Browse the corpus

were screened with conventional PCR-based sequencing (appendix pp 4–5); choice of screening method was based on the availability of funding. All patients were prospectively followed up per the standard Australian National Health and Medical Research Council guidelines, with a median follow-up of 60 months (IQR 36–69). We investigated the prognostic associations of KRAS and BRAF mutations in relation to MSI status by pooling data from the QUASAR 2 gene panel, the Australian validation set, and additional QUASAR 2 and stage II or III Australian colorectal cancers that had been analysed for MSI and by Sanger sequencing for KRAS or BRAF mutations (appendix p 8) for an extended set of patients. Similar analyses were also done in the extended cohorts, whereby TP53 status derived from either next-generation sequencing or Sanger sequencing was added.

Introduction There is increasing recognition that treatment of common cancers can be modified according to a patient's expected prognosis or response to therapy. For some new molecularly guided therapies, powerful biomarkers of response are available, which often comprise mutations in the specific protein that is targeted. However, for conventional cytotoxic therapies, predictive markers of response are rare. In view of the modest survival benefits that conventional cytotoxic therapies provide for patients with common solid malignancies, biomarkers of prognosis still have substantial potential clinical importance. Such markers could guide the use of more or less aggressive treatment regimens and enable clinicians to balance expected outcomes against early and late therapeutic toxicities. Research in context Evidence before this study The decision to give adjuvant chemotherapy after resection of stage II or III colorectal cancer is based mainly on pathological factors such as tumour and nodal stage. Microsatellite instability (MSI) is the only molecular marker used routinely in this setting. However, patient outcomes remain variable, and stratification needs to be improved. We searched PubMed with the terms “prognosis”, “colorectal”, “colon”, and “rectal” for articles published in English up to Feb 16, 2017. Only two large studies (>400 profiled patients) in the adjuvant setting had screened more than four molecular markers. Added value of this study

The decision to give adjuvant chemotherapy after resection of stage II or III colorectal cancer is based mainly on pathological factors such as tumour and nodal stage. Microsatellite instability (MSI) is the only molecular marker used routinely in this setting. However, patient outcomes remain variable, and stratification needs to be improved. We searched PubMed with the terms “prognosis”, “colorectal”, “colon”, and “rectal” for articles published in English up to Feb 16, 2017. Only two large studies (>400 profiled patients) in the adjuvant setting had screened more than four molecular markers. Added value of this study We used next-generation sequencing to analyse a panel of 82 genes in colorectal cancer from the QUASAR 2 clinical trial, and validated our findings in an Australian community-based colorectal cancer cohort. We identified high mutation burden as an independent marker of good prognosis, even after omitting hypermutant tumours with defects in DNA mismatch or polymerase proofreading repair. We hypothesise that this finding resulted from high neo-epitope levels genome-wide. TP53, KRAS, and BRAF mutations were additionally independently associated with poor prognosis, although the association with BRAF and KRAS was restricted to MSI-negative tumours. Implications of all the available evidence

We used next-generation sequencing to analyse a panel of 82 genes in colorectal cancer from the QUASAR 2 clinical trial, and validated our findings in an Australian community-based colorectal cancer cohort. We identified high mutation burden as an independent marker of good prognosis, even after omitting hypermutant tumours with defects in DNA mismatch or polymerase proofreading repair. We hypothesise that this finding resulted from high neo-epitope levels genome-wide. TP53, KRAS, and BRAF mutations were additionally independently associated with poor prognosis, although the association with BRAF and KRAS was restricted to MSI-negative tumours. Implications of all the available evidence Although the 15% of stage II or III colorectal cancers with hypermutation caused by DNA repair defects have previously been shown to have a good prognosis in the non-metastatic setting, we have shown that increased mutation burden among non-hypermutated colorectal cancers is also associated with favourable outcomes. Our data additionally show that the prognostic value of MSI is improved by a model based on mutation burden and KRAS, BRAF, and TP53 mutations. Use of even a modestly sized gene panel provides superior prognostic information to tests based on a handful of genes, and could allow for existing and novel therapies to be targeted to subgroups of patients with poor prognosis, thereby sparing patients with good outcomes unnecessary and toxic treatment.

P53 mutations. Use of even a modestly sized gene panel provides superior prognostic information to tests based on a handful of genes, and could allow for existing and novel therapies to be targeted to subgroups of patients with poor prognosis, thereby sparing patients with good outcomes unnecessary and toxic treatment. Biomarkers can be based on several different types of molecule, and high-profile work has highlighted the potential use of mRNA profiling for identification of groups of colorectal cancers with varying prognoses.1 Other biomarkers are based on DNA, which is more stable and thus generally easier to analyse than mRNA. For colorectal cancers treated with curative intent, the biomarker most consistently used in clinical practice is microsatellite instability (MSI), which usually results from defective DNA mismatch repair.2 For stage II colorectal cancers, MSI predicts good recurrence-free survival, with hazard ratios (HRs) as low as 0·6.3, 4 This association is less strong for stage III cancers, and, in stage IV colorectal cancers, MSI positivity is probably associated with poor prognosis.5

ly results from defective DNA mismatch repair.2 For stage II colorectal cancers, MSI predicts good recurrence-free survival, with hazard ratios (HRs) as low as 0·6.3, 4 This association is less strong for stage III cancers, and, in stage IV colorectal cancers, MSI positivity is probably associated with poor prognosis.5 The availability of a few large datasets (>500 participants) from clinical trials has begun to clarify the associations between some somatic mutations and prognosis of colorectal cancers. However, most of these analyses have been restricted to KRAS mutations, BRAF mutations, or MSI (appendix pp 6–7). Overall, for colorectal cancers treated with curative intent (generally stage II or III), data support an association between MSI and good prognosis, and weaker evidence suggests that KRAS and BRAF mutations, which are mutually exclusive, indicate poor prognosis in MSI-negative tumours.6, 7, 8, 9, 10, 11 However, MSI-positive colorectal cancers tend to be BRAF-mutant and KRAS-wild-type, so statistical interactions could exist between these prognostic biomarkers. Furthermore, whether combinations of other genetic biomarkers provide useful prognostic information is unclear.

osis in MSI-negative tumours.6, 7, 8, 9, 10, 11 However, MSI-positive colorectal cancers tend to be BRAF-mutant and KRAS-wild-type, so statistical interactions could exist between these prognostic biomarkers. Furthermore, whether combinations of other genetic biomarkers provide useful prognostic information is unclear. Screening has been restricted to only a few genes in large genetic biomarker studies for two main reasons: suboptimal sample quality or quantity, and the cost of mutation screening. Because somatic mutations tend to co-occur in molecular pathways of tumorigenesis, screening of many potentially prognostic mutations in the same dataset would be highly desirable to identify the primary determinants of tumour behaviour. However, the few studies in which such analyses were done did not have standardised recruitment and follow-up. The prime example is the exome or genome sequencing of over 600 colorectal cancers by the Cancer Genome Atlas group.12 This work provided an excellent dataset for discovery of driver mutations, but is of little use for biomarker discovery owing to the heterogeneity of the sample set and associated variability in clinical data.

e example is the exome or genome sequencing of over 600 colorectal cancers by the Cancer Genome Atlas group.12 This work provided an excellent dataset for discovery of driver mutations, but is of little use for biomarker discovery owing to the heterogeneity of the sample set and associated variability in clinical data. In this exploratory study, we aimed to retain the advantages of a large clinical trial dataset while assessing several prognostic biomarkers for colorectal cancers. To this end, we used an 82-gene panel to identify somatic mutations in all the major colorectal cancer driver genes in more than 500 tumours from the QUASAR 2 clinical trial of stage II and III colorectal cancers. We also assessed MSI and the ultramutator phenotype resulting from POLE mutations.4 We also tested a larger QUASAR 2 sample set for KRAS and BRAF mutations and MSI. Variables associated with survival in QUASAR 2 were replication tested in an independent community-based cohort, and subjected to a combined analysis.

ers. We also assessed MSI and the ultramutator phenotype resulting from POLE mutations.4 We also tested a larger QUASAR 2 sample set for KRAS and BRAF mutations and MSI. Variables associated with survival in QUASAR 2 were replication tested in an independent community-based cohort, and subjected to a combined analysis. Methods Study design and participants In this exploratory study, we assessed prognostic biomarkers for colorectal cancers in a large clinical trial dataset from a phase 3 clinical trial (QUASAR 2) and an independent community-based validation cohort. QUASAR 2 was an open-label, randomised phase 3 clinical trial13 comprising 1952 patients with high-risk stage II or stage III colorectal cancer, who were randomly assigned to capecitabine alone or capecitabine plus bevacizumab, without radiotherapy. Median follow-up was 4·92 years (IQR 4·00–5·16). Overall or disease-free survival did not differ significantly between the two groups at 3 years' follow-up.13 Similar results have been recorded in two other trials.14, 15 We obtained clinicopathological data (appendix p 8) from the QUASAR 2 trial database. Some data were converted to binary variables—ie, sex, location (proximal vs distal), and depth of invasion (T4 vs T1, T2, or T3) and lymph node metastasis (N2 or N1 vs N0) according to the TNM grading system. Age and grade were assessed as continuous variables.

ical data (appendix p 8) from the QUASAR 2 trial database. Some data were converted to binary variables—ie, sex, location (proximal vs distal), and depth of invasion (T4 vs T1, T2, or T3) and lymph node metastasis (N2 or N1 vs N0) according to the TNM grading system. Age and grade were assessed as continuous variables. The community-based series included 657 patients with stage II or III colorectal cancer who were treated at the Royal Melbourne Hospital (Parkville, VIC, Australia), Western Hospital Footscray (Footscray, VIC, Australia) or St Vincent's Hospital (Sydney, NSW, Australia) between Jan 1, 1993, and Dec 31, 2009 (appendix p 8). Individuals with hereditary colorectal cancer syndromes were excluded. All patients received standard neoadjuvant or adjuvant fluorouracil-based chemotherapy or concurrent chemoradiotherapy. In this patient series, stage II disease was deemed low risk when tumours were T3/N0; otherwise it was judged high risk. All patients provided written informed consent, and the study was approved by medical ethics committees at all three sites.

r adjuvant fluorouracil-based chemotherapy or concurrent chemoradiotherapy. In this patient series, stage II disease was deemed low risk when tumours were T3/N0; otherwise it was judged high risk. All patients provided written informed consent, and the study was approved by medical ethics committees at all three sites. Procedures Colorectal cancer samples from UK QUASAR 2 were collected for molecular analysis. 40 μm scrolls were cut from formalin-fixed paraffin-embedded specimens of colorectal cancers that had greater than 80% estimated purity, and from healthy bowel; 10 μm sections were cut from the remaining colorectal cancers and needle microdissected to enrich for tumours with a haematoxylin and eosin section as a guide. Peripheral blood samples were also available from most patients. DNA was extracted from formalin-fixed paraffin-embedded tissue with the DNeasy kit (Qiagen, Hilden, Germany) and from blood with the Maxwell 16 Blood DNA Purification Kit (Promega, Madison, WI, USA). The whole cohort was analysed by Sanger sequencing for selected mutations and for MSI (appendix pp 1–3), and a subset of tumours was also analysed with an Ion Torrent (Life Technologies, Guildorf, CT, USA) sequencing gene panel for 82 genes (appendix p 9). We eliminated mutations with a high probability of being artifacts and cancers with high levels of artifactual hypermutation owing to ex-vivo cytosine deamination (appendix pp 2–3, 32).

umours was also analysed with an Ion Torrent (Life Technologies, Guildorf, CT, USA) sequencing gene panel for 82 genes (appendix p 9). We eliminated mutations with a high probability of being artifacts and cancers with high levels of artifactual hypermutation owing to ex-vivo cytosine deamination (appendix pp 2–3, 32). We identified all probable driver mutations (appendix p 14) and selected the 13 most commonly mutated genes (ie, mutated in eight or more tumours) for further analysis to identify mutations tending to occur together in genetic pathways (appendix pp 17, 35–36). High-depth sequencing allowed us to identify tumours carrying somatic mutations at substantially reduced allele frequency (suggestive of subclonal status). From the community-based series, fresh-frozen tumours and matched normal specimens were retrieved from hospital tissue banks. A subset of these tumours was screened in 113 genes by targeted next-generation sequencing, the others were screened with conventional PCR-based sequencing (appendix pp 4–5); choice of screening method was based on the availability of funding. All patients were prospectively followed up per the standard Australian National Health and Medical Research Council guidelines, with a median follow-up of 60 months (IQR 36–69).

III Australian colorectal cancers that had been analysed for MSI and by Sanger sequencing for KRAS or BRAF mutations (appendix p 8) for an extended set of patients. Similar analyses were also done in the extended cohorts, whereby TP53 status derived from either next-generation sequencing or Sanger sequencing was added. Statistical analysis Individual driver gene mutations, combinations of mutations, or global measures such as MSI or mutation burden (total number of non-synonymous mutations and coding indels) were tested for associations with relapse-free survival in univariable and multivariable models, principally Cox proportional hazards models in accordance with published guidelines (appendix p 10).16 We used the likelihood ratio test to compare a prognostic model based on the gold standard of clinicopathological variables and MSI with our new model, and did 10% leave-out cross-validation analysis to confirm the robustness of these results. To test whether the prognostic effect of mutation burden was due to hypermutation only, the same model was run in the subset of tumours without MSI or pathogenic POLE mutations. All survival analyses were two-sided and were deemed significant if p values were less than or equal to 0·05. Univariable results with p values less than 0·1 were taken forward to be tested in multivariable models. Further details of patients and analytic methods are in the appendix (p 5).

or pathogenic POLE mutations. All survival analyses were two-sided and were deemed significant if p values were less than or equal to 0·05. Univariable results with p values less than 0·1 were taken forward to be tested in multivariable models. Further details of patients and analytic methods are in the appendix (p 5). Because several mutations co-varied, we searched for primary associations by multivariable regression, hierarchical clustering, and Bayesian networks (appendix p 4). All analyses were done in STATA (version 10), R (version 3.4.1), or Banjo (version 2.2.0). Research materials supporting this publication can be accessed by contacting the corresponding author. Role of the funding source The study funders had no role in the study design; data collection, analysis, or interpretation; or writing of the report. The corresponding author had full access to all study data and final responsibility for the decision to submit for publication. Results 598 tumours from the QUASAR 2 clinical trial were sequenced for 82 genes. After exclusion of mutations with a high probability of being artifacts and cancers with high levels of artifactual hypermutation owing to ex-vivo cytosine deamination, 511 tumours remained for further analysis (appendix pp 2–3).

Role of the funding source The study funders had no role in the study design; data collection, analysis, or interpretation; or writing of the report. The corresponding author had full access to all study data and final responsibility for the decision to submit for publication. Results 598 tumours from the QUASAR 2 clinical trial were sequenced for 82 genes. After exclusion of mutations with a high probability of being artifacts and cancers with high levels of artifactual hypermutation owing to ex-vivo cytosine deamination, 511 tumours remained for further analysis (appendix pp 2–3). The 13 most commonly mutated genes (APC, TP53, KRAS, PIK3CA, BRAF, FBXW7, SMAD4, ATM, PTEN, NF1, CTNNB1, GNAS, and NRAS)—ie, mutated in eight or more tumours—were selected for further analysis to identify mutations tending to occur together in genetic pathways (appendix p 14). In addition to known associations, such as those between BRAF mutation and MSI or between mutations of KRAS and PIK3CA, new unreported ones were found. Multivariable regression, hierarchical clustering, and Bayesian networks showed that mutations in NF1, a negative regulator of the Ras pathway, were positively associated with NRAS mutations, but not with mutations in KRAS or BRAF (appendix pp 17, 35–36). SMAD4 mutations were associated with BRAF mutations but not with KRAS or NRAS changes (appendix pp 17, 35–36), suggesting possible synergy between BRAF and the TGFβ or BMP pathways. Additionally, logistic regression and Bayesian network analyses showed a strong negative association between driver mutations in TP53 and ATM (appendix pp 17, 35–36). Clustering and Bayesian network analysis suggested a positive association between ATM and PTEN mutations (appendix pp 17, 35–36). Regression analysis between molecular and clinical variables showed that KRAS mutations were associated with female sex (similar to BRAF mutations;12, 17 appendix pp 17, 35–36). Additionally, mutations in FBXW7 and CTNNB1 were associated with high-grade disease (appendix pp 17, 35–36).

utations (appendix pp 17, 35–36). Regression analysis between molecular and clinical variables showed that KRAS mutations were associated with female sex (similar to BRAF mutations;12, 17 appendix pp 17, 35–36). Additionally, mutations in FBXW7 and CTNNB1 were associated with high-grade disease (appendix pp 17, 35–36). High-depth sequencing identified 58 (11%) tumours carrying somatic mutations at substantially reduced allele frequency, suggesting subclonal status. Of the 13 most commonly mutated genes, PIK3CA (p=0·001), ATM (p=0·002), and SMAD4 (p=0·05) had lower driver mutation allele frequencies than the other genes, suggesting they were more often subclonal (appendix p 18). Mutation burden, clonal diversity (presence of any identified mutation at low allele frequency), and driver mutations in the 13 genes were tested for prediction of bevacizumab treatment response, with no significant associations identified (data not shown).

es, suggesting they were more often subclonal (appendix p 18). Mutation burden, clonal diversity (presence of any identified mutation at low allele frequency), and driver mutations in the 13 genes were tested for prediction of bevacizumab treatment response, with no significant associations identified (data not shown). In QUASAR 2, overall mutation burden and mutations in four specific genes (TP53, KRAS, BRAF, and GNAS) showed promising individual associations with relapse-free survival (predefined p<0·10) and were thus selected for multivariable analysis, together with T stage, N stage, treatment group (because bevacizumab had previously been associated with poor prognosis in our patient subgroup, although not the whole trial), and MSI (which co-varied with mutation burden and is probably the best established prognostic factor for colorectal cancer; table 1, appendix p 19). Mutation burden (HR 0·81 [95% CI 0·68–0·96]; p=0·014), mutations in TP53, KRAS, BRAF, and GNAS, T stage, N stage, and use of bevacizumab were all independently associated with poor prognosis (ie, p≤0·05), but MSI was not (HR 1·12 [95% CI 0·57–2·19]; p=0·75; table 1). To test whether the prognostic effect of mutation burden was due to hypermutation only, the same model was run in the subset of tumours without MSI or pathogenic POLE mutations. Mutation burden was no longer significantly associated with outcome (HR 0·85 [95% CI 0·73–1·00]; p=0·051), although the HR was similar. The other variables retained significance similar to that previously shown (table 1).Table 1 Associations between clinicopathological molecular variables and relapse-free survival in the QUASAR 2 cohort

s no longer significantly associated with outcome (HR 0·85 [95% CI 0·73–1·00]; p=0·051), although the HR was similar. The other variables retained significance similar to that previously shown (table 1).Table 1 Associations between clinicopathological molecular variables and relapse-free survival in the QUASAR 2 cohort All cases univariable (n=511) All cases multivariable (n=511) MSI-negative and non-pathogenic POLE multivariable (n=443) HR 95% CI p value HR 95% CI p value HR 95% CI p value KRAS mutation 1·48 1·07–2·05 0·018 1·99 1·37–2·91 3·44 × 10−4 2·25 1·51–3·35 6·07 × 10−5 BRAF mutation 1·42 0·94–2·13 0·093 2·46 1·51–4·03 3·31 × 10−4 2·88 1·70–4·85 7·50 × 10−5 TP53 mutation 1·53 1·08–2·18 0·018 1·63 1·12–2·38 0·011 1·61 1·09–2·38 0·025 GNAS mutation 2·19 0·89–5·35 0·087 2·76 1·08–7·04 0·034 4·00 1·42–11·3 0·009 Mutation burden (quartiles) 0·87 0·75–1·00 0·055 0·81 0·68–0·96 0·014 0·85 0·73–1·00 0·051 MSI 0·73 0·42–1·28 0·271 1·12 0·57–2·19 0·75 .. .. .. Chemotherapy (bevacizumab plus capecitabine vs capecitabine) 1·37 0·98–1·92 0·065 1·43 1·02–2·00 0·039 1·55 1·09–2·22 0·015 T4 vs T1, T2, or T3* 2·11 1·52–2·94 8·59 × 10−6 2·10 1·50–2·93 1·36 × 10−5 2·29 1·61–3·25 3·66 × 10−6 N1 or N2 vs N0* 1·80 1·22–2·63 0·003 1·85 1·25–2·73 0·002 2·03 1·33–3·09 0·001 Cox proportional hazards analysis was done. The univariable analyses were adjusted by T stage, N stage, and treatment arm (or two of these if the adjustment variable itself was being assessed). Multivariable analysis was based on all variables shown. Mutation burden was derived from total number of non-synonymous mutations and coding indels, which are most likely to be functionally relevant, but similar results were obtained when other somatic variants were also included (appendix). POLE proofreading mutation is not shown as a prognostic variable because of the low frequency of those cancers (appendix). MSI=microsatellite instability. HR=hazard ratio.

els, which are most likely to be functionally relevant, but similar results were obtained when other somatic variants were also included (appendix). POLE proofreading mutation is not shown as a prognostic variable because of the low frequency of those cancers (appendix). MSI=microsatellite instability. HR=hazard ratio. * According to TNM tumour classification. In the Australian community-based cohort, 379 patients received adjuvant fluorouracil treatment, of whom 47 also received oxaliplatin (no data for oxaliplatin use were available for 38). We replication tested our prognostic markers in 296 tumours from the Australian cohort (appendix pp 8, 37–38), in which all prognostic markers identified in QUASAR 2 (except GNAS mutations) had been assessed. A multivariable analysis incorporating the same clinical and molecular variables and co-variables showed that, in agreement with the QUASAR 2 analysis, BRAF mutation, TP53 mutation, and mutation burden were associated (p≤0·05) with relapse-free survival, whereas MSI was not (table 2). KRAS mutation also showed a similar prognostic association in the Australian patients to that present in QUASAR 2, but this was not statistically significant. When MSI-positive and ultramutator tumours were excluded from the Australian analysis, KRAS mutation was significantly associated with prognosis, but BRAF mutation was not (table 2).Table 2 Associations between clinicopathological molecular variables and relapse-free survival in the Australian community-based series

t. When MSI-positive and ultramutator tumours were excluded from the Australian analysis, KRAS mutation was significantly associated with prognosis, but BRAF mutation was not (table 2).Table 2 Associations between clinicopathological molecular variables and relapse-free survival in the Australian community-based series All cases univariable (n=296)* All cases multivariable (n=253) MSI negative and non-pathogenic POLE multivariable (n=209) HR 95% CI p value HR 95% CI p value HR 95% CI p value KRAS mutation 1·31 0·92–1·87 0·136 1·51 0·97–2·38 0·066 1·61 1·02–2·59 0·040 BRAF mutation 0·91 0·52–1·64 0·780 2·18 1·08–4·56 0·029 1·79 0·73–4·24 0·204 TP53 mutation 1·19 0·83–1·71 0·334 1·82 1·12–2·73 0·014 1·81 1·09–2·82 0·020 Mutation burden (quartiles) 0·72 0·62–0·85 8·62 ×    10−5 0·78 0·63–0·95 0·014 0·82 0·64–0·93 0·008 MSI 0·39 0·18–0·71 0·003 0·62 0·24–1·44 0·247 .. .. .. Chemotherapy (yes vs no) 1·01 0·71–1·44 0·946 0·60 0·34–0·91 0·019 0·51 0·18–0·90 0·018 Radiotherapy (yes vs no) 1·21 0·50–3·02 0·653 1·33 0·53–3·32 0·546 1·29 0·51–3·20 0·603 T4 vs T1, T2, or T3† 2·19 1·54–3·22 2·01 ×    10−5 2·38 1·57–3·75 6·34 ×    10−5 2·67 1·73–4·21 1·62 ×    10−5 N1 or N2 vs N0† 1·40 0·97–2·08 0·070 1·21 0·71–2·04 0·493 1·19 0·66–2·05 0·597 Cox proportional hazards analysis was done. The univariable analyses were adjusted by T stage, N stage, and treatment group (or two of these if the adjustment variable itself was being assessed). Multivariable analysis was based on all variables shown. Mutation burden was derived from total number of non-synonymous mutations and coding indels, which are most likely to be functionally relevant, but similar results were obtained when other somatic variants were also included (appendix). POLE proofreading mutation is not shown as a prognostic variable because of the low frequency of those cancers (appendix). BRAF was tested only for the common V600E variant. GNAS was not tested. MSI=microsatellite instability. HR=hazard ratio.

lts were obtained when other somatic variants were also included (appendix). POLE proofreading mutation is not shown as a prognostic variable because of the low frequency of those cancers (appendix). BRAF was tested only for the common V600E variant. GNAS was not tested. MSI=microsatellite instability. HR=hazard ratio. * Missing data for KRAS (n=9), BRAF (n=11), TP53 (n=10), mutation burden (n=11), MSI (n=1), and radiotherapy (n=21). † According to TNM tumour classification. A combined analysis of the QUASAR 2 and Australian cohorts (n=807), showed that mutations in KRAS, BRAF, and TP53, and lower mutation burden were all independently associated with poor prognosis, whereas MSI was not (figure 1; table 3; appendix p 20). Exclusion of MSI-positive and ultramutator cancers did not affect our findings (table 3). No significant heterogeneity was noted between cohorts and our model persisted in Australian patients treated with chemotherapy (data not shown).Figure 1 Relapse-free survival in the combined QUASAR 2 and Australian cohorts by mutation burden from gene-panel analysis (n=672) Burden data are shown by quartile (highest burden in quartile 4). Cancers that were positive for microsatellite instability or with pathogenic POLE mutations were excluded. Cox proportional hazards model results are shown for univariable and multivariable analyses with quartile 1–4 as a continuous variable and other co-variables as per table 3. The numbers in each quartile are not equal because of ties in mutation burden. HR=hazard ratio.

y or with pathogenic POLE mutations were excluded. Cox proportional hazards model results are shown for univariable and multivariable analyses with quartile 1–4 as a continuous variable and other co-variables as per table 3. The numbers in each quartile are not equal because of ties in mutation burden. HR=hazard ratio. Table 3 Associations between clinicopathological molecular variables and relapse-free survival in the combined QUASAR 2 and Australian community-based series population

y or with pathogenic POLE mutations were excluded. Cox proportional hazards model results are shown for univariable and multivariable analyses with quartile 1–4 as a continuous variable and other co-variables as per table 3. The numbers in each quartile are not equal because of ties in mutation burden. HR=hazard ratio. Table 3 Associations between clinicopathological molecular variables and relapse-free survival in the combined QUASAR 2 and Australian community-based series population All cases univariable (n=807)* All cases multivariable (n=764) MSI-negative and non-pathogenic POLE multivariable (n=652) HR 95% CI p value HR 95% CI p value HR 95% CI p value KRAS mutation 1·40 1·10–1·78 0·006 1·74 1·31–2·29 1·21 × 10−4 1·88 1·40–2·51 2·11 × 10−5 BRAF mutation 1·23 0·88–1·72 0·231 2·21 1·47–3·29 1·02 × 10−4 2·32 1·50–3·58 1·49 × 10−4 TP53 mutation 1·30 1·01–1·67 0·039 1·65 1·24–2·19 4·67 × 10−4 1·68 1·24–2·26 0·001 Mutation burden (quartiles) 0·82 0·74–0·92 5·1 × 10−4 0·8 0·70–0·91 0·001 0·84 0·74–0·94 0·004 MSI 0·58 0·38–0·89 0·012 0·8 0·46–1·35 0·399 .. .. .. Cohort plus treatment QUASAR 2 capecitabine Reference .. .. Reference .. .. Reference .. .. Cohort plus treatment QUASAR 2 bevacizumab plus capecitabine 1·45 1·04–2·03 0·029 1·44 1·02–2·01 0·034 1·53 1·07–2·18 0·019 Cohort plus treatment Australia no chemotherapy 2·04 1·4–2·98 2·2 × 10−4 3·48 2·28–5·30 7·04 × 10−9 4·05 2·58–6·34 9·96 × 10−10 Cohort plus treatment Australia chemotherapy 2·06 1·45–2·93 5·61 × 10−6 1·75 1·18–2·58 0·005 1·88 1·25–2·83 0·002 Radiotherapy (yes vs no) 1·56 0·64–3·78 0·326 1·37 0·54–3·41 0·503 1·3 0·51–3·24 0·579 T4 vs T1, T2, or T3† 1·81 1·42–2·29 1·30 × 10−6 2·19 1·68–2·83 3·03 × 10−9 2·36 1·80–3·09 4·38 × 10−10 N1 or N2 vs N0† 1·45 1·11–1·89 0·006 1·63 1·21–2·20 0·001 1·68 1·21–2·30 0·002 Cox proportional hazards analysis was done. The univariable analyses were adjusted by T stage, N stage, and treatment group (or two of these if the adjustment variable itself was being assessed). Multivariable analysis was based on all variables shown. Mutation burden was derived from total number of non-synonymous mutations and coding indels, which are most likely to be functionally relevant, but similar results were obtained when other somatic variants were also included (appendix). POLE proofreading mutation is not shown as a prognostic variable because of the low frequency of those cancers (appendix). Mutation burden quartile was derived separately for the QUASAR 2 and Australian cohorts because of the different content of the two panels. The cohort/treatment variables are categorical. MSI=microsatellite instability. HR=hazard ratio.

as a prognostic variable because of the low frequency of those cancers (appendix). Mutation burden quartile was derived separately for the QUASAR 2 and Australian cohorts because of the different content of the two panels. The cohort/treatment variables are categorical. MSI=microsatellite instability. HR=hazard ratio. * Missing data from Australian cohort for KRAS (n=9), BRAF (n=11), TP53 (n=10), mutation burden (n=11), MSI (n=1), and radiotherapy (n=21). † According to TNM tumour classification. We compared a prognostic model based on the gold standard of clinicopathological variables and MSI with our new model incorporating clinical variables, mutation burden, and driver mutations in KRAS, BRAF, and TP53. In both QUASAR 2 and the Australian cohort, our new model was significantly better (p=0·00004 and p=0·0057, respectively, based on the likelihood ratio test). A 10% leave-out cross-validation analysis showed these analyses to be robust (appendix p 5).

bles, mutation burden, and driver mutations in KRAS, BRAF, and TP53. In both QUASAR 2 and the Australian cohort, our new model was significantly better (p=0·00004 and p=0·0057, respectively, based on the likelihood ratio test). A 10% leave-out cross-validation analysis showed these analyses to be robust (appendix p 5). We explored the prognostic model separately in stage II (n=266) and stage III (n=499) colorectal cancers and found that the model was significant (p=7·3 × 10−8) only in stage III disease (appendix pp 21–22), but HRs were similar in both stages. Correspondingly, despite inherently reduced power, an analysis by tumour location (proximal colon, distal colon, rectum) showed similar HRs for all biomarkers across sites, even after exclusion of hypermutated tumours (appendix pp 23–25). Additionally, formal assessment of interactions between individual biomarkers and stage or tumour location showed no evidence of significant deviation from a log-additive model (data not shown).

rectum) showed similar HRs for all biomarkers across sites, even after exclusion of hypermutated tumours (appendix pp 23–25). Additionally, formal assessment of interactions between individual biomarkers and stage or tumour location showed no evidence of significant deviation from a log-additive model (data not shown). On the basis of previous reports,6, 7, 8, 9, 10, 11 we investigated the prognostic associations of KRAS and BRAF mutations in relation to MSI status by pooling data from an extended set of the QUASAR 2 and Australian cohorts, including an additional 676 colorectal cancers from QUASAR 2 and 362 stage II or III colorectal cancers from the Australian cohort (n=1732). In multivariable analysis, MSI was associated with good prognosis (HR 0·45 [95% CI 0·31–0·64]; p=0·00001), and KRAS (1·22 [1·01–1·48]; p=0·035] and BRAF (1·53 [1·14–2·04]; p=0·004) mutations were both associated with poor prognosis (appendix p 26). Because the strong co-variation of these biomarkers could have confounded or obscured prognostic effects, we added multiplicative interaction terms between MSI and mutations in KRAS and BRAF to the multivariable model. Both of these interactions were significant (p=0·003 and p=0·023, respectively), suggesting differential prognostic effects.

ariation of these biomarkers could have confounded or obscured prognostic effects, we added multiplicative interaction terms between MSI and mutations in KRAS and BRAF to the multivariable model. Both of these interactions were significant (p=0·003 and p=0·023, respectively), suggesting differential prognostic effects. Accordingly, we explored different combinations of MSI, KRAS mutation, and BRAF mutation. Compared with triple-negative (ie, MSI-negative, KRAS and BRAF wild-type) cancers, MSI-negative tumours with KRAS (HR 1·35 [95% CI 1·11–1·64]; p=0·003) or BRAF (2·02 [1·47–2·76]; 1·20 × 10−5) mutations were associated with worse prognosis (table 4, figure 2). By contrast, MSI-positive colorectal cancers with KRAS (HR 0·28 [95% CI 0·09–0·89]; p=0·028) or BRAF (0·55 [0·35–0·90]; p=0·017) mutations were associated with a significantly better prognosis than the triple negatives (table 4), although the difference was not significant compared with MSI-positive colorectal cancers without KRAS or BRAF mutations. The six main subgroups combining MSI, KRAS, and BRAF had consistent effects between the QUASAR 2 and Australian cohorts (data not shown).Figure 2 Relapse-free survival by combinations of MSI and mutations in KRAS and BRAF in the combined extended QUASAR 2 and Australian cohorts Cancers with pathogenic POLE mutations were excluded. MSI=microsatellite instability. Table 4 Prognosis associated with subgroups by KRAS mutation, V600E BRAF mutation, and MSI in all cohorts (n=1732)

Accordingly, we explored different combinations of MSI, KRAS mutation, and BRAF mutation. Compared with triple-negative (ie, MSI-negative, KRAS and BRAF wild-type) cancers, MSI-negative tumours with KRAS (HR 1·35 [95% CI 1·11–1·64]; p=0·003) or BRAF (2·02 [1·47–2·76]; 1·20 × 10−5) mutations were associated with worse prognosis (table 4, figure 2). By contrast, MSI-positive colorectal cancers with KRAS (HR 0·28 [95% CI 0·09–0·89]; p=0·028) or BRAF (0·55 [0·35–0·90]; p=0·017) mutations were associated with a significantly better prognosis than the triple negatives (table 4), although the difference was not significant compared with MSI-positive colorectal cancers without KRAS or BRAF mutations. The six main subgroups combining MSI, KRAS, and BRAF had consistent effects between the QUASAR 2 and Australian cohorts (data not shown).Figure 2 Relapse-free survival by combinations of MSI and mutations in KRAS and BRAF in the combined extended QUASAR 2 and Australian cohorts Cancers with pathogenic POLE mutations were excluded. MSI=microsatellite instability. Table 4 Prognosis associated with subgroups by KRAS mutation, V600E BRAF mutation, and MSI in all cohorts (n=1732) Hazard ratio 95% CI p value KRAS wild-type, BRAF wild-type, MSI negative Reference .. .. KRAS mutated, BRAF wild-type, MSI negative 1·35 1·11–1·64 0·003 KRAS wild-type, BRAF mutated, MSI negative 2·02 1·47–2·76 1·20 × 10−5 KRAS wild-type, BRAF wild-type, MSI positive 0·90 0·56–1·45 0·670 KRAS mutated, BRAF wild-type, MSI positive 0·28 0·09–0·89 0·028 KRAS wild-type, BRAF mutated, MSI positive 0·55 0·35–0·90 0·017 T4 vs T1, T2, or T3* 2·26 1·88–2·71 3·32 × 10−18 N1 or N2 vs N0* 2·07 1·65–2·59 2·62 × 10−10 The p value for the interaction between MSI and BRAF is 0·003; the p value for the interaction between MSI and KRAS is 0·023. Results are from multivariable analysis adjusted by cohort groups. Six patients in very rare subgroups are not shown. MSI=microsatellite instability.

N1 or N2 vs N0* 2·07 1·65–2·59 2·62 × 10−10 The p value for the interaction between MSI and BRAF is 0·003; the p value for the interaction between MSI and KRAS is 0·023. Results are from multivariable analysis adjusted by cohort groups. Six patients in very rare subgroups are not shown. MSI=microsatellite instability. * According to TNM tumour classification. Although MSI was not an independent prognostic marker when mutation burden was also assessed, it was prognostic in the absence of information about mutation burden (appendix p 26). We therefore explored whether new prognostic groups within the larger MSI-negative subset could be identified with KRAS, BRAF, and TP53, given that TP53 mutation remained an independent prognostic marker when MSI-positive and ultramutator colorectal cancers were excluded from the main analysis based on gene panels (table 1). Within the MSI-negative colorectal cancer set (n=991), tumours with BRAF and TP53 mutations had a particularly poor prognosis (HR 3·08 [95% CI 1·88–5·03]; p=7·12 × 10−6; figure 3; appendix p 27). Neither the interaction between TP53 and BRAF (HR 2·21 [95% CI 0·97–5·03]; p=0·058), nor that between TP53 and KRAS (1·13 [0·71–1·80]; p=0·62) were significant.Figure 3 Relapse-free survival by combinations of mutations in KRAS, BRAF, and TP53 in MSI-negative tumours in the combined extended QUASAR 2 and Australian cohorts Cancers that were MSI positive or with pathogenic POLE mutations were excluded. MSI=microsatellite instability.

Although MSI was not an independent prognostic marker when mutation burden was also assessed, it was prognostic in the absence of information about mutation burden (appendix p 26). We therefore explored whether new prognostic groups within the larger MSI-negative subset could be identified with KRAS, BRAF, and TP53, given that TP53 mutation remained an independent prognostic marker when MSI-positive and ultramutator colorectal cancers were excluded from the main analysis based on gene panels (table 1). Within the MSI-negative colorectal cancer set (n=991), tumours with BRAF and TP53 mutations had a particularly poor prognosis (HR 3·08 [95% CI 1·88–5·03]; p=7·12 × 10−6; figure 3; appendix p 27). Neither the interaction between TP53 and BRAF (HR 2·21 [95% CI 0·97–5·03]; p=0·058), nor that between TP53 and KRAS (1·13 [0·71–1·80]; p=0·62) were significant.Figure 3 Relapse-free survival by combinations of mutations in KRAS, BRAF, and TP53 in MSI-negative tumours in the combined extended QUASAR 2 and Australian cohorts Cancers that were MSI positive or with pathogenic POLE mutations were excluded. MSI=microsatellite instability. Discussion In this study, we used overlapping cancer gene mutation panels to analyse a cohort from a high-quality clinical trial of colorectal cancers treated with curative intent and a validation cohort. In multivariable analysis incorporating known clinicopathological prognostic factors, we showed that low overall mutation burden and mutations in KRAS, BRAF, and TP53 were independently associated with decreased relapse-free survival after colorectal cancer treated with curative intent. These findings were present both in the clinical trial cohort and in the Australian validation set of community-based patients. The fact that we found no molecular marker for bevacizumab response in QUASAR 2 or chemotherapy response in the Australian cohorts suggests that the markers we identified are prognostic, although formal demonstration of this hypothesis is difficult because most patients received fluorouracil-based chemotherapy.

tients. The fact that we found no molecular marker for bevacizumab response in QUASAR 2 or chemotherapy response in the Australian cohorts suggests that the markers we identified are prognostic, although formal demonstration of this hypothesis is difficult because most patients received fluorouracil-based chemotherapy. Use of prognostic molecular markers in management of solid tumours is still not widespread, partly because of a lack of validated markers and partly because of differences between studies, leading to uncertainty about which markers to use and their estimated effect sizes. Although molecular indicators of colorectal cancer prognosis have been assessed in several large studies, analyses in most cases have been restricted to a handful of markers.

ed markers and partly because of differences between studies, leading to uncertainty about which markers to use and their estimated effect sizes. Although molecular indicators of colorectal cancer prognosis have been assessed in several large studies, analyses in most cases have been restricted to a handful of markers. The complexity of associations between mutations and colorectal cancer prognosis is arguably reflected by the generally stronger associations of markers in our multivariable than in univariable analyses. Furthermore, MSI was generally not prognostic in our analyses, because its effects were captured by mutation burden (somatic single nucleotide variants and small indels). However, mutation burden not only strongly co-varied with MSI and POLE, but also provided prognostic information in MSI-negative colorectal cancers. Although high mutation burden has been associated with good colorectal cancer prognosis in the context of MSI and POLE proofreading deficiency,4 this relation has not previously been shown for colorectal cancers without those forms of genomic instability. Similar data for other tumour types are few,18, 19, 20 although in other cancers with generally high mutation burdens but without specific forms of genomic instability, such as lung carcinoma and melanoma, mutation burden has predicted response to immune checkpoint inhibitors.21, 22

those forms of genomic instability. Similar data for other tumour types are few,18, 19, 20 although in other cancers with generally high mutation burdens but without specific forms of genomic instability, such as lung carcinoma and melanoma, mutation burden has predicted response to immune checkpoint inhibitors.21, 22 In our study, undetected hypermutator or ultramutator cancers could have contributed to the mutation burden association, although the frequencies of MSI and POLE mutations that we recorded were typical of other studies,4 and we identified a monotonic relationship between mutation burden quartile and relapse-free survival. Another potential cause of the mutation burden association was non-excluded deamination artifacts if they happened to be associated with an unknown factor correlated with good prognosis. However, we made strenuous efforts to exclude those artifacts, no plausible explanatory causes such as tumour age were detectable within QUASAR 2, and the Australian validation cohort analyses were done in fresh frozen tissue, which was unlikely to have deamination. In our study, the association between prognosis and mutation burden was sufficiently strong that even a modestly sized gene panel should pick it up, suggesting that it was representative of mutation burden in the exome.23 The underlying reason for that association is unclear, although anti-tumour immune responses are evidently the prime candidate.18, 19, 20

rognosis and mutation burden was sufficiently strong that even a modestly sized gene panel should pick it up, suggesting that it was representative of mutation burden in the exome.23 The underlying reason for that association is unclear, although anti-tumour immune responses are evidently the prime candidate.18, 19, 20 We showed a strong negative association between driver mutations in TP53 and ATM, two key mediators in the DNA damage response, suggesting that these mutations are alternative DNA damage response inactivators. We also found a positive association between ATM and PTEN mutations; PTEN is phosphorylated by ATM in response to DNA-damaging agents, thus inducing autophagy.24 Mutations in FBXW7 and CTNNB1 were associated with high-grade disease, the latter suggesting that activation of the Wnt pathway through CTNNB1 rather than APC mutation might predispose to poorly differentiated colorectal cancers.

EN is phosphorylated by ATM in response to DNA-damaging agents, thus inducing autophagy.24 Mutations in FBXW7 and CTNNB1 were associated with high-grade disease, the latter suggesting that activation of the Wnt pathway through CTNNB1 rather than APC mutation might predispose to poorly differentiated colorectal cancers. The interplay between KRAS, BRAF, and TP53 mutations, MSI, and mutation burden in our data set is intriguing. These mutations co-vary strongly (appendix), and are additionally associated with other molecular variables. Thus, to decipher primary associations is extremely challenging. Nevertheless, our study strongly supports the reported poor prognosis of MSI-negative colorectal cancers with KRAS or BRAF mutations6, 7, 8, 9, 10, 11 compared with MSI-negative colorectal cancers wild-type for these genes and unselected MSI-positive colorectal cancers. Additionally, we showed that KRAS or BRAF mutation could be associated with improved prognosis in MSI-positive colorectal cancers. TP53 has not previously been consistently reported as a prognostic marker for colorectal cancer in the curative setting, but very few large studies have included a sufficiently comprehensive molecular analysis of KRAS, BRAF, TP53, and MSI. Notably, addition of these four prognostic markers improved outcome prediction compared with current clinical guidelines based on MSI.

a prognostic marker for colorectal cancer in the curative setting, but very few large studies have included a sufficiently comprehensive molecular analysis of KRAS, BRAF, TP53, and MSI. Notably, addition of these four prognostic markers improved outcome prediction compared with current clinical guidelines based on MSI. The strengths of our study are that several potential biomarkers were screened in a large, high-quality clinical trial and a community-based cohort. We have carefully done quality-control analysis to derive high-quality mutation calls. For mutation burden, the study is arguably limited by the size of the gene panels used, and a larger panel or exome and genome sequencing might detect even stronger associations with prognosis. Limitations include the low numbers of patients with stage II disease in the sample set, which means that the utility of our model in such patients remains formally unproven. Although we found our model to be significant only in stage III disease (appendix p 22), HRs were similar in both stages, suggesting that the lack of significance for stage II disease was the result of lower power in that set. Furthermore, we cannot formally distinguish between the model being prognostic or predictive for fluorouracil response. Another potential weakness is the different treatment regimens used in each cohort, although regimen was incorporated as a co-variable in the analyses. Finally, our study might have suboptimal power to draw firm conclusions about outcomes in small patient groups or subgroups, such as those with combinations of several molecular variables.

akness is the different treatment regimens used in each cohort, although regimen was incorporated as a co-variable in the analyses. Finally, our study might have suboptimal power to draw firm conclusions about outcomes in small patient groups or subgroups, such as those with combinations of several molecular variables. Advances in molecular testing hold considerable promise for the delivery of precision cancer medicine, but their clinical use to date has largely been limited to analysis of small numbers of actionable variants. In colorectal cancer, these include KRAS and NRAS mutation testing for prediction of resistance to anti-EGFR therapies,25 and MSI, which identifies stage II tumours with excellent prognosis26 and stage IV tumours likely to respond to immune checkpoint inhibition. Our findings show that the use of even a modest-sized gene panel can provide clinically useful information beyond individual driver mutations. Tumour mutation burden displaced MSI and POLE as a marker of prognosis in multivariable analysis, thus extending the group of colorectal cancers with good prognoses to include those with high mutation burden in the absence of a specific underlying mutator phenotype. Although we were unable to test whether mutational load is predictive for immunotherapy response, this correlation is well documented in other tumour types, including melanoma and lung and ovarian cancers.27 Accordingly, our results suggest that the use of tumour mutation burden as a prognostic and predictive marker in colorectal cancer is worthy of further exploration, beyond tumours with MSI or POLE mutation. Other genome-wide molecular phenotypes, such as mutational signatures,28 are also likely to have a role in cancer management in the future.

uggest that the use of tumour mutation burden as a prognostic and predictive marker in colorectal cancer is worthy of further exploration, beyond tumours with MSI or POLE mutation. Other genome-wide molecular phenotypes, such as mutational signatures,28 are also likely to have a role in cancer management in the future. Supplementary Material Supplementary appendix

uggest that the use of tumour mutation burden as a prognostic and predictive marker in colorectal cancer is worthy of further exploration, beyond tumours with MSI or POLE mutation. Other genome-wide molecular phenotypes, such as mutational signatures,28 are also likely to have a role in cancer management in the future. Supplementary Material Supplementary appendix Acknowledgments This study was funded in part by the UK Technology Strategy Board and supported by the National Institute for Health Research Oxford Biomedical Research Centre, Cancer Research UK, a Cancer Australia Project Grant (APP1120882), a Cancer Council Victoria Grant-in-Aid (APP1060964), Melbourne Bioinformatics at the University of Melbourne (VR0310), the Ludwig Institute for Cancer Research, and the Victorian Government's Operational Infrastructure Support Program. The views expressed are the authors' and not necessarily those of the Department of Health, National Institute for Health Research, or Oxford Biomedical Research Centre. ED is supported by the UK Medical Research Council and Cancer Research UK stratified medicine consortium for colorectal cancer (S:CORT). DNC is supported by an Academy of Medical Sciences–Health Foundation Clinician Scientist Fellowship. OS is a National Health and Medical Research Council R D Wright Biomedical Fellow (APP1062226). MJP is supported by a Cancer Therapeutics CRC Top Up PhD Scholarship and an Australian Government Research Training Program Scholarship. We thank the patients from QUASAR 2 and the Australian cohort who consented to tumour analysis, the Victorian Cancer BioBank for provision of specimens, and Biogrid Australia for access to clinical data.

a Cancer Therapeutics CRC Top Up PhD Scholarship and an Australian Government Research Training Program Scholarship. We thank the patients from QUASAR 2 and the Australian cohort who consented to tumour analysis, the Victorian Cancer BioBank for provision of specimens, and Biogrid Australia for access to clinical data. Contributors ED, JCT, and IT designed the study. RLW, JS, OS, JCT, and IT acquired funding. RLW, NJH, PG, DK, RK, and OS provided resources. ED, CC, PJK, MJP, MMP, SM, MP, RLW, NJH, PG, HA, DO, HW, JW, ET, YB, KK, ECJ, CP, DNC, MN, HED, and OS collected the data, which were curated by ED, RLW, and OS, analysed by ED, DM, MMP, OS, and IT, and interpreted by ED, OS, JCT, and IT. ED and IT wrote the Article, which was read and approved by all authors. Declaration of interests RK reports personal fees from Oxford Cancer Biomarkers (Oxford, UK), outside the submitted work. All other authors declare no competing interests.

Introduction Crohn's disease is a chronic, relapsing inflammatory bowel disease. Estimates of the frequency of surgical resection in Crohn's disease vary. Historical data suggest that up to 60% of patients need a major abdominal resection within 10 years of diagnosis.1 However, more recent population-based data suggest this figure is as low as 29% at 7 years.2 Postoperative recurrence is common within 2 years, in the form of endoscopic signs (72–98% of patients) or clinical symptoms (37–70%), with the proportion of patients needing surgery increasing by 5% per year.3, 4 Research in context Evidence before this study

Introduction Crohn's disease is a chronic, relapsing inflammatory bowel disease. Estimates of the frequency of surgical resection in Crohn's disease vary. Historical data suggest that up to 60% of patients need a major abdominal resection within 10 years of diagnosis.1 However, more recent population-based data suggest this figure is as low as 29% at 7 years.2 Postoperative recurrence is common within 2 years, in the form of endoscopic signs (72–98% of patients) or clinical symptoms (37–70%), with the proportion of patients needing surgery increasing by 5% per year.3, 4 Research in context Evidence before this study There remains uncertainty about the efficacy of thiopurines in patients with postoperative Crohn's disease. We searched the Cochrane Central Register of Controlled Trials until May 24, 2016, and PubMed from Jan 1, 1974, to May 24, 2016, with the terms “(azathioprine OR mercaptopurine OR thiopurine) AND Crohn's AND (postoperative OR resection OR hemicolectomy OR ileectomy OR surgical procedures OR surgery) AND trial”, with no language restrictions. We identified two previous systematic reviews with meta-analyses comparing thiopurines with placebo, both published by the Cochrane Collaboration. These reviews differed in their choice of timepoint to assess outcome and in their handling of loss to follow-up. The earlier Cochrane review compared clinical recurrence at a standard time of 12 months across all studies and used the number of patients with a clinical relapse as the outcome measure. Clinical relapse differed significantly between thiopurines and placebo (risk ratio 0·59, 95% CI 0·38–0·92, favouring thiopurines). The more recent Cochrane meta-analysis used the end of study, which varied between 1 year and 2 years, and regarded anyone who did not complete the study as a treatment failure. This study reported a benefit for thiopurines compared with placebo (risk ratio 0·74, 95% CI 0·58–0·94). The Grading of Recommendations Assessment, Development and Evaluation score for the evidence was low. No further published randomised controlled trials were identified that compared thiopurines with placebo since this meta-analysis.

ed a benefit for thiopurines compared with placebo (risk ratio 0·74, 95% CI 0·58–0·94). The Grading of Recommendations Assessment, Development and Evaluation score for the evidence was low. No further published randomised controlled trials were identified that compared thiopurines with placebo since this meta-analysis. Added value of this study TOPPIC is, to our knowledge, the largest randomised controlled study of thiopurines for postoperative prevention of Crohn's disease, and the largest interventional study of any kind for this indication, with 240 patients enrolled. We found no significant difference between mercaptopurine and placebo for the primary endpoint of clinical recurrence of Crohn's disease (Crohn's Disease Activity Index >150 plus 100-point increase in score) and the need for anti-inflammatory rescue treatment or primary surgical intervention. Smoking was confirmed as the only factor predictive of disease recurrence. A subgroup analysis revealed that mercaptopurine was effective at reducing the incidence of clinical recurrence within 3 years of surgery in smokers, but not in non-smokers (pinteraction=0·018). Implications of all the available evidence

TOPPIC is, to our knowledge, the largest randomised controlled study of thiopurines for postoperative prevention of Crohn's disease, and the largest interventional study of any kind for this indication, with 240 patients enrolled. We found no significant difference between mercaptopurine and placebo for the primary endpoint of clinical recurrence of Crohn's disease (Crohn's Disease Activity Index >150 plus 100-point increase in score) and the need for anti-inflammatory rescue treatment or primary surgical intervention. Smoking was confirmed as the only factor predictive of disease recurrence. A subgroup analysis revealed that mercaptopurine was effective at reducing the incidence of clinical recurrence within 3 years of surgery in smokers, but not in non-smokers (pinteraction=0·018). Implications of all the available evidence Combining our data with those included in the previous Cochrane meta-analyses derives a risk ratio of 0·57 (0·38–0·85) in favour of mercaptopurine for the prevention of post-operative Crohn's disease (appendix p 11). We confirm that smoking affects the clinical course of Crohn's disease, as well as response to treatment, whereas no differences were reported by age, sex, or a history of previous surgery. Smoking cessation should be a priority in patients with Crohn's disease after surgery.

-operative Crohn's disease (appendix p 11). We confirm that smoking affects the clinical course of Crohn's disease, as well as response to treatment, whereas no differences were reported by age, sex, or a history of previous surgery. Smoking cessation should be a priority in patients with Crohn's disease after surgery. Strategies to prevent or delay postoperative recurrence of Crohn's disease are of major clinical importance. However, there is a paucity of evidence to support any particular drug treatment strategy.5, 6 Azathioprine and mercaptopurine have an established role in inducing remission, and in the maintenance of medically induced remission, in patients with Crohn's disease. These drugs are recommended in treatment algorithms for patients at high risk of postoperative relapse,4 but the evidence to support this, and the evidence that clinical parameters can predict patients at high risk of relapse, is weak. A meta-analysis7 showed that efficacy data for thiopurines in this setting were inconclusive and, aside from smoking, there were no consistent predictors of postoperative relapse. A Cochrane review8 also concluded that the evidence supporting thiopurines for the reduction of endoscopic and clinical recurrence was insufficient because of the small numbers of patients included and flawed study designs. The value of thiopurine metabolites in postoperative Crohn's disease is unknown. The role of biological treatments in postoperative Crohn's disease has received substantial attention. After smaller randomised studies of infliximab,9, 10 findings from the POCER study11 showed that targeted escalation of immune-modulatory treatment (ie, thiopurines followed by adalimumab) in patients with early endoscopic evidence of recurrence might delay subsequent endoscopic, although not clinical, recurrence.

ion. After smaller randomised studies of infliximab,9, 10 findings from the POCER study11 showed that targeted escalation of immune-modulatory treatment (ie, thiopurines followed by adalimumab) in patients with early endoscopic evidence of recurrence might delay subsequent endoscopic, although not clinical, recurrence. We therefore aimed to establish whether mercaptopurine, compared with placebo, can prevent or delay postoperative clinical recurrence of Crohn's disease that needs anti-inflammatory rescue treatment or surgery. Methods Study design and participants TOPPIC was a randomised, placebo-controlled, double-blind study done at 29 secondary and tertiary UK hospitals. The study was approved by the Scotland “A” Research Ethics Committee.

We therefore aimed to establish whether mercaptopurine, compared with placebo, can prevent or delay postoperative clinical recurrence of Crohn's disease that needs anti-inflammatory rescue treatment or surgery. Methods Study design and participants TOPPIC was a randomised, placebo-controlled, double-blind study done at 29 secondary and tertiary UK hospitals. The study was approved by the Scotland “A” Research Ethics Committee. Patients aged at least 16 years (Scotland) or 18 years (England and Wales) who had a diagnosis of Crohn's disease12 and an ileocolic or small bowel resection within the preceding 3 months were eligible for inclusion. Key exclusion criteria were residual active Crohn's disease present after surgery, known intolerance or hypersensitivity to thiopurines, known need for further surgery, strictureplasty alone, formation of a stoma, active or untreated malignancy, absent thiopurine methyltransferase activity, substantial abnormalities of liver function tests or full blood count, and pregnancy. The appendix (p 4) lists all inclusion and exclusion criteria. Before randomisation, postoperative infections were treated and existing treatments for Crohn's disease stopped. The protocol was amended on Sept 28, 2010, to include patients successfully treated for a malignancy and in remission for at least 5 years and to exclude those receiving treatment for active Crohn's disease at random allocation. All patients provided written informed consent before enrolment.

r Crohn's disease stopped. The protocol was amended on Sept 28, 2010, to include patients successfully treated for a malignancy and in remission for at least 5 years and to exclude those receiving treatment for active Crohn's disease at random allocation. All patients provided written informed consent before enrolment. Randomisation and masking Patients were randomly assigned (1:1) to mercaptopurine or identical matched placebo using a computer-generated web-based randomisation system managed by the Edinburgh Clinical Trials Unit (University of Edinburgh, Edinburgh, UK) and stratified according to smoking status at baseline and recruiting site (block sizes of two or four). Patients' details were entered into the randomisation system before random allocation and were concealed at randomisation. Patients and their carers and physicians were masked to the treatment allocation. Blood monitoring results were reviewed by an independent central clinician masked to treatment allocation and to mean corpuscular volume results. The appendix (p 5) details the dose reduction algorithm. To protect masking, investigators were informed that sham dose reductions were planned for patients on placebo. However, on the advice of the data monitoring committee, sham dose reductions did not occur; the investigators were not informed of this. Procedures Patients received once daily oral mercaptopurine, at a dose of 1 mg/kg bodyweight rounded to the nearest 25 mg, or identical matched placebo. Patients with low thiopurine methyltransferase activity were prescribed half the normal dose.

Randomisation and masking Patients were randomly assigned (1:1) to mercaptopurine or identical matched placebo using a computer-generated web-based randomisation system managed by the Edinburgh Clinical Trials Unit (University of Edinburgh, Edinburgh, UK) and stratified according to smoking status at baseline and recruiting site (block sizes of two or four). Patients' details were entered into the randomisation system before random allocation and were concealed at randomisation. Patients and their carers and physicians were masked to the treatment allocation. Blood monitoring results were reviewed by an independent central clinician masked to treatment allocation and to mean corpuscular volume results. The appendix (p 5) details the dose reduction algorithm. To protect masking, investigators were informed that sham dose reductions were planned for patients on placebo. However, on the advice of the data monitoring committee, sham dose reductions did not occur; the investigators were not informed of this. Procedures Patients received once daily oral mercaptopurine, at a dose of 1 mg/kg bodyweight rounded to the nearest 25 mg, or identical matched placebo. Patients with low thiopurine methyltransferase activity were prescribed half the normal dose. Baseline assessments included Crohn's Disease Activity Index (CDAI); patient-reported outcome measures, including the Inflammatory Bowel Disease Questionnaire (IBDQ); a physical examination; and a blood sample for 6-thioguanine nucleotide concentrations (6-thioguanine and 6-methylmercaptopurine; appendix p 6). We also took additional blood samples for genetic and serological analysis and will report those results separately. Treatment was planned for 3 years, with dose adjusted for changes in bodyweight. The appendix (p 6) describes which procedures were done at which timepoints. Blood monitoring was done weekly for the first 6 weeks and thereafter at 6-weekly intervals. Patients with abnormal results had a dose reduction, temporary cessation, or cessation as per a study algorithm (appendix p 5). Patients with persistent nausea or persistent influenza-like symptoms also received a dose reduction, according to the protocol. If abnormal parameters improved after a temporary cessation, treatment was recommenced at a lower dose. At each study visit, the following data were collected: CDAI, physical examination, concomitant medications, and patient-reported outcomes, including the IBDQ (appendix p 6). Samples for assay of faecal calprotectin, 6-thioguanine, and 6-methylmercaptopurine were collected at randomisation and weeks 13, 49, 103, and 157. Faecal samples were stored on site at −80°C and then shipped on dry ice to a central laboratory (Gastrointestinal Laboratory, Western General Hospital, Edinburgh, UK) for analysis with the CALPRO Calprotectin ELISA test (Calpro AS, Lysaker, Norway). Samples were stored in a freezer until the patient exited the study, and all samples for an individual were then tested at the same time. The Edinburgh laboratory has a coefficient of variation of 10% for faecal calprotectin (based on assessments of the entire sample processing pipeline).

(Calpro AS, Lysaker, Norway). Samples were stored in a freezer until the patient exited the study, and all samples for an individual were then tested at the same time. The Edinburgh laboratory has a coefficient of variation of 10% for faecal calprotectin (based on assessments of the entire sample processing pipeline). 6-thioguanine nucleotide and 6-methylmercaptopurine were analysed by the Viapath Purine Research Laboratory (St Thomas' Hospital, London, UK) using a method adapted from Dervieux and colleagues.13 Briefly, thioguanine and methylated mercaptopurine nucleotides in whole blood were hydrolysed to the base by boiling in 15% perchloric acid and detected by ultraviolet absorption after separation on a Waters Ultra-Performance Liquid Chromatography system (Waters Limited, Elstree, Hertfordshire, UK). Colonoscopy was done at 49 and 157 weeks after randomisation.

topurine nucleotides in whole blood were hydrolysed to the base by boiling in 15% perchloric acid and detected by ultraviolet absorption after separation on a Waters Ultra-Performance Liquid Chromatography system (Waters Limited, Elstree, Hertfordshire, UK). Colonoscopy was done at 49 and 157 weeks after randomisation. Outcomes The primary outcome was clinical recurrence, defined as a CDAI score of over 150 and a 100-point increase from baseline, and the need for anti-inflammatory rescue treatment or primary surgical intervention. Secondary outcomes were clinical recurrence, defined as reaching either of the individual components of the primary outcome (ie, either a CDAI score of >150 and a 100-point increase from baseline, or the need for anti-inflammatory rescue treatment or primary surgical intervention); endoscopic recurrence, defined as a Rutgeerts score of at least i2; Crohn's Disease Endoscopic Index of Severity (CDEIS) score;14, 15 and quality of life, measured by changes in IBDQ scores. Adverse events were assessed by investigators at the participating sites using criteria defined in the trial protocol. Statistical analysis A sample size of 234 patients was needed to give 80% power to detect a reduction in the frequency of recurrence from 50% in the placebo group to 30% in the treatment group by 3 years at the 5% level of significance.

Outcomes The primary outcome was clinical recurrence, defined as a CDAI score of over 150 and a 100-point increase from baseline, and the need for anti-inflammatory rescue treatment or primary surgical intervention. Secondary outcomes were clinical recurrence, defined as reaching either of the individual components of the primary outcome (ie, either a CDAI score of >150 and a 100-point increase from baseline, or the need for anti-inflammatory rescue treatment or primary surgical intervention); endoscopic recurrence, defined as a Rutgeerts score of at least i2; Crohn's Disease Endoscopic Index of Severity (CDEIS) score;14, 15 and quality of life, measured by changes in IBDQ scores. Adverse events were assessed by investigators at the participating sites using criteria defined in the trial protocol. Statistical analysis A sample size of 234 patients was needed to give 80% power to detect a reduction in the frequency of recurrence from 50% in the placebo group to 30% in the treatment group by 3 years at the 5% level of significance. Analyses were by intention to treat. For the primary analysis, we used a Cox proportional hazards model with terms for treatment and the variables on which the randomisation was stratified (smoking status and recruitment site), adjusted for baseline values of previous treatment with mercaptopurine and previous treatment with azathioprine. We present adjusted and unadjusted Cox proportional hazard ratios (HRs) for the comparison of mercaptopurine versus placebo (reference), with an HR of less than one suggesting a treatment effect in favour of mercaptopurine. For both primary and secondary outcomes, the adjusted analysis was judged to be the primary analysis. We did predefined subgroup analyses of the primary and secondary outcomes to assess treatment effect in terms of previous medical treatment, previous surgery, smoking status, duration of disease, and age at diagnosis. The interaction between subgroup and treatment was included in the Cox regression model to establish whether the treatment effect differed by subgroup.

y and secondary outcomes to assess treatment effect in terms of previous medical treatment, previous surgery, smoking status, duration of disease, and age at diagnosis. The interaction between subgroup and treatment was included in the Cox regression model to establish whether the treatment effect differed by subgroup. We compared endoscopic recurrence between treatment groups using a χ2 test. CDEIS results at week 157 were compared between treatment groups using a two-sided t test. The same subgroups analysed for the primary and secondary outcomes were also analysed with respect to endoscopic recurrence and CDEIS scores. We produced receiver operating characteristic (ROC) curves to calculate the diagnostic accuracy of faecal calprotectin to predict endoscopic recurrence and remission. The optimum cutoff point was calculated by maximising Youden's J statistic. We incorporated faecal calprotectin and 6-thioguanine separately into a Cox proportional hazards model as time-varying covariates. Quality of life, as measured by the IBDQ, was analysed using a change from baseline repeated measures ANCOVA to assess the effect of treatment over time for the overall mean IBDQ score and also the overall total IBDQ score. Quality of life as measured by the EQ-5D system was summarised by treatment group across study visits.

y of life, as measured by the IBDQ, was analysed using a change from baseline repeated measures ANCOVA to assess the effect of treatment over time for the overall mean IBDQ score and also the overall total IBDQ score. Quality of life as measured by the EQ-5D system was summarised by treatment group across study visits. We excluded missing data from any formal statistical analyses, with the exception of a secondary sensitivity analysis to the complete case analysis of IBDQ data, for which we used several imputation techniques, as described in the statistical analysis plan. A data monitoring committee oversaw the trial. Data were analysed in SAS version 9.4. In a post-hoc analysis, we used the absolute risk reduction experienced by patients on mercaptopurine versus placebo to calculate the number needed to treat (NNT), for benefit and harm, for smokers and non-smokers. This trial is registered with the International Standard Randomised Controlled Trial Register (ISRCTN89489788) and the European Clinical Trials Database (EudraCT number 2006-005800-15). Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

This trial is registered with the International Standard Randomised Controlled Trial Register (ISRCTN89489788) and the European Clinical Trials Database (EudraCT number 2006-005800-15). Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results Between June 6, 2008, and April 23, 2012, 328 patients were screened at 29 centres (appendix p 7), 240 of whom met eligibility criteria, consented to inclusion, and were enrolled and randomly assigned: 128 to mercaptopurine and 112 to placebo (figure 1). There was low patient recruitment (only one or two patients) at several centres, which resulted in only one treatment being assigned at these centres, which created the imbalance in recruitment numbers between treatment groups. All patients received at least one dose of study drug. 146 (61%) were women and 55 (23%) were smokers (table 1). Baseline characteristics were similar between study groups (table 1). 104 (43%) of 240 patients received study drug at the initial dose for the entire 3-year treatment period. The mean treatment period was 22·6 months (SD 13·7): 23·4 months (14·0) in the mercaptopurine group versus 21·8 months (13·4) in the placebo group. 50 (39%) of 128 patients in the mercaptopurine group and 18 (16%) of 112 in the placebo group had a dose reduction in accordance with the trial protocol. Study drug was discontinued in 66 (52%) of 128 patients in the mercaptopurine groups versus 70 (63%) of 112 in the placebo group for the following reasons: adverse events in 80 patients (59%; 39 in the mercaptopurine group and 41 in the placebo group), abnormal blood test results in 18 (13%; 12 and six), early withdrawal in 21 (15%; eight and 13), loss to follow-up in 16 (12%; seven and nine), and death in one (1%; in the placebo group). The appendix (p 8) summarises data completeness for the study visits. Median follow-up was 36·0 months (IQR 27·5–36·0) in the mercaptopurine group and 36·0 months (19·5–36·0) in the placebo group.Figure 1 TOPPIC trial profile

), loss to follow-up in 16 (12%; seven and nine), and death in one (1%; in the placebo group). The appendix (p 8) summarises data completeness for the study visits. Median follow-up was 36·0 months (IQR 27·5–36·0) in the mercaptopurine group and 36·0 months (19·5–36·0) in the placebo group.Figure 1 TOPPIC trial profile Table 1 Demographics and baseline characteristics before randomisation

), loss to follow-up in 16 (12%; seven and nine), and death in one (1%; in the placebo group). The appendix (p 8) summarises data completeness for the study visits. Median follow-up was 36·0 months (IQR 27·5–36·0) in the mercaptopurine group and 36·0 months (19·5–36·0) in the placebo group.Figure 1 TOPPIC trial profile Table 1 Demographics and baseline characteristics before randomisation Mercaptopurine (n=128) Placebo (n=112) Sex Female 79 (62%) 67 (60%) Male 49 (38%) 45 (40%) Age (years) Mean (SD) 39·2 (12·8) 38·2 (13·4) Median (IQR) 38 (28–50) 36 (28–48) Range 17–67 17–75 Age at diagnosis ≤40 years 103 (80%) 87 (78%) >40 years 25 (20%) 23 (21%) Unknown 0 2 (2%) Present smoker Yes* 29 (23%) 26 (23%) No 99 (77%) 86 (77%) Duration of Crohn's disease from diagnosis ≤1 year 37 (29%) 41 (37%) >1 year 91 (71%) 69 (62%) Unknown 0 2 (2%) Duration of Crohn's disease from diagnosis (years)† Mean (SD) 7·7 (9·7) 7·6 (9·5) Median (IQR) 3 (0–11) 4 (0–11) Range 0–39 0–47 Crohn's disease location‡ Ileal 54 (42%) 39 (35%) Colonic 4 (3%) 2 (2%) Ileocolonic 70 (55%) 70 (63%) Previous treatments Mercaptopurine‡ Yes 14 (11%) 5 (4%) No 114 (89%) 106 (95%) Azathioprine‡ Yes 80 (63%) 47 (42%) No 48 (38%) 64 (57%) Either thiopurine‡ Yes 81 (63%) 50 (45%) No 47 (37%) 61 (54%) Infliximab§ Yes 21 (16%) 15 (13%) No 104 (81%) 96 (86%) Methotrexate‡ Yes 8 (6%) 7 (6%) No 120 (94%) 104 (93%) Other corticosteroids‡ Yes 97 (76%) 79 (71%) No 31 (24%) 32 (29%) Any immunosuppressants‡ Yes 112 (88%) 86 (77%) No 16 (13%) 25 (22%) Previous surgery‡ Yes 46 (36%) 28 (25%) No 82 (64%) 83 (74%) Data are number (%), unless otherwise specified. Some percentages do not add up to 100 because of rounding.

