had died and 4751 (<1%) were lost to follow-up, mostly because participants moved out of the study area. Of 512 715 individuals in the CKB, 23 129 had a previous history of stroke, transient ischaemic attack, or ischaemic heart disease at baseline and were excluded from these analyses. 489 586 participants were included, of whom 45 732 had a stroke during the study period and 443 854 had no stroke (table 1).Table 1 Baseline characteristics of participants by type of stroke

ious history of stroke, transient ischaemic attack, or ischaemic heart disease at baseline and were excluded from these analyses. 489 586 participants were included, of whom 45 732 had a stroke during the study period and 443 854 had no stroke (table 1).Table 1 Baseline characteristics of participants by type of stroke Stroke type Any stroke (n=45 732) No stroke (n=443 854) IS (n=36 588) ICH (n=7440) SAH (n=702) Unspecified (n=1002) Age group, years 30–39 974 (2·7%) 285 (3·8%) 52 (7·4%) 36 (3·6%) 1347 (2·9%) 76 013 (17·1%) 40–49 5662 (15·5%) 1122 (15·1%) 148 (21·1%) 166 (16·6%) 7098 (15·5%) 143 628 (32·4%) 50–59 12 096 (33·1%) 2202 (29·6%) 259 (36·9%) 282 (28·1%) 14 839 (32·4%) 135 833 (30·6%) 60–69 12 122 (33·1%) 2467 (33·2%) 172 (24·5%) 323 (32·2%) 15 084 (33·0%) 67 199 (15·1%) 70–79 5734 (15·7%) 1364 (18·3%) 71 (10·1%) 195 (19·5%) 7364 (16·1%) 21 181 (4·8%) Age, years 59·3 (9·6) 59·6 (10·2) 56 (10·1) 59·6 (10·4) 59·3 (9·8) 50·8 (10·3) Sex Men 16 380 (44·8%) 3913 (52·6%) 265 (37·7%) 455 (45·4%) 21 013 (45·9%) 179 122 (40·4%) Women 20 208 (55·2%) 3527 (47·4%) 437 (62·3%) 547 (54·6%) 24 719 (54·1%) 264 732 (59·6%) Area Rural 17 606 (48·1%) 5618 (75·5%) 447 (63·7%) 530 (52·9%) 24 201 (52·9%) 253 985 (57·2%) Urban 18 982 (51·9%) 1822 (24·5%) 255 (36·3%) 472 (47·1%) 21 531 (47·1%) 189 869 (42·8%) Highest education Primary school or no formal education 20 079 (54·9%) 5317 (71·5%) 433 (61·7%) 649 (64·8%) 26 478 (57·9%) 222 431 (50·1%) Middle school or high school 13 938 (38·1%) 1931 (26·0%) 244 (34·8%) 317 (31·6%) 16 430 (35·9%) 196 457 (44·3%) College or university 2571 (7·0%) 192 (2·6%) 25 (3·6%) 36 (3·6%) 2824 (6·2%) 24 966 (5·6%) Annual household income, ¥ <9999 11 301 (30·9%) 3335 (44·8%) 209 (29·8%) 377 (37·6%) 15 222 (33·3%) 123 610 (27·8%) 10 000–19 999 11 822 (32·3%) 2095 (28·2%) 224 (31·9%) 280 (27·9%) 14 421 (31·5%) 126 910 (28·6%) 20 000–34 999 8000 (21·9%) 1291 (17·4%) 159 (22·6%) 199 (19·9%) 9649 (21·1%) 111 274 (25·1%) ≥35 000 5465 (14·9%) 719 (9·7%) 110 (15·7%) 146 (14·6%) 6440 (14·1%) 82 060 (18·5%) Current smoker Men 9272 (56·6%) 2414 (61·7%) 160 (60·4%) 254 (55·8%) 12 100 (57·6%) 111 930 (62·5%) Women 802 (4·0%) 133 (3·8%) 23 (5·3%) 20 (3·7%) 978 (4·0%) 5642 (2·1%) Current alcohol drinker Men 11 915 (72·7%) 2525 (64·5%) 182 (68·7%) 305 (67·0%) 14 927 (71·0%) 138 459 (77·3%) Women 7596 (37·6%) 982 (27·8%) 165 (37·8%) 175 (32·0%) 8918 (36·1%) 95 510 (36·1%) Regular dietary intake* Meat 28 811 (78·7%) 5427 (72·9%) 582 (82·9%) 793 (79·1%) 35 613 (77·9%) 369 906 (83·3

·0%) 5642 (2·1%) Current alcohol drinker Men 11 915 (72·7%) 2525 (64·5%) 182 (68·7%) 305 (67·0%) 14 927 (71·0%) 138 459 (77·3%) Women 7596 (37·6%) 982 (27·8%) 165 (37·8%) 175 (32·0%) 8918 (36·1%) 95 510 (36·1%) Regular dietary intake* Meat 28 811 (78·7%) 5427 (72·9%) 582 (82·9%) 793 (79·1%) 35 613 (77·9%) 369 906 (83·3 %) Fish 15 636 (42·7%) 2427 (32·6%) 311 (44·3%) 438 (43·7%) 18 812 (41·1%) 209 947 (47·3%) Dairy 9029 (24·7%) 859 (11·5%) 129 (18·4%) 181 (18·1%) 10 198 (22·3%) 86 145 (19·4%) Fruit 20 468 (55·9%) 3295 (44·3%) 415 (59·1%) 522 (52·1%) 24 700 (54·0%) 266 159 (60·0%) Prevalent disease† Diabetes 2583 (7·1%) 310 (4·2%) 20 (2·8%) 71 (7·1%) 2984 (6·5%) 10 329 (2·3%) Hypertension 7960 (21·8%) 1844 (24·8%) 117 (16·7%) 246 (24·6%) 10 167 (22·2%) 38 392 (8·6%) Systolic blood pressure, mm Hg 141·5 (24·2) 151·2 (26·6) 138·3 (22·6) 143 (25·7) 143 (24·9) 129·3 (20·2) Body-mass index (kg/m2) 24·3 (3·5) 23·3 (3·6) 23·7 (3·5) 23·8 (3·6) 24·1 (3·5) 23·5 (3·3) Random plasma glucose, mmol/L 6·7 (3·2) 6·5 (3·1) 6·2 (2·7) 6·6 (3·3) 6·7 (3·2) 6·0 (2·2) Data are mean (SD) or n (%). 23 129 participants with self-reported previous ischaemic heart disease, stroke, or transient ischaemic attack at baseline were excluded. IS=ischaemic stroke. ICH=intracerebral haemorrhage. SAH=subarachnoid haemorrhage. * 1–3 days per week or more often. † Self-reported at baseline survey.

%) Fish 15 636 (42·7%) 2427 (32·6%) 311 (44·3%) 438 (43·7%) 18 812 (41·1%) 209 947 (47·3%) Dairy 9029 (24·7%) 859 (11·5%) 129 (18·4%) 181 (18·1%) 10 198 (22·3%) 86 145 (19·4%) Fruit 20 468 (55·9%) 3295 (44·3%) 415 (59·1%) 522 (52·1%) 24 700 (54·0%) 266 159 (60·0%) Prevalent disease† Diabetes 2583 (7·1%) 310 (4·2%) 20 (2·8%) 71 (7·1%) 2984 (6·5%) 10 329 (2·3%) Hypertension 7960 (21·8%) 1844 (24·8%) 117 (16·7%) 246 (24·6%) 10 167 (22·2%) 38 392 (8·6%) Systolic blood pressure, mm Hg 141·5 (24·2) 151·2 (26·6) 138·3 (22·6) 143 (25·7) 143 (24·9) 129·3 (20·2) Body-mass index (kg/m2) 24·3 (3·5) 23·3 (3·6) 23·7 (3·5) 23·8 (3·6) 24·1 (3·5) 23·5 (3·3) Random plasma glucose, mmol/L 6·7 (3·2) 6·5 (3·1) 6·2 (2·7) 6·6 (3·3) 6·7 (3·2) 6·0 (2·2) Data are mean (SD) or n (%). 23 129 participants with self-reported previous ischaemic heart disease, stroke, or transient ischaemic attack at baseline were excluded. IS=ischaemic stroke. ICH=intracerebral haemorrhage. SAH=subarachnoid haemorrhage. * 1–3 days per week or more often. † Self-reported at baseline survey. The mean age at baseline was 59·3 years (SD 9·8) for participants who had a stroke during the study period and 50·8 years (10·3) for participants who had no stroke. 54·1% with stroke and 59·6% without stroke were women, and 47·1% with stroke and 42·8% without stroke resided in urban areas (table 1). During 4·72 million person-years of follow-up, 45 732 participants had a first incident stroke, including 36 588 (80%) ischaemic strokes, 7440 (16%) intracerebral haemorrhages, 702 (2%) subarachnoid haemorrhages, and 1002 (2%) unspecified stroke types (table 1). 42 073 (92%) of 45 732 stroke diagnoses were confirmed by brain imaging using CT or MRI.

s of follow-up, 45 732 participants had a first incident stroke, including 36 588 (80%) ischaemic strokes, 7440 (16%) intracerebral haemorrhages, 702 (2%) subarachnoid haemorrhages, and 1002 (2%) unspecified stroke types (table 1). 42 073 (92%) of 45 732 stroke diagnoses were confirmed by brain imaging using CT or MRI. The standardised event rates were 778 (95% CI 107–113) per 100 000 person-years for ischaemic stroke (110 [107–113] lacunar infarction, 668 [661–676] non-lacunar infarction), 156 (153–160) for intracerebral haemorrhage, 15 (14–16) for subarachnoid haemorrhage, and 21 (20–22) for unspecified stroke types (appendix p 4). The event rate of ischaemic stroke was higher in those living in urban areas rather than rural areas (appendix p 4). In contrast, the event rate of intracerebral haemorrhage were higher in those living in rural areas rather than urban areas (appendix p 4).

ge, and 21 (20–22) for unspecified stroke types (appendix p 4). The event rate of ischaemic stroke was higher in those living in urban areas rather than rural areas (appendix p 4). In contrast, the event rate of intracerebral haemorrhage were higher in those living in rural areas rather than urban areas (appendix p 4). Compared with those who had no stroke, individuals who developed a first stroke during the study period tended to be older, male, to reside in urban areas, and to have lower levels of education and household income at baseline (table 1). Compared with ischaemic stroke, individuals who had intracerebral haemorrhage were more likely to be men, to live in rural areas, and to have lower levels of education and income. Individuals who had intracerebral haemorrhage also tended not to regularly consume meat, fish, fruit, and dairy products, and tended to be current smokers and have higher mean systolic blood pressure than individuals who had ischaemic stroke. However, the two groups had a similar mean body-mass index and random plasma glucose concentration. The baseline characteristics of individuals who had subarachnoid haemorrhage were generally similar to those of individuals who had intracerebral haemorrhage, and those of individuals who had unspecified stroke were similar to those of individuals who had ischaemic stroke (table 1).

random plasma glucose concentration. The baseline characteristics of individuals who had subarachnoid haemorrhage were generally similar to those of individuals who had intracerebral haemorrhage, and those of individuals who had unspecified stroke were similar to those of individuals who had ischaemic stroke (table 1). 28-day mortality of individuals with first stroke was 11% (95% CI 11–11; figure 1), with higher age-specific mortality in men than women (13% [12–13] vs 10% [9–10]; appendix p 13), and in rural areas than urban areas (16% [15–16] vs 6% [6–6]; appendix p 13). Mortality increased with age, particularly for intracerebral haemorrhage (figure 1). 28-day mortality was highest for intracerebral haemorrhage (47% [46–48]) and lowest for ischaemic stroke (3% [3–4]). For ischaemic stroke, 28-day mortality was higher for non-lacunar infarction than for lacunar infarction (4% [3–4] vs 0·4% [0·2–0·6]; data not shown).Figure 1 Age-specific 28-day mortality after a first stroke of different types Mortality was calculated as the proportion of participants dying from any cause within 28 days after a first stroke. Whiskers indicate 95% CIs. IS=ischaemic stroke. ICH=intracerebral haemorrhagic stroke. SAH=subarachnoid haemorrhage.