0 (94%) 104 (93%) Other corticosteroids‡ Yes 97 (76%) 79 (71%) No 31 (24%) 32 (29%) Any immunosuppressants‡ Yes 112 (88%) 86 (77%) No 16 (13%) 25 (22%) Previous surgery‡ Yes 46 (36%) 28 (25%) No 82 (64%) 83 (74%) Data are number (%), unless otherwise specified. Some percentages do not add up to 100 because of rounding. * Smoked >1 cigarette per day at study entry. † Data missing for two patients in the placebo group. ‡ Data missing for one patient in the placebo group. § Data missing for three patients in the mercaptopurine group and one in the placebo group.

0 (94%) 104 (93%) Other corticosteroids‡ Yes 97 (76%) 79 (71%) No 31 (24%) 32 (29%) Any immunosuppressants‡ Yes 112 (88%) 86 (77%) No 16 (13%) 25 (22%) Previous surgery‡ Yes 46 (36%) 28 (25%) No 82 (64%) 83 (74%) Data are number (%), unless otherwise specified. Some percentages do not add up to 100 because of rounding. * Smoked >1 cigarette per day at study entry. † Data missing for two patients in the placebo group. ‡ Data missing for one patient in the placebo group. § Data missing for three patients in the mercaptopurine group and one in the placebo group. The primary endpoint of clinical recurrence of Crohn's disease and the need for anti-inflammatory rescue treatment or primary surgical intervention occurred in 42 (18%) of 240 patients: 16 (13%) of 128 in the mercaptopurine group versus 26 (23%) of 112 in the placebo group (adjusted HR 0·54, 95% CI 0·27–1·06; p=0·07; unadjusted HR 0·53, 95% CI 0·28–0·99; p=0·046; figure 2). All 42 patients met the CDAI criteria for recurrence and had rescue treatment, five (12%) of whom subsequently went on to have surgery. In predefined subgroup analyses, 15 (27%) of 55 smokers had a clinical recurrence, three (10%) of 29 in the mercaptopurine group and 12 (46%) of 26 in the placebo group (HR 0·13, 95% CI 0·04–0·46), compared with 27 (15%) of 185 non-smokers, 13 (13%) of 99 in the mercaptopurine group and 14 (16%) of 86 in the placebo group (0·90, 0·42–1·94; pinteraction=0·018; figure 3). In a post-hoc analysis, the NNT for benefit was calculated as three for smokers (95% CI 1·7–7·3) and 31 for non-smokers (95% CI NNTbenefit 7·5 to ∞ to NNTharm 14·1) across the entire follow-up period. Previous exposure to treatment, previous surgery, thiopurine status, duration of disease, and age at diagnosis had no effect on the response to study drug (figure 3).Figure 2 Kaplan-Meier plot for the primary outcome of time to clinical recurrence of postoperative Crohn's disease

·1) across the entire follow-up period. Previous exposure to treatment, previous surgery, thiopurine status, duration of disease, and age at diagnosis had no effect on the response to study drug (figure 3).Figure 2 Kaplan-Meier plot for the primary outcome of time to clinical recurrence of postoperative Crohn's disease Figure 3 Unadjusted subgroup analyses of the primary outcome of clinical recurrence of postoperative Crohn's disease 34 (27%) of 128 patients in the mercaptopurine group versus 40 (36%) of 112 in the placebo group experienced the secondary endpoint of clinical recurrence, defined as a CDAI rise or need for rescue treatment or surgery (adjusted HR 0·74, 95% CI 0·44–1·23; p=0·24; unadjusted 0·75, 0·47–1·18; p=0·21). In subgroup analyses, mercaptopurine reduced recurrence in smokers only (pinteraction=0·033; figure 4).Figure 4 Unadjusted subgroup analyses of the secondary outcome of clinical recurrence

CDAI rise or need for rescue treatment or surgery (adjusted HR 0·74, 95% CI 0·44–1·23; p=0·24; unadjusted 0·75, 0·47–1·18; p=0·21). In subgroup analyses, mercaptopurine reduced recurrence in smokers only (pinteraction=0·033; figure 4).Figure 4 Unadjusted subgroup analyses of the secondary outcome of clinical recurrence Of the 208 patients who remained in the study 49 weeks after randomisation, 172 attended for colonoscopy (95 in the mercaptopurine group and 77 in the placebo group), and a Rutgeerts score was available for 168 (91 in the mercaptopurine group and 77 in the placebo group). Of these, 121 (72%; 58 in the mercaptopurine group and 63 in the placebo group) had some form of endoscopic recurrence (score >i0). Of the 161 patients who remained in the study at week 157 after randomisation, 128 (69 in the mercaptopurine group and 59 in the placebo group) underwent a colonoscopy, and a Rutgeerts score was available for 124 (67 in the mercaptopurine group and 57 in the placebo group). Of these, 95 (77%; 47 in the mercaptopurine group and 48 in the placebo group) had some form of endoscopic recurrence (score >i0). 29 (43%) of 67 patients in the mercaptopurine group and 28 (49%) of 57 in the placebo group had endoscopic recurrence with a Rutgeerts score of i2 or greater (p=0·38). We noted a similar pattern at week 49, although this pattern was not formally analysed. Week 157 CDEIS scores did not differ significantly between groups (data not shown). Similarly, none of the CDEIS subgroup analyses showed a significant interaction with treatment (data not shown).

ts score of i2 or greater (p=0·38). We noted a similar pattern at week 49, although this pattern was not formally analysed. Week 157 CDEIS scores did not differ significantly between groups (data not shown). Similarly, none of the CDEIS subgroup analyses showed a significant interaction with treatment (data not shown). Of the 168 patients who had a Rutgeerts score calculated at week 49, faecal calprotectin concentrations were available in 126 patients (71 in the mercaptopurine group and 55 in the placebo group). Of the 124 patients who had a Rutgeerts score calculated at week 157, faecal calprotectin concentrations were available in 88 patients (46 in the mercaptopurine group and 42 in the placebo group; appendix p 12). These data were combined to generate ROC curves to examine test accuracy at predicting endoscopic recurrence and remission. In both scenarios, the faecal calprotectin measurement proved to be an unreliable test, with an area under the curve of 0·70 (95% CI 0·63–0·77) for recurrence and 0·66 (0·58–0·75) for remission. Selecting a faecal calprotectin concentration of 50 μg/g (commonly proposed as an appropriate cutoff concentration to detect inflammation) to predict endoscopic recurrence produced a sensitivity of 84·4% (95% CI 77·0–91·9), specificity of 44·4% (35·6–53·1), positive predictive value (PPV) of 52·4% (44·3–60·5), and negative predictive value (NPV) of 79·7% (70·2–89·2). Increasing the cutoff concentration to 100 μg/g produced a sensitivity of 72·2% (95% CI 63·0–81·5), specificity of 62·1% (53·6–70·6), PPV of 58·0% (48·9–67·2), and NPV of 75·5% (67·1–83·8). The NPV for the prediction of endoscopic remission with a faecal calprotectin concentration of 50 μg/g was 81·4% (95% CI 75·0–87·7) and with a concentration of 100 μg/g it was 83·9% (77·1–90·7; appendix p 9). Analysis of faecal calprotectin as a time-varying covariate suggested that, for every 100-unit increase in faecal calprotectin, the HR for the primary endpoint (data available for 31 [74%] of 42 patients who reached the primary endpoint) increased by 18% (HR 1·18, 95% CI 1·08–1·28; p=0·0002).

(77·1–90·7; appendix p 9). Analysis of faecal calprotectin as a time-varying covariate suggested that, for every 100-unit increase in faecal calprotectin, the HR for the primary endpoint (data available for 31 [74%] of 42 patients who reached the primary endpoint) increased by 18% (HR 1·18, 95% CI 1·08–1·28; p=0·0002). 102 (92%) of 111 patients had 6-thioguanine nucleotide concentrations measured at week 49, and 64 (72%) of 89 who remained on mercaptopurine had concentrations measured at week 157. 6-thioguanine nucleotide concentrations were grouped according to the target therapeutic range (235–450 pmol per 8 × 108 red blood cells). At week 49, 61 (60%) of 102 patients had subtherapeutic concentrations, versus 40 (63%) of 64 at week 157 (appendix p 13). In the corresponding time-varying covariate analysis of 6-thioguanine nucleotide concentrations in patients receiving mercaptopurine, the association with the primary outcome was not significant (HR 0·80, 95% CI 0·57–1·13; p=0·21). IBDQ data were available for 203 (85%) of 240 randomly assigned patients at week 49 (109 in the mercaptopurine group and 94 in the placebo group) and 155 (65%) at week 157 (86 in the mercaptopurine group and 69 in the placebo group). Overall mean or total IBDQ scores did not seem to differ significantly between groups (data not shown).

ailable for 203 (85%) of 240 randomly assigned patients at week 49 (109 in the mercaptopurine group and 94 in the placebo group) and 155 (65%) at week 157 (86 in the mercaptopurine group and 69 in the placebo group). Overall mean or total IBDQ scores did not seem to differ significantly between groups (data not shown). The incidence and types of adverse events were similar in the mercaptopurine and placebo groups (table 2 and appendix p 10). Adverse events caused discontinuation of treatment in 80 (33%) of 240 patients: 39 (30%) of 128 in the mercaptopurine group versus 41 (37%) of 112 in the placebo group. Of the 1747 reported adverse events, four (<1%) were malignancies (three in the mercaptopurine group [two cases of lentigo maligna in the same individual and one case of basal cell carcinoma] and one in the placebo group [breast cancer]), and one patient on placebo died of ischaemic heart disease. 171 (18%) of 947 events in the mercaptopurine group and 184 (23%) of 798 in the placebo group were infections. Mercaptopurine was temporarily stopped in 32 patients because of abnormal blood monitoring results or other side-effects. Placebo was temporarily discontinued in 35 patients for similar reasons.Table 2 Adverse events

%) of 947 events in the mercaptopurine group and 184 (23%) of 798 in the placebo group were infections. Mercaptopurine was temporarily stopped in 32 patients because of abnormal blood monitoring results or other side-effects. Placebo was temporarily discontinued in 35 patients for similar reasons.Table 2 Adverse events Mercaptopurine (n=128) Placebo (n=112) Mild Moderate Severe Total Mild Moderate Severe Total Cancers 0 2 (2%) 0 2 (2%) 0 0 1 (1%) 1 (1%) Abnormal liver function test 0 4 (3%) 0 4 (3%) 0 5 (4%) 0 5 (4%) Gastrointestinal symptoms Abdominal pain 15 (12%) 32 (25%) 19 (15%) 66 (52%) 10 (9%) 42 (38%) 15 (13%) 67 (60%) Constipation or diarrhoea 12 (9%) 19 (15%) 6 (5%) 37 (29%) 7 (6%) 23 (21%) 7 (6%) 37 (33%) Nausea or vomiting 13 (10%) 24 (19%) 8 (6%) 45 (35%) 12 (11%) 16 (14%) 2 (2%) 30 (27%) Other 14 (11%) 16 (13%) 4 (3%) 34 (27%) 12 (11%) 16 (14%) 0 28 (25%) Headache 9 (7%) 17 (13%) 0 26 (20%) 6 (5%) 11 (10%) 3 (3%) 20 (18%) Infections 31 (24%) 45 (35%) 5 (4%) 81 (63%) 24 (21%) 38 (34%) 6 (5%) 68 (61%) Joint pain or arthralgia 13 (10%) 23 (18%) 4 (3%) 40 (31%) 8 (7%) 26 (23%) 2 (2%) 36 (32%) Other 22 (17%) 55 (43%) 8 (6%) 85 (66%) 18 (16%) 35 (31%) 9 (8%) 62 (55%) Pain 7 (5%) 8 (6%) 3 (2%) 18 (14%) 4 (4%) 10 (9%) 3 (3%) 17 (15%) Pancreatitis 0 1 (1%) 0 1 (1%) 0 0 1 (1%) 1 (1%) Rash 14 (11%) 9 (7%) 1 (1%) 24 (19%) 12 (11%) 2 (2%) 0 14 (13%) Worsening of Crohn's disease 6 (5%) 13 (10%) 5 (4%) 24 (19%) 3 (3%) 20 (18%) 6 (5%) 29 (26%) Data are number of patients with one or more adverse event in that category. Patients who had more than one adverse effect in the same category but different severity are counted in the most severe category.

) 2 (2%) 0 14 (13%) Worsening of Crohn's disease 6 (5%) 13 (10%) 5 (4%) 24 (19%) 3 (3%) 20 (18%) 6 (5%) 29 (26%) Data are number of patients with one or more adverse event in that category. Patients who had more than one adverse effect in the same category but different severity are counted in the most severe category. 14 pregnancies were reported during the course of the trial, with 12 healthy outcomes (ie, successful birth and healthy infant). One spontaneous abortion occurred at about 21 weeks gestation in the mercaptopurine group and one congenital anomaly (heart murmur, septal defect, and hydrocephalus) occurred in a child born to a patient in the placebo group. In a post-hoc analysis, complete endoscopic remission (Rutgeerts score i0) was maintained in proportionally more patients on mercaptopurine than placebo at both week 49 and week 157 (appendix pp 14–15). In a subgroup analysis, mercaptopurine was more effective at preventing endoscopic recurrence in patients with previous thiopurine exposure (odds ratio 0·25, 95% CI 0·09–0·70) than in thiopurine-naive patients (3·00, 1·00–9·04; p=0·001) Endoscopic recurrence, defined as Rutgeerts score greater than i0 (ie, anything other than complete remission), was present in 58 (64%) of 91 patients in the mercaptopurine group versus 62 (82%) of 76 in the placebo group at week 49 (p=0·01), and in 47 (70%) of 67 in the mercaptopurine group versus 48 (84%) of 57 in the placebo group at week 157 (p=0·07).

geerts score greater than i0 (ie, anything other than complete remission), was present in 58 (64%) of 91 patients in the mercaptopurine group versus 62 (82%) of 76 in the placebo group at week 49 (p=0·01), and in 47 (70%) of 67 in the mercaptopurine group versus 48 (84%) of 57 in the placebo group at week 157 (p=0·07). Discussion To our knowledge, this is the largest randomised, double-blind study to assess the efficacy of mercaptopurine in the prevention of postoperative Crohn's disease. The primary outcome of clinical recurrence of Crohn's disease (Crohn's Disease Activity Index >150 plus 100-point increase in score) and the need for anti-inflammatory rescue treatment or primary surgical intervention, occurred in 13% of patients in the mercaptopurine group versus 23% in the placebo group (adjusted p=0·07); however, clinical recurrence was significantly more common in smokers, in whom mercaptopurine proved beneficial, with an NNT of three. The secondary outcome of endoscopic recurrence was recorded in a third of patients in each group, with no significant difference between groups; however, in a post-hoc analysis, mercaptopurine was significantly more effective than placebo at maintaining complete endoscopic remission. Although there was no significant difference in the prespecified primary clinical efficacy endpoint, the subgroup analyses provide relevant insights in terms of clinical prediction of response and outcome, and in terms of the challenges in assessing outcome by endoscopic or clinical criteria.

lete endoscopic remission. Although there was no significant difference in the prespecified primary clinical efficacy endpoint, the subgroup analyses provide relevant insights in terms of clinical prediction of response and outcome, and in terms of the challenges in assessing outcome by endoscopic or clinical criteria. Findings from this study confirm that smoking affects the clinical course of Crohn's disease, as well as response to treatment. Of the factors assessed, the primary endpoint was only significantly different between smokers and non-smokers, whereas no differences were reported by age, sex, a history of previous surgery, duration of disease, previous treatments, or thiopurine status. The data highlight the importance of smoking cessation in disease management and support findings from previous studies16, 17, 18 that showed that surgical recurrence increases with the number of cigarettes smoked each day, and that smoking cessation reduces clinical and surgical recurrence. In the present study, treatment with mercaptopurine to delay or prevent postoperative recurrence was particularly effective in people who continue to smoke; thus, in smokers, thiopurine treatment seems to be justified in the early postoperative period. In non-smokers, the data do not provide a sufficiently strong rationale for immediate initiation of treatment in the postoperative period. A considered approach involving colonoscopy in the first 6–12 months is likely to be beneficial in this group.

ine treatment seems to be justified in the early postoperative period. In non-smokers, the data do not provide a sufficiently strong rationale for immediate initiation of treatment in the postoperative period. A considered approach involving colonoscopy in the first 6–12 months is likely to be beneficial in this group. This study is one of the largest to report on endoscopic recurrence of Crohn's disease, and is important for several reasons. First, the incidence of any endoscopic recurrence was 77% at 3 years, which is similar to the 85% reported previously.14 Second, over a 3-year period, there was a poor association between endoscopic and clinical recurrence. There are several possible explanations for this finding, and there is no consensus as to whether to prioritise clinical outcomes over endoscopic outcomes. Third, mercaptopurine seems to maintain complete endoscopic remission (i0), whereas using a cutoff score of at least i2 to define endoscopic recurrence revealed no difference between treatment and placebo groups. Fourth, our study is, to our knowledge, the largest comprehensive assessment of faecal calprotectin in postoperative Crohn's disease. Using a cutoff of 50 μg/g, the NPV for recurrence was 79·7%, which decreased to 75·5% by increasing the cutoff to 100 μg/g. The corresponding NPVs for the prediction of endoscopic remission were 81·4% and 83·9%, respectively. If mucosal integrity is the goal, these values might not provide the confidence to abandon endoscopic assessment. The performance of faecal calprotectin was poorer than reported in the POCER study (NPV of 94%).11 Reported differences between studies are probably due to differences in study methods, since in the POCER study, an endoscopic score of i2 was imputed for all missing values; no imputations were made in our study.

ssment. The performance of faecal calprotectin was poorer than reported in the POCER study (NPV of 94%).11 Reported differences between studies are probably due to differences in study methods, since in the POCER study, an endoscopic score of i2 was imputed for all missing values; no imputations were made in our study. Several important factors should be considered when interpreting these findings. Based on existing data, power calculations estimated a 20% difference between mercaptopurine and placebo groups. Clinical recurrence rates were 23·2% in the placebo group and 12·5% in the treatment group: a difference of 10·7%. These rates are lower than those in studies by Hanauer and colleagues19 (50% treatment and 77% placebo) and Ardizzone and colleagues20 (17% treatment and 28% control), on which the power calculations were based. This marked difference between recurrence rates is probably a result of differing primary outcome definitions; clinical scoring systems advocated to identify disease relapse in clinical trials such as CDAI have flaws, especially in the postoperative setting.21 We used an outcome that was based on a disease activity score (CDAI >150 and a 100-point increase from baseline) and the need for medical treatment. In view of the difficulties of using the CDAI postoperatively, we judged this definition of clinical recurrence to be robust. The Rutgeerts score attempts to make the assessment of endoscopic recurrence an objective exercise, but has never been prospectively validated.14 We selected a score of at least i2 as a secondary endpoint, in line with previous studies, including the study by Hanauer and colleagues.19 However, the limitations of this approach are well recognised; there is little difference between i1 (≤5 aphthous ulcers) and i2 (>5 aphthous ulcers or larger lesions confined to the anastomosis), and inter-observer variation is an issue. In a substudy22 of the TOPPIC trial, inter-observer agreement on 43 endoscopic images was measured by five investigators; complete agreement occurred in only 79% of cases. Although centralised reading could overcome some of the inter-observer variation in endoscopic scoring in future studies, an alternative approach might be to regard complete mucosal healing as the preferred therapeutic target in Crohn's disease, and to identify maintenance of endoscopic remission (i0) as the target in postoperative Crohn's disease. A review of the scoring of endoscopic recurrence is warranted.

c scoring in future studies, an alternative approach might be to regard complete mucosal healing as the preferred therapeutic target in Crohn's disease, and to identify maintenance of endoscopic remission (i0) as the target in postoperative Crohn's disease. A review of the scoring of endoscopic recurrence is warranted. Optimum dosing in all patients was also difficult to achieve in the context of a double-blind study and a protocol-led dose adjustment strategy. Of the 240 patients, 104 (43%) received treatment at the initial dose for the duration of the study. Data available at the end of the study show that about 60% of patients randomly assigned to mercaptopurine were on subtherapeutic doses, and a stronger treatment effect might have been noted had 6-thioguanine nucleotide results been available to optimise the dose during the study. The rates of discontinuation of treatment and withdrawal or loss to follow-up in this study are similar to those in previous work. For example, Ardizzone and colleagues20 reported treatment discontinuations owing to adverse events in 15 (22%) of 69 patients in the azathioprine group versus six (9%) of 69 in the mesalazine group, although data on treatment discontinuation within other trials are not clearly documented. Analysis of these data was on an intention-to-treat basis and therefore the effect of the drug taken at full dose in an individual patient is likely to have been underestimated. In clinical practice, many patients taking thiopurines might be inadvertently under-dosed initially, but are identified on the basis of mean corpuscular volumes or, more recently, available metabolite testing.23

re the effect of the drug taken at full dose in an individual patient is likely to have been underestimated. In clinical practice, many patients taking thiopurines might be inadvertently under-dosed initially, but are identified on the basis of mean corpuscular volumes or, more recently, available metabolite testing.23 Adverse events were noted frequently in both groups but were generally mild. Rates of pancreatitis and malignancy were lower than expected. Unusually for a clinical trial, we did not remove patients who became pregnant during the trial, in keeping with accepted clinical practice. We noted 14 pregnancies, with 12 healthy outcomes. No fetal malformations occurred in the mercaptopurine-treated group. Masked safety monitoring contributed to the validity of the results.

for a clinical trial, we did not remove patients who became pregnant during the trial, in keeping with accepted clinical practice. We noted 14 pregnancies, with 12 healthy outcomes. No fetal malformations occurred in the mercaptopurine-treated group. Masked safety monitoring contributed to the validity of the results. The strengths of this study include the double-blind design, the comparison of symptom scores with endoscopic findings, the assessment of faecal calprotectin in a large number of patients in the postoperative setting, and a demonstration of the potential usefulness of 6-thioguanine nucleotide concentrations in patient management. The study also included patients from 29 UK centres, both secondary and tertiary hospitals, which makes it generalisable beyond specialist centres. Limitations include the absence of therapeutic drug monitoring with dose adjustment, missing colonoscopy data in 20% of eligible patients, and the absence of centralised endoscopy reading. Furthermore, we included CDAI measurement within the primary outcome even though it has been previously criticised in this setting. The 36-item Short Form quality-of-life instrument underwent internal text changes at the time of trial start-up. The reporting of these results was deemed not to be compliant with 36-item Short Form licensing terms and these results are therefore not presented.

though it has been previously criticised in this setting. The 36-item Short Form quality-of-life instrument underwent internal text changes at the time of trial start-up. The reporting of these results was deemed not to be compliant with 36-item Short Form licensing terms and these results are therefore not presented. A definitive study of postoperative prevention of Crohn's disease has proved difficult to undertake. The PREVENT study24 was terminated early because of small numbers of patients reaching the primary outcome. The PREVENT study24 had selective inclusion criteria, and no difference was reported in clinical relapse between those on infliximab and those on placebo at week 76, although an endoscopic effect was noted. A smaller study25 that compared early azathioprine initiation with azathioprine driven by endoscopic findings at week 26 was stopped after 6 years because of slow recruitment, with no meaningful conclusions. Although our study was underpowered to detect the reported treatment effect, and many patients were under-dosed with mercaptopurine, our data nonetheless provide some evidence of efficacy of mercaptopurine in the context of postoperative prevention. Indeed, a meta-analysis of these data with the two other randomised placebo-controlled trials of thiopurines in the postoperative setting16, 19 shows a significant reduction in postoperative clinical relapse at 12 months (relative risk 0·57, 95% CI 0·38–0·85; appendix p 11). Taken with the other recent data, our study helps to make progress towards a treatment algorithm for all patients after surgery for Crohn's disease, with smoking habit the key determinant affecting management.

eduction in postoperative clinical relapse at 12 months (relative risk 0·57, 95% CI 0·38–0·85; appendix p 11). Taken with the other recent data, our study helps to make progress towards a treatment algorithm for all patients after surgery for Crohn's disease, with smoking habit the key determinant affecting management. Several areas now require further clinical studies, including putative mechanisms for the effect of smoking on Crohn's disease,26 and smoking intervention studies. The efficacy and safety of mercaptopurine compared with anti-tumour-necrosis-factor (TNF) as postoperative preventive treatment is a key issue to investigate. At present, anti-TNF treatment is reserved for patients who are intolerant or unresponsive to thiopurine, but the safety, efficacy, and cost of these drugs is under continuous reassessment. Endoscopic findings and faecal calprotectin remain important components of disease assessment, but the exact parameters that best define postoperative recurrent disease remain to be elucidated. Supplementary Material Supplementary appendix Acknowledgments We thank the patients who agreed to participate in this trial, the study contributors, and the investigators who recruited patients (appendix pp 2–3). The trial was supported by the Medical Research Council and National Institute of Health Research's Efficacy and Mechanism Evaluation Programme, Scottish Government Chief Scientist Office, and the National Institute of Health Research National Portfolio. This paper is dedicated to the memory of Keith Leiper, Liverpool gastroenterologist, who died on Oct 21, 2011.

ch Council and National Institute of Health Research's Efficacy and Mechanism Evaluation Programme, Scottish Government Chief Scientist Office, and the National Institute of Health Research National Portfolio. This paper is dedicated to the memory of Keith Leiper, Liverpool gastroenterologist, who died on Oct 21, 2011. Contributors CMo, IA, AC, JM, MGD, MC, AM, RJP, and JS conceived, designed, and formulated the statistical analysis of the study. CMo, IA, HE, CK, SLe, NAK, and JS prepared the first draft of the paper. All other authors contributed to data acquisition, reviewed the paper, and approved the submitted version. Declaration of interests We declare no competing interests.

Introduction Globally, more than 70 million people are estimated to have hepatitis C virus (HCV) infection.1 Prevalence of HCV is particularly high in the eastern Mediterranean region and Europe, where approximately 2·3% and 1·5% of the general population have HCV infection, respectively.1 In the USA, the estimated prevalence of past or current HCV infection is 1·4%, affecting 4·6 million people, of whom at least 3·5 million have active HCV infection (1% of the general population).2 The number of new incident cases of HCV infections in the USA has been increasing since 2010.3 After acute HCV infection, most patients develop chronic infection.4 Usually, these patients remain asymptomatic, with less than a third progressing to liver cirrhosis in the subsequent 20–30 years.4 Although mortality due to cirrhosis and hepatocellular carcinoma are well recognised long-term complications of chronic HCV infection,5, 6 patients with chronic infection are also at increased risk of non-liver-related mortality, including cancer and circulatory death.7

o liver cirrhosis in the subsequent 20–30 years.4 Although mortality due to cirrhosis and hepatocellular carcinoma are well recognised long-term complications of chronic HCV infection,5, 6 patients with chronic infection are also at increased risk of non-liver-related mortality, including cancer and circulatory death.7 Atherosclerotic cardiovascular disease is the most common cause of death worldwide and the burden of disease is projected to rise substantially over the next few decades, particularly in low-income and middle-income countries (LMICs).8 HCV transmission is also projected to rise considerably in LMICs due to unsafe health-care practices and injection drug use.9, 10, 11 Published data12, 13, 14 suggest that the long period of chronic HCV infection might lead to the development of atherosclerotic cardiovascular disease because of derangements in metabolic pathways and chronic inflammation. However, the direction and strength of the association between HCV infection and cardiovascular disease remains uncertain.15, 16, 17, 18, 19 Research in context Evidence before this study

Atherosclerotic cardiovascular disease is the most common cause of death worldwide and the burden of disease is projected to rise substantially over the next few decades, particularly in low-income and middle-income countries (LMICs).8 HCV transmission is also projected to rise considerably in LMICs due to unsafe health-care practices and injection drug use.9, 10, 11 Published data12, 13, 14 suggest that the long period of chronic HCV infection might lead to the development of atherosclerotic cardiovascular disease because of derangements in metabolic pathways and chronic inflammation. However, the direction and strength of the association between HCV infection and cardiovascular disease remains uncertain.15, 16, 17, 18, 19 Research in context Evidence before this study We searched PubMed from database inception to Jan 1, 2018, for systematic reviews and meta-analyses evaluating the association between hepatitis C virus (HCV) infection and atherosclerotic cardiovascular disease using the search terms “myocardial infarction”, “stroke”, “cerebrovascular disease”, “cardiovascular disease”, and “hepatitis C”. We found no studies that assessed the risk of cardiovascular disease or calculated the burden from all major atherosclerotic cardiovascular events associated with hepatitis C. Previous meta-analyses have evaluated the association between HCV infection and stroke and surrogate markers of subclinical atherosclerotic disease. Added value of this study

We searched PubMed from database inception to Jan 1, 2018, for systematic reviews and meta-analyses evaluating the association between hepatitis C virus (HCV) infection and atherosclerotic cardiovascular disease using the search terms “myocardial infarction”, “stroke”, “cerebrovascular disease”, “cardiovascular disease”, and “hepatitis C”. We found no studies that assessed the risk of cardiovascular disease or calculated the burden from all major atherosclerotic cardiovascular events associated with hepatitis C. Previous meta-analyses have evaluated the association between HCV infection and stroke and surrogate markers of subclinical atherosclerotic disease. Added value of this study To our knowledge, our study is the first meta-analysis to investigate the risk of major atherosclerotic cardiovascular disease in people with HCV infection and to estimate the burden of atherosclerotic cardiovascular disease attributed to HCV infection at the global, regional, and national level. Implication of all the available evidence

lopment (contract number 175087). SA acknowledges International Centre for Casemix and Clinical Coding, Faculty of Medicine, National University of Malaysia, and Department of Health Policy and Management, Faculty of Public Health, Kuwait University, for the approval and support to participate in this research project. Editorial note: The Lancet Group takes a neutral position with respect to territorial claims in published maps and institutional affiliations.

lopment (contract number 175087). SA acknowledges International Centre for Casemix and Clinical Coding, Faculty of Medicine, National University of Malaysia, and Department of Health Policy and Management, Faculty of Public Health, Kuwait University, for the approval and support to participate in this research project. Editorial note: The Lancet Group takes a neutral position with respect to territorial claims in published maps and institutional affiliations. GBD 2017 Colorectal Cancer Collaborators Saeid Safiri, Sadaf G Sepanlou, Kevin S Ikuta, Catherine Bisignano, Hamideh Salimzadeh, Alireza Delavari, Reza Ansari, Gholamreza Roshandel, Shahin Merat, Christina Fitzmaurice, Lisa M Force, Molly R Nixon, Hedayat Abbastabar, Kedir Hussein Abegaz, Mohsen Afarideh, Ayat Ahmadi, Muktar Beshir Ahmed, Tomi Akinyemiju, Fares Alahdab, Raghib Ali, Mahtab Alikhani, Vahid Alipour, Syed Mohamed Aljunid, Majid Abdulrahman Hamad Almadi, Amir Almasi-Hashiani, Rajaa M Al-Raddadi, Nelson Alvis-Guzman, Saeed Amini, Nahla Hamed Anber, Alireza Ansari-Moghaddam, Jalal Arabloo, Zohreh Arefi, Mohammad Asghari Jafarabadi, Abbas Azadmehr, Alaa Badawi, Nafiseh Baheiraei, Till Winfried Bärnighausen, Huda Basaleem, Masoud Behzadifar, Meysam Behzadifar, Yaschilal Muche Belayneh, Kidanemaryam Berhe, Krittika Bhattacharyya, Belete Biadgo, Ali Bijani, Antonio Biondi, Tone Bjørge, Antonio M Borzì, Cristina Bosetti, Ibrahim R Bou-Orm, Hermann Brenner, Andrey Nikolaevich Briko, Nikolay Ivanovich Briko, Giulia Carreras, Félix Carvalho, Carlos A Castañeda-Orjuela, Ester Cerin, Peggy Pei-Chia Chiang, Onyema Greg Chido-Amajuoyi, Ahmad Daryani, Dragos Virgil Davitoiu, Alireza Delavari, Gebre Teklemariam Demoz, Rupak Desai, Mostafa Dianati Nasab, Aziz Eftekhari, Iman El Sayed, Iffat Elbarazi, Mohammad Hassan Emamian, Aman Yesuf Endries, Firooz Esmaeilzadeh, Alireza Esteghamati, Arash Etemadi, Farshad Farzadfar, Eduarda Fernandes, João C Fernandes, Irina Filip, Florian Fischer, Masoud Foroutan, Mohamed M Gad, Silvano Gallus, Fatemeh Ghaseni-Kebria, Ahmad Ghashghaee, Giuseppe Gorini, Nima Hafezi-Nejad, Arvin Haj-Mirzaian, Arya Haj-Mirzaian, Susan Hasanpour-Heidari, Amir Hasanzadeh, Soheil Hassanipour, Simon I Hay, Chi Linh Hoang, Mihaela Hostiuc, Mowafa Househ, Olayinka Stephen Ilesanmi, Milena D Ilic, Kaire Innos, Seyed Sina Naghibi Irvani, Farhad Islami, Anelisa Jaca, Nader Jafari Balalami, Nastaran Jafari Delouei, Morteza Jafarinia, Mohammad Ali Jahani, Mihajlo Jakovljevic, Spencer L James, Mehdi Javanbakht, Ensiyeh Jenabi, Ravi Prakash Jha, Farahnaz Joukar, Amir Kasaeian, Tesfaye Dessale Kassa, Mesfin Wudu Kassaw, Andre Pascal Kengne, Yousef Saleh Khader, M

d Islami, Anelisa Jaca, Nader Jafari Balalami, Nastaran Jafari Delouei, Morteza Jafarinia, Mohammad Ali Jahani, Mihajlo Jakovljevic, Spencer L James, Mehdi Javanbakht, Ensiyeh Jenabi, Ravi Prakash Jha, Farahnaz Joukar, Amir Kasaeian, Tesfaye Dessale Kassa, Mesfin Wudu Kassaw, Andre Pascal Kengne, Yousef Saleh Khader, M ojtaba Khaksarian, Rovshan Khalilov, Ejaz Ahmad Khan, Maryam Khayamzadeh, Maryam Khazaee-Pool, Salman Khazaei, Fatemeh Khosravi Shadmani, Jagdish Khubchandani, Daniel Kim, Adnan Kisa, Sezer Kisa, Jonathan M Kocarnik, Hamidreza Komaki, Jacek A Kopec, Ai Koyanagi, Ernst J Kuipers, Vivek Kumar, Carlo La Vecchia, Faris Hasan Lami, Alan D Lopez, Platon D Lopukhov, Raimundas Lunevicius, Azeem Majeed, Maryam Majidinia, Amir Manafi, Navid Manafi, Ana-Laura Manda, Fariborz Mansour-Ghanaei, Lorenzo Giovanni Mantovani, Dhruv Mehta, Toni Meier, Hagazi Gebre Meles, Walter Mendoza, Tomislav Mestrovic, Bartosz Miazgowski, Tomasz Miazgowski, Seyed Mostafa Mir, Hamed Mirzaei, Karzan Abdulmuhsin Mohammad, Naser Mohammad Gholi Mezerji, Abdollah Mohammadian-Hafshejani, Milad Mohammadoo-Khorasani, Shafiu Mohammed, Farnam Mohebi, Ali H Mokdad, Lorenzo Monasta, Maryam Moossavi, Ghobad Moradi, Farhad Moradpour, Rahmatollah Moradzadeh, Azin Nahvijou, Gurudatta Naik, Farid Najafi, Javad Nazari, Ionut Negoi, Serban Negru, Cuong Tat Nguyen, Trang Huyen Nguyen, Dina Nur Anggraini Ningrum, Felix Akpojene Ogbo, Andrew T Olagunju, Tinuke O Olagunju, Adrian Pana, David M Pereira, Majid Pirestani, Akram Pourshams, Hossein Poustchi, Mostafa Qorbani, Mohammad Rabiee, Navid Rabiee, Amir Radfar, Marveh Rahmati, Fatemeh Rajati, Samira Raoofi, David Laith Rawaf, Salman Rawaf, Robert C Reiner Jr, Andre M N Renzaho, Nima Rezaei, Aziz Rezapour, Anas M Saad, Seyedmohammad Saadatagah, Basema Saddik, Farkhonde Salehi, Saleh Salehi Zahabi, Inbal Salz, Abdallah M Samy, Juan Sanabria, Milena M Santric Milicevic, Arash Sarveazad, Maheswar Satpathy, Ione J C Schneider, Mario Sekerija, Faramarz Shaahmadi, Hosein Shabaninejad, Morteza Shamsizadeh, Zeinab Sharafi, Mehdi Sharif, Amrollah Sharifi, Sara Sheikhbahaei, Reza Shirkoohi, Sudeep K Siddappa Malleshappa, Diego Augusto Santos Silva, Mekonnen Sisay, Catalin-Gabriel Smarandache, Moslem Soofi, Kjetil Soreide, Sergey Soshnikov, Vladimir I Starodubov, Rafael Tabarés-Seisdedos, Mark Sullman, Amir Taherkhani, Berhe Etsay Tesfay, Roman Topor-Madry, Eugenio Traini, Bach Xuan Tran, Khanh Bao Tran, Irfan Ullah, Olalekan A Uthman, Marco Vacante, Amir Vahedian-Azimi, Aless

To our knowledge, our study is the first meta-analysis to investigate the risk of major atherosclerotic cardiovascular disease in people with HCV infection and to estimate the burden of atherosclerotic cardiovascular disease attributed to HCV infection at the global, regional, and national level. Implication of all the available evidence Our findings show that people with HCV infection have a higher risk of cardiovascular disease than those without. The global burden of cardiovascular disease attributable to HCV accounted for a substantial number of disability-adjusted life-years in 2015, and the majority of the burden was borne by low-income and middle-income countries. These finding highlights the importance of public health strategies to eradicate HCV infection to reduce the burden of not only hepatic, but extrahepatic complications (such as cardiovascular disease), especially in regions with high HCV prevalence. Our study aimed to determine the association between HCV infection and the risk of cardiovascular disease to establish the global burden of cardiovascular disease attributable to HCV.

Our findings show that people with HCV infection have a higher risk of cardiovascular disease than those without. The global burden of cardiovascular disease attributable to HCV accounted for a substantial number of disability-adjusted life-years in 2015, and the majority of the burden was borne by low-income and middle-income countries. These finding highlights the importance of public health strategies to eradicate HCV infection to reduce the burden of not only hepatic, but extrahepatic complications (such as cardiovascular disease), especially in regions with high HCV prevalence. Our study aimed to determine the association between HCV infection and the risk of cardiovascular disease to establish the global burden of cardiovascular disease attributable to HCV. Methods Search strategy and selection criteria We searched MEDLINE, EMBASE, Ovid Global Health, and Web of Science from database inception to May 9, 2018, for original peer-reviewed articles using the search terms “myocardial infarction”, “stroke”, “cerebrovascular disease”, “cardiovascular disease”, and “hepatitis C” with no language restrictions. Full search terms are in the appendix (pp 2, 3). Additionally, we manually searched relevant review articles and bibliographic reference lists of studies selected for inclusion in our meta-analysis.

arction”, “stroke”, “cerebrovascular disease”, “cardiovascular disease”, and “hepatitis C” with no language restrictions. Full search terms are in the appendix (pp 2, 3). Additionally, we manually searched relevant review articles and bibliographic reference lists of studies selected for inclusion in our meta-analysis. We included all longitudinal studies (case-control studies, cohort studies, and randomised controlled trials) that reported risk ratios (RRs) for hospital admission due to atherosclerotic cardiovascular disease or cardiovascular mortality in people with HCV compared with people without HCV. When there were multiple publications using data from the same cohort, we selected the article that reported the longest follow-up period. Detailed full-text review and data extraction was done independently by at least two investigators (KKL, DS, RB, or MA) and any disagreements were resolved by a third investigator (ASVS). We contacted authors for additional data or clarification if required. This study was done in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (appendix pp 4, 5).20 The study protocol is available online.

sagreements were resolved by a third investigator (ASVS). We contacted authors for additional data or clarification if required. This study was done in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (appendix pp 4, 5).20 The study protocol is available online. For studies that stratified the study population according to the presence of HCV RNA viraemia, we used the RR estimates pertaining to individuals who were HCV RNA positive. We defined a cardiovascular event as hospital admission with, or mortality from, acute myocardial infarction or stroke. Studies that evaluated a composite of acute cardiovascular events that included myocardial infarction or stroke but were not exclusive to these conditions were also included. For studies that stratified stroke events into haemorrhagic and ischaemic strokes, we included only ischaemic strokes in the analysis because haemorrhagic strokes have distinct pathophysiological mechanisms that are unrelated to atherosclerosis.21

or stroke but were not exclusive to these conditions were also included. For studies that stratified stroke events into haemorrhagic and ischaemic strokes, we included only ischaemic strokes in the analysis because haemorrhagic strokes have distinct pathophysiological mechanisms that are unrelated to atherosclerosis.21 Data analysis We extracted RR estimates comparing cardiovascular events in people with HCV versus those without HCV from published reports using a standardised data extraction sheet. We estimated pooled RRs with 95% CIs. Since this outcome was relatively uncommon, we pooled studies that reported odds ratio and RR. We also assumed independence between risk estimates for different endpoints reported within studies, consistent with our previous analysis.22 We did a subgroup analysis stratified by outcome, HIV co-infection, publication year, risk of bias, definition of outcome event, and geographical location.

y, Catalin-Gabriel Smarandache, Moslem Soofi, Kjetil Soreide, Sergey Soshnikov, Vladimir I Starodubov, Rafael Tabarés-Seisdedos, Mark Sullman, Amir Taherkhani, Berhe Etsay Tesfay, Roman Topor-Madry, Eugenio Traini, Bach Xuan Tran, Khanh Bao Tran, Irfan Ullah, Olalekan A Uthman, Marco Vacante, Amir Vahedian-Azimi, Aless andro Valli, Elena Varavikova, Isidora S Vujcic, Ronny Westerman, Vahid Yazdi-Feyzabadi, Engida Yisma, Chuanhua Yu, Vesna Zadnik, Telma Zahirian Moghadam, Leila Zaki, Hamed Zandian, Zhi-Jiang Zhang, Christopher J L Murray, Mohsen Naghavi*, and Reza Malekzadeh*. *These authors jointly supervised the study.

y, Catalin-Gabriel Smarandache, Moslem Soofi, Kjetil Soreide, Sergey Soshnikov, Vladimir I Starodubov, Rafael Tabarés-Seisdedos, Mark Sullman, Amir Taherkhani, Berhe Etsay Tesfay, Roman Topor-Madry, Eugenio Traini, Bach Xuan Tran, Khanh Bao Tran, Irfan Ullah, Olalekan A Uthman, Marco Vacante, Amir Vahedian-Azimi, Aless andro Valli, Elena Varavikova, Isidora S Vujcic, Ronny Westerman, Vahid Yazdi-Feyzabadi, Engida Yisma, Chuanhua Yu, Vesna Zadnik, Telma Zahirian Moghadam, Leila Zaki, Hamed Zandian, Zhi-Jiang Zhang, Christopher J L Murray, Mohsen Naghavi*, and Reza Malekzadeh*. *These authors jointly supervised the study. Affiliations Aging Research Institute (S Safiri PhD), Department of Community Medicine (S Safiri PhD), Department of Pharmacology and Toxicology (A Eftekhari PhD), Department of Biostatistics and Epidemiology (Prof M Asghari Jafarabadi PhD), Tabriz University of Medical Sciences, Tabriz, Iran; Cancer Biology Research Center (M Rahmati PhD, R Shirkoohi PhD), Cancer Research Center (A Nahvijou PhD), Cancer Research Institute (R Shirkoohi PhD), Department of Cardiology (S Saadatagah MD), Department of Health Promotion and Education (Z Arefi PhD), Department of Microbiology (A Hasanzadeh PhD), Department of Pharmacology (Arv Haj-Mirzaian MD, Ary Haj-Mirzaian MD), Digestive Diseases Research Institute (S G Sepanlou MD, H Salimzadeh PhD, Prof A Delavari MD, R Ansari MD, G Roshandel PhD, Prof S Merat MD, Prof A Pourshams MD, H Poustchi PhD, Prof R Malekzadeh MD), Endocrinology and Metabolism Research Center (M Afarideh MD, Prof A Esteghamati MD, S Sheikhbahaei MD), Hematologic Malignancies Research Center (A Kasaeian PhD), Hematology-Oncology and Stem Cell Transplantation Research Center (A Kasaeian PhD), Iran National Institute of Health Research (F Mohebi MD), Iranian Center of Neurological Research (H Abbastabar PhD), Knowledge Utilization Research Center (A Ahmadi PhD), Non-communicable Diseases Research Center (F Farzadfar MD, F Mohebi MD), Research Center for Immunodeficiencies (Prof N Rezaei PhD), Department of Internal Medicine (Prof A Delavari MD), School of Medicine (N Hafezi-Nejad MD), Tehran University of Medical Sciences, Tehran, Iran (A Etemadi PhD); Department of Epidemiology (M Dianati Nasab MSc), Non-communicable Diseases Research Center (S G Sepanlou MD, Prof R Malekzadeh MD), Shiraz University of Medical Sciences, Shiraz, Iran; Department of Health Metrics Sciences, School of Medicine (Prof S I Hay FMedSci, Prof A H Mokdad PhD, R C Reiner Jr PhD, Prof C J L Murray DPhil, Prof M Naghavi PhD), Division of Allergy and Infectious Diseases (K S Ikuta MD), Division of Hematology (C Fitzmaurice MD), Institute for Health Metrics and Evaluation (K S Ikuta MD, C Bisignano MPH, C Fitzmaurice MD, L M Force MD, M R Nixon PhD, Prof S I Hay FMedSci, S L James MD, J

ted odds ratio and RR. We also assumed independence between risk estimates for different endpoints reported within studies, consistent with our previous analysis.22 We did a subgroup analysis stratified by outcome, HIV co-infection, publication year, risk of bias, definition of outcome event, and geographical location. Two independent investigators (KKL and DS) assessed individual studies for risk of bias, using the degree of adjustment for confounders as the primary domain, and any disagreements were adjudicated by a third investigator. Studies that had adjusted for age, sex, and at least one other confounder were classified as being at low risk of bias. Studies that adjusted for fewer confounders than this were classified as moderate or high risk: studies that adjusted for either age or sex without any other confounders were classified as moderate risk of bias and those that did not adjust for both age or sex were classified as high risk of bias. We did the subgroup analysis stratified by risk of bias.

founders than this were classified as moderate or high risk: studies that adjusted for either age or sex without any other confounders were classified as moderate risk of bias and those that did not adjust for both age or sex were classified as high risk of bias. We did the subgroup analysis stratified by risk of bias. We estimated the burden of cardiovascular disease attributable to HCV at the national, regional, and global level. We obtained 2015 global prevalence estimates of viraemic HCV (HCV RNA positive) for 100 countries from the Polaris Observatory, with estimates stratified by Global Burden of Disease region.23 These 100 countries represent more than 85% of the global population and where more than 89% of all HCV viral infections (HCV RNA positive) are estimated to occur worldwide.23 The national prevalence estimates obtained were age-specific and sex-specific. We obtained age-specific and sex-specific disability-adjusted life-year (DALY) estimates for cardiovascular disease (DALYs due to ischaemic heart disease and stroke) for all adults aged older than 20 years in 2015 from the Institute of Health Metrics and Evaluation.24 The extraction databases from the systematic review and the data from the Polaris Observatory and Institute of Health Metrics and Evaluation used to derive the pooled estimates and the burden estimates alongside the R code script are available online.

years in 2015 from the Institute of Health Metrics and Evaluation.24 The extraction databases from the systematic review and the data from the Polaris Observatory and Institute of Health Metrics and Evaluation used to derive the pooled estimates and the burden estimates alongside the R code script are available online. We estimated the population attributable risk fraction at the national, regional, and global level using the pooled RR for cardiovascular disease in patients with hepatitis C and the prevalence estimates of HCV. The population attributable fraction (PAF) for cardiovascular disease attributable to HCV was calculated as described previously:25, 26 Population attributable fraction=Prevalence×(RR-1)1+(Prevalence×RR) We then used national, regional, and global level attributable fractions to calculate the burden as previously described (appendix pp 6–10): DALYs attributable to HCV=Cardiovascular DALYs×PAF We provided estimates of PAF and burden in 5-year age groups and presented data graphically using a linear model to interpolate the intervening years. We further provided burden estimates by income of nation stratified by high-income versus LMICs. National income status was defined according to the 2018 World Bank classification.27

stimates of PAF and burden in 5-year age groups and presented data graphically using a linear model to interpolate the intervening years. We further provided burden estimates by income of nation stratified by high-income versus LMICs. National income status was defined according to the 2018 World Bank classification.27 We anticipated heterogeneity in the RRs across studies because of differences in study design, patient population, geographical location, statistical methods, and adjustment for confounders. We pooled RRs using a random effects model to account for within and between study heterogeneity. We assessed heterogeneity in the pooled meta-estimate of the RR using the I2 statistic. We assessed publication bias using visual inspection of funnel plots of the RR estimates and using Egger's regression test for asymmetry.28 We corrected for asymmetry using Duval and Tweedie's trim and fill method.29 Full statistical methods are in the appendix (pp 6–10). All analyses were done using R (version 3.4.1). A two-sided p value of less than 0·05 was considered to indicate statistical significance. 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. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

We anticipated heterogeneity in the RRs across studies because of differences in study design, patient population, geographical location, statistical methods, and adjustment for confounders. We pooled RRs using a random effects model to account for within and between study heterogeneity. We assessed heterogeneity in the pooled meta-estimate of the RR using the I2 statistic. We assessed publication bias using visual inspection of funnel plots of the RR estimates and using Egger's regression test for asymmetry.28 We corrected for asymmetry using Duval and Tweedie's trim and fill method.29 Full statistical methods are in the appendix (pp 6–10). All analyses were done using R (version 3.4.1). A two-sided p value of less than 0·05 was considered to indicate statistical significance. 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. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results We identified 16 650 articles, of which 2270 were duplicates (figure 1). 343 full-text articles were assessed for eligibility. After full-text review, 36 studies,7, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64 which provided 47 estimates, were included in our analyses (table). These studies included 341 739 people with HCV. 31 (86%) of 36 studies were done in North America, Europe, and east Asia. Only two studies30, 31 originated from LMICs. Most studies used the International Classification of Diseases coding or physician diagnosis to define the outcome events.Figure 1 Study selection

ble). These studies included 341 739 people with HCV. 31 (86%) of 36 studies were done in North America, Europe, and east Asia. Only two studies30, 31 originated from LMICs. Most studies used the International Classification of Diseases coding or physician diagnosis to define the outcome events.Figure 1 Study selection Table Baseline characteristics of studies included in the meta-analysis

ble). These studies included 341 739 people with HCV. 31 (86%) of 36 studies were done in North America, Europe, and east Asia. Only two studies30, 31 originated from LMICs. Most studies used the International Classification of Diseases coding or physician diagnosis to define the outcome events.Figure 1 Study selection Table Baseline characteristics of studies included in the meta-analysis Cohort name Country or region Study type Data source Participants, n Events, n Men, n (%) Mean age at baseline, years Study period Hepatitis C virus status Outcome Outcome definition Heo et al, 201833 .. USA Cohort study Organ Procurement and Transplant Network 2728 117 1996 (73%) 50·9 2004–14 Seropositive Cardiovascular disease Undefined Alvaro-Meca et al, 201734 .. Spain Case-control study Spanish Minimum Basic Data Set 4091 369 3248 (79%) 45 1997–2013 Seropositive Stroke* ICD-9 Butt et al, 201735 ERCHIVES USA Cohort study Veterans Health Administration 171 726 5949 171 726 (100%) 54 2001–15 Viraemic Myocardial infarction ICD-9 Chew et al, 201736 ERCHIVES USA Cohort study Veterans Health Administration 168 256 11 753 168 256 (100%) 55 2001–14 Seropositive Cardiovascular disease ICD-9 Goodkin et al, 201730 DOPPS Multiple† Cohort study Hospital records 76 689 6790 45 016 (59%) 62·5 1996–2015 Seropositive Cardiovascular disease, myocardial infarction, stroke Undefined Kovari et al, 201737 .. Switzerland Cohort study Swiss HIV Cohort Study 5006 143 3624 (72%) 50 1994–2014 Seropositive Cardiovascular disease* Physician diagnosis Piazza et al, 201638 .. USA Cohort study Hospital records 143 19 101 (71%) 55 2005–10 Undefined Cardiovascular disease Undefined Fernandez-Montero et al, 201539 .. Spain Cohort study Hospital register 1066 29 842 (79%) 42·7 2004–15 Viraemic Cardiovascular disease† Physician diagnosis Tsai et al, 201540 NHIRD–HCV Taiwan Cohort study National Health Insurance Research Database 69 915 848 35 936 (51%) 54·7 1998–2008 Undefined Myocardial infarction Undefined Vajdic et al, 201541 .. Australia Cohort study Pharmaceutical Drugs of Addiction System 29 571 122 20 403 (69%) 26 1993–2007 Seropositive Cardiovascular disease ICD-9, ICD-10 Enger et al, 201442 ORD USA Cohort study Optum Research Database (insurance plans) 90 931 534 56 740 (62%) 49 2000–06 Seropositive Myocardial infarction, stroke ICD-9 Gillis et al, 201443 OCS Canada Cohort study Clinic register 4152 167 3483 (84%) 36 1995–2011 Seropositive Cardiovascular disease† Physician diagnosis Hsu et al, 201444 ..

2 ORD USA Cohort study Optum Research Database (insurance plans) 90 931 534 56 740 (62%) 49 2000–06 Seropositive Myocardial infarction, stroke ICD-9 Gillis et al, 201443 OCS Canada Cohort study Clinic register 4152 167 3483 (84%) 36 1995–2011 Seropositive Cardiovascular disease† Physician diagnosis Hsu et al, 201444 .. Taiwan Cohort study National Health Insurance Research Database 7055 429 4599 (65%) 54·9 2003–11 Seropositive Myocardial infarction, stroke ICD-9 Pothineni et al, 201445 UAMS USA Cohort study Enterprise Data Warehouse at University of Arkansas for Medical Sciences 23 050 951 12 631 (55%) 50·9 2001–13 Viraemic Cardiovascular disease ICD-9 Tripathi et al, 201446 .. USA Cohort study Medicaid 13 632 1284 7661 (56%) 38 1994–2011 Seropositive Cardiovascular disease ICD-9 Womack et al, 201447 Veterans Aging Cohort Study–virtual cohort USA Cohort study Veterans Health Administration, Medicare, Medicaid, and Quality Enhancement Research Initiative in ischaemic heart disease 2187 86 0 43·6 2003–09 Seropositive Cardiovascular disease* ICD-9 Adinolfi et al, 201348 .. Italy Case-control study Hospital records 820 123 524 (64%) 76 2010–12 Seropositive Stroke Physician diagnosis Hsu et al, 201349 LHID2000 Taiwan Cohort study Longitudinal Health Insurance Database 2000 15 565 NR 8078 (52%) Not reported 2004–07 Viraemic Stroke ICD-9 Younossi et al, 201350 NHANES III USA Cohort study National Health and Nutrition Examination Survey 8985 NR 4178 (46%) Not reported 1988–2006 Viraemic Cardiovascular disease ICD-10 Campbell et al, 201251 ..

study Longitudinal Health Insurance Database 2000 15 565 NR 8078 (52%) Not reported 2004–07 Viraemic Stroke ICD-9 Younossi et al, 201350 NHANES III USA Cohort study National Health and Nutrition Examination Survey 8985 NR 4178 (46%) Not reported 1988–2006 Viraemic Cardiovascular disease ICD-10 Campbell et al, 201251 .. UK Cohort study Hospital records 4068 32 4068 (100%) 36·5 2004–09 Seropositive Cardiovascular disease* Physician diagnosis Carrieri et al, 201252 APROCO-COPILOTE France Cohort study Medical questionnaires 1154 49 900 (78%) 37·7 1997–2010 Seropositive Cardiovascular disease* ICD-10 Forde et al, 201253 THIN UK Cohort study General practice medical records 76 477 264 46 727 (61%) 38·6 1996–2008 Undefined Myocardial infarction Read diagnostic code Lee et al, 201254 REVEAL–HCV Taiwan Cohort study Questionnaires and interviews 19 636 477 9523 (48%) 47·6 1991–2008 Viraemic Cardiovascular disease ICD-9 Liao et al, 201255 NHIRD Taiwan Cohort study National Health Insurance Research Database 20470 1981 10235 (50%) 52 2002–08 Viraemic Stroke ICD-9 Freiberg et al, 201156 Veterans Aging Cohort Study–virtual cohort USA Cohort study Veterans Aging Cohort Study and Large Health Study of Veteran Enrollees 8579 194 8579 (100%) 48·1 2000–07 Seropositive Cardiovascular disease* ICD-9 Kristiansen et al, 201157 .. Norway Cohort study Department of Microbiology, University Hospital of North Norway 1010 5 686 (68%) 40 1990–2000 Seropositive Cardiovascular disease ICD-10 Ohsawa et al, 201158 KAREN Japan Cohort study KAREN cohort 1077 194 682 (63%) 60·4 2003–08 Seropositive Cardiovascular disease ICD-10 Bedimo et al, 201059 HIV Clinical Care Registry USA Cohort study Veterans Registry 19 424 1146 18 938 (97%) 46·2 1984–2004 Viraemic Myocardial infarction, stroke* ICD-9 Belloso et al, 201031 LATINA Brazil, Mexico Cohort study LATINA cohort 160 40 Not reported Not reported 1997–2007 Seropositive Cardiovascular disease* Physician diagnosis DAD Study Group, 201060 DAD study Europe, USA, Australia Cohort study DAD cohort 21 815 517 16 143 (74%) 38 1999–2007 Seropositive Myocardial infarction* WHO MONICA Project Lee et al, 201061 .. Taiwan Cohort study National Death Certification Registry 23 665 22 11 879 (50%) 47·1 1991–92 Viraemic Stroke ICD-9 Tsui et al, 200932 The Heart and Soul study USA Cohort study Veterans Administration electronic records 981 151 803 (82%) 66·3 2000–06 Seropositive Cardiovascular disease Physician diagnosis Guiltinan et al, 200862 ..

study National Death Certification Registry 23 665 22 11 879 (50%) 47·1 1991–92 Viraemic Stroke ICD-9 Tsui et al, 200932 The Heart and Soul study USA Cohort study Veterans Administration electronic records 981 151 803 (82%) 66·3 2000–06 Seropositive Cardiovascular disease Physician diagnosis Guiltinan et al, 200862 .. USA Cohort study Blood Systems 20 518 88 13 254 (65%) Not reported 1991–2002 Seropositive Cardiovascular disease ICD-9 CM, ICD-10 CM Kalantar-Zadeh et al, 200763 .. USA Cohort study DaVita outpatient dialysis database 13 664 NR 7433 (54%) 60·1 2001–04 Seropositive Cardiovascular disease Undefined Arcari et al, 200664 .. USA Case-control study Clinical registry 75 834 292 47 775 (63%) 40·2 1991–2000 Seropositive Myocardial infarction ICD-9 Amin et al, 20067 .. Australia Cohort study New South Wales Health Department Notifiable Diseases Database 582 450 582 (100%) 34 1990–2002 Seropositive Cardiovascular disease ICD-9 and ICD-10 ICD=International Classification of Diseases. ERCHIVES=Electronically Retrieved Cohort of Hepatitis C Virus Infected Veterans. DOPPS=Dialysis Outcomes and Practice Patterns Study. NHIRD–HCV=National Health Insurance Research Database–Hepatitis C Virus. ORD=Optum Research Database. OCS=Ontario HIV Treatment Network Cohort Study. UAMS=University of Arkansas for Medical Sciences. LHID2000=Longitudinal Health Insurance Database 2000. NR=not reported. NHANES III=National Health and Nutrition Examination Survey III. APROCO-COPILOTE=Antiprotéases Cohorte. THIN=The Health Improvement Network. REVEAL–HCV=Risk Evaluation of Viral Load Elevation and Associated Liver Disease/Cancer–Hepatitis C Virus. DAD=Data Collection on Adverse Events of Anti-HIV Drugs. MONICA=Multinational Monitoring of Trends and Determinants in Cardiovascular Disease. CM=Clinical Modification.

Cohorte. THIN=The Health Improvement Network. REVEAL–HCV=Risk Evaluation of Viral Load Elevation and Associated Liver Disease/Cancer–Hepatitis C Virus. DAD=Data Collection on Adverse Events of Anti-HIV Drugs. MONICA=Multinational Monitoring of Trends and Determinants in Cardiovascular Disease. CM=Clinical Modification. * Risk ratio reported for hepatitis C virus and HIV co-infection versus HIV infection only. † Australia, Belgium, Canada, mainland China, France, Germany, Bahrain, Kuwait, Oman, Qatar, Saudi Arabia, United Arab Emirates, Italy, Japan, New Zealand, Spain, Russia, Sweden, Turkey, the UK, and the USA.

Cohorte. THIN=The Health Improvement Network. REVEAL–HCV=Risk Evaluation of Viral Load Elevation and Associated Liver Disease/Cancer–Hepatitis C Virus. DAD=Data Collection on Adverse Events of Anti-HIV Drugs. MONICA=Multinational Monitoring of Trends and Determinants in Cardiovascular Disease. CM=Clinical Modification. * Risk ratio reported for hepatitis C virus and HIV co-infection versus HIV infection only. † Australia, Belgium, Canada, mainland China, France, Germany, Bahrain, Kuwait, Oman, Qatar, Saudi Arabia, United Arab Emirates, Italy, Japan, New Zealand, Spain, Russia, Sweden, Turkey, the UK, and the USA. 24 studies defined HCV infection as anti-HCV antibody seropositivity, nine studies used detectable HCV RNA levels, and three studies did not explicitly define their approach (table). Overall, the meta-analysis showed that individuals with HCV had a higher risk of cardiovascular disease than individuals without HCV (pooled RR 1·28, 95% CI 1·18–1·39; figure 2). When stratified by outcome the risk ratio was 1·13 (95% CI 1·00–1·28) for myocardial infarction, 1·38 (1·19–1·60) for stroke, and 1·39 (1·24–1·55) for cardiovascular mortality (appendix p 11). Individuals with HCV and HIV co-infection had a higher risk of cardiovascular disease than those with HIV mono-infection (RR 1·20, 1·09–1·32). Post-hoc analyses showed that the RR estimates from studies published before 2014, which was the median publication year, were marginally higher than those from studies published after 2014 (1·39 [1·25–1·54] vs 1·22 [1·11–1·34]). Studies that ascertained outcome with physician diagnosis had higher RRs than did those that used International Classification of Diseases codes (1·68 [1·24–2·29] vs 1·31 [1·20–1·42]). Nine studies in patients with HCV viraemia had marginally higher RRs than the overall pooled RR (1·32, 1·15–1·51). Nearly two-thirds of studies originated from the USA or Taiwan (21 [58%] of 36 studies). In the subgroup analysis, studies from these two countries had a similar pooled RR as those from the rest of the world (1·28 [1·16–1·40] vs 1·29 [1·12–1·48] respectively).Figure 2 Forest plots of pooled RRs for cardiovascular disease in people with hepatitis C virus versus people without

aiwan (21 [58%] of 36 studies). In the subgroup analysis, studies from these two countries had a similar pooled RR as those from the rest of the world (1·28 [1·16–1·40] vs 1·29 [1·12–1·48] respectively).Figure 2 Forest plots of pooled RRs for cardiovascular disease in people with hepatitis C virus versus people without Pooled RRs for composite cardiovascular events (A), myocardial infarction (B), and stroke (C). n=number of events. N=number of participants. RR=risk ratio. NR=not reported. df=degrees of freedom. DAD=Data Collection on Adverse Events of Anti-HIV Drugs. There was significant heterogeneity (I2=77·5%) and publication bias in the overall estimate (Egger's test p=0·003). Using the trim and fill method to correct for funnel plot asymmetry did not change the direction of effect but did attenuate the effect size (appendix p 20). 11 studies were at moderate or high risk of bias (appendix pp 12, 13). Compared with studies with a low risk of bias, those with moderate or high risk of bias had a similar pooled RR (1·30 [95% CI 1·10–1·55] for studies with moderate or high risk of bias vs 1·29 [1·19–1·40] for studies with low risk of bias). We estimated that in 2015, 1·5 million (95% CI 0·9–2·1) DALYs from cardiovascular disease were attributable to HCV, with marked geographical variation in the estimated burden. LMICs had the highest disease burden, with South Asia, eastern Europe, north Africa, and the Middle East accounting for nearly two-thirds of the global burden of cardiovascular disease attributable to HCV in 2015 (920·7 thousand DALYs; appendix p 14).

to HCV, with marked geographical variation in the estimated burden. LMICs had the highest disease burden, with South Asia, eastern Europe, north Africa, and the Middle East accounting for nearly two-thirds of the global burden of cardiovascular disease attributable to HCV in 2015 (920·7 thousand DALYs; appendix p 14). Of the 100 countries with available age-specific and sex-specific viraemic HCV prevalence estimates for 2015, the highest burden (ie, cardiovascular DALYs attributable to HCV) was in Ukraine, Mongolia, Gabon, and Egypt (figure 3; appendix pp 15–17). Worldwide, the PAF of cardiovascular disease attributable to HCV was highest in people aged 55–59 years (appendix p 18). DALYs from cardiovascular disease attributable to HCV was highest in people aged 70–74 years (appendix p 18). The burden of cardiovascular disease attributable to HCV was higher in LMICs than in high-income countries (1·4 million [95% CI 0·88–1·95] DALYs vs 0·1 million [95% CI 0·07–0·12] DALYs; figure 4; appendix p 19).Figure 3 DALYs per 100 000 people for cardiovascular disease attributable to HCV Grey colour denotes regions for which no HCV prevalence data were available to estimate burden. HCV=hepatitis C virus. DALYs=disability-adjusted life-years. Figure 4 DALYs for cardiovascular disease attributable to hepatitis C virus Solid lines show the central estimate and dashed lines show the 95% CI. DALYs=disability-adjusted life-years.

Grey colour denotes regions for which no HCV prevalence data were available to estimate burden. HCV=hepatitis C virus. DALYs=disability-adjusted life-years. Figure 4 DALYs for cardiovascular disease attributable to hepatitis C virus Solid lines show the central estimate and dashed lines show the 95% CI. DALYs=disability-adjusted life-years. Discussion In this systematic review and meta-analysis, we assessed the association between HCV and cardiovascular disease and estimated the global, regional, and national burden of cardiovascular disease attributable to HCV. We made several key observations. First, people with HCV have an increased risk of cardiovascular disease compared with those without HCV (RR 1·28). When stratified by type of cardiovascular event, the overall pooled estimate was higher for stroke than for myocardial infarction. Second, our pooled risk estimate was derived from 341 739 people with HCV infection included in 36 studies from 51 countries. Only two studies30, 31 reported findings from populations of LMICs, highlighting the paucity of data from these regions. Third, the most up-to-date annual global burden of cardiovascular disease attributable to HCV was 1·5 million DALYs. Most of this burden was concentrated in the 55–75 year age group, reflecting more premature development of cardiovascular disease in people with HCV. Fourth, considerable geographical variation was identified in the burden of cardiovascular disease attributable to HCV, with the highest burden observed in south Asia, eastern Europe, north Africa, and the Middle East. The majority of the burden was borne by LMICs rather than high-income countries. This observation is likely to reflect both a high prevalence of chronic hepatitis C in these regions and an increasing burden of cardiovascular disease.

hest burden observed in south Asia, eastern Europe, north Africa, and the Middle East. The majority of the burden was borne by LMICs rather than high-income countries. This observation is likely to reflect both a high prevalence of chronic hepatitis C in these regions and an increasing burden of cardiovascular disease. Our analysis has several strengths. We included longitudinal studies that evaluated the association between HCV and hospital admissions with, or mortality from, cardiovascular disease. Furthermore, the endpoint of our analysis was major adverse cardiovascular events, which enabled accurate risk estimation and assessment of cardiovascular burden. Previous systematic reviews and meta-analyses,65, 66, 67 which included cross-sectional studies and studies that used surrogate endpoints that might not be fully reflective of a causal relationship, have reported divergent findings. We also analysed burden using age-specific, sex-specific, and country-specific cardiovascular burden and HCV prevalence estimates, allowing us to provide HCV attributable burden estimates for specific age groups, which could be useful for policy makers. Moreover, our estimates for HCV prevalence obtained from the Polaris Observatory23 reflect viraemia rather than just seropositivity alone, and the countries for which we had prevalence estimates accounted for over 89% of all chronic HCV infections globally. Our estimates for the PAF and subsequent HCV associated cardiovascular burden are therefore based on a high-risk population with active HCV infection, in whom both long-term hepatic and extrahepatic complications remain common.

h we had prevalence estimates accounted for over 89% of all chronic HCV infections globally. Our estimates for the PAF and subsequent HCV associated cardiovascular burden are therefore based on a high-risk population with active HCV infection, in whom both long-term hepatic and extrahepatic complications remain common. This study has a number of limitations. Most studies included in this meta-analysis originated from high-income countries in North America and western Europe, but estimates were applied to all regions. This approach is commonly used in this type of analysis because of paucity of data from LMICs.68, 69 This highlights an ongoing need for research in these low-resource settings, in which the disease prevalence of HCV and cardiovascular disease is high, to improve the accuracy of the burden estimates in these regions. We also observed significant heterogeneity in our RR estimates. However, the direction of effect was consistent and robust across all subgroup analyses. The observed heterogeneity is likely to reflect the diverse patient population, viraemic status of the study population, differences in health-care systems, access to treatment, and geographical location of the studies pooled in this analysis. Many studies did not fully account for the competing risk of non-cardiovascular mortality, thus some methodological heterogeneity exists. Most people with HCV infection die from non-cardiovascular causes,7, 70 therefore this is an important competing risk that might distort the exposure–outcome association with cardiovascular disease. People with HIV and HCV co-infection have a higher risk for cardiovascular disease than those with HIV mono-infection. This increased risk highlights the importance of risk stratification in this patient population considering that people with HIV are twice as likely to have cardiovascular events than those without HIV.22 Although most studies evaluating the RR of cardiovascular disease adjusted for risk factors for cardiovascular disease, a substantial possibility of residual confounding remains. Furthermore, there was substantial publication bias in the literature, which might have influenced the risk estimates. However, there was little attenuation of the RRs when analysis was restricted to studies without moderate to high risk of bias or after accounting for publication bias using the trim and fill method.

. Furthermore, there was substantial publication bias in the literature, which might have influenced the risk estimates. However, there was little attenuation of the RRs when analysis was restricted to studies without moderate to high risk of bias or after accounting for publication bias using the trim and fill method. Additionally, we pooled RR estimates of myocardial infarction or stroke to estimate the PAF and combined this with the DALYs for ischaemic heart disease and cerebrovascular disease to estimate the burden of cardiovascular disease attributable to HCV. We were unable to estimate the burden of angina or peripheral artery disease attributable to HCV since these conditions are often diagnosed in the outpatient setting and are less likely to be captured by electronic health record systems. Therefore, it is possible that we have underestimated the cardiovascular burden associated with HCV. However, the 2010 Global Burden of Disease study24, 71 showed that angina and peripheral artery disease contributed a relatively small proportion of the overall cardiovascular disease burden. The studies included in this meta-analysis are likely to be exposed to a degree of outcome misclassification bias because most studies used routine diagnostic coding to define cardiovascular events rather than clinical adjudication. All of the included studies were observational studies, and thus we are unable to establish causality.

luded in this meta-analysis are likely to be exposed to a degree of outcome misclassification bias because most studies used routine diagnostic coding to define cardiovascular events rather than clinical adjudication. All of the included studies were observational studies, and thus we are unable to establish causality. The underlying pathophysiological mechanism for the association between HCV and cardiovascular disease remains unclear.15 HCV infection has been associated with conditions such as type 2 diabetes, a well known cardiovascular risk factor.72 Evidence has emerged showing direct effects of HCV on the development of atherosclerosis,73 beyond that attributable to metabolic derangements alone. Chronic HCV infection results in a chronic state of immune stimulation and inflammation evidenced by increased circulating levels of proinflammatory cytokines, such as interleukin 6, tumour necrosis factor-α, C-reactive protein, and fibrinogen, all of which are associated with the development of atherosclerotic cardiovascular disease.32, 74, 75 Interferon-based antiviral treatments for HCV reduce markers of inflammation, endothelial dysfunction, and diabetes mellitus.76, 77, 78 Sustained viral response with direct-acting antivirals have also been associated with a lower risk of cardiovascular events.79 Whether eradication of HCV infection reduces future risk of adverse cardiovascular events should be further explored in randomised controlled trials of direct-acting antivirals to investigate this finding.

ral response with direct-acting antivirals have also been associated with a lower risk of cardiovascular events.79 Whether eradication of HCV infection reduces future risk of adverse cardiovascular events should be further explored in randomised controlled trials of direct-acting antivirals to investigate this finding. The link between HCV and cardiovascular disease has important implications for the formulation of health policies and resource allocation, particularly in regions with limited health-care resources, where chronic HCV infection remains prevalent and cardiovascular disease burden is increasing. Globally, prevalence of HCV is projected to increase substantially, particularly in LMICs, as a result of transmission via unsafe health-care related injections and injection drug use.9, 10, 11 Consequently, mortality due to hepatic and extrahepatic complications of HCV is likely to increase considerably if efforts to improve early testing and treatment are not implemented. A disproportionate burden of cardiovascular disease is borne by LMICs, where currently more than 80% of the global burden is concentrated.80 Our estimates suggest that more than 90% of the global cardiovascular burden attributable to HCV occurs in LMICs.

forts to improve early testing and treatment are not implemented. A disproportionate burden of cardiovascular disease is borne by LMICs, where currently more than 80% of the global burden is concentrated.80 Our estimates suggest that more than 90% of the global cardiovascular burden attributable to HCV occurs in LMICs. The introduction of direct-acting antiviral therapies with the ability to achieve sustained virological response in more than 90% of treated individuals should be a cause for optimism. These new therapeutic options enable the prevention of both hepatic and extrahepatic complications of HCV infection with a shorter duration of treatment and fewer adverse events than previous generations of antiviral therapies. However, at present, public health programmes and access to health-care services for people with HCV lags behind other comparable infectious diseases, such as HIV or malaria.1 The provision of direct-acting antiviral therapies in patients with HCV infection remains low on the global scale, with only one in 15 patients1 currently being treated, the majority of whom reside in high-income countries.81, 82 To have the greatest impact on HCV morbidity and mortality, the delivery of curative HCV treatment needs to be coupled with efficient health systems to provide chronic care services for patients with both hepatic and extrahepatic complications of HCV. Considering our study findings, investment in greater strategic integration and linkage of viral hepatitis services with other relevant services, including cardiovascular disease prevention, might be a cost-effective method of facilitating the prevention and management of concurrent major health conditions. These innovative approaches to health-care delivery might require further research to evaluate feasibility and efficacy in a real-world clinical setting.

es, including cardiovascular disease prevention, might be a cost-effective method of facilitating the prevention and management of concurrent major health conditions. These innovative approaches to health-care delivery might require further research to evaluate feasibility and efficacy in a real-world clinical setting. People with HCV have a higher risk of developing cardiovascular disease than those without HCV. HCV accounted for 1·5 million DALYs due to cardiovascular disease worldwide in 2015, with the highest burden in South Asia, eastern Europe, north Africa, and the Middle East. Most of the disease burden is borne by LMICs, where HCV prevalence is projected to rise substantially. Our findings are of public health importance and could inform future research and health-care policies to improve risk stratification and treatment strategies aimed at reducing the combined global burden of HCV and extrahepatic sequelae such as cardiovascular disease. For the extraction databases, data used to derive pooled estimates, burden estimates, and R code script see https://github.com/kk-lee/hcv Supplementary Material Supplementary appendix

People with HCV have a higher risk of developing cardiovascular disease than those without HCV. HCV accounted for 1·5 million DALYs due to cardiovascular disease worldwide in 2015, with the highest burden in South Asia, eastern Europe, north Africa, and the Middle East. Most of the disease burden is borne by LMICs, where HCV prevalence is projected to rise substantially. Our findings are of public health importance and could inform future research and health-care policies to improve risk stratification and treatment strategies aimed at reducing the combined global burden of HCV and extrahepatic sequelae such as cardiovascular disease. For the extraction databases, data used to derive pooled estimates, burden estimates, and R code script see https://github.com/kk-lee/hcv Supplementary Material Supplementary appendix Acknowledgments This study was supported by the British Heart Foundation through a Clinical Research Training Fellowship (FS/18/25/33454), Intermediate Clinical Research Fellowship (FS/19/17/34172), Senior Clinical Research Fellowship (FS/16/14/32023) and a Research Excellence Award (RE/18/5/34216). DAM is funded by a Wellcome Trust Intermediate Clinical Fellowship (201492/Z/16/Z). DEN is funded by a Senior Investigator Award (WT103782AIA). The Polaris Observatory is supported by the John C Martin Foundation.

72), Senior Clinical Research Fellowship (FS/16/14/32023) and a Research Excellence Award (RE/18/5/34216). DAM is funded by a Wellcome Trust Intermediate Clinical Fellowship (201492/Z/16/Z). DEN is funded by a Senior Investigator Award (WT103782AIA). The Polaris Observatory is supported by the John C Martin Foundation. Contributors KKL, DS, and ASVS conceived and designed the study. KKL, DS, RB, MA, FS, SoB, and ASVS acquired the data. KKL and ASVS analysed and interpreted the data. KKL and ASVS drafted the initial manuscript. KKL, DS, RB, DEN, JSS, MHC, GSB, CTL, ShB, SK, SaB, HR, PRM, NLM, DAM, and ASVS critically reviewed the manuscript for intellectual content. All authors approved the final version of the report. Declaration of interests CTL reports grants from Gilead Sciences, outside the submitted work. SaB and HR report grants from Gilead and AbbVie, outside the submitted work. All other authors declare no competing interests.

Introduction In 2016, cancer accounted for more than 213 million disability-adjusted life-years (DALYs) and 8·9 million deaths globally.1, 2 The burden of cancer is usually reported in aggregated form,1, 3 but cancer-specific reports allow a more detailed exploration of the problem by providing information that is useful for the development and evaluation of cancer-specific prevention programmes, screening strategies, treatment, and resource allocation. An understanding of the geographical and temporal trends in colorectal cancer is important because it was the second leading cause of death (age-standardised and all ages) among cancers globally in 2017 and the 16th leading cause of death among all diseases and injuries.4 Trends in the burden of colorectal cancer have been subject to substantial changes across the world because of the expansion of screening programmes, with wide recommendation of colonoscopy in the late 1990s, as well as changes in risk factors associated with colorectal cancer.5, 6 Research in context Evidence before this study

Introduction In 2016, cancer accounted for more than 213 million disability-adjusted life-years (DALYs) and 8·9 million deaths globally.1, 2 The burden of cancer is usually reported in aggregated form,1, 3 but cancer-specific reports allow a more detailed exploration of the problem by providing information that is useful for the development and evaluation of cancer-specific prevention programmes, screening strategies, treatment, and resource allocation. An understanding of the geographical and temporal trends in colorectal cancer is important because it was the second leading cause of death (age-standardised and all ages) among cancers globally in 2017 and the 16th leading cause of death among all diseases and injuries.4 Trends in the burden of colorectal cancer have been subject to substantial changes across the world because of the expansion of screening programmes, with wide recommendation of colonoscopy in the late 1990s, as well as changes in risk factors associated with colorectal cancer.5, 6 Research in context Evidence before this study This study is part of the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD), which is the most comprehensive effort to date to measure epidemiological levels and trends. In its most up-to-date iteration, 359 diseases and injuries; 282 causes of death; and 84 behavioural, environmental and occupational, and metabolic risk factors were studied. The International Agency for Research on Cancer generates periodically updated estimates for all cancers including colorectal cancer in the Global Cancer Incidence, Mortality and Prevalence (GLOBOCAN) project. The burden of colorectal cancer has been investigated in previous research using GLOBOCAN data, but these studies have several limitations. The global burden of colorectal cancer is reported in terms of incidence and mortality, but important measures such as years of life lost, years lived with disability, and disability-adjusted life-years are not reported. The measures GLOBOCAN produces do not allow for comparability of the burden of disability or premature mortality between countries or with other causes. The temporal trends in GLOBOCAN estimates begin in 2002 and have occurred globally at 4-year or 6-year intervals with 95% uncertainty intervals provided only for the 2018 estimates. Using a consistent methodology to produce annual estimates dating back to 1990 provides a rich context for the burden estimates. Finally, the burden of colorectal cancer attributable to risk factors has not previously been calculated.

6-year intervals with 95% uncertainty intervals provided only for the 2018 estimates. Using a consistent methodology to produce annual estimates dating back to 1990 provides a rich context for the burden estimates. Finally, the burden of colorectal cancer attributable to risk factors has not previously been calculated. Added value of this study To our knowledge, this study is the first to report the incidence, mortality, and disability from colorectal cancer and its attributable risk factors from 1990 to 2017 in 195 countries and territories, by age, sex, Socio-demographic Index (a composite of sociodemographic factors), and Healthcare Access and Quality Index, an indicator of health system performance. Implications of all the available evidence Colorectal cancer remains a substantial public health challenge across the globe. Age-standardised incidence rates increased in most countries from 1990 to 2017, and the age-standardised death rate decreased at the global level and decreased particularly in countries high on the Socio-demographic Index. The burden of colorectal cancer was mainly attributed to dietary risks, alcohol use, and smoking. Further research is required to better understand the increases in incidence of colorectal cancer and to improve prevention, early detection, and treatment of this disease.

in countries high on the Socio-demographic Index. The burden of colorectal cancer was mainly attributed to dietary risks, alcohol use, and smoking. Further research is required to better understand the increases in incidence of colorectal cancer and to improve prevention, early detection, and treatment of this disease. Whereas colorectal cancer age-standardised death rates have stabilised or declined in many high-income countries, which historically had the highest burden of colorectal cancer in the world,7 the burden is increasing in most low-income and middle-income countries,8 possibly as a result of ageing populations, urbanisation, and increased prevalence of westernised lifestyle risk factors, such as alcohol consumption, obesity, smoking, and suboptimal diet.9, 10 The global burden of colorectal cancer attributable to various modifiable risk factors has not been described elsewhere and is an important estimate to report because it has implications for policy making and prevention efforts.

tyle risk factors, such as alcohol consumption, obesity, smoking, and suboptimal diet.9, 10 The global burden of colorectal cancer attributable to various modifiable risk factors has not been described elsewhere and is an important estimate to report because it has implications for policy making and prevention efforts. Studies reporting the global burden of colorectal cancer have been published previously but have several limitations. Specifically, previous estimates reported the global burden of colorectal cancer in terms of incidence and mortality but did not aim to calculate important measures such as years of life lost (YLLs), years lived with disability (YLDs), and DALYs.3, 7, 11, 12, 13, 14 Moreover, although the burden of colorectal cancer and trends associated with this disease have been reported up to 2018, the temporal trends occur at 4-year or 6-year intervals for most countries and 95% uncertainty intervals (UIs) have been provided only for the most recent global estimates in 2018.7, 11, 12, 13, 15 Finally, the association between countries' development status and colorectal cancer burden has previously been described using Global Cancer Incidence, Mortality and Prevalence (GLOBOCAN) data from only a subset of countries.16 We aimed to report the incidence, mortality, and disability due to colorectal cancer and its attributable risk factors from 1990 to 2017 in 195 countries and territories, by age, sex, Socio-demographic Index (SDI; a composite of socio-demographic factors), and Healthcare Access and Quality (HAQ) Index, an indicator of health system performance.

idence, mortality, and disability due to colorectal cancer and its attributable risk factors from 1990 to 2017 in 195 countries and territories, by age, sex, Socio-demographic Index (SDI; a composite of socio-demographic factors), and Healthcare Access and Quality (HAQ) Index, an indicator of health system performance. Methods Overview This study is part of the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD), which covers seven super-regions, consisting of 21 regions containing 195 countries and territories. The most up-to-date iteration, GBD 2017, reported estimates for 359 diseases and injuries; 282 causes of death; and 84 behavioural, environmental and occupational, and metabolic risk factors. The general methodology used and updates to the methodology have been previously presented in GBD 2017 papers.4, 17, 18, 19, 20, 21 Briefly, the mortality-to-incidence ratio (MIR) estimation was updated from GBD 2016, with use of the HAQ Index rather than the SDI in the data cleaning and modelling process, and the spatiotemporal Gaussian process regression approach was also updated. Covariate inputs for the Cause of Death Ensemble model (CODEm) were updated and changed on the basis of recommendations from GBD collaborators. The rates were standardised according to the GBD world population and reported per 100 000 person-years.17 The method for propagating uncertainty in this paper is similar to that used in previous GBD 2017 papers.4, 19 The distribution of every step in the computation process is stored in 1000 draws that are used for every other step in the process. The distributions are characterised from the sampling error of data inputs, the uncertainty of the model coefficients, MIRs, and age-specific death rates. GBD assumes that uncertainty in the MIR is independent of uncertainty in the estimated age-specific death rates. Final estimates were computed using the mean estimate across 1000 draws, and the 95% UIs were specified on the basis of the 25th and 975th ranked values across all 1000 draws. The GBD study is compliant with the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER).

mated age-specific death rates. Final estimates were computed using the mean estimate across 1000 draws, and the 95% UIs were specified on the basis of the 25th and 975th ranked values across all 1000 draws. The GBD study is compliant with the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER). Data sources All cancers coded as C18–C21, D01.0–D01.2, and D12–D12.8 in the 10th revision of the International Classification of Diseases were considered to be colorectal cancer.19 Vital registration (18 857 site-years of data), sample vital registration (761 site-years), verbal autopsy (660 site-years), and cancer registry (4474 site-years) data from GBD 2017 were used in this study.4 Vital registration is the system by which governments record the vital events of their residents, including causes of death. In sample vital registration, vital events are recorded in nationally representative cluster samples to estimate birth rates, deaths rates, and causes of death for the total population in countries where high coverage of vital registration is not available. Verbal autopsy is a method by which trained interviewers collect information about the signs, symptoms, and demographic characteristics of a recently deceased person from an individual familiar with the deceased to determine individuals' causes of death and cause-specific mortality fractions in populations without a complete vital registration system. Finally, a cancer registry gathers data on every person with cancer in a defined population, usually comprising residents in a well defined geographical region. The details on data quality rating for 195 countries and territories are provided in the appendix (pp 11–17). More detailed information about the data sources used for each country can be found on the GBD 2017 Data Input Sources Tool website.

pulation, usually comprising residents in a well defined geographical region. The details on data quality rating for 195 countries and territories are provided in the appendix (pp 11–17). More detailed information about the data sources used for each country can be found on the GBD 2017 Data Input Sources Tool website. Mortality estimates Mortality data from vital registration, sample vital registration, and verbal autopsy were sparse. Therefore, incidence data from cancer registries were converted into mortality data by modelling the MIRs independently. We modelled MIRs using the locations that had both incidence and mortality data available for the same year. The initial MIR model used a linear-step mixed-effects model with logit link functions, as well as the HAQ Index, age, and sex as covariates. The resulting estimates were then smoothed over space and time, and adjusted with spatiotemporal Gaussian process regression.18 We used the observed mortality (from vital registration and verbal autopsy) and mortality estimates (computed from the MIRs and incidence data) as inputs for a CODEm.4 Country-level covariates used for the CODEm and the assumed directions are described in the appendix (p 18). We used CODEm to select which predictors produce the best fit to the data. We used the CoDCorrect algorithm to adjust the sum of predicted single-cause mortalities in an age–sex–location–year group to be consistent with the results from all-cause mortality estimation.4

d directions are described in the appendix (p 18). We used CODEm to select which predictors produce the best fit to the data. We used the CoDCorrect algorithm to adjust the sum of predicted single-cause mortalities in an age–sex–location–year group to be consistent with the results from all-cause mortality estimation.4 Non-fatal estimates The final mortality estimates were divided by the MIR to compute colorectal cancer incidence.19 Colorectal cancer prevalence was calculated by estimating 10-year survival based on MIRs and adjusting for expected background mortality. The cohort members who had survived more than 10 years were assumed to be cured, and one of the two sequelae were assigned to them: the diagnosis and primary therapy phase or the controlled phase. The controlled phase included all patients who survived more than 10 years and who had finished primary therapy. The prevalence for the cohort in which people died during the 10-year period was categorised into four sequelae (appendix p 20). The diagnosis and primary therapy phase was defined as 4·0 months, the metastatic phase as 9·7 months, and terminal phase as 1 month.22, 23 The remaining time was assigned to the controlled phase. The duration of sequela one (diagnosis and primary therapy) described by Allgar and colleagues22 was used and 2 months were added to account for the average treatment duration. Duration of sequela two (controlled phase) was 10 years for the survivors minus the duration of the other sequelae. Duration of sequela three (metastatic phase) was based on Surveillance, Epidemiology, and End Results (SEER) data for median survival of patients with stage IV disease. A duration of 1 month for sequela four (terminal phase) was used for all cancers.22

0 years for the survivors minus the duration of the other sequelae. Duration of sequela three (metastatic phase) was based on Surveillance, Epidemiology, and End Results (SEER) data for median survival of patients with stage IV disease. A duration of 1 month for sequela four (terminal phase) was used for all cancers.22 To estimate procedure-related disability for all locations by age, sex, and year, we used hospital data on the proportion of patients that undergo ostomies (ie, the procedure proportion) as our input for a DisMod-MR 2.1 proportion model.19 We determined through a literature review that an average of 58% of all ostomies are for colorectal cancer, so we multiplied the all-cause ostomies by 0·58.24, 25, 26 We applied these procedure proportions to the number of incident cases of colorectal cancer and multiplied that by the proportion of the incident population that had survived for 10 years. This process gave us the number of incident cases of colorectal cancer that involved an ostomy procedure and survived beyond 10 years. We then input these cases into DisMod-MR 2.1. This model produced estimates of incidence and lifetime prevalent cases of people with colorectal cancer-related stomas who have survived beyond 10 years.19

us the number of incident cases of colorectal cancer that involved an ostomy procedure and survived beyond 10 years. We then input these cases into DisMod-MR 2.1. This model produced estimates of incidence and lifetime prevalent cases of people with colorectal cancer-related stomas who have survived beyond 10 years.19 Following this process, to estimate the sequela-specific YLDs, procedure sequelae prevalence and general sequela prevalence rates were multiplied by the sequela-specific disability weight. The disability weights for four sequelae and one procedure can be found in the appendix (p 19).19 The disability weights ranged from 0 (perfect health) to 1 (equivalent to death). GBD uses different disability weights for the four phases of colorectal cancer, but these weights are the same for all cancers. YLLs were calculated by multiplying the estimated number of deaths by age with a standard life expectancy at that age. Finally, DALYs were calculated by summing YLDs and YLLs.

Following this process, to estimate the sequela-specific YLDs, procedure sequelae prevalence and general sequela prevalence rates were multiplied by the sequela-specific disability weight. The disability weights for four sequelae and one procedure can be found in the appendix (p 19).19 The disability weights ranged from 0 (perfect health) to 1 (equivalent to death). GBD uses different disability weights for the four phases of colorectal cancer, but these weights are the same for all cancers. YLLs were calculated by multiplying the estimated number of deaths by age with a standard life expectancy at that age. Finally, DALYs were calculated by summing YLDs and YLLs. SDI and HAQ Index We used the GBD 2017 SDI and GBD 2016 HAQ Index to determine the association a country's development level had with colorectal cancer age-standardised DALY rates. Examining the association of development level (SDI) and health system performance (HAQ Index) with colorectal cancer burden is important because these factors affect the prevalence of cancer risk factors. In GBD 2017, the SDI was revised to better reflect the development status of each country.4, 18, 19, 20, 21 The SDI ranges from 0 (worst) to 1 (best) and incorporates the total fertility rate in women under the age of 25 years, mean education for individuals aged 15 years and older, and lag-distributed income per person. The HAQ Index reflects the personal health-care access and quality for 195 countries and territories and was calculated on the basis of amenable mortality (ie, deaths from causes that should not occur in the presence of effective medical care). The HAQ Index ranges from 0 (worst) to 100 (best). Further details on the HAQ Index are presented elsewhere.27

health-care access and quality for 195 countries and territories and was calculated on the basis of amenable mortality (ie, deaths from causes that should not occur in the presence of effective medical care). The HAQ Index ranges from 0 (worst) to 100 (best). Further details on the HAQ Index are presented elsewhere.27 Risk factors We selected risk factors that had evidence of causation with colorectal cancer. We extracted the relative risks and exposure estimates from all available data sources. We calculated a population attributable fraction as the proportional reduction in a health outcome that would occur if exposure to a risk factor was reduced to the theoretical minimum exposure level. We reported the proportion of DALYs due to colorectal cancer that were attributable to smoking, high body-mass index, high fasting plasma glucose, low physical activity, and five dietary risks (diets low in calcium, milk, and fibre, and diets high in red meat and processed meat). Details on definitions of these risk factors and their relative risk for colorectal cancer, prevalence of risk factors, and methods for quantifying the proportion of the burden of colorectal cancer attributable to these risk factors are described elsewhere.18 The DALYs due to colorectal cancer that were attributable to each risk factor were estimated by multiplying the total DALYs for colorectal cancer by the population attributable fraction for the risk–outcome pair for each age group, sex, location, and year.

r attributable to these risk factors are described elsewhere.18 The DALYs due to colorectal cancer that were attributable to each risk factor were estimated by multiplying the total DALYs for colorectal cancer by the population attributable fraction for the risk–outcome pair for each age group, sex, location, and year. Role of the funding source The funder of the study had no role in study design; the collection, analysis, or interpretation of the data; or the writing of the report. The corresponding authors had full access to the data and had responsibility for final submission of the manuscript. Results In 2017, there were 1·8 million (95% UI 1·8–1·9) incident cases of colorectal cancer, with an age-standardised incidence rate of 23·2 (22·7–23·7) per 100 000 person-years. The age-standardised incidence rate showed an increase of 9·5% (4·5–13·5) from 1990 to 2017 (table). Colorectal cancer also accounted for 896 000 (876 300–915 700) deaths globally, with an age-standardised death rate of 11·5 (11·3–11·8) per 100 000 person-years and a decrease in age-standardised death rates from 1990 to 2017 (−13·5% [–18·4 to −10·0]; appendix pp 21–29). Colorectal cancer was responsible for 19·0 million (18·5–19·5) DALYs globally, with an age-standardised rate of 235·7 (229·7–242·0) DALYs per 100 000 person-years. The age-standardised DALY rate decreased from 1990 to 2017 (−14·5% [–20·4 to −10·3]; appendix pp 30–39).Table Incident cases of colorectal cancer for both sexes and percentage change in age-standardised rates by location, 1990–2017

ly, with an age-standardised rate of 235·7 (229·7–242·0) DALYs per 100 000 person-years. The age-standardised DALY rate decreased from 1990 to 2017 (−14·5% [–20·4 to −10·3]; appendix pp 30–39).Table Incident cases of colorectal cancer for both sexes and percentage change in age-standardised rates by location, 1990–2017 1990 2017 Percentage change in age-standardised incidence rates, 1990–2017 Incident cases Age-standardised incidence rate (per 100 000 person-years) Incident cases Age-standardised incidence rate (per 100 000 person-years) Global 826 357 (807 380 to 854 834) 21·2 (20·7 to 21·9) 1 833 451 (1 791 865 to 1 873 464) 23·2 (22·7 to 23·7) 9·5% (4·5 to 13·5) Central Europe, eastern Europe, and central Asia Central Asia 5534 (5430 to 5645) 11·2 (11·0 to 11·4) 8977 (8558 to 9410) 12·3 (11·8 to 12·9) 10·0% (5·3 to 14·8) Armenia 418 (397 to 441) 14·7 (14·0 to 15·5) 772 (728 to 815) 18·7 (17·7 to 19·8) 27·1% (18·1 to 36·5) Azerbaijan 536 (506 to 568) 9·9 (9·3 to 10·4) 1210 (1028 to 1383) 12·9 (11·0 to 14·6) 30·1% (11·6 to 49·0) Georgia 737 (698 to 779) 11·7 (11·1 to 12·4) 901 (836 to 964) 15·7 (14·6 to 16·8) 34·2% (23·4 to 46·0) Kazakhstan 2064 (1992 to 2146) 15·5 (15·0 to 16·1) 2773 (2566 to 3009) 16·4 (15·2 to 17·8) 6·2% (−1·9 to 13·6) Kyrgyzstan 387 (364 to 411) 12·4 (11·7 to 13·2) 372 (344 to 421) 8·5 (7·9 to 9·5) −31·5% (−37·3 to −22·6) Mongolia 91 (83 to 102) 8·5 (7·7 to 9·4) 183 (158 to 206) 8·2 (7·1 to 9·3) −2·6% (−19·4 to 14·1) Tajikistan 229 (216 to 242) 7·5 (7·1 to 7·9) 430 (385 to 482) 8·0 (7·2 to 8·9) 6·8% (−4·8 to 18·8) Turkmenistan 155 (148 to 163) 7·4 (7·1 to 7·8) 353 (325 to 384) 9·6 (8·8 to 10·4) 28·8% (16·4 to 42·3) Uzbekistan 917 (882 to 953) 7·4 (7·1 to 7·7) 1982 (1757 to 2219) 9·5 (8·4 to 10·6) 28·2% (14·1 to 42·8) Central Europe 41 719 (41 148 to 42 319) 27·7 (27·3 to 28·1) 72 984 (70 812 to 75 162) 34·6 (33·5 to 35·6) 24·8% (20·8 to 29·0) Albania 184 (171 to 218) 8·1 (7·5 to 10·0) 454 (371 to 552) 11·2 (9·2 to 13·5) 37·5% (10·6 to 69·3) Bosnia and Herzegovina 671 (627 to 778) 16·2 (15·2 to 18·9) 1735 (1584 to 1896) 29·4 (27·0 to 32·0) 81·8% (61·7 to 100·4) Bulgaria 3092 (2983 to 3202) 24·1 (23·3 to 24·9) 5156 (4765 to 5545) 35·5 (32·8 to 38·1) 47·5% (35·6 to 59·5) Croatia 2211 (2126 to 2297) 34·4 (33·1 to 35·7) 3993 (3720 to 4278) 45·9 (42·9 to 49·2) 33·6% (23·0 to 44·4) Czech Republic 6800 (6579 to 7013) 48·6 (47·1 to 50·1) 8320 (7750 to 8966) 40·1 (37·4 to 43·2) −17·5% (−23·9 to −10·3) Hungary 6117 (5932 to 6300) 40·8 (39·6 to 41·9) 8454 (7883 to 90

38·1) 47·5% (35·6 to 59·5) Croatia 2211 (2126 to 2297) 34·4 (33·1 to 35·7) 3993 (3720 to 4278) 45·9 (42·9 to 49·2) 33·6% (23·0 to 44·4) Czech Republic 6800 (6579 to 7013) 48·6 (47·1 to 50·1) 8320 (7750 to 8966) 40·1 (37·4 to 43·2) −17·5% (−23·9 to −10·3) Hungary 6117 (5932 to 6300) 40·8 (39·6 to 41·9) 8454 (7883 to 90 40) 44·7 (41·7 to 47·7) 9·8% (2·1 to 18·0) Macedonia 304 (284 to 335) 15·9 (14·8 to 17·8) 812 (722 to 914) 24·1 (21·4 to 27·1) 51·6% (26·6 to 74·6) Montenegro 124 (113 to 136) 19·7 (17·9 to 21·5) 238 (215 to 265) 23·8 (21·6 to 26·6) 21·1% (6·2 to 38·2) Poland 10 892 (10 582 to 11 200) 24·1 (23·4 to 24·8) 20 482 (19 092 to 22 015) 29·7 (27·7 to 31·8) 23·3% (14·4 to 32·6) Romania 4736 (4576 to 4913) 16·5 (16·0 to 17·1) 10 989 (10 254 to 11 753) 30·5 (28·5 to 32·6) 84·4% (70·8 to 98·9) Serbia 3475 (3188 to 3840) 30·1 (27·7 to 33·1) 5971 (5507 to 6494) 38·4 (35·4 to 41·7) 27·5% (14·4 to 41·3) Slovakia 2275 (2178 to 2372) 37·5 (36·0 to 39·1) 4739 (4289 to 5177) 52·4 (47·5 to 57·1) 39·8% (24·3 to 54·4) Slovenia 837 (801 to 877) 33·6 (32·2 to 35·1) 1639 (1508 to 1785) 39·4 (36·2 to 43·0) 17·4% (7·3 to 29·1) Eastern Europe 68 421 (66 610 to 71 088) 23·8 (23·2 to 24·7) 103 116 (100 177 to 106 623) 30·2 (29·3 to 31·2) 26·8% (23·4 to 30·6) Belarus 2904 (2798 to 2999) 21·9 (21·1 to 22·6) 4478 (4078 to 5121) 28·3 (25·7 to 32·5) 29·1% (17·1 to 47·1) Estonia 588 (563 to 613) 27·9 (26·8 to 29·1) 929 (801 to 1065) 34·8 (30·1 to 40·2) 24·8% (6·8 to 44·4) Latvia 892 (862 to 925) 24·2 (23·3 to 25·0) 1205 (1066 to 1360) 29·8 (26·2 to 33·7) 23·2% (7·9 to 40·5) Lithuania 1061 (1026 to 1098) 22·9 (22·1 to 23·6) 1683 (1558 to 1806) 29·2 (27·1 to 31·4) 27·9% (17·5 to 38·6) Moldova 981 (941 to 1018) 21·1 (20·2 to 21·9) 1437 (1349 to 1539) 25·2 (23·6 to 26·9) 19·3% (10·8 to 28·5) Russia 42 907 (41 400 to 45 268) 23·1 (22·3 to 24·4) 69 283 (67 424 to 71 061) 29·9 (29·2 to 30·7) 29·5% (23·5 to 34·9) Ukraine 19 089 (18 412 to 19 805) 26·0 (25·1 to 26·9) 24 101 (22 571 to 25 877) 31·5 (29·6 to 33·8) 21·5% (13·4 to 30·2) High income Australasia 11 968 (11 694 to 12 218) 50·2 (49·1 to 51·2) 22 266 (20 408 to 24 232) 46·4 (42·5 to 50·6) −7·4% (−15·7 to 1·0) Australia 9497 (9253 to 9741) 47·8 (46·6 to 48·9) 18 429 (16 592 to 20 418) 45·7 (41·2 to 50·8) −4·3% (−14·6 to 6·3) New Zealand 2472 (2365 to 2589) 62·1 (59·5 to 64·9) 3837 (3562 to 4144) 50·2 (46·6 to 54·2) −19·1% (−25·9 to −12·1) High-income Asia Pacific 67 498 (66 180 to 68 809) 33·2 (32·6 to 33·9) 183 789 (175 950 to 193 063) 41·9 (40·2 t

741) 47·8 (46·6 to 48·9) 18 429 (16 592 to 20 418) 45·7 (41·2 to 50·8) −4·3% (−14·6 to 6·3) New Zealand 2472 (2365 to 2589) 62·1 (59·5 to 64·9) 3837 (3562 to 4144) 50·2 (46·6 to 54·2) −19·1% (−25·9 to −12·1) High-income Asia Pacific 67 498 (66 180 to 68 809) 33·2 (32·6 to 33·9) 183 789 (175 950 to 193 063) 41·9 (40·2 t o 44·1) 26·1% (20·8 to 32·1) Brunei 32 (28 to 38) 31·2 (26·7 to 36·4) 139 (127 to 154) 43·8 (39·8 to 48·6) 40·5% (16·6 to 66·7) Japan 62 351 (61 081 to 63 664) 36·4 (35·6 to 37·1) 153 905 (146 718 to 161 765) 45·0 (43·1 to 47·3) 23·8% (18·3 to 30·0) Singapore 778 (751 to 808) 34·1 (32·9 to 35·3) 2394 (2213 to 2622) 34·9 (32·2 to 38·1) 2·4% (−6·3 to 12·4) South Korea 4337 (4184 to 4495) 14·3 (13·8 to 14·8) 27 351 (24 820 to 30 076) 32·5 (29·5 to 35·7) 127·3% (105·1 to 150·7) High-income North America 165 322 (163 317 to 167 704) 45·6 (45·0 to 46·2) 234 927 (228 060 to 241 844) 39·1 (37·9 to 40·3) −14·2% (−17·2 to −11·4) Canada 13 301 (12 833 to 13 790) 40·1 (38·7 to 41·5) 25 661 (23 835 to 27 580) 38·5 (35·8 to 41·4) −3·8% (−11·7 to 4·3) Greenland 15 (13 to 16) 45·4 (40·4 to 50·3) 25 (23 to 28) 39·0 (35·5 to 42·4) −14·2% (−26·5 to −1·2) USA 152 002 (150 137 to 154 241) 46·1 (45·6 to 46·8) 209 237 (203 167 to 215 912) 39·1 (38·0 to 40·4) −15·1% (−18·2 to −12·0) Southern Latin America 9098 (8881 to 9339) 19·5 (19·1 to 20·1) 20 898 (19 394 to 22 657) 25·5 (23·6 to 27·6) 30·4% (20·2 to 41·5) Argentina 6650 (6439 to 6875) 20·4 (19·8 to 21·0) 13 927 (12 487 to 15 469) 26·1 (23·4 to 29·0) 28·1% (14·8 to 43·3) Chile 1341 (1285 to 1396) 13·4 (12·9 to 14·0) 5154 (4626 to 5746) 22·2 (19·9 to 24·8) 65·5% (46·7 to 86·6) Uruguay 1106 (1067 to 1145) 27·8 (26·8 to 28·7) 1817 (1620 to 2025) 33·3 (29·6 to 37·3) 20·0% (6·1 to 34·5) Western Europe 220 737 (217 920 to 223 500) 37·3 (36·8 to 37·7) 347 288 (332 898 to 361 454) 38·7 (37·1 to 40·3) 3·8% (−0·6 to 8·1) Andorra 21 (17 to 26) 36·1 (29·6 to 44·0) 52 (42 to 63) 38·3 (30·8 to 46·1) 6·1% (−14·8 to 28·9) Austria 4883 (4718 to 5065) 40·8 (39·4 to 42·2) 5592 (5201 to 6011) 31·5 (29·3 to 33·9) −22·6% (−28·6 to −16·3) Belgium 6258 (6013 to 6521) 39·7 (38·2 to 41·2) 8141 (7518 to 8809) 35·5 (32·7 to 38·4) −10·5% (−18·0 to −2·3) Cyprus 170 (150 to 197) 19·8 (17·5 to 23·0) 551 (492 to 620) 29·0 (26·0 to 32·6) 46·1% (20·5 to 75·4) Denmark 2330 (2261 to 2399) 28·4 (27·6 to 29·2) 5175 (4762 to 5593) 45·6 (42·0 to 49·3) 60·5% (47·4 to 74·7) Finland 1784 (1731 to 1838) 24·7 (23·9 to 25·4) 3437 (3197 to 3725) 28·9 (26·9 to 31·3) 17·3% (7·5 to

o −2·3) Cyprus 170 (150 to 197) 19·8 (17·5 to 23·0) 551 (492 to 620) 29·0 (26·0 to 32·6) 46·1% (20·5 to 75·4) Denmark 2330 (2261 to 2399) 28·4 (27·6 to 29·2) 5175 (4762 to 5593) 45·6 (42·0 to 49·3) 60·5% (47·4 to 74·7) Finland 1784 (1731 to 1838) 24·7 (23·9 to 25·4) 3437 (3197 to 3725) 28·9 (26·9 to 31·3) 17·3% (7·5 to 27·9) France 29 412 (28 397 to 30 488) 34·3 (33·1 to 35·5) 45 501 (41 853 to 49 486) 33·0 (30·4 to 36·0) −3·7% (−11·7 to 5·4) Germany 59 179 (57 557 to 60 958) 45·4 (44·2 to 46·7) 76 179 (68 038 to 84 803) 41·1 (36·7 to 45·8) −9·4% (−19·0 to 0·9) Greece 2661 (2540 to 2784) 17·3 (16·5 to 18·0) 6556 (6083 to 7025) 27·6 (25·6 to 29·6) 60·1% (46·8 to 73·3) Iceland 87 (82 to 92) 29·7 (28·0 to 31·5) 169 (157 to 182) 31·7 (29·3 to 34·0) 6·5% (−3·4 to 16·7) Ireland 1643 (1582 to 1705) 39·7 (38·2 to 41·1) 2948 (2661 to 3280) 40·6 (36·7 to 45·2) 2·5% (−8·2 to 14·0) Israel 1307 (1251 to 1380) 26·7 (25·6 to 28·1) 3165 (2921 to 3438) 27·9 (25·8 to 30·4) 4·5% (−4·1 to 13·6) Italy 30 748 (29 557 to 31 888) 34·2 (32·9 to 35·4) 52 228 (48 427 to 56 835) 37·2 (34·3 to 40·4) 8·8% (−0·5 to 18·7) Luxembourg 233 (221 to 247) 41·8 (39·7 to 44·2) 409 (359 to 475) 42·1 (37·0 to 49·3) 0·9% (−11·7 to 17·5) Malta 109 (104 to 116) 25·4 (24·1 to 27·0) 306 (281 to 333) 34·4 (31·7 to 37·2) 35·2% (22·9 to 48·4) Netherlands 8553 (8241 to 8849) 41·9 (40·4 to 43·4) 16 948 (15 727 to 18 222) 50·9 (47·1 to 54·7) 21·3% (11·9 to 31·7) Norway 2861 (2811 to 2917) 41·7 (40·9 to 42·5) 4556 (4316 to 4796) 48·4 (46·0 to 51·0) 16·2% (9·8 to 22·7) Portugal 4052 (3901 to 4207) 29·5 (28·5 to 30·6) 9390 (8696 to 10288) 41·4 (38·4 to 45·3) 40·3% (28·5 to 54·8) Spain 17 169 (16 664 to 17 708) 30·8 (29·9 to 31·7) 41 133 (38 218 to 44 436) 43·4 (40·2 to 47·0) 40·8% (29·2 to 53·7) Sweden 5106 (4972 to 5255) 33·0 (32·1 to 33·9) 7130 (6693 to 7575) 34·7 (32·7 to 36·8) 5·1% (−1·6 to 12·2) Switzerland 2227 (2137 to 2321) 20·9 (20·1 to 21·8) 5032 (4597 to 5547) 29·4 (26·9 to 32·5) 40·3% (26·6 to 56·1) UK 39 729 (39 124 to 40 372) 42·7 (42·1 to 43·4) 52 331 (51 067 to 53 737) 41·7 (40·7 to 42·9) −2·3% (−5·2 to 1·0) Latin America and Caribbean Andean Latin America 1770 (1608 to 1997) 8·7 (7·9 to 9·7) 7635 (6901 to 8372) 14·2 (12·9 to 15·6) 64·3% (41·6 to 89·3) Bolivia 324 (201 to 527) 10·3 (6·5 to 16·5) 1092 (799 to 1460) 13·2 (9·7 to 17·6) 28·0% (−9·2 to 77·0) Ecuador 415 (400 to 432) 7·7 (7·4 to 8·0) 1954 (1769 to 2160) 13·4 (12·2 to 14·8) 73·8% (56·4 to 93·4) Peru 1031 (947 to 1121) 8·6 (7·9 to 9·4) 4589 (3917 to 5349)

01 to 8372) 14·2 (12·9 to 15·6) 64·3% (41·6 to 89·3) Bolivia 324 (201 to 527) 10·3 (6·5 to 16·5) 1092 (799 to 1460) 13·2 (9·7 to 17·6) 28·0% (−9·2 to 77·0) Ecuador 415 (400 to 432) 7·7 (7·4 to 8·0) 1954 (1769 to 2160) 13·4 (12·2 to 14·8) 73·8% (56·4 to 93·4) Peru 1031 (947 to 1121) 8·6 (7·9 to 9·4) 4589 (3917 to 5349) 15·0 (12·8 to 17·5) 73·6% (45·2 to 107·2) Caribbean 4453 (4299 to 4655) 17·1 (16·5 to 17·8) 11 943 (11 109 to 12 868) 23·5 (21·8 to 25·3) 37·6% (29·2 to 47·0) Antigua and Barbuda 7 (7 to 8) 13·9 (13·0 to 15·0) 20 (18 to 21) 19·9 (18·2 to 21·7) 42·9% (27·1 to 59·8) The Bahamas 33 (30 to 35) 20·5 (19·2 to 22·0) 95 (85 to 105) 25·8 (23·2 to 28·6) 25·7% (9·6 to 44·1) Barbados 66 (62 to 71) 22·0 (20·7 to 23·3) 153 (138 to 168) 31·8 (28·6 to 34·8) 44·6% (28·4 to 61·7) Belize 7 (7 to 8) 7·7 (7·0 to 8·5) 30 (27 to 32) 11·4 (10·5 to 12·5) 47·6% (29·9 to 65·3) Bermuda 20 (19 to 22) 32·3 (30·0 to 34·4) 46 (41 to 50) 36·0 (32·4 to 39·6) 11·4% (−1·7 to 28·7) Cuba 2285 (2210 to 2369) 22·0 (21·2 to 22·7) 5629 (4988 to 6293) 29·9 (26·5 to 33·4) 36·0% (21·3 to 52·9) Dominica 9 (8 to 10) 12·0 (11·2 to 12·9) 16 (15 to 18) 17·4 (15·9 to 19·1) 44·8% (27·8 to 61·8) Dominican Republic 283 (260 to 307) 7·4 (6·8 to 8·0) 1277 (1100 to 1467) 13·9 (11·9 to 16·0) 87·7% (57·2 to 120·5) Grenada 11 (10 to 12) 15·6 (14·6 to 16·6) 29 (27 to 32) 18·7 (17·1 to 20·3) 19·6% (7·5 to 32·1) Guyana 41 (39 to 44) 10·6 (10·0 to 11·2) 79 (70 to 89) 13·2 (11·7 to 14·8) 23·9% (8·2 to 41·6) Haiti 333 (236 to 517) 10·7 (7·8 to 16·1) 803 (577 to 1164) 12·8 (9·4 to 18·1) 19·2% (−4·1 to 51·4) Jamaica 243 (229 to 262) 13·3 (12·5 to 14·2) 639 (538 to 743) 22·0 (18·5 to 25·6) 66·1% (36·5 to 96·9) Puerto Rico 726 (694 to 757) 19·5 (18·7 to 20·3) 2084 (1921 to 2254) 30·4 (28·1 to 32·9) 55·5% (43·2 to 69·2) Saint Lucia 11 (11 to 12) 12·7 (12·0 to 13·4) 31 (29 to 34) 15·1 (13·9 to 16·4) 18·9% (7·5 to 31·4) Saint Vincent and the Grenadines 10 (9 to 10) 12·9 (12·1 to 13·9) 23 (21 to 25) 16·5 (15·1 to 18·0) 27·7% (13·9 to 43·0) Suriname 33 (30 to 35) 12·9 (12·0 to 13·8) 107 (96 to 119) 19·0 (17·1 to 21·0) 46·7% (29·5 to 66·4) Trinidad and Tobago 158 (150 to 167) 18·6 (17·7 to 19·6) 373 (308 to 447) 20·9 (17·3 to 24·9) 12·4% (−7·2 to 35·7) Virgin Islands 23 (21 to 25) 27·7 (25·1 to 30·5) 80 (69 to 90) 43·5 (37·6 to 49·1) 57·2% (31·8 to 83·3) Central Latin America 7618 (7492 to 7774) 8·9 (8·8 to 9·1) 35 294 (33 818 to 36 661) 15·2 (14·6 to 15·8) 70·4% (62·5 to 77·8) Colombia 2012 (1933 to 2094) 11·4 (10·9 to 11·8) 8683 (7757

to 24·9) 12·4% (−7·2 to 35·7) Virgin Islands 23 (21 to 25) 27·7 (25·1 to 30·5) 80 (69 to 90) 43·5 (37·6 to 49·1) 57·2% (31·8 to 83·3) Central Latin America 7618 (7492 to 7774) 8·9 (8·8 to 9·1) 35 294 (33 818 to 36 661) 15·2 (14·6 to 15·8) 70·4% (62·5 to 77·8) Colombia 2012 (1933 to 2094) 11·4 (10·9 to 11·8) 8683 (7757 to 9798) 16·1 (14·4 to 18·2) 41·5% (25·3 to 60·4) Costa Rica 267 (255 to 279) 15·1 (14·4 to 15·7) 1397 (1264 to 1518) 28·4 (25·7 to 30·9) 88·8% (69·4 to 108·3) El Salvador 184 (172 to 203) 6·1 (5·7 to 6·6) 869 (731 to 1022) 15·2 (12·8 to 17·8) 149·8% (105·5 to 197·4) Guatemala 185 (177 to 193) 5·2 (4·9 to 5·4) 1010 (906 to 1118) 9·3 (8·4 to 10·3) 79·4% (59·3 to 101·3) Honduras 125 (111 to 140) 5·8 (5·1 to 6·5) 582 (443 to 718) 9·7 (7·4 to 11·9) 67·1% (29·4 to 110·8) Mexico 3381 (3313 to 3458) 7·7 (7·5 to 7·8) 16 550 (15 933 to 17 026) 14·5 (13·9 to 14·9) 88·9% (80·2 to 95·3) Nicaragua 115 (105 to 126) 7·0 (6·4 to 7·6) 501 (439 to 573) 10·9 (9·6 to 12·5) 56·8% (34·3 to 82·6) Panama 194 (184 to 203) 12·8 (12·2 to 13·4) 737 (673 to 802) 18·6 (17·0 to 20·2) 45·0% (31·2 to 59·6) Venezuela 1155 (1109 to 1204) 11·8 (11·3 to 12·2) 4965 (4272 to 5735) 17·9 (15·4 to 20·5) 52·0% (29·8 to 77·8) Tropical Latin America 9583 (9343 to 9871) 10·5 (10·3 to 10·8) 37 656 (36 473 to 38 850) 16·2 (15·7 to 16·8) 54·2% (46·9 to 60·5) Brazil 9426 (9184 to 9708) 10·6 (10·4 to 10·9) 36 934 (35 748 to 38 099) 16·3 (15·8 to 16·8) 53·5% (46·3 to 59·9) Paraguay 157 (143 to 171) 7·2 (6·6 to 7·9) 722 (599 to 860) 13·9 (11·5 to 16·5) 91·7% (56·2 to 131·7) North Africa and Middle East North Africa and Middle East 15 515 (13 256 to 19 992) 8·8 (7·6 to 11·2) 52 224 (49 748 to 54 659) 12·4 (11·8 to 12·9) 39·9% (7·3 to 65·8) Afghanistan 747 (306 to 1691) 10·6 (4·6 to 23·4) 1458 (806 to 2773) 12·9 (7·7 to 22·9) 21·0% (−10·0 to 101·4) Algeria 857 (763 to 956) 6·9 (6·1 to 7·7) 2821 (2488 to 3132) 8·6 (7·6 to 9·6) 25·7% (3·0 to 48·8) Bahrain 22 (20 to 25) 11·6 (10·0 to 13·6) 110 (96 to 126) 11·3 (10·0 to 12·6) −2·8% (−20·8 to 22·0) Egypt 1476 (1357 to 1618) 4·9 (4·6 to 5·5) 4500 (3742 to 5195) 7·3 (6·1 to 8·4) 48·6% (16·9 to 76·2) Iran 2280 (1950 to 2824) 8·6 (7·4 to 10·5) 9784 (8702 to 10304) 14·0 (12·5 to 14·7) 63·3% (27·5 to 94·0) Iraq 631 (505 to 809) 8·1 (6·5 to 10·1) 1309 (1183 to 1430) 5·6 (5·0 to 6·0) −31·1% (−47·4 to −12·1) Jordan 191 (156 to 230) 12·6 (10·2 to 15·0) 940 (782 to 1095) 15·9 (13·3 to 18·5) 26·8% (−5·6 to 65·9) Kuwait 63 (59 to 68) 8·2 (7·7 to 8·9) 290 (249 to 344) 10·9 (9·3 to 12

304) 14·0 (12·5 to 14·7) 63·3% (27·5 to 94·0) Iraq 631 (505 to 809) 8·1 (6·5 to 10·1) 1309 (1183 to 1430) 5·6 (5·0 to 6·0) −31·1% (−47·4 to −12·1) Jordan 191 (156 to 230) 12·6 (10·2 to 15·0) 940 (782 to 1095) 15·9 (13·3 to 18·5) 26·8% (−5·6 to 65·9) Kuwait 63 (59 to 68) 8·2 (7·7 to 8·9) 290 (249 to 344) 10·9 (9·3 to 12 ·9) 32·9% (15·8 to 58·7) Lebanon 383 (323 to 449) 17·3 (14·6 to 20·2) 1692 (1404 to 1994) 28·1 (23·4 to 33·2) 62·7% (26·4 to 101·4) Libya 281 (227 to 357) 14·2 (11·6 to 17·8) 1053 (889 to 1240) 21·8 (18·4 to 25·4) 52·8% (15·2 to 102·0) Morocco 897 (768 to 1063) 6·2 (5·3 to 7·3) 2754 (2248 to 3298) 8·8 (7·2 to 10·5) 41·8% (3·5 to 85·6) Oman 53 (42 to 68) 7·5 (5·9 to 9·4) 222 (180 to 265) 10·9 (9·1 to 12·8) 46·3% (2·3 to 95·9) Palestine 149 (117 to 191) 16·1 (12·7 to 20·5) 432 (387 to 474) 17·0 (15·2 to 18·6) 5·5% (−22·2 to 40·8) Qatar 19 (15 to 24) 17·1 (14·3 to 20·8) 158 (132 to 189) 17·8 (15·0 to 21·0) 4·1% (−21·0 to 36·4) Saudi Arabia 438 (339 to 576) 6·7 (5·2 to 8·7) 3000 (2539 to 3528) 16·6 (14·2 to 18·9) 149·2% (76·9 to 242·9) Sudan 627 (407 to 1087) 6·6 (4·4 to 11·1) 1509 (1111 to 2104) 8·3 (6·3 to 11·4) 25·8% (−11·2 to 81·4) Syria 384 (318 to 477) 6·9 (5·8 to 8·6) 1237 (1018 to 1501) 9·7 (8·0 to 11·7) 39·5% (2·1 to 79·1) Tunisia 431 (378 to 499) 8·8 (7·7 to 10·1) 1476 (1164 to 1833) 12·3 (9·7 to 15·2) 40·5% (0·8 to 86·2) Turkey 5162 (4082 to 6691) 14·0 (11·2 to 18·0) 15 436 (13 838 to 17 433) 17·6 (15·8 to 20·0) 26·1% (−6·9 to 58·6) United Arab Emirates 60 (43 to 83) 13·3 (9·4 to 18·6) 759 (598 to 940) 19·9 (16·3 to 24·1) 50·3% (−0·1 to 121·4) Yemen 354 (189 to 634) 6·9 (4·0 to 11·8) 1234 (868 to 1820) 9·5 (6·9 to 13·5) 38·2% (−5·6 to 123·3) South Asia South Asia 36 162 (31 934 to 43 729) 6·2 (5·5 to 7·4) 104 958 (93 845 to 113 041) 8·1 (7·2 to 8·7) 31·6% (1·8 to 55·6) Bangladesh 4935 (4048 to 6317) 9·6 (8·0 to 12·3) 10 188 (8726 to 12 073) 8·4 (7·2 to 10·0) −12·5% (−35·5 to 12·9) Bhutan 17 (12 to 25) 6·6 (4·9 to 9·7) 48 (36 to 62) 8·1 (6·2 to 10·3) 22·1% (−18·4 to 79·2) India 26 950 (23 572 to 33 017) 5·8 (5·1 to 7·0) 82 775 (74 559 to 89 201) 7·9 (7·1 to 8·6) 37·5% (6·0 to 64·0) Nepal 547 (373 to 843) 5·8 (4·0 to 8·8) 1438 (1157 to 1841) 7·0 (5·6 to 8·9) 20·1% (−13·0 to 63·7) Pakistan 3713 (3334 to 4094) 6·4 (5·7 to 7·0) 10 509 (7826 to 12 968) 9·4 (7·0 to 11·4) 47·1% (10·5 to 82·3) Southeast Asia, east Asia, and Oceania East Asia 114 366 (107 795 to 125 264) 12·3 (11·6 to 13·5) 462 088 (438 223 to 483 591) 22·8 (21·6 to 23·9) 85·2% (63·9 to 102·6) Chin

·0 (5·6 to 8·9) 20·1% (−13·0 to 63·7) Pakistan 3713 (3334 to 4094) 6·4 (5·7 to 7·0) 10 509 (7826 to 12 968) 9·4 (7·0 to 11·4) 47·1% (10·5 to 82·3) Southeast Asia, east Asia, and Oceania East Asia 114 366 (107 795 to 125 264) 12·3 (11·6 to 13·5) 462 088 (438 223 to 483 591) 22·8 (21·6 to 23·9) 85·2% (63·9 to 102·6) Chin a 107 038 (100 408 to 117 587) 12·2 (11·4 to 13·4) 431 951 (408 225 to 452 721) 22·4 (21·2 to 23·5) 84·1% (62·0 to 102·2) North Korea 2095 (1683 to 2551) 12·2 (9·9 to 14·8) 4483 (3552 to 5524) 14·3 (11·3 to 17·5) 16·8% (−9·7 to 51·5) Taiwan (province of China) 3327 (3242 to 3418) 19·9 (19·4 to 20·4) 18 209 (17 062 to 19 442) 48·0 (45·1 to 51·3) 141·9% (126·3 to 158·5) Oceania 308 (252 to 447) 10·0 (8·5 to 14·2) 745 (617 to 1031) 11·2 (9·8 to 14·8) 12·1% (−3·2 to 27·8) American Samoa 4 (3 to 4) 15·9 (14·1 to 17·6) 8 (7 to 9) 18·5 (16·5 to 20·7) 16·7% (−0·5 to 38·5) Federated States of Micronesia 6 (5 to 8) 11·9 (9·7 to 15·5) 9 (7 to 12) 13·7 (10·9 to 16·9) 15·3% (−6·7 to 39·9) Fiji 34 (29 to 41) 9·3 (7·9 to 10·9) 82 (68 to 95) 11·8 (9·8 to 13·5) 26·6% (−0·4 to 57·5) Guam 16 (14 to 18) 18·8 (16·9 to 21·0) 43 (38 to 47) 23·8 (21·6 to 26·4) 26·5% (7·3 to 50·9) Kiribati 4 (3 to 4) 9·5 (8·4 to 10·5) 7 (6 to 8) 10·3 (8·5 to 12·2) 9·0% (−14·1 to 33·0) Marshall Islands 2 (2 to 3) 14·1 (10·6 to 20·0) 6 (4 to 7) 17·2 (13·7 to 22·2) 21·9% (2·0 to 47·8) Northern Mariana Islands 3 (3 to 4) 16·5 (14·2 to 19·9) 9 (8 to 10) 17·8 (15·7 to 20·0) 7·3% (−12·7 to 28·8) Papua New Guinea 187 (139 to 298) 9·3 (7·2 to 14·6) 469 (355 to 737) 10·0 (7·9 to 15·3) 8·0% (−11·0 to 31·5) Samoa 9 (7 to 11) 10·2 (8·3 to 12·9) 15 (12 to 18) 11·2 (9·2 to 13·5) 9·6% (−14·4 to 39·0) Solomon Islands 11 (9 to 17) 7·9 (6·2 to 11·8) 29 (23 to 38) 9·0 (7·4 to 11·7) 13·8% (−7·9 to 37·3) Tonga 4 (4 to 5) 7·9 (7·0 to 9·1) 7 (6 to 8) 9·4 (7·9 to 10·9) 19·4% (−6·8 to 46·8) Vanuatu 8 (6 to 11) 11·5 (8·9 to 16·0) 21 (16 to 28) 12·9 (9·8 to 17·1) 12·5% (−15·1 to 45·0) Southeast Asia 27 105 (23 553 to 32 801) 10·4 (9·1 to 12·5) 85 149 (80 680 to 90 557) 14·7 (14·0 to 15·6) 40·9% (15·4 to 63·2) Cambodia 572 (339 to 1013) 12·4 (7·5 to 21·6) 1445 (1086 to 1940) 13·1 (10·0 to 17·4) 5·6% (−25·7 to 56·9) Indonesia 7946 (6669 to 10 070) 7·8 (6·6 to 9·9) 18 739 (17 443 to 20 172) 9·3 (8·6 to 10·0) 18·0% (−9·8 to 43·7) Laos 251 (154 to 408) 11·9 (7·5 to 19·1) 488 (377 to 650) 11·7 (9·2 to 15·4) −1·6% (−29·0 to 40·8) Malaysia 1800 (1555 to 2184) 20·5 (17·5 to 24·7) 6605 (5777 to 7568) 26·9 (23·6 to 30·6) 31·1% (5·5 to 55·9)

donesia 7946 (6669 to 10 070) 7·8 (6·6 to 9·9) 18 739 (17 443 to 20 172) 9·3 (8·6 to 10·0) 18·0% (−9·8 to 43·7) Laos 251 (154 to 408) 11·9 (7·5 to 19·1) 488 (377 to 650) 11·7 (9·2 to 15·4) −1·6% (−29·0 to 40·8) Malaysia 1800 (1555 to 2184) 20·5 (17·5 to 24·7) 6605 (5777 to 7568) 26·9 (23·6 to 30·6) 31·1% (5·5 to 55·9) Maldives 7 (5 to 12) 7·8 (5·3 to 11·9) 27 (24 to 31) 9·1 (7·9 to 10·2) 15·6% (−31·1 to 78·3) Mauritius 74 (70 to 78) 9·9 (9·4 to 10·5) 322 (294 to 351) 19·4 (17·6 to 21·1) 95·4% (76·0 to 115·6) Myanmar 3328 (1899 to 5522) 14·2 (8·3 to 23·3) 6560 (4900 to 9053) 14·9 (11·2 to 20·6) 5·4% (−21·4 to 55·0) Philippines 2039 (1903 to 2160) 6·5 (6·1 to 6·9) 13 472 (11 799 to 15 373) 18·4 (16·1 to 20·8) 181·9% (145·4 to 227·0) Sri Lanka 654 (610 to 705) 6·0 (5·6 to 6·4) 2451 (1929 to 2976) 10·1 (8·0 to 12·2) 69·4% (32·7 to 110·5) Seychelles 9 (8 to 10) 15·0 (13·6 to 17·9) 39 (35 to 42) 36·4 (32·6 to 39·8) 142·4% (84·6 to 176·7) Thailand 4671 (4283 to 5179) 12·4 (11·3 to 13·7) 15 598 (13 999 to 17 415) 16·0 (14·3 to 17·8) 29·3% (10·2 to 47·1) Timor-Leste 20 (15 to 32) 7·1 (5·4 to 10·6) 79 (63 to 102) 10·2 (8·1 to 13·0) 43·1% (1·8 to 94·3) Vietnam 5697 (4925 to 6498) 13·8 (12·0 to 15·8) 19 210 (16 530 to 22 243) 21·0 (18·2 to 24·1) 51·9% (21·3 to 87·3) Sub-Saharan Africa Central sub-Saharan Africa 1904 (1499 to 2534) 8·7 (7·2 to 11·1) 4416 (3711 to 5434) 9·2 (7·9 to 10·9) 5·2% (−11·9 to 25·7) Angola 373 (246 to 563) 9·7 (6·8 to 14·0) 1049 (857 to 1289) 10·3 (8·4 to 12·4) 6·0% (−26·2 to 53·2) Central African Republic 111 (64 to 180) 9·8 (6·1 to 15·2) 202 (115 to 335) 9·8 (6·1 to 15·4) 0·0% (−20·2 to 24·5) Congo (Brazzaville) 129 (96 to 171) 12·2 (9·5 to 15·4) 300 (234 to 386) 12·8 (10·3 to 15·7) 5·2% (−18·2 to 38·8) Democratic Republic of the Congo 1205 (953 to 1563) 8·0 (6·6 to 10·1) 2676 (2098 to 3493) 8·4 (6·8 to 10·5) 4·2% (−15·7 to 29·5) Equatorial Guinea 19 (11 to 31) 9·9 (6·4 to 15·2) 54 (35 to 78) 12·2 (8·0 to 17·1) 22·1% (−35·7 to 107·4) Gabon 66 (51 to 88) 12·0 (9·4 to 15·5) 134 (101 to 168) 13·2 (9·9 to 16·3) 10·2% (−21·1 to 42·6) Eastern sub-Saharan Africa 7703 (6131 to 9924) 10·5 (8·6 to 13·3) 16 007 (14 839 to 17 000) 10·7 (9·9 to 11·3) 1·2% (−20·0 to 26·6) Burundi 184 (141 to 247) 8·7 (6·7 to 11·4) 328 (258 to 429) 8·4 (6·8 to 10·8) −2·8% (−22·1 to 22·1) Comoros 24 (19 to 31) 11·8 (9·4 to 15·4) 52 (43 to 64) 12·0 (9·8 to 14·6) 1·9% (−20·9 to 34·3) Djibouti 20 (13 to 30) 13·4 (8·9 to 19·9) 77 (53 to 109) 14·4 (10·2 to 19·9) 7·2% (−25·2 to 60·7) Eritrea 124 (87

0 to 26·6) Burundi 184 (141 to 247) 8·7 (6·7 to 11·4) 328 (258 to 429) 8·4 (6·8 to 10·8) −2·8% (−22·1 to 22·1) Comoros 24 (19 to 31) 11·8 (9·4 to 15·4) 52 (43 to 64) 12·0 (9·8 to 14·6) 1·9% (−20·9 to 34·3) Djibouti 20 (13 to 30) 13·4 (8·9 to 19·9) 77 (53 to 109) 14·4 (10·2 to 19·9) 7·2% (−25·2 to 60·7) Eritrea 124 (87 to 180) 13·0 (9·5 to 18·7) 320 (248 to 412) 14·3 (11·4 to 17·9) 9·9% (−20·3 to 58·5) Ethiopia 2566 (1412 to 3859) 14·1 (8·3 to 20·6) 4375 (3873 to 4821) 11·7 (10·4 to 12·8) −17·1% (−44·2 to 45·3) Kenya 680 (565 to 832) 8·2 (6·8 to 10·1) 1966 (1754 to 2229) 9·6 (8·6 to 10·9) 17·5% (2·2 to 32·4) Madagascar 527 (403 to 726) 10·1 (7·7 to 13·7) 1021 (771 to 1385) 10·0 (7·6 to 13·4) −1·1% (−19·1 to 22·1) Malawi 194 (128 to 235) 5·0 (3·5 to 5·9) 438 (352 to 523) 6·1 (4·9 to 7·2) 22·3% (−5·5 to 79·8) Mozambique 697 (599 to 803) 12·1 (10·4 to 13·9) 1503 (1236 to 1830) 14·4 (12·1 to 17·3) 19·2% (−9·8 to 55·4) Rwanda 239 (179 to 322) 8·4 (6·2 to 11·2) 448 (285 to 604) 8·2 (5·3 to 11·0) −2·0% (−25·6 to 28·8) Somalia 281 (144 to 472) 11·2 (6·4 to 18·3) 766 (495 to 1188) 12·7 (8·4 to 19·4) 13·6% (−17·4 to 72·5) South Sudan 244 (146 to 383) 10·5 (6·7 to 15·9) 402 (283 to 588) 11·1 (7·8 to 16·0) 5·7% (−22·6 to 52·3) Tanzania 1055 (805 to 1321) 10·2 (8·2 to 12·7) 2307 (1954 to 2715) 10·1 (8·6 to 11·8) −1·3% (−25·3 to 27·9) Uganda 504 (430 to 588) 7·7 (6·6 to 9·0) 1263 (1044 to 1499) 9·5 (7·9 to 11·2) 22·1% (−2·8 to 53·2) Zambia 361 (287 to 444) 13·1 (10·7 to 15·8) 731 (622 to 841) 12·2 (10·4 to 13·9) −7·0% (−30·2 to 21·9) Southern sub-Saharan Africa 2591 (2398 to 2816) 9·3 (8·5 to 10·2) 6002 (5469 to 6404) 11·1 (10·1 to 11·8) 19·1% (10·4 to 26·8) Botswana 51 (42 to 61) 9·1 (7·6 to 10·7) 134 (111 to 170) 10·5 (8·8 to 13·0) 15·9% (−10·1 to 44·4) Lesotho 70 (58 to 95) 7·4 (6·1 to 9·8) 122 (95 to 153) 10·7 (8·4 to 13·3) 43·8% (11·1 to 81·6) Namibia 56 (46 to 72) 7·8 (6·5 to 10·0) 118 (100 to 139) 8·6 (7·3 to 10·1) 9·8% (−21·3 to 45·5) South Africa 1989 (1803 to 2212) 9·3 (8·3 to 10·5) 4774 (4223 to 5154) 11·1 (9·8 to 12·0) 18·7% (10·5 to 27·2) Swaziland (eSwatini) 30 (25 to 39) 10·7 (8·9 to 13·5) 72 (56 to 91) 13·3 (10·5 to 16·4) 24·8% (−6·0 to 59·6) Zimbabwe 395 (349 to 446) 9·7 (8·6 to 10·9) 782 (667 to 915) 11·7 (10·0 to 13·6) 21·0% (1·3 to 44·7) Western sub-Saharan Africa 6983 (5677 to 9178) 8·2 (6·7 to 10·7) 15 089 (12 862 to 17 883) 9·0 (7·7 to 10·5) 8·8% (−18·3 to 39·8) Benin 132 (105 to 159) 6·6 (5·3 to 7·9) 350 (280 to 444) 7·9 (6·4 to 9·9) 20·0% (−4·6 to 53·3) Burkina Faso

to 446) 9·7 (8·6 to 10·9) 782 (667 to 915) 11·7 (10·0 to 13·6) 21·0% (1·3 to 44·7) Western sub-Saharan Africa 6983 (5677 to 9178) 8·2 (6·7 to 10·7) 15 089 (12 862 to 17 883) 9·0 (7·7 to 10·5) 8·8% (−18·3 to 39·8) Benin 132 (105 to 159) 6·6 (5·3 to 7·9) 350 (280 to 444) 7·9 (6·4 to 9·9) 20·0% (−4·6 to 53·3) Burkina Faso 508 (408 to 592) 12·3 (10·1 to 14·3) 1146 (955 to 1366) 13·7 (11·5 to 16·1) 11·2% (−13·0 to 43·3) Cameroon 365 (308 to 425) 8·5 (7·2 to 9·9) 1015 (746 to 1290) 9·4 (7·0 to 11·9) 11·3% (−13·6 to 37·9) Cape Verde 9 (8 to 10) 4·0 (3·6 to 4·4) 40 (36 to 44) 8·9 (8·0 to 9·8) 124·9% (95·5 to 160·7) Chad 182 (141 to 252) 6·5 (5·0 to 8·9) 422 (320 to 564) 8·2 (6·3 to 10·8) 27·1% (5·0 to 54·9) Côte d'Ivoire 220 (193 to 250) 5·7 (5·0 to 6·4) 554 (438 to 698) 5·8 (4·7 to 7·3) 2·5% (−21·4 to 31·0) The Gambia 20 (17 to 24) 5·9 (5·0 to 6·9) 60 (43 to 80) 6·6 (4·8 to 8·8) 12·8% (−26·3 to 56·6) Ghana 484 (397 to 613) 7·9 (6·6 to 9·8) 1384 (1113 to 1656) 9·5 (7·6 to 11·2) 19·5% (−18·0 to 58·4) Guinea 198 (175 to 221) 6·0 (5·4 to 6·7) 403 (323 to 512) 7·7 (6·2 to 9·7) 28·7% (1·5 to 64·7) Guinea-Bissau 45 (24 to 72) 11·4 (6·4 to 17·9) 72 (50 to 98) 10·8 (7·8 to 14·5) −5·0% (−29·9 to 37·4) Liberia 87 (69 to 115) 7·8 (6·2 to 10·2) 164 (122 to 230) 9·0 (6·8 to 12·5) 15·8% (−8·9 to 45·2) Mali 298 (264 to 339) 7·5 (6·7 to 8·5) 625 (454 to 856) 7·8 (5·7 to 10·6) 3·8% (−26·6 to 44·8) Mauritania 92 (68 to 126) 9·1 (6·8 to 12·4) 181 (138 to 234) 9·5 (7·3 to 12·2) 4·6% (−20·3 to 44·4) Niger 183 (127 to 267) 6·7 (4·7 to 9·7) 449 (327 to 650) 6·6 (4·9 to 9·5) −1·0% (−18·3 to 19·3) Nigeria 3623 (2486 to 5445) 8·5 (5·9 to 12·7) 7096 (5194 to 9621) 9·2 (6·9 to 12·3) 8·5% (−27·9 to 65·9) São Tomé and Príncipe 6 (5 to 6) 8·4 (7·5 to 9·5) 13 (10 to 18) 13·8 (10·6 to 17·9) 63·1% (22·4 to 115·0) Senegal 293 (231 to 381) 9·3 (7·3 to 12·0) 564 (438 to 697) 8·3 (6·5 to 10·2) −10·8% (−39·8 to 28·9) Sierra Leone 158 (113 to 220) 8·2 (5·9 to 11·3) 303 (229 to 403) 9·3 (7·1 to 12·3) 14·1% (−9·6 to 44·1) Togo 80 (63 to 96) 6·5 (5·2 to 7·9) 247 (190 to 316) 7·7 (6·0 to 9·6) 17·3% (−4·9 to 43·2) Data in parentheses are 95% uncertainty intervals.