28-day mortality of individuals with first stroke was 11% (95% CI 11–11; figure 1), with higher age-specific mortality in men than women (13% [12–13] vs 10% [9–10]; appendix p 13), and in rural areas than urban areas (16% [15–16] vs 6% [6–6]; appendix p 13). Mortality increased with age, particularly for intracerebral haemorrhage (figure 1). 28-day mortality was highest for intracerebral haemorrhage (47% [46–48]) and lowest for ischaemic stroke (3% [3–4]). For ischaemic stroke, 28-day mortality was higher for non-lacunar infarction than for lacunar infarction (4% [3–4] vs 0·4% [0·2–0·6]; data not shown).Figure 1 Age-specific 28-day mortality after a first stroke of different types Mortality was calculated as the proportion of participants dying from any cause within 28 days after a first stroke. Whiskers indicate 95% CIs. IS=ischaemic stroke. ICH=intracerebral haemorrhagic stroke. SAH=subarachnoid haemorrhage. Among those who survived a first stroke, the cumulative recurrence rates from 28 days after the first stroke were 17% (95% CI 16–17) at 1 year, 41% (41–42) at 5 years, and 53% (52–54) at 9 years (table 2). All-cause mortality (predominantly from vascular causes) was 4% (4–4) at 1 year, 17% (17–18) at 5 years, and 30% (30–31) at 9 years. Individuals with intracerebral haemorrhage had higher post-28-day mortality than any other stroke type, with mortality rates of 11% (10–12) at 1 year, 28% (26–29) at 5 years, and 41% (39–44) at 9 years (appendix p 6), compared with 4% (3–4) at 1 year, 16% (15–16) at 5 years, and 29% (28–30) at 9 years for those with ischaemic stroke (appendix p 5). At 5 years, the all-cause mortality for those with intracerebral haemorrhage (28% [26–29]) was almost twice as high as that for other stroke types (figure 2). Stroke recurrence rates at 5 years were similar for those with intracerebral haemorrhage (44% [42–46]), ischaemic stroke (41% [41–42]), and unspecified stroke type (40% [35–44]), but were lower for those with subarachnoid haemorrhage (22% [18–27]). Mortality at 5 years was almost twice as high for individuals with non-lacunar infarction compared with lacunar infarction (17% [16–17] vs 9% [8–10]), and the stroke recurrence rates were similar (41% [40–42] vs 45% [43–47]; figure 3).Table 2 Cumulative event rate of recurrent stroke, major vascular events, vascular mortality, and all-cause mortality (n=45 732)

gh for individuals with non-lacunar infarction compared with lacunar infarction (17% [16–17] vs 9% [8–10]), and the stroke recurrence rates were similar (41% [40–42] vs 45% [43–47]; figure 3).Table 2 Cumulative event rate of recurrent stroke, major vascular events, vascular mortality, and all-cause mortality (n=45 732) 28 days 1 year 2 years 3 years 4 years 5 years 6 years 7 years 8 years 9 years Recurrent stroke Events 0 5616 8140 9685 10 623 11 298 11 722 11 995 12 161 12 230 No events or death 35 654 24 916 18 535 13 723 9953 7008 4799 2937 1565 668 Deaths 0 655 1067 1391 1634 1786 1910 2002 2064 2077 Censored 0 4467 7912 10 855 13 444 15 562 17 223 18 720 19 864 20 679 Cumulative event rate, % (95% CI) 0 17% (16–17) 26% (25–26) 32% (31–33) 37% (36–37) 41% (41–42) 45% (44–45) 48% (47–49) 51% (50–52) 53% (52–54) Major vascular event* Events 0 5963 8731 10 461 11 552 12 324 12 816 13 148 13 346 13 426 No events or death 35 605 24 839 18 448 13 646 9872 6933 4738 2894 1547 660 Deaths 0 347 548 698 807 877 937 975 1001 1005 Censored 0 4456 7878 10 800 13 374 15 471 17 114 18 588 19 711 20 514 Cumulative event rate, % (95% CI) 0 18% (17–18) 27% (27–28) 35% (34–35) 40% (40–41) 45% (45–46) 49% (49–50) 53% (53–54) 57% (56–58) 60% (59–61) Vascular mortality Events 0 1208 1972 2597 3113 3545 3885 4105 4284 4362 No events or death 40 370 33 334 27 525 22 097 17 001 12 562 9078 5984 3321 1368 Deaths 0 426 729 962 1168 1298 1416 1496 1550 1569 Censored 0 5402 10 144 14 714 19 088 22 965 25 991 28 785 31 215 33 071 Cumulative event rate, % (95% CI) 0 3% (3–3) 6% (5–6) 8% (8–8) 10% (10–10) 13% (12–13) 15% (15–16) 17% (17–18) 20% (20–21) 23% (22–24) All-cause mortality Events 0 1634 2701 3559 4281 4843 5301 5601 5834 5931 No events or death 40 370 33 334 27 525 22 097 17 001 12 562 9078 5984 3321 1368 Censored 0 5402 10 144 14 714 19 088 22 965 25 991 28 785 31 215 33 071 Cumulative event rate, % (95% CI) 0 4% (4–4) 8% (7–8) 11% (10–11) 14% (14–14) 17% (17–18) 21% (20–21) 24% (23–24) 27% (27–28) 30% (30–31) Time-points are time since first stroke.

r death 40 370 33 334 27 525 22 097 17 001 12 562 9078 5984 3321 1368 Censored 0 5402 10 144 14 714 19 088 22 965 25 991 28 785 31 215 33 071 Cumulative event rate, % (95% CI) 0 4% (4–4) 8% (7–8) 11% (10–11) 14% (14–14) 17% (17–18) 21% (20–21) 24% (23–24) 27% (27–28) 30% (30–31) Time-points are time since first stroke. * Stroke, myocardial infarction, and vascular mortality. Figure 2 Estimated cumulative event rates of recurrent stroke, major vascular events, vascular mortality, and all-cause mortality from 28 days after first stroke of different types Plotted lines indicate the cumulative incidence, starting at the date of first stroke. Whiskers indicate 95% CIs. Deaths from any cause were treated as competing risks. Participants experiencing an event or death within 28 days from first stroke were excluded. IS=ischaemic stroke. ICH=intracerebral haemorrhagic stroke. SAH=subarachnoid haemorrhage. Figure 3 Estimated cumulative event rates of recurrent stroke, major vascular events, vascular mortality, and all-cause mortality from 28 days after first ischaemic stroke of different subtypes Plotted lines show the cumulative incidence, starting with the date of first stroke. Deaths from any cause were treated as competing risks. Participants experiencing an event or death within 28 days following a first stroke were excluded. Only participants whose first stroke was ischaemic stroke were included. LACI=lacunar infarction.

ow the cumulative incidence, starting with the date of first stroke. Deaths from any cause were treated as competing risks. Participants experiencing an event or death within 28 days following a first stroke were excluded. Only participants whose first stroke was ischaemic stroke were included. LACI=lacunar infarction. For all types of first stroke, the cumulative recurrent stroke rate at 5 years was higher in men than women (except for subarachnoid haemorrhage) and in urban than rural areas (appendix p 14). All-cause mortality at 5 years was higher in men than women (except for subarachnoid haemorrhage), but lower in urban than rural areas for all stroke types (appendix p 15). After a first ischaemic stroke, 91% of recurrent strokes were ischaemic stroke (urban 94%, rural 86%), 7% were intracerebral haemorrhage (urban 3%, rural 11%), and 2% were unspecified stroke type (urban 2%, rural 2%; appendix p 10). After a first intracerebral haemorrhage, 56% of recurrent strokes were intracerebral haemorrhage (urban 43%, rural 62%), 41% were ischaemic stroke (urban 53%, rural 35%), and 3% were unspecified stroke type (urban 3%, rural 3%). After a first subarachnoid haemorrhage, 43% of recurrent strokes were ischaemic stroke, 29% were intracerebral haemorrhage, and 27% were subarachnoid haemorrhage. After a stroke of unspecified type, 69% of recurrent strokes were ischaemic stroke, 22% were unspecified stroke, and 9% were intracerebral haemorrhage (appendix p 10).

rst subarachnoid haemorrhage, 43% of recurrent strokes were ischaemic stroke, 29% were intracerebral haemorrhage, and 27% were subarachnoid haemorrhage. After a stroke of unspecified type, 69% of recurrent strokes were ischaemic stroke, 22% were unspecified stroke, and 9% were intracerebral haemorrhage (appendix p 10). After a first non-lacunar infarction, 80% of recurrent strokes were non-lacunar infarction (urban 78%, rural 82%), 10% were lacunar infarction, and 7% were intracerebral haemorrhage (appendix p 10). After a lacunar infarction, 47% of recurrent strokes were lacunar infarction (urban 50%, rural 23%), 48% were non-lacunar infarction (urban 47%, rural 64%), and 3% were intracerebral haemorrhage (urban 2%, rural 13%). Compared with recurrent stroke (41%), the event rate of myocardial infarction at 5 years after first stroke was much lower, at about 2% (appendix p 16). The 5-year event rate for non-myocardial-infarction cardiac events was 15% after ischaemic stroke and 7% after intracerebral haemorrhage. Among individuals who survived a first stroke at 28 days, the cumulative event rates for major vascular events (ie, non-fatal recurrent stroke, non-fatal myocardial infarction, or vascular death) were 18% at 1 year, 45% at 5 years, and 60% at 9 years (table 2). After combining both short-term (≤28 days) and long-term (>28 days) risks, the cumulative mortality after a first stroke was 15% at 1 year (ischaemic stroke 7%, intracerebral haemorrhage 53%) and 26% at 5 years (ischaemic stroke 19%, intracerebral haemorrhage 62%; appendix pp 8, 9).

years, and 60% at 9 years (table 2). After combining both short-term (≤28 days) and long-term (>28 days) risks, the cumulative mortality after a first stroke was 15% at 1 year (ischaemic stroke 7%, intracerebral haemorrhage 53%) and 26% at 5 years (ischaemic stroke 19%, intracerebral haemorrhage 62%; appendix pp 8, 9). After excluding individuals with stroke as secondary diagnosis, analyses restricted to 21 817 individuals with an adjudicated diagnosis of a non-fatal first stroke event as their main diagnosis showed similar 5-year cumulative event rates for recurrent stroke, major vascular events, vascular and all-cause mortality (appendix p 11). Further sensitivity analyses excluding individuals with recurrent stroke events who also had comorbidities only slightly reduced the overall 5-year recurrence rates (from 41% to 35%; appendix p 12).

ear cumulative event rates for recurrent stroke, major vascular events, vascular and all-cause mortality (appendix p 11). Further sensitivity analyses excluding individuals with recurrent stroke events who also had comorbidities only slightly reduced the overall 5-year recurrence rates (from 41% to 35%; appendix p 12). Discussion In this large community-based study in China, the incidence of stroke in the adult population was high, with around 9% of included participants experiencing a stroke during 9-year follow-up. Prognosis was also notably poor, with one in ten individuals dying within 28 days after onset of a first stroke, almost one in five dying by 5 years and more than 40% experiencing a recurrent stroke within 5 years of a first stroke. Among stroke types, the short-term and long-term prognosis for individuals with intracerebral haemorrhage was substantially worse than for those with ischaemic stroke, particularly in rural areas. After a first ischaemic stroke, although the 28-day mortality was low, the long-term risks of recurrent stroke and mortality were high. After a first ischaemic stroke, most of the incidences of recurrent stroke were also ischaemic stroke, whereas after a first intracerebral haemorrhage, a higher proportion of recurrent strokes were ischaemic stroke than were intracerebral haemorrhage.

ow, the long-term risks of recurrent stroke and mortality were high. After a first ischaemic stroke, most of the incidences of recurrent stroke were also ischaemic stroke, whereas after a first intracerebral haemorrhage, a higher proportion of recurrent strokes were ischaemic stroke than were intracerebral haemorrhage. The event rate of first stroke in the present study population was considerably higher than the incidence reported in a recent national representative survey (345·1 per 100 000 person-years).3 The discrepant results in the present study might reflect the older mean age of the CKB study participants or the inclusion of areas with a high incidence of stroke (eg, Harbin and Hunan).11 Because the present study was not designed to be nationally representative, it is not appropriate to make direct comparisons of stroke incidence in CKB with those in representative national surveys. Nevertheless, the observed north-to-south gradients and rural-urban differences in stroke incidence observed in CKB were generally consistent with those reported in the most recent representative national survey.3

riate to make direct comparisons of stroke incidence in CKB with those in representative national surveys. Nevertheless, the observed north-to-south gradients and rural-urban differences in stroke incidence observed in CKB were generally consistent with those reported in the most recent representative national survey.3 In the present study, 28-day mortality of intracerebral haemorrhage was considerably higher than that of ischaemic stroke (47% vs 3%), which is expected and could reflect differences in both pathology and effects of treatment.17 A systematic review of 8145 individuals with intracerebral haemorrhage from 36 studies in 21 countries also reported a similar 28-day mortality of 40%,18 and a previous study involving 16 031 Chinese adults who were enrolled between 1991 and 2000 reported 28-day mortality of 49·4% for intracerebral haemorrhage.19 Similar studies in western populations have reported lower 28-day mortality of intracerebral haemorrhage (eg, 29·6% in France20 and 30·7% in Sweden21). The higher mortality of intracerebral haemorrhage in China might reflect differences in stroke severity (due to poor detection and control of hypertension in China),11, 22 or delays in access to hospital care (including access to neurosurgery), particularly in rural areas.11 Similarly, differences in lifestyle factors might also account for the higher mortality observed in rural areas than urban areas in CKB. Unexpectedly, the present study showed lower mortality for subarachnoid haemorrhage than for intracerebral haemorrhage. Because subarachnoid haemorrhage is relatively rare among stroke types, the possibility of misclassification of subarachnoid haemorrhage as intracerebral haemorrhage for some individuals cannot be entirely excluded. The lower mortality of subarachnoid haemorrhage in Asian countries (Japan and China) than in other regions was also reported in a recent meta-analysis of subarachnoid haemorrhage involving 33 studies in 19 countries recruited after 1995.23 In contrast with intracerebral haemorrhage, the 28-day mortality for ischaemic stroke observed in the present study and in previous studies in China were considerably lower than those reported in western populations,21, 24, 25, 26 possibly reflecting differences in the incidence of ischaemic stroke subtypes (eg, a higher proportion of individuals with small vessel ischaemic stroke) or better detection of less severe stroke types.27, 28

in previous studies in China were considerably lower than those reported in western populations,21, 24, 25, 26 possibly reflecting differences in the incidence of ischaemic stroke subtypes (eg, a higher proportion of individuals with small vessel ischaemic stroke) or better detection of less severe stroke types.27, 28 Few previous studies in China have examined the long-term prognosis after the onset of first stroke. In the China National Stroke Registry study, which involved 11 560 individuals with ischaemic stroke recruited in 2007–08 from major hospitals, 1-year cumulative mortality after a first ischaemic stroke was twice as high as that reported in the present study (14% vs 7%),29 but the stroke recurrence rate at 1 year was comparable to that observed in the present study (17% vs 16%). Less evidence is available on the long-term prognosis after first intracerebral haemorrhage. In a meta-analysis of 2408 individuals with intracerebral haemorrhage from 122 studies in different countries,30 the overall cumulative mortality was 54% at 1 year and 71% at 5 years. The results of the present study were similar to those observed in a recent Swedish study, which reported mortality rates of 42% at 1 year and 62% at 5 years.21 A meta-analysis involving 1102 individuals with intracerebral haemorrhage enrolled between 1981 and 2007 reported higher stroke recurrence rates than the present study (46% vs 23%), but the differences might reflect improvements in health-care during the period between the studies.30

42% at 1 year and 62% at 5 years.21 A meta-analysis involving 1102 individuals with intracerebral haemorrhage enrolled between 1981 and 2007 reported higher stroke recurrence rates than the present study (46% vs 23%), but the differences might reflect improvements in health-care during the period between the studies.30 The poor long-term prognosis after a first stroke in China might also reflect the inadequate use of established medications (eg, antiplatelet therapy, blood pressure-lowering, and lipid-lowering treatments), as shown in a previous report from the CKB study population.22 Other community-based studies in China have reported that only 10% of individuals with ischaemic stroke used aspirin, 2% used LDL cholesterol-lowering therapy, and around 35% used antihypertensive medication.11, 31 The scarce use of such treatments with proven efficacy might reflect inadequate reimbursement policies, poor access to primary health care services,32 or poor awareness among patients and their doctors of the need for long-term use of such treatments for stroke survivors.11, 33

% used antihypertensive medication.11, 31 The scarce use of such treatments with proven efficacy might reflect inadequate reimbursement policies, poor access to primary health care services,32 or poor awareness among patients and their doctors of the need for long-term use of such treatments for stroke survivors.11, 33 The present study showed that nearly all incidences of recurrent stroke after ischaemic stroke were also ischaemic stroke, which was consistent with previous studies done in western populations.34 In contrast with previous reports that suggested that lacunar infarction and non-lacunar infarction have a distinct pathology,28, 35 the present study showed that subsequent stroke can be both lacunar infarction and non-lacunar infarction after first lacunar infarction or non-lacunar infarction indicating overlapping causes underlying lacunar infarction and non-lacunar infarction. Few large studies have specifically examined the type of recurrent stroke events after intracerebral haemorrhage. A systematic review of ten studies, involving 1880 individuals with intracerebral haemorrhage recorded between 1982 and 2000, reported that 75% of the individuals who had recurrent strokes had intracerebral haemorrhage.36 By contrast, a meta-analysis involving 2408 individuals with intracerebral haemorrhage reported similar proportions of recurrent ischaemic stroke and recurrent intracerebral haemorrhage after first intracerebral haemorrhage.30 The present study, involving three-times as many individuals with intracerebral haemorrhage as the previous meta-analysis,30 also reported that more than 40% of recurrent strokes after intracerebral haemorrhage were ischaemic stroke.

troke and recurrent intracerebral haemorrhage after first intracerebral haemorrhage.30 The present study, involving three-times as many individuals with intracerebral haemorrhage as the previous meta-analysis,30 also reported that more than 40% of recurrent strokes after intracerebral haemorrhage were ischaemic stroke. Current guidelines recommend use of antiplatelet therapy and statins for ischaemic stroke, but not for intracerebral haemorrhage.37 However, most of the available evidence about the use of such treatment in intracerebral haemorrhage is based on studies done in western populations in which the incidence of intracerebral haemorrhage was low. Given the high incidence of intracerebral haemorrhage in China, large randomised trials are needed to assess the efficacy and safety of long-term antiplatelet and cholesterol-lowering therapy for secondary prevention of ischaemic stroke. A previous estimate suggested that the benefits of LDL cholesterol-lowering treatments on ischaemic stroke and ischaemic heart disease would exceed the adverse effects on intracerebral haemorrhage by more than five times in Chinese adults.38

and cholesterol-lowering therapy for secondary prevention of ischaemic stroke. A previous estimate suggested that the benefits of LDL cholesterol-lowering treatments on ischaemic stroke and ischaemic heart disease would exceed the adverse effects on intracerebral haemorrhage by more than five times in Chinese adults.38 This study had several strengths, including the prospective study design, the large number of participants, detailed data on stroke types, and complete long-term follow-up for both fatal and non-fatal incidences of stroke and other vascular diseases. Of note, more than 90% of the reported first strokes were confirmed by brain imaging. However, the study also had several limitations. Firstly, the CKB cohort was established mainly to investigate the determinants of common chronic diseases. Therefore, the study design focused on diversity rather than representativeness of study populations. As such, the findings may not be directly generalisable to the overall Chinese population. Secondly, recurrent strokes were defined as any reported stroke occurring more than 24 h after a first incident stroke that led to new hospitalisation.35 It is possible that some of the recurrent stroke cases might not be new stroke events, but were recorded as new cases for various reasons (including transfers between different hospitals). However, most of the recurrent strokes occurred many months or years after their initial event, frequently involving a different stroke pathological type, suggesting that they were unlikely to be readmissions for the same stroke or an evolving stroke after a first event. Thirdly, stroke survivors admitted to hospitals subsequently for reasons other than a new stroke could still have a stroke diagnosis recorded in their medical records, leading to false positive diagnoses for recurrent strokes. However, sensitivity analyses using recurrent stroke as the primary diagnosis or excluding those that also had comorbidities recorded at hospital admission did not notably alter the high rates of stroke recurrence and mortality. Nevertheless, the possibility of overestimation or underestimation of stroke recurrence after a first stroke cannot be entirely excluded and the verification of at least some recurrent stroke events in future studies would help to clarify this.

on did not notably alter the high rates of stroke recurrence and mortality. Nevertheless, the possibility of overestimation or underestimation of stroke recurrence after a first stroke cannot be entirely excluded and the verification of at least some recurrent stroke events in future studies would help to clarify this. This study showed a high incidence of stroke in Chinese adults, and a surprisingly poor long-term prognosis after a first stroke of any type, particularly for individuals living in rural areas. The findings highlight the need for improved primary prevention of stroke through lifestyle modification, and for urgent improvements to the delivery of effective secondary prevention treatments for stroke (eg, anti-platelet, lipid-lowering, and blood pressure-lowering medication) in both urban and rural areas in China and in other low-income and middle-income countries. Supplementary Material Supplementary appendix Contributors YC, LL, RCl, and ZC had full access to the data. All authors were involved in study design, conduct, long-term follow-up, analysis of data, interpretation, and writing of the report, and had final responsibility for the decision to submit for publication. Declaration of interests We declare no competing interests.