·3 (7·3 to 12·0) 564 (438 to 697) 8·3 (6·5 to 10·2) −10·8% (−39·8 to 28·9) Sierra Leone 158 (113 to 220) 8·2 (5·9 to 11·3) 303 (229 to 403) 9·3 (7·1 to 12·3) 14·1% (−9·6 to 44·1) Togo 80 (63 to 96) 6·5 (5·2 to 7·9) 247 (190 to 316) 7·7 (6·0 to 9·6) 17·3% (−4·9 to 43·2) Data in parentheses are 95% uncertainty intervals. Australasia (46·4 [95% UI 42·5–50·6] per 100 000 person-years), high-income Asia Pacific (41·9 [40·2–44·1] per 100 000 person-years), and high-income North America (39·1 [37·9–40·3] per 100 000 person-years) had the highest age-standardised incidence rates in 2017. By contrast, south Asia (8·1 [7·2–8·7] per 100 000 person-years), western sub-Saharan Africa (9·0 [7·7–10·5] per 100 000 person-years), and central sub-Saharan Africa (9·2 [7·9–10·9] per 100 000 person-years) had the lowest age-standardised incidence rates in 2017 (table). In all regions except Andean Latin America, the age-standardised incidence rate was higher among males than females in 2017 (figure 1A). The age-standardised death rates in 2017 were highest in central Europe (20·9 [20·3–21·6] per 100 000 person-years), eastern Europe (16·4 [16·0–16·9] per 100 000 person-years), and southern Latin America (16·1 [14·9–17·4] per 100 000 person-years). By contrast, south Asia (7·1 [6·4–7·6] per 100 000 person-years), north Africa and the Middle East (8·0 [7·6–8·3] per 100 000 person-years), and central Latin America (8·0 [7·7–8·3] per 100 000 person-years) had the lowest age-standardised death rates in 2017 (appendix pp 21–29). The age-standardised death rates in 2017 were higher for males in all GBD regions (figure 1B).Figure 1 The age-standardised incidence (A) and death (B) rates of colorectal cancer for 21 GBD regions by sex, 2017

[7·7–8·3] per 100 000 person-years) had the lowest age-standardised death rates in 2017 (appendix pp 21–29). The age-standardised death rates in 2017 were higher for males in all GBD regions (figure 1B).Figure 1 The age-standardised incidence (A) and death (B) rates of colorectal cancer for 21 GBD regions by sex, 2017 Error bars indicate 95% uncertainty intervals. GBD=Global Burden of Diseases, Injuries, and Risk Factors Study.

[7·7–8·3] per 100 000 person-years) had the lowest age-standardised death rates in 2017 (appendix pp 21–29). The age-standardised death rates in 2017 were higher for males in all GBD regions (figure 1B).Figure 1 The age-standardised incidence (A) and death (B) rates of colorectal cancer for 21 GBD regions by sex, 2017 Error bars indicate 95% uncertainty intervals. GBD=Global Burden of Diseases, Injuries, and Risk Factors Study. The percentage change in age-standardised incidence rates from 1990 to 2017 differed substantially between the GBD regions, with east Asia (85·2% [95% UI 63·9 to 102·6]), central Latin America (70·4% [62·5 to 77·8]), and Andean Latin America (64·3% [41·6 to 89·3]) showing the largest increases. By contrast, high-income North America (−14·2% [–17·2 to −11·4]) and Australasia (−7·4% [–15·7 to 1·0]) showed decreasing trends during this period, although the decrease for Australasia was not significant (table). The percentage change in age-standardised death rates from 1990 to 2017 also differed between the GBD regions. The largest increases were seen in south Asia (20·4% [–6·2 to 42·8]), central Latin America (20·4% [15·0 to 25·4]), and tropical Latin America (18·2% [12·9 to 22·6]). By contrast, the largest decreases during this period were found in Australasia (−34·0% [–39·2 to −28·6]), high-income North America (−30·0% [–32·2 to −27·8]), and western Europe (−26·1% [–29·1 to −23·1]; appendix pp 21–29). Percentage change increments in age-standardised incidence rates of colorectal cancer from 1990 to 2017 were higher among males in most regions except Andean Latin America and south Asia (figure 2A). Similarly, percentage change increments for colorectal cancer age-standardised death rates in this period were highest in males in most regions, except for south Asia (figure 2B). In 2017, the highest number of incident cases were found in east Asia, western Europe, and high-income North America (table; appendix p 1). The highest numbers of deaths were in east Asia, western Europe, and high-income North America in 2017 (appendix pp 2, 21–28).Figure 2 The percentage change in age-standardised incidence (A) and death (B) rates of colorectal cancer for 21 GBD regions by sex, 1990–2017

nd high-income North America (table; appendix p 1). The highest numbers of deaths were in east Asia, western Europe, and high-income North America in 2017 (appendix pp 2, 21–28).Figure 2 The percentage change in age-standardised incidence (A) and death (B) rates of colorectal cancer for 21 GBD regions by sex, 1990–2017 Error bars indicate 95% uncertainty intervals. GBD=Global Burden of Diseases, Injuries, and Risk Factors Study. In 2017, the age-standardised incidence rates for colorectal cancer were highest in Slovakia (52·4 [95% UI 47·5–57·1] per 100 000 person-years), the Netherlands (50·9 [47·1–54·7] per 100 000 person-years), and New Zealand (50·2 [46·6–54·2] per 100 000 person-years). The lowest age-standardised rates in 2017 were found in Iraq (5·6 [5·0–6·0] per 100 000 person-years), Côte d'Ivoire (5·8 [4·7–7·3] per 100 000 person-years), and Malawi (6·1 [4·9–7·2] per 100 000 person-years; figure 3A; table). In 2017, the age-standardised death rates were highest in Greenland (26·5 [24·2–28·8] per 100 000 person-years), Hungary (26·1 [24·5–27·8] per 100 000 person-years), and Slovakia (24·5 [21·9–26·4] per 100 000 person-years). Conversely, Iraq (4·5 [4·1–4·9] per 100 000 person-years), Maldives (5·1 [4·4–5·7] per 100 000 person-years), and Egypt (5·3 [4·3–6·1] per 100 000 person-years) had the lowest age-standardised death rates in 2017 (figure 3B; appendix pp 21–29).Figure 3 Age-standardised incidence (A) and death (B) rate of colorectal cancer per 100 000 person-years by country and territory, 2017

Maldives (5·1 [4·4–5·7] per 100 000 person-years), and Egypt (5·3 [4·3–6·1] per 100 000 person-years) had the lowest age-standardised death rates in 2017 (figure 3B; appendix pp 21–29).Figure 3 Age-standardised incidence (A) and death (B) rate of colorectal cancer per 100 000 person-years by country and territory, 2017 ATG=Antigua and Barbuda. FSM=Federated States of Micronesia. Isl=Islands. LCA=Saint Lucia. TLS=Timor-Leste. TTO=Trinidad and Tobago. VCT=Saint Vincent and the Grenadines. The percentage change in age-standardised incidence rates from 1990 to 2017 differed substantially between countries, with the Philippines (181·9% [95% UI 145·4 to 227·0]), El Salvador (149·8% [105·5 to 197·4]), and Saudi Arabia (149·2% [76·9 to 242·9]) showing the largest increases. By contrast, Kyrgyzstan (−31·5% [–37·3 to −22·6]), Iraq (−31·1% [–47·4 to −12·1]), and Austria (−22·6% [–28·6 to −16·3]) showed the largest decreases in age-standardised incidence during this period (table). The percentage change in age-standardised death rates from 1990 to 2017 also differed between countries. The largest increases were seen in the Philippines (139·8% [109·4 to 176·2]), Cape Verde (108·5% [80·7 to 143·9]), and Seychelles (82·9% [40·7 to 107·9]). By contrast, the largest decreases during this period were found in Austria (−42·7% [–46·7 to −38·8]), the Czech Republic (−38·3% [–42·6 to −33·4]), and Singapore (−37·5% [–42·2 to −31·8]; appendix pp 21–29).

ippines (139·8% [109·4 to 176·2]), Cape Verde (108·5% [80·7 to 143·9]), and Seychelles (82·9% [40·7 to 107·9]). By contrast, the largest decreases during this period were found in Austria (−42·7% [–46·7 to −38·8]), the Czech Republic (−38·3% [–42·6 to −33·4]), and Singapore (−37·5% [–42·2 to −31·8]; appendix pp 21–29). Our study found that, in 2017, the incidence rate increased in a non-linear manner with increasing age and was higher in males than in females across all age groups (figure 4). The difference in incidence rates between males and females increased with each increasing age group up to the ages of 85–89 years, after which the gap started to decrease again. The number of incident cases was also higher in males than in females up to the ages of 80–84 years and peaked at ages 65–69 years (figure 4). A relatively similar pattern was also observed for death rates and death counts (appendix p 3). The highest rates of incidence and death observed were in the oldest age group (≥95 years) for both sexes in 2017. The pattern for DALY rates was slightly different, such that the age-standardised DALY rate started decreasing after the ages of 80–84 years for males and after the ages of 85–89 years for females (appendix p 4). The number of DALYs was also higher in males than in females up to the ages of 80–84 years, and then females had slightly higher numbers of DALYs for the older age groups. The number of DALYs followed a normal distribution and peaked at ages 65–69 years (appendix p 4). Decomposition of the DALY rate into YLLs and YLDs showed that YLLs were the primary contributor to DALYs, with the 2017 YLL rate peaking at the ages of 80–84 years (appendix p 5).Figure 4 Global number of incident cases and incidence rate of colorectal cancer per 100 000 person-years by age and sex, 2017

appendix p 4). Decomposition of the DALY rate into YLLs and YLDs showed that YLLs were the primary contributor to DALYs, with the 2017 YLL rate peaking at the ages of 80–84 years (appendix p 5).Figure 4 Global number of incident cases and incidence rate of colorectal cancer per 100 000 person-years by age and sex, 2017 Error bars indicate the 95% uncertainty interval for incident cases. Shading indicates the 95% uncertainty interval for the incidence rate. Figure 5 presents the global and regional-level observed age-standardised DALY rates from 1990 to 2017 versus the expected level based only on the SDI values of the global regions. The expected pattern was non-linear in nature, peaking at an SDI value of approximately 0·75, before decreasing with increasing SDI values. However, there were large regional differences. Australasia, central Europe, western Europe, and high-income North America showed the largest decreases in observed age-standardised DALY rates with increases in SDI value, whereas the Caribbean and central Latin American regions showed increases in observed age-standardised DALY rates with increasing SDI value. The observed age-standardised DALY rate for some regions, such as southern sub-Saharan Africa, initially increased and then decreased with an improvement in SDI value over time. At the global level, the age-standardised DALY rate dropped below the expected level for 2015–17.Figure 5 Age-standardised DALY rates per 100 000 person-years for colorectal cancer for 21 GBD regions by SDI, 1990–2017

sub-Saharan Africa, initially increased and then decreased with an improvement in SDI value over time. At the global level, the age-standardised DALY rate dropped below the expected level for 2015–17.Figure 5 Age-standardised DALY rates per 100 000 person-years for colorectal cancer for 21 GBD regions by SDI, 1990–2017 Expected values based on SDI and age-standardised DALY rates in all locations are shown as the black line. For each region, points from left to right depict estimates from each year from 1990 to 2017. DALY=disability-adjusted life-year. GBD=Global Burden of Diseases, Injuries, and Risk Factors Study. SDI=Socio-demographic Index. Figure 6 shows the national-level observed age-standardised DALY rates and their association with the SDI and HAQ Index. The expected patterns were non-linear in nature, peaking at an SDI value of approximately 0·81 and HAQ Index value of approximately 84, before decreasing with increasing SDI and HAQ Index values. However, there were large national differences. Several countries, including Hungary, Greenland, Slovakia, Serbia, and Brunei, had a higher than expected age-standardised DALY rate, whereas others, such as Iraq, Maldives, Sri Lanka, Kuwait, and Oman, had much lower than expected age-standardised DALY rates based only on the SDI. This pattern was also observed based on the HAQ Index.Figure 6 Age-standardised DALY rates of colorectal cancer for 195 countries and territories by SDI and HAQ Index, 2017

hereas others, such as Iraq, Maldives, Sri Lanka, Kuwait, and Oman, had much lower than expected age-standardised DALY rates based only on the SDI. This pattern was also observed based on the HAQ Index.Figure 6 Age-standardised DALY rates of colorectal cancer for 195 countries and territories by SDI and HAQ Index, 2017 (A) Age-standardised DALY rates of colorectal cancer by 195 countries and the HAQ Index, 2016. (B) Expected values are shown as the black line. DALY=disability-adjusted life-year. HAQ Index=Healthcare Access and Quality Index. SDI=Socio-demographic Index.

hereas others, such as Iraq, Maldives, Sri Lanka, Kuwait, and Oman, had much lower than expected age-standardised DALY rates based only on the SDI. This pattern was also observed based on the HAQ Index.Figure 6 Age-standardised DALY rates of colorectal cancer for 195 countries and territories by SDI and HAQ Index, 2017 (A) Age-standardised DALY rates of colorectal cancer by 195 countries and the HAQ Index, 2016. (B) Expected values are shown as the black line. DALY=disability-adjusted life-year. HAQ Index=Healthcare Access and Quality Index. SDI=Socio-demographic Index. Although the proportions of age-standardised DALYs that were attributable to colorectal cancer risk factors differed in the GBD regions, diet low in calcium (20·5% [95% UI 12·9–28·9]), alcohol use (15·2% [12·1–18·3]), and diet low in milk (14·3% [5·1–24·8]) had the three highest percentages of attributable age-standardised DALYs for both sexes globally (figure 7; appendix p 11). This global pattern was different in males and females: alcohol use (21·5% [17·4–25·9]), diet low in calcium (19·8% [12·3–28·2]), and smoking (19·2% [12·8–25·3]) were the risk factors that contributed most to age-standardised DALYs in males, whereas diets low in calcium (21·3% [13·7–29·9]), milk (14·4% [5·1–24·0]), and fibre (12·5% [6·6–19·3]) were the risk factors that contributed most to age-standardised DALYs in females (appendix pp 6–7). The percentage of DALYs attributable to colorectal cancer risk factors also differed across age groups, especially for alcohol use, smoking, and high fasting plasma glucose. The highest percentage of global attributable DALYs were in the 55–59 years age group for alcohol use, 65–69 years age group for smoking, and 85–89 years age group for high fasting plasma glucose for both sexes combined (appendix p 8). The sex-specific estimates of global DALYs attributable to studied risk factors by age are reported in appendix (pp 9–10).Figure 7 Percentage of age-standardised DALYs due to colorectal cancer attributable to risk factors for 21 GBD regions, both sexes, 2017

asting plasma glucose for both sexes combined (appendix p 8). The sex-specific estimates of global DALYs attributable to studied risk factors by age are reported in appendix (pp 9–10).Figure 7 Percentage of age-standardised DALYs due to colorectal cancer attributable to risk factors for 21 GBD regions, both sexes, 2017 DALY=disability-adjusted life-year. GBD=Global Burden of Diseases, Injuries, and Risk Factors Study. Discussion From 1990 to 2017, the age-standardised incidence rates of colorectal cancer increased globally, with substantial regional and national heterogeneity. By contrast, the age-standardised death and DALY rates decreased across the study period. On the basis of our DALY estimates, colorectal cancer is the 36th leading cause of disease burden globally for 2017, and is the fourth leading cause of cancer burden, behind only lung cancer, liver cancer, and stomach cancer. The most recent GLOBOCAN report3 in 2018 estimated that there were 1 800 977 incident cases and 861 663 deaths from colorectal cancer, which are relatively consistent with our 2017 estimates (1 833 451 [95% UI 1 791 865–1 873 464] incident cases and 896 040 [876 279–915 720] deaths). Similar to the GLOBOCAN report,3 we found that the highest age-standardised incidence rates in 2017 were in Australasia, high-income Asia Pacific, and high-income North America, and the highest age-standardised death rates were found in central Europe, eastern Europe, and southern Latin America.

6 279–915 720] deaths). Similar to the GLOBOCAN report,3 we found that the highest age-standardised incidence rates in 2017 were in Australasia, high-income Asia Pacific, and high-income North America, and the highest age-standardised death rates were found in central Europe, eastern Europe, and southern Latin America. We also investigated heterogeneous trends in age-standardised incidence, death, and DALY rates from 1990 to 2017 at the national level. Most countries showed an increase in the age-standardised incidence rate of colorectal cancer during 1990–2017, such that only Australasia and high-income North America experienced a decrease in age-standardised incidence rate at the regional level. One potential explanation for this global increase in age-standardised incidence is that the introduction of screening tests might have led to increased detection and thus increased incidence, but this increase might be short-lived because of the removal of precancerous polyps during colonoscopies.5 Similarly, in countries where screening programmes were established two or three decades ago, reductions in death rates were observed that support the benefits attributable to screening interventions.28 Improving survival by adopting the best practices in cancer treatment and management can also lead to reduced death rates. On the basis of the data from high-income countries, several factors might have contributed to the decrease in the number of deaths due to colorectal cancer, such as enhanced access to screening colonoscopy and early stage detection, as well as improved surgical techniques, radiotherapy, chemotherapy, targeted therapy, and palliative care.29, 30, 31, 32 Key interventions to decrease deaths from colorectal cancer include the removal of polyps and early detection interventions, such as colonoscopy, flexible sigmoidoscopy, faecal occult blood testing, and faecal immunochemical testing.

ques, radiotherapy, chemotherapy, targeted therapy, and palliative care.29, 30, 31, 32 Key interventions to decrease deaths from colorectal cancer include the removal of polyps and early detection interventions, such as colonoscopy, flexible sigmoidoscopy, faecal occult blood testing, and faecal immunochemical testing. Previous research has investigated the association between a country's development level and the incidence and mortality rates of colorectal cancer, using the Human Development Index.7 Because one of the components of the Human Development Index is health related, it is not optimal to use this index when comparing the health outcomes of countries. To avoid this problem, we used the SDI, which does not contain any health-related measures. Our analysis of the association between the SDI and age-standardised DALY rate of colorectal cancer produced results that have not been previously reported. In some regions, such as Australasia and central Europe, the age-standardised DALY rates were higher than expected from 1990 to 2017, whereas in other regions they were lower than expected, such as in central Latin America and south Asia. Several regions also fluctuated between higher and lower than expected age-standardised DALY rates during the study period. Therefore, regional trends in age-standardised rates of deaths, DALYs, and incidence should not just be considered in isolation. Instead, their observed rates should be compared with their expected rates to determine whether regions have managed colorectal cancer better or worse than expected. Additionally, dividing regions or countries into developed and undeveloped regions to evaluate the association between development and colorectal cancer burden might be an oversimplification, since the findings from this study suggest that the association is complex and non-linear in nature. Furthermore, at the national level, age-standardised DALY rates of countries on both ends of the SDI range were at higher than expected levels based on the SDI, so countries at all development levels need to enhance their prevention programmes.

his study suggest that the association is complex and non-linear in nature. Furthermore, at the national level, age-standardised DALY rates of countries on both ends of the SDI range were at higher than expected levels based on the SDI, so countries at all development levels need to enhance their prevention programmes. Our report indicates that the colorectal cancer burden attributable to risk factors is different in males and females, and this difference should be considered in national policy makers' prevention programmes. Alcohol use, smoking, and diets low in calcium, milk, and fibre had considerable attributable colorectal cancer burden in males. By contrast, dietary risks, but not alcohol use or smoking, were found to have considerable attributable burden in females. The results of this study highlight the role of certain dietary risk factors, which are responsible for a greater burden than smoking or alcohol use globally. Specifically, diet low in calcium has been previously described as a risk factor for colorectal cancer.33 However, the large burden attributable to this dietary risk factor has not been described before at the global level. This large burden attributable to a diet low in calcium is likely due in part to the high prevalence of this risk factor. The results of our study underscore the importance of improving diet through public health interventions.

attributable to this dietary risk factor has not been described before at the global level. This large burden attributable to a diet low in calcium is likely due in part to the high prevalence of this risk factor. The results of our study underscore the importance of improving diet through public health interventions. A previous study showed that alcohol use is responsible for nearly 10% of global deaths in the population aged 15–49 years and will lead to remarkable health loss in the absence of appropriate policy action.34 The same study also reported that the safest level of alcohol consumption is zero, which is in contrast to current health guidelines.34 Decreasing population-level alcohol consumption should be considered in prevention strategies to effectively minimise the corresponding health loss.34 Smoking is another important risk factor. The global prevalence of smoking has decreased by 28% in males and 34% in females since 1990.18 Taxation, advertising bans, and educational programmes about smoking and its toll on health are suggested as strategies to decrease smoking prevalence more substantially.18

A previous study showed that alcohol use is responsible for nearly 10% of global deaths in the population aged 15–49 years and will lead to remarkable health loss in the absence of appropriate policy action.34 The same study also reported that the safest level of alcohol consumption is zero, which is in contrast to current health guidelines.34 Decreasing population-level alcohol consumption should be considered in prevention strategies to effectively minimise the corresponding health loss.34 Smoking is another important risk factor. The global prevalence of smoking has decreased by 28% in males and 34% in females since 1990.18 Taxation, advertising bans, and educational programmes about smoking and its toll on health are suggested as strategies to decrease smoking prevalence more substantially.18 Diets low in calcium, milk, and fibre should also be addressed in colorectal cancer prevention strategies along with addressing low physical activity. To improve diet, increase physical activity, and reduce smoking, the American Heart Association suggests following population-based approaches, such as media and educational campaigns; labelling and consumer information; taxation, subsidies, and other economic incentives; school and workplace approaches; local environmental changes; and direct restrictions and mandates.35

e smoking, the American Heart Association suggests following population-based approaches, such as media and educational campaigns; labelling and consumer information; taxation, subsidies, and other economic incentives; school and workplace approaches; local environmental changes; and direct restrictions and mandates.35 Our findings indicate that although high body-mass index was not among the top three risk factors for attributable DALYs, it is an important risk factor that has a higher attributable percentage of DALYs due to colorectal cancer in males than in females. One study showed that the prevalence of obesity has doubled in more than 70 countries and has continuously increased in most other countries since 1980.36 Effective prevention programmes are needed to decrease exposure to this important risk factor through appropriate strategies, such as restricting the advertisement of unhealthy foods, using taxation to reduce consumption of unhealthy foods, providing subsidies to increase intake of healthy foods, and using supply-chain incentives to increase the production of healthy foods.37 Fasting plasma glucose can also be controlled mainly through physical activity and healthy diets.38

ement of unhealthy foods, using taxation to reduce consumption of unhealthy foods, providing subsidies to increase intake of healthy foods, and using supply-chain incentives to increase the production of healthy foods.37 Fasting plasma glucose can also be controlled mainly through physical activity and healthy diets.38 This study had several limitations, including the fact that some of the variations in age-standardised incidence and mortality rates might be due to detection biases as well as changes in screening protocols. For example, the low age-standardised incidence and death rates in Iraq might be due to low detection rates. In addition, a major limitation of cancer burden research is the scarcity of data for many countries. Although different data sources, such as cancer registries, vital registration systems, and verbal autopsies, are used to produce cancer estimates, some countries do not have any of these sources available so their estimates are based on predictive covariates or trends from neighbouring countries. Moreover, estimates for the most recent years are usually based on past trends and covariates because there is a lag in data availability. As GBD is an iterative study, additional data sources for different locations will be added in future rounds and make the estimates more data driven, particularly in data-sparse locations.

r, estimates for the most recent years are usually based on past trends and covariates because there is a lag in data availability. As GBD is an iterative study, additional data sources for different locations will be added in future rounds and make the estimates more data driven, particularly in data-sparse locations. This study found large country and regional variations in the burden of colorectal cancer in 2017. Whereas age-standardised incidence rates increased in most countries and territories over the measurement period, age-standardised death rates decreased at the global level, and in particular in high SDI countries, possibly due to fast improvement in diagnostics and interventions in these countries. Further research is required to expand our knowledge of additional factors associated with colorectal cancer incidence and to improve early detection and treatment of this disease, especially in developing countries. Clearly, colorectal cancer remains a substantial public health challenge across the globe. The results of GBD 2017 can be valuable for policy makers to implement cost-effective interventions and address modifiable risk factors and for researchers to design and carry out further research on proper modalities for prevention, early detection, and treatment of colorectal cancer. This online publication has been corrected. The corrected version first appeared at thelancet.com/gastrohep on Feb 12, 2020 Supplementary Material Supplementary appendix

This study found large country and regional variations in the burden of colorectal cancer in 2017. Whereas age-standardised incidence rates increased in most countries and territories over the measurement period, age-standardised death rates decreased at the global level, and in particular in high SDI countries, possibly due to fast improvement in diagnostics and interventions in these countries. Further research is required to expand our knowledge of additional factors associated with colorectal cancer incidence and to improve early detection and treatment of this disease, especially in developing countries. Clearly, colorectal cancer remains a substantial public health challenge across the globe. The results of GBD 2017 can be valuable for policy makers to implement cost-effective interventions and address modifiable risk factors and for researchers to design and carry out further research on proper modalities for prevention, early detection, and treatment of colorectal cancer. This online publication has been corrected. The corrected version first appeared at thelancet.com/gastrohep on Feb 12, 2020 Supplementary Material Supplementary appendix Acknowledgments This study was funded by the Bill & Melinda Gates Foundation. AB is supported by the Public Health Agency of Canada. TB was supported by the Alexander von Humboldt Foundation through the Alexander von Humboldt Professor award, funded by the German Federal Ministry of Education and Research. AMS was supported by a fellowship from the Egyptian Fulbright Mission Program. FC and EF acknowledge support from UID/MULTI/04378/2019 and UID/QUI/50006/2019 with funding from Fundação para a Ciência e a Tecnologia/Ministério da Ciência, Tecnologia e Ensino Superior through Portuguese national funds. TM acknowledges institutional support from the Competence Cluster for Nutrition and Cardiovascular Health (nutriCARD), Jena-Halle-Leipzig (Germany). WM is Program Analyst Population and Development at the UN Population Fund-UNFPA Country Office in Peru, which does not necessarily endorse this study. MJ's contribution to this GBD study was co-financed by the Serbian Ministry of Education Science and Technological Development through grant OI 175 014. MSM acknowledges the support from the Serbian Ministry of Education, Science and Technological Development (contract number 175087). SA acknowledges International Centre for Casemix and Clinical Coding, Faculty of Medicine, National University of Malaysia, and Department of Health Policy and Management, Faculty of Public Health, Kuwait University, for the approval and support to participate in this research project.

er Jr PhD, Prof C J L Murray DPhil, Prof M Naghavi PhD), Division of Allergy and Infectious Diseases (K S Ikuta MD), Division of Hematology (C Fitzmaurice MD), Institute for Health Metrics and Evaluation (K S Ikuta MD, C Bisignano MPH, C Fitzmaurice MD, L M Force MD, M R Nixon PhD, Prof S I Hay FMedSci, S L James MD, J M Kocarnik PhD, Prof A D Lopez PhD, Prof A H Mokdad PhD, R C Reiner Jr PhD, Prof C J L Murray DPhil, Prof M Naghavi PhD), University of Washington, Seattle, WA, USA; Golestan Research Center of Gastroenterology and Hepatology (G Roshandel PhD, F Ghaseni-Kebria MSc, S Hasanpour-Heidari MSc, N Jafari Delouei MSc, A Sharifi PhD), Golestan University of Medical Sciences, Gorgan, Iran (S Mir MSc); Department of Oncology (L M Force MD), Department of Global Pediatric Medicine (L M Force MD), St Jude Children's Research Hospital, Memphis, TN, USA; Biostatistics and Health Informatics Department (K H Abegaz MPH), Madda Walabu University, Bale Robe, Oromia, Ethiopia; Radiotherapy Center (K H Abegaz MPH), School of Allied Health Sciences (E Yisma MPH), Addis Ababa University, Addis Ababa, Ethiopia (G T Demoz MPharm); Department of Epidemiology (M B Ahmed MPH), Jimma University, Jimma, Oromia, Ethiopia; Department of Population Health Sciences (T Akinyemiju PhD), Duke Global Health Institute (T Akinyemiju PhD), Duke University, Durham, North Carolina, USA; Evidence Based Practice Center (F Alahdab MD), Mayo Clinic Foundation for Medical Education and Research, Rochester, MN, USA; Public Health Research Center (R Ali MPH), New York University Abu Dhabi, Abu Dhabi, United Arab Emirates; Nuffield Department of Population Health (R Ali MPH), University of Oxford, Oxford, UK; Colorectal Research Center (A Sarveazad PhD), Department of Health Policy (H Shabaninejad PhD), Department of Health Services Management (M Alikhani PhD, A Ghashghaee BSc, S Raoofi MSc), Health Economics Department (V Alipour PhD), Health Management and Economics Research Center (V Alipour PhD, J Arabloo PhD, A Rezapour PhD, T Zahirian Moghadam PhD), Ophthalmology Department (N Manafi MD), Iran University of Medical Sciences, Tehran, Iran; Department of Health Policy and Management (Prof S M Aljunid PhD), Kuwait University, Safat, Kuwait City, Kuwait; International Centre for Casemix and Clinical Coding (Prof S M Aljunid PhD), National University of Malaysia, Bandar Tun Razak, Malaysia; College of Medicine (M A H Almadi FRCPC), King Saud University, Riyadh, Central, Saudi Arabia; Division of Gastroenterol

Aljunid PhD), Kuwait University, Safat, Kuwait City, Kuwait; International Centre for Casemix and Clinical Coding (Prof S M Aljunid PhD), National University of Malaysia, Bandar Tun Razak, Malaysia; College of Medicine (M A H Almadi FRCPC), King Saud University, Riyadh, Central, Saudi Arabia; Division of Gastroenterol ogy & Hepatology (M A H Almadi FRCPC), McGill University, Montreal, QC, Canada; Department of Epidemiology (A Almasi-Hashiani PhD, R Moradzadeh PhD), Department of Pediatrics (J Nazari PhD), Health Services Management Department (S Amini PhD), Arak University of Medical Sciences, Arak, Iran; Department of Family and Community Medicine (Prof R M Al-Raddadi PhD), King Abdulaziz University, Jeddah, Saudi Arabia; Research Group in Health Economics (Prof N Alvis-Guzman PhD), University of Cartagena, Cartagena, Colombia; Research Group in Hospital Management and Health Policies (Prof N Alvis-Guzman PhD), University of the Coast, Barranquilla, Colombia; Faculty of Medicine (N H Anber PhD), Mansoura University, Mansoura, Egypt (N H Anber PhD); Department of Epidemiology and Biostatistics (Prof A Ansari-Moghaddam PhD), Health Promotion Research Center, Zahedan, Iran; Department of Biostatistics and Epidemiology (Prof M Asghari Jafarabadi PhD), Zanjan University of Medical Sciences, Zanjan, Iran; Cellular and Molecular Biology Research Center (A Azadmehr PhD), Clinical Biochemistry (S Mir MSc), Faculty of Medicine (M Jahani PhD), Social Determinants of Health Research Center (A Bijani PhD), Babol University of Medical Sciences, Babol, Iran; Public Health Risk Sciences Division (A Badawi PhD), Public Health Agency of Canada, Toronto, ON, Canada; Department of Nutritional Sciences (A Badawi PhD), University of Toronto, Toronto, ON, Canada; Department of Clinical Biochemistry (M Mohammadoo-Khorasani PhD), Department of Parasitology and Entomology (M Pirestani PhD, L Zaki PhD), Faculty of Medical Sciences, Tarbiat Modares University, Tehran, Iran (N Baheiraei PhD); Heidelberg Institute of Global Health (HIGH) (Prof T W Bärnighausen MD), Institute of Public Health (S Mohammed PhD), Heidelberg University, Heidelberg, Germany; Department of Medicine Brigham and Women's Hospital (V Kumar MD), T.H.

aculty of Medical Sciences, Tarbiat Modares University, Tehran, Iran (N Baheiraei PhD); Heidelberg Institute of Global Health (HIGH) (Prof T W Bärnighausen MD), Institute of Public Health (S Mohammed PhD), Heidelberg University, Heidelberg, Germany; Department of Medicine Brigham and Women's Hospital (V Kumar MD), T.H. Chan School of Public Health (Prof T W Bärnighausen MD), Harvard University, Boston, MA, USA; University of Aden, Aden, Yemen (H Basaleem PhD); Department of Physiology (M Khaksarian PhD), Hepatitis Research Center (Me Behzadifar MS), Razi Herbal Medicines Research Center (Z Sharafi PhD), Social Determinants of Health Research Center (Ma Behzadifar PhD), Lorestan University of Medical Sciences, Khorramabad, Iran; Department of Pharmacy (Y M Belayneh MSc), Wollo University, Dessie, Ethiopia; Clinical Pharmacy Unit (T D Kassa MSc), Department of Nutrition and Dietetics (K Berhe MPH), Mekelle University, Mekelle, Ethiopia (H G Meles MPH); Department of Statistical and Computational Genomics (K Bhattacharyya MSc), National Institute of Biomedical Genomics, Kalyani, West Bengal, India; Department of Statistics (K Bhattacharyya MSc), University of Calcutta, Kolkata, India; Department of Clinical Chemistry (B Biadgo MSc), University of Gondar, Gondar, Ethiopia; Department of Clinical and Molecular Biomedicine (MEDBIO) (A M Borzì MD), Department of General Surgery and Medical-Surgical Specialties (Prof A Biondi PhD, M Vacante PhD, A Valli PhD), University of Catania, Catania, Italy; Department of Clinical Medicine (Prof K Soreide PhD), Department of Global Public Health and Primary Care (Prof T Bjørge PhD), University of Bergen, Bergen, Norway; Cancer Registry of Norway, Oslo, Norway (Prof T Bjørge PhD); Department of Environmental Health Science (S Gallus DSc), Department of Oncology (C Bosetti PhD), Mario Negri Institute for Pharmacological Research, Milan, Italy; Ministry of Public Health, Beirut, Lebanon (I R Bou-Orm MD); Division of Clinical Epidemiology and Aging Research (Prof H Brenner MD), German Cancer Research Center, Heidelberg, Germany; Biomedical Technologies (A N Briko MSc), Bauman Moscow State Technical University, Moscow, Russia; Department of Epidemiology and Evidence-Based Medicine (P D Lopukhov Cand of Sci [Med]), Epidemiology and Evidence Based Medicine (Prof N I Briko DSc), I.M.

Brenner MD), German Cancer Research Center, Heidelberg, Germany; Biomedical Technologies (A N Briko MSc), Bauman Moscow State Technical University, Moscow, Russia; Department of Epidemiology and Evidence-Based Medicine (P D Lopukhov Cand of Sci [Med]), Epidemiology and Evidence Based Medicine (Prof N I Briko DSc), I.M. Sechenov First Moscow State Medical University, Moscow, Russia; Institute for Cancer Research, Prevention and Clinical Network, Florence, Italy (G Carreras PhD); Applied Molecular Biosciences Unit (Prof F Carvalho PhD), Institute of Public Health (Prof F Carvalho PhD), REQUIMTE/LAQV (Prof E Fernandes PhD, Prof D M Pereira PhD), University of Porto, Porto, Portugal; Colombian National Health Observatory (C A Castañeda-Orjuela MD), National Institute of Health, Bogota, Colombia; Epidemiology and Public Health Evaluation Group (C A Castañeda-Orjuela MD), National University of Colombia, Bogota, Colombia; Mary MacKillop Institute for Health Research (Prof E Cerin PhD), Australian Catholic University, Melbourne, VIC, Australia; School of Public Health (Prof E Cerin PhD), University of Hong Kong, Hong Kong, China; Clinical Governance (P P Chiang PhD), Gold Coast Health, Gold Coast, QLD, Australia; Department of Epidemiology, Human Genetics, and Environmental Sciences (O G Chido-Amajuoyi MD), University of Texas, Houston, Texas, USA; Department of Public Health (M Khazaee-Pool PhD), Toxoplasmosis Research Center (Prof A Daryani PhD), Mazandaran University of Medical Sciences, Sari, Mazandaran, Iran; Department of General Surgery (D V Davitoiu PhD, M Hostiuc PhD), Emergency Hospital of Bucharest (I Negoi PhD), General Surgery Department (I Negoi PhD), Surgery 2nd Department-SUUB (C Smarandache MD), Carol Davila University of Medicine and Pharmacy, Bucharest, 2nd sector, Romania; Department of Surgery (D V Davitoiu PhD), Clinical Emergency Hospital Sf.

vitoiu PhD, M Hostiuc PhD), Emergency Hospital of Bucharest (I Negoi PhD), General Surgery Department (I Negoi PhD), Surgery 2nd Department-SUUB (C Smarandache MD), Carol Davila University of Medicine and Pharmacy, Bucharest, 2nd sector, Romania; Department of Surgery (D V Davitoiu PhD), Clinical Emergency Hospital Sf. Pantelimon, Bucharest, Romania; School of Pharmacy (G T Demoz MPharm), Aksum University, Aksum, Ethiopia; Division of Cardiology (R Desai MBBS), Atlanta Veterans Affairs Medical Center, Decatur, Georgia, USA; Department of Microbiology (A Hasanzadeh PhD), Pharmacology and Toxicology Department (A Eftekhari PhD), Maragheh University of Medical Sciences, Maragheh, Iran (F Esmaeilzadeh PhD); Biomedical Informatics and Medical Statistics (I El Sayed PhD), Alexandria University, Alexandria, Egypt; Institute of Public Health (I Elbarazi DrPH), United Arab Emirates University, Al Ain, United Arab Emirates; Ophthalmic Epidemiology Research Center (M Emamian PhD), Shahroud University of Medical Sciences, Shahroud, Semnan, Iran; Public Health Department (A Y Endries MPH), Saint Paul's Hospital Millennium Medical College, Addis Ababa, Ethiopia; Division of Cancer Epidemiology and Genetics (A Etemadi PhD), National Cancer Institute, Bethesda, MD, USA; Center for Biotechnology and Fine Chemistry (J C Fernandes PhD), Catholic University of Portugal, Porto, Portugal; Psychiatry Department (I Filip MD), Kaiser Permanente, Fontana, CA, USA; College of Graduate Health Sciences (A Radfar MD), Department of Health Sciences (I Filip MD), A.T.

er Institute, Bethesda, MD, USA; Center for Biotechnology and Fine Chemistry (J C Fernandes PhD), Catholic University of Portugal, Porto, Portugal; Psychiatry Department (I Filip MD), Kaiser Permanente, Fontana, CA, USA; College of Graduate Health Sciences (A Radfar MD), Department of Health Sciences (I Filip MD), A.T. Still University, Mesa, AZ, USA; Department of Public Health Medicine (F Fischer PhD), Bielefeld University, Bielefeld, Germany; Abadan School of Medical Sciences, Abadan, Iran (M Foroutan PhD); Department of Cardiovascular Medicine (M M Gad MD), Cleveland Clinic, Cleveland, OH, USA; Gillings School of Global Public Health (M M Gad MD), University of North Carolina Chapel Hill, Chapel Hill, NC, USA; Occupational and Environmental Epidemiology Section (G Gorini MD), Cancer Prevention and Research Institute, Florence, Italy; Department of Radiology (N Hafezi-Nejad MD, A Haj-Mirzaian MD), Department of Radiology and Radiological Sciences (S Sheikhbahaei MD), Johns Hopkins University, Baltimore, MD, USA; Cancer Research Center (M Khayamzadeh MD), Obesity Research Center (A Haj-Mirzaian MD), Research Institute for Endocrine Sciences (S N Irvani MD), Shahid Beheshti University of Medical Sciences, Tehran, Iran; Gastrointestinal and Liver Disease Research Center (S Hassanipour PhD, F Joukar PhD, Prof F Mansour-Ghanaei PhD), Guilan University of Medical Sciences, Rasht, Guilan, Iran (S Hassanipour PhD); Center of Excellence in Behavioral Medicine (C L Hoang BMedSc, T H Nguyen BMedSc), Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam; Department of Internal Medicine (M Hostiuc PhD), Surgery 2nd Department (C Smarandache MD), Bucharest Emergency Hospital, Bucharest, Romania; Division of Information and Computing Technology (Prof M Househ PhD), Hamad Bin Khalifa University, Doha, Qatar; Qatar Foundation for Education, Science, and Community Development, Doha, Qatar (Prof M Househ PhD); Department of Community Medicine (O S Ilesanmi PhD), University of Ibadan, Ibadan, Nigeria; Department of Epidemiology (Prof M D Ilic PhD), Department of Global Health, Economics, and Policy (Prof M Jakovljevic PhD), University of Kragujevac, Kragujevac, Serbia; Department of Epidemiology and Biostatistics (K Innos PhD), National Institute for Health Development, Tallinn, Estonia; Surveillance and Health Services Research (F Islami PhD), American Cancer Society, Atlanta, GA, USA; Cochrane South Africa (A Jaca PhD), Non-communicable Diseases Research Unit (Prof A P Kengne PhD), Medical

t of Epidemiology and Biostatistics (K Innos PhD), National Institute for Health Development, Tallinn, Estonia; Surveillance and Health Services Research (F Islami PhD), American Cancer Society, Atlanta, GA, USA; Cochrane South Africa (A Jaca PhD), Non-communicable Diseases Research Unit (Prof A P Kengne PhD), Medical Research Council South Africa, Cape Town, Western Cape, South Africa; Centre for Evidence Based Health Care (A Jaca PhD), Stellenbosch University, Cape Town, Western Cape, South Africa; Department of Psychosis (N Jafari Balalami PhD), Babol Nushirvani University of Technology, Babol, Iran; Clinical Medicine and Community Health, University of Milan, Milano, Italy (Prof C La Vecchia MD); Department of Immunology, School of Medicine, Isfahan University of Medical Sciences, Isfahan, Iran (M Jafarnia MSc); Newcastle University, Tyne, UK (M Javanbakht PhD); Autism Spectrum Disorders Research Center (E Jenabi PhD), Chronic Diseases (Home Care) Research Center (M Shamsizadeh MSc), Department of Biostatistics (N Mohammad Gholi Mezerji MSc), Department of Epidemiology (S Khazaei PhD), Neurophysiology Research Center (H Komaki MD), Research Center for Molecular Medicine (A Taherkhani PhD), Hamadan University of Medical Sciences, Hamadan, Iran; Department of Community Medicine (R P Jha MSc), Banaras Hindu University, Varanasi, Uttar Pradesh, India; Department of Nursing (M W Kassaw MSc), Woldia University, Woldia, Amhara, Ethiopia; Amhara Public Health Institute, Bair Dar, Amhara, Ethiopia (M W Kassaw MSc); Department of Medicine (Prof A P Kengne PhD), University of Cape Town, Cape Town, Western Cape, South Africa; Department of Public Health and Community Medicine (Prof Y S Khader PhD), Jordan University of Science and Technology, Ramtha, Irbid, Jordan; Department of Physiology (R Khalilov PhD), Baku State University, Baku, Azerbaijan; Epidemiology and Biostatistics Department (E A Khan MPH), Health Services Academy, Islamabad, Islamabad Capital Territory, Pakistan; Department of Community and Family Medicine (F H Lami PhD), Academy of Medical Science, Tehran, Iran (M Khayamzadeh MD); Department of Epidemiology & Biostatistics (F Khosravi Shadmani PhD, Prof F Najafi PhD), Department of Health Education & Promotion (F Rajati PhD), Radiology and Nuclear Medicine (S Salehi Zahabi PhD), Social Development and Health Promotion Research Center (M Soofi PhD), Taleghani Hospital (F Salehi MA), Kermanshah University of Medical Sciences, Kermanshah, Iran; Department of Nutrition an

F Najafi PhD), Department of Health Education & Promotion (F Rajati PhD), Radiology and Nuclear Medicine (S Salehi Zahabi PhD), Social Development and Health Promotion Research Center (M Soofi PhD), Taleghani Hospital (F Salehi MA), Kermanshah University of Medical Sciences, Kermanshah, Iran; Department of Nutrition an d Health Science (Prof J Khubchandani PhD), Ball State University, Muncie, IN, USA; Department of Health Sciences (Prof D Kim DrPH), Northeastern University, Boston, Massachusetts, USA; Department of Health Management and Health Economics (Prof A Kisa PhD), Kristiania University College, Oslo, Norway; Department of Health Services Policy and Management (Prof A Kisa PhD), University of South Carolina, Columbia, SC, USA; Department of Nursing and Health Promotion (S Kisa PhD), Oslo Metropolitan University, Oslo, Norway; Public Health Sciences Division (J M Kocarnik PhD), Fred Hutchinson Cancer Research Center, Seattle, WA, USA; Brain Engineering Research Center (H Komaki MD), Institute for Research in Fundamental Sciences, Tehran, Iran; University of British Columbia, Vancouver, BC, Canada (J A Kopec PhD); Arthritis Research Canada, Richmond, BC, Canada (J A Kopec PhD); CIBERSAM (A Koyanagi MD), San Juan de Dios Sanitary Park, Sant Boi de Llobregat, Barcelona, Spain; Catalan Institution for Research and Advanced Studies (ICREA), Barcelona, Spain (A Koyanagi MD); Department of Gastroenterology and Hepatology (Prof E J Kuipers MD), Erasmus University Medical Center, Rotterdam, Netherlands; University of Melbourne, Melbourne, QLD, Australia (Prof A D Lopez PhD); General Surgery Department (R Lunevicius PhD), Aintree University Hospital National Health Service (NHS) Foundation Trust, Liverpool, Merseyside, UK; Surgery Department (R Lunevicius PhD), University of Liverpool, Liverpool, Merseyside, UK; Department of Primary Care and Public Health (Prof A Majeed MD, Prof S Rawaf MD), WHO Collaborating Centre for Public Health Education and Training (D L Rawaf MD), Imperial College London, London, England, UK; Solid Tumor Research Center (M Majidinia PhD), Urmia University of Medical Science, Urmia, Iran; Department of Surgery (A Manafi MD), University of Virginia, Charlottesville, Virginia, USA; Ophthalmology Department (N Manafi MD), University of Manitoba, Winnipeg, Manitoba, Canada; Surgery Department (A Manda MD), Emergency University Hospital Bucharest, Bucharest, 5th sector, Romania; School of Medicine and Surgery (Prof L G Mantovani DSc), University of Milan Bicocc

ginia, Charlottesville, Virginia, USA; Ophthalmology Department (N Manafi MD), University of Manitoba, Winnipeg, Manitoba, Canada; Surgery Department (A Manda MD), Emergency University Hospital Bucharest, Bucharest, 5th sector, Romania; School of Medicine and Surgery (Prof L G Mantovani DSc), University of Milan Bicocc a, Monza, MB, Italy; Division of Gastroenterology and Hepatobiliary Disease (D Mehta MD), New York Medical College, Valhalla, New York, USA; Institute for Agricultural and Nutritional Sciences (T Meier PhD), Martin Luther University Halle-Wittenberg, Halle, Sachsen-Anhalt, Germany; Innovation Office (T Meier PhD), Competence Cluster for Nutrition and Cardiovascular Health (nutriCARD), Halle, Sachsen-Anhalt, Germany; Peru Country Office (W Mendoza MD), United Nations Population Fund (UNFPA), Lima, Lima, Peru; Clinical Microbiology and Parasitology Unit (T Mestrovic PhD), Dr.

chsen-Anhalt, Germany; Innovation Office (T Meier PhD), Competence Cluster for Nutrition and Cardiovascular Health (nutriCARD), Halle, Sachsen-Anhalt, Germany; Peru Country Office (W Mendoza MD), United Nations Population Fund (UNFPA), Lima, Lima, Peru; Clinical Microbiology and Parasitology Unit (T Mestrovic PhD), Dr. Zora Profozic Polyclinic, Zagreb, Croatia; University Centre Varazdin (T Mestrovic PhD), University North, Varazdin, Croatia; Center for Innovation in Medical Education (B Miazgowski MD), Department of Propedeutics of Internal Diseases & Arterial Hypertension (Prof T Miazgowski MD), Pomeranian Medical University, Szczecin, Zachodniopomorskie, Poland (B Miazgowski MD); Research Center for Biochemistry and Nutrition in Metabolic Diseases (H Mirzaei PhD), Kashan University of Medical Sciences, Kashan, Isfahan, Iran; Department of Biology (K A Mohammad PhD), Salahaddin University, Erbil, Iraq; ISHIK University, Erbil, Iraq (K A Mohammad PhD); Department of Epidemiology and Biostatistics (A Mohammadian-Hafshejani PhD), Shahrekord University of Medical Sciences, Shahrekord, Iran; Health Systems and Policy Research Unit (S Mohammed PhD), Ahmadu Bello University, Zaria, Nigeria; Clinical Epidemiology and Public Health Research Unit (L Monasta DSc, E Traini MSc), Burlo Garofolo Institute for Maternal and Child Health, Trieste, Italy; Department of Molecular Medicine (M Moossavi PhD), Birjand University of Medical Sciences, Birjand, Iran; Department of Epidemiology and Biostatistics (G Moradi PhD), Social Determinants of Health Research Center (G Moradi PhD, F Moradpour PhD), Kurdistan University of Medical Sciences, Sanandaj, Kurdistan, Iran; Department of Epidemiology (G Naik MPH), University of Alabama at Birmingham, Birmingham, AL, USA; Iranian Ministry of Health and Medical Education, Tehran, Iran (J Nazari PhD); Department of Oncology (S Negru MD), University of Medicine and Pharmacy, Timisoara, Romania; Institute for Global Health Innovations (C T Nguyen MPH), Duy Tan University, Hanoi, Vietnam; Public Health Science Department (D N A Ningrum MPH), State University of Semarang, Kota Semarang, Indonesia; Graduate Institute of Biomedical Informatics (D N A Ningrum MPH), Taipei Medical University, Taipei City, Taiwan; School of Social Sciences and Psychology (Prof A M N Renzaho PhD), Translational Health Research Institute (F A Ogbo PhD), Western Sydney University, Penrith, NSW, Australia; Department of Pathology and Molecular Medicine (T O Olagunju MD), Department of

N A Ningrum MPH), Taipei Medical University, Taipei City, Taiwan; School of Social Sciences and Psychology (Prof A M N Renzaho PhD), Translational Health Research Institute (F A Ogbo PhD), Western Sydney University, Penrith, NSW, Australia; Department of Pathology and Molecular Medicine (T O Olagunju MD), Department of Psychiatry and Behavioural Neurosciences (A T Olagunju MD), McMaster University, Hamilton, ON, Canada; Department of Psychiatry (A T Olagunju MD), University of Lagos, Lagos, Nigeria; Department of Statistics and Econometrics (A Pana MD), Bucharest University of Economic Studies, Bucharest, Romania; Center for Health Outcomes & Evaluation, Bucharest, Romania (A Pana MD); Cartagena University, Cartagena, Colombia (Prof D M Pereira PhD); Department of Health Promotion and Education (F Shaahmadi PhD), Non-communicable Diseases Research Center (M Qorbani PhD), Alborz University of Medical Sciences, Karaj, Alborz, Iran; Biomedical Engineering Department (Prof M Rabiee PhD), Amirkabir University of Technology, Tehran, Iran; Department of Chemistry (N Rabiee PhD), Sharif University of Technology, Tehran, Iran; Medichem, Barcelona, Spain (A Radfar MD); University College London Hospitals, London, UK (D L Rawaf MD); Academic Public Health (Prof S Rawaf MD), Public Health England, London, UK; Network of Immunity in Infection, Malignancy, and Autoimmunity (NIIMA) (Prof N Rezaei PhD), Universal Scientific Education and Research Network (USERN), Tehran, Iran; Department of Entomology (A M Samy PhD), Faculty of Medicine (A M Saad MBBCh), Ain Shams University, Cairo, Egypt; Medical Department (B Saddik PhD), University of Sharjah, Sharjah, United Arab Emirates; Research Deputy (S Salehi Zahabi PhD), Taleghani Hospital, Kermanshah, Iran; Health and Disability Intelligence Group (I Salz MD), Ministry of Health, Wellington, New Zealand; Department of Surgery (Prof J Sanabria MD), Marshall University, Huntington, WV, USA; Department of Nutrition and Preventive Medicine (Prof J Sanabria MD), Case Western Reserve University, Cleveland, OH, USA; Centre School of Public Health and Health Management (Prof M M Santric Milicevic PhD), Faculty of Medicine Institute of Epidemiology (I S Vujcic PhD), University of Belgrade, Belgrade, Serbia; UGC Centre of Advanced Study in Psychology (M Satpathy PhD), Utkal University, Bhubaneswar, India; Udyam-Global Association for Sustainable Development, Bhubaneswar, Odisha, India (M Satpathy PhD); School of Health Sciences (Prof I J C Schneider PhD, P

miology (I S Vujcic PhD), University of Belgrade, Belgrade, Serbia; UGC Centre of Advanced Study in Psychology (M Satpathy PhD), Utkal University, Bhubaneswar, India; Udyam-Global Association for Sustainable Development, Bhubaneswar, Odisha, India (M Satpathy PhD); School of Health Sciences (Prof I J C Schneider PhD, P rof D A S Silva PhD), Federal University of Santa Catarina, Ararangua, Brazil; Department of Medical Statistics, Epidemiology and Medical Informatics (M Sekerija PhD), University of Zagreb, Zagreb, Croatia; Division of Epidemiology and Prevention of Chronic Non-communicable Diseases (M Sekerija PhD), Croatian Institute of Public Health, Zagreb, Croatia; Department of Basic Sciences (Prof M Sharif PhD), Department of Laboratory Sciences (Prof M Sharif PhD), Islamic Azad University, Sari, Iran; Department of Hematology-Oncology (S K Siddappa Malleshappa MD), Baystate Medical Center, Springfield, MA, USA; School of Pharmacy (M Sisay MSc), Haramaya University, Harar, Ethiopia; Department of Gastrointestinal Surgery (Prof K Soreide PhD), Stavanger University Hospital, Stavanger, Rogaland, Norway; Research Development (S Soshnikov PhD), Central Research Institute of Cytology and Genetics (E Varavikova PhD), Federal Research Institute for Health Organization and Informatics of the Ministry of Health (FRIHOI), Moscow, Russia (Prof V I Starodubov DSc); Department of Social Sciences (Prof M Sullman PhD), University of Nicosia, Nicosia, Cyprus; Department of Medicine (Prof R Tabarés-Seisdedos PhD), University of Valencia, Valencia, Spain; Carlos III Health Institute (Prof R Tabarés-Seisdedos PhD), Biomedical Research Networking Center for Mental Health Network (CiberSAM), Madrid, Spain; Department of Public Health (B E Tesfay MPH), Adigrat University, Adigrat, Ethiopia; Faculty of Health Sciences (R Topor-Madry PhD), Jagiellonian University Medical College, Krakow, Poland; The Agency for Health Technology Assessment and Tariff System, Warsaw, Poland (R Topor-Madry PhD); Department of Health Economics (B X Tran PhD), Hanoi Medical University, Hanoi, Vietnam; Molecular Medicine and Pathology (K B Tran MD), University of Auckland, Auckland, New Zealand; Clinical Hematology and Toxicology (K B Tran MD), Military Medical University, Hanoi, Vietnam; Gomal Center of Biochemistry and Biotechnology (I Ullah PhD), Gomal University, Dera Ismail Khan, Pakistan; TB Culture Laboratory (I Ullah PhD), Mufti Mehmood Memorial Teaching Hospital, Dera Ismail Khan, Pakistan; Division of Health

Hematology and Toxicology (K B Tran MD), Military Medical University, Hanoi, Vietnam; Gomal Center of Biochemistry and Biotechnology (I Ullah PhD), Gomal University, Dera Ismail Khan, Pakistan; TB Culture Laboratory (I Ullah PhD), Mufti Mehmood Memorial Teaching Hospital, Dera Ismail Khan, Pakistan; Division of Health Sciences (O A Uthman PhD), University of Warwick, Coventry, UK; Baqiyatallah University of Medical Sciences, Tehran, Iran (A Vahedian-Azimi PhD); Competence Center of Mortality-Follow-Up, German National Cohort (R Westerman DSc), Federal Institute for Population Research, Wiesbaden, Hesse, Germany; Department of Health Management, Policy and Economics (V Yazdi-Feyzabadi PhD), Health Services Management Research Center (V Yazdi-Feyzabadi PhD), Kerman University of Medical Sciences, Kerman, Iran; Department of Epidemiology and Biostatistics (Prof C Yu PhD), Department of Preventive Medicine (Z Zhang PhD), Global Health Institute (Prof C Yu PhD), Wuhan University, Wuhan, Hubei Province, China; Epidemiology and Cancer Registry Sector (Prof V Zadnik PhD), Institute of Oncology Ljubljana, Ljubljana, Slovenia; and Department of Community Medicine (H Zandian PhD), Social Determinants of Health Research Center (T Zahirian Moghadam PhD, H Zandian PhD), Ardabil University of Medical Science, Ardabil, Iran.

e, China; Epidemiology and Cancer Registry Sector (Prof V Zadnik PhD), Institute of Oncology Ljubljana, Ljubljana, Slovenia; and Department of Community Medicine (H Zandian PhD), Social Determinants of Health Research Center (T Zahirian Moghadam PhD, H Zandian PhD), Ardabil University of Medical Science, Ardabil, Iran. Contributors SS, SGS, HS, GR, CF, MN, and CJLM prepared the first draft. RM, AD, RA, SM, CF, MN, and CJLM provided overall guidance. RM, SGS, SS, CF, MN, and CJLM managed the project. SGS, SS, HS, and CF analysed data. RM, SGS, SS, SM, HS, CF, MN, and CJLM finalised the manuscript on the basis of comments from other authors and reviewer feedback. All other authors provided data, developed models, reviewed results, provided guidance on methods, or reviewed and contributed to the manuscript. Declaration of interests SLJ reports grants from Sanofi Pasteur, outside the submitted work. All other authors declare no competing interests.

Introduction Inflammatory bowel disease (IBD) is characterised by non-infectious chronic inflammation of the gastrointestinal tract, and primarily includes Crohn's disease (which can affect any segment of the gastrointestinal tract from the mouth to the anus), ulcerative colitis (which is limited to the colonic mucosa), and indeterminate colitis.1 Crohn's disease and ulcerative colitis both commonly present with abdominal pain and diarrhoea. Rectal bleeding occurs more frequently with ulcerative colitis than with Crohn's disease, but patients with Crohn's disease often experience weight loss and perianal disease.2 The peak of IBD's occurrence, followed by a chronic relapsing pattern, usually happens in the second to fourth decade of life, the most productive time of adulthood.1 IBD can adversely affect all aspects of an individual's life.1 Although the cause of IBD is still not completely understood, IBD is suggested to be a result of uncontrolled immune response to a trigger in genetically prone individuals.1 The role of environmental factors either as the triggers or causes of the uncontrolled immune response continues to be debated.1 Research in context Evidence before this study

Introduction Inflammatory bowel disease (IBD) is characterised by non-infectious chronic inflammation of the gastrointestinal tract, and primarily includes Crohn's disease (which can affect any segment of the gastrointestinal tract from the mouth to the anus), ulcerative colitis (which is limited to the colonic mucosa), and indeterminate colitis.1 Crohn's disease and ulcerative colitis both commonly present with abdominal pain and diarrhoea. Rectal bleeding occurs more frequently with ulcerative colitis than with Crohn's disease, but patients with Crohn's disease often experience weight loss and perianal disease.2 The peak of IBD's occurrence, followed by a chronic relapsing pattern, usually happens in the second to fourth decade of life, the most productive time of adulthood.1 IBD can adversely affect all aspects of an individual's life.1 Although the cause of IBD is still not completely understood, IBD is suggested to be a result of uncontrolled immune response to a trigger in genetically prone individuals.1 The role of environmental factors either as the triggers or causes of the uncontrolled immune response continues to be debated.1 Research in context Evidence before this study Inflammatory bowel disease (IBD) imposes health and economic burdens on communities worldwide, and substantially reduces patients' quality of life. It is estimated that more than 3 million people in the USA and Europe have IBD, and its prevalence is estimated to exceed 0·3% in North America, Oceania, and many countries in Europe. Evidence from systematic reviews points to a changing epidemiology of IBD, with stable or decreasing incidence in North America and Europe, and increasing incidence in newly industrialised countries. Most of these studies have originated in high-income countries, and countries with lower socioeconomic status have produced few, if any, population-based studies that report prevalence, incidence, or IBD-related deaths. These studies also do not include estimation over time and usually do not include estimates for years of life lost (YLLs), years lived with disability (YLDs), or disability-adjusted life-years (DALYs) across either countries or geographically related regions.

es that report prevalence, incidence, or IBD-related deaths. These studies also do not include estimation over time and usually do not include estimates for years of life lost (YLLs), years lived with disability (YLDs), or disability-adjusted life-years (DALYs) across either countries or geographically related regions. Added value of this study The Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2017 used an integrated modelling approach to estimate not only the epidemiological parameters for regions with available data, but also for countries and territories, as well as regions in which sufficient data are not available. GBD 2017 provided estimates of the burden of IBD for seven super-regions, 21 regions, and 195 countries and territories from 1990 to 2017. There has been no previous dedicated and detailed publication of GBD methods and estimates for IBD. We report the burden of IBD, including prevalence, mortality, YLDs, and DALYs, by age, sex, and Socio-demographic Index (SDI) from 1990 to 2017, using all available data and based on standardised GBD methods at the global, regional, and national levels. In addition to exploring country-level variation in the burden of IBD by development level, we assessed the temporal patterns and changes in geographical patterns of the burden of IBD. We believe this analysis is the most comprehensive picture of IBD burden to date. Implication of all the available evidence

The Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2017 used an integrated modelling approach to estimate not only the epidemiological parameters for regions with available data, but also for countries and territories, as well as regions in which sufficient data are not available. GBD 2017 provided estimates of the burden of IBD for seven super-regions, 21 regions, and 195 countries and territories from 1990 to 2017. There has been no previous dedicated and detailed publication of GBD methods and estimates for IBD. We report the burden of IBD, including prevalence, mortality, YLDs, and DALYs, by age, sex, and Socio-demographic Index (SDI) from 1990 to 2017, using all available data and based on standardised GBD methods at the global, regional, and national levels. In addition to exploring country-level variation in the burden of IBD by development level, we assessed the temporal patterns and changes in geographical patterns of the burden of IBD. We believe this analysis is the most comprehensive picture of IBD burden to date. Implication of all the available evidence Between 1990 and 2017, the global number of prevalent cases of IBD increased. After many years of sharp increases in IBD incidence in North America and western European countries, and because of improved survival, a pattern of increased prevalence emerged in these regions. An alarming trend for health systems is the observed rise in prevalence in newly industrialised countries. The full effect of this rise might not yet have been fully appreciated, because IBD symptoms persist throughout life, producing prominent disability and morbidity. The information provided in this study could be crucial for researchers, clinicians, and health policy makers to prepare their clinical infrastructure and educate specialised personnel to be able to confront the burden of this complex, and socially and economically costly disease. Moreover, the rising pattern of IBD provides a unique opportunity for researchers to focus on identifying the environmental risk factors contributing to IBD.

e their clinical infrastructure and educate specialised personnel to be able to confront the burden of this complex, and socially and economically costly disease. Moreover, the rising pattern of IBD provides a unique opportunity for researchers to focus on identifying the environmental risk factors contributing to IBD. Traditionally, IBD has been regarded as a disease of high-income nations, but a shift in the epidemiological pattern has been reported, indicating stabilising incidence in high-income countries with a high burden and prevalence, and a rapid rise in newly industrialised countries in South America, eastern Europe, Asia, and Africa.3 These rapid changes call for global estimates to provide insight into the burden and trends. Moreover, defining the varying incidence, prevalence, and prognosis of IBD in different geographical regions might provide researchers with clues to the cause of the disease.4 We report the IBD burden in 195 countries and territories from 1990 to 2017, based on the most recent Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2017 estimates in terms of prevalence, mortality, years lived with disability (YLDs), years of life lost (YLLs), and disability-adjusted life-years (DALYs).

eport the IBD burden in 195 countries and territories from 1990 to 2017, based on the most recent Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2017 estimates in terms of prevalence, mortality, years lived with disability (YLDs), years of life lost (YLLs), and disability-adjusted life-years (DALYs). Methods Overview This study is part of GBD 2017, which, to the best of our knowledge, is the most comprehensive and systematic effort to estimate the burden of diseases, injuries, and risk factors at global, regional, and national levels to date. GBD 2017 estimated 359 diseases and injuries; 282 causes of death; and 84 behavioural, environmental and occupational, and metabolic risk factors. The detailed methods are published elsewhere.3, 5, 6 Crohn's disease and ulcerative colitis (ie, the two types of IBD we investigated) are diagnosed by endoscopy, imaging studies, or biopsy in a patient with relevant clinical signs and symptoms. In some cases of IBD, neither Crohn's disease nor ulcerative colitis can be definitely diagnosed, and a diagnosis of indeterminate colitis is applied indefinitely or until definitive features of Crohn's disease or ulcerative colitis are identifiable.7 International Classification of Disease version 10 (ICD-10) codes were K50 for Crohn's disease, K51 for ulcerative colitis, and K52 for indeterminate colitis.

gnosed, and a diagnosis of indeterminate colitis is applied indefinitely or until definitive features of Crohn's disease or ulcerative colitis are identifiable.7 International Classification of Disease version 10 (ICD-10) codes were K50 for Crohn's disease, K51 for ulcerative colitis, and K52 for indeterminate colitis. Mortality estimates To model IBD mortality, we used the causes of death database, which includes data from vital registration and verbal autopsy data. The data processing for the causes of death data has been described previously.5 We marked data as outliers if garbage code redistribution and noise reduction in combination with small sample sizes resulted in unreasonable cause fractions, as well as data that violated well-established time or age trends.

autopsy data. The data processing for the causes of death data has been described previously.5 We marked data as outliers if garbage code redistribution and noise reduction in combination with small sample sizes resulted in unreasonable cause fractions, as well as data that violated well-established time or age trends. We modelled deaths due to IBD with a standard Cause of Death Ensemble model (CODEm), using the causes of death database and location-level covariates as inputs. We hybridised separate global and data-rich models for each sex to obtain unadjusted results. We then finalised and adjusted estimates to be consistent with all-cause mortality levels for each age–sex–year location using the cause of death correct procedure (CODCorrect) to reach final YLLs due to IBD.5 The method for propagating uncertainty was similar to that used in previous GBD papers.5 The distribution of every step in the computation process was stored in 1000 draws that were used for every other step in the process. Final estimates were computed using the mean estimate across 1000 draws, and the 95% uncertainty intervals (UIs) were specified on the basis of the 25th and 975th ranked values across all 1000 draws. The percentage change between any 2 years of estimates in GBD was calculated at the draw level. Every one of 1000 draws for 2017 was compared with the corresponding draw for 1990 to generate 1000 percent change draws. The mean of the draws and the 25th and 975th ordered draws were then used as the mean, lower UI limit, and upper UI limit for reporting the percentage change.

n GBD was calculated at the draw level. Every one of 1000 draws for 2017 was compared with the corresponding draw for 1990 to generate 1000 percent change draws. The mean of the draws and the 25th and 975th ordered draws were then used as the mean, lower UI limit, and upper UI limit for reporting the percentage change. Non-fatal estimates To estimate the non-fatal burden of IBD, we used two separate databases, one for Crohn's disease and another for ulcerative colitis. Both included data from literature, hospital discharges, and claims data (the latter available only from the USA in 2000, 2010, and 2012; further information on IBD data is provided in the appendix, p 21). Claims data link multiple inpatient and outpatient claims to a single individual; prevalent cases were extracted if an individual had at least one inpatient or outpatient encounter with an appropriate ICD code as any diagnosis. Data from hospital discharges were adjusted using correction factors from claims, converting encounters to estimates of cases, correcting for some facilities providing only primary diagnostic codes, and estimating outpatient cases from inpatient cases. Literature data came from a systematic review done for GBD 2016.8 In brief, this systematic review of literature was done to capture studies of the prevalence and incidence of IBD. Studies were excluded if they were not representative of the national population, or if they had insufficient or inappropriate sampling methods. Reviews were excluded from the search results.

or GBD 2016.8 In brief, this systematic review of literature was done to capture studies of the prevalence and incidence of IBD. Studies were excluded if they were not representative of the national population, or if they had insufficient or inappropriate sampling methods. Reviews were excluded from the search results. The prevalence and incidence data described earlier were entered into separate models for ulcerative colitis and Crohn's disease in DisMod-MR 2.1, a Bayesian meta-regression tool, as the main method of estimation, ensuring consistency between rates of prevalence, incidence, remission, and cause of death for each non-fatal condition. Outputs from DisMod were then adjusted to account for IBD due to indeterminate colitis.

olitis and Crohn's disease in DisMod-MR 2.1, a Bayesian meta-regression tool, as the main method of estimation, ensuring consistency between rates of prevalence, incidence, remission, and cause of death for each non-fatal condition. Outputs from DisMod were then adjusted to account for IBD due to indeterminate colitis. For both ulcerative colitis and Crohn's disease, the DisMod models used prevalence and incidence data, as described above. Reference data were claims from the USA from 2012, and study-level covariates were used to mark data from literature, hospital discharges, US claims in 2000 and 2010, and the Medical Expenditure Panel Survey. The study-level covariate for hospital discharges for Crohn's disease was found to have no significant effect and was later dropped during data analysis. For ulcerative colitis, a prior value on remission was set to zero for all age groups, and an incidence prior value was set to zero only for ages zero to 1 year. For Crohn's disease, a prior value on remission was set to zero for all age groups, and on incidence a prior value was set to zero only for ages zero to 2 years. Location-level covariates were log-transformed lag-distributed income and Healthcare Access and Quality Index (both on excess mortality) and log-transformed age-standardised death rate due to both types of IBD (on prevalence).

all age groups, and on incidence a prior value was set to zero only for ages zero to 2 years. Location-level covariates were log-transformed lag-distributed income and Healthcare Access and Quality Index (both on excess mortality) and log-transformed age-standardised death rate due to both types of IBD (on prevalence). The proportion of IBD cases categorised as indeterminate colitis was determined to be 0·624 (95% UI 0·0549–0·0699) via meta-analysis. To account for all IBD cases, an adjustment of 1·0624 (1·0549–1·0699) was applied to the outputs of our ulcerative colitis and Crohn's disease models. This approach assumes that all cases initially diagnosed as indeterminant ultimately declare themselves to be one of these two defined diseases. Estimates of prevalence were combined with disability weights to estimate YLDs. The basis of the GBD disability weight survey assessments are lay descriptions of sequelae highlighting major functional consequences and symptoms. For GBD 2017, we used the Medical Expenditure Panel Survey to find the proportion of ulcerative colitis and Crohn's disease that was asymptomatic versus symptomatic during a given 4-week period. The lay description for either of these diseases in case they were symptomatic was defined as a person who has cramping abdominal pain, has diarrhoea several times a day, and feels very tired for 2 months every year, and when the person does not have symptoms, there is anxiety about them returning. The disability weight for symptomatic Crohn's disease and ulcerative colitis was 0·231 (95% UI 0·156–0·320).

ed as a person who has cramping abdominal pain, has diarrhoea several times a day, and feels very tired for 2 months every year, and when the person does not have symptoms, there is anxiety about them returning. The disability weight for symptomatic Crohn's disease and ulcerative colitis was 0·231 (95% UI 0·156–0·320). We estimated the burden of IBD in terms of mortality, prevalence, YLLs, YLDs, and DALYs, which are the sum of YLLs and YLDs, for both sexes, 20 age groups, and 195 countries and territories from 1990 to 2017. The rates were age-standardised according to the GBD world population and are reported per 100 000 population.9 95% UIs were reported for all estimates. This study is compliant with the Guidelines for Accurate and Transparent Health Estimates Reporting. Role of the funding source The funder of the study had no role in study design; the collection, analysis, or interpretation of the data; or the writing of the report. The corresponding author had full access to all of the data and the final responsibility to submit for publication.

We estimated the burden of IBD in terms of mortality, prevalence, YLLs, YLDs, and DALYs, which are the sum of YLLs and YLDs, for both sexes, 20 age groups, and 195 countries and territories from 1990 to 2017. The rates were age-standardised according to the GBD world population and are reported per 100 000 population.9 95% UIs were reported for all estimates. This study is compliant with the Guidelines for Accurate and Transparent Health Estimates Reporting. Role of the funding source The funder of the study had no role in study design; the collection, analysis, or interpretation of the data; or the writing of the report. The corresponding author had full access to all of the data and the final responsibility to submit for publication. Results Between 1990 and 2017, the number of individuals with IBD increased from 3·7 million (95% UI 3·5–3·9) to more than 6·8 million (6·4–7·3), an increase of 85·1% (79·5–89·9) in global prevalent cases of IBD (appendix p 22). However, the global age-standardised prevalence rate of IBD showed only a 6·1% (3·3–8·6) increase, from 79·5 (75·9–83·5) per 100 000 population in 1990 to 84·3 (79·2–89·9) per 100 000 population in 2017. The global map of age-standardised prevalence rate of IBD and percentage change in age-standardised prevalence at the country level are presented in figure 1. Both the number of prevalent cases and age-standardised prevalence rate were significantly higher in females than males in all years from 1990 to 2017 (figure 2). Overall, nearly 3·9 million (3·6–4·1) prevalent cases (57%) occurred among females in 2017, and nearly 3·0 million (2·8–3·2) cases (43%) occurred in males. The age-standardised prevalence rate was 75·0 (70·3–79·7) per 100 000 population in males and 93·8 (87·8–100·0) per 100 000 population in females in 2017. The highest peak of IBD age-specific prevalence rate occurred at age 60–64 years in females, whereas the peak was at age 70–74 years in males (figure 3).Figure 1 (A) Age-standardised prevalence rate (per 100 000 population) of IBD, both sexes, for 195 countries and territories, 2017. (B) Percentage change in age-standardised prevalence rate (per 100 000 population) of IBD, both sexes, for 195 countries and territories, 1990–2017

s at age 70–74 years in males (figure 3).Figure 1 (A) Age-standardised prevalence rate (per 100 000 population) of IBD, both sexes, for 195 countries and territories, 2017. (B) Percentage change in age-standardised prevalence rate (per 100 000 population) of IBD, both sexes, for 195 countries and territories, 1990–2017 IBD=inflammatory bowel disease. ATG=Antigua and Barbuda. VCT=Saint Vincent and the Grenadines. LCA=Saint Lucia. TTO=Trinidad and Tobago. TLS=Timor-Leste. FSM=Federated States of Micronesia. Figure 2 Trends from 1990 to 2017 in number and age-standardised prevalence rates of IBD at the global level Error bars indicate the 95% uncertainty interval (UI) for prevalent cases. Shading indicates the 95% UI for the age-standardised prevelance rate. IBD=inflammatory bowel disease. Figure 3 Age patterns by sex in 2017 of the total number of prevalent cases and age-specific prevalence rate of IBD at the global level IBD=inflammatory bowel disease.

Error bars indicate the 95% uncertainty interval (UI) for prevalent cases. Shading indicates the 95% UI for the age-standardised prevelance rate. IBD=inflammatory bowel disease. Figure 3 Age patterns by sex in 2017 of the total number of prevalent cases and age-specific prevalence rate of IBD at the global level IBD=inflammatory bowel disease. The total number of IBD-related deaths increased by 67·0% (95% UI 23·6–96·1) from 1990 to 2017, from 23 000 (20 000–27 000) to 38 000 (32 000–41 000; figure 4; appendix p 30). Despite this rise, the global age-standardised death rate decreased from 0·61 (0·55–0·69) per 100 000 population in 1990 to 0·51 (0·42–0·54) per 100 000 population in 2017, a rate that corresponded with a 16·4% (36·0–4·7) decrease in age-standardised death rate over the study period (appendix p 30). The total number of deaths caused by IBD constituted 0·07% (0·06–0·07) of total all-cause deaths in 2017 (estimates available through the GBD results tool). In 2017, the number of deaths from IBD was highest in females aged 85–89 years and males aged 80–84 years, whereas the age-specific rate of death was highest in the group aged 95 years and older for both sexes (appendix p 3).Figure 4 Trends from 1990 to 2017 in number and age-standardised death rates of IBD at the global level Error bars indicate the 95% uncertainty interval (UI) for number of deaths. Shading indicates the 95% UI for the age-standardised death rate. IBD=inflammatory bowel disease.

The total number of IBD-related deaths increased by 67·0% (95% UI 23·6–96·1) from 1990 to 2017, from 23 000 (20 000–27 000) to 38 000 (32 000–41 000; figure 4; appendix p 30). Despite this rise, the global age-standardised death rate decreased from 0·61 (0·55–0·69) per 100 000 population in 1990 to 0·51 (0·42–0·54) per 100 000 population in 2017, a rate that corresponded with a 16·4% (36·0–4·7) decrease in age-standardised death rate over the study period (appendix p 30). The total number of deaths caused by IBD constituted 0·07% (0·06–0·07) of total all-cause deaths in 2017 (estimates available through the GBD results tool). In 2017, the number of deaths from IBD was highest in females aged 85–89 years and males aged 80–84 years, whereas the age-specific rate of death was highest in the group aged 95 years and older for both sexes (appendix p 3).Figure 4 Trends from 1990 to 2017 in number and age-standardised death rates of IBD at the global level Error bars indicate the 95% uncertainty interval (UI) for number of deaths. Shading indicates the 95% UI for the age-standardised death rate. IBD=inflammatory bowel disease. The total YLDs attributed to IBD almost doubled over the study period, from 0·56 million (95% UI 0·39–0·77) in 1990 to 1·02 million (0·71–1·38) in 2017 (estimates available through the GBD results tool). However, the age-standardised YLDs rate did not have the same sharp increase (12·0 [8·4–16·5] per 100 000 population in 1990 to 12·6 [8·7–17·0] per 100 000 population in 2017). The number of YLDs in 2017 peaked in the 50–54 years age group (0·12 million [0·08–0·16]; appendix p 4) and then declined in the older age groups. Among the diseases in the GBD digestive disease category, IBD rose in YLD rank (both number of YLDs and age-standardised rate) from fifth in 1990 to fourth in 2017, after upper digestive diseases, hernia, and cirrhosis (estimates available through the online data visualisation tool).

n declined in the older age groups. Among the diseases in the GBD digestive disease category, IBD rose in YLD rank (both number of YLDs and age-standardised rate) from fifth in 1990 to fourth in 2017, after upper digestive diseases, hernia, and cirrhosis (estimates available through the online data visualisation tool). The total YLLs attributed to IBD in 1990 was 0·68 million (95% UI 0·54–0·96), which increased to 0·83 million (0·71–0·90) in 2017. The age-standardised rates of YLLs declined over time for both sexes, from 14·5 YLLs (11·8–19·0) per 100 000 population in 1990 to 10·7 YLLs (9·1–11·7) per 100 000 population in 2017, for both sexes combined. In 2017, the highest number of YLLs occurred in the group aged 65–69 years (72 000 [60 000–77 000]), while the highest number of YLDs occurred in the 50–54 year age group (119 000 [80 000–163 000]; figure 5). The 1–4 year age group had the third highest number of YLLs, because each death at a young age results in more YLLs than at older ages, but had the second lowest number of YLDs (figure 5).Figure 5 Global counts and age-specific rates of YLLs and YLDs due to IBD across age groups, 2017 Error bars indicate the 95% uncertainty interval (UI) for YLLs and YLDs. Shading indicates the 95% UI for the rates. YLLs=years of life lost. YLDs=years of life lived with disability. IBD=inflammatory bowel disease.

The total YLLs attributed to IBD in 1990 was 0·68 million (95% UI 0·54–0·96), which increased to 0·83 million (0·71–0·90) in 2017. The age-standardised rates of YLLs declined over time for both sexes, from 14·5 YLLs (11·8–19·0) per 100 000 population in 1990 to 10·7 YLLs (9·1–11·7) per 100 000 population in 2017, for both sexes combined. In 2017, the highest number of YLLs occurred in the group aged 65–69 years (72 000 [60 000–77 000]), while the highest number of YLDs occurred in the 50–54 year age group (119 000 [80 000–163 000]; figure 5). The 1–4 year age group had the third highest number of YLLs, because each death at a young age results in more YLLs than at older ages, but had the second lowest number of YLDs (figure 5).Figure 5 Global counts and age-specific rates of YLLs and YLDs due to IBD across age groups, 2017 Error bars indicate the 95% uncertainty interval (UI) for YLLs and YLDs. Shading indicates the 95% UI for the rates. YLLs=years of life lost. YLDs=years of life lived with disability. IBD=inflammatory bowel disease. The age-standardised rate of DALYs decreased from 26·5 (95% UI 21·0–33·0) per 100 000 population in 1990 to 23·2 (19·1–27·8) per 100 000 population in 2017 (appendix p 38). The total DALYs caused by IBD increased between 1990 (1·25 million [0·97–1·61]) and 2017 (1·85 million [1·51–2·23]; appendix p 38). Of the total DALYs caused by IBD in 2017, 45% were due to YLLs and 55% were due to YLDs.

per 100 000 population in 1990 to 23·2 (19·1–27·8) per 100 000 population in 2017 (appendix p 38). The total DALYs caused by IBD increased between 1990 (1·25 million [0·97–1·61]) and 2017 (1·85 million [1·51–2·23]; appendix p 38). Of the total DALYs caused by IBD in 2017, 45% were due to YLLs and 55% were due to YLDs. In 2017, the highest age-standardised prevalence rate among all seven super-regions was observed in the high-income super-region (206·1 [95% UI 195·3–216·8] per 100 000 population; appendix p 8). Moreover, this region had the highest increase in the age-standardised prevalence rate from 1990 to 2017 (31·3% [26·4–36·6]; percentage changes for other super-regions available from the GBD results tool).

ns was observed in the high-income super-region (206·1 [95% UI 195·3–216·8] per 100 000 population; appendix p 8). Moreover, this region had the highest increase in the age-standardised prevalence rate from 1990 to 2017 (31·3% [26·4–36·6]; percentage changes for other super-regions available from the GBD results tool). The high-income super-region also had the largest increase in age-standardised death rate from 1990 to 2017 (17·6% [95% UI −35·7 to 34·7]; figure 6). Sub-Saharan Africa was the only other super-region to experience an increase in age-standardised death rate over the study period, a 0·7% (−32·5 to 43·6) increase (figure 6). The total number of deaths caused by IBD in the high-income super-region increased from 7440 (95% UI 6910–9550) in 1990 to 16 900 (10 300–18 800) in 2017 (super-region data available from the GBD results tool). North Africa and the Middle East remained the super-region with the lowest number of deaths, with 872 (781–999) deaths in 2017. The number of deaths was 372 (317–432) in females and 500 (436–605) in males. The 2017 age-standardised death rate in this super-region was also the lowest among all GBD super-regions (0·21 [0·19–0·25] per 100 000 population). A sharp decrease in the age-standardised death rate also occurred in southeast Asia, east Asia, and Oceania from 1990 to 2017 (figure 6).Figure 6 Trends from 1990 to 2017 in age-standardised death rate of IBD in seven GBD super-regions IBD=inflammatory bowel disease. GBD=Global Burden of Disease, Injuries, and Risk Factors Study.

The high-income super-region also had the largest increase in age-standardised death rate from 1990 to 2017 (17·6% [95% UI −35·7 to 34·7]; figure 6). Sub-Saharan Africa was the only other super-region to experience an increase in age-standardised death rate over the study period, a 0·7% (−32·5 to 43·6) increase (figure 6). The total number of deaths caused by IBD in the high-income super-region increased from 7440 (95% UI 6910–9550) in 1990 to 16 900 (10 300–18 800) in 2017 (super-region data available from the GBD results tool). North Africa and the Middle East remained the super-region with the lowest number of deaths, with 872 (781–999) deaths in 2017. The number of deaths was 372 (317–432) in females and 500 (436–605) in males. The 2017 age-standardised death rate in this super-region was also the lowest among all GBD super-regions (0·21 [0·19–0·25] per 100 000 population). A sharp decrease in the age-standardised death rate also occurred in southeast Asia, east Asia, and Oceania from 1990 to 2017 (figure 6).Figure 6 Trends from 1990 to 2017 in age-standardised death rate of IBD in seven GBD super-regions IBD=inflammatory bowel disease. GBD=Global Burden of Disease, Injuries, and Risk Factors Study. High-income North America was the region with the highest age-standardised prevalence rate for both sexes between 1990 and 2017 (344·8 [95% UI 331·7–359·3] per 100 000 population in 1990 and 422·0 [398·7–466·1] per 100 000 population in 2017 for both sexes combined; figure 7; appendix pp 11, 22). The lowest age-standardised prevalence rate in 2017 was observed in the Caribbean (6·7 [6·3–7·2] per 100 000 population; appendix pp 22–29), followed by Andean Latin America and the four sub-Saharan Africa regions.Figure 7 Age-standardised prevalence rate of IBD globally and for 21 GBD regions by SDI, 1990–2017

1, 22). The lowest age-standardised prevalence rate in 2017 was observed in the Caribbean (6·7 [6·3–7·2] per 100 000 population; appendix pp 22–29), followed by Andean Latin America and the four sub-Saharan Africa regions.Figure 7 Age-standardised prevalence rate of IBD globally and for 21 GBD regions by SDI, 1990–2017 For each region, points from left to right depict estimates from each year from 1990 to 2017. IBD=inflammatory bowel disease. GBD=Global Burden of Disease, Injuries, and Risk Factors Study. The western Europe region had the highest age-standardised death rate in 2017 (0·97 [95% UI 0·54–1·11] per 100 000 population), followed by high-income North America (0·83 [0·55–0·91] per 100 000 population; appendix pp 13, 30–37), driven primarily in high-income North America by high number of deaths in the USA (appendix p 30). Both western Europe (33·6% [–37·5 to 59·0]) and high-income North America (36·9% [–22·3 to 57·1]) experienced an increase in age-standardised death rate over the study period (figure 8). The age-standardised death rate decreased sharply in east Asia (59·6% [66·3–27·3] decrease from 1990 to 2017; appendix pp 14, 33). High-income Asia Pacific (0·16 [0·13–0·24] per 100 000 population), north Africa and the Middle East (0·21 [0·19–0·25] per 100 000 population), and Andean Latin America (0·26 [0·22–0·29] per 100 000 population) had the lowest age-standardised death rate among regions in 2017 (figure 8; appendix p 13).Figure 8 Age-standardised death rate of IBD globally and for 21 GBD regions by SDI, 1990–2017

frica and the Middle East (0·21 [0·19–0·25] per 100 000 population), and Andean Latin America (0·26 [0·22–0·29] per 100 000 population) had the lowest age-standardised death rate among regions in 2017 (figure 8; appendix p 13).Figure 8 Age-standardised death rate of IBD globally and for 21 GBD regions by SDI, 1990–2017 The expected age-standardised death rate in 2017 based solely on SDI is represented by the black line. For each region, points from left to right depict estimates from each year from 1990 to 2017. IBD=inflammatory bowel disease. SDI=Socio-demographic Index. Higher SDI was associated with higher age-standardised prevalence rates of IBD, with values that were higher than the global rate in the two highest SDI quintiles, and lower than the global rate in the three lowest SDI quintiles (figure 9). High SDI and low SDI quintiles had the highest and lowest age-standardised prevalence rates (213·0 [95% UI 202·3–223·8] per 100 000 population and 13·8 [12·6–15·2] per 100 000 population), respectively, in 2017 (figure 8). By contrast, while the high SDI quintile (0·71 [0·44–0·78] per 100 000 population) also had the highest age-standardised death rate in 2017, the low SDI quintile (0·52 [0·42–0·64]) had the second-highest rate. The largest decrease in age-standardised death rate from 1990 to 2017 occurred in the middle SDI quintile (46·2% [51·8–22·4] decrease; GBD results tool).Figure 9 Trends from 1990 to 2017 in age-standardised prevalence rates of IBD by SDI quintile IBD=inflammatory bowel disease. SDI=Socio-demographic Index.

Higher SDI was associated with higher age-standardised prevalence rates of IBD, with values that were higher than the global rate in the two highest SDI quintiles, and lower than the global rate in the three lowest SDI quintiles (figure 9). High SDI and low SDI quintiles had the highest and lowest age-standardised prevalence rates (213·0 [95% UI 202·3–223·8] per 100 000 population and 13·8 [12·6–15·2] per 100 000 population), respectively, in 2017 (figure 8). By contrast, while the high SDI quintile (0·71 [0·44–0·78] per 100 000 population) also had the highest age-standardised death rate in 2017, the low SDI quintile (0·52 [0·42–0·64]) had the second-highest rate. The largest decrease in age-standardised death rate from 1990 to 2017 occurred in the middle SDI quintile (46·2% [51·8–22·4] decrease; GBD results tool).Figure 9 Trends from 1990 to 2017 in age-standardised prevalence rates of IBD by SDI quintile IBD=inflammatory bowel disease. SDI=Socio-demographic Index. The age-standardised prevalence rate of IBD in the two most populous countries, China and India, were 136·2 (95% UI 125·4 to 147·4) per 100 000 population and 16·2 (14·7 to 17·9) per 100 000 population, respectively, in 2017 (see appendix p 22–29 for the full national-level data for age-standardised prevalence rates). The highest age-standardised prevalence rate of IBD was observed in the USA (464·5 [438·6 to 490·9] per 100 000 population), followed by the UK (449·6 [420·6 to 481·6] per 100 000 population). At a national level, the largest change in age-standardised prevalence rate of IBD from 1990 to 2017 occurred in the Solomon Islands, where the age-standardised prevalence rate increased by 139·8% (106·3 to 177·7). In 1990, the age-standardised prevalence rate of IBD in this country was 10·3 (9·2 to 11·5) per 100 000 population, increasing to 24·6 (20·7 to 29·5) per 100 000 population in 2017. Vanuatu had the highest age-standardised death rate in 2017 (1·8 [0·8 to 3·2] per 100 000 population), and Singapore had the lowest (0·08 [0·06 to 0·14] per 100 000 population; appendix pp 17, 30–37; see appendix pp 30–37 for the full national level data for age-standardised death rates). At the national level, Italy had the largest increase in age-standardised death rate over the study period, from 0·40 (0·34 to 0·63) per 100 000 population in 1990 to 0·79 (0·34 to 0·95) per 100 000 population in 2017—a 99% (−39·4 to 170·6) increase. South Korea had the largest decrease in age-standardised death rate over the study period (77·8% [84·6 to 8·7] decrease per 100 000 population (appendix pp 18, 30–37). The age-standardised death rate in South Korea decreased from 1·62 (0·48 to 2·07) per 100 000 population in 1990 to 0·36 (0·30 to 0·49) per 100 000 population in 2017. Age-standardised DALY rates caused by IBD in 2017, for 195 countries and territories, are presented in figure 10 and in the appendix (pp 38–45).Figure 10 Age-standardised DALYs rates from IBD by SDI for 195 countries and territories in 2017

0 000 population in 1990 to 0·36 (0·30 to 0·49) per 100 000 population in 2017. Age-standardised DALY rates caused by IBD in 2017, for 195 countries and territories, are presented in figure 10 and in the appendix (pp 38–45).Figure 10 Age-standardised DALYs rates from IBD by SDI for 195 countries and territories in 2017 The black line represents the expected age-standardised DALY rate of IBD based solely on SDI. DALY=disability-adjusted life-year. IBD=inflammatory bowel disease. SDI=Socio-demographic Index. Discussion In this study, we used a standardised approach to describe burden due to IBD at the global, super-region, regional, and national levels. We report that currently, approximately nearly 3·9 million females and nearly 3·0 million males are living with IBD worldwide and the number of prevalent cases is on the rise. This is important for health-care delivery systems and economies in the global context of treating chronic diseases like IBD, because standard care for these conditions, particularly immunotherapies, is extremely costly.

y 3·0 million males are living with IBD worldwide and the number of prevalent cases is on the rise. This is important for health-care delivery systems and economies in the global context of treating chronic diseases like IBD, because standard care for these conditions, particularly immunotherapies, is extremely costly. Historically, IBD has been considered a condition of high-income countries.1, 10 We found that the region of high-income North America, specifically the USA, makes a prominent contribution to the global number of patients with IBD. The USA had the highest age-standardised prevalence rate globally, with nearly a quarter of total global patients with IBD living there in 2017. Among European countries, the UK had the highest age-standardised prevalence. The prevalence of IBD was reported to range from 252 to 439 cases per 100 000 population in the USA.11 A 2018 systematic review3 evaluating more than 200 population-based studies reported that the highest prevalence rate of IBD occurred in North America. In the UK, a prevalence as high as 373 per 100 000 population has been reported.12 Our estimates confirmed these findings and showed an age-standardised prevalence rate of 464·5 (95% UI 438·6–490·9) per 100 000 population for the USA.

d studies reported that the highest prevalence rate of IBD occurred in North America. In the UK, a prevalence as high as 373 per 100 000 population has been reported.12 Our estimates confirmed these findings and showed an age-standardised prevalence rate of 464·5 (95% UI 438·6–490·9) per 100 000 population for the USA. We also noticed a clear trend in prevalence of IBD from low to high SDI quintiles, with higher prevalence in countries in the high SDI quintile. This pattern was preserved over time, suggesting that the burden of IBD was consistently greater in countries with a high index of development such as the UK, the USA, Canada, and Australia. This correlation, suggested by many studies,13, 14, 15 might indicate that there are common environmental pressures across these regions that act as important risk factors for IBD, although we did not evaluate the role of potential risk factors in IBD prevalence for this study. These risk factors might include urbanisation, more hygienic environments, and diets low in dietary fibre and high in meat.15, 16 Based on the immune disease development model, the higher prevalence of IBD among people with higher socioeconomic status has been suggested to be due to a delay in or low level of exposure to common infectious agents during childhood. Consequently, the immune response is altered in genetically susceptible individuals.17, 18 Therefore, the primary health-care indicators of high-income and low-income countries might be correlated with prevalence rates, because in many low-income countries in Asia and Africa, basic sanitation is still an issue. The higher prevalence of IBD in regions with higher SDI might also be interpreted as an indication that individuals with higher socio-demographic status are at higher risk of IBD and need more investigation during their routine check-ups. Another possible explanation for the higher prevalence of IBD in higher sociodemographic regions might be better access to diagnostic testing tools, resulting in higher rates of diagnosis.16

iduals with higher socio-demographic status are at higher risk of IBD and need more investigation during their routine check-ups. Another possible explanation for the higher prevalence of IBD in higher sociodemographic regions might be better access to diagnostic testing tools, resulting in higher rates of diagnosis.16 We report an increase in age-standardised prevalence rate of IBD in regions that formerly had low prevalence, including east and south Asia, Oceania, and sub-Saharan Africa. It is probable that a combination of factors, including improvements in the socioeconomic status of newly industrialised countries, changes in diet and other lifestyle changes, improved sanitation, changed microbiota, and environmental factors, increase the risk of developing IBD.10, 13, 19 Behavioural and environmental factors might play an increasingly critical role in the development of IBD.20 Different factors that might increase the risk of developing IBD include smoking, lifestyle choices, discontinued breastfeeding, enteric infections, appendicectomy, and air pollution.20, 21 Improvement in access to health-care systems, more widely available diagnostic tools, and increased awareness on the part of both patients and physicians19 might also contribute to higher rates of diagnosis.

ing, lifestyle choices, discontinued breastfeeding, enteric infections, appendicectomy, and air pollution.20, 21 Improvement in access to health-care systems, more widely available diagnostic tools, and increased awareness on the part of both patients and physicians19 might also contribute to higher rates of diagnosis. The continuing increase in prevalence of IBD in previously low-prevalence areas has important implications for both health-care providers and those responsible for health-care policy planning. Almost three-quarters of all people (about 3–5 billion) live in developing countries.22 Almost 2·7 billion people live in India and China. Therefore, even a small increase in the occurrence of chronic diseases such as IBD, which has low mortality but high disability, could have devastating effects in developing countries in the coming years.3 Of note, the economic effects of IBD are not limited to its burden on health-care systems. A German study23 reported that, each year, about 9% and 3% of all German employees with IBD had rehabilitation or were granted a disability pension, respectively. More than 50% of the total social costs of IBD are indirect costs such as early retirement or sick leave.24 Therefore, IBD contributes to an ever-increasing burden, not only on health-care systems but also on the economy as a whole.

ployees with IBD had rehabilitation or were granted a disability pension, respectively. More than 50% of the total social costs of IBD are indirect costs such as early retirement or sick leave.24 Therefore, IBD contributes to an ever-increasing burden, not only on health-care systems but also on the economy as a whole. We report that the age-specific death rate at the global and regional levels attenuated from 1990 to 2017. The overall decrease in mortality probably reflects the improved survival of patients with IBD, which might be due to increased use of immunomodulators, earlier introduction of biological agents, improvements in surgical techniques, and increased awareness of colorectal cancer surveillance.25, 26 Because the prevalence of IBD is much lower in low SDI countries, we expected a lower age-standardised death rate in these countries than in high SDI locations. However, the low quality of death registries in these countries might be another reason behind the low number of reported deaths in low SDI countries. Moreover, it is clear that IBD is not an easy disease to diagnose—ie, it requires a colonoscopy, and this is not available for the majority of people living in low SDI countries. It is possible that the fewer reported deaths due to IBD in low SDI countries might be due to under-detection of IBD-related deaths.

untries. Moreover, it is clear that IBD is not an easy disease to diagnose—ie, it requires a colonoscopy, and this is not available for the majority of people living in low SDI countries. It is possible that the fewer reported deaths due to IBD in low SDI countries might be due to under-detection of IBD-related deaths. Although the fatal burden of IBD remains relatively low, the non-fatal burden continues to increase, climbing from the fifth-leading cause of YLDs among digestive diseases in 1990 to the fourth in 2017. IBD can substantially compromise the physical, psychological, familial, and social dimensions of life. As a result, the secondary effects of the disease can be seen in the increased rates of anxiety, depression, and other emotional effects. A 2016 study27 showed a notable association between symptoms of depression and clinical disease activity in patients with IBD, regardless of IBD subtypes. However, only IBD-specific symptoms are accounted for in disability weights in GBD, and not the social stigma, depression, anxiety, and other inflammatory conditions.

2016 study27 showed a notable association between symptoms of depression and clinical disease activity in patients with IBD, regardless of IBD subtypes. However, only IBD-specific symptoms are accounted for in disability weights in GBD, and not the social stigma, depression, anxiety, and other inflammatory conditions. Sex-stratified global and national incidence rates of IBD, reported either by GBD or by other studies, are similar, suggesting that the disease is not sex-specific.28 However, the age-standardised death rate of IBD is lower, but the prevalence rate is higher, among females. It is possible that differences in environmental determinants derived from biological, social, and economic exposures between males and females might be responsible for this difference. The higher prevalence of smoking, as one of the most consistently studied environmental factors of IBD, in males compared with females might have contributed to the higher mortality rate in male patients.29 Alternatively, some studies point to the influence of hormones on the brain–gut–microbiota axis as the reason for sex differences in IBD prognosis, but the mechanism underlying this complex pathophysiology is still not completely understood.

emales might have contributed to the higher mortality rate in male patients.29 Alternatively, some studies point to the influence of hormones on the brain–gut–microbiota axis as the reason for sex differences in IBD prognosis, but the mechanism underlying this complex pathophysiology is still not completely understood. As is the case with all GBD research, our study was limited by low availability and quality of data, which could only partially be overcome by using statistical methods. In data-scarce locations, we had to rely on predictive covariates and spatiotemporal trends. For non-fatal models, for which data were especially scarce, this resulted in estimates for some regions being determined entirely by global trends and associations with income and health-care access (in the case of Oceania and western, central, and southern sub-Saharan Africa), and by these factors plus data for a single country (in the case of eastern sub-Saharan Africa and South Asia). This is reflected in wider UIs in these locations, and suggests extra caution should be applied in interpreting estimates for these locations. It might also explain an unexpected discrepancy in age-standardised prevalence rate between China (136·2 [95% UI 125·4–147·4] per 100 000 population), where we had many data inputs, and India (16·2 [14·7–17·9] per 100 000 population) in 2017, where estimates were heavily influenced by a single Nepalese data source. Our results for India were lower than previous reports, mainly for the northern parts of the country. The age-standardised prevalence rate estimates for India also contrast previous reports,30, 31, 32 suggesting that a single Nepalese study in the GBD 2017 database for south Asia may be poorly representative of the region. Primary data from these previous reports should be incorporated into future rounds of GBD to improve these estimates. Likewise, special effort should be made to obtain more input data from other data-scarce regions.

ingle Nepalese study in the GBD 2017 database for south Asia may be poorly representative of the region. Primary data from these previous reports should be incorporated into future rounds of GBD to improve these estimates. Likewise, special effort should be made to obtain more input data from other data-scarce regions. Additionally, some prevalence estimates could have been improved by imposing an upper bound on the prior value for excess mortality rate or providing data inputs on excess mortality to our compartmental models. For example, a large population-based study33 in Canada reported a prevalence of IBD around 520 per 100 000 population, notably higher than the GBD 2017 prevalence estimates, which ranged from 46 to 57 cases per 100 000 population; this lower estimate was influenced by a high previous distribution of excess mortality in our high-income North America model, which was overcome by abundant prevalence data inputs for the USA, but not by the relatively fewer prevalence data inputs for Canada.

s, which ranged from 46 to 57 cases per 100 000 population; this lower estimate was influenced by a high previous distribution of excess mortality in our high-income North America model, which was overcome by abundant prevalence data inputs for the USA, but not by the relatively fewer prevalence data inputs for Canada. Lastly, because of greater availability of data from facilities and insurance claims, we relied on ICD10-based diagnosis for our case definition, and used fixed effects from our global mixed-effects compartmental model to adjust data with other case definitions toward this reference. In future rounds of GBD, we will use previously published estimates of the predictive value of ICD codes in administrative data to adjust our administrative data inputs toward the magnitude we would expect with stringent diagnosis, thus improving the specificity of our case definition but maintaining the geographical coverage afforded by administrative data. We also hope to include an analysis of the burden of IBD that is attributable to certain risk factors in future iterations of GBD, once the links between IBD and potential risk factors are better understood.

he specificity of our case definition but maintaining the geographical coverage afforded by administrative data. We also hope to include an analysis of the burden of IBD that is attributable to certain risk factors in future iterations of GBD, once the links between IBD and potential risk factors are better understood. The natural course of IBD, with low mortality, as well as improved survival, caused an increase in prevalence of the disease from 1990 to 2017. In keep with this trend, prevalence is expected to continue increasing in the future. Rising prevalence, plus the increase in incidence in historically low-incidence regions, will have important health and economic effects. Our findings could be useful for health service planners and policy makers to justify and prioritise resource allocation to be able to respond to the growing number of patients with IBD. This study will motivate health planners to develop cost-effective and simple community-based interventions for implementation by health-care professionals at the primary-care level. This is necessary because IBD can last for many years and the ageing population is increasing. We emphasise that understanding the shared and different environmental determinants of IBD across various regions is essential to implement interventions that will slow down the rising global burden of IBD. Supplementary Material Supplementary appendix

The natural course of IBD, with low mortality, as well as improved survival, caused an increase in prevalence of the disease from 1990 to 2017. In keep with this trend, prevalence is expected to continue increasing in the future. Rising prevalence, plus the increase in incidence in historically low-incidence regions, will have important health and economic effects. Our findings could be useful for health service planners and policy makers to justify and prioritise resource allocation to be able to respond to the growing number of patients with IBD. This study will motivate health planners to develop cost-effective and simple community-based interventions for implementation by health-care professionals at the primary-care level. This is necessary because IBD can last for many years and the ageing population is increasing. We emphasise that understanding the shared and different environmental determinants of IBD across various regions is essential to implement interventions that will slow down the rising global burden of IBD. Supplementary Material Supplementary appendix Acknowledgments This study is funded by the Bill & Melinda Gates Foundation. AA is supported by the Department of Science and Technology, Government of India, New Delhi through INSPIRE Faculty programme. AB and JF acknowledge support with funding from Fundacao para a Ciencia e a Tecnologia/Ministerio da Ciencia, Tecnologia e Ensino Superior (FCT/MCTES) through Portuguese national funds, though UID/MULTI/04378/2019 (AB), UID/QUI/50006/2019 (AB), and UID/Multi/50016/2019 (JF) grants. AD would like to thank the College of Science and Engineering, Hamad Bin Khalifa University, Qatar, for providing him with the time and support to work on this important publication.

CTES) through Portuguese national funds, though UID/MULTI/04378/2019 (AB), UID/QUI/50006/2019 (AB), and UID/Multi/50016/2019 (JF) grants. AD would like to thank the College of Science and Engineering, Hamad Bin Khalifa University, Qatar, for providing him with the time and support to work on this important publication. GBD 2017 Inflammatory Bowel Disease Collaborators Sudabeh Alatab, Sadaf G Sepanlou, Kevin Ikuta, Homayoon Vahedi, Catherine Bisignano, Saeid Safiri, Anahita Sadeghi, Molly R Nixon, Amir Abdoli, Hassan Abolhassani, Vahid Alipour, Majid A H Almadi, Amir Almasi-Hashiani, Amir Anushiravani, Jalal Arabloo, Suleman Atique, Ashish Awasthi, Alaa Badawi, Atif A A Baig, Neeraj Bhala, Ali Bijani, Antonio Biondi, Antonio M Borzì, Kristin E Burke, Félix Carvalho, Ahmad Daryani, Manisha Dubey, Aziz Eftekhari, Eduarda Fernandes, João C Fernandes, Florian Fischer, Arvin Haj-Mirzaian, Arya Haj-Mirzaian, Amir Hasanzadeh, Maryam Hashemian, Simon I Hay, Chi L Hoang, Mowafa Househ, Olayinka S Ilesanmi, Nader Jafari Balalami, Spencer L James, Andre P Kengne, Masoud M Malekzadeh, Shahin Merat, Tuomo J Meretoja, Tomislav Mestrovic, Erkin M Mirrakhimov, Hamed Mirzaei, Karzan A Mohammad, Ali H Mokdad, Lorenzo Monasta, Ionut Negoi, Trang H Nguyen, Cuong T Nguyen, Akram Pourshams, Hossein Poustchi, Mohammad Rabiee, Navid Rabiee, Kiana Ramezanzadeh, David L Rawaf, Salman Rawaf, Nima Rezaei, Stephen R Robinson, Luca Ronfani, Sonia Saxena, Masood Sepehrimanesh, Masood A Shaikh, Zeinab Sharafi, Mehdi Sharif, Soraya Siabani, Ali Reza Sima, Jasvinder A Singh, Amin Soheili, Rasoul Sotoudehmanesh, Hafiz Ansar Rasul Suleria, Berhe E Tesfay, Bach Tran, Derrick Tsoi, Marco Vacante, Adam B Wondmieneh, Afshin Zarghi, Zhi-Jiang Zhang, Mae Dirac, Reza Malekzadeh*, and Mohsen Naghavi*. *These authors jointly supervised the study.

rif, Soraya Siabani, Ali Reza Sima, Jasvinder A Singh, Amin Soheili, Rasoul Sotoudehmanesh, Hafiz Ansar Rasul Suleria, Berhe E Tesfay, Bach Tran, Derrick Tsoi, Marco Vacante, Adam B Wondmieneh, Afshin Zarghi, Zhi-Jiang Zhang, Mae Dirac, Reza Malekzadeh*, and Mohsen Naghavi*. *These authors jointly supervised the study. Affiliations Department of Medical Immunology (H Mirzaei PhD), Department of Microbiology (A Hasanzadeh PhD), Department of Pharmacology (A Haj-Mirzaian MD, A Haj-Mirzaian MD), Digestive Disease Research Institute (A Anushiravani MD), Digestive Diseases Research Institute (S Alatab PhD, S G Sepanlou MD, H Vahedi MD, A Sadeghi MD, M M Malekzadeh MD, Prof S Merat MD, Prof A Pourshams MD, H Poustchi PhD, A Sima MD, Prof R Sotoudehmanesh BHlthSci, Prof R Malekzadeh MD), Research Center for Immunodeficiencies (Prof N Rezaei PhD), Research center for Immunodeficiencies (H Abolhassani PhD), Shariati Hospital (A Sima MD), Tehran University of Medical Sciences, Tehran, Iran; Non-Communicable Diseases Research Center (S G Sepanlou MD), Non-communicable Diseases Research Center (Prof R Malekzadeh MD), Shiraz University of Medical Sciences, Shiraz, Iran; Department of Health Metrics Sciences, School of Medicine (Prof S I Hay FMedSci, Prof A H Mokdad PhD, Prof M Naghavi MD), Division of Allergy and Infectious Diseases (K Ikuta MD), Institute for Health Metrics and Evaluation (K Ikuta MD, C Bisignano MPH, Prof S I Hay FMedSci, S L James MD, Prof A H Mokdad PhD, M R Nixon PhD, D Tsoi BS, M Dirac MD, Prof M Naghavi MD), University of Washington, Seattle, WA, USA; Aging Research Institute (S Safiri PhD), Department of Community Medicine (S Safiri PhD), Department of Pharmacology and Toxicology (A Eftekhari PhD), Tabriz University of Medical Sciences, Tabriz, Iran; Department of Parasitology and Mycology (A Abdoli PhD), Jahrom University of Medical Sciences, Jahrom, Iran; LABMED (H Abolhassani PhD), Karolinska University Hospital, Stockholm, Sweden; Department of Epidemiology (A Almasi-Hashiani PhD), Health Management and Economics Research Center (V Alipour PhD), Tehran, Iran; Health Economics Department (V Alipour PhD), Health Management and Economics Research Center (J Arabloo PhD), Iran University of Medical Sciences, Tehran, Iran; College of Medicine (M A H Almadi FRCPC), King Saud University, Riyadh, Saudi Arabia; Division of Gastroenterology & Hepatology (M A H Almadi FRCPC), McGill University, Montreal, QB, Canada; University Institute of Public Health (S Atique PhD), The

ter (J Arabloo PhD), Iran University of Medical Sciences, Tehran, Iran; College of Medicine (M A H Almadi FRCPC), King Saud University, Riyadh, Saudi Arabia; Division of Gastroenterology & Hepatology (M A H Almadi FRCPC), McGill University, Montreal, QB, Canada; University Institute of Public Health (S Atique PhD), The University of Lahore, Lahore, Pakistan; College of Public Health (S Atique PhD), University of Hail, Hail, Saudi Arabia; Indian Institute of Public Health, Gandhinagar, India (A Awasthi PhD); Public Health Foundation of India, Gurugram, India (A Awasthi PhD); Public Health Risk Sciences Division (A Badawi PhD), Public Health Agency of Canada, Toronto, ON, Canada; Department of Nutritional Sciences (A Badawi PhD), University of Toronto, Toronto, ON, Canada; Biomedicine Department (A A A Baig PhD), Unit of Biochemistry, Faculty of Medicine (A A A Baig PhD), Universiti Sultan Zainal Abidin, Kuala Terengganu, Malaysia; Institutes of Applied Health Research and Translational Medicine (N Bhala DPhil), Queen Elizabeth Hospital Birmingham, Birmingham, UK; IAHR/ITM (N Bhala DPhil), University of Birmingham, Birmingham, UK; Social Determinants of Health Research Center (A Bijani PhD), Babol University of Medical Sciences, Babol, Iran; Department of General Surgery and Medical-Surgical Specialties (Prof A Biondi PhD, M Vacante PhD), General Surgery and Medical-Surgical Specialties (A M Borzì MD), University of Catania, Catania, Italy; Division Gastroenterology (K E Burke MD), Massachusetts General Hospital, Boston, MA, USA; Applied Molecular Biosciences Unit (Prof F Carvalho PhD), Institute of Public Health (Prof F Carvalho PhD), University of Porto, Porto, Portugal; Toxoplasmosis Research Center (Prof A Daryani PhD), Mazandaran University of Medical Sciences, Sari, Iran; United Nations World Food Programme (M Dubey PhD), United Nations World Food Programme, New Delhi, India; Department of Microbiology (A Hasanzadeh PhD), Pharmacology and Toxicology Department (A Eftekhari PhD), Maragheh University of Medical Sciences, Maragheh, Iran; Center for Biotechnology and Fine Chemistry (J C Fernandes PhD), Catholic University of Portugal, Porto, Portugal; REQUIMTE/LAQV (Prof E Fernandes PhD), University of Porto, Porto, Portugal; School of Public Health Medicine (F Fischer PhD), Bielefeld University, Bielefeld, Germany; Department of Medicinal and Pharmaceutical Chemistry (Prof A Zarghi PhD), Department of Pharmacology (K Ramezanzadeh PharmD), Obesity Research Center (A Haj-Mirza

AQV (Prof E Fernandes PhD), University of Porto, Porto, Portugal; School of Public Health Medicine (F Fischer PhD), Bielefeld University, Bielefeld, Germany; Department of Medicinal and Pharmaceutical Chemistry (Prof A Zarghi PhD), Department of Pharmacology (K Ramezanzadeh PharmD), Obesity Research Center (A Haj-Mirza ian MD), Shahid Beheshti University of Medical Sciences, Tehran, Iran; Department of Radiology (A Haj-Mirzaian MD), Johns Hopkins University, Baltimore, MD, USA; Division of Cancer Epidemiology and Genetics (M Hashemian PhD), National Cancer Institute, Rockville, MD, USA; Department of Biology (M Hashemian PhD), Utica College, Utica, NY, USA; Center of Excellence in Behavioral Medicine (C L Hoang BMedSci, T H Nguyen BMedSci), Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam; Division of Information and Computing Technology, College of Science and Engineering (Prof M Househ PhD), Hamad Bin Khalifa University, Doha, Qatar; Qatar Foundation for Education, Science, and Community Development, Doha, Qatar (Prof M Househ PhD); Department of Community Medicine (O S Ilesanmi PhD), University of Ibadan, Ibadan, Nigeria; Department of Psychosis (N Jafari Balalami PhD), Babol Noshirvani University of Technology, Babol, Iran; Non-Communicable Diseases Research Unit (Prof A P Kengne PhD), Medical Research Council South Africa, Cape Town, South Africa; Department of Medicine (Prof A P Kengne PhD), University of Cape Town, Cape Town, South Africa; Breast Surgery Unit (T J Meretoja MD), Helsinki University Hospital, Helsinki, Finland; University of Helsinki, Helsinki, Finland (T J Meretoja MD); Clinical Microbiology and Parasitology Unit (T Mestrovic PhD), Zora Profozic Polyclinic, Zagreb, Croatia; University Centre Varazdin (T Mestrovic PhD), University North, Varazdin, Croatia; Faculty of General Medicine (Prof E M Mirrakhimov MD), Kyrgyz State Medical Academy, Bishkek, Kyrgyzstan; Department of Atherosclerosis and Coronary Heart Disease (Prof E M Mirrakhimov MD), National Center of Cardiology and Internal Disease, Bishkek, Kyrgyzstan; Research Center for Biochemistry and Nutrition in Metabolic Diseases (H Mirzaei PhD), Kashan University of Medical Sciences, Kashan, Iran; Department of Biology (K A Mohammad PhD), Salahaddin University, Erbil, Iraq; ISHIK University, Erbil, Iraq (K A Mohammad PhD); Clinical Epidemiology and Public Health Research Unit (L Monasta DSc, L Ronfani PhD), Burlo Garofolo Institute for Maternal and Child Health, Trieste, Italy; Emergency Hospital

, Kashan, Iran; Department of Biology (K A Mohammad PhD), Salahaddin University, Erbil, Iraq; ISHIK University, Erbil, Iraq (K A Mohammad PhD); Clinical Epidemiology and Public Health Research Unit (L Monasta DSc, L Ronfani PhD), Burlo Garofolo Institute for Maternal and Child Health, Trieste, Italy; Emergency Hospital of Bucharest (I Negoi PhD), General Surgery Department (I Negoi PhD), Carol Davila University of Medicine and Pharmacy, Bucharest, Romania; Institute for Global Health Innovations (C T Nguyen MPH), Duy Tan University, Hanoi, Vietnam; Biomedical Engineering Department (Prof M Rabiee PhD), Amirkabir University of Technology, Tehran, Iran; Department of Chemistry (N Rabiee PhD), Sharif University of Technology, Tehran, Iran; Department of Primary Care and Public Health (Prof S Rawaf MD), School of Public Health (Prof S Saxena MD), WHO Collaborating Centre for Public Health Education and Training (D L Rawaf MD), Imperial College London, London, UK; University College London Hospitals, London, UK (D L Rawaf MD); Academic Public Health (Prof S Rawaf MD), Public Health England, London, UK; Network of Immunity in Infection, Malignancy and Autoimmunity (NIIMA) (Prof N Rezaei PhD), Universal Scientific Education and Research Network (USERN), Tehran, Iran; Department of Psychology (Prof S R Robinson PhD), Royal Melbourne Institute of Technology University, Bundoora, VIC, Australia; New Iberia Research Center (M Sepehrimanesh PhD), University of Louisiana at Lafayette, Lafayette, LA, USA; Guilan University of Medical Sciences, Rasht, Iran (M Sepehrimanesh PhD); Independent Consultant, Karachi, Pakistan (M A Shaikh MD); Razi Herbal Medicines Research Center (Z Sharafi PhD), Lorestan University of Medical Sciences, Khorramabad, Iran; Department of Basic Sciences (Prof M Sharif PhD), Department of Laboratory Sciences (Prof M Sharif PhD), Islamic Azad University, Sari, Iran; Imam Ali Cardiovascular Research Center (S Siabani PhD), Kermanshah University of Medical Sciences, Kermanshah, Iran; School of Health (S Siabani PhD), University of Technology Sydney, Sydney, NSW, Australia; Rheumatology/Medicine Department (J A Singh MD), University of Alabama at Birmingham, Birmingham, AL, United States; Medical Surgical Nursing Department (A Soheili PhD), Urmia University of Medical Science, Urmia, Iran; Emergency Nursing Department (A Soheili PhD), Zanjan University of Medical Sciences, Iran; Department of Agriculture and Food Systems (H Suleria PhD), University of Melbourne, Melbourn

, Birmingham, AL, United States; Medical Surgical Nursing Department (A Soheili PhD), Urmia University of Medical Science, Urmia, Iran; Emergency Nursing Department (A Soheili PhD), Zanjan University of Medical Sciences, Iran; Department of Agriculture and Food Systems (H Suleria PhD), University of Melbourne, Melbourn e, VC, Australia; Department of Public Health (B E Tesfay MPH), Adigrat University, Adigrat, Ethiopia; Department of Health Economics (B Tran PhD), Hanoi Medical University, Hanoi, Vietnam; Nursing Department (A B Wondmieneh MSc), Wollo University, Dessie, Ethiopia; Addis Ababa University, Addis Ababa, Ethiopia (A B Wondmieneh MSc); and Department of Preventive Medicine (Z Zhang PhD), Wuhan University, Wuhan, China. Contributors SA, SGS, SS, AS, AS, AA, and MMM prepared the first draft. RM, HV, SM, and MN provided overall guidance. RM, SGS, and SS managed the project. SGS, SS, MD, and DT analysed data. RM, SGS, SA, SM, HV, CB, and MD finalised Article on the basis of comments from other authors and reviewer feedback. All other authors provided data, developed models, reviewed results, provided guidance on methods, or reviewed and contributed to the Article.

d SS managed the project. SGS, SS, MD, and DT analysed data. RM, SGS, SA, SM, HV, CB, and MD finalised Article on the basis of comments from other authors and reviewer feedback. All other authors provided data, developed models, reviewed results, provided guidance on methods, or reviewed and contributed to the Article. Declaration of interests SLJ reports grants from Sanofi Pasteur, outside the submitted work. JS reports personal fees for consulting work from Crealta/Horizon, Medisys, Fidia, UBM LLC, Medscape, WebMD, Clinical Care Options, Clearview Healthcare Partners, Putnam Associates, Spherix, The National Institutes of Health, and The American College of Rheumatology, outside the submitted work; and reports owning stock options in Amarin Pharmaceuticals and Viking Pharmaceuticals, outside the submitted work. All other authors declare no competing interests.