Contributors YC, LL, RCl, and ZC had full access to the data. All authors were involved in study design, conduct, long-term follow-up, analysis of data, interpretation, and writing of the report, and had final responsibility for the decision to submit for publication. Declaration of interests We declare no competing interests. Acknowledgments The baseline survey was supported by the Kadoorie Charitable Foundation, Hong Kong. Long-term follow-up was supported by grants from the UK Wellcome Trust (212946/Z/18/Z, 202922/Z/16/Z, 104085/Z/14/Z, 088158/Z/09/Z), National Natural Science Foundation of China (81390540, 81390541, 81390544), and the National Key Research and Development Program of China (2016YFC 0900500, 0900501, 0900504, 1303904). The British Heart Foundation, UK Medical Research Council, and Cancer Research UK provided core support to the Clinical Trial Service Unit. We thank the participants, the project staff, and the China National Centre for Disease Control and Prevention and its regional offices for access to death and disease registries. The Chinese National Health Insurance scheme provided electronic linkage to all hospitalised individuals.

Introduction China has seen a rapid demographic and epidemiological transition in recent decades, resulting in an increasing burden of chronic non-communicable diseases and escalating financial pressures on the health-care system.1 Cardiovascular diseases, chiefly stroke and ischaemic heart disease, are the leading causes of premature death and permanent disability in China, accounting for about 100 million prevalent cases and 4 million deaths in 2017.2 Due to the scarcity of effective primary care and limited implementation of primary and secondary prevention treatments for cardiovascular disease, hospitalisation (referring to admission to hospital) costs for cardiovascular disease have increased substantially in China.1, 3, 4

ent cases and 4 million deaths in 2017.2 Due to the scarcity of effective primary care and limited implementation of primary and secondary prevention treatments for cardiovascular disease, hospitalisation (referring to admission to hospital) costs for cardiovascular disease have increased substantially in China.1, 3, 4 In response to growing inequalities in health-care use and disease outcomes between urban and rural areas, new health insurance programmes were initiated in China in 2003.5 In 2009, the Chinese Government launched a major health-care reform aiming to provide affordable and equal access to high quality basic health services for all by 2020.6 The first cycle of this reform (2009–12) included an expansion of health insurance to the entire population and increase in benefits, strengthening of the primary health-care and public health services, and establishment of a national essential medicines system, while the second cycle (2013–16) focused more on improving the efficiency and quality of public hospital services.6 However, little is known about trends in health-care use and outcomes for affected individuals with stroke and ischaemic heart disease in urban and rural areas, and in different socioeconomic groups in the past decade.6, 7 Research in context Evidence before this study

In response to growing inequalities in health-care use and disease outcomes between urban and rural areas, new health insurance programmes were initiated in China in 2003.5 In 2009, the Chinese Government launched a major health-care reform aiming to provide affordable and equal access to high quality basic health services for all by 2020.6 The first cycle of this reform (2009–12) included an expansion of health insurance to the entire population and increase in benefits, strengthening of the primary health-care and public health services, and establishment of a national essential medicines system, while the second cycle (2013–16) focused more on improving the efficiency and quality of public hospital services.6 However, little is known about trends in health-care use and outcomes for affected individuals with stroke and ischaemic heart disease in urban and rural areas, and in different socioeconomic groups in the past decade.6, 7 Research in context Evidence before this study We searched for studies investigating trends in hospital use for cardiovascular diseases and any cause, by socioeconomic characteristics in the past decade in China, from Jan 1, 2007, to March 15, 2019, using MEDLINE and Embase. The search strategy included terms related to hospital admissions (“hospitalisation”, “admission”, and “inpatient”), performance measures (“utilisation”, “rate”, “length of stay”, “trends”, “case fatality”, “mortality”, “quality”, “efficiency”, “equity”), and socioeconomic characteristics influencing the use of care (“insurance”, “socioeconomic”, “education”, “occupation”, “income”, “urban”, and “rural”). Multiple possible synonyms and spellings of search terms were also used. We included studies that were based in mainland China and involved hospital admissions for stroke, ischaemic heart disease, or any cause in the general population. Most studies had a cross-sectional or retrospective design, focused on specific regions (eg, low-income regions), populations (urban or rural) or health insurance enrollees (mainly new rural cooperative medical scheme). These studies investigated the effect of health insurance on health-care use or measured inequalities in the use of inpatient or outpatient care for any cause. A few studies explored the disparities in health outcomes and health-care use across health insurance groups, but only among patients with acute myocardial infarction, heart failure, and intracranial haemorrhage. Studies that examined trends in hospital use for stroke and ischaemic heart disease did not include variations by socioeconomic characteristics. Previous studies using data from the 2003, 2008, and 2011 National Health Services Survey and medical records from 11 hospitals nationwide, reported increasing trends in hospitalisation rates for any cause before and after the 2009 health-care reforms. On the basis of data from over 100 tertiary hospitals in 2007–10, it was reported that the number of hospitalisations for stroke increased and in-hospital mortality decreased.

s from 11 hospitals nationwide, reported increasing trends in hospitalisation rates for any cause before and after the 2009 health-care reforms. On the basis of data from over 100 tertiary hospitals in 2007–10, it was reported that the number of hospitalisations for stroke increased and in-hospital mortality decreased. The China PEACE study reported increased admission rates for acute myocardial infarction, decreased mean length of stay, and stable adjusted in-hospital mortality in the period 2001–11. A study in Beijing, China, in 2007–12 reported that hospitalisation rates for acute myocardial infarction increased, but mean length of stay and in-hospital mortality decreased. Added value of this study

The China PEACE study reported increased admission rates for acute myocardial infarction, decreased mean length of stay, and stable adjusted in-hospital mortality in the period 2001–11. A study in Beijing, China, in 2007–12 reported that hospitalisation rates for acute myocardial infarction increased, but mean length of stay and in-hospital mortality decreased. Added value of this study This is the first large study, to our knowledge, to provide a comprehensive analysis of trends in hospital admissions for common cardiovascular diseases in China between 2009 and 2016, both overall and across different socioeconomic groups. In men and women aged 30–79 years at entry in the China Kadoorie Biobank, after adjusting for demographic, socioeconomic, lifestyle, and morbidity factors, the hospitalisation rates for stroke, ischaemic heart disease, and any cause increased by approximately 4–5% per year in this period. Individuals living in rural areas, with lower education or income levels, and enrolled in the urban or rural resident health insurance scheme had the greatest annual increases in rates of hospitalisation for stroke and ischaemic heart disease, and the greatest reductions in 28-day case fatality rates. The mean length of stay for stroke, ischaemic heart disease, and any cause decreased by about 2% each year and decreased to a greater extent in higher socioeconomic groups during 2009–16. Implications of all the available evidence

This is the first large study, to our knowledge, to provide a comprehensive analysis of trends in hospital admissions for common cardiovascular diseases in China between 2009 and 2016, both overall and across different socioeconomic groups. In men and women aged 30–79 years at entry in the China Kadoorie Biobank, after adjusting for demographic, socioeconomic, lifestyle, and morbidity factors, the hospitalisation rates for stroke, ischaemic heart disease, and any cause increased by approximately 4–5% per year in this period. Individuals living in rural areas, with lower education or income levels, and enrolled in the urban or rural resident health insurance scheme had the greatest annual increases in rates of hospitalisation for stroke and ischaemic heart disease, and the greatest reductions in 28-day case fatality rates. The mean length of stay for stroke, ischaemic heart disease, and any cause decreased by about 2% each year and decreased to a greater extent in higher socioeconomic groups during 2009–16. Implications of all the available evidence The available evidence showed substantial increases in hospital admissions and reductions in case fatality rates and mean length of stay for stroke and ischaemic heart disease in Chinese adults aged 30–90 years between 2009 and 2016. Although socioeconomic inequalities in hospital use persist, improvements in health-care use and health outcomes during this period have been greatest in rural areas and among lower socioeconomic groups. These results should inform strategies for health promotion and equitable distribution of health-care resources. Further improvements in prevention and management of cardiovascular disease are needed to reverse the trends in incidence and achieve further reductions in inequalities in hospital use and outcomes for stroke and ischaemic heart disease.

rategies for health promotion and equitable distribution of health-care resources. Further improvements in prevention and management of cardiovascular disease are needed to reverse the trends in incidence and achieve further reductions in inequalities in hospital use and outcomes for stroke and ischaemic heart disease. Previous studies of the burden of stroke and ischaemic heart disease in China have mainly focused on trends in incidence, prevalence, mortality, or disability-adjusted life-years.3, 8 Most of the available studies of the use of health-care services for stroke and ischaemic heart disease in China have been constrained by short duration of follow-up, restriction to individual hospitals, or inability to study the effects of person-level characteristics on disease outcomes.5, 9, 10, 11, 12, 13, 14, 15, 16 To address these limitations, the aims of the present study were to: (1) examine trends in annual rates of hospitalisations for stroke, ischaemic heart disease, and all causes combined together (any cause) between 2009 and 2016; (2) investigate differences in hospitalisation rates between different socioeconomic groups over this period; and (3) assess whether trends in case fatality rates and mean length of stay for stroke and ischaemic heart disease differ between socioeconomic groups.

uses combined together (any cause) between 2009 and 2016; (2) investigate differences in hospitalisation rates between different socioeconomic groups over this period; and (3) assess whether trends in case fatality rates and mean length of stay for stroke and ischaemic heart disease differ between socioeconomic groups. Methods Study design and participants We used data from the China Kadoorie Biobank (CKB), a prospective cohort study of adults, who were recruited from five urban (Qingdao, Harbin, Haikou, Suzhou, Liuzhou) and five rural (Gansu, Sichuan, Henan, Zhejiang, Hunan) areas in China between June 25, 2004, and July 15, 2008.17 Official local residential records were used to identify eligible individuals (adults aged 30–79 years). All participants completed an interviewer-administered electronic questionnaire, providing detailed information on demographic and socioeconomic characteristics, medical history, and lifestyle factors (appendix 2 p 5). In addition, physical measurements were recorded, and a blood sample was collected for long-term storage. Individuals who were alive on Jan 1, 2009 contributed to the present study. All participants in the CKB provided written informed consent for participation and long-term follow-up for health status by electronic linkage to relevant medical records. Prior international, national, and local ethical approvals were obtained. The protocol for the CKB has been previously published.17, 18

sent study. All participants in the CKB provided written informed consent for participation and long-term follow-up for health status by electronic linkage to relevant medical records. Prior international, national, and local ethical approvals were obtained. The protocol for the CKB has been previously published.17, 18 Procedures Data for all hospital admissions were obtained by ongoing linkage via unique national identification number, to electronic hospital records from the nationwide health insurance system (which has >98% coverage across the ten study areas), to registers of stroke and ischaemic heart disease established by CKB's regional centres, and to death registers monitored through China's Disease Surveillance Points and Information System (appendix 2 p 5). Diagnoses associated with each hospital admission were reviewed, integrated centrally and standardised using the International Classification of Diseases 10th revision (ICD-10) codes using a bespoke programme. Hospital admissions were identified using ICD-10 codes I60-I61 and I63-I64 for stroke, and I20-I25 for ischaemic heart disease. Overall, 92% of reported incident strokes had their diagnoses confirmed by CT or MRI and 97% of incident ischaemic heart disease cases had electrocardiogram reports. Hospital admissions included all first and subsequent admissions occurring in public and private hospitals across all hospital tiers.