Research in context Evidence before this study In 1994, an analysis of official death certification data derived from the WHO database was conducted, and reports were released on deaths due to cirrhosis from 1955 to 1990 in high-income countries. More recently, a 2014 report on findings from the Global Burden of Disease Study (GBD) 2010 reported the mortality of cirrhosis and other chronic liver diseases in 187 countries from 1980 to 2010. Both estimates showed increasing numbers and decreasing rates across almost all countries. Other local studies have also shown the contributions of vaccination and lifestyle modification to reductions in death rates, incidence, and prevalence of cirrhosis. Added value of this study In this study we present the results of the latest iteration of GBD (2017), regarding cirrhosis and other chronic liver diseases (hereafter referred to as cirrhosis). Compared with GBD 2016, we had access to more data points, we used novel and more robust methods, we made separate estimates for decompensated and compensated cirrhosis, and we differentiated non-alcoholic steatohepatitis as the fifth cause of cirrhosis along with hepatitis B and C, alcohol-related liver disease, and the broad category of other causes. Our study provides a valuable insight into the shifting burden of cirrhosis due to different causes. Although hepatitis B caused the greatest proportion of cirrhosis deaths in 2017, our findings suggest that non-alcoholic steatohepatitis might become the leading cause of cirrhosis in the near future. Implications of all the available evidence

In this study we present the results of the latest iteration of GBD (2017), regarding cirrhosis and other chronic liver diseases (hereafter referred to as cirrhosis). Compared with GBD 2016, we had access to more data points, we used novel and more robust methods, we made separate estimates for decompensated and compensated cirrhosis, and we differentiated non-alcoholic steatohepatitis as the fifth cause of cirrhosis along with hepatitis B and C, alcohol-related liver disease, and the broad category of other causes. Our study provides a valuable insight into the shifting burden of cirrhosis due to different causes. Although hepatitis B caused the greatest proportion of cirrhosis deaths in 2017, our findings suggest that non-alcoholic steatohepatitis might become the leading cause of cirrhosis in the near future. Implications of all the available evidence The number of deaths and prevalent cases of both decompensated and compensated cirrhosis are increasing, despite a decrease in age-standardised death rates. The contribution of different causes to mortality and morbidity is expected to change as a result of available prevention and treatment modalities. Because of variation in the contribution of the five causes of cirrhosis assessed in this study and varying amounts and trends of burden across nations, policies should be targeted at national and even subnational levels accordingly. We intend for our findings to be used by policy makers and the public to address the increasing burden of chronic liver diseases.

the five causes of cirrhosis assessed in this study and varying amounts and trends of burden across nations, policies should be targeted at national and even subnational levels accordingly. We intend for our findings to be used by policy makers and the public to address the increasing burden of chronic liver diseases. Introduction Cirrhosis is the leading cause of liver-related death globally.1 It is the end stage of progressive liver fibrosis, in which the hepatic architecture is distorted.2 In the initial stages, cirrhosis is compensated. Most patients are asymptomatic at this stage, and cirrhosis is usually discovered incidentally during medical encounters for other reasons. Thus, reports on the prevalence of compensated cirrhosis are almost always underestimated. Decompensation in patients with compensated cirrhosis is usually defined as the first occurrence of ascites, oesophageal variceal bleeding, hepatic encephalopathy, and, in some individuals, increased bilirubin concentration.3, 4 Because of the nature of decompensation, these patients are rapidly brought to medical attention, and thus reports on the prevalence of decompensated cirrhosis are probably much more accurate than those of compensated cirrhosis.3 Once decompensation occurs, the mortality and morbidity resulting from cirrhosis increase sharply, and depending on the cause of decompensation, the 1-year case-fatality rate can be as high as 80%.5, 6 Almost all of the mortality and morbidity resulting from cirrhosis is caused by the decompensated type. Such patients need frequent medical attention and an increasing amount of medication over the disease course. Quality of life is affected and frequent hospitalisations (admissions and stays) are required.7 As the disease progresses, hospital stays become more frequent and more prolonged. Finally, patients either die or receive a liver transplant, which is a high-burden option for patients, health-care systems, and health financing and governance.7

fected and frequent hospitalisations (admissions and stays) are required.7 As the disease progresses, hospital stays become more frequent and more prolonged. Finally, patients either die or receive a liver transplant, which is a high-burden option for patients, health-care systems, and health financing and governance.7 Compensated cirrhosis is more benign. Patients often have a life expectancy similar to that of healthy adults if cirrhosis remains compensated.5 However, identifying these patients is important because if they are not treated appropriately, they are at risk of progressing to decompensated cirrhosis.3 Cirrhosis is generally considered to be irreversible at later stages, although reversal has been documented in many individuals with compensated cirrhosis after treating the underlying cause.8, 9, 10 The most common causes of cirrhosis are chronic hepatitis B and C, alcohol-related liver disease, and non-alcoholic steatohepatitis (NASH).7 Patients with decompensated cirrhosis are susceptible to complications and a reduction in life expectancy.11

ith compensated cirrhosis after treating the underlying cause.8, 9, 10 The most common causes of cirrhosis are chronic hepatitis B and C, alcohol-related liver disease, and non-alcoholic steatohepatitis (NASH).7 Patients with decompensated cirrhosis are susceptible to complications and a reduction in life expectancy.11 Despite being a global health challenge, estimates of cirrhosis mortality and morbidity are not widely available, especially at national levels, because of data sparsity in many regions where cirrhosis is fatal, particularly in Africa.12 Accurate mortality data are also scarce, because official death records under-report cirrhosis as a cause of death.13 The burden of cirrhosis differs considerably across locations, sexes, races and ethnicities, and socioeconomic strata, and the burden has also varied substantially over time. Moreover, the incidence and prevalence of cirrhosis are not well established, even in population-based studies. Studies are further limited by referral bias; the structure of the population under study (inpatient vs outpatient); absence of standardised definitions; and diversity in methods of assessment, including self-reporting, laboratory tests and non-invasive biomarkers, imaging, liver biopsies, and death certificates.12

es. Studies are further limited by referral bias; the structure of the population under study (inpatient vs outpatient); absence of standardised definitions; and diversity in methods of assessment, including self-reporting, laboratory tests and non-invasive biomarkers, imaging, liver biopsies, and death certificates.12 Data by cause are also inaccurate and diverse. Clarification of the underlying cause and stage of cirrhosis is challenging, even in high-income countries.14 It is generally well-recognised that the contribution of underlying causes is variable across locations and has changed over time. With the availability of methods for controlling hepatitis B and curing hepatitis C,15 along with rising prevalence of metabolic syndrome and obesity as a plausible underlying cause, the proportion of cirrhosis mortality and morbidity caused by NASH is expected to increase in the near future.16, 17

s changed over time. With the availability of methods for controlling hepatitis B and curing hepatitis C,15 along with rising prevalence of metabolic syndrome and obesity as a plausible underlying cause, the proportion of cirrhosis mortality and morbidity caused by NASH is expected to increase in the near future.16, 17 Previous studies have investigated the burden of cirrhosis at the global level. In 1994, La Vecchia and colleagues18 analysed official death certification data from 1955 to 1990 derived from the WHO database to produce estimates covering 38 high-income countries. In 2014, Mokdad and colleagues19 reported the results of the 2010 iteration of the Global Burden of Disease Study (GBD) on mortality due to cirrhosis by four causes in 187 countries from 1980 to 2010. Both studies reported increasing numbers of deaths and decreasing age-standardised mortality rates in most countries. However, updated estimates on mortality and morbidity caused by cirrhosis and the prevalence of decompensated and compensated types by cause were not available, to our knowledge, before this study. Using data from GBD 2017, this study provides estimates of the prevalence, mortality, and disability-adjusted life-years (DALYs) of cirrhosis and other chronic liver diseases by cause, sex, and age at global, regional, and national levels across 195 countries and territories from 1990 to 2017.

ledge, before this study. Using data from GBD 2017, this study provides estimates of the prevalence, mortality, and disability-adjusted life-years (DALYs) of cirrhosis and other chronic liver diseases by cause, sex, and age at global, regional, and national levels across 195 countries and territories from 1990 to 2017. Methods Overview This study is part of GBD 2017,1, 20, 21, 22 which was a systematic effort to estimate the levels and trends of burden caused by 359 diseases and injuries by sex, age, year (1990–2017), and location, including seven super-regions, 21 regions, and 195 countries and territories. We modelled the mortality and prevalence of cirrhosis and other chronic liver diseases, hereafter collectively referred to as cirrhosis. We report numbers and age-standardised and age-specific rates for mortality, prevalence, and DALYs, which are the sum of years of life lost (YLLs) due to premature death and years lived with disability (YLDs).1, 20, 21, 22 This study is compliant with the Guidelines for Accurate and Transparent Health Estimates Reporting.

report numbers and age-standardised and age-specific rates for mortality, prevalence, and DALYs, which are the sum of years of life lost (YLLs) due to premature death and years lived with disability (YLDs).1, 20, 21, 22 This study is compliant with the Guidelines for Accurate and Transparent Health Estimates Reporting. We considered diagnoses coded as B18 and K70–77 in the International Classification of Diseases 10th revision (ICD-10) to be compensated and decompensated cirrhosis. ICD-10 codes B18.0–18.2 were mapped to chronic viral hepatitis B and C in the GBD cause list, B18.8 and 18.9 to other causes, K70 to alcohol-related liver disease, K75.81 to NASH, and the rest of the codes to the category of other causes, which included but was not limited to autoimmune hepatitis (K75.4). The detailed list of ICD-10 codes mapped to the GBD cause list is reported in the appendix (p 4). The causes grouped in the “other chronic liver diseases” category mainly included autoimmune hepatitis, toxic liver diseases, other inflammatory liver diseases, chronic hepatitis not specified, and other diseases of the liver (K76). ICD-10 codes for acute hepatitis were not included in this study. Non-alcoholic fatty liver disease was considered to be a separate entity from compensated and decompensated cirrhosis, and codes for diabetes were also excluded. Finally, deaths caused by hepatocellular carcinoma were excluded because: ICD-10 codes define hepatocellular carcinoma as a cause irrespective of liver cirrhosis; hepatocellular carcinoma can be distinguished from cirrhosis in countries with adequate data; and implications on natural history and management of the two causes are not similar.

by hepatocellular carcinoma were excluded because: ICD-10 codes define hepatocellular carcinoma as a cause irrespective of liver cirrhosis; hepatocellular carcinoma can be distinguished from cirrhosis in countries with adequate data; and implications on natural history and management of the two causes are not similar. Existing evidence shows that ICD-10 coding is valid for defining overall cirrhosis and chronic liver diseases in most of the data sources used to assemble the cause-of-death database but does not have the required accuracy for reporting cirrhosis by the five causes estimated in the current study.23, 24 The models that we used to split the parent cirrhosis mortality and morbidity into the five causes are described below.

seases in most of the data sources used to assemble the cause-of-death database but does not have the required accuracy for reporting cirrhosis by the five causes estimated in the current study.23, 24 The models that we used to split the parent cirrhosis mortality and morbidity into the five causes are described below. Mortality estimates Details of the methods used to estimate mortality have been published previously.1, 20, 21, 22 We modelled decompensated cirrhosis mortality due to the five causes defined by the aforementioned ICD-10 codes using all available data in the cause-of-death database. We obtained data from vital registrations, vital registration samples, and verbal autopsies. Vital registrations are systems that governments use to record the vital events of their residents, including causes of death. Vital registration samples are nationally representative cluster samples in countries where vital registration has low coverage or is unavailable. Verbal autopsy is a method by which causes of death and cause-specific mortality fractions in populations without a complete vital registration system are determined. To record data through verbal autopsy, trained interviewers collect information about the symptoms, signs, and demographic characteristics of a recently deceased person from an individual familiar with the deceased person. The majority of the cause-of-death data are vital registration data obtained from the WHO Mortality Database, which consists of data submitted to WHO by individual countries. Vital registration data are also obtained from country-specific mortality databases operated by official offices. We collected 19 329 data points from vital registrations, 793 data points from vital registration samples, and 1263 points from verbal autopsies. To address problems of zero counts in vital registration and verbal autopsy for a given age group in a given year, we used a Bayesian noise-reduction algorithm.1 We used Cause of Death Ensemble models (CODEm) to estimate liver cirrhosis mortality with uncertainty by age, sex, country or territory, and year. CODEm is the framework used to model most cause-specific death rates in GBD.1 First, all available data were identified and gathered. Second, a diverse set of plausible models were developed to capture well-documented associations in the estimates.

mortality with uncertainty by age, sex, country or territory, and year. CODEm is the framework used to model most cause-specific death rates in GBD.1 First, all available data were identified and gathered. Second, a diverse set of plausible models were developed to capture well-documented associations in the estimates. The use of a wide range of individual models to create an ensemble predictive model has been shown to outperform techniques that use only a single model, both in cause-of-death estimation and in more general prediction applications.25 Third, the out-of-sample predictive validity was assessed for all individual models, which were then ranked for use in the ensemble modelling stage. Finally, differently weighted combinations of individual models were evaluated to select the ensemble model with the highest out-of-sample predictive validity. The methods we used for propagating uncertainty are the same as those used in previous GBD iterations. At each step of the computation process, we took 1000 draws and then computed final estimates using the mean estimate across the draws. 95% uncertainty intervals (UIs) were calculated as the 25th and 975th ranked values across all 1000 draws.

ropagating uncertainty are the same as those used in previous GBD iterations. At each step of the computation process, we took 1000 draws and then computed final estimates using the mean estimate across the draws. 95% uncertainty intervals (UIs) were calculated as the 25th and 975th ranked values across all 1000 draws. In GBD 2017, NASH was included as a fifth cause of cirrhosis for the first time. Cirrhosis caused by NASH was previously nested within the category of other cirrhosis. As discussed previously, we were able to use the cause-of-death database only for estimating overall cirrhosis and not its five causes. Instead, we used DisMod-MR 2.1 to split the parent cirrhosis mortality estimates and used liver cancer aetiological proportion models as covariates. These covariates included alcohol consumption (L per capita), HBsAg seroprevalence, hepatitis C (anti-hepatitis C virus antibody) seroprevalence, and obesity. Proportions from the five aetiological models were then rescaled to sum to one at the draw level and used to split the parent cirrhosis mortality estimates.

tes. These covariates included alcohol consumption (L per capita), HBsAg seroprevalence, hepatitis C (anti-hepatitis C virus antibody) seroprevalence, and obesity. Proportions from the five aetiological models were then rescaled to sum to one at the draw level and used to split the parent cirrhosis mortality estimates. Morbidity estimates We modelled total cirrhosis prevalence and decompensated and compensated cirrhosis on the basis of hospital data and claims data, using different definitions for each model. Total cirrhosis prevalence was defined as any diagnosis of cirrhosis. The distinction between decompensated and compensated cirrhosis was based on the diagnosis made upon hospital admission. If cirrhosis was the primary diagnosis upon hospital admission, we considered the case to be decompensated. If cirrhosis was diagnosed in outpatient settings or was detected among patients admitted to the hospital for other causes, we considered the case to be compensated.

on the diagnosis made upon hospital admission. If cirrhosis was the primary diagnosis upon hospital admission, we considered the case to be decompensated. If cirrhosis was diagnosed in outpatient settings or was detected among patients admitted to the hospital for other causes, we considered the case to be compensated. A systematic review of the literature was done to identify studies on the proportion of cirrhosis cases attributable to alcohol-related liver disease, hepatitis B, hepatitis C, NASH, and other causes. We re-extracted all literature within our databases on cirrhosis cause proportions for cirrhosis due to NASH. We searched the peer-reviewed literature via PubMed and solicited sources from GBD 2017 collaborators. The inclusion criteria stipulated that the publication year was 1980 or later; the sample was representative of patients with decompensated cirrhosis (eg, studies of patients with both hepatocellular carcinoma and hepatitis were excluded); sufficient information was provided on study method and sample characteristics to assess the quality of the study; and hepatitis B and C were confirmed via HBsAg in the case of hepatitis B, and anti-hepatitis C virus antibody in the case of hepatitis C. The number of site-years of data by cause is provided in the appendix (p 4).

information was provided on study method and sample characteristics to assess the quality of the study; and hepatitis B and C were confirmed via HBsAg in the case of hepatitis B, and anti-hepatitis C virus antibody in the case of hepatitis C. The number of site-years of data by cause is provided in the appendix (p 4). We modelled cirrhosis prevalence using hospital data and cause-specific mortality rates, assuming no remission.20 Similar to the mortality estimates, we developed aetiological proportion models using DisMod-MR 2.1 and used the results of these models to split the parent total cirrhosis prevalence estimates. We adopted the aetiological proportion model from liver cancer. We first developed five single-parameter DisMod models, each to estimate the proportion of liver cancer due to a given cause (ie, alcohol, hepatitis B, hepatitis C, NASH, and other). Estimates from these liver cancer models were then used as covariates in the five corresponding cirrhosis aetiological models. Proportions from the five aetiological models were then rescaled to sum to one at the draw level and used to split the parent cirrhosis estimates. We multiplied these fractions by prevalence separately for decompensated and compensated cirrhosis by age, sex, year, and location, and estimated the prevalence due to cirrhosis by each cause. Upon calculation of cause fractions, we encountered some reports on cirrhosis due to more than one cause. In these cases, we redistributed the numbers proportionally on the number of cases diagnosed with only one cause.

rhosis by age, sex, year, and location, and estimated the prevalence due to cirrhosis by each cause. Upon calculation of cause fractions, we encountered some reports on cirrhosis due to more than one cause. In these cases, we redistributed the numbers proportionally on the number of cases diagnosed with only one cause. To calculate YLDs, prevalence was multiplied by a disability weight, which represents the magnitude of the health loss associated with disease. Disability weights are measured on a scale from 0 to 1, where 0 is a state of full health and 1 is death. Compensated cirrhosis has a disability weight of 0 because of its asymptomatic nature. Decompensated cirrhosis from any cause has a disability weight of 0·178 (95% UI 0·122–0·250).20 We used the Socio-demographic Index (SDI) to determine the relationship between the development level of a region or country and cirrhosis mortality, prevalence, and DALYs. In GBD 2017, the SDI was revised to better reflect the development status of each country. The SDI ranges from 0 (worst) to 1 (best) and is a composite measure of the total fertility rate in women under the age of 25 years, mean education for individuals aged 15 years and older, and lag-distributed income per capita.

. In GBD 2017, the SDI was revised to better reflect the development status of each country. The SDI ranges from 0 (worst) to 1 (best) and is a composite measure of the total fertility rate in women under the age of 25 years, mean education for individuals aged 15 years and older, and lag-distributed income per capita. All estimates are reported in terms of counts, rates per 100 000 population, and percentages. Population estimates independently produced by GBD 2017 were used as references for calculating age-standardised death rates and age-standardised prevalence.26 95% UIs, including all sources of uncertainty arising from measurement error, systematic biases, and modelling, are reported for all estimates. All estimates of mortality, prevalence, DALYs, YLDs, and YLLs are reported by sex and age, and across location and time period studied (1990–2017). Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or the writing of the report. The corresponding authors had full access to the data in the study and had final responsibility for the decision to submit for publication.

All estimates are reported in terms of counts, rates per 100 000 population, and percentages. Population estimates independently produced by GBD 2017 were used as references for calculating age-standardised death rates and age-standardised prevalence.26 95% UIs, including all sources of uncertainty arising from measurement error, systematic biases, and modelling, are reported for all estimates. All estimates of mortality, prevalence, DALYs, YLDs, and YLLs are reported by sex and age, and across location and time period studied (1990–2017). Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or the writing of the report. The corresponding authors had full access to the data in the study and had final responsibility for the decision to submit for publication. Results Globally, cirrhosis caused more than 1·32 million (95% UI 1·27–1·45) deaths in 2017, with 440 000 deaths (416 000–518 000; 33·3%) in females and 883 000 (838 000–967 000; 66·7%) in males, compared with less than 899 000 (829 000–948 000) deaths for both sexes in 1990 (figure 1; appendix p 5). These deaths constituted 2·4% (2·3–2·6) of all deaths globally in 2017 compared with 1·9% (1·8–2·0) in 1990. The age-standardised death rate at the global level decreased from 21·0 (19·2–22·3) per 100 000 population in 1990 to 16·5 (15·8–18·1) per 100 000 population in 2017 (figure 1). The age-standardised death rate was consistently lowest in the high-income super-region (10·1 [9·8–10·5] per 100 000 in 2017) and highest in the sub-Saharan Africa super-region (32·2 [25·8–38·6] per 100 000 in 2017) in all years from 1990 to 2017 (appendix p 237). The pattern of age-standardised death rates between 1990 and 2017 was similar in males and females, although the rates were consistently higher in males than females in all super-regions (estimates available through the GBD online results tool and data visualisation tool) and all years from 1990 to 2017 (figure 1).Figure 1 Counts and age-standardised rates of cirrhosis death at the global level by sex, 1990–2017

les, although the rates were consistently higher in males than females in all super-regions (estimates available through the GBD online results tool and data visualisation tool) and all years from 1990 to 2017 (figure 1).Figure 1 Counts and age-standardised rates of cirrhosis death at the global level by sex, 1990–2017 Error bars indicate 95% UIs for number of deaths. Shading indicates 95% UIs for age-standardised death rates. UI=uncertainty interval. Cirrhosis led to nearly 41·4 million (95% UI 39·6–45·1) DALYs in 2017, which was an increase from just over 30·5 million (28·6–32·2) in 1990. The age-standardised rate of DALYs decreased from 656·4 (612·8–689·2) per 100 000 population in 1990 to 510·7 (487·6–557·1) per 100 000 population in 2017 (appendix p 15). In 2017, cirrhosis caused 28·8 million DALYs (27·3–31·4) in males and 12·6 million DALYs (11·9–14·7) in females, compared with 20·6 million DALYs (18·7–21·8) in males and 9·9 million (9·1–11·1) DALYs in females in 1990. The age-standardised DALY rates in 2017 were 719·3 (683·0–784·9) per 100 000 population in males and 307·6 (288·5–359.6) per 100 000 in females, which was a substantial decrease from 1990 (903·1 [820·1–956·1] per 100 000 in males and 415·5 [382·5–457·1] per 100 000 in females in 1990; estimates available through the GBD online results tool).

LY rates in 2017 were 719·3 (683·0–784·9) per 100 000 population in males and 307·6 (288·5–359.6) per 100 000 in females, which was a substantial decrease from 1990 (903·1 [820·1–956·1] per 100 000 in males and 415·5 [382·5–457·1] per 100 000 in females in 1990; estimates available through the GBD online results tool). The number of prevalent cases of decompensated cirrhosis globally increased from more than 5·20 million (95% UI 5·08–5·32) in 1990 to over 10·6 million (10·3–10·9) in 2017 (appendix p 23), of which 6·42 million (6·23–6·60; 60·3%) prevalent cases were in males and 4·23 million (4·10–4·35; 39·7%) were in females (data not shown). The age-standardised prevalence of decompensated cirrhosis increased from 110·6 (108·0–113·0) per 100 000 population in 1990 to 132·5 (128·6–136·2) per 100 000 population in 2017 (appendix p 23). In males, the age-standardised prevalence in 2017 was 162·7 (157·9–167·3) per 100 000, with a 21·1% (19·7–22·4) increase from 1990 to 2017. In females, the age-standardised prevalence in 2017 was 103·5 (100·6–106·3) per 100 000, with an 18·1% (16·9–19·2) increase from 1990 to 2017 (data not shown).

017 (appendix p 23). In males, the age-standardised prevalence in 2017 was 162·7 (157·9–167·3) per 100 000, with a 21·1% (19·7–22·4) increase from 1990 to 2017. In females, the age-standardised prevalence in 2017 was 103·5 (100·6–106·3) per 100 000, with an 18·1% (16·9–19·2) increase from 1990 to 2017 (data not shown). The number of global prevalent cases of compensated cirrhosis increased from 65·9 million (95% UI 63·4–68·7) in 1990 to just over 112 million (107–119) in 2017 (appendix p 32), of which 66·1 million (62·7–69·7; 58·8%) prevalent cases were in males and 46·3 million (43·8–48·8; 41·2%) were in females (data not shown). The age-standardised prevalence of compensated cirrhosis increased from 1354·5 (1300·6–1411·7) per 100 000 in 1990 to 1395·0 (1323·5–1470·5) in 2017 (appendix p 32). The age-standardised prevalence of compensated cirrhosis in males was 1605·4 (95% UI 1543·3–1671·2) per 100 000 in 1990 and 1651·4 (1567·5–1742·8) per 100 000 in 2017. The age-standardised prevalence of compensated cirrhosis in females was 1103·3 (1055·6–1152·4) per 100 000 in 1990 and 1142·4 (1079·9–1204·0) per 100 000 in 2017 (data not shown). The percentage change in age-standardised prevalence rate of compensated cirrhosis from 1990 to 2017 was 2·9% (0·4–5·2) in males and 3·5% (1·3–5·6) in females (data not shown).

ompensated cirrhosis in females was 1103·3 (1055·6–1152·4) per 100 000 in 1990 and 1142·4 (1079·9–1204·0) per 100 000 in 2017 (data not shown). The percentage change in age-standardised prevalence rate of compensated cirrhosis from 1990 to 2017 was 2·9% (0·4–5·2) in males and 3·5% (1·3–5·6) in females (data not shown). At the regional level, central Asia had the highest age-standardised death rate due to cirrhosis in 2017 for males, females, and both sexes combined (39·0 [95% UI 36·2–41·5] per 100 000 for both sexes combined; figure 2A), with deaths largely driven by alcohol-related liver disease (36·6% of all cirrhosis deaths). Western, eastern, and central sub-Saharan Africa had the next highest age-standardised death rates for both sexes combined in 2017, with rates of 35·8 (23·9–49·9) per 100 000 population in western Africa, 34·8 (26·4–42·2) per 100 000 population in eastern Africa, and 34·3 (25·9–45·2) per 100 000 population in central sub-Saharan Africa (figure 2A). In central and eastern sub-Saharan Africa in 2017, the most common cause of death due to cirrhosis was hepatitis C (32·4% and 28·9%, respectively), followed by hepatitis B (31·2% and 25·9%, respectively; appendix p 41). However, in western sub-Saharan Africa, the most common cause of death due to cirrhosis was hepatitis B (48·9%) in all countries in the region, whereas the proportion of deaths from hepatitis C was much lower in this region (7·8%) than in the other sub-Saharan African regions (appendix p 41).Figure 2 Age-standardised rates for cirrhosis by region and sex, 2017

e most common cause of death due to cirrhosis was hepatitis B (48·9%) in all countries in the region, whereas the proportion of deaths from hepatitis C was much lower in this region (7·8%) than in the other sub-Saharan African regions (appendix p 41).Figure 2 Age-standardised rates for cirrhosis by region and sex, 2017 (A) Age-standardised death rate. (B) Age-standardised prevalence rate of compensated and decompensated cirrhosis. Error bars indicate 95% uncertainty intervals for age-standardised rates. Southeast Asia ranked fifth in terms of age-standardised death rate from cirrhosis across regions in 2017 (29·5 [95% UI 27·7–31·9] per 100 000) but had low age-standardised prevalence rates of compensated and decompensated cirrhosis (figure 2A, B); both deaths and prevalent cases were mainly caused by hepatitis B and C (appendix pp 41, 48, 55).

h in terms of age-standardised death rate from cirrhosis across regions in 2017 (29·5 [95% UI 27·7–31·9] per 100 000) but had low age-standardised prevalence rates of compensated and decompensated cirrhosis (figure 2A, B); both deaths and prevalent cases were mainly caused by hepatitis B and C (appendix pp 41, 48, 55). Eastern Europe ranked sixth in terms of age-standardised death rate due to cirrhosis in 2017 (25·9 [95% UI 25·0–26·7] per 100 000), and, as in central Asia, deaths were primarily caused by alcohol-related liver disease (figure 2A; appendix pp 5, 41). These two regions were the only ones in which age-standardised death rates significantly increased over the study period: by 137·0% (129·5–146·2) in eastern Europe and by 50·5% (38·6–59·9) in central Asia (appendix p 5). The age-standardised death rate increased by more than 100% in all but one country (Moldova) in eastern Europe (appendix p 5). Eastern Europe had a high and increasing age-standardised prevalence of both decompensated and compensated cirrhosis from 1990 to 2017; prevalent cases were predominantly caused by alcohol-related liver disease (figure 2B; appendix pp 48, 55).

than 100% in all but one country (Moldova) in eastern Europe (appendix p 5). Eastern Europe had a high and increasing age-standardised prevalence of both decompensated and compensated cirrhosis from 1990 to 2017; prevalent cases were predominantly caused by alcohol-related liver disease (figure 2B; appendix pp 48, 55). The age-standardised death rate was lowest in Australasia (5·4 [95% UI 4·9–6·0] per 100 000 population), east Asia (8·3 [7·6–10·7] per 100 000 population), and high-income Asia Pacific (8·6 [7·9–9·1] per 100 000 population; figure 2A). Australasia was among the regions with the lowest age-standardised prevalence of both compensated and decompensated cirrhosis (figure 2B). Deaths due to cirrhosis and prevalent cases of compensated and decompensated cirrhosis in this region were primarily caused by hepatitis C (appendix pp 48, 55). East Asia had the second-lowest age-standardised death rate, but it had a relatively high age-standardised prevalence rate for compensated and decompensated cirrhosis combined (figure 2A, B). Deaths due to cirrhosis and prevalent cases of compensated and decompensated cirrhosis in this region were driven primarily by hepatitis B (appendix pp 41, 48, 55).

standardised death rate, but it had a relatively high age-standardised prevalence rate for compensated and decompensated cirrhosis combined (figure 2A, B). Deaths due to cirrhosis and prevalent cases of compensated and decompensated cirrhosis in this region were driven primarily by hepatitis B (appendix pp 41, 48, 55). High-income Asia Pacific had the third lowest age-standardised death rate in 2017 and all countries in this region experienced a steep decline over the study period, particularly South Korea (figure 2A; appendix p 5). However, this region had the highest age-standardised prevalence for both compensated cirrhosis (2455·0 [95% UI 2344·9–2575·8] per 100 000) and decompensated cirrhosis (267·4 [259·8–275·1] per 100 000) in 2017 (figure 2B). The high age-standardised prevalence in this region was primarily driven by high numbers of prevalent cases in Japan, where the highest proportions of compensated and decompensated cirrhosis were due to hepatitis C (appendix pp 23, 32, 48, 55).

ompensated cirrhosis (267·4 [259·8–275·1] per 100 000) in 2017 (figure 2B). The high age-standardised prevalence in this region was primarily driven by high numbers of prevalent cases in Japan, where the highest proportions of compensated and decompensated cirrhosis were due to hepatitis C (appendix pp 23, 32, 48, 55). Western Europe, southern sub-Saharan Africa, and high-income North America had the fourth to sixth lowest age-standardised death rates of cirrhosis in 2017 (figure 2A; appendix p 5). In western Europe, deaths were primarily caused by alcohol-related liver disease (41·7%), whereas in southern sub-Saharan Africa and high-income North America, deaths were primarily caused by hepatitis C (33·9% and 34·7%, respectively; appendix p 41). Western Europe and southern sub-Saharan Africa had modest age-standardised prevalence rates of both compensated and decompensated cirrhosis, whereas high-income North America had the lowest age-standardised prevalence of compensated cirrhosis of all regions; prevalent cases in this region were mainly caused by hepatitis C (figure 2B; appendix pp 48, 55). South Asia and Oceania had modest age-standardised death rates but were among the regions with the lowest age-standardised prevalence of compensated and decompensated cirrhosis (figure 2A, B). Deaths due to cirrhosis and prevalent cases of compensated and decompensated cirrhosis were primarily caused by hepatitis B in these regions (appendix p 41).

est age-standardised death rates but were among the regions with the lowest age-standardised prevalence of compensated and decompensated cirrhosis (figure 2A, B). Deaths due to cirrhosis and prevalent cases of compensated and decompensated cirrhosis were primarily caused by hepatitis B in these regions (appendix p 41). Central Latin America was among the regions with the highest age-standardised prevalence of both compensated (2272·3 [95% UI 2167·0–2383·4] per 100 000) and decompensated (206·6 [200·4–212·6] per 100 000) cirrhosis (figure 2B; appendix pp 23, 32), although age-standardised death rates were modest in this region. Other regions in Latin America had modest age-standardised death rates and age-standardised prevalence rates of both types of cirrhosis. Cirrhosis deaths in Latin American regions were primarily driven by alcohol-related liver disease, except for tropical Latin America (38·1% in Andean Latin America, 36·6% in central Latin America, 34·8% in the Caribbean, and 34·6% in southern Latin America; appendix p 41). Cirrhosis deaths in tropical Latin America were mainly attributable to hepatitis C. In every Latin American region, the cause of the highest proportion of cirrhosis deaths was also the cause of the highest proportion of compensated and decompensated cirrhosis prevalent cases (appendix pp 48, 55). Finally, north Africa and the Middle East had modest age-standardised death and prevalence rates from cirrhosis, with the largest proportion of deaths due to cirrhosis and prevalent cases of compensated and decompensated cirrhosis driven by hepatitis B and C.

Central Latin America was among the regions with the highest age-standardised prevalence of both compensated (2272·3 [95% UI 2167·0–2383·4] per 100 000) and decompensated (206·6 [200·4–212·6] per 100 000) cirrhosis (figure 2B; appendix pp 23, 32), although age-standardised death rates were modest in this region. Other regions in Latin America had modest age-standardised death rates and age-standardised prevalence rates of both types of cirrhosis. Cirrhosis deaths in Latin American regions were primarily driven by alcohol-related liver disease, except for tropical Latin America (38·1% in Andean Latin America, 36·6% in central Latin America, 34·8% in the Caribbean, and 34·6% in southern Latin America; appendix p 41). Cirrhosis deaths in tropical Latin America were mainly attributable to hepatitis C. In every Latin American region, the cause of the highest proportion of cirrhosis deaths was also the cause of the highest proportion of compensated and decompensated cirrhosis prevalent cases (appendix pp 48, 55). Finally, north Africa and the Middle East had modest age-standardised death and prevalence rates from cirrhosis, with the largest proportion of deaths due to cirrhosis and prevalent cases of compensated and decompensated cirrhosis driven by hepatitis B and C. Specific country and territory data can be found in the appendix (pp 5–14 for age-standardised death rates; pp 15–22 for DALYs, pp 23–40 for prevalence, and pp 41–68 for cause). Egypt had the highest age-standardised death rate of cirrhosis in all years from 1990 to 2017, despite a 22·4% (95% UI 1·5–42·1) decrease from 1990 (133·1 [76·4–150·8] per 100 000) to 2017 (103·3 [64·4–133·4] per 100 000). The second highest rate was in Rwanda in 1990 (88·1 [71·9–104·8] per 100 000) and in Cambodia in 2017 (79·4 [67·4–96·1] per 100 000; figure 3A; appendix p 5). In 2017, 41·5% of deaths due to cirrhosis in Egypt were caused by hepatitis B and 34·4% were caused by hepatitis C (appendix p 41).Figure 3 Age-standardised death rate for cirrhosis in 2017 (A), age-standardised prevalence for decompensated cirrhosis (B), and age-standardised prevalence for compensated cirrhosis (C), 2017

2017, 41·5% of deaths due to cirrhosis in Egypt were caused by hepatitis B and 34·4% were caused by hepatitis C (appendix p 41).Figure 3 Age-standardised death rate for cirrhosis in 2017 (A), age-standardised prevalence for decompensated cirrhosis (B), and age-standardised prevalence for compensated cirrhosis (C), 2017 ATG=Antigua and Barbuda. FSM=Federated States of Micronesia. Isl=Islands. LCA=Saint Lucia. TLS=Timor-Leste. TTO=Trinidad and Tobago. VCT=Saint Vincent and the Grenadines. The lowest age-standardised death rate of cirrhosis in 2017 was in Singapore (3·7 [95% UI 3·3–4·0] per 100 000 population). The largest increase in age-standardised death rate from 1990 to 2017 was in Lithuania (176·7% [154·1–203·2]; appendix p 5). The eight countries with the largest increases in age-standardised death rate from cirrhosis over the study period (in order: Lithuania, Ukraine, Belarus, Russia, Kazakhstan, Estonia, Latvia, and Armenia) were all in eastern Europe or central Asia (appendix p 5). In all of these countries, the largest proportion of deaths in 2017 was caused by alcohol-related liver disease (appendix p 41).

rate from cirrhosis over the study period (in order: Lithuania, Ukraine, Belarus, Russia, Kazakhstan, Estonia, Latvia, and Armenia) were all in eastern Europe or central Asia (appendix p 5). In all of these countries, the largest proportion of deaths in 2017 was caused by alcohol-related liver disease (appendix p 41). Among the ten countries with the highest age-standardised prevalence of decompensated cirrhosis in 2017, eight were from central and eastern Europe (in order: Slovakia [349·6, 95% UI 338·4–361·2] per 100 000, Moldova, Romania, Poland, Ukraine, Lithuania, Albania, and Hungary), in addition to Japan (third-highest; 279·8 [272·0–287·8] per 100 000) and South Korea (eighth-highest; 245·9 [237·8–253·9] per 100 000; figure 3B; appendix p 23). The Philippines had the lowest age-standardised prevalence of decompensated cirrhosis in 2017 (41·8 [40·2–43·4 per 100 000 population]), with the largest proportion of prevalent cases due to hepatitis B. Despite decreasing age-standardised prevalence of compensated cirrhosis since 1990, South Korea and Japan were among the ten countries with the highest rates in 2017 (figure 3C; appendix p 32). The three countries or territories with the highest age-standardised prevalence of compensated cirrhosis were Moldova, Taiwan (Province of China), and Slovakia.

d prevalence of compensated cirrhosis since 1990, South Korea and Japan were among the ten countries with the highest rates in 2017 (figure 3C; appendix p 32). The three countries or territories with the highest age-standardised prevalence of compensated cirrhosis were Moldova, Taiwan (Province of China), and Slovakia. Globally, in 2017, 31·5% of cirrhosis deaths in males were caused by hepatitis B, 25·5% were caused by hepatitis C, 27·3% were caused by alcohol-related liver disease, 7·7% were caused by NASH, and 8·0% resulted from other causes. In females, the proportions of cirrhosis deaths caused by hepatitis B (24·0%) and alcohol-related liver disease (20·6%) were lower than in males, the proportion caused by hepatitis C (26·7%) was similar to that in males, and the proportion caused by NASH (11·3%) and by other causes (17·3%) was higher than in males (figure 4).Figure 4 Proportion of deaths due to five causes of cirrhosis at global and regional levels by sex, 2017

(20·6%) were lower than in males, the proportion caused by hepatitis C (26·7%) was similar to that in males, and the proportion caused by NASH (11·3%) and by other causes (17·3%) was higher than in males (figure 4).Figure 4 Proportion of deaths due to five causes of cirrhosis at global and regional levels by sex, 2017 Hepatitis B caused around 287 000 (95% UI 252 000–318 000) deaths due to cirrhosis in 1990, which increased to almost 384 000 (349 000–442 000) in 2017. 278 000 (252 000–321 000; 72·5%) of these deaths in 2017 occurred in males, and 106 000 (94 000–130 000; 27·5%) occurred in females (appendix p 69). However, the age-standardised death rate for both sexes combined decreased steadily from 6·7 (5·8–7·4) per 100 000 in 1990 to 4·8 (4·3–5·5) per 100 000 in 2017 (figure 5; appendix p 69). The age-standardised death rate of cirrhosis due to hepatitis B was 7·2 (6·5–8·3) per 100 000 in males and 2·5 (2·2–3·1) per 100 000 in females in 2017. Hepatitis B caused 36·7 million (95% UI 34·1–39·5) prevalent cases of compensated cirrhosis and 2·97 million (2·81–3·12) cases of decompensated cirrhosis in 2017. The age-standardised prevalence of decompensated cirrhosis caused by hepatitis B increased from 30·9 (95% UI 29·3–32·2) per 100 000 in 1990 to 36·6 (34·7–38·4) per 100 000 in 2017 (appendix p 139). The age-standardised prevalence of compensated cirrhosis caused by hepatitis B did not change notably from 1990 (461·8 [435·8–491·2] per 100 000) to 2017 (451·9 [95% UI 420·0–485·9] per 100 000; appendix p 184). Across regions, the proportion of deaths from cirrhosis due to hepatitis B was highest in western sub-Saharan Africa (48·9%) and lowest in high-income North America (5·7%; figure 4). The proportion of cirrhosis mortality that was caused by hepatitis B ranged from 5·7% in the USA to 54·8% in Chad (appendix pp 41).Figure 5 Number of deaths and age-standardised death rates at the global level by cause of cirrhosis, 1990–2017

aran Africa (48·9%) and lowest in high-income North America (5·7%; figure 4). The proportion of cirrhosis mortality that was caused by hepatitis B ranged from 5·7% in the USA to 54·8% in Chad (appendix pp 41).Figure 5 Number of deaths and age-standardised death rates at the global level by cause of cirrhosis, 1990–2017 Bars refer to number of deaths in each year. Lines refer to age-standardised death rate in each year. NASH=non-alcoholic steatohepatitis.

aran Africa (48·9%) and lowest in high-income North America (5·7%; figure 4). The proportion of cirrhosis mortality that was caused by hepatitis B ranged from 5·7% in the USA to 54·8% in Chad (appendix pp 41).Figure 5 Number of deaths and age-standardised death rates at the global level by cause of cirrhosis, 1990–2017 Bars refer to number of deaths in each year. Lines refer to age-standardised death rate in each year. NASH=non-alcoholic steatohepatitis. Hepatitis C caused more than 225 000 (95% UI 202 000–249 000) cirrhosis deaths in 1990, which increased to over 342 000 (313 000–381 000) in 2017. 225 000 (203 000–251 000; 65·8%) of these deaths in 2017 occurred in males, and 117 000 (107 000–135 000; 34·2%) occurred in females (appendix p 83). The age-standardised death rate decreased from 5·3 (4·8–5·9) per 100 000 in 1990 to 4·2 (3·9–4·7) per 100 000 in 2017 (figure 5; appendix p 83). The age-standardised death rate due to hepatitis C was 5·8 (5·3–6·5) per 100 000 in males and 2·8 (2·5–3·2) in females in 2017. Hepatitis C caused 27·72 million (25·52–30·00) cases of compensated cirrhosis and 2·64 million (2·49–2·81) cases of decompensated cirrhosis in 2017. The age-standardised prevalence of compensated cirrhosis caused by hepatitis C was 327·0 (303·9–349·7) per 100 000 in 1990 and 341·1 (314·1–368·7) per 100 000 in 2017. The age-standardised prevalence of decompensated cirrhosis increased from 27·2 (25·7–28·7) per 100 000 in 1990 to 32·5 (30·6–34·5) per 100 000 in 2017 (appendix pp 148, 193). Generally, the lowest proportions of cirrhosis mortality due to hepatitis C were observed in countries in western sub-Saharan Africa (7·8%) and the highest among countries in high-income Asia-Pacific (41·3%; figure 4). The proportion of cirrhosis mortality that was caused by hepatitis C ranged from 7·1% in Niger to 55·7% in Tunisia (appendix p 41).

ons of cirrhosis mortality due to hepatitis C were observed in countries in western sub-Saharan Africa (7·8%) and the highest among countries in high-income Asia-Pacific (41·3%; figure 4). The proportion of cirrhosis mortality that was caused by hepatitis C ranged from 7·1% in Niger to 55·7% in Tunisia (appendix p 41). Alcohol-related liver disease caused just over 215 000 (95% UI 195 000–235 000) cirrhosis deaths in 1990, which increased to just over 332 000 (303 000–373 000) deaths in 2017. 241 000 (220 000–268 000; 72·6%) of these deaths in 2017 occurred in males, and 91 000 (82 000–110 000; 27·4%) occurred in females. The age-standardised death rate decreased from 5·1 (4·6–5·5) per 100 000 in 1990 to 4·1 (3·7–4·6) per 100 000 in 2017 (figure 5; appendix p 97). The age-standardised death rate due to alcohol-related liver disease was 6·2 (5·7–6·9) per 100 000 in males and 2·1 (1·9–2·6) per 100 000 in females in 2017. Alcohol-related liver disease caused 23·6 million (21·9–25·5) cases of compensated cirrhosis and 2·46 million (2·32–2·61) cases of decompensated cirrhosis in 2017. The age-standardised prevalence of decompensated cirrhosis caused by alcohol-related liver disease increased from 25·3 (23·9–26·7) per 100 000 in 1990 to 30·0 (28·2–31·8) per 100 000 in 2017. The age-standardised prevalence of compensated cirrhosis caused by alcohol-related liver disease did not change notably from 1990 (290·0 [271·9–309·9] per 100 000) to 2017 (288·1 [267·5–311·3] per 100 000; appendix pp 157, 202). Among GBD regions, the highest proportions of cirrhosis deaths due to alcohol-related liver disease were in central Europe (44·0%), western Europe (41·7%), and Andean Latin America (38·1%; figure 4; appendix p 41). The proportion of cirrhosis deaths due to alcohol-related liver disease was lowest in north Africa and the Middle East (5·3%) and in Egypt specifically (4·8%), and highest in Belgium (53·5%; appendix p 41).

ere in central Europe (44·0%), western Europe (41·7%), and Andean Latin America (38·1%; figure 4; appendix p 41). The proportion of cirrhosis deaths due to alcohol-related liver disease was lowest in north Africa and the Middle East (5·3%) and in Egypt specifically (4·8%), and highest in Belgium (53·5%; appendix p 41). NASH caused almost 61 900 (95% UI 55 400–68 000) cirrhosis deaths in 1990, which increased to around 118 000 (109 000–129 000) in 2017. 68 000 (62 000–75 000; 57·6%) of these deaths in 2017 occurred in males, and 50 000 (46 000–56 000; 42·4%) occurred in females. NASH was the only cause for which the age-standardised death rate did not decrease: the rate was 1·5 (1·3–1·6) per 100 000 in 1990 and 1·5 (1·3–1·6) per 100 000 in 2017 (figure 5; appendix p 111). The age-standardised death rate due to NASH was 1·8 (1·6–1·9) per 100 000 in males and 1·2 (1·1–1·3) per 100 000 in females in 2017. NASH caused 4·06 million (3·70–4·45) prevalent cases of compensated cirrhosis in 1990 and 9·42 million (8·57–10·34) cases in 2017 (more than doubling over the study period). NASH caused 325 000 (302 000–349 000) cases of decompensated cirrhosis in 1990 and 917 000 (850 000–986 000) cases in 2017 (nearly tripling over the study period). The age-standardised prevalence of compensated cirrhosis caused by NASH increased from 86·7 (79·0–94·6) per 100 000 in 1990 to 115·5 (105–126·5) per 100 000 in 2017, a 33·2% increase. The age-standardised prevalence of decompensated cirrhosis increased from 7·3 (6·8–7·8) per 100 000 in 1990 to 11·3 (10·4–12·1) per 100 000 in 2017, a 54·8% increase (appendix pp 166, 211). At the regional level, the proportion of deaths due to NASH was highest in Latin American regions, with the highest proportion in Tropical Latin America (22·6%) and Andean Latin America (22·2%). The lowest proportion was in high-income Asia Pacific (4·7%; figure 4; appendix p 41). At the national level, the proportion of deaths due to NASH was highest in Ecuador (25·2%) and lowest in Singapore (2·7%).

rican regions, with the highest proportion in Tropical Latin America (22·6%) and Andean Latin America (22·2%). The lowest proportion was in high-income Asia Pacific (4·7%; figure 4; appendix p 41). At the national level, the proportion of deaths due to NASH was highest in Ecuador (25·2%) and lowest in Singapore (2·7%). Finally, other causes led to almost 110 000 (95% UI 97 000–127 000) cirrhosis deaths in 1990, which increased to around 146 000 (131 000–165 000) in 2017. 70 000 (62 000–80 000; 48·0%) of these deaths in 2017 occurred in males, and 76 000 (67 000–90 000) occurred in females (52·0%). The age-standardised death rate decreased from 2·5 (2·2–2·8) per 100 000 in 1990 to 1·9 (1·7–2·1) per 100 000 in 2017 (figure 5; appendix p 125). Other causes led to 15·0 million (13·6–16·2) cases of compensated cirrhosis and 1·65 million (1·55–1·76) cases of decompensated cirrhosis in 2017. The age-standardised prevalence of decompensated cirrhosis due to other causes was 20·0 (18·9–21·1) per 100 000 population in 1990 and 22·2 (20·9–23·5) per 100 000 in 2017. The age-standardised prevalence of compensated cirrhosis due to other causes was 189·0 (173·0–204·6) per 100 000 population in 1990 and 198·4 (180·1–215·4) per 100 000 in 2017 (appendix pp 175, 220). At the regional level, the proportion of deaths due to other causes of cirrhosis was highest in high-income North America (20·6%) and lowest in east Asia (3·8%; figure 4; appendix p 41). At the national level, the highest proportion of deaths was in the UK (30·0%) and the lowest was in China (3·8%) and the United Arab Emirates (3·8%).

level, the proportion of deaths due to other causes of cirrhosis was highest in high-income North America (20·6%) and lowest in east Asia (3·8%; figure 4; appendix p 41). At the national level, the highest proportion of deaths was in the UK (30·0%) and the lowest was in China (3·8%) and the United Arab Emirates (3·8%). Patterns in burden by age were similar between males and females and across years from 1990 to 2017, although the numbers and rates were consistently higher in males than in females. The number of deaths peaked at 60–64 years in both sexes combined in the years 1990–2004 and 2011–17, and at 55–59 years for all other years (estimates available through the GBD results tool). In 2017, the number of deaths peaked in the 60–64 year age group for males and the 65–69 year age group for females, whereas death rates increased steadily with age in every year of the study period (figure 6A). In 2017, the number of DALYs peaked at 50–54 years in males and at 55–59 years in females, whereas DALY rates peaked at 60–64 years in males and 70–74 years in females (figure 6B). The number of prevalent cases of compensated cirrhosis peaked at 45–49 years in females and 40–44 years in males in 2017. The prevalence rate of compensated cirrhosis peaked at 45–49 years in males and at 50–54 years in females. The number of prevalent cases of decompensated cirrhosis peaked at 50–54 years in both sexes, whereas the prevalence rates increased up to 65–69 years for males and 75–79 years for females, decreased until 90–94 years for males before increasing again, and decreased through the oldest age groups for females (figure 6C).Figure 6 Age patterns of deaths, DALYs, and prevalence of cirrhosis by sex, 2017

both sexes, whereas the prevalence rates increased up to 65–69 years for males and 75–79 years for females, decreased until 90–94 years for males before increasing again, and decreased through the oldest age groups for females (figure 6C).Figure 6 Age patterns of deaths, DALYs, and prevalence of cirrhosis by sex, 2017 (A) Number of deaths and age-specific death rate per 100 000 population. (B) Number of DALYs and age-specific DALY rate per 100 000 population. (C) Number of prevalent cases and age-specific prevalence rates per 100 000 population of decompensated and compensated cirrhosis. Error bars indicate 95% UIs for number of deaths, DALYs, and prevalent cases. Shading indicates 95% UIs for death rates, DALY rates, and prevalence rates. DALYs=disability-adjusted life-years. UI=uncertainty interval.

s and age-specific prevalence rates per 100 000 population of decompensated and compensated cirrhosis. Error bars indicate 95% UIs for number of deaths, DALYs, and prevalent cases. Shading indicates 95% UIs for death rates, DALY rates, and prevalence rates. DALYs=disability-adjusted life-years. UI=uncertainty interval. Age patterns across regions are provided in the appendix (pp 241–61). In high-income North America in 2017, the number of deaths was highest in the age groups from 50 to 69 years for both sexes combined, with an increasing trend in both the number and rate of deaths in these age groups from 1990 to 2017. In both 1990 and 2017, death rates increased with age, up to and including the 95 years and older age group. The number of cirrhosis deaths in 2017 was also highest in the middle-aged age groups (50–74 years) in all other regions except Oceania (highest in the 40–55 year age groups) and high-income Asia Pacific (highest in the 75–84 year age groups). However, unlike high-income North America, these regions, with the exception of eastern Europe, all experienced a decreasing trend in death rates in the middle-aged age groups over the study period.

xcept Oceania (highest in the 40–55 year age groups) and high-income Asia Pacific (highest in the 75–84 year age groups). However, unlike high-income North America, these regions, with the exception of eastern Europe, all experienced a decreasing trend in death rates in the middle-aged age groups over the study period. The age patterns in central and eastern Europe were markedly different from other regions. Although the number of deaths in these regions peaked at 55–69 years in 2017, similar to other regions, death rates did not consistently increase with age. In 2017, death rates in central Europe declined from 70–74 years to 90–94 years, and in eastern Europe from 65–69 years to 80–84 years and increased again in the oldest age groups in both regions. In all other regions, death rates increased with increasing in age in all years from 1990 to 2017.

ently increase with age. In 2017, death rates in central Europe declined from 70–74 years to 90–94 years, and in eastern Europe from 65–69 years to 80–84 years and increased again in the oldest age groups in both regions. In all other regions, death rates increased with increasing in age in all years from 1990 to 2017. Between 1990 and 2017, the age-standardised death rate decreased or remained steady in all regions except central Asia and eastern Europe (figure 7A). In these two regions, the observed age-standardised death rates not only increased over the study period, but increased to much higher rates than those that would be expected based solely on SDI. The observed trend in these two regions was an increase in age-standardised death rate from 1990 to the late 2000s, then a slight decrease until 2017. Overall, the age-standardised death rate was lower at the regional level at higher SDI levels, with expected rates following a roughly linear decreasing trend until SDI equalled 0·70, an increase in age-standardised death rate until an SDI of 0·75, then a decline through the highest SDI levels. High and increasing rates in eastern Europe explain the rise in expected rates at 0·70 SDI. At the national level, age-standardised death rates were also lower at higher SDI levels in 2017 (appendix p 262).Figure 7 Age-standardised rates of cirrhosis globally and for 21 regions by SDI, 1990–2017

the highest SDI levels. High and increasing rates in eastern Europe explain the rise in expected rates at 0·70 SDI. At the national level, age-standardised death rates were also lower at higher SDI levels in 2017 (appendix p 262).Figure 7 Age-standardised rates of cirrhosis globally and for 21 regions by SDI, 1990–2017 (A) Age-standardised death rate per 100 000 population. (B) Age-standardised DALY rate per 100 0000 population. (C) Age-standardised prevalence per 100 000 population of decompensated cirrhosis. (D) Age-standardised prevalence per 100 000 population of compensated cirrhosis. For each region, points from left to right depict estimates from each year from 1990 to 2017. Black lines show the expected death, DALY, or prevalence rates on the basis of SDI alone. DALYs=disability-adjusted life-years. SDI=Socio-demographic Index. The patterns of age-standardised DALY rates versus SDI were similar to those of age-standardised death rates (figure 7B; appendix p 262). In most regions, the observed age-standardised death and DALY rates were close to their expected value on the basis of SDI. The exceptions were western sub-Saharan Africa, southeast Asia, central Asia, and central Latin America, where observed levels were higher than expected in every year of the study period.

p 262). In most regions, the observed age-standardised death and DALY rates were close to their expected value on the basis of SDI. The exceptions were western sub-Saharan Africa, southeast Asia, central Asia, and central Latin America, where observed levels were higher than expected in every year of the study period. Age-standardised prevalence of decompensated cirrhosis was slightly higher at higher SDI levels, although it decreased at an SDI of 0·45 (because of low age-standardised prevalence in Oceania and south Asia) and peaked at an SDI of 0·75 (because of high age-standardised prevalence in central and eastern Europe; figure 7C). Southern sub-Saharan Africa was the only region where age-standardised prevalence of decompensated cirrhosis decreased over time. The pattern of compensated cirrhosis versus SDI was similar to that of decompensated cirrhosis, although most regions showed little change in rate over time. The high-income Asia Pacific region showed a decline in age-standardised prevalence of compensated cirrhosis over the study period, but it was still much higher than expected based on SDI (figure 7D).

ersus SDI was similar to that of decompensated cirrhosis, although most regions showed little change in rate over time. The high-income Asia Pacific region showed a decline in age-standardised prevalence of compensated cirrhosis over the study period, but it was still much higher than expected based on SDI (figure 7D). Discussion In this paper, we described the results of GBD 2017 regarding cirrhosis mortality and morbidity by the four most common causes, and a fifth category of other causes. Of note, the four major causes of cirrhosis mortality and morbidity can generally be prevented by vaccination or lifestyle modification and are readily treatable if diagnosed early enough.7 It is essential for national health-care policy makers to know the death and prevalence rates of each cause to be able to plan and implement systematic interventions that prevent premature deaths and morbidity due to cirrhosis. The age-standardised death, DALY, and prevalence rates were universally lower in females than in males in both 1990 and 2017. Generally, females had a lower proportion of cirrhosis deaths caused by hepatitis B and alcohol-related liver disease than males, and had a higher proportion due to NASH and other causes. This trend could be driven by hormonal factors; lower prevalence of high-risk behaviours; higher prevalence of obesity;27 lower consumption of alcohol;28 and diseases that almost exclusively affect women, such as autoimmune hepatitis, which has a simple and effective treatment and low mortality if treated in time.29

s. This trend could be driven by hormonal factors; lower prevalence of high-risk behaviours; higher prevalence of obesity;27 lower consumption of alcohol;28 and diseases that almost exclusively affect women, such as autoimmune hepatitis, which has a simple and effective treatment and low mortality if treated in time.29 Our results show that in almost all regions, the number of deaths peaked in the middle-aged age groups (approximately 50–74 years) and rates increased steadily with increasing age. Central and eastern Europe were exceptions to these trends, and showed lower death rates in the older age groups (approximately 70–84 years) than in the middle-aged age groups throughout the study period. These findings suggest that cirrhosis incidence in these regions, where cirrhosis deaths are primarily driven by alcohol-related liver disease, was higher in the 1990s, coinciding with peak consumption of alcohol following the dissolution of the Soviet Union, than in 2017.28, 30 Among high-income regions, high-income North America showed an unusual age pattern, with an increase in death rates in the 50–69 year age group from 1990 to 2017. This finding is compatible with previous reports for the USA and Canada.31, 32

of alcohol following the dissolution of the Soviet Union, than in 2017.28, 30 Among high-income regions, high-income North America showed an unusual age pattern, with an increase in death rates in the 50–69 year age group from 1990 to 2017. This finding is compatible with previous reports for the USA and Canada.31, 32 Our results show that the numbers of deaths and DALYs due to cirrhosis increased globally between 1990 and 2017. The increase in numbers was primarily driven by population growth and ageing across the globe, specifically in low-income and middle-income countries.26 By contrast, age-standardised death and DALY rates decreased, and were lower at higher SDI levels. Accordingly, we observed high age-standardised death rates in low-income sub-Saharan African regions in 2017, despite a substantial decline since 1990, which is compatible with previous reports.33 However, we observed lower death rates in higher SDI countries and territories (eg, the lowest rate was in Singapore). These findings are expected because most cirrhosis-related deaths can be avoided in high-income countries through better access to health care and stronger health infrastructure. Notable exceptions are countries located in eastern Europe and central Asia, where, unlike in other regions, the age-standardised death rates increased from 1990 to 2017, primarily driven by alcohol-related liver disease, which has also been reported previously.34 Similarly, in Latin American regions, a high proportion of deaths were due to alcohol-related liver disease, although age-standardised death rates were not as high as those in eastern Europe and central Asia. These patterns closely follow the distribution of alcohol consumption in these regions.28, 30

ted previously.34 Similarly, in Latin American regions, a high proportion of deaths were due to alcohol-related liver disease, although age-standardised death rates were not as high as those in eastern Europe and central Asia. These patterns closely follow the distribution of alcohol consumption in these regions.28, 30 In western Europe, despite low age-standardised death rates, a high proportion of deaths were due to alcohol-related liver disease. This finding is consistent with previous evidence on high alcohol use in these regions28, 30, 34 and demonstrates a need for a systematic approach to reducing alcohol use in these countries. To this end, in 2010, the member states of WHO reached a consensus at the World Health Assembly on a global strategy to confront the harmful use of alcohol.35 High-income regions had low age-standardised rates of cirrhosis deaths. Most of these regions, including Australasia, high-income North America, and western Europe, also had low age-standardised prevalences of both compensated and decompensated cirrhosis as a result of concerted efforts to treat and prevent viral hepatitis.34, 36 The estimated age-standardised prevalence and death rates caused by cirrhosis in this study were slightly higher but generally compatible with previous reports from the USA and released by the US Centers for Disease Control and Prevention.37 The estimated age-standardised death rates in the UK were also consistent with previous reports.38

andardised prevalence and death rates caused by cirrhosis in this study were slightly higher but generally compatible with previous reports from the USA and released by the US Centers for Disease Control and Prevention.37 The estimated age-standardised death rates in the UK were also consistent with previous reports.38 The high-income Asia Pacific countries, especially Japan and South Korea, had a high but declining age-standardised prevalence rate of compensated cirrhosis, despite having low mortality rates.39 Prevalence trends could indicate a high incidence in preceding decades followed by a recent decline in mortality because of improved health care and access to affordable and effective antiviral treatment. Accordingly, in 2005, the WHO Western Pacific region set a goal to reduce HBsAg seroprevalence in children 5 years of age and older to less than 2% by 2012. A supranational goal was selected to create a sense of political urgency to establish routine immunisation and improve access to health care.39 A similar approach might be successful in African and other Asian countries, as well as in the region of north Africa and the Middle East, to control the incidence of hepatitis B; cirrhosis deaths in these regions were mainly due to hepatitis B and C.

establish routine immunisation and improve access to health care.39 A similar approach might be successful in African and other Asian countries, as well as in the region of north Africa and the Middle East, to control the incidence of hepatitis B; cirrhosis deaths in these regions were mainly due to hepatitis B and C. Southeast Asia had higher age-standardised death rates than expected, and south Asia and Oceania had modest death rates, but these three regions were among those with the lowest age-standardised prevalence of compensated and decompensated cirrhosis. Mortality was primarily due to hepatitis B in these regions. The low age-standardised prevalence of compensated cirrhosis in these regions might be because of a low detection rate, whereas the low age-standardised prevalence of decompensated cirrhosis might be because of a relatively poor medical infrastructure, leading to a short survival time once the cirrhosis starts to decompensate.40

e-standardised prevalence of compensated cirrhosis in these regions might be because of a low detection rate, whereas the low age-standardised prevalence of decompensated cirrhosis might be because of a relatively poor medical infrastructure, leading to a short survival time once the cirrhosis starts to decompensate.40 In north Africa and the Middle East, alcohol-related liver disease constituted the lowest proportion of age-standardised prevalence and death rates.19 This finding could be expected, because alcohol is prohibited in many of the countries in this region, which could lead to both decreased use and the possibility of under-reporting. Instead, hepatitis B and C constitute larger proportions. Despite a decrease in age-standardised death rates in this region, hepatitis B is still the major cause of cirrhosis deaths, and although most countries in these regions have nationwide infant vaccination programmes for hepatitis B, none have been in place for more than 30 years.41 Most of the mortality of hepatitis B is in patients older than 40 years, so we expect hepatitis B to remain a major cause of cirrhosis death for another decade or two, even in countries with good vaccine coverage.

ationwide infant vaccination programmes for hepatitis B, none have been in place for more than 30 years.41 Most of the mortality of hepatitis B is in patients older than 40 years, so we expect hepatitis B to remain a major cause of cirrhosis death for another decade or two, even in countries with good vaccine coverage. Hepatitis C, by contrast, now has an effective cure, which should have an impact much sooner. With the increased availability of cheap generic anti-hepatitis C virus medicines,42 we expect the death rate for hepatitis C to rapidly decrease in the near future, if countries are able to meet the goals set by WHO to eliminate the burden of hepatitis C by 2030.43 For example, in Egypt, which currently has the highest death rate due to hepatitis C,19 we expect a rapid decrease in the hepatitis C death rate within the next 5–10 years because of an active patient-finding and treatment programme enacted by the government in 2014.44 Australia has also been successful in finding and treating patients with hepatitis C and will probably have very low hepatitis C death rates in the near future.45

se in the hepatitis C death rate within the next 5–10 years because of an active patient-finding and treatment programme enacted by the government in 2014.44 Australia has also been successful in finding and treating patients with hepatitis C and will probably have very low hepatitis C death rates in the near future.45 Between 1990 and 2017, we observed a doubling in the number of deaths, a more than tripling in prevalent cases of decompensated cirrhosis, and a more than doubling in prevalent cases of compensated cirrhosis due to NASH. Unlike the other four causes studied, NASH was the only one not to show a decreasing trend in age-standardised death rates. The highest proportions of deaths due to NASH were in Latin America and north Africa and the Middle East, and the lowest was in high-income Asia Pacific, as previously reported.46 The epidemiology of NASH closely follows the distribution of overweight and obesity and components of metabolic syndrome in these regions.27, 47 Because of the absence of practical means of preventing or treating NASH and a global increase in metabolic syndrome and obesity, we should expect an increase in the burden of this disease.16, 17

NASH closely follows the distribution of overweight and obesity and components of metabolic syndrome in these regions.27, 47 Because of the absence of practical means of preventing or treating NASH and a global increase in metabolic syndrome and obesity, we should expect an increase in the burden of this disease.16, 17 Despite the availability of an effective vaccine for hepatitis B for decades, the availability of an effective treatment for hepatitis C, and an increase in obesity worldwide, the proportional contribution of different causes to cirrhosis remained fairly constant between 1990 and 2017 at the global level. The development of cirrhosis and its progression to decompensation takes decades, and it might still be too early to see the effects of vaccination coverage or increased prevalence of obesity and metabolic syndrome.

ion of different causes to cirrhosis remained fairly constant between 1990 and 2017 at the global level. The development of cirrhosis and its progression to decompensation takes decades, and it might still be too early to see the effects of vaccination coverage or increased prevalence of obesity and metabolic syndrome. The limitations of the methods of GBD impose biases on our estimates in the current study, as with all GBD research. The most important limitation is the low availability and low quality of data, although we used robust statistical methods to overcome data scarcity in countries with low data availability. In these locations, we relied on predictive covariates, trends in neighbouring countries, or a combination of these. The wide variation in the availability of high-quality data across locations is reflected in the uncertainty associated with all estimates. As a result of lower availability of population-based data on the prevalence of cirrhosis, dependence on hospital and claims data, and various definitions and diagnostic criteria for decompensated and compensated cirrhosis, it is possible that both types, especially compensated cirrhosis, have been underestimated. We made the assumption that patients admitted to hospitals had decompensated cirrhosis and that total cases detected upon admission constituted all cases of cirrhosis. Of note, although ICD-10 codes have been validated for overall cirrhosis and chronic liver diseases, detailed codes for the five causes have not been fully validated. To address this challenge, we combined the aetiological models for cirrhosis and liver cancer, as these two diseases share the same causes. Aetiological proportion models of liver cancer were used as covariates for cirrhosis aetiological proportion models. Finally, advances in imaging modalities and the availability of liver biopsy as the gold standard for detecting cirrhosis in population-based studies can considerably improve the accuracy of estimations.

Aetiological proportion models of liver cancer were used as covariates for cirrhosis aetiological proportion models. Finally, advances in imaging modalities and the availability of liver biopsy as the gold standard for detecting cirrhosis in population-based studies can considerably improve the accuracy of estimations. Further work is needed to disentangle the separate entities that were categorised as other causes in this study, including autoimmune hepatitis as a sixth cause for cirrhosis. A detailed study on the epidemiology of NASH, particularly in north Africa and the Middle East and Latin America, is needed because it disproportionately affects age-standardised death rate as compared with other causes. Future studies should focus on evaluating common risk factors, dietary patterns, and screening practices affecting the incidence, mortality, and prevalence of cirrhosis by age, sex, and race, especially among countries and territories with the highest numbers of deaths. The financial burden of cirrhosis also merits attention in future studies.

us on evaluating common risk factors, dietary patterns, and screening practices affecting the incidence, mortality, and prevalence of cirrhosis by age, sex, and race, especially among countries and territories with the highest numbers of deaths. The financial burden of cirrhosis also merits attention in future studies. In conclusion, the major causes of cirrhosis are preventable and treatable; however, setting cost-effective policies to prevent and treat cirrhosis requires high-quality, localised data. GBD provides the most up-to-date estimates on the burden of cirrhosis by cause to guide policy makers in designing effective preventive plans and implementing interventions at national and even subnational levels. In line with the Global Health Sector Strategy on Viral Hepatitis 2016 to 2021, the Sustainable Development Goals, and the WHO Global Strategy to Reduce Harmful Use of Alcohol, there is a need to implement comprehensive prevention efforts to achieve a sustained reduction in cirrhosis burden. Supplementary Material Supplementary appendix

In conclusion, the major causes of cirrhosis are preventable and treatable; however, setting cost-effective policies to prevent and treat cirrhosis requires high-quality, localised data. GBD provides the most up-to-date estimates on the burden of cirrhosis by cause to guide policy makers in designing effective preventive plans and implementing interventions at national and even subnational levels. In line with the Global Health Sector Strategy on Viral Hepatitis 2016 to 2021, the Sustainable Development Goals, and the WHO Global Strategy to Reduce Harmful Use of Alcohol, there is a need to implement comprehensive prevention efforts to achieve a sustained reduction in cirrhosis burden. Supplementary Material Supplementary appendix Acknowledgments This study is funded by the Bill & Melinda Gates Foundation. AB is supported by the Public Health Agency of Canada. LAC acknowledges the Argentine Society of Medicine for supporting our research. FC acknowledges support through Portuguese national funds (UID/MULTI/04378/2019 and UID/QUI/50006/2019 with FCT/MCTES). VMC acknowledges her grant (SFRH/BHD/110001/2015), received by Portuguese national funds through Fundaco para a Ciencia e Tecnologia (FCT), IP, under the Norma Transitaria DL57/2016/CP1334/CT0006. SLJ works on an influenza/RSV grant funded by Sanofi Pasteur. PJ is supported by Wellcome Trust/DBT India Alliance Fellowship [grant number IA/CPHI/14/1/501497]. YJK acknowledges support from Xiamen University Malaysia Research Fund (Grant No. XMUMRF/2018-2/ITCM/0001). SL acknowledges institutional support from the Competence Cluster for Nutrition and Cardiovascular Health (nutriCARD) Halle-Jena-Leipzig (Germany; German Federal Ministry of Education and Research; grant agreement number 01EA1808A). WM is currently Program Analyst Population and Development at the United Nations Population Fund-UNFPA Country Office in Peru, which not necessarily endorses this study. PM-Z acknowledges the support awarded in 2015 by the National Council of Science and Technology of Mexico, the Inter-American Development Bank, the Organization of American States, and the National Council for Scientific and Technological Research of Costa Rica for offering them a postgraduate scholarship to be able to participate in the Global Burden of Disease technical training celebrated in Greece. AMS was supported by a fellowship from the Egyptian Fulbright Mission Program (EFMP).

n States, and the National Council for Scientific and Technological Research of Costa Rica for offering them a postgraduate scholarship to be able to participate in the Global Burden of Disease technical training celebrated in Greece. AMS was supported by a fellowship from the Egyptian Fulbright Mission Program (EFMP). Editorial note: The Lancet Group takes a neutral position with respect to territorial claims in published maps and institutional affiliations.

n States, and the National Council for Scientific and Technological Research of Costa Rica for offering them a postgraduate scholarship to be able to participate in the Global Burden of Disease technical training celebrated in Greece. AMS was supported by a fellowship from the Egyptian Fulbright Mission Program (EFMP). Editorial note: The Lancet Group takes a neutral position with respect to territorial claims in published maps and institutional affiliations. GBD 2017 Cirrhosis Collaborators Sadaf G Sepanlou, Saeid Safiri, Catherine Bisignano, Kevin S Ikuta, Shahin Merat, Mehdi Saberifiroozi, Hossein Poustchi, Derrick Tsoi, Danny V Colombara, Amir Abdoli, Rufus Adesoji Adedoyin, Mohsen Afarideh, Sutapa Agrawal, Sohail Ahmad, Elham Ahmadian, Ehsan Ahmadpour, Tomi Akinyemiju, Chisom Joyqueenet Akunna, Vahid Alipour, Amir Almasi-Hashiani, Abdulaziz M Almulhim, Rajaa M Al-Raddadi, Nelson Alvis-Guzman, Nahla Hamed Anber, Colin Angus, Amir Anoushiravani, Jalal Arabloo, Ephrem Mebrahtu Araya, Daniel Asmelash, Bahar Ataeinia, Zerihun Ataro, Maha Moh'd Wahbi Atout, Floriane Ausloos, Ashish Awasthi, Alaa Badawi, Maciej Banach, Diana Fernanda Bejarano Ramirez, Akshaya Srikanth Bhagavathula, Neeraj Bhala, Krittika Bhattacharyya, Antonio Biondi, Srinivasa Rao Bolla, Archith Boloor, Antonio M Borzì, Zahid A Butt, Luis LA Alberto Cámera, Ismael R Campos-Nonato, Félix Carvalho, Dinh-Toi Chu, Sheng-Chia Chung, Paolo Angelo Cortesi, Vera M Costa, Benjamin C Cowie, Ahmad Daryani, Barbora de Courten, Gebre Teklemariam Demoz, Rupak Desai, Samath Dhamminda Dharmaratne, Shirin Djalalinia, Hoa Thi Do, Fariba Dorostkar, Thomas M Drake, Manisha Dubey, Bruce B Duncan, Andem Effiong, Aziz Eftekhari, Aisha Elsharkawy, Arash Etemadi, Mohammad Farahmand, Farshad Farzadfar, Eduarda Fernandes, Irina Filip, Florian Fischer, Ketema Bizuwork Bizuwork Gebremedhin, Birhanu Geta, Syed Amir Gilani, Paramjit Singh Gill, Reyna Alma Gutiérrez, Michael Tamene Haile, Arvin Haj-Mirzaian, Saeed S Hamid, Milad Hasankhani, Amir Hasanzadeh, Maryam Hashemian, Hamid Yimam Hassen, Simon I Hay, Khezar Hayat, Behnam Heidari, Andualem Henok, Chi Linh Hoang, Mihaela Hostiuc, Sorin Hostiuc, Vivian Chia-rong Hsieh, Ehimario U Igumbor, Olayinka Stephen Ilesanmi, Seyed Sina Naghibi Irvani, Nader Jafari Balalami, Spencer L James, Panniyammakal Jeemon, Ravi Prakash Jha, Jost B Jonas, Jacek Jerzy Jozwiak, Ali Kabir, Amir Kasaeian, Hagazi Gebremedhin Kassaye, Adane Teshome Kefale, Rovshan Khalilov, Muhammad Ali Khan, Ejaz Ahmad Khan, Amir Khater, Yun Jin Kim, Ai Koyanagi, Carlo La Vecchia, Lee-Ling Lim, Alan D Lopez, Stefan Lorkowski, Paulo A Lotufo, Rafael Lozano, Muhammed Magdy

Jha, Jost B Jonas, Jacek Jerzy Jozwiak, Ali Kabir, Amir Kasaeian, Hagazi Gebremedhin Kassaye, Adane Teshome Kefale, Rovshan Khalilov, Muhammad Ali Khan, Ejaz Ahmad Khan, Amir Khater, Yun Jin Kim, Ai Koyanagi, Carlo La Vecchia, Lee-Ling Lim, Alan D Lopez, Stefan Lorkowski, Paulo A Lotufo, Rafael Lozano, Muhammed Magdy Abd El Razek, Hue Thi Mai, Navid Manafi, Amir Manafi, Mohammad Ali Mansournia, Lorenzo Giovanni Mantovani, Giampiero Mazzaglia, Dhruv Mehta, Walter Mendoza, Ritesh G Menezes, Melkamu Merid Mengesha, Tuomo J Meretoja, Tomislav Mestrovic, Bartosz Miazgowski, Ted R Miller, Erkin M Mirrakhimov, Prasanna Mithra, Babak Moazen, Masoud Moghadaszadeh, Abdollah Mohammadian-Hafshejani, Shafiu Mohammed, Ali H Mokdad, Pablo A Montero-Zamora, Ghobad Moradi, Mukhammad David Naimzada, Vinod Nayak, Ionut Negoi, Trang Huyen Nguyen, Richard Ofori-Asenso, In-Hwan Oh, Tinuke O Olagunju, Jagadish Rao Padubidri, Keyvan Pakshir, Adrian Pana, Mona Pathak, Akram Pourshams, Navid Rabiee, Amir Radfar, Alireza Rafiei, Kiana Ramezanzadeh, Saleem Muhammad M Rana, Salman Rawaf, David Laith Rawaf, Robert C Reiner Jr, Leonardo Roever, Robin Room, Gholamreza Roshandel, Saeed Safari, Abdallah M Samy, Juan Sanabria, Benn Sartorius, Maria Inês Schmidt, Subramanian Senthilkumaran, Masood Ali Shaikh, Mehdi Sharif, Amrollah Sharifi, Mika Shigematsu, Jasvinder A Singh, Amin Soheili, Hafiz Ansar Rasul Suleria, Berhane Fseha Teklehaimanot, Berhe Etsay Tesfay, Marco Vacante, Amir Vahedian-Azimi, Pascual R Valdez, Tommi Juhani Vasankari, Giang Thu Vu, Yasir Waheed, Kidu Gidey Weldegwergs, Andrea Werdecker, Ronny Westerman, Dawit Zewdu Wondafrash, Adam Belay Wondmieneh, Yordanos Gizachew Yeshitila, Naohiro Yonemoto, Chuanhua Yu, Zoubida Zaidi, Afshin Zarghi, Shira Zelber-Sagi, Kaleab Alemayehu Zewdie, Zhi-Jiang Zhang, Xiu-Ju Zhao, Mohsen Naghavi, and Reza Malekzadeh.