aemic heart disease. Overall, 92% of reported incident strokes had their diagnoses confirmed by CT or MRI and 97% of incident ischaemic heart disease cases had electrocardiogram reports. Hospital admissions included all first and subsequent admissions occurring in public and private hospitals across all hospital tiers. Statistical analysis The study outcomes were the annual number of hospital admissions, 28-day case fatality, and length of stay of admissions for stroke, ischaemic heart disease, and any cause. Since recruitment to the CKB was completed on July 15, 2008, the present analyses were restricted to all hospital admissions occurring between Jan 1, 2009, and Dec 31, 2016. We defined the length of stay as the difference in days between the admission date and the discharge date of the hospitalisation or date of death, when relevant. Day case admissions (ie, participant admitted and discharged on the same day) were counted as having a length of stay of 0·5 days. Missing values for length of stay were imputed using multiple imputation (appendix 2 p 6). The 28-day case fatality was defined as death within 28 days from the first-ever event for stroke or ischaemic heart disease for participants without history of stroke, transient ischaemic attack, and ischaemic heart disease before 2009.

values for length of stay were imputed using multiple imputation (appendix 2 p 6). The 28-day case fatality was defined as death within 28 days from the first-ever event for stroke or ischaemic heart disease for participants without history of stroke, transient ischaemic attack, and ischaemic heart disease before 2009. Annual rates of hospital admissions were estimated using generalised linear models (GLM) with log-link function and negative binomial distribution. Annual 28-day case fatality rates were estimated using GLM with logit link function and binomial distribution. Annual mean length of stay per admission was estimated using GLM with log link function and gamma distribution (appendix 2 p 6).19

linear models (GLM) with log-link function and negative binomial distribution. Annual 28-day case fatality rates were estimated using GLM with logit link function and binomial distribution. Annual mean length of stay per admission was estimated using GLM with log link function and gamma distribution (appendix 2 p 6).19 We examined hospitalisation rates, case fatality rates, and mean length of stay for stroke, ischaemic heart disease, and any cause, overall and by the following factors: urban and rural area of residence, highest level of education attained (no formal school, primary or middle school, high school or above), annual household income in renminbi (<¥10 000, ¥10 000–19 999, ¥20 000–34 999, ≥¥35 000), and health insurance type (urban employee basic medical insurance [UEBMI], urban resident basic medical insurance [URBMI], or new rural cooperative medical scheme [NRCMS]: appendix 2 p 8). The annual rates of hospitalisation, 28-day case fatality, and mean length of stay, and their annual trends (percentage change per calendar year) were estimated with adjustment for annually updated age (ie, continuous age, age-squared, age-cubed), sex, and region (ie, minimally adjusted models). In further analyses, socioeconomic, lifestyle, and morbidity factors were also included (ie, fully adjusted models), to evaluate annual rates and trends most indicative of changes in the health-care system, including changes in the organisation and delivery of health care. Socioeconomic factors included marital status, household size, highest level of education attained, annual household income, and health insurance type. Lifestyle factors included smoking and alcohol consumption, body-mass index, physical activity assessed using metabolic equivalents of task, and self-rated health status. Classification of socioeconomic (except health insurance type) and lifestyle factors was restricted to information collected at entry into the study (2004–08). Morbidity factors included history of self-reported doctor-diagnosed diseases at entry into CKB, with previous history of major diseases (cerebrovascular disease, ischaemic heart disease, malignant neoplasms, respiratory diseases, infectious and parasitic diseases, diabetes, chronic kidney disease, and tuberculosis) updated annually at the start of each year from recruitment to Dec 31, 2016, using linked CKB data.

with previous history of major diseases (cerebrovascular disease, ischaemic heart disease, malignant neoplasms, respiratory diseases, infectious and parasitic diseases, diabetes, chronic kidney disease, and tuberculosis) updated annually at the start of each year from recruitment to Dec 31, 2016, using linked CKB data. Additional adjustments were made for stroke type (ischaemic, haemorrhagic, or unspecified) and ischaemic heart disease type (acute myocardial infarction or other ischaemic heart disease) in analyses of case fatality and mean length of stay.

with previous history of major diseases (cerebrovascular disease, ischaemic heart disease, malignant neoplasms, respiratory diseases, infectious and parasitic diseases, diabetes, chronic kidney disease, and tuberculosis) updated annually at the start of each year from recruitment to Dec 31, 2016, using linked CKB data. Additional adjustments were made for stroke type (ischaemic, haemorrhagic, or unspecified) and ischaemic heart disease type (acute myocardial infarction or other ischaemic heart disease) in analyses of case fatality and mean length of stay. Data for health insurance types for each participant were identified annually in 2012–16, but were unavailable for the years 2009–11; the missing data for health insurance type were imputed based on the insurance scheme in which participants were enrolled in 2012 (appendix 2 pp 6–7). Health insurance types were classified as UEBMI, URBMI, NRCMS, or uninsured. These health insurance schemes differed in their target population, administration, source of funding, and benefits, with UEBMI providing the most comprehensive coverage (appendix 2 p 8).6 In the analyses, URBMI and NRCMS were combined, as they provided similar benefits, and in four of the ten CKB regions, the two schemes merged into a single scheme from 2012–13 onwards. Participants who were uninsured were excluded from analyses by health insurance type because of the small number of cases. Models for trends by urban or rural areas did not include adjustments for health insurance type, because of substantial correlation between these two factors. Heterogeneity or, if relevant (ie, more than two ordered categories), trends in annual rates of hospitalisation, case fatalities, and mean length of stay between categories of participants by urban and rural areas, education, income, and health insurance type were assessed using χ2 tests for heterogeneity or trend. Standardised predicted admission rates (per 1000 person-years), case fatality rates (per 100 events), and mean length of stay (in days) for each calendar year of follow-up were derived using characteristics of all CKB study participants in 2009, except for models by urban and rural area. As each CKB region contributed data for only urban or only rural area, the estimates for urban and rural areas, in which adjustment for region were used, were standardised using the separate CKB populations in urban and rural areas in 2009. Annual absolute differences in rates of hospitalisation, case fatality rate, and mean length of stay between the two most extreme groups of each socioeconomic characteristic, and an annual slope index of inequality for each characteristic except urban and rural area, were also estimated, together with a test for linear trend.

absolute differences in rates of hospitalisation, case fatality rate, and mean length of stay between the two most extreme groups of each socioeconomic characteristic, and an annual slope index of inequality for each characteristic except urban and rural area, were also estimated, together with a test for linear trend. The slope index of inequality is a weighted measure of inequality that takes into account the entire distribution of each socioeconomic characteristic, rather than only comparing the two most extreme groups.20, 21 Cluster-adjusted robust SEs were estimated to control for absence of independence between hospital admissions of the same participant during follow-up (appendix 2 p 8). In sensitivity analyses, hospital admissions were restricted to first-ever admissions for stroke or ischaemic heart disease in the study, and the annual differences in hospitalisation rates, case fatality rates, and mean length of stay, and slope index of inequality between urban and rural area were standardised for the entire CKB study population in 2009, but without adjustment for region. In further analyses, trends in hospitalisation rates, case fatality rates, and mean length of stay were assessed during the two cycles of the health-care reforms (2009–12, 2013–16), by stroke and ischaemic heart disease type, and by age group (<60, 60–69, ≥70 years). All analyses were done using Stata 15 or R 3.6.0.

n. In further analyses, trends in hospitalisation rates, case fatality rates, and mean length of stay were assessed during the two cycles of the health-care reforms (2009–12, 2013–16), by stroke and ischaemic heart disease type, and by age group (<60, 60–69, ≥70 years). All analyses were done using Stata 15 or R 3.6.0. Role of the funding source The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. ML, YC, RC, ZC, and BM had full access to all the data in the study and had final responsibility for the decision to submit for publication.

n. In further analyses, trends in hospitalisation rates, case fatality rates, and mean length of stay were assessed during the two cycles of the health-care reforms (2009–12, 2013–16), by stroke and ischaemic heart disease type, and by age group (<60, 60–69, ≥70 years). All analyses were done using Stata 15 or R 3.6.0. Role of the funding source The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. ML, YC, RC, ZC, and BM had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results Enrolment in the CKB took place between June 25, 2004, and July 15, 2008. 1 801 167 eligible individuals were identified, and 512 715 (28%) of those invited responded and were enrolled. Among the 512 715 participants in the CKB, 6469 died and 251 were lost to follow-up before 2009, leaving 505 995 (99%) participants contributing data between Jan 1, 2009, and Dec 31, 2016, in the present analyses. Among the 505 995 participants, the mean age was 54·7 years (SD 10·6) in 2009 and 59·2% (299 793) were women (table). Socioeconomic characteristics varied substantially between the ten regions in China, particularly between the five urban and five rural areas (appendix 2 p 9). The proportion of participants who had completed high school education was higher in those living in urban (n=80 593, 36·0%) than in rural (n=26 164, 9·3%) areas, as was the proportion of participants with an annual household income of ¥35 000 or greater in urban (n=56 974, 25·4%) compared with rural areas (n=34 758, 12·3%). In 2012, 280 967 (56·8%) participants were enrolled in URBMI or NRCMS, 186 369 (37·7%) were enrolled in UEBMI, and only 27 027 (5·5%) were uninsured. Most individuals living in urban areas were enrolled in the UEBMI, while most participants living in rural areas were enrolled in the NRCMS (table). During 3·92 million person-years of follow-up between 2009 and 2016, 794 824 hospital admissions were recorded for 268 070 participants, which included 74 313 (9·3%) admissions for stroke and 69 446 (8·7%) admissions for ischaemic heart disease.Table Background characteristics in the CKB