Yasir Waheed, Kidu Gidey Weldegwergs, Andrea Werdecker, Ronny Westerman, Dawit Zewdu Wondafrash, Adam Belay Wondmieneh, Yordanos Gizachew Yeshitila, Naohiro Yonemoto, Chuanhua Yu, Zoubida Zaidi, Afshin Zarghi, Shira Zelber-Sagi, Kaleab Alemayehu Zewdie, Zhi-Jiang Zhang, Xiu-Ju Zhao, Mohsen Naghavi, and Reza Malekzadeh. Affiliations Digestive Diseases Research Institute (S G Sepanlou MD, Prof S Merat MD, H Poustchi PhD, A Anoushiravani MD, Prof A Pourshams MD, G Roshandel PhD, Prof R Malekzadeh MD), Digestive Diseases Research Institute, SHARIATI Hospital (Prof M Saberifiroozi MD), Department of Epidemiology and Biostatistics (M Mansournia PhD), Department of Microbiology (A Hasanzadeh PhD), Department of Pharmacology (A Haj-Mirzaian MD), Endocrinology and Metabolism Research Center (M Afarideh MD, B Heidari MD), Hematology-Oncology and Stem Cell Transplantation Research Center (A Kasaeian PhD), Non-Communicable Diseases Research Center, Endocrinology and Metabolism Research Institute (B Ataeinia MD), Non-communicable Diseases Research Center (F Farzadfar MD), School of Public Health (M Farahmand PhD), Tehran University of Medical Sciences, Tehran, Iran (A Etemadi PhD); Non-Communicable Diseases Research Center (S G Sepanlou MD, Prof R Malekzadeh MD), Parasitology and Mycology (Prof K Pakshir PhD), Shiraz University of Medical Sciences, Shiraz, Iran; Aging Research Institute (S Safiri PhD), Department of Community Medicine (S Safiri PhD), Biotechnology Research Center (M Moghadaszadeh PhD), Department of Parasitology (E Ahmadpour PhD), Department of Pharmacology and Toxicology (A Eftekhari PhD), Infectious and Tropical Disease Research Center (E Ahmadpour PhD), Molecular Medicine Research Center (M Moghadaszadeh PhD), Pharmacology and Toxicology (E Ahmadian PhD), School of Nutrition and Food Sciences (M Hasankhani MSc), Tabriz University of Medical Sciences, Tabriz, Iran; Institute for Health Metrics and Evaluation (C Bisignano MPH, K S Ikuta MD, D Tsoi BS, D V Colombara PhD, S D Dharmaratne MD, Prof S I Hay FMedSci, S L James MD, Prof R Lozano MD, Prof A H Mokdad PhD, R C Reiner Jr PhD, Prof M Naghavi MD), Department of Health Metrics Sciences, School of Medicine (Prof S I Hay FMedSci, Prof R Lozano MD, Prof A H Mokdad PhD, R C Reiner Jr PhD, Prof B Sartorius PhD, Prof M Naghavi MD), Division of Allergy and Infectious Diseases (K S Ikuta MD), University of Washington, Seattle, WA, USA; Department of Parasitology and Mycology (A Abdoli PhD), Jahrom University of Medical Science

(Prof S I Hay FMedSci, Prof R Lozano MD, Prof A H Mokdad PhD, R C Reiner Jr PhD, Prof B Sartorius PhD, Prof M Naghavi MD), Division of Allergy and Infectious Diseases (K S Ikuta MD), University of Washington, Seattle, WA, USA; Department of Parasitology and Mycology (A Abdoli PhD), Jahrom University of Medical Science s, Jahrom, Iran; Department of Medical Rehabilitation (Prof R A Adedoyin PhD), Obafemi Awolowo University, Ile-Ife, Nigeria; Public Health Foundation of India, Gurugram, India (S Agrawal PhD, A Awasthi PhD); Vital Strategies, Gurugram, India (S Agrawal PhD); Department of Clinical Pharmacy (S Ahmad MSc), Faculty of Pharmacy, Universiti Teknologi MARA (UiTM) (S Ahmad MSc), MAHSA University, Kuala Langat, Malaysia; Department of Physiology (R Khalilov PhD), Institute of Radiation Problems of Azerbaijan (E Ahmadian PhD), Baku State University, Baku, Azerbaijan; Department of Population Health Sciences (T Akinyemiju PhD), Duke Global Health Institute (T Akinyemiju PhD), Duke University, Durham, NC, USA; Department of Public Health (C J Akunna DMD), The Intercountry Centre for Oral Health (ICOH) for Africa, Jos, Nigeria; Department of Public Health (C J Akunna DMD), Federal Ministry of Health, Garki, Nigeria; Health Management and Economics Research Center (V Alipour PhD), Tehran, Iran; Faculty of Allied Medicine (F Dorostkar PhD), Health Economics Department (V Alipour PhD), Health Management and Economics Research Center (J Arabloo PhD), Minimally Invasive Surgery Research Center (A Kabir MD), Ophthalmology Department (N Manafi MD), Pars Advanced and Minimally Invasive Medical Manners Research Center (A Kasaeian PhD), Iran University of Medical Sciences, Tehran, Iran; Department of Epidemiology (A Almasi-Hashiani PhD), Arak University of Medical Sciences, Arak, Iran; College of Medicine (A M Almulhim MBBS), Forensic Medicine Division, Department of Pathology, College of Medicine (Prof R G Menezes MD), Imam Abdulrahman Bin Faisal University, Dammam, Saudi Arabia; Department of Family and Community Medicine (Prof R M Al-Raddadi PhD), King Abdulaziz University, Jeddah, Saudi Arabia; Grupo de Investigación en Economía de la Salud (GIES) (Prof N Alvis-Guzman PhD), University of Cartagena, Cartagena, Colombia; Research Group in Hospital Management and Health Policies (Prof N Alvis-Guzman PhD), University of the Coast, Barranquilla, Colombia; Faculty of Medicine (N H Anber PhD), Mansoura University, Mansoura, Egypt (N H Anber PhD); School of Health and Related Research (C

an PhD), University of Cartagena, Cartagena, Colombia; Research Group in Hospital Management and Health Policies (Prof N Alvis-Guzman PhD), University of the Coast, Barranquilla, Colombia; Faculty of Medicine (N H Anber PhD), Mansoura University, Mansoura, Egypt (N H Anber PhD); School of Health and Related Research (C Angus MSc), University of Sheffield, Sheffield, UK; Department of Pharmacy (H G Kassaye MSc), Department of Public Health (B F Teklehaimanot MPH, B E Tesfay MPH), Department of Pharmacy (E Araya MSc), Adigrat University, Adigrat, Ethiopia; Clinical Chemistry Department (D Asmelash MSc), University of Gondar, Gondar, Ethiopia; Department of Epidemiology and Biostatistics (M Mengesha MPH), Department of Medical Laboratory Science (Z Ataro MSc), Haramaya University, Harar, Ethiopia; School of Nursing (M M W Atout PhD), University of Nottingham, Amman, Jordan; Gastro-enterology Department (F Ausloos MD), University of Liège, Liège, Belgium; Public Health Risk Sciences Division (A Badawi PhD), Public Health Agency of Canada, Toronto, ON, Canada; Department of Nutritional Sciences (A Badawi PhD), University of Toronto, Toronto, ON, Canada; Department of Hypertension (Prof M Banach PhD), Medical University of Lodz, Lodz, Poland; Polish Mothers' Memorial Hospital Research Institute, Lodz, Poland (Prof M Banach PhD); Department of Medicine (D F Bejarano Ramirez BN), El Bosque University, Bogotá, Colombia; Transplant Service (D F Bejarano Ramirez BN), University Hospital Foundation Santa Fe de Bogotá, Bogotá, Colombia; Internal Medicine (A S Bhagavathula PharmD), United Arab Emirates University, Al Ain, United Arab Emirates; Social and Clinical Pharmacy (A S Bhagavathula PharmD), Charles University, Hradec Kralove, Czech Republic; Institutes of Applied Health Research and Translational Medicine (N Bhala DPhil), Queen Elizabeth Hospital Birmingham, Birmingham, UK; IAHR/ITM (N Bhala DPhil), University of Birmingham, Birmingham, UK; Department of Statistical and Computational Genomics (K Bhattacharyya MSc), National Institute of Biomedical Genomics, Kalyani, India; Department of Statistics (K Bhattacharyya MSc), University of Calcutta, Kolkata, India; Department of General Surgery and Medical-Surgical Specialties (Prof A Biondi PhD, A M Borzì MD, M Vacante PhD), University of Catania, Catania, Italy; Department of Biomedical Sciences (S R Bolla PhD), Nazarbayev University, Nur-Sultan city, Kazakhstan; Community Medicine (P Mithra MD), Department of Forensic Medicine and Tox

nt of General Surgery and Medical-Surgical Specialties (Prof A Biondi PhD, A M Borzì MD, M Vacante PhD), University of Catania, Catania, Italy; Department of Biomedical Sciences (S R Bolla PhD), Nazarbayev University, Nur-Sultan city, Kazakhstan; Community Medicine (P Mithra MD), Department of Forensic Medicine and Tox icology (V Nayak MD), Department of Forensic Medicine, Kasturba Medical College (J Padubidri MD), Internal Medicine (A Boloor MD), Manipal Academy of Higher Education, Mangalore, India; School of Public Health and Health Systems (Z A Butt PhD), University of Waterloo, Waterloo, ON, Canada; Al Shifa School of Public Health (Z A Butt PhD), Al Shifa Trust Eye Hospital, Rawalpindi, Pakistan; Medicina Interna (Prof L L A Cámera MD), Hospital Italiano de Buenos Aires, Ciudad Autónoma de Buenos Aires, Argentina; Comisión Directiva (Prof L L A Cámera MD), Argentine Society of Medicine, Ciudad Autonoma Buenos Aires, Argentina (Prof P R Valdez MEd); Center for Health Systems Research (P A Montero-Zamora MSc), National Institute of Public Health, Cuernavaca, Mexico (I R Campos-Nonato PhD); REQUIMTE/LAQV (Prof E Fernandes PhD), UCIBIO (Prof F Carvalho PhD), UCIBIO, REQUIMTE, Laboratory of Toxicology, Faculty of Pharmacy (Prof V M Costa PharmD), University of Porto, Porto, Portugal; Faculty of Biology (D Chu PhD), Hanoi National University of Education, Hanoi, Vietnam; Department of Health Informatics (S Chung PhD), University College London, London, UK; Health Data Research UK, London, UK (S Chung PhD); Department of Medicine (G Mazzaglia PhD), School of Medicine and Surgery (P A Cortesi PhD, Prof L G Mantovani DSc), University of Milan Bicocca, Monza, Italy; WHO Collaborating Centre for Viral Hepatitis (B C Cowie PhD), The Peter Doherty Institute for Infection and Immunity, Melbourne, VIC, Australia; Department of Agriculture and Food Systems (H Suleria PhD), Department of Medicine (B C Cowie PhD), School of Population and Global Health (Prof A D Lopez PhD), University of Melbourne, Parkville, VIC, Australia; Department of Immunology (Prof A Rafiei PhD), Molecular and Cell Biology Research Center (Prof A Rafiei PhD), Toxoplasmosis Research Center (Prof A Daryani PhD), Mazandaran University of Medical Sciences, Sari, Iran; Centre of Cardiovascular Research and Education in Therapeutics (R Ofori-Asenso MSc), Monash Centre for Health Research and Implementation (B de Courten PhD), Monash University, Melbourne, VIC, Australia; Department of Diabetes and Vascular Medicine (B de

ani PhD), Mazandaran University of Medical Sciences, Sari, Iran; Centre of Cardiovascular Research and Education in Therapeutics (R Ofori-Asenso MSc), Monash Centre for Health Research and Implementation (B de Courten PhD), Monash University, Melbourne, VIC, Australia; Department of Diabetes and Vascular Medicine (B de Courten PhD), Monash Health, Melbourne, VIC, Australia; School of Pharmacy (G T Demoz MPharm), Aksum University, Aksum, Ethiopia; Department of Pharmacology (D Z Wondafrash MS), Nursing Department (K B B Gebremedhin MSc), Addis Ababa University, Addis Ababa, Ethiopia (G T Demoz MPharm, A B Wondmieneh MSc); Division of Cardiology (R Desai MBBS), Atlanta Veterans Affairs Medical Center, Decatur, GA, USA; Department of Community Medicine (S D Dharmaratne MD), University of Peradeniya, Peradeniya, Sri Lanka; Deputy of Research and Technology (S Djalalinia PhD), Ministry of Health and Medical Education, Tehran, Iran; Institute for Global Health Innovations (H T Do MD, H T Mai MPH), Duy Tan University, Hanoi, Vietnam; Department of Clinical Surgery (T M Drake MD), University of Edinburgh, Edinburgh, UK; World Food Programme, New Delhi, India (M Dubey PhD); Postgraduate Program in Epidemiology (B B Duncan MD, Prof M I Schmidt PhD), Federal University of Rio Grande do Sul, Porto Alegre, Brazil; Clinical Epidemiology and Biostatistics (A Effiong MB), University of Newcastle, Newcastle, NSW, Australia; Department of Microbiology (A Hasanzadeh PhD), Pharmacology and Toxicology Department (A Eftekhari PhD), Maragheh University of Medical Sciences, Maragheh, Iran; Endemic Medicine and Hepatogastroentrology Department (A Elsharkawy MD), Cairo University, Cairo, Egypt; Division of Cancer Epidemiology and Genetics (A Etemadi PhD), National Cancer Institute, Bethesda, MD, USA; Psychiatry Department (I Filip MD), Kaiser Permanente, Fontana, CA, USA; College of Graduate Health Sciences (A Radfar MD), School of Health Sciences (I Filip MD), A.T.

sulting estimates were then smoothed over place and time, and adjusted using spatiotemporal Gaussian process regression (see appendix of reference 10).10 The vital registration mortality, as well as the cancer registry mortality estimates computed from MIRs, were used as inputs for a Cause of Death Ensemble model.7, 11 Non-fatal estimates Pancreatic cancer incidence was computed by dividing the final mortality estimates by the MIR. Four sequelae were defined for pancreatic cancer—diagnosis and primary therapy phase, controlled phase, metastatic phase, and terminal phase.1 The diagnosis and primary therapy phase was defined as 4·1 months, the disseminated and metastatic phase as 2·54 months, and terminal phase as 1·0 month.12, 13 The remaining time was assigned to the controlled phase. Following this process, to estimate the sequelae-specific YLDs, we multiplied each sequela-specific prevalence rate by a sequela-specific disability weight. Each of the four sequelae had defined disability weights that ranged from 0·049 to 0·540 (appendix p 8). DALYs were calculated as the sum of YLDs and YLLs.

arkawy MD), Cairo University, Cairo, Egypt; Division of Cancer Epidemiology and Genetics (A Etemadi PhD), National Cancer Institute, Bethesda, MD, USA; Psychiatry Department (I Filip MD), Kaiser Permanente, Fontana, CA, USA; College of Graduate Health Sciences (A Radfar MD), School of Health Sciences (I Filip MD), A.T. Still University, Mesa, AZ, USA; Department of Population Medicine and Health Services Research (F Fischer PhD), Bielefeld University, Bielefeld, Germany; Nursing (A B Wondmieneh MSc), Pharmacy (B Geta MSc), Wollo University, Dessie, Ethiopia; Faculty of Allied Health Sciences (Prof S Gilani PhD), The University of Lahore, Lahore, Pakistan; Chairman BOG (Prof S Gilani PhD), Afro-Asian Institute, Lahore, Pakistan; Unit of Academic Primary Care (Prof P S Gill DM), University of Warwick, Coventry, UK; Department of Epidemiology and Psychosocial Research (R A Gutiérrez PhD), Ramón de la Fuente Muñiz National Institute of Psychiatry, Mexico City, Mexico; Department of Nursing (M T Haile MSc), St Paul's Hospital Millennium Medical College, Addis Ababa, Ethiopia; Department of Medicinal and Pharmaceutical Chemistry (Prof A Zarghi PhD), Department of Pharmacology (K Ramezanzadeh PharmD), Emergency Department (S Safari MD), Obesity Research Center (A Haj-Mirzaian MD), Research Institute for Endocrine Sciences (S N Irvani MD), Shahid Beheshti University of Medical Sciences, Tehran, Iran; Medicine (Prof S S Hamid FRCP), Aga Khan University, Karachi, Pakistan; Department of Biology (M Hashemian PhD), Utica College, Utica, NY, USA; Department of Public Health (H Y Hassen MPH, A Henok MPH), Pharmacy Department (A T Kefale MSc), Mizan-Tepi University, Tepi, Ethiopia; Unit of Epidemiology and Social Medicine (H Y Hassen MPH), University Hospital Antwerp, Wilrijk, Belgium; Institute of Pharmaceutical Sciences (K Hayat MS), University of Veterinary and Animal Sciences, Lahore, Pakistan; Department of Pharmacy Administration and Clinical Pharmacy (K Hayat MS), Xian Jiaotong University, Xian, China; Center of Excellence in Behavioral Medicine (C L Hoang BMedSc, T H Nguyen BMedSc, G T Vu BA), Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam; Department of General Surgery (M Hostiuc PhD), Faculty of Dentistry, Department of Legal Medicine and Bioethics (S Hostiuc PhD), General Surgery (I Negoi PhD), Carol Davila University of Medicine and Pharmacy, Bucharest, Romania; Department of Internal Medicine (M Hostiuc PhD), Bucharest Emergency Hospital, Bucharest, Romania; Clinical Leg

Surgery (M Hostiuc PhD), Faculty of Dentistry, Department of Legal Medicine and Bioethics (S Hostiuc PhD), General Surgery (I Negoi PhD), Carol Davila University of Medicine and Pharmacy, Bucharest, Romania; Department of Internal Medicine (M Hostiuc PhD), Bucharest Emergency Hospital, Bucharest, Romania; Clinical Leg al Medicine Department (S Hostiuc PhD), National Institute of Legal Medicine Mina Minovici, Bucharest, Romania; Department of Health Services Administration (V Hsieh PhD), China Medical University, Taichung, Taiwan; School of Public Health (Prof E U Igumbor PhD), University of the Western Cape, Cape Town, South Africa; Department of Public Health (Prof E U Igumbor PhD), Walter Sisulu University, Mthatha, South Africa; Department of Community Medicine (O S Ilesanmi PhD), University of Ibadan, Ibadan, Nigeria; Department of Psychosis (N Jafari Balalami PhD), Babol Noshirvani University of Technology, Babol, Iran; Achutha Menon Centre for Health Science Studies (P Jeemon PhD), Sree Chitra Tirunal Institute for Medical Sciences and Technology, Trivandrum, India; Department of Community Medicine (R P Jha MSc), Banaras Hindu University, Varanasi, India; Department of Ophthalmology (Prof J B Jonas MD), Heidelberg Institute of Global Health (HIGH) (B Moazen MSc, S Mohammed PhD), Heidelberg University, Mannheim, Germany; Beijing Institute of Ophthalmology, Beijing Ophthalmology & Visual Science Key Laboratory (Prof J B Jonas MD), Beijing Tongren Hospital, Beijing, China; Department of Family Medicine and Public Health (J J Jozwiak PhD), University of Opole, Opole, Poland; Department of Medicine (Prof J A Singh MD), University of Alabama at Birmingham, Birmingham, AL, USA (M Khan MD); University of Tennessee, Knoxville, TN, USA (M Khan MD); Epidemiology and Biostatistics Department (E A Khan MPH), Health Services Academy, Islamabad, Pakistan; Internal Medicine and Gastroenterology Department (A Khater MD), National Hepatology and Tropical Research Institute, Cairo, Egypt; School of Medicine (Y Kim PhD), Xiamen University Malaysia, Sepang, Malaysia; CIBERSAM (A Koyanagi MD), San Juan de Dios Sanitary Park, Sant Boi de Llobregat, Spain; Catalan Institution for Research and Advanced Studies (ICREA), Barcelona, Spain (A Koyanagi MD); Clinical Medicine and Community Health (Prof C La Vecchia MD), University of Milan, Milano, Italy; Department of Medicine (L Lim MRCP), University of Malaya, Kuala Lumpur, Malaysia; Department of Medicine and Therapeutics (L Lim MRCP), The Chines

Research and Advanced Studies (ICREA), Barcelona, Spain (A Koyanagi MD); Clinical Medicine and Community Health (Prof C La Vecchia MD), University of Milan, Milano, Italy; Department of Medicine (L Lim MRCP), University of Malaya, Kuala Lumpur, Malaysia; Department of Medicine and Therapeutics (L Lim MRCP), The Chines e University of Hong Kong, Shatin, Hong Kong, China; Institute of Nutrition (Prof S Lorkowski PhD), Friedrich Schiller University Jena, Jena, Germany; Competence Cluster for Nutrition and Cardiovascular Health (nutriCARD), Jena, Germany (Prof S Lorkowski PhD); Department of Medicine (Prof P A Lotufo DrPH), University of São Paulo, São Paulo, Brazil; Ophthalmology Department (M Magdy Abd El Razek MB), Aswan Faculty of Medicine, Aswan, Egypt; Ophthalmology Department (N Manafi MD), University of Manitoba, Winnipeg, MB, Canada; Department of Surgery (A Manafi MD), University of Virginia, Charlottesville, VA, USA; Value-Based Healthcare Unit (Prof L G Mantovani DSc), IRCCS MultiMedica, Sesto San Giovanni, Italy; Division of Gastroenterology and Hepatobiliary disease (D Mehta MD), New York Medical College, Valhalla, NY, USA; Peru Country Office (W Mendoza MD), United Nations Population Fund (UNFPA), Lima, Peru; Breast Surgery Unit (T J Meretoja MD), Helsinki University Hospital, Helsinki, Finland; University of Helsinki, Helsinki, Finland (T J Meretoja MD); Clinical Microbiology and Parasitology Unit (T Mestrovic PhD), Zora Profozic Polyclinic, Zagreb, Croatia; University Centre Varazdin (T Mestrovic PhD), University North, Varazdin, Croatia; Center for Innovation in Medical Education (B Miazgowski MD), Pomeranian Medical University, Szczecin, Poland (B Miazgowski MD); Pacific Institute for Research & Evaluation, Calverton, MD, USA (T R Miller PhD); School of Public Health (T R Miller PhD), Curtin University, Perth, WA, Australia; Faculty of Internal Medicine (Prof E M Mirrakhimov MD), Kyrgyz State Medical Academy, Bishkek, Kyrgyzstan; Department of Atherosclerosis and Coronary Heart Disease (Prof E M Mirrakhimov MD), National Center of Cardiology and Internal Disease, Bishkek, Kyrgyzstan; Institute of Addiction Research (ISFF) (B Moazen MSc), Frankfurt University of Applied Sciences, Frankfurt, Germany; Department of Epidemiology and Biostatistics (A Mohammadian-Hafshejani PhD), Shahrekord University of Medical Sciences, Shahrekord, Iran; Health Systems and Policy Research Unit (S Mohammed PhD), Ahmadu Bello University, Zaria, Nigeria; Department of Public Health Sc

University of Applied Sciences, Frankfurt, Germany; Department of Epidemiology and Biostatistics (A Mohammadian-Hafshejani PhD), Shahrekord University of Medical Sciences, Shahrekord, Iran; Health Systems and Policy Research Unit (S Mohammed PhD), Ahmadu Bello University, Zaria, Nigeria; Department of Public Health Sc iences (P A Montero-Zamora MSc), University of Miami, Miami, FL, USA; Department of Epidemiology and Biostatistics (G Moradi PhD), Social Determinants of Health Research Center (G Moradi PhD), Kurdistan University of Medical Sciences, Sanandaj, Iran; Laboratory of Public Health Indicators Analysis and Health Digitalization (M Naimzada MD), Moscow Institute of Physics and Technology, Dolgoprudny, Russia; Experimental Surgery and Oncology Laboratory (M Naimzada MD), Kursk State Medical University of the Ministry of Health of the Russian Federation, Kursk, Russia; General Surgery (I Negoi PhD), Emergency Hospital of Bucharest, Bucharest, Romania; Independent Consultant, Accra, Ghana (R Ofori-Asenso MSc); Department of Preventive Medicine (I Oh PhD), Kyung Hee University, Dongdaemun-gu, South Korea; Department of Pathology and Molecular Medicine (T O Olagunju MD), McMaster University, Hamilton, ON, Canada; Department of Statistics and Econometrics (A Pana MD), Bucharest University of Economic Studies, Bucharest, Romania; Center for Health Outcomes & Evaluation, Bucharest, Romania (A Pana MD); Research & Publication Cell (M Pathak PhD), Kalinga Institute of Medical Sciences, Bhubaneswar, Bhubaneswar, India; Department of Chemistry (N Rabiee PhD), Sharif University of Technology, Tehran, Iran; College of Medicine (A Radfar MD), University of Central Florida, Orlando, FL, USA; University Institute of Public Health (Prof S M M Rana PhD), University of Lahore, Lahore, Pakistan; Public Health Department (Prof S M M Rana PhD), University of Health Sciences, Lahore, Pakistan; Department of Primary Care and Public Health (Prof S Rawaf MD), WHO Collaborating Centre for Public Health Education and Training (D L Rawaf MD), Imperial College London, London, UK; Academic Public Health Department (Prof S Rawaf MD), Public Health England, London, UK; University College London Hospitals, London, UK (D L Rawaf MD); Department of Clinical Research (L Roever PhD), Federal University of Uberlândia, Uberlândia, Brazil; Centre for Alcohol Policy Research (Prof R Room PhD), La Trobe University, Melbourne, VIC, Australia; Centre for Social Research on Alcohol and Drugs (Prof R Room PhD), Sto

London Hospitals, London, UK (D L Rawaf MD); Department of Clinical Research (L Roever PhD), Federal University of Uberlândia, Uberlândia, Brazil; Centre for Alcohol Policy Research (Prof R Room PhD), La Trobe University, Melbourne, VIC, Australia; Centre for Social Research on Alcohol and Drugs (Prof R Room PhD), Sto ckholm University, Stockholm, Sweden; Golestan Research Center of Gastroenterology and Hepatology (G Roshandel PhD), Golestan Research Center of Gastroenterology and Hepatology (GRCGH) (A Sharifi PhD), Golestan University of Medical Sciences, Gorgan, Iran; Department of Entomology (A M Samy PhD), Ain Shams University, Cairo, Egypt; Department of Surgery (Prof J Sanabria MD), Marshall University, Huntington, WV, USA; Department of Nutrition and Preventive Medicine (Prof J Sanabria MD), Case Western Reserve University, Cleveland, OH, USA; Faculty of Infectious and Tropical Diseases (Prof B Sartorius PhD), London School of Hygiene & Tropical Medicine, London, UK; Emergency Department (S Senthilkumaran MD), Manian Medical Centre, Erode, India; Department of Basic Sciences (Prof M Sharif PhD), Department of Laboratory Sciences (Prof M Sharif PhD), Islamic Azad University, Sari, Iran; Independent Consultant, Sindh, Pakistan (M A Shaikh MD); National Institute of Infectious Diseases, Tokyo, Japan (M Shigematsu PhD); Medicine Service (Prof J A Singh MD), US Department of Veterans Affairs (VA), Birmingham, AL, USA; Medical Surgical Nursing Department (A Soheili PhD), Urmia University of Medical Science, Urmia, Iran; Emergency Nursing Department (A Soheili PhD), Semnan University of Medical Sciences, Iran; Trauma Research Center, Nursing Facility (A Vahedian-Azimi PhD), Baqiyatallah University of Medical Sciences, Tehran, Iran; Velez Sarsfield Hospital, Buenos Aires, Argentina (Prof P R Valdez MEd); UKK Institute, Tampere, Finland (Prof T J Vasankari MD); Foundation University Medical College (Y Waheed PhD), Foundation University, Islamabad, Pakistan; Clinical Pharmacy Unit (K G Weldegwergs MSc), Department of Pharmacology and Toxicology, College of Health Sciences, School of Pharmacy (D Z Wondafrash MS), Pharmacy (K A Zewdie MSc), Mekelle University, Mekelle, Ethiopia; Competence Center of Mortality-Follow-Up, German National Cohort (R Westerman DSc), Demographic Change and Ageing Research Area (A Werdecker PhD), Federal Institute for Population Research, Wiesbaden, Germany; Center of Population and Health, Wiesbaden, Germany (A Werdecker PhD); Nursing (Y G Yeshitila MSc

e, Ethiopia; Competence Center of Mortality-Follow-Up, German National Cohort (R Westerman DSc), Demographic Change and Ageing Research Area (A Werdecker PhD), Federal Institute for Population Research, Wiesbaden, Germany; Center of Population and Health, Wiesbaden, Germany (A Werdecker PhD); Nursing (Y G Yeshitila MSc ), Arba Minch University, Arba Minch, Ethiopia; Department of Psychopharmacology (N Yonemoto MPH), National Center of Neurology and Psychiatry, Tokyo, Japan; Department of Epidemiology and Biostatistics (Prof C Yu PhD), Department of Preventive Medicine (Z Zhang PhD), Global Health Institute (Prof C Yu PhD), Wuhan University, Wuhan, China; Department of Epidemiology (Prof Z Zaidi PhD), University Hospital of Setif, Setif, Algeria; School of Public Health, Faculty of Social Welfare and Health Sciences (Prof S Zelber-Sagi PhD), University of Haifa, Haifa, Israel; Gastroenterology Department (Prof S Zelber-Sagi PhD), Tel Aviv Saurasky Medical Center, Tel Aviv, Israel; and Wuhan Polytechnic University, Wuhan, China (X Zhao PhD). Contributors SGS, SM, SS, AP, MS, and HP prepared the first draft. RM, MSF, DC, and MN provided overall guidance. RM, SGS, SM, SS, and KS managed the project. SGS, SS, and DT analysed data. RM, SGS, SM, MSF, and CB finalised the manuscript on the basis of comments from other authors and reviewer feedback. All other authors provided data, developed models, reviewed results, provided guidance on methods, or reviewed and contributed to the manuscript.

nd KS managed the project. SGS, SS, and DT analysed data. RM, SGS, SM, MSF, and CB finalised the manuscript on the basis of comments from other authors and reviewer feedback. All other authors provided data, developed models, reviewed results, provided guidance on methods, or reviewed and contributed to the manuscript. Declaration of interests SLJ reports grants from Sanofi Pasteur, outside the submitted work. PJ reports receiving the Wellcome Trust/DBT India Alliance Fellowship [grant number IA/CPHI/14/1/501497]. JJJ reports personal fees from Alab Laboratoria and Teva Polska, and non-financial support from Servier, Microlife, Superpharm, and Medicover, outside the submitted work. SL reports personal fees from Akcea Therapeutics, Amgen, Berlin-Chemie, Boehringer Ingelheim Pharma, Daiichi Sankyo, MSD Sharp & Dohme, Novo Nordisk, Sanofi-Aventis, Synlab, Unilever, and Upfield, as well as non-financial support from Preventicus outside the submitted work. PAL reports grants from Fundação Vale, Rio de Janeiro, Brazil. JAS is on the steering committee of OMERACT, an international organization that develops measures for clinical trials and receives funding from 12 pharmaceutical companies, serves on the FDA Arthritis Advisory Committee, is a member of the Veterans Affairs Rheumatology Field Advisory Committee, is the editor and the Director of the UAB Cochrane Musculoskeletal Group Satellite Center on Network Meta-analysis, is on the speaker's bureau of Simply Speaking, consults with Crealta/Horizon, Medisys, Fidia, UBM LLC, Trio health, Medscape, WebMD, Clinical Care options, Clearview healthcare partners, Putnam associates, Spherix, Practice Point communications, the National Institutes of Health and the American College of Rheumatology, and has stock options in Amarin Pharmaceuticals and Viking Pharmaceuticals. All other authors declare no competing interests.

Introduction Cancer incidence and mortality are rapidly increasing worldwide.1, 2 This increase is thought to be due to population growth and ageing, as well as changes in the prevalence of the main risk factors for cancer, several of which are associated with socioeconomic development.1, 2 Pancreatic cancer remains one of the cancers with the poorest prognosis, with an overall 5-year survival rate of about 5%, without much difference between high-income countries and low-income and middle-income countries.3 On the basis of the results of the previous iteration of the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD), pancreatic cancer ranked eighth among cancers in mortality and 14th in incidence in 2016.1 Pancreatic cancer incidence and mortality vary considerably in the world.1 The highest incidence and mortality rates of pancreatic cancer are found in high-income countries.2 Although the causes of pancreatic cancer are still insufficiently understood, certain risk factors have been identified, such as smoking, obesity, and diabetes.4 These risk factors probably explain some of the national variation. Research in context Evidence before this study

Pancreatic cancer remains one of the cancers with the poorest prognosis, with an overall 5-year survival rate of about 5%, without much difference between high-income countries and low-income and middle-income countries.3 On the basis of the results of the previous iteration of the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD), pancreatic cancer ranked eighth among cancers in mortality and 14th in incidence in 2016.1 Pancreatic cancer incidence and mortality vary considerably in the world.1 The highest incidence and mortality rates of pancreatic cancer are found in high-income countries.2 Although the causes of pancreatic cancer are still insufficiently understood, certain risk factors have been identified, such as smoking, obesity, and diabetes.4 These risk factors probably explain some of the national variation. Research in context Evidence before this study Pancreatic cancer was estimated as the seventh leading cause of cancer death in both sexes worldwide in 2018, on the basis of the Global Cancer Incidence, Mortality and Prevalence 2018 estimates, from 185 countries, using subregional rather than national data. Because of the poor prognosis of pancreatic cancer, there were almost as many deaths (n=432 000) as there were cases (n=459 000). The rates reported were three times to four times higher in higher Human Development Index countries, with incidence rates being highest in Europe, North America, Australia, and New Zealand, and lowest in south central Asia. To our knowledge, there were no estimates of temporal patterns, trends, age patterns, years of life lost, disability-adjusted life-years, and associated risk factors of pancreatic cancer at national, regional, global, and socioeconomic levels before the Global Burden of Disease Study (GBD).

owest in south central Asia. To our knowledge, there were no estimates of temporal patterns, trends, age patterns, years of life lost, disability-adjusted life-years, and associated risk factors of pancreatic cancer at national, regional, global, and socioeconomic levels before the Global Burden of Disease Study (GBD). Added value of this study We present estimates of the global burden of pancreatic cancer based on results from GBD 2017, which are reported by sex and age groups for 195 countries and territories from 1990 to 2017. We also investigated the association of socioeconomic development status with incidence and mortality caused by pancreatic cancer at the national level. We believe that this analysis provides the most comprehensive picture of the burden of pancreatic cancer to date. Examining trends of pancreatic cancer from 1990 to 2017 and comparisons across populations offers important information about the changing burden of pancreatic cancer to aid in the allocation of necessary resources at local levels to help control this lethal cancer. Implications of all the available evidence

We present estimates of the global burden of pancreatic cancer based on results from GBD 2017, which are reported by sex and age groups for 195 countries and territories from 1990 to 2017. We also investigated the association of socioeconomic development status with incidence and mortality caused by pancreatic cancer at the national level. We believe that this analysis provides the most comprehensive picture of the burden of pancreatic cancer to date. Examining trends of pancreatic cancer from 1990 to 2017 and comparisons across populations offers important information about the changing burden of pancreatic cancer to aid in the allocation of necessary resources at local levels to help control this lethal cancer. Implications of all the available evidence The incidence and mortality rates of pancreatic cancer increased in almost all countries and territories from 1990 to 2017. With population growth and increases in longevity, clinicians and policy makers might expect a further substantial rise in the absolute number of pancreatic cancer cases, particularly in low-income and middle-income nations. Existing data gaps are a major challenge for policy making at the regional and national scale. To our knowledge, this study is the first effort to provide comprehensive worldwide estimates of the burden, epidemiological features, and risk factors of pancreatic cancer. Future studies should explore the predictors of these epidemiological trends to help policy makers implement cost-effective interventions for prevention, early detection, and control of pancreatic cancer.

e comprehensive worldwide estimates of the burden, epidemiological features, and risk factors of pancreatic cancer. Future studies should explore the predictors of these epidemiological trends to help policy makers implement cost-effective interventions for prevention, early detection, and control of pancreatic cancer. Data about incidence and trends of pancreatic cancer and its risk factors are scarce, specifically in nuanced time and location dimensions. GBD is the first comprehensive and systematic effort to report the incidence of and mortality and disability caused by pancreatic cancer and its risk factors, using an extensive set of data sources and novel statistical methods in seven super-regions, 21 regions, and 195 countries and territories, for both sexes and 20 age groups, from 1990 to 2017. To our knowledge, this study is the first to investigate the association between development status (measured by the Socio-demographic Index [SDI]) and pancreatic cancer incidence and mortality at the national level. Methods Overview This study is part of GBD. In the latest iteration, GBD 2017, 359 diseases and injuries, 282 causes of death, and 84 risk factors were estimated. The rationale, methods, and summary results of GBD 2017 have been published previously.5, 6, 7, 8 Rates and numbers of deaths, incident cases, years of life lost (YLLs) as a result of premature death, years lived with disability (YLDs), and disability-adjusted life-years (DALYs) were reported for both males and females, 17 age groups, and 195 countries and territories.

2017 have been published previously.5, 6, 7, 8 Rates and numbers of deaths, incident cases, years of life lost (YLLs) as a result of premature death, years lived with disability (YLDs), and disability-adjusted life-years (DALYs) were reported for both males and females, 17 age groups, and 195 countries and territories. The rates were age-standardised according to the world population estimated by the GBD study.9 95% uncertainty intervals (UIs) were reported for all estimates, including all sources of uncertainty arising from measurement error, systematic biases, and modelling. This study is compliant with the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER). Data sources We considered all cancers coded as C25–C25.9 in the 10th revision of the International Classification of Diseases to be pancreatic cancer and mapped them to the GBD cause list.5, 7 For this study, we used GBD 2017 vital registration and sample vital registration (19 321 site-years of data) and cancer registry (4472 site-years) data.7 Vital registration systems include vital event data from all residents in a population, including causes of death. Sample vital registration systems include nationally representative data from which birth rates, death rates, and causes of death can be estimated. Cancer registries include data on all cancer patients in a defined population, typically from a particular location. Detailed information on data sources used in this study can be found on the GBD 2017 Data Input Sources Tool website.

ative data from which birth rates, death rates, and causes of death can be estimated. Cancer registries include data on all cancer patients in a defined population, typically from a particular location. Detailed information on data sources used in this study can be found on the GBD 2017 Data Input Sources Tool website. Mortality estimates Data coverage and quality were higher for mortality data than for other measures of pancreatic cancer burden. The cancer registry mortality estimates that were uploaded into the causes of death database were derived from cancer registry incidence data that had been transformed to mortality estimates through the use of mortality-to-incidence ratios (MIRs). We modelled MIRs using the locations that had both incidence and mortality data for the same year. The initial MIR model used a linear-step mixed-effects model with logit link functions. The resulting estimates were then smoothed over place and time, and adjusted using spatiotemporal Gaussian process regression (see appendix of reference 10).10 The vital registration mortality, as well as the cancer registry mortality estimates computed from MIRs, were used as inputs for a Cause of Death Ensemble model.7, 11

phase. Following this process, to estimate the sequelae-specific YLDs, we multiplied each sequela-specific prevalence rate by a sequela-specific disability weight. Each of the four sequelae had defined disability weights that ranged from 0·049 to 0·540 (appendix p 8). DALYs were calculated as the sum of YLDs and YLLs. SDI We used the SDI to determine the relationship between pancreatic cancer incidence and mortality rates with development status at national and regional levels. The SDI ranges from 0 (worst) to 1 (best) and is composed of the total fertility rate among women under the age of 25 years, mean education for individuals aged 15 years and older, and lag-distributed income per capita.5, 6, 7 Components were extracted using principal components analysis. Each component was given equal weight, and the final SDI score was computed as the geometric mean of each of the components. Risk factors We used the comparative risk assessment framework to estimate the proportion of deaths and DALYs attributable to three recognised risk factors for pancreatic cancer: smoking, high fasting plasma glucose, and high body-mass index (BMI). We used the counterfactual scenario of theoretical minimum risk exposure level to model the population attributable fraction. The definitions of the framework have already been published.8

table to three recognised risk factors for pancreatic cancer: smoking, high fasting plasma glucose, and high body-mass index (BMI). We used the counterfactual scenario of theoretical minimum risk exposure level to model the population attributable fraction. The definitions of the framework have already been published.8 Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and the final responsibility for the decision to submit for publication. Results The number of incident cases of pancreatic cancer in both sexes increased 2·3 times from 195 000 (95% UI 192 000–199 000) incident cases in 1990 to 448 000 (439 000–456 000) cases in 2017 globally (appendix p 19). In 2017, 51·9% (232 000 [210 000–221 000]) of the total incident cases occurred in males, compared with 52·1% (102 000 [99 000–106 000]) in 1990. The global age-standardised incidence rate was 5·0 (95% UI 4·9–5·1) per 100 000 person-years in 1990, which increased to 5·7 (5·6–5·8) per 100 000 person-years in 2017 (appendix p 19). Globally, there were 9·1 million (8·9–9·3) DALYs due to pancreatic cancer in 2017. This was a 2·1 times increase from 4·4 million (4·3–4·5) DALYs in 1990 (appendix p 27). 99% of all DALYs in all years were due to YLLs (appendix p 7).

90, which increased to 5·7 (5·6–5·8) per 100 000 person-years in 2017 (appendix p 19). Globally, there were 9·1 million (8·9–9·3) DALYs due to pancreatic cancer in 2017. This was a 2·1 times increase from 4·4 million (4·3–4·5) DALYs in 1990 (appendix p 27). 99% of all DALYs in all years were due to YLLs (appendix p 7). In 2017, pancreatic cancer caused 441 000 (95% UI 433 000–449 000) deaths globally, including 226 000 (51·3%; 219 000–233 000) deaths among males and 215 000 (48·7%; 211 000–220 000) deaths among females. There was a 2·3 times (125% [118–131]) increase in the number of deaths globally from 1990 to 2017, increasing from 196 000 (193 000–200 000) deaths for both sexes combined in 1990. The age-standardised death rate increased by 10·4% (7·0–13·0), from 5·1 (5·0–5·2) per 100 000 person-years in 1990 to 5·6 (5·5–5·7) per 100 000 person-years in 2017 (appendix p 11). The age-standardised death rate in males was 5·7 (5·6–5·9) per 100 000 person-years in 1990 and 6·3 (6·1–6·5) per 100 000 person-years in 2017. The equivalent findings for females were 4·5 (4·5–4·6) per 100 000 person-years in 1990 and 5·0 (4·9–5·1) per 100 000 person-years in 2017.

erson-years in 2017 (appendix p 11). The age-standardised death rate in males was 5·7 (5·6–5·9) per 100 000 person-years in 1990 and 6·3 (6·1–6·5) per 100 000 person-years in 2017. The equivalent findings for females were 4·5 (4·5–4·6) per 100 000 person-years in 1990 and 5·0 (4·9–5·1) per 100 000 person-years in 2017. The age-standardised death rate was highest in the high-income super-region across all years from 1990 to 2017: 8·1 (8·1–8·2) per 100 000 person-years in 1990 and 8·6 (8·5–8·8) per 100 000 person-years in 2017 (appendix p 2). Central Europe, eastern Europe, and central Asia ranked second at 6·8 (6·5–7·0) per 100 000 person-years in 1990 and 7·6 (7·5–7·8) per 100 000 person-years in 2017. South Asia had the lowest rates: 1·6 (1·4–1·8) per 100 000 person-years in 1990 and 2·9 (2·7–3·0) per 100 000 person-years in 2017. The pattern of age-standardised incidence rates in super-regions was similar to the pattern we observed for age-standardised death rate (appendix p 2). The pattern of age-standardised incidence and death rates was also similar between sexes (data not shown).

s in 1990 and 2·9 (2·7–3·0) per 100 000 person-years in 2017. The pattern of age-standardised incidence rates in super-regions was similar to the pattern we observed for age-standardised death rate (appendix p 2). The pattern of age-standardised incidence and death rates was also similar between sexes (data not shown). Age-standardised incidence and death rates increased in all GBD regions from 1990 to 2017 (figure 1). High-income North America and western Europe were among the top three regions for highest age-standardised rates of both incidence and deaths in 2017, with high-income Asia Pacific and central Europe also in the top three for highest age-standardised rate of incidence and death, respectively (figure 1). These regions all had smaller increases in age-standardised rates of incidence and deaths from 1990 to 2017 than many other regions (figure 1B, D; appendix pp 11–26). The lowest age-standardised incidence and death rates in 2017 were observed in south Asia and eastern and central sub-Saharan Africa (figure 1A, C). The Caribbean, Andean Latin America, and central Asia had the highest percentage change in both incidence and death rates from 1990 to 2017 (figure 1B, D). The age-standardised rates for both incidence and death were higher among males than females in almost all regions and all years from 1990 to 2017, with the exception of Andean Latin America and western sub-Saharan Africa (figure 1A, C).Figure 1 Levels and trends in age-standardised incidence and death rates of pancreatic cancer across 21 GBD regions by sex

idence and death were higher among males than females in almost all regions and all years from 1990 to 2017, with the exception of Andean Latin America and western sub-Saharan Africa (figure 1A, C).Figure 1 Levels and trends in age-standardised incidence and death rates of pancreatic cancer across 21 GBD regions by sex (A) The age-standardised incidence rates of pancreatic cancer in 2017. (B) The percentage change in age-standardised incidence rate of pancreatic cancer from 1990 to 2017. (C) The age-standardised death rates of pancreatic cancer in 2017. (D) The percentage change in age-standardised death rate of pancreatic cancer from 1990 to 2017. GBD=Global Burden of Diseases, Injuries, and Risk Factors Study.

ercentage change in age-standardised incidence rate of pancreatic cancer from 1990 to 2017. (C) The age-standardised death rates of pancreatic cancer in 2017. (D) The percentage change in age-standardised death rate of pancreatic cancer from 1990 to 2017. GBD=Global Burden of Diseases, Injuries, and Risk Factors Study. Age-specific rates for both incidence and deaths increased with increasing age; this trend was similar between males and females (figure 2A, B). The number of both deaths and incident cases peaked at the ages of 65–69 years in males, whereas the peak in females was observed at the ages of 75–79 years. Additionally, the numbers of deaths and incident cases were lower in females younger than 75 years than in males in the same age group, whereas the numbers were higher in females than in males in age groups of 75 years and older (figure 2A, B). Until the ages of 90–94 years, the incidence, death, and DALYs rates were higher in males than in females in the same age group (figure 2). The age pattern for number of DALYs showed a similar trend to number of deaths and incident cases for total counts, but the rates decreased in age groups older than 80 years. Similar to the number of incident cases and deaths, in 2017 the number of DALYs was much higher in males than in females in all age groups younger 75 years, after which female DALY numbers were higher (although with overlapping uncertainty in the age group of 75–79 years).Figure 2 Age-specific counts and rates of incident cases (A), deaths (B), and DALYs (C) of pancreatic cancer by sex, 2017

DALYs was much higher in males than in females in all age groups younger 75 years, after which female DALY numbers were higher (although with overlapping uncertainty in the age group of 75–79 years).Figure 2 Age-specific counts and rates of incident cases (A), deaths (B), and DALYs (C) of pancreatic cancer by sex, 2017 DALYs=disability-adjusted life-years.

DALYs was much higher in males than in females in all age groups younger 75 years, after which female DALY numbers were higher (although with overlapping uncertainty in the age group of 75–79 years).Figure 2 Age-specific counts and rates of incident cases (A), deaths (B), and DALYs (C) of pancreatic cancer by sex, 2017 DALYs=disability-adjusted life-years. In both 1990 and 2017, the highest age-standardised death rates were observed in Greenland: 19·7 (95% UI 17·8–21·8) per 100 000 person-years in 1990 and 17·4 (15·8–19·0) per 100 000 person-years in 2017 (figure 3A; appendix p 11). Yet the number of deaths due to pancreatic cancer in Greenland was among the lowest in the world (11·1 [10·2–12·1] in 2017). Uruguay was the next leading country for highest age-standardised death rates from pancreatic cancer, although it was substantially behind Greenland, with an age-standardised death rate of 12·1 (10·9–13·5) per 100 000 person-years in 2017. Bangladesh (1·9 [1·5–2·3] per 100 000 person-years) had the lowest age-standardised rate in 2017, whereas São Tomé and Príncipe (1·3 [1·1–1·5] per 100 000 person-years) had the lowest rate in 1990. The incidence rates followed a very similar pattern: highest in Greenland in both 1990 and 2017, lowest in São Tomé and Príncipe in 1990, and lowest in Bangladesh in 2017 (figure 3B). All estimates were similar between males and females (data not shown). Specific country and territory data for incidence, deaths, and DALYs can be found in the appendix (pp 11–34).Figure 3 Age-standardised rates of incidence (A) and death (B) of pancreatic cancer across 195 countries and territories in both sexes, 2017

All estimates were similar between males and females (data not shown). Specific country and territory data for incidence, deaths, and DALYs can be found in the appendix (pp 11–34).Figure 3 Age-standardised rates of incidence (A) and death (B) of pancreatic cancer across 195 countries and territories in both sexes, 2017 The average annualised percentage change in both age-standardised incidence and death rates was highest in Grenada (5·5%) and lowest in Bahrain (−1·2%) from 1990 to 2017. For males, the percentage change in both age-standardised incidence and death rates was highest in Bermuda (6·0%) and lowest in Qatar (−1·5% for age-standardised incidence rate and −1·6% for age-standardised death rate). For females, the highest percentage change in both age-standardised incidence and death rates was in Grenada (5·4% for age-standardised incidence rate and 5·5% for age-standardised death rate) and the lowest in Bahrain (−1·2%).

Qatar (−1·5% for age-standardised incidence rate and −1·6% for age-standardised death rate). For females, the highest percentage change in both age-standardised incidence and death rates was in Grenada (5·4% for age-standardised incidence rate and 5·5% for age-standardised death rate) and the lowest in Bahrain (−1·2%). 93 600 (95% UI 82 500–108 000) pancreatic cancer deaths, equivalent to 21·1% (18·8–23·7) of all age-standardised deaths from pancreatic cancer, were attributable to smoking for both sexes combined in 2017. The age-standardised proportions of all pancreatic cancer deaths that were attributable to smoking in 2017 were 25·9% (22·2–29·6) for males and 16·1% (13·2–18·8) for females (figure 4A). 59 000 (63·1%; 50 000–68 000) of these deaths occurred in males and 33 500 (36·1%; 28 000–41 000) in females. In 1990, the proportion of pancreatic cancer age-standardised deaths attributable to smoking was 26·6% (23·8–29·5) for both sexes combined.Figure 4 Fraction of pancreatic cancer age-standardised deaths attributable to smoking, high fasting plasma glucose, and high body-mass index by region (A) and fraction of pancreatic cancer age-specific deaths attributable to smoking, high fasting plasma glucose, and high body-mass index by age group (B) for males and females, 2017

tion of pancreatic cancer age-standardised deaths attributable to smoking, high fasting plasma glucose, and high body-mass index by region (A) and fraction of pancreatic cancer age-specific deaths attributable to smoking, high fasting plasma glucose, and high body-mass index by age group (B) for males and females, 2017 Globally, in 2017, 8·9% (2·1–19·4) of pancreatic cancer age-standardised deaths were attributable to high fasting plasma glucose, including 9·3% (1·7–21·3) in males and 8·6% (1·4–19·6) in females (figure 4A; appendix pp 8–9), compared with 7·7% (1·8–16·8) for both sexes combined in 1990. Likewise, 6·2% (2·5–11·4) of pancreatic cancer age-standardised deaths were attributable to high BMI, including 5·0% (0·0–12·1) in males and 7·4% (2·6–13·0) in females (figure 4A; appendix pp 8–9), compared with 5·0% (1·9–9·6) for both sexes combined in 1990.

mpared with 7·7% (1·8–16·8) for both sexes combined in 1990. Likewise, 6·2% (2·5–11·4) of pancreatic cancer age-standardised deaths were attributable to high BMI, including 5·0% (0·0–12·1) in males and 7·4% (2·6–13·0) in females (figure 4A; appendix pp 8–9), compared with 5·0% (1·9–9·6) for both sexes combined in 1990. In 2017, the proportion of age-standardised deaths attributable to smoking for males was highest in eastern Europe (35·7% of all pancreatic cancer deaths) and east Asia (31·3%); for females it was highest in high-income North America (29·3%) and southern Latin America (27·6%). The lowest age-standardised attributable proportion for smoking was observed in western sub-Saharan Africa for both males (8·0%) and females (2·1%). In 2017, the highest proportion of age-standardised deaths attributable to high fasting plasma glucose was observed in Oceania in both males (16·0%) and females (17·3%), and the highest fraction attributable to high BMI was observed in high-income North America for both males (8·6%) and females (11·7%). Additionally, in 2017, the lowest proportion of age-standardised deaths attributable to high fasting plasma glucose was observed in east Asia in males (5·8%) and in tropical Latin America in females (6·1%). As for proportion of deaths attributable to high BMI, the lowest proportion in males was observed in eastern sub-Saharan Africa (2·1%), and in females the lowest proportion was observed in high-income Asia Pacific (3·5%; figure 4A).

observed in east Asia in males (5·8%) and in tropical Latin America in females (6·1%). As for proportion of deaths attributable to high BMI, the lowest proportion in males was observed in eastern sub-Saharan Africa (2·1%), and in females the lowest proportion was observed in high-income Asia Pacific (3·5%; figure 4A). Across age groups, the proportion of age-standardised deaths attributable to smoking was higher than 25% in males aged between 55 and 84 years and higher than 16% in females in the same age group (figure 4B). The highest proportion attributable to high fasting plasma glucose in both sexes was observed in the 85–89 year age group. Although higher proportions of pancreatic cancer deaths attributable to high BMI were observed between the ages of 45 years and 79 years, the proportions were more similar between all age groups than for the leading risk factors, with attributable deaths starting to occur at the ages of 20–24 years (figure 4B). From 1990 to 2017, the age-standardised rates of both deaths and incidence of pancreatic cancer increased, along with increases in SDI. That is, the lowest rates were observed in low SDI countries and higher rates were detected in countries with respectively higher SDI across all years from 1990 to 2017 (appendix p 3).

From 1990 to 2017, the age-standardised rates of both deaths and incidence of pancreatic cancer increased, along with increases in SDI. That is, the lowest rates were observed in low SDI countries and higher rates were detected in countries with respectively higher SDI across all years from 1990 to 2017 (appendix p 3). Figure 5 demonstrates the trend in age-standardised death rates across SDI by region, from 1990 to 2017. Regions generally followed the trend of increasing death, incidence, and DALY rates along with increases in SDI (figure 5; appendix pp 4–5). Several regions, including Andean Latin America and southern sub-Saharan Africa, showed a decrease in age-standardised death rate late in the study period, but not down to 1990 levels. Among high-income regions, Australasia had a rising age-standardised death rate but it was well below the expected levels in all years, while other high-income regions were either near or above the levels expected on the basis of SDI. Although south Asia, southeast Asia, and east Asia had rising age-standardised death rates, they were among the lowest in the world and were below the expected levels in all years from 1990 to 2017. Figure 6 demonstrates the association between age-standardised death rate and SDI across countries and territories in 2017. Similar to regional trends, there was a trend at the national level of increasing age-standardised death rates along with increases in SDI. As mentioned previously, the observed levels were much higher than expected in Greenland and Uruguay and much lower than expected in many countries, including Bangladesh, Kuwait, and Singapore based solely on SDI.Figure 5 The trend in age-standardised death rates of pancreatic cancer across 21 GBD regions by SDI for both sexes combined, 1990–2017

observed levels were much higher than expected in Greenland and Uruguay and much lower than expected in many countries, including Bangladesh, Kuwait, and Singapore based solely on SDI.Figure 5 The trend in age-standardised death rates of pancreatic cancer across 21 GBD regions by SDI for both sexes combined, 1990–2017 For each region, points from left to right depict estimates from each year from 1990 to 2017. GBD=Global Burden of Diseases, Injuries, and Risk Factors Study. SDI=Socio-demographic Index. Figure 6 The age-standardised death rates of pancreatic cancer across 195 countries and territories by SDI in both sexes, 2017 SDI=Socio-demographic Index.

For each region, points from left to right depict estimates from each year from 1990 to 2017. GBD=Global Burden of Diseases, Injuries, and Risk Factors Study. SDI=Socio-demographic Index. Figure 6 The age-standardised death rates of pancreatic cancer across 195 countries and territories by SDI in both sexes, 2017 SDI=Socio-demographic Index. Discussion Our results showed that there were approximately 441 000 deaths due to pancreatic cancer worldwide in 2017. The incidence and death rates of pancreatic cancer for both sexes varied greatly across GBD regions. The 2017 age-standardised death rates for pancreatic cancer were highest in Greenland and Uruguay. The lowest were observed in Bangladesh and São Tomé and Príncipe. Although there was not a clear explanation for such great differences in pancreatic cancer mortality in different regions of the world, this national variation could be attributed to exposure to known or suspected lifestyle and environmental risk factors related to pancreatic cancer as well as scarcity of efficient diagnostic tools in low-income and middle-income countries.14, 15 The substantial increase in worldwide pancreatic cancer suggests that change in ageing populations, especially in low and middle SDI countries, and environmental and behavioural changes, more so than genetic factors, are related to its cause. Differences in pancreatic cancer death and incidence rates across countries could also reflect variation in quality of cancer registry data and tools for pancreatic cancer diagnosis.16, 17

in low and middle SDI countries, and environmental and behavioural changes, more so than genetic factors, are related to its cause. Differences in pancreatic cancer death and incidence rates across countries could also reflect variation in quality of cancer registry data and tools for pancreatic cancer diagnosis.16, 17 Since 1990, regional age-standardised death and incidence rates of pancreatic cancer generally increased with increasing SDI. During the past three decades, these rates were consistently higher in high SDI regions and lower in low SDI regions. Higher incidence of pancreatic cancer in high SDI countries could be due to the ageing population and to lifestyle choices that increase exposure to risk factors; some of the risk factors for pancreatic cancer are more prevalent in high SDI countries than in low ones.2, 18 There was a 2·3 times increase in global number of incident cases and deaths of pancreatic cancer in both sexes from 1990 to 2017, reflecting both ageing and growth of the population, especially in low and middle SDI countries. The age-standardised death and incidence rates increased from 1990 to 2017 at the global level. This increase was probably related to an increase in the prevalence of obesity and diabetes, as reflected by the risk factors of high BMI and fasting plasma glucose, two of the known risk factors for pancreatic cancer.15, 19, 20 This increase occurred despite a mild-to-moderate reduction in smoking rates in high-income countries over the study period.21

to an increase in the prevalence of obesity and diabetes, as reflected by the risk factors of high BMI and fasting plasma glucose, two of the known risk factors for pancreatic cancer.15, 19, 20 This increase occurred despite a mild-to-moderate reduction in smoking rates in high-income countries over the study period.21 The incidence and mortality of pancreatic cancer were somewhat higher in males than in females across age groups lower than 75 years in all years from 1990 to 2017; 51·9% of pancreatic cancer deaths occurred in males and 48·1% in females. This pattern is consistent with other worldwide studies.3, 22 The reasons for lower incidence of pancreatic cancer in females are not completely understood. Females are either less prone to pancreatic cancer or have less exposure to smoking as the main environmental risk factor of pancreatic cancer.4 Smoking in males (25·0%) is five times more frequent than in females (5·4%) worldwide.21

ons for lower incidence of pancreatic cancer in females are not completely understood. Females are either less prone to pancreatic cancer or have less exposure to smoking as the main environmental risk factor of pancreatic cancer.4 Smoking in males (25·0%) is five times more frequent than in females (5·4%) worldwide.21 High-income regions had the highest death and incidence rates of pancreatic cancer for both sexes combined in all years from 1990 to 2017. Similar to other comparable studies, we found that pancreatic cancer is mainly a disease of high-income countries, where overall rates are nearly three times higher than in middle-income and low-income countries, but age-standardised rates are increasing at similar rates among countries with different SDI levels.2, 22 A forecasting study has predicted that pancreatic cancer will escalate from the fourth to the second leading cause of cancer deaths in the USA by 2030,23 and another study forecasted that 111 500 deaths from pancreatic cancer will occur in the European region by 2025, an almost 50% increase in the number of recorded deaths from pancreatic cancer in 2010.24, 25

creatic cancer will escalate from the fourth to the second leading cause of cancer deaths in the USA by 2030,23 and another study forecasted that 111 500 deaths from pancreatic cancer will occur in the European region by 2025, an almost 50% increase in the number of recorded deaths from pancreatic cancer in 2010.24, 25 Pancreatic cancer is typically a disease of older people, since 90% of newly diagnosed patients are aged older than 55 years, with most in their seventh and eighth decades of life.3, 22 Although pancreatic cancer rarely presents before the age of 45 years, the incidence rises sharply thereafter. As survival for chronic diseases improves, the number of older patients diagnosed with pancreatic cancer is increasing. Age is the most important risk factor in the development of pancreatic cancer. We found the number of both deaths and incident cases had a peak in the age group of 65–69 years in males, whereas the peak in females was observed in the age group of 75–79 years. Using a life table approach is necessary to assess differences in exposure and genetic susceptibility among males and females with pancreatic cancer. It has not been reported previously that females experience pancreatic cancer at older ages than males.

the peak in females was observed in the age group of 75–79 years. Using a life table approach is necessary to assess differences in exposure and genetic susceptibility among males and females with pancreatic cancer. It has not been reported previously that females experience pancreatic cancer at older ages than males. Pancreatic cancer remains one of the deadliest cancers, with a 5% 5-year survival rate.22 5-year survival rates from pancreatic cancer have changed little over the past decades. The current rate of 5% is only slightly improved from 3% in 1970.26 Pancreatic cancer is often diagnosed at advanced stages and responds poorly to chemotherapy, leading to low treatment success rates.22

a 5% 5-year survival rate.22 5-year survival rates from pancreatic cancer have changed little over the past decades. The current rate of 5% is only slightly improved from 3% in 1970.26 Pancreatic cancer is often diagnosed at advanced stages and responds poorly to chemotherapy, leading to low treatment success rates.22 The estimated population attributable fraction of pancreatic cancer deaths to tobacco smoking is 11–32%.27 We found that the proportion of age-standardised deaths attributable to smoking decreased slightly from 1990 (26·6% [95% UI 23·8–29·5]) to 2017 (21·1% [18·8–23·7]) in both sexes combined, but it was still higher than the proportion of age-standardised deaths attributable to high fasting plasma glucose (8·9% [2·1–19·6]) and high BMI (6·2% [2·5–11·4]) in 2017. Reductions in the proportion of pancreatic cancer cases attributable to smoking are similar worldwide reductions in smoking rates. Globally, the age-standardised prevalence of daily smoking decreased by 28·4% for males and 34·4% for females between 1990 and 2015; however, four countries had significant annualised increases in smoking prevalence between 2005 and 2015 (Congo and Azerbaijan for males and Kuwait and Timor-Leste for females).21 On the basis of a 2012 study28 by the International Pancreatic Cancer Case-Control Consortium (PanC4; 6507 pancreatic cancer cases, 12 890 controls), former smokers, in comparison with never smokers, had an odds ratio (OR) of 1·2 (95% CI 1·0–1·3), and current smokers, in comparison with never smokers, had an OR of 2·2 (1·7–2·8), for risk of pancreatic cancer, with a trend of significantly increasing risk of pancreatic cancer with increasing number of cigarettes among current smokers (OR 3·4 for ≥35 cigarettes per day, ptrend<0·0001). Risk increased in relation to duration of cigarette smoking up to 40 years of smoking (OR 2·4).28 No trend in risk was observed for age at starting cigarette smoking, whereas risk decreased with increasing time since cigarette cessation, with an OR of 0·98 after 20 years.28 We found that the highest proportion of pancreatic cancer deaths attributable to smoking for both sexes was observed in the 55–84 year age group, which is consistent with the findings from the PanC4.28 The International Agency for Research on Cancer confirmed that smoking is causally associated with pancreatic cancer.29

e found that the highest proportion of pancreatic cancer deaths attributable to smoking for both sexes was observed in the 55–84 year age group, which is consistent with the findings from the PanC4.28 The International Agency for Research on Cancer confirmed that smoking is causally associated with pancreatic cancer.29 Type 2 diabetes has been linked with an excess risk of pancreatic cancer in several studies.30, 31, 32 We found that 8·8% of pancreatic cancer deaths were attributable to high fasting plasma glucose in both sexes in 2017. By comparison, a population study in Italy estimated that 9·7% of pancreatic cancer occurrence was attributable to diabetes.33 The US National Cancer Institute estimated that diabetes is associated with a 1·8 times increased risk of pancreatic cancer.34 From 1980 to 2014, in all countries, diabetes prevalence in adults either increased, especially in low and middle SDI locations, or at best remained unchanged; worldwide, the number of adults with diabetes has quadrupled,19 so it is expected that diabetes will have a greater contribution to pancreatic cancer occurrence in the future.