enrolled in the NRCMS (table). During 3·92 million person-years of follow-up between 2009 and 2016, 794 824 hospital admissions were recorded for 268 070 participants, which included 74 313 (9·3%) admissions for stroke and 69 446 (8·7%) admissions for ischaemic heart disease.Table Background characteristics in the CKB All (N=505 995*) Urban (n=224 093) Rural (n=281 902) Age,† years 54·7 (10·6) 55·4 (10·9) 54·2 (10·4) Sex Women 299 793 (59·2%) 133 933 (59·8%) 165 860 (58·8%) Men 206 202 (40·8%) 90 160 (40·2%) 116 042 (41·2%) Medical history Diabetes 15 604 (3·1%) 10 255 (4·6%) 5349 (1·9%) Hypertension 58 237 (11·5%) 30 625 (13·7%) 27 612 (9·8%) Coronary heart disease 15 018 (3·0%) 10 267 (4·6%) 4751 (1·7%) Stroke or transient ischaemic attack 8454 (1·7%) 5081 (2·3%) 3373 (1·2%) Body-mass index,‡ kg/m2 Normal, <25·0 338 770 (67·0%) 134 233 (59·9%) 204 537 (72·6%) Overweight, 25·0–29·9 146 475 (28·9%) 77 721 (34·7%) 68 754 (24·4%) Obese, ≥30·0 20 748 (4·1%) 12 139 (5·4%) 8609 (3·1%) Physical activity, metabolic equivalent h per day 21·2 (13·9) 18·5 (12·6) 23·3 (14·4) Systolic blood pressure, mm Hg 130·9 (21·1) 129·2 (21·0) 132·3 (21·1) Current smoker Men 126 084 (61·1%) 49 517 (54·9%) 76 567 (66·0%) Women 6981 (2·3%) 2713 (2·0%) 4268 (2·6%) Current alcohol drinker§ Men 92 090 (44·7%) 45 067 (50·0%) 47 023 (40·5%) Women 11 695 (3·9%) 5749 (4·3%) 5946 (3·6%) Self-reported poor health status¶ 51 187 (10·1%) 20 205 (9·0%) 30 982 (11·0%) Married 459 051 (90·7%) 200 278 (89·4%) 258 773 (91·8%) Household size 3·8 (1·5) 3·2 (1·3) 4·2 (1·5) Education No formal school 92 965 (18·4%) 26 434 (11·8%) 66 531 (23·6%) Primary or middle school 306 273 (60·5%) 117 066 (52·2%) 189 207 (67·1%) High school and above 106 757 (21·1%) 80 593 (36·0%) 26 164 (9·3%) Household income, ¥ per year <¥10 000 141 424 (28·0%) 30 188 (13·5%) 111 236 (39·5%) ¥10 000–19 999 147 143 (29·1%) 65 436 (29·2%) 81 707 (29·0%) ¥20 000–34 999 125 696 (24·8%) 71 495 (31·9%) 54 201 (19·2%) ≥¥35 000 91 732 (18·1%) 56 974 (25·4%) 34 758 (12·3%) Health insurance type‖ UEBMI 186 369 (37·7%) 158 288 (71·9%) 28 081 (10·2%) URBMI or NRCMS** 280 967 (56·8%) 44 505 (20·2%) 236 462 (86·2%) Uninsured 27 027 (5·5%) 17 321 (7·9%) 9706 (3·5%) Data are mean (SD) or n (%). UEBMI=urban employee basic medical insurance. URBMI=urban resident basic medical insurance. NRCMS=new rural cooperative medical scheme. CKB=China Kadoorie Biobank.

8 081 (10·2%) URBMI or NRCMS** 280 967 (56·8%) 44 505 (20·2%) 236 462 (86·2%) Uninsured 27 027 (5·5%) 17 321 (7·9%) 9706 (3·5%) Data are mean (SD) or n (%). UEBMI=urban employee basic medical insurance. URBMI=urban resident basic medical insurance. NRCMS=new rural cooperative medical scheme. CKB=China Kadoorie Biobank. * From the 512 715 participants initially recruited in the CKB, 6469 died and 251 were lost to follow-up in 2005–08. In 2009–16, 4500 participants were lost to follow-up. † Age is for the year 2009. Mean age at entry in CKB was 51·7 years (SD 10·6) for all, 52·8 (10·8) for urban, and 51·2 (10·4) for rural participants. ‡ 2 participants had missing data on body-mass index. Underweight individuals were grouped as “normal”. § Regular alcohol drinkers were those who drank alcohol at least monthly. ¶ Self-reported poor health status was defined as a response of “poor” from either excellent, good, fair, or poor by participants when answering the question, “how is your current general health status”. ‖ Information on health insurance type is for the year 2012 (n=494 363), as annual linkage to participants' schemes were only available from 2012 and missing health insurance type in 2009–11 was imputed based on the scheme participants were enrolled on in 2012. Uninsured participants include 307 (1·1%) participants with missing health insurance type. ** In the ten CKB centres, 263 916 (53·4%) individuals were enrolled in NRCMS and 17 038 (3·5%) in URBMI. In urban areas, 27 720 (12·6%) individuals were enrolled in NRCMS, while in rural areas 236 196 (86·1%) were enrolled in NRCMS.

‖ Information on health insurance type is for the year 2012 (n=494 363), as annual linkage to participants' schemes were only available from 2012 and missing health insurance type in 2009–11 was imputed based on the scheme participants were enrolled on in 2012. Uninsured participants include 307 (1·1%) participants with missing health insurance type. ** In the ten CKB centres, 263 916 (53·4%) individuals were enrolled in NRCMS and 17 038 (3·5%) in URBMI. In urban areas, 27 720 (12·6%) individuals were enrolled in NRCMS, while in rural areas 236 196 (86·1%) were enrolled in NRCMS. The hospitalisation rates adjusted for age, sex, and region increased by 8·2% for stroke (95% CI 7·8–8·6), 10·0% for ischaemic heart disease (9·5–10·6), and 10·3% for any cause (10·2–10·5) per year. Following further adjustment for individual socioeconomic, lifestyle, and morbidity factors, the corresponding annual increases in hospitalisation rates decreased by about half to 3·6% for stroke (95% CI 3·2–4·0), 5·4% for ischaemic heart disease (4·9–5·9), and 4·2% for any cause (4·0–4·3) (figure 1). The fully adjusted annual hospital admission rates increased from 10·9 (95% CI 10·5–11·2) in 2009 to 14·5 (14·2–14·9) admissions per 1000 person-years in 2016 for stroke; from 9·3 (8·9–9·6) to 14·1 (13·7–14·4) for ischaemic heart disease, and from 112·2 (111·0–113·3) to 151·2 (150·0–152·5) for any cause (appendix 2 p 16). The annual increases in rates of hospitalisation for stroke, ischaemic heart disease, and any cause were higher during the first cycle of the reforms (2009–12) than the second cycle (2013–16). The annual increases in rates of hospitalisation for stroke were 5·7% (95% CI 4·4 to 7·0) in the first cycle versus 0·5% (–0·4 to 1·5) in the second cycle; for ischaemic heart disease were 7·0% (5·4 to 8·5) versus 2·6% (1·5 to 3·6); and for any cause were 5·0% (4·6 to 5·5) versus −0·7% (–1·1 to −0·4; appendix 2 p 10).Figure 1 Change in rates of hospitalisation for stroke, ischaemic heart disease, and any cause, since 2009

cle versus 0·5% (–0·4 to 1·5) in the second cycle; for ischaemic heart disease were 7·0% (5·4 to 8·5) versus 2·6% (1·5 to 3·6); and for any cause were 5·0% (4·6 to 5·5) versus −0·7% (–1·1 to −0·4; appendix 2 p 10).Figure 1 Change in rates of hospitalisation for stroke, ischaemic heart disease, and any cause, since 2009 Generalised linear regression models with negative binomial distribution and log link function were used. The minimally adjusted models included adjustments for age, sex, and region. The fully adjusted models included adjustments for demographic factors (age and sex), socioeconomic factors (marital status, household size, education, income, and health insurance type), lifestyle factors (smoking, alcohol consumption, BMI, physical activity, self-reported health), morbidity factors, and region. Two participants had missing data for BMI. Rates are shown with 95% CIs, which are based on floating absolute risks. The area of each square is inversely proportional to the variance. BMI=body-mass index.

actors (smoking, alcohol consumption, BMI, physical activity, self-reported health), morbidity factors, and region. Two participants had missing data for BMI. Rates are shown with 95% CIs, which are based on floating absolute risks. The area of each square is inversely proportional to the variance. BMI=body-mass index. In 2009, the rates of hospitalisation were higher in urban than in rural areas for stroke (13·5 vs 8·7 admissions per 1000 person-years) and for ischaemic heart disease (13·0 vs 6·3; figure 2). The absolute differences in hospitalisation rates between individuals living in urban and rural areas in 2009–16 only decreased for ischaemic heart disease (trend: p=0·013; appendix 2 p 11). The fully adjusted annual increases in hospitalisation rates were higher for individuals living in rural than in urban areas for stroke (4·5% vs 3·3%) and ischaemic heart disease (8·4% vs 3·6%; figure 3). Overall, the fully adjusted hospitalisation rates for stroke and ischaemic heart disease were similar by levels of education. However, individuals with no formal education had the greatest annual increase in hospitalisation rates for both stroke (7·1%) and ischaemic heart disease (8·2%). Similarly, individuals in the lowest income group had the highest annual increase in fully adjusted hospitalisation rates for stroke (7·0%) and ischaemic heart disease (9·8%). Individuals enrolled in UEBMI had higher rates of hospitalisation for stroke and ischaemic heart disease than those enrolled in URBMI or NRCMS (stroke: 12·7 vs 10·3 per 1000 person-years in 2009; ischaemic heart disease: 11·3 vs 8·1 per 1000 person-years in 2009; figure 2). The differences and slope index of inequality in hospitalisation rates between UEBMI and URBMI or NRCMS enrollees decreased over time for stroke (trend: differences p=0·039; slope p=0·014) and ischaemic heart disease (both p<0·0001; appendix 2 p 11). The annual increases in rates were higher for individuals enrolled in URBMI or NRCMS than in UEBMI (stroke: 4·2% vs 2·5%; ischaemic heart disease: 7·4% vs 3·6%; figure 3).Figure 2 Annual rates of hospitalisation for stroke and ischaemic heart disease, by urban and rural area, education, and health insurance type

p 11). The annual increases in rates were higher for individuals enrolled in URBMI or NRCMS than in UEBMI (stroke: 4·2% vs 2·5%; ischaemic heart disease: 7·4% vs 3·6%; figure 3).Figure 2 Annual rates of hospitalisation for stroke and ischaemic heart disease, by urban and rural area, education, and health insurance type Generalised linear regression models with negative binomial distribution and log link function were used. Fully adjusted models included adjustments for demographic, socioeconomic, lifestyle, and morbidity factors, as well as region. In analyses by health insurance type, uninsured participants were excluded because of the small number of cases. Rates by education and health insurance type were standardised for the overall CKB participant population in 2009 and rates by urban and rural area were standardised separately for the CKB participant population living in urban or rural areas, in 2009. Rates per 1000 person-years are shown from 2009 to 2016. The area of each square is inversely proportional to the variance. 95% CIs are shown. Two participants had missing data for body-mass index. CKB=China Kadoorie Biobank. UEBMI=urban employee basic medical insurance. URBMI=urban resident basic medical insurance. NRCMS=new rural cooperative medical scheme. Figure 3 Annual change in rates of hospitalisation, 28-day case fatality, and mean length of hospital stay for stroke and ischaemic heart disease, by urban and rural area, education, income, and health insurance type

Generalised linear regression models with negative binomial distribution and log link function were used. Fully adjusted models included adjustments for demographic, socioeconomic, lifestyle, and morbidity factors, as well as region. In analyses by health insurance type, uninsured participants were excluded because of the small number of cases. Rates by education and health insurance type were standardised for the overall CKB participant population in 2009 and rates by urban and rural area were standardised separately for the CKB participant population living in urban or rural areas, in 2009. Rates per 1000 person-years are shown from 2009 to 2016. The area of each square is inversely proportional to the variance. 95% CIs are shown. Two participants had missing data for body-mass index. CKB=China Kadoorie Biobank. UEBMI=urban employee basic medical insurance. URBMI=urban resident basic medical insurance. NRCMS=new rural cooperative medical scheme. Figure 3 Annual change in rates of hospitalisation, 28-day case fatality, and mean length of hospital stay for stroke and ischaemic heart disease, by urban and rural area, education, income, and health insurance type For rates of hospitalisation: GLM with negative binomial distribution and log link function were used. For 28-day case fatality rates: GLM with binomial distribution and logit link function were used. For mean length of stay: GLM with gamma distribution and log link were used. Fully adjusted models included adjustments for demographic, socioeconomic, lifestyle, and morbidity factors, and region. Models for case fatality rates and mean length of stay also included adjustment for case mix (stroke types: ischaemic, haemorrhagic, and unspecified; and ischaemic heart disease types: acute myocardial infarction and other ischaemic heart disease). In analyses by health insurance type, uninsured participants were excluded due to the small number of cases. Two participants had missing data for BMI in models for rates of hospitalisation for both outcomes and one participant had missing BMI in models for case fatality rates for ischaemic heart disease. The area of each square is inversely proportional to the variance. GLM=generalised linear regression models. UEBMI=urban employee basic medical insurance. URBMI=urban resident basic medical insurance. NRCMS=new rural cooperative medical scheme. BMI=body-mass index.

s for case fatality rates for ischaemic heart disease. The area of each square is inversely proportional to the variance. GLM=generalised linear regression models. UEBMI=urban employee basic medical insurance. URBMI=urban resident basic medical insurance. NRCMS=new rural cooperative medical scheme. BMI=body-mass index. The fully adjusted 28-day case fatality rates for stroke decreased over time, from 14·4 to 9·3 deaths per 100 events (by 12·0% per year, 95% CI 10·3–13·7), and for ischaemic heart disease from 15·4 to 8·7 deaths per 100 events (by 14·5% per year, 12·6–16·3; appendix 2 p 17). The annual decreases in case fatality rates for stroke and ischaemic heart disease did not vary significantly between 2009–12 and 2013–16 (appendix 2 p 10). Case fatality rates for stroke and ischaemic heart disease were higher for individuals living in rural than in urban areas from 2009–16 (stroke: 13·0 vs 5·3 deaths per 100 events, ischaemic heart disease: 11·5 vs 6·7 in 2016; figure 4). However, the annual decrease in case fatality rates was around two-times higher for rural residents (stroke: 11·9% vs 6·4%; ischaemic heart disease: 16·7% vs 8·3%; figure 3). During the study period, case fatality rates decreased to a greater extent in individuals with lower levels of education or income (figure 3; appendix 2 p 18). Likewise, while 28-day case fatality rates for URBMI or NRCMS enrollees were higher than for UEBMI enrollees (stroke: 16·1 vs 9·1 deaths per 100 events; ischaemic heart disease: 21·3 vs 8·1 in 2009; figure 4), the annual decrease in case fatality rates was higher in URBMI or NRCMS enrollees (stroke: 14·2% vs 7·8%, ischaemic heart disease: 17·2% vs 11·2%; figure 3). Both the absolute differences and slope index of inequality in case fatality rates for stroke and ischaemic heart disease also decreased over time between individuals living in rural versus urban areas, those with the lowest versus highest levels of education, and those enrolled in URBMI or NRCMS versus UEBMI (all trends p<0·05; appendix 2 p 12).Figure 4 Annual 28-day case fatality rates for stroke and ischaemic heart disease, by urban and rural area, education, and health insurance type

g in rural versus urban areas, those with the lowest versus highest levels of education, and those enrolled in URBMI or NRCMS versus UEBMI (all trends p<0·05; appendix 2 p 12).Figure 4 Annual 28-day case fatality rates for stroke and ischaemic heart disease, by urban and rural area, education, and health insurance type Generalised linear regression models with binomial distribution and logit link function were used. Fully adjusted models included adjustments for demographic, socioeconomic, lifestyle, and morbidity factors, as well as region and case mix. In analyses by health insurance type, uninsured participants were excluded because of the small number of cases. Rates by education and health insurance type were standardised for the overall CKB participant population in 2009 and rates by urban and rural area were standardised separately for the CKB participant population living in urban or rural areas, in 2009. One participant had missing data for body-mass index in models for ischaemic heart disease. Rates per 100 events are shown for the years 2009 and 2016. The area of each square is inversely proportional to the variance. 95% CIs are shown. UEBMI=urban employee basic medical insurance. URBMI=urban resident basic medical insurance. NRCMS=new rural cooperative medical scheme.

in models for ischaemic heart disease. Rates per 100 events are shown for the years 2009 and 2016. The area of each square is inversely proportional to the variance. 95% CIs are shown. UEBMI=urban employee basic medical insurance. URBMI=urban resident basic medical insurance. NRCMS=new rural cooperative medical scheme. Between 2009 and 2016, the annual mean length of stay decreased by 1·9% per year for stroke (from 14·3 days in 2009 to 12·5 in 2016), 1·6% for ischaemic heart disease (from 11·2 days in 2009 to 10·1 in 2016), and 1·9% for any cause (from 12·3 days in 2009 to 10·1 in 2016; figure 5). Urban residents had a longer mean length of stay for stroke and ischaemic heart disease than rural residents (stroke: 13·0 vs 12·0 days; ischaemic heart disease: 10·8 vs 8·8, in 2016), and likewise for UEBMI versus URBMI or NRCMS enrollees (appendix 2 p 19). The annual decrease in mean length of stay was greater for individuals with the highest levels of education or income for stroke, ischaemic heart disease, and any cause for individuals living in urban areas and those enrolled in UEBMI for stroke and ischaemic heart disease (figure 3; appendix 2 p 20). The absolute differences in mean length of stay for stroke and ischaemic heart disease between the highest and lowest category and slope index of inequality decreased over time for most socioeconomic characteristics (appendix 2 p 13).Figure 5 Mean length of hospital stay for stroke, ischaemic heart disease, and any cause, by year

e differences in mean length of stay for stroke and ischaemic heart disease between the highest and lowest category and slope index of inequality decreased over time for most socioeconomic characteristics (appendix 2 p 13).Figure 5 Mean length of hospital stay for stroke, ischaemic heart disease, and any cause, by year Generalised linear regression models with gamma distribution and log link function were used. Fully adjusted models included adjustments for demographic, socioeconomic, lifestyle, and morbidity factors, as well as region and case mix. Mean length of stay was standardised for the overall China Kadoorie Biobank participant population in 2009. Two participants had missing data for body-mass index in models for any cause. Numbers above the squares are mean length of stay in days. 95% CIs are shown. The area of each square is inversely proportional to the variance.

of stay was standardised for the overall China Kadoorie Biobank participant population in 2009. Two participants had missing data for body-mass index in models for any cause. Numbers above the squares are mean length of stay in days. 95% CIs are shown. The area of each square is inversely proportional to the variance. In sensitivity analyses, the fully adjusted probabilities of first-ever hospitalisation for stroke increased by 3·8% (95% CI 3·3–4·3) per year and by 5·8% (5·2–6·4) per year for ischaemic heart disease (appendix 2 p 21), which was similar to the increase for all admissions. Likewise, the mean length of stay of first-ever hospitalisation decreased by 2·4% (95% CI 2·0–2·9) per year for stroke and by 3·9% (3·0–4·9) for ischaemic heart disease. The annual decrease for first-ever hospitalisation was similar to that for all admissions for stroke, but was greater for ischaemic heart disease. Annual differences in the rates of hospital admissions and case fatalities between urban and rural areas were smaller when standardised across the entire study population, but without adjustment for region (appendix 2 p 14).

was similar to that for all admissions for stroke, but was greater for ischaemic heart disease. Annual differences in the rates of hospital admissions and case fatalities between urban and rural areas were smaller when standardised across the entire study population, but without adjustment for region (appendix 2 p 14). The annual increase in hospitalisation rates for stroke was mainly due to increases in admissions for ischaemic rather than haemorrhagic stroke (4·9%, 95% CI 3·5 to 4·4 vs 1·1%, −0·2 to 2·3; appendix 2 p 22), while the annual increase in rates of hospitalisation for ischaemic heart disease was due to increases in both acute myocardial infarction (6·4%, 95% CI 4·6 to 8·3) and other ischaemic heart disease (5·3%, 4·8 to 5·8). Fully adjusted annual rates of hospitalisation, case fatality rates, and mean length of stay for stroke, ischaemic heart disease, and any cause were similar across age groups, and annual percentage changes by age group were also broadly similar to the overall annual percentage change (appendix 2 pp 15, 23).

·8 to 5·8). Fully adjusted annual rates of hospitalisation, case fatality rates, and mean length of stay for stroke, ischaemic heart disease, and any cause were similar across age groups, and annual percentage changes by age group were also broadly similar to the overall annual percentage change (appendix 2 pp 15, 23). Discussion We studied around 0·5 million Chinese adults who were followed up for an 8-year period from 2009 to 2016. We found that, after controlling for demographic, socioeconomic, lifestyle, and morbidity factors, hospital admissions for stroke and ischaemic heart disease increased by approximately 4–5% per year. Moreover, although hospital use was higher among individuals living in urban areas and those enrolled in UEBMI, annual increases in hospitalisation rates for stroke and ischaemic heart disease were greater among individuals living in rural areas, those with lower levels of education or income, and those enrolled in URBMI or NRCMS than UEBMI. In addition, the study showed that case fatality rates for stroke and ischaemic heart disease decreased over the period. While case fatality rates were higher in lower than higher socioeconomic groups, annual reductions in case fatality rates were greater in the lower socioeconomic groups. Over the same period, the mean length of stay decreased by around 2% per year for stroke, ischaemic heart disease, and any cause. The mean length of stay was longer for individuals living in urban areas and enrolled in UEBMI, and the annual decrease in mean length of stay was higher in these groups and in individuals with higher education or income levels. These trends indicate modest improvements in use, quality, and efficiency of hospital care for stroke, ischaemic heart disease, and any cause in China.