14, in all countries, diabetes prevalence in adults either increased, especially in low and middle SDI locations, or at best remained unchanged; worldwide, the number of adults with diabetes has quadrupled,19 so it is expected that diabetes will have a greater contribution to pancreatic cancer occurrence in the future. Large studies have indicated a positive association between increasing BMI and risk of pancreatic cancer.18, 35 A pooled study35 of seven prospective cohorts showed that compared with normal weight (BMI 18·5 to <25), the adjusted relative risk for pancreatic cancer was 1·13 for overweight (BMI 25 to <30 kg/m2) and 1·19 for obesity class I (BMI 30 to <35 kg/m2). A pooled analysis from the Pancreatic Cancer Cohort Consortium18 showed that in males, the adjusted OR for pancreatic cancer for the highest versus lowest quartile of BMI was 1·33, and in females it was 1·34. The prevalence of obesity (BMI ≥30 kg/m2) is increasing at an alarming rate in many parts of the world. The number of obese people has risen globally from 105 million in 1975 to 641 million in 2014. Since 1975, the prevalence of obese males has more than tripled, and that of obese females has more than doubled.36 More than 2 billion people are overweight, and a third of them are obese. By 2025, global obesity prevalence is projected to reach 18% in males and surpass 21% in females; severe obesity is likely to surpass 6% in males and 9% in females.19 We found that 6·2% of pancreatic cancer in both sexes (5·0% in males, 7·4% in females) was attributable to obesity, which is inconsistent with estimated population attributable fractions (3–16%).27 Although obesity carries a modest risk for pancreatic cancer, its rapid increase makes it a serious risk factor for pancreatic cancer, especially in females, who are more commonly obese than males.

in females) was attributable to obesity, which is inconsistent with estimated population attributable fractions (3–16%).27 Although obesity carries a modest risk for pancreatic cancer, its rapid increase makes it a serious risk factor for pancreatic cancer, especially in females, who are more commonly obese than males. Future strategies should include comprehensive policies to control tobacco use and reduce the burden of obesity and diabetes across the world. Additionally, efforts must be made to identify other modifiable risk factors for pancreatic cancer, such as opium use in North Africa and the Middle East.37, 38

in females) was attributable to obesity, which is inconsistent with estimated population attributable fractions (3–16%).27 Although obesity carries a modest risk for pancreatic cancer, its rapid increase makes it a serious risk factor for pancreatic cancer, especially in females, who are more commonly obese than males. Future strategies should include comprehensive policies to control tobacco use and reduce the burden of obesity and diabetes across the world. Additionally, efforts must be made to identify other modifiable risk factors for pancreatic cancer, such as opium use in North Africa and the Middle East.37, 38 This study has several limitations. Generally, as for estimation of all diseases and cancers in the GBD study, the major limitation of the current study is the lack of high-quality data in many regions and countries, particularly in low-income locations. Although we did a sensitive search to take advantage of all available data sources comprising cancer registries and vital registration systems, in many locations data were either sparse or entirely unavailable. For estimating pancreatic cancer burden in these locations, we had to base our estimations on covariates and spatiotemporal smoothing. Ascertainment bias, detection bias, and diagnostic inaccuracy were additional limitations in low-income locations. The reported high incidence and mortality of pancreatic cancer in high-income versus low-income locations might be partly related to availability of accurate diagnostic modalities for pancreatic cancer and richness of data in high-income regions. Additionally, all of the general limitations in estimating the burden attributable to risk factors through comparative risk assessment methods were challenging. Finally, because of lags in data availability, recent estimates of trends relied on covariates and past trends, leading to wider UIs.

a in high-income regions. Additionally, all of the general limitations in estimating the burden attributable to risk factors through comparative risk assessment methods were challenging. Finally, because of lags in data availability, recent estimates of trends relied on covariates and past trends, leading to wider UIs. Pancreatic cancer deaths and incidence more than doubled over the study period. Much of this increase was due to increases in population and longevity, but even after accounting for population changes, incidence and death rates increased from 1990 to 2017, probably due to changes in associated risk factors. Pancreatic cancer is an aggressive cancer, predicted to become the second leading cause of cancer deaths in some regions. It often presents in old age, at an advanced stage, and has a poor prognosis. Major risk factors associated with pancreatic cancer (smoking, diabetes, and obesity) are potentially modifiable, affording a unique opportunity for preventing one of the deadliest cancers. The results of our study can be used by policy makers to allocate resources efficiently for developing methods for early diagnosis of pancreatic cancer, reducing its modifiable risk factors, and evaluating novel treatment strategies to reduce its case-fatality rate by proper treatment strategies. This online publication has been corrected. The corrected version first appeared at thelancet.com/gastrohep on Feb 12, 2020 Supplementary Material Supplementary appendix

Pancreatic cancer deaths and incidence more than doubled over the study period. Much of this increase was due to increases in population and longevity, but even after accounting for population changes, incidence and death rates increased from 1990 to 2017, probably due to changes in associated risk factors. Pancreatic cancer is an aggressive cancer, predicted to become the second leading cause of cancer deaths in some regions. It often presents in old age, at an advanced stage, and has a poor prognosis. Major risk factors associated with pancreatic cancer (smoking, diabetes, and obesity) are potentially modifiable, affording a unique opportunity for preventing one of the deadliest cancers. The results of our study can be used by policy makers to allocate resources efficiently for developing methods for early diagnosis of pancreatic cancer, reducing its modifiable risk factors, and evaluating novel treatment strategies to reduce its case-fatality rate by proper treatment strategies. This online publication has been corrected. The corrected version first appeared at thelancet.com/gastrohep on Feb 12, 2020 Supplementary Material Supplementary appendix Acknowledgments This study was supported by the Bill & Melinda Gates Foundation. AB is supported by the Public Health Agency of Canada. FC and EF acknowledge UID/MULTI/04378/2019 and UID/QUI/50006/2019 support with funding from Fundação para a Ciência e a Tecnologia/Ministério da Ciência, Tecnologia e Ensino Superior (FCT/MCTES) through Portuguese national funds. TM acknowledges institutional support from the Competence Cluster for Nutrition and Cardiovascular Health (nutriCARD), Jena-Halle-Leipzig. AMS acknowledges support by a fellowship from the Egyptian Fulbright Mission Program (EFMP). RT-S acknowledges support in part by grant number PROMETEOII/2015/021 from Generalitat Valenciana and the national grant PI17/00719 from ISCIII-FEDER.

ter for Nutrition and Cardiovascular Health (nutriCARD), Jena-Halle-Leipzig. AMS acknowledges support by a fellowship from the Egyptian Fulbright Mission Program (EFMP). RT-S acknowledges support in part by grant number PROMETEOII/2015/021 from Generalitat Valenciana and the national grant PI17/00719 from ISCIII-FEDER. Editorial note: The Lancet Group takes a neutral position with respect to territorial claims in published maps and institutional affiliations.

ter for Nutrition and Cardiovascular Health (nutriCARD), Jena-Halle-Leipzig. AMS acknowledges support by a fellowship from the Egyptian Fulbright Mission Program (EFMP). RT-S acknowledges support in part by grant number PROMETEOII/2015/021 from Generalitat Valenciana and the national grant PI17/00719 from ISCIII-FEDER. Editorial note: The Lancet Group takes a neutral position with respect to territorial claims in published maps and institutional affiliations. GBD 2017 Pancreatic Cancer Collaborators Akram Pourshams, Sadaf G Sepanlou, Kevin S Ikuta, Catherine Bisignano, Saeid Safiri, Gholamreza Roshandel, Mehdi Sharif, Morteza Khatibian, Christina Fitzmaurice, Molly R Nixon, Nooshin Abbasi, Mohsen Afarideh, Elham Ahmadian, Tomi Akinyemiju, Fares Alahdab, Tahiya Alam, Vahid Alipour, Christine A Allen, Nahla Hamed Anber, Alireza Ansari-Moghaddam, Jalal Arabloo, Alaa Badawi, Mojtaba Bagherzadeh, Yaschilal Muche Belayneh, Belete Biadgo, Ali Bijani, Antonio Biondi, Tone Bjørge, Antonio M Borzì, Cristina Bosetti, Andrey Nikolaevich Briko, Nikolay Ivanovich Briko, Giulia Carreras, Félix Carvalho, Jee-Young J Choi, Dinh-Toi Chu, Anh Kim Dang, Ahmad Daryani, Dragos Virgil Davitoiu, Gebre Teklemariam Demoz, Rupak Desai, Subhojit Dey, Hoa Thi Do, Huyen Phuc Do, Aziz Eftekhari, Alireza Esteghamati, Farshad Farzadfar, Eduarda Fernandes, Irina Filip, Florian Fischer, Masoud Foroutan, Mohamed M Gad, Silvano Gallus, Birhanu Geta, Giuseppe Gorini, Nima Hafezi-Nejad, James D Harvey, Milad Hasankhani, Amir Hasanzadeh, Soheil Hassanipour, Simon I Hay, Hagos D Hidru, Chi Linh Hoang, Sorin Hostiuc, Mowafa Househ, Olayinka Stephen Ilesanmi, Milena D Ilic, Seyed Sina Naghibi Irvani, Nader Jafari Balalami, Spencer L James, Farahnaz Joukar, Amir Kasaeian, Tesfaye Dessale Kassa, Andre Pascal Kengne, Rovshan Khalilov, Ejaz Ahmad Khan, Amir Khater, Fatemeh Khosravi Shadmani, Jonathan M Kocarnik, Hamidreza Komaki, Ai Koyanagi, Vivek Kumar, Carlo La Vecchia, Platon D Lopukhov, Farzad Manafi, Navid Manafi, Ana-Laura Manda, Fariborz Mansour-Ghanaei, Dhruv Mehta, Varshil Mehta, Toni Meier, Hagazi Gebre Meles, Getnet Mengistu, Tomasz Miazgowski, Mehdi Mohamadnejad, Abdollah Mohammadian-Hafshejani, Milad Mohammadoo-Khorasani, Shafiu Mohammed, Farnam Mohebi, Ali H Mokdad, Lorenzo Monasta, Maryam Moossavi, Rahmatollah Moradzadeh, Gurudatta Naik, Ionut Negoi, Cuong Tat Nguyen, Long Hoang Nguyen, Trang Huyen Nguyen, Andrew T Olagunju, Tinuke O Olagunju, Alyssa Pennini, Mohammad Rabiee, Navid Rabiee, Amir Radfar, Mahdi Rahimi, Goura Kishor Rath, David Laith Rawaf, Salman Rawaf, Robert C Reiner Jr, Nima Rezaei, Aziz Rezapour, Anas M Saad, Seyedmohammad Saadat

aik, Ionut Negoi, Cuong Tat Nguyen, Long Hoang Nguyen, Trang Huyen Nguyen, Andrew T Olagunju, Tinuke O Olagunju, Alyssa Pennini, Mohammad Rabiee, Navid Rabiee, Amir Radfar, Mahdi Rahimi, Goura Kishor Rath, David Laith Rawaf, Salman Rawaf, Robert C Reiner Jr, Nima Rezaei, Aziz Rezapour, Anas M Saad, Seyedmohammad Saadat agah, Amirhossein Sahebkar, Hamideh Salimzadeh, Abdallah M Samy, Juan Sanabria, Arash Sarveazad, Monika Sawhney, Mario Sekerija, Pavel Shabalkin, Masood Ali Shaikh, Rajesh Sharma, Sara Sheikhbahaei, Reza Shirkoohi, Sudeep K Siddappa Malleshappa, Mekonnen Sisay, Kjetil Soreide, Sergey Soshnikov, Rasoul Sotoudehmanesh, Vladimir I Starodubov, Michelle L Subart, Rafael Tabarés-Seisdedos, Degena Bahray Bahrey Tadesse, Eugenio Traini, Bach Xuan Tran, Khanh Bao Tran, Irfan Ullah, Marco Vacante, Amir Vahedian-Azimi, Elena Varavikova, Ronny Westerman, Dawit Zewdu Wondafrash, Rixing Xu, Naohiro Yonemoto, Vesna Zadnik, Zhi-Jiang Zhang, Reza Malekzadeh*, and Mohsen Naghavi*. *These authors jointly supervised the study.

agah, Amirhossein Sahebkar, Hamideh Salimzadeh, Abdallah M Samy, Juan Sanabria, Arash Sarveazad, Monika Sawhney, Mario Sekerija, Pavel Shabalkin, Masood Ali Shaikh, Rajesh Sharma, Sara Sheikhbahaei, Reza Shirkoohi, Sudeep K Siddappa Malleshappa, Mekonnen Sisay, Kjetil Soreide, Sergey Soshnikov, Rasoul Sotoudehmanesh, Vladimir I Starodubov, Michelle L Subart, Rafael Tabarés-Seisdedos, Degena Bahray Bahrey Tadesse, Eugenio Traini, Bach Xuan Tran, Khanh Bao Tran, Irfan Ullah, Marco Vacante, Amir Vahedian-Azimi, Elena Varavikova, Ronny Westerman, Dawit Zewdu Wondafrash, Rixing Xu, Naohiro Yonemoto, Vesna Zadnik, Zhi-Jiang Zhang, Reza Malekzadeh*, and Mohsen Naghavi*. *These authors jointly supervised the study. Affiliations Digestive Diseases Research Institute (Prof A Pourshams MD, S G Sepanlou MD, G Roshandel PhD, M Khatibian MD, M Mohamadnejad MD, H Salimzadeh PhD, Prof R Sotoudehmanesh BHlthSci, Prof R Malekzadeh MD), Cancer Biology Research Center (R Shirkoohi PhD), Cancer Research Institute (R Shirkoohi PhD), Department of Cardiology (S Saadatagah MD), Department of Microbiology (A Hasanzadeh PhD), Endocrinology and Metabolism Research Center (M Afarideh MD, Prof A Esteghamati MD, S Sheikhbahaei MD), Hematologic Malignancies Research Center (A Kasaeian PhD), Hematology-Oncology and Stem Cell Transplantation Research Center (A Kasaeian PhD), Iran National Institute of Health Research (F Mohebi MD), Liver and Pancreatobiliary Diseases Research Center (M Mohamadnejad MD), Non-communicable Diseases Research Center (F Farzadfar MD, F Mohebi MD), Research Center for Immunodeficiencies (Prof N Rezaei PhD), School of Medicine (N Hafezi-Nejad MD), Tehran University of Medical Sciences, Tehran, Iran; Non-communicable Diseases Research Center (S G Sepanlou MD, Prof R Malekzadeh MD), Shiraz University of Medical Sciences, Shiraz, Iran; Institute for Health Metrics and Evaluation (K S Ikuta MD, C Bisignano MPH, C Fitzmaurice MD, M R Nixon PhD, T Alam MPH, C A Allen BA, J D Harvey BS, Prof S I Hay FMedSci, S L James MD, J M Kocarnik PhD, Prof A H Mokdad PhD, A Pennini MSc, R C Reiner Jr PhD, M L Subart BA, M L Subart BA, R Xu BS, Prof M Naghavi MD), Division of Allergy and Infectious Diseases (K S Ikuta MD), Division of Hematology (C Fitzmaurice MD), Department of Health Metrics Sciences, School of Medicine (Prof S I Hay FMedSci, Prof A H Mokdad PhD, R C Reiner Jr PhD, Prof M Naghavi MD), University of Washington, Seattle, WA, USA; Aging Research Institute (S Safiri PhD), Department of Community Medicine (S Safiri PhD), Department of Pharmacology and Toxicology (E Ahmadian PhD, A Eftekhari PhD), Drug Applied Research Center (M Rahimi PhD), School of Nutrition and Food Sciences (M Hasankhani MSc), Tabriz University of Medical Sciences, Tabriz, Iran; Golestan Research Center of Gastroenterology and Hepatology (G Roshandel PhD), Golestan University of Medical Sciences, Gorgan, Ir

E Ahmadian PhD, A Eftekhari PhD), Drug Applied Research Center (M Rahimi PhD), School of Nutrition and Food Sciences (M Hasankhani MSc), Tabriz University of Medical Sciences, Tabriz, Iran; Golestan Research Center of Gastroenterology and Hepatology (G Roshandel PhD), Golestan University of Medical Sciences, Gorgan, Ir an; Department of Basic Sciences (Prof M Sharif PhD), Department of Laboratory Sciences (Prof M Sharif PhD), Islamic Azad University, Sari, Iran; Montreal Neurological Institute (N Abbasi MD), McGill University, Montreal, QC, Canada; Department of Physiology (R Khalilov PhD), Institute of Radiation Problems of Azerbaijan, Baku State University, Baku, Azerbaijan (E Ahmadian PhD); Department of Population Health Sciences (T Akinyemiju PhD), Duke Global Health Institute (T Akinyemiju PhD), Duke University, Durham, NC, USA; Evidence-Based Practice Center (F Alahdab MD), Mayo Clinic Foundation for Medical Education and Research, Rochester, MN, USA; Colorectal Research Center (A Sarveazad PhD), Health Economics Department (V Alipour PhD), Health Management and Economics Research Center (V Alipour PhD, J Arabloo PhD, A Rezapour PhD), Ophthalmology Department (N Manafi MD), Iran University of Medical Sciences, Tehran, Iran; Faculty of Medicine (N H Anber PhD), Mansoura University, Mansoura, Egypt (N H Anber PhD); Department of Epidemiology and Biostatistics (Prof A Ansari-Moghaddam PhD), Health Promotion Research Center, Zahedan, Iran; Public Health Risk Sciences Division (A Badawi PhD), Public Health Agency of Canada, Toronto, ON, Canada; Department of Nutritional Sciences (A Badawi PhD), Joint Centre for Bioethics (F Manafi MD), University of Toronto, Toronto, ON, Canada; Department of Chemistry (Prof M Bagherzadeh PhD, N Rabiee PhD), Sharif University of Technology, Tehran, Iran; Department of Pharmacy (Y M Belayneh MSc, B Geta MSc, G Mengistu MSc), Wollo University, Dessie, Ethiopia; Department of Clinical Chemistry (B Biadgo MSc), University of Gondar, Gondar, Ethiopia; Social Determinants of Health Research Center (A Bijani PhD), Babol University of Medical Sciences, Babol, Iran; Department of Clinical and Molecular Biomedicine (MEDBIO) (A M Borzì MD), Department of General Surgery and Medical-Surgical Specialties (Prof A Biondi PhD, M Vacante PhD), University of Catania, Catania, Italy; Department of Clinical Medicine (Prof K Soreide PhD), Department of Global Public Health and Primary Care (Prof T Bjørge PhD), University of Bergen, Bergen, Norway; Cancer Registr

), Department of General Surgery and Medical-Surgical Specialties (Prof A Biondi PhD, M Vacante PhD), University of Catania, Catania, Italy; Department of Clinical Medicine (Prof K Soreide PhD), Department of Global Public Health and Primary Care (Prof T Bjørge PhD), University of Bergen, Bergen, Norway; Cancer Registr y of Norway, Oslo, Norway (Prof T Bjørge PhD); Department of Environmental Health Science (S Gallus DSc), Department of Oncology (C Bosetti PhD), Mario Negri Institute for Pharmacological Research, Milan, Italy; Department of Biomedical Technologies (A N Briko MSc), Bauman Moscow State Technical University, Moscow, Russia; Department of Epidemiology and Evidence-Based Medicine (Prof N I Briko DSc, P D Lopukhov Cand of Sci [Med]), I.M.

f Oncology (C Bosetti PhD), Mario Negri Institute for Pharmacological Research, Milan, Italy; Department of Biomedical Technologies (A N Briko MSc), Bauman Moscow State Technical University, Moscow, Russia; Department of Epidemiology and Evidence-Based Medicine (Prof N I Briko DSc, P D Lopukhov Cand of Sci [Med]), I.M. Sechenov First Moscow State Medical University, Moscow, Russia; Institute for Cancer Research, Prevention and Clinical Network, Oncological Network, Prevention and Research Institute (ISPRO), Florence, Italy (G Carreras PhD); Applied Molecular Biosciences Unit (Prof F Carvalho PhD), Institute of Public Health (Prof F Carvalho PhD), REQUIMTE/LAQV (Prof E Fernandes PhD), University of Porto, Porto, Portugal; Department of Biochemistry, Biomedical Science (J J Choi PhD), Seoul National University Hospital, Seoul, South Korea; Faculty of Biology (D Chu PhD), Hanoi National University of Education, Hanoi, Vietnam; Institute for Global Health Innovations (A K Dang MD, C T Nguyen MPH), Duy Tan University, Hanoi, Vietnam; Toxoplasmosis Research Center (Prof A Daryani PhD), Mazandaran University of Medical Sciences, Sari, Iran; Department of General Surgery (I Negoi PhD, D V Davitoiu PhD), Emergency Hospital of Bucharest (I Negoi PhD), Faculty of Dentistry, Department of Legal Medicine and Bioethics (S Hostiuc PhD), Carol Davila University of Medicine and Pharmacy, Bucharest, Romania; Department of Surgery (D V Davitoiu PhD), Clinical Emergency Hospital Sf. Pantelimon, Bucharest, Romania; School of Pharmacy (G T Demoz MPharm), Aksum University, Aksum, Ethiopia; Department of Pharmacology (D Z Wondafrash MSc), Addis Ababa University, Addis Ababa, Ethiopia (G T Demoz MPharm); Division of Cardiology (R Desai MBBS), Atlanta Veterans Affairs Medical Center, Decatur, GA, USA; Disha Foundation, Gurgaon, India (S Dey PhD); Center of Excellence in Behavioral Medicine (H P Do PhD, C L Hoang BMedSc, L H Nguyen PhD, T H Nguyen BMedSc), Center of Excellence in Public Health Nutrition (H T Do MD), Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam; Department of Microbiology (A Hasanzadeh PhD), Pharmacology and Toxicology Department (A Eftekhari PhD), Maragheh University of Medical Sciences, Maragheh, Iran; Psychiatry Department (I Filip MD), Kaiser Permanente, Fontana, CA, USA; College of Graduate Health Sciences (A Radfar MD), Department of Health Sciences (I Filip MD), A.T.

Microbiology (A Hasanzadeh PhD), Pharmacology and Toxicology Department (A Eftekhari PhD), Maragheh University of Medical Sciences, Maragheh, Iran; Psychiatry Department (I Filip MD), Kaiser Permanente, Fontana, CA, USA; College of Graduate Health Sciences (A Radfar MD), Department of Health Sciences (I Filip MD), A.T. Still University, Mesa, AZ, USA; Department of Public Health Medicine (F Fischer PhD), Bielefeld University, Bielefeld, Germany; Abadan School of Medical Sciences, Abadan, Iran (M Foroutan PhD); Department of Cardiovascular Medicine (M M Gad MD), Cleveland Clinic, Cleveland, OH, USA; Gillings School of Global Public Health (M M Gad MD), University of North Carolina Chapel Hill, Chapel Hill, NC, USA; Occupational and Environmental Epidemiology Section (G Gorini MD), Cancer Prevention and Research Institute, Florence, Italy; Department of Radiology and Radiological Sciences (N Hafezi-Nejad MD, S Sheikhbahaei MD), Johns Hopkins University, Baltimore, MD, USA; Gastrointestinal and Liver Disease Research Center (S Hassanipour PhD, F Joukar PhD, Prof F Mansour-Ghanaei PhD), Guilan University of Medical Sciences, Rasht, Iran (S Hassanipour PhD); Department of Epidemiology (H D Hidru MPH), Adigrat University, Adigrat, Ethiopia; Clinical Legal Medicine (S Hostiuc PhD), National Institute of Legal Medicine Mina Minovici, Bucharest, Romania; Division of Information and Computing Technology (Prof M Househ PhD), Hamad Bin Khalifa University, Doha, Qatar; Qatar Foundation for Education, Science, and Community Development, Doha, Qatar (Prof M Househ PhD); Department of Community Medicine (O S Ilesanmi PhD), University of Ibadan, Ibadan, Nigeria; Department of Epidemiology (Prof M D Ilic PhD), University of Kragujevac, Kragujevac, Serbia; Research Institute for Endocrine Sciences (S N Irvani MD), Shahid Beheshti University of Medical Sciences, Tehran, Iran; Department of Psychosis (N Jafari Balalami PhD), Babol Nushirvani University of Technology, Babol, Iran; Clinical Pharmacy Unit (T D Kassa MSc), Department of Pharmacology and Toxicology (D Z Wondafrash MSc), Mekelle University, Mekelle, Ethiopia (H G Meles MPH); Non-communicable Diseases Research Unit (Prof A P Kengne PhD), Medical Research Council South Africa, Cape Town, South Africa; Department of Medicine (Prof A P Kengne PhD), University of Cape Town, Cape Town, South Africa; Epidemiology and Biostatistics Department (E A Khan MPH), Health Services Academy, Islamabad, Pakistan; Internal Medicine and Gastroenterology De

P Kengne PhD), Medical Research Council South Africa, Cape Town, South Africa; Department of Medicine (Prof A P Kengne PhD), University of Cape Town, Cape Town, South Africa; Epidemiology and Biostatistics Department (E A Khan MPH), Health Services Academy, Islamabad, Pakistan; Internal Medicine and Gastroenterology De partment (A Khater MD), National Hepatology and Tropical Research Institute, Cairo, Egypt; Department of Epidemiology (F Khosravi Shadmani PhD), Kermanshah University of Medical Sciences, Kermanshah, Iran; Public Health Sciences Division (J M Kocarnik PhD), Fred Hutchinson Cancer Research Center, Seattle, WA, USA; Neurophysiology Research Center (H Komaki MD), Hamadan University of Medical Sciences, Hamadan, Iran; Brain Engineering Research Center (H Komaki MD), Institute for Research in Fundamental Sciences, Tehran, Iran; CIBERSAM (A Koyanagi MD), San Juan de Dios Sanitary Park, Sant Boi de Llobregat, Spain; Catalan Institution for Research and Advanced Studies (ICREA), Barcelona, Spain (A Koyanagi MD); Department of Medicine, Brigham and Women's Hospital (V Kumar MD), Harvard University, Boston, MA, USA; Clinical Medicine and Community Health, University of Milan, Milano, Italy (Prof C La Vecchia MD); Department of Nursing, Aksum University, Aksum, Ethiopia (D B B Tadesse CAP); Ophthalmology Department (N Manafi MD), University of Manitoba, Winnipeg, MB, Canada; Surgery Department (A Manda MD), Emergency University Hospital Bucharest, Bucharest, Romania; Division of Gastroenterology and Hepatobiliary Disease (D Mehta MD), New York Medical College, Valhalla, NY, USA; Department of Internal Medicine (V Mehta MD), SevenHills Hospital, Mumbai, India; Institute for Agricultural and Nutritional Sciences (T Meier PhD), Martin Luther University Halle-Wittenberg, Halle, Germany; Innovation Office (T Meier PhD), Competence Cluster for Nutrition and Cardiovascular Health (nutriCARD), Halle, Germany; School of Pharmacy (G Mengistu MSc, M Sisay MSc), Haramaya University, Harar, Ethiopia; Department of Propedeutics of Internal Diseases & Arterial Hypertension (Prof T Miazgowski MD), Pomeranian Medical University, Szczecin, Poland; Department of Epidemiology and Biostatistics (A Mohammadian-Hafshejani PhD), Shahrekord University of Medical Sciences, Shahrekord, Iran; Department of Clinical Biochemistry (M Mohammadoo-Khorasani PhD), Tarbiat Modares University, Tehran, Iran; Health Systems and Policy Research Unit (S Mohammed PhD), Ahmadu Bello University, Zaria, Nigeria; Inst

d Biostatistics (A Mohammadian-Hafshejani PhD), Shahrekord University of Medical Sciences, Shahrekord, Iran; Department of Clinical Biochemistry (M Mohammadoo-Khorasani PhD), Tarbiat Modares University, Tehran, Iran; Health Systems and Policy Research Unit (S Mohammed PhD), Ahmadu Bello University, Zaria, Nigeria; Inst itute of Public Health (S Mohammed PhD), Heidelberg University, Heidelberg, Germany; Clinical Epidemiology and Public Health Research Unit (L Monasta DSc, E Traini MSc), Burlo Garofolo Institute for Maternal and Child Health, Trieste, Italy; Department of Molecular Medicine (M Moossavi PhD), Birjand University of Medical Sciences, Birjand, Iran; Department of Epidemiology (R Moradzadeh PhD), Arak University of Medical Sciences, Arak, Iran; Department of Epidemiology (G Naik MPH), University of Alabama at Birmingham, Birmingham, AL, USA; Department of Pathology and Molecular Medicine (T O Olagunju MD), Department of Psychiatry and Behavioural Neurosciences (A T Olagunju MD), McMaster University, Hamilton, ON, Canada; Department of Psychiatry (A T Olagunju MD), University of Lagos, Lagos, Nigeria; Biomedical Engineering Department (Prof M Rabiee PhD), Amirkabir University of Technology, Tehran, Iran; Medichem, Barcelona, Spain (A Radfar MD); Department of Radiation Oncology (Prof G K Rath MD), All India Institute of Medical Sciences, New Delhi, India; Department of Primary Care and Public Health (Prof S Rawaf MD), WHO Collaborating Centre for Public Health Education and Training (D L Rawaf MD), Imperial College London, London, UK; University College London Hospitals, London, UK (D L Rawaf MD); Academic Public Health Department (Prof S Rawaf MD), Public Health England, London, UK; Network of Immunity in Infection, Malignancy and Autoimmunity (NIIMA) (Prof N Rezaei PhD), Universal Scientific Education and Research Network (USERN), Tehran, Iran; Department of Entomology (A M Samy PhD), Faculty of Medicine (A M Saad MBBCh), Ain Shams University, Cairo, Egypt; Biotechnology Research Center (A Sahebkar PhD), Neurogenic Inflammation Research Center (A Sahebkar PhD), Mashhad University of Medical Sciences, Mashhad, Iran; Department of Surgery (Prof J Sanabria MD), Marshall University, Huntington, WV, USA; Department of Nutrition and Preventive Medicine (Prof J Sanabria MD), Case Western Reserve University, Cleveland, OH, USA; Department of Public Health Sciences (M Sawhney PhD), University of North Carolina at Charlotte, Charlotte, NC, USA; Department of Medical Statistic

a MD), Marshall University, Huntington, WV, USA; Department of Nutrition and Preventive Medicine (Prof J Sanabria MD), Case Western Reserve University, Cleveland, OH, USA; Department of Public Health Sciences (M Sawhney PhD), University of North Carolina at Charlotte, Charlotte, NC, USA; Department of Medical Statistic s, Epidemiology and Medical Informatics (M Sekerija PhD), University of Zagreb, Zagreb, Croatia; Division of Epidemiology and Prevention of Chronic Noncommunicable Diseases (M Sekerija PhD), Croatian Institute of Public Health, Zagreb, Croatia; Cancer Research and Development Department (Prof P Shabalkin MD), N.N.

a MD), Marshall University, Huntington, WV, USA; Department of Nutrition and Preventive Medicine (Prof J Sanabria MD), Case Western Reserve University, Cleveland, OH, USA; Department of Public Health Sciences (M Sawhney PhD), University of North Carolina at Charlotte, Charlotte, NC, USA; Department of Medical Statistic s, Epidemiology and Medical Informatics (M Sekerija PhD), University of Zagreb, Zagreb, Croatia; Division of Epidemiology and Prevention of Chronic Noncommunicable Diseases (M Sekerija PhD), Croatian Institute of Public Health, Zagreb, Croatia; Cancer Research and Development Department (Prof P Shabalkin MD), N.N. Blokhin National Medical Cancer Research Center, Moscow, Russia; independent consultant, Karachi, Pakistan (M A Shaikh MD); University School of Management and Entrepreneurship (R Sharma PhD), Delhi Technological University, New Delhi, India; Department of Hematology-Oncology (S K Siddappa Malleshappa MD), Baystate Medical Center, Springfield, MA, USA; Department of Gastrointestinal Surgery (Prof K Soreide PhD), Stavanger University Hospital, Stavanger, Norway; Research Development Department (S Soshnikov PhD), Central Research Institute of Cytology and Genetics (E Varavikova PhD), Federal Research Institute for Health Organization and Informatics of the Ministry of Health (FRIHOI), Moscow, Russia (Prof V I Starodubov DSc); Department of Medicine (Prof R Tabarés-Seisdedos PhD), University of Valencia, Valencia, Spain; Carlos III Health Institute (Prof R Tabarés-Seisdedos PhD), Biomedical Research Networking Center for Mental Health Network (CiberSAM), Madrid, Spain; Department of Health Economics (B X Tran PhD), Hanoi Medical University, Hanoi, Vietnam; Department of Molecular Medicine and Pathology (K B Tran MD), University of Auckland, Auckland, New Zealand; Department of Clinical Hematology and Toxicology (K B Tran MD), Military Medical University, Hanoi, Vietnam; Gomal Center of Biochemistry and Biotechnology (I Ullah PhD), Gomal University, Dera Ismail Khan, Pakistan; TB Culture Laboratory (I Ullah PhD), Mufti Mehmood Memorial Teaching Hospital, Dera Ismail Khan, Pakistan; Baqiyatallah University of Medical Sciences, Tehran, Iran (A Vahedian-Azimi PhD); Competence Center of Mortality-Follow-Up, German National Cohort (R Westerman DSc), Federal Institute for Population Research, Wiesbaden, Germany; Department of Psychopharmacology (N Yonemoto MPH), National Center of Neurology and Psychiatry, Tokyo, Japan; Epidemiology and Cancer Registry Sector (Prof V Zadnik PhD), Institute of Oncology Ljubljana, Ljubljana, Slovenia; and Department of Preventive Medicine (Z Zhang PhD), Wuhan University, Wuhan, China.

, Germany; Department of Psychopharmacology (N Yonemoto MPH), National Center of Neurology and Psychiatry, Tokyo, Japan; Epidemiology and Cancer Registry Sector (Prof V Zadnik PhD), Institute of Oncology Ljubljana, Ljubljana, Slovenia; and Department of Preventive Medicine (Z Zhang PhD), Wuhan University, Wuhan, China. Contributors AP, SGS, SS, GR, CF, MN, and CJLM prepared the first draft. RM, MM, RS, MK, CF, MN, and CJLM provided overall guidance. RM, SGS, AP, CF, MN, and CJLM managed the project. SGS, SS, and CF analysed data. RM, SGS, AK, RS, SS, CF, MN, and CJLM finalised the manuscript on the basis of comments from other authors and reviewer feedback. All other authors provided data, developed models, reviewed results, provided guidance on methods, or reviewed and contributed to the manuscript. Declaration of interests SLJ reports grants from Sanofi Pasteur, outside the submitted work. All other authors declare no competing interests.

Introduction Stomach cancer is an important contributor to the global burden of cancer,1 and less than a century ago it was the most common cancer in the world.2 Since then, the incidence and mortality rates of stomach cancer have fallen.3 However, this trend has shown signs of change; for example, some researchers suggest that in the USA, the rates of stomach cancer might be increasing among younger age groups (ie, <50 years) and predict that this increase might reverse the overall decline in the incidence of stomach cancer.4 More than 90% of stomach cancers are adenocarcinomas, and, depending on whether the tumour is located near the gastro-oesophageal junction (cardia) or away from it, they are subdivided into cardia and non-cardia tumours, respectively.1 The decreasing trend of stomach cancer incidence and mortality in most populations is due to the falling rates of non-cardia stomach cancer and has been linked to a decline in Helicobacter pylori infection rates.5, 6 H pylori is a known carcinogen7 for non-cardia stomach cancer, which probably once infected most adults during their life course.8 Improved socio-economic status, hygienic practices, and widespread antibiotic use have led to a decrease in infection rates.9

linked to a decline in Helicobacter pylori infection rates.5, 6 H pylori is a known carcinogen7 for non-cardia stomach cancer, which probably once infected most adults during their life course.8 Improved socio-economic status, hygienic practices, and widespread antibiotic use have led to a decrease in infection rates.9 The epidemiology of stomach cancer has substantial geographical heterogeneity, and its incidence can vary 5-fold to 10-fold between high-risk and low-risk countries.10 Part of this geographical variation correlates with H pylori infection rates across populations; however, a number of environmental factors also contribute to the risk of stomach cancer. Cigarette smoking has been shown to be a risk factor for both cardia and non-cardia stomach cancers.11 Evidence suggests that salt and salt-preserved foods might increase the risk of stomach cancer.12, 13 Both types of stomach cancer are more common among males, which might be due to the higher prevalence of risk factors, such as smoking, or hormonal factors contributing to this difference.2 Research in context Evidence before this study

The epidemiology of stomach cancer has substantial geographical heterogeneity, and its incidence can vary 5-fold to 10-fold between high-risk and low-risk countries.10 Part of this geographical variation correlates with H pylori infection rates across populations; however, a number of environmental factors also contribute to the risk of stomach cancer. Cigarette smoking has been shown to be a risk factor for both cardia and non-cardia stomach cancers.11 Evidence suggests that salt and salt-preserved foods might increase the risk of stomach cancer.12, 13 Both types of stomach cancer are more common among males, which might be due to the higher prevalence of risk factors, such as smoking, or hormonal factors contributing to this difference.2 Research in context Evidence before this study The age-standardised incidence and death rates of stomach cancer have declined in most parts of the world, but it remains a major health problem in many countries. Understanding the current burden of stomach cancer and trends across different locations is essential for formulating effective preventive strategies. The International Agency for Research on Cancer has regularly provided cancer estimates in the Global Cancer Incidence, Mortality and Prevalence (GLOBOCAN) project; however, GLOBOCAN does not provide estimates over time for all locations, correlations with risk factors, or estimates for disability-adjusted life-years (DALYs). We used estimates from the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2017 to examine trends of incidence, mortality, and burden of disease across 195 countries and territories in seven super-regions and 21 regions from 1990 to 2017.

ctors, or estimates for disability-adjusted life-years (DALYs). We used estimates from the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2017 to examine trends of incidence, mortality, and burden of disease across 195 countries and territories in seven super-regions and 21 regions from 1990 to 2017. Added value of this study Using results from GBD 2017, we studied the global, regional, and national trends of stomach cancer using comprehensive data collected from sources around the world, and reported geographical variation and trends over time. Our study adds value to the available evidence because, to our knowledge, it provides the most comprehensive assessment of the burden of stomach cancer by age, sex, socio-demographic status, and location over time. Implications of all the available evidence

Using results from GBD 2017, we studied the global, regional, and national trends of stomach cancer using comprehensive data collected from sources around the world, and reported geographical variation and trends over time. Our study adds value to the available evidence because, to our knowledge, it provides the most comprehensive assessment of the burden of stomach cancer by age, sex, socio-demographic status, and location over time. Implications of all the available evidence Stomach cancer is a significant cause of morbidity and mortality in many parts or the world, and the total numbers of incident cases and deaths are increasing worldwide. East Asia, particularly China, contributes the largest number of incident cases, deaths, and DALYs of stomach cancer in the world. The age-standardised rates of incidence and death, however, have declined steadily, and are much lower than the corresponding rates in 1990. The falling rates globally, and within each region, are associated with rises in the Socio-demographic Index. Our findings provide insight into the changing burden of stomach cancer at the global, regional, and national levels, which can be used by policy makers to develop location-specific programmes aimed at further reducing the burden of stomach cancer.

each region, are associated with rises in the Socio-demographic Index. Our findings provide insight into the changing burden of stomach cancer at the global, regional, and national levels, which can be used by policy makers to develop location-specific programmes aimed at further reducing the burden of stomach cancer. Although survival rates have generally improved over the past several decades, prognosis remains poor.14 The 5-year survival rate is around 20%, with the notable exceptions of 65% in Japan15 and 71·5% in South Korea,16 where population screening has led to the effective diagnosis of tumours at early stages.17 Given this poor survival and the considerable burden associated with stomach cancer, we evaluated the burden of stomach cancer using the results of the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2017. We highlight global and regional trends that can help to inform global and local interventions to lower disease burden and, perhaps, curtail the increasing number of incident cases. Methods Overview The methods used for GBD 2017 have been described previously and are briefly summarised here.18, 19, 20, 21 Cancers in GBD 2017 are classified into 29 groups according to the International Classification of Diseases 10th edition (ICD-10). Stomach cancer included all diagnoses coded C16·0–C16·9 (malignant neoplasm of stomach), Z12·0, and Z85·02–Z85·028, and did not include tumours of gastro-oesophageal junction.22 This study is compliant with the Guidelines for Accurate and Transparent Health Estimates Reporting.

cation of Diseases 10th edition (ICD-10). Stomach cancer included all diagnoses coded C16·0–C16·9 (malignant neoplasm of stomach), Z12·0, and Z85·02–Z85·028, and did not include tumours of gastro-oesophageal junction.22 This study is compliant with the Guidelines for Accurate and Transparent Health Estimates Reporting. Data sources For this study, we used GBD 2017 vital registration (19 618 site-years of data), verbal autopsy (374 site-years), and cancer registry (4474 site-years) data sources that provided a representative partial or complete sample of incidence or mortality. Information about the data sources used for each location in this study can be found on the GBD 2017 Data Input Sources Tool website.

e-years of data), verbal autopsy (374 site-years), and cancer registry (4474 site-years) data sources that provided a representative partial or complete sample of incidence or mortality. Information about the data sources used for each location in this study can be found on the GBD 2017 Data Input Sources Tool website. Mortality We derived mortality estimates from the data source described above and, when necessary, registry incidence data were multiplied by the corresponding, independently modelled, mortality-to-incidence ratio to produce mortality estimates. Mortality-to-incidence ratios were modelled using locations where same-year cancer mortality and incidence data were available. The mortality-to-incidence ratio model started with a linear-step mixed-effects model with logit link functions, with the Healthcare Access and Quality (HAQ) Index,23 age, and sex as covariates. The estimates produced by this model were then smoothed over space and time and adjusted using spatiotemporal Gaussian process regression.24 All the estimates computed from the mortality-to-incidence ratios and incidence data were used as inputs for a Cause of Death Ensemble model.25 Potential covariates in these models were selected at three levels, pertaining to a possible predictive relationship between the covariate and stomach cancer mortality. Level 1 covariates (those with a proven strong relationship with stomach cancer; eg, biological or causative link) included smoking prevalence, mean cigarettes per capita, cumulative cigarettes (5, 10, 15, and 20 mean pack-years), diet high in sodium, log-transformed summary exposure value (SEV) scalar for stomach cancer, SEV of unsafe water, and SEV of unsafe sanitation. Level 2 covariates (strong relationship without a direct biological link) were mean body-mass index, indoor air pollution (all cooking fuels), outdoor air pollution (particulate matter concentration of 2·5 μm dimeter), HAQ Index, adjusted fruit and vegetable intake (grams), sanitation (proportion with access), and improved water source (proportion with access). Finally, Level 3 (more distal covariates mediated through Level 1 or 2 covariates) included education (years per capita), lag-distributed income (US$ per capita), and Socio-demographic Index (SDI).

uit and vegetable intake (grams), sanitation (proportion with access), and improved water source (proportion with access). Finally, Level 3 (more distal covariates mediated through Level 1 or 2 covariates) included education (years per capita), lag-distributed income (US$ per capita), and Socio-demographic Index (SDI). Non-fatal modelling The final mortality estimates were divided by the estimated mortality-to-incidence ratios to compute stomach cancer incidence. Disability-adjusted life-years (DALYs) were calculated by summing years lived with disability (YLDs) and years of life lost (YLLs). The contributions of YLDs and YLLs to stomach cancer DALYs were 2% and 98%, respectively. YLDs were estimated by classifying 10-year cancer prevalence into four sequelae and multiplying these prevalences by corresponding disability weights: diagnosis and treatment, remission, disseminated and metastatic, and terminal phase. The durations of the four prevalence phases for stomach cancer were 5·2 months26 of diagnosis and treatment, 3·88 months27 of disseminated and metastatic disease, and 1 month of terminal phase. Remission durations were calculated on the basis of the remainder of time after attributing other sequelae. The YLLs were estimated by multiplying the estimated number of deaths by age with a standard life expectancy at that age. Details of estimation methods and data sources have been published before.22

rminal phase. Remission durations were calculated on the basis of the remainder of time after attributing other sequelae. The YLLs were estimated by multiplying the estimated number of deaths by age with a standard life expectancy at that age. Details of estimation methods and data sources have been published before.22 Rates were standardised to the GBD world population and reported per 100 000 population as age-standardised incidence rates, age-standardised death rates, and age-standardised DALY rates. All estimates were generated with 95% uncertainty intervals (UIs), including all sources of uncertainty arising from measurement error, biases, and modelling. The 95% UIs were derived from the 2·5th and 97·5th percentiles of 1000 draws. SDI and risk factors Risk factor quantification was based on the GBD 2017 comparative risk assessment described earlier.18 The SDI is a composite indicator of socio-development status that includes fertility, education, and income, and which has shown a strong association with health outcomes. SDI ranges from zero (worst) to one (best).22 We used linear correlation and fitted regression lines to determine the relationship between countries' development level (ie, SDI) and incidence, death, and DALY rates of stomach cancer. We reported the percentage of DALYs due to stomach cancer that were attributable to high-sodium diet and smoking by multiplying stomach cancer DALYs by the risk factor's population attributable fraction for a given age, sex, location, and year.18

t level (ie, SDI) and incidence, death, and DALY rates of stomach cancer. We reported the percentage of DALYs due to stomach cancer that were attributable to high-sodium diet and smoking by multiplying stomach cancer DALYs by the risk factor's population attributable fraction for a given age, sex, location, and year.18 Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or the writing of the report. All authors had full access to the data in the study and had final responsibility for the decision to submit for publication.

t level (ie, SDI) and incidence, death, and DALY rates of stomach cancer. We reported the percentage of DALYs due to stomach cancer that were attributable to high-sodium diet and smoking by multiplying stomach cancer DALYs by the risk factor's population attributable fraction for a given age, sex, location, and year.18 Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or the writing of the report. All authors had full access to the data in the study and had final responsibility for the decision to submit for publication. Results In 2017, more than 1·22 million (95% UI 1·19–1·25) incident cases of stomach cancer occurred worldwide, and nearly 865 000 people (848 000–885 000) died of stomach cancer (table). The age-standardised incidence rate of stomach cancer was 15·4 per 100 000 population (15·0–15·8), with an age-standardised death rate of 11·0 per 100 000 population (10·8–11·2). Stomach cancer contributed to 19·1 million (18·7–19·6) DALYs worldwide in 2017. For males, both the age-standardised incidence and death rates of stomach cancer were more than twice the rates for females (21·7 [21·0–22·6] vs 9·9 [9·6–10·2], and 15·2 [14·8–15·7] vs 7·5 [7·3–7·7] per 100 000 population, respectively). The male–female incidence and death gap existed in all regions, but was narrower in Andean Latin America and south Asia (figure 1). Specific country and territory data for incidence, deaths, and DALYs can be found in the appendix (p 2–25).Table Incident cases of deaths and DALYs of stomach cancer in 2017, and percentage change of age-standardised rates by sex and GBD region

egions, but was narrower in Andean Latin America and south Asia (figure 1). Specific country and territory data for incidence, deaths, and DALYs can be found in the appendix (p 2–25).Table Incident cases of deaths and DALYs of stomach cancer in 2017, and percentage change of age-standardised rates by sex and GBD region Incidence Deaths DALYs Number of incident cases Age-standardised incidence rate (per 100 000 population) Percentage change in rates, 1990–2017 Number of deaths Age-standardised death rate (per 100 000 population) Percentage change in rates, 1990–2017 Number of DALYs Age-standardised DALYs rate (per 100 000 population) Percentage change in rates, 1990–2017 Global 1 220 662 (1 189 032 to 1 254 563) 15·4 (15·0 to 15·8) –28·0 (−30·5 to −25·4) 864 989 (848 254 to 884 655) 11·0 (10·8 to 11·2) –43·2 (−45·1 to −41·4) 19 130 771 (18 738 585 to 19 569 409) 235·9 (231·1 to 241·3) –47·1 (−49·0 to −45·3) Males 799 309 (771 025 to 830 413) 21·7 (21·0 to 22·6) –24·0 (−27·0 to −20·7) 546 441 (530 918 to 564 028) 15·2 (14·8 to 15·7) –41·4 (−43·4 to −38·9) 12 248 716 (11 898 092 to 12 645 097) 317·8 (308·8 to 327·9) –45·4 (−47·6 to −43·0) Females 421 353 (408 084 to 434 424) 9·9 (9·6 to 10·2) –35·6 (−39·3 to −32·3) 318 548 (309 796 to 327 854) 7·5 (7·3 to 7·7) –46·9 (−50·0 to −44·4) 6 882 055 (6 683 617 to 7 083 864) 163·0 (158·3 to 167·8) –50·0 (−53·0 to −47·5) Andean Latin America 8925 (8194 to 9658) 16·6 (15·2 to 18·0) –33·2 (−39·2 to −26·6) 9130 (8408 to 9901) 17·1 (15·7 to 18·5) –35·9 (−41·4 to −29·7) 193 905 (176 702 to 210 939) 353·5 (322·4 to 384·4) –38·3 (−44·2 to −31·9) Australasia 4263 (3845 to 4692) 8·8 (7·9 to 9·7) –24·9 (−32·4 to −17·1) 2233 (2052 to 2429) 4·4 (4·1 to 4·8) –47·4 (−51·7 to −42·8) 39 703 (36 268 to 43 401) 88·4 (80·7 to 96·6) –47·6 (−52·2 to −42·5) Caribbean 4033 (3769 to 4342) 7·9 (7·4 to 8·5) –25·2 (−30·4 to −19·6) 3684 (3438 to 3966) 7·3 (6·8 to 7·8) –34·1 (−38·6 to −29·3) 83 446 (77 493 to 90 753) 164·4 (152·5 to 178·7) –31·1 (−36·2 to −25·2) Central Asia 10 513 (10 059 to 10 965) 14·1 (13·5 to 14·7) –38·9 (−41·6 to −36·2) 10 331 (9891 to 10 769) 14·3 (13·8 to 14·9) –39·0 (−41·7 to −36·4) 273 093 (260 966 to 285 850) 340·0 (325·0 to 354·6) –42·7 (−45·3 to −40·1) Central Europe 19 794 (19 194 to 20 462) 9·4 (9·1 to 9·7) –43·3 (−45·4 to −41·2) 18 570 (18 014 to 19 135) 8·6 (8·4 to 8·9) –48·7 (−50·5 to −46·8) 379 211 (367 809 to 391 374) 189·9 (184·0 to 196·0) –50·0 (−51·8 to −48·0) Central Latin America 29 601 (28 255 to 31 008) 12·9 (12·3 to 13·5)

354·6) –42·7 (−45·3 to −40·1) Central Europe 19 794 (19 194 to 20 462) 9·4 (9·1 to 9·7) –43·3 (−45·4 to −41·2) 18 570 (18 014 to 19 135) 8·6 (8·4 to 8·9) –48·7 (−50·5 to −46·8) 379 211 (367 809 to 391 374) 189·9 (184·0 to 196·0) –50·0 (−51·8 to −48·0) Central Latin America 29 601 (28 255 to 31 008) 12·9 (12·3 to 13·5) –16·9 (−20·8 to −12·7) 21 226 (20 432 to 22 007) 9·3 (9·0 to 9·7) –41·6 (−43·8 to −39·4) 487 298 (468 124 to 506 160) 203·8 (195·8 to 211·7) –38·9 (−41·4 to −36·6) Central sub-Saharan Africa 3555 (3106 to 4009) 7·1 (6·3 to 8·0) –34·3 (−43·2 to −24·4) 3551 (3110 to 3990) 7·6 (6·7 to 8·5) –33·9 (−42·7 to −24·2) 100 529 (87 599 to 113 903) 173·0 (151·3 to 195·0) –34·9 (−44·5 to −24·0) East Asia 583 758 (554 933 to 612 688) 28·6 (27·3 to 30·0) –14·7 (−21·5 to −8·9) 371 288 (356 398 to 387 201) 18·7 (17·9 to 19·5) –44·5 (−48·7 to −41·0) 8 175 270 (7 834 190 to 8 543 708) 389·5 (373·5 to 407·2) –49·7 (−53·3 to −46·6) Eastern Europe 59 809 (57 983 to 61 663) 17·7 (17·2 to 18·3) –39·5 (−41·4 to −37·5) 43 943 (42 870 to 45 173) 12·8 (12·5 to 13·1) –51·9 (−52·9 to −50·8) 1 014 257 (987 340 to 1 043 869) 308·7 (300·7 to 317·3) –53·5 (−54·5 to −52·5) Eastern sub-Saharan Africa 10 056 (9361 to 10 804) 6·4 (5·9 to 6·8) –40·2 (−46·5 to −32·8) 10 087 (9391 to 10 828) 6·8 (6·3 to 7·3) –39·3 (−45·3 to −32·5) 280 901 (261 364 to 301 544) 155·8 (145·0 to 167·4) –41·9 (−48·7 to −33·7) High-income Asia Pacific 131 636 (125 691 to 138 437) 29·5 (28·2 to 31·0) –48·7 (−51·1 to −45·9) 68 042 (65 688 to 71 099) 14·2 (13·7 to 14·8) –56·7 (−58·3 to −54·7) 1 099 094 (1 056 842 to 1147 281) 280·0 (268·6 to 292·3) –62·4 (−63·8 to −60·6) High-income North America 39 247 (37 998 to 40 539) 6·5 (6·3 to 6·7) –20·4 (−23·3 to −17·4) 22 159 (21 591 to 22 750) 3·6 (3·5 to 3·7) –40·0 (−41·7 to −38·3) 419 819 (407 273 to 431 498) 74·5 (72·3 to 76·7) –39·2 (−41·3 to −37·2) North Africa and Middle East 35 755 (33 988 to 37 539) 8·7 (8·3 to 9·1) –35·6 (−40·7 to −29·7) 34 530 (32 838 to 36 201) 8·8 (8·4 to 9·2) –37·4 (−42·1 to −31·9) 857 927 (809 034 to 906 399) 189·0 (179·0 to 198·9) –42·2 (−46·9 to −36·5) Oceania 988 (815 to 1177) 13·8 (11·7 to 16·1) –14·5 (−25·6 to −3·0) 913 (758 to 1077) 14·0 (11·9 to 16·1) –14·5 (−24·8 to −3·4) 30 288 (24 766 to 36 707) 358·4 (297·2 to 423·6) –15·4 (−28·2 to −1·9) South Asia 96 577 (91 436 to 101 062) 7·2 (6·8 to 7·5) –30·9 (−37·2 to −25·0) 96 652 (91 276 to 101 052) 7·5 (7·0 to 7·8) –30·2 (−36·6 to −24·6) 2 626 634 (2 487 690 to 2 747 242) 178·2 (168·5 to 186·2) –33·0

4·0 (11·9 to 16·1) –14·5 (−24·8 to −3·4) 30 288 (24 766 to 36 707) 358·4 (297·2 to 423·6) –15·4 (−28·2 to −1·9) South Asia 96 577 (91 436 to 101 062) 7·2 (6·8 to 7·5) –30·9 (−37·2 to −25·0) 96 652 (91 276 to 101 052) 7·5 (7·0 to 7·8) –30·2 (−36·6 to −24·6) 2 626 634 (2 487 690 to 2 747 242) 178·2 (168·5 to 186·2) –33·0 (−39·4 to −27·3) Southeast Asia 39 191 (36 559 to 42 265) 6·8 (6·4 to 7·3) –43·8 (−49·0 to −38·2) 38 871 (36 283 to 41 869) 7·0 (6·6 to 7·6) –44·7 (−49·7 to −39·4) 960 904 (893 460 to 1038 037) 154·1 (143·5 to 166·3) –48·1 (−53·3 to −42·3) Southern Latin America 10 145 (9436 to 10 956) 12·4 (11·5 to 13·4) –34·1 (−39·0 to −28·5) 10 203 (9515 to 10 988) 12·3 (11·5 to 13·2) –38·3 (−42·8 to −33·1) 203 392 (188 154 to 220 430) 253·8 (234·7 to 275·1) –40·4 (−45·0 to −35·0) Southern sub-Saharan Africa 2839 (2706 to 2979) 5·2 (5·0 to 5·4) –27·3 (−32·3 to −22·2) 2910 (2779 to 3052) 5·5 (5·3 to 5·8) –26·2 (−31·3 to −21·0) 73 206 (69 472 to 77 098) 123·3 (117·3 to 129·6) –30·7 (−35·8 to −25·5) Tropical Latin America 21 823 (21 369 to 22 331) 9·5 (9·3 to 9·7) –40·2 (−41·8 to −38·5) 21 140 (20 734 to 21 550) 9·3 (9·1 to 9·4) –44·9 (−46·3 to −43·6) 482 998 (473 844 to 492 741) 203·0 (199·3 to 207·1) –44·1 (−45·6 to −42·8) Western Europe 95 266 (90 462 to 100 111) 10·5 (10·0 to 11·0) –37·4 (−40·6 to −34·2) 62 213 (59 851 to 64 525) 6·4 (6·2 to 6·7) –54·3 (−56·0 to −52·6) 1 025 787 (984 851 to 1 066 633) 125·4 (120·2 to 130·4) –55·1 (−56·9 to −53·3) Western sub-Saharan Africa 12 890 (11 719 to 14 193) 7·7 (7·1 to 8·5) –20·5 (−27·9 to −11·3) 13 311 (12 153 to 14 701) 8·4 (7·7 to 9·2) –19·5 (−27·1 to −9·9) 323 109 (293 082 to 359 119) 167·9 (152·9 to 185·7) –23·8 (−30·9 to −14·8) Data in parentheses are 95% uncertainty intervals.

0·2 to 130·4) –55·1 (−56·9 to −53·3) Western sub-Saharan Africa 12 890 (11 719 to 14 193) 7·7 (7·1 to 8·5) –20·5 (−27·9 to −11·3) 13 311 (12 153 to 14 701) 8·4 (7·7 to 9·2) –19·5 (−27·1 to −9·9) 323 109 (293 082 to 359 119) 167·9 (152·9 to 185·7) –23·8 (−30·9 to −14·8) Data in parentheses are 95% uncertainty intervals. DALY=disability-adjusted life-year. GBD=Global Burden of Diseases, Injuries, and Risk Factors Study. Figure 1 The age-standardised incidence (A) and death (B) rates of stomach cancer in 2017 for 21 GBD regions, by sex GBD=Global Burden of Diseases, Injuries, and Risk Factors Study.

0·2 to 130·4) –55·1 (−56·9 to −53·3) Western sub-Saharan Africa 12 890 (11 719 to 14 193) 7·7 (7·1 to 8·5) –20·5 (−27·9 to −11·3) 13 311 (12 153 to 14 701) 8·4 (7·7 to 9·2) –19·5 (−27·1 to −9·9) 323 109 (293 082 to 359 119) 167·9 (152·9 to 185·7) –23·8 (−30·9 to −14·8) Data in parentheses are 95% uncertainty intervals. DALY=disability-adjusted life-year. GBD=Global Burden of Diseases, Injuries, and Risk Factors Study. Figure 1 The age-standardised incidence (A) and death (B) rates of stomach cancer in 2017 for 21 GBD regions, by sex GBD=Global Burden of Diseases, Injuries, and Risk Factors Study. The world map of age-standardised incidence rates of stomach cancer in 2017 is shown in figure 2. The highest age-standardised incidence rate was seen in the high-income Asia Pacific region (29·5 per 100 000 population [95% UI 28·2–31·0]), particularly in Japan and South Korea, and east Asia (28·6 per 100 000 population [27·3–30·0]; table; appendix pp 2–9, 18–25). In east Asia, China alone had nearly half of the global incident cases in 2017 (562 000 [533 000–591 000]), and contributed to 7·8 million (7·5–8·2) DALYs (appendix pp 2–9, 18–25). The eastern Europe (17·7 [17·2–18·3]) and Andean Latin America (16·6 [15·2–18·0]) regions had the next highest age-standardised incidence rates. Two countries outside these high-incidence regions—Mongolia (35·6 [31·9–39·6]) and Afghanistan (32·8, [26·5–39·6])—had the overall highest age-standardised incidence rates. The lowest incidence rates were seen in southern and eastern sub-Saharan Africa and high-income North America (table). East Asia had the highest age-standardised death rate (18·7 [17·9–19·5]), followed by Andean Latin America (17·1 [15·7–18·5]) and central Asia (14·3 [13·8–14·9]). The high-income Asia Pacific region, which ranked first in age-standardised incidence rate, had the fourth highest age-standardised death rate and the sixth highest DALY rate among all GBD regions in 2017. The two countries with the highest age-standardised incidence rate also had the highest age-standardised death rates: Mongolia (37·6 [33·8–41·8]) and Afghanistan (33·6 [27·2–40·2]). The lowest age-standardised death rates were seen in high-income North America and Australasia.Figure 2 Age-standardised incidence rate of stomach cancer per 100 000 population in 2017, for 195 countries and territories

ighest age-standardised death rates: Mongolia (37·6 [33·8–41·8]) and Afghanistan (33·6 [27·2–40·2]). The lowest age-standardised death rates were seen in high-income North America and Australasia.Figure 2 Age-standardised incidence rate of stomach cancer per 100 000 population in 2017, for 195 countries and territories Compared with 1990, in 2017 the number of incident cases of stomach cancer increased from about 864 000 (95% UI 847 000–890 000) to 1·22 million (1·19–1·25; appendix p 2)—an increase of around 356 000 cases. The number of deaths increased from around 769 000 (752 000–795 000) to 865 000 (848 000–885 000; appendix p 10)—ie, an increase of around 96 000 deaths. But stomach cancer contributed to almost the same number of DALYs in 2017 as in 1990 (appendix pp 18–25). The majority of increases in the absolute number of cases and deaths came from east Asia: from 1990 to 2017, incident cases rose from about 308 000 (296 000–326 000) to almost 584 000 (555 000–613 000) and the number of deaths increased from around 296 000 (285 000–313 000) to more than 371 000 (356 000–387 000). Again, the bulk of these increases occurred in China (appendix pp 2–17). Other regions that made large contributions to increased numbers of cases and deaths included south Asia, central Latin America, and north Africa and the Middle East (figure 3). In high-income Asia Pacific countries, the number of incident cases increased from over 117 000 (115 000–119 000) to almost 132 000 (126 000–138 000), but the number of deaths due to stomach cancer showed very little change in the same period (66 000 [65 300–66 800] in 1990 to 68 000 [65 700–71 100] in 2017; appendix pp 2–17). Both the number of incident cases and number of deaths due to stomach cancer decreased in European regions (central, eastern, and western) between 1990 and 2017.Figure 3 Absolute number of incident cases of (A) and deaths due to (B) stomach cancer, 1990 to 2017, for 21 GBD regions

00–71 100] in 2017; appendix pp 2–17). Both the number of incident cases and number of deaths due to stomach cancer decreased in European regions (central, eastern, and western) between 1990 and 2017.Figure 3 Absolute number of incident cases of (A) and deaths due to (B) stomach cancer, 1990 to 2017, for 21 GBD regions GBD=Global Burden of Diseases, Injuries, and Risk Factors Study.

00–71 100] in 2017; appendix pp 2–17). Both the number of incident cases and number of deaths due to stomach cancer decreased in European regions (central, eastern, and western) between 1990 and 2017.Figure 3 Absolute number of incident cases of (A) and deaths due to (B) stomach cancer, 1990 to 2017, for 21 GBD regions GBD=Global Burden of Diseases, Injuries, and Risk Factors Study. Despite the increases in absolute numbers, the global age-standardised incidence and death rates of stomach cancer decreased compared with 1990 (figure 4). During this period, the age-standardised incidence rate decreased by 28·0% (95% UI 25·4–30·5) globally, while the age-standardised death rate dropped at a faster rate of 43·2% (41·0–45·1), and age-standaradised DALY rate by 47·1% (45·3–49·0; table; figure 4). The downward trend in age-standardised incidence rates did, however, plateau at the global level and for some regions in the last 5 years of the study period. The decreases in age-standardised incidence and mortality from 1990 to 2017 were greater for females than for males at the global level and in many regions (appendix pp 31–32). The high-income Asia Pacific region had the sharpest drop in age-standardised rates between 1990 and 2017 compared with all other regions (decrease in age-standardised incidence by 48·7% [45·9–51·1]; age-standardised deaths by 56·7% [54·7–58·3]; and age-standardised DALYs by 62·4% [60·6–63·8]). In east Asia, the drop in age-standardised incidence rate was not as sharp (14·7% [8·9–21·5]); however, a 44·5% (41·2–48·7) decrease in age-standardised deaths caused the age-standardised DALY rate to reduce by nearly half (49·7%, 46·6–53·3; table).Figure 4 Secular trends of age-standardised incidence (A), death (B), and DALY (C) rates of stomach cancer, 1990–2017, global and for 21 GBD regions

4·7% [8·9–21·5]); however, a 44·5% (41·2–48·7) decrease in age-standardised deaths caused the age-standardised DALY rate to reduce by nearly half (49·7%, 46·6–53·3; table).Figure 4 Secular trends of age-standardised incidence (A), death (B), and DALY (C) rates of stomach cancer, 1990–2017, global and for 21 GBD regions DALYs=disability-adjusted life-years. GBD=Global Burden of Diseases, Injuries, and Risk Factors Study.

4·7% [8·9–21·5]); however, a 44·5% (41·2–48·7) decrease in age-standardised deaths caused the age-standardised DALY rate to reduce by nearly half (49·7%, 46·6–53·3; table).Figure 4 Secular trends of age-standardised incidence (A), death (B), and DALY (C) rates of stomach cancer, 1990–2017, global and for 21 GBD regions DALYs=disability-adjusted life-years. GBD=Global Burden of Diseases, Injuries, and Risk Factors Study. The percentage of age-standardised DALYs attributable to high-sodium diet and smoking in each region are shown in figure 5. Globally, 38·2% (95% UI 21·1–57·8) of the age-standardised DALYs were attributable to a high-sodium diet, which was moderately higher in males than in females (41·2% [23·6–60·9] for males; and 32·7% [16·4–52·7] for females). In east Asia, this figure was almost twice that of all other regions, with 61·3% (42·8–78·2) of age-standardised DALYs attributable to a high-sodium diet. For males, 24·5% (20·0–28·9) of the age-standardised DALYs globally were attributable to smoking. Eastern Europe (33·0%, 27·5–37·9) and east Asia (29·0%, 23·5–34·2) had the highest percentage of age-standardised DALYs attributable to smoking among males. In females, smoking did not account for a sizeable fraction of age-standardised DALYs globally, but in parts of Europe (particularly western and central), high-income North America, Australasia, and parts of Latin America (particularly tropical and southern), the contribution of smoking to stomach cancer age-standardised DALYs among females was higher than 10%.Figure 5 Percentage of stomach cancer age-standardised DALYs attributable to high-sodium diet and smoking in 2017, by sex, globally and for 21 GBD regions

d parts of Latin America (particularly tropical and southern), the contribution of smoking to stomach cancer age-standardised DALYs among females was higher than 10%.Figure 5 Percentage of stomach cancer age-standardised DALYs attributable to high-sodium diet and smoking in 2017, by sex, globally and for 21 GBD regions DALYs=disability-adjusted life-years. GBD=Global Burden of Diseases, Injuries, and Risk Factors Study. Discussion Stomach cancer is an important cause of morbidity and mortality in many parts or the world, and the total numbers of incident cases and deaths are increasing worldwide. East Asia, particularly China, contributes the largest number of incident cases, deaths, and DALYs from stomach cancer globally. However, the age-standardised incidence and death rates have declined steadily.

tality in many parts or the world, and the total numbers of incident cases and deaths are increasing worldwide. East Asia, particularly China, contributes the largest number of incident cases, deaths, and DALYs from stomach cancer globally. However, the age-standardised incidence and death rates have declined steadily. Our findings generally agree with those of the Global Cancer Incidence, Mortality and Prevalence (GLOBOCAN) project,28 although our estimates were somewhat higher than theirs, possibly as a result of differences in data sources and estimation methods. From 1990 to 2017, stomach cancer dropped from the fifth leading incident cancer worldwide to the seventh, and from the second leading cause of cancer deaths to third (after lung and colorectal cancers).22 As a result, stomach cancer accounts for the third highest cancer-related DALYs after lung and liver cancers. However, this decline in burden relative to other cancers and the dramatic decline in age-standardised rates have not necessarily led to a lower burden of stomach cancer on the health systems in high-risk countries. This is because changes in the population age structure and population growth have meant that numbers of incident cases and deaths of stomach cancer have continued to increase in many locations.22 Based on our findings, a further decrease in the absolute number of cases and deaths could be possible if the rates in east Asia, where almost half of the incident cases and deaths occur, are further reduced.

ant that numbers of incident cases and deaths of stomach cancer have continued to increase in many locations.22 Based on our findings, a further decrease in the absolute number of cases and deaths could be possible if the rates in east Asia, where almost half of the incident cases and deaths occur, are further reduced. Based on migrant studies29 and the secular trends in stomach cancer rates, environmental factors are thought to play a significant role in the pathogenesis of stomach cancer.3 In contrast, only about 10% of cases aggregate in families, and only 1–3% occur as part of known hereditary syndromes.1 Our results suggest lifestyle factors play a significant role in stomach cancer burden, in particular high-sodium diet in east Asian populations and smoking among males. Both of these are also risk factors for other non-communicable diseases30 and minimal exposure to them is generally suggested in guidelines for a healthy lifestyle. Reducing high-salt foods in the diet is one of the ways proposed to tackle the stomach cancer problem in high-risk Asian countries.31

smoking among males. Both of these are also risk factors for other non-communicable diseases30 and minimal exposure to them is generally suggested in guidelines for a healthy lifestyle. Reducing high-salt foods in the diet is one of the ways proposed to tackle the stomach cancer problem in high-risk Asian countries.31 H pylori infection is the most important established risk factor for stomach cancer, and, as a result, most of the prevention strategies against this cancer focus on this infection. H pylori was once a ubiquitous infection,32 and in populations where infection rates are high, stomach cancer is a significant public health problem despite other interventions.3, 33 In some countries in western Europe (eg, Germany, the UK, and Spain) and in the USA, the previously declining burden of stomach cancer has plateaued, especially in the middle-aged (ie, 50–64 years) population, probably due to a low and stable prevalence of H pylori infection.34 For our study, we were unable to evaluate the role of H pylori infection in stomach cancer burden: such data are not included in registries and population-level data sources because they are costly and difficult to obtain at large scales. However, most of the risk reduction due to improved socio-economic status (even in the absence of specific preventive strategies) is thought to stem from reduced H pylori infection rates.35 Beyond general sanitation and improved socio-economic status, H pylori can also be effectively eradicated by different drug regimens. There is some controversy over whether population eradication of H pylori would be a cost-effective strategy to lower the burden of stomach cancer, and a trial to address this question would be logistically challenging and resource-intensive because of the period required for follow-up. In a systematic review and meta-analysis, Lee and colleagues36 reported that mass eradication of H pylori