urban areas and enrolled in UEBMI, and the annual decrease in mean length of stay was higher in these groups and in individuals with higher education or income levels. These trends indicate modest improvements in use, quality, and efficiency of hospital care for stroke, ischaemic heart disease, and any cause in China. National statistics and previous studies in China also reported increased rates of hospital admissions for any cause and cardiovascular disease over the calendar period of the present study.5, 10, 14, 16, 22 However, age, sex, and region-adjusted admission rates for any cause in the CKB study were higher (23% for CKB vs 17% for national data in 2016), which probably reflects the older population studied in the CKB study.23 Annual increases in hospitalisation rates for stroke, ischaemic heart disease, and any cause were reduced by around half after adjustment for socioeconomic, lifestyle, and morbidity factors. The observed reduction was consistent with reports that a large proportion of the increasing cardiovascular disease burden in China is likely to reflect poor control of established cardiovascular disease risk factors (eg, smoking, systolic blood pressure, LDL cholesterol, blood glucose, body-mass index, and physical activity).24

ed reduction was consistent with reports that a large proportion of the increasing cardiovascular disease burden in China is likely to reflect poor control of established cardiovascular disease risk factors (eg, smoking, systolic blood pressure, LDL cholesterol, blood glucose, body-mass index, and physical activity).24 Apart from established cardiovascular disease risk factors and morbidities, health system factors could also play an important role in the increasing trends in hospital use.5 Although the present trends cannot attribute causality between health-care reforms and changes in hospital use and outcomes between different socioeconomic groups, the present findings could inform future public health policy to reduce inequalities in health-care delivery in China. These results suggest greater increases in hospital use during the first cycle of the reforms (2009–12), coinciding with a greater expansion of health insurance coverage.25 During the reform period, inpatient reimbursement rates and ceilings (maximum level of patient reimbursement) for NRCMS and URBMI increased rapidly and deductibles decreased, which might partly account for the higher annual increases in hospital use for enrollees of such schemes compared with UEBMI enrollees.5 In 2012, China launched a critical illness insurance, which covers patients enrolled in NRCMS or URBMI with diseases whose annual health-care costs exceed the annual mean disposable income per capita in the local area.26 Individuals who have critical illnesses, including stroke and ischaemic heart disease subtypes, receive extra reimbursements for treatment costs,5 which increases affordable access to hospital care. As part of the reforms, local civil affairs bureaus designed medical assistance programmes to cover any remaining co-payments for priority diseases and individual contributions for low-income households.6 Our study showed that the most disadvantaged groups had the highest annual increases in rates of hospitalisation and greatest reductions in case fatality rates. However, inequalities in hospitalisation and case fatality rates between different socioeconomic groups have persisted between 2009 and 2016, possibly reflecting further differences in health-care seeking and affordability.

est annual increases in rates of hospitalisation and greatest reductions in case fatality rates. However, inequalities in hospitalisation and case fatality rates between different socioeconomic groups have persisted between 2009 and 2016, possibly reflecting further differences in health-care seeking and affordability. The increasing rates of hospital use, although at a lower rate for urban residents and higher socioeconomic groups, might reflect the financial incentives of hospitals to promote greater use of hospital services and underdeveloped and underused primary health care in China.25, 27, 28 Patients' distrust of primary health-care services combined with the design of health insurance packages, which continue to provide low levels of coverage for outpatient care compared with inpatient care, are key contributors.5, 6 Moreover, between 2009 and 2016, the number of hospitals in China increased by 44% nationally, and the number of beds in medical institutions increased by 42% in urban areas and by 50% in rural areas, which might also have contributed to the increased hospital use.23

pared with inpatient care, are key contributors.5, 6 Moreover, between 2009 and 2016, the number of hospitals in China increased by 44% nationally, and the number of beds in medical institutions increased by 42% in urban areas and by 50% in rural areas, which might also have contributed to the increased hospital use.23 The decreasing trends in case fatality rates among all socioeconomic groups might partly reflect improvements in quality of care and more effective hospital treatment.29, 30 The findings in the present study are consistent with the substantial reductions in case fatality rates for stroke and ischaemic heart disease reported in western countries.30, 31, 32, 33, 34 Case fatality rates for stroke in the present study were lower than those previously reported by other studies in China (<5–50%).27, 35 Lower rates in the present study might reflect differences in disease severity, types of stroke, quality of care, or lower thresholds for diagnosis over time.27, 35

.30, 31, 32, 33, 34 Case fatality rates for stroke in the present study were lower than those previously reported by other studies in China (<5–50%).27, 35 Lower rates in the present study might reflect differences in disease severity, types of stroke, quality of care, or lower thresholds for diagnosis over time.27, 35 Decreasing annual trends in mean length of stay for stroke, ischaemic heart disease, and any cause probably reflect the government introduction of a target mean length of stay as a performance indicator to reduce health-care expenditure, and other efforts by health insurance agencies to reduce costs.36 China has also implemented standard clinical pathways for major cardiovascular disease outcomes (eg, acute ischaemic stroke and myocardial infarction) and established an official monitoring system for these pathways in 2009.29 Such reforms appear to have contributed to reductions in mean length of stay, and improvements in survival after stroke and ischaemic heart disease.37 The greater reduction in mean length of stay for higher socioeconomic groups might reflect the greater reductions in mean length of stay in large urban hospitals, mostly used by these groups.38 However, mean length of stay in China is still greater than that in Organisation for Economic Co-operation and Development member countries,34 with the typically conservative behaviour of Chinese doctors seeking to comply with patient wishes; the adverse incentives generated by the use of fee-for-service reimbursement;12, 36 and the limited availability of rehabilitation services for post-acute care identified as contributors.29

evelopment member countries,34 with the typically conservative behaviour of Chinese doctors seeking to comply with patient wishes; the adverse incentives generated by the use of fee-for-service reimbursement;12, 36 and the limited availability of rehabilitation services for post-acute care identified as contributors.29 The main strengths of the present study include a long duration of follow-up, which covers the two major waves of the reforms (2009–12 and 2013–16);25 the diversity in regions and hospitals included; and the extensive data collected for individual participants, which enabled adjustment for a wide range of confounders. However, the study also had several limitations. First, the CKB was designed to include diverse areas, socioeconomic groups, and risk factors in China rather than to be representative of the Chinese population, so results should be interpreted with caution. However, the diverse socioeconomic groups studied suggest that the findings are likely to be generalisable to the full range of such exposures in the population. Second, despite our ability to adjust for participant's characteristics at entry and their age and morbidities during follow-up, the study was constrained by having a longitudinal fixed cohort study design. In the CKB study, periodic resurveys only included a random subset of participants (around 5%), so some of the covariates included in the analyses could not be updated during follow-up. Third, data for participants' health insurance type were only available from 2012 onwards and, therefore, an assumption had to be made that the participants' health insurance type in 2009–11 was the same as that recorded in 2012. However, data from resurveys of CKB participants indicated that 97% of participants reported being insured in both 2008 and 2014, and the proportion of participants enrolled in the same scheme between 2012 and 2016 remained stable. Fourth, the data available in the present study did not allow for a more nuanced measure of hospital care access with hospital use capturing exclusively health demand presenting for hospital care. Fifth, interpreting fully adjusted trends in hospitalisation rates as health system effects relies on the assumption that adjustments for demographic (eg, age and sex), socioeconomic, lifestyle, and morbidity factors were successful.

ss with hospital use capturing exclusively health demand presenting for hospital care. Fifth, interpreting fully adjusted trends in hospitalisation rates as health system effects relies on the assumption that adjustments for demographic (eg, age and sex), socioeconomic, lifestyle, and morbidity factors were successful. Fully adjusted trends also need to be interpreted in light of other possible supply and demand factors, including details of health insurance coverage and health-care costs, hospital-level characteristics, distance between household and hospitals, outpatient care provision and use, physician-related factors, and changes in admission thresholds or use of diagnostic tests for cardiovascular disease.27 Sixth, as participants in each CKB region were either all urban or all rural residents, absolute differences in hospital use and disease outcomes between urban and rural areas, although subject to regional variation, could not be adjusted for differences across all regions. Overall, the associations reported in this prospective cohort study cannot be causally attributed to the health-care reforms as a different study design would be required to address causality. Longer duration of follow-up might also be needed to reliably assess the longer-term effects of the health-care reforms in China.

erall, the associations reported in this prospective cohort study cannot be causally attributed to the health-care reforms as a different study design would be required to address causality. Longer duration of follow-up might also be needed to reliably assess the longer-term effects of the health-care reforms in China. In conclusion, this large prospective study examined changes in hospital care use and outcomes for individuals with stroke and ischaemic heart disease between 2009 and 2016, both overall and across different socioeconomic groups. The robust findings of increasing hospital admission rates and decreasing case fatality rates, particularly in rural areas and for individuals with lower levels of education or income, and enrolled in the urban or rural resident health insurance scheme, reflect improvements in use and quality of care in 2009–16. The continuous reductions in mean length of stay also indicate improvements in hospital efficiency. Nevertheless, there is substantial scope to achieve further reductions in socioeconomic inequalities, by altering health insurance benefits, strengthening primary care, and implementing cost-effective interventions to decrease the cardiovascular disease burden and control escalating hospital care costs. Supplementary Materials Chinese translation of the abstract Supplementary appendix

In conclusion, this large prospective study examined changes in hospital care use and outcomes for individuals with stroke and ischaemic heart disease between 2009 and 2016, both overall and across different socioeconomic groups. The robust findings of increasing hospital admission rates and decreasing case fatality rates, particularly in rural areas and for individuals with lower levels of education or income, and enrolled in the urban or rural resident health insurance scheme, reflect improvements in use and quality of care in 2009–16. The continuous reductions in mean length of stay also indicate improvements in hospital efficiency. Nevertheless, there is substantial scope to achieve further reductions in socioeconomic inequalities, by altering health insurance benefits, strengthening primary care, and implementing cost-effective interventions to decrease the cardiovascular disease burden and control escalating hospital care costs. Supplementary Materials Chinese translation of the abstract Supplementary appendix Acknowledgments Our main acknowledgment is to the participants, the project staff, and staff of the Chinese Center for Disease Control and Prevention and its regional offices for access to death and disease registries. The Chinese National Health Insurance scheme provided electronic linkage to all hospitalisation data. The CKB study is jointly coordinated by the University of Oxford and the Chinese Academy of Medical Sciences. The funding body for the baseline survey was the Kadoorie Charitable Foundation, Hong Kong, China and the funding sources for the long-term continuation of the study include Wellcome Trust (202922/Z/16/Z, 104085/Z/14/Z, 088158/Z/09/Z), Chinese National Natural Science Foundation (81390540, 81390541, 81390544), and the National Key Research and Development Program of China (2016YFC0900500, 2016YFC0900501, 2016YFC0900504, 2016YFC1303904). Core funding was provided to the Clinical Trial Service Unit and Epidemiological Studies Unit, University of Oxford, by the British Heart Foundation, the UK Medical Research Council, and Cancer Research UK. ML is supported by a Wellcome Trust Doctoral Studentship in Society and Ethics (208126/Z/17/Z). BM is supported by the National Institute for Health Research Oxford Biomedical Research Centre.

l Studies Unit, University of Oxford, by the British Heart Foundation, the UK Medical Research Council, and Cancer Research UK. ML is supported by a Wellcome Trust Doctoral Studentship in Society and Ethics (208126/Z/17/Z). BM is supported by the National Institute for Health Research Oxford Biomedical Research Centre. Contributors ML, YC, RC, WY, and BM designed and planned the study. ML did data analyses and wrote the first draft of the manuscript. YC, RC, WY, ZC, and BM provided critical comments on the scientific interpretation of the results and on revised versions of the manuscript. RC, LL, and ZC, as members of CKB steering committee, designed and supervised the overall conduct of the study including obtaining funding. YC, YT, YG, ZB, and ZC coordinated the data acquisition (for baseline and long-term follow-up). All authors provided critical comments on the manuscript. Declaration of interests We declare no competing interests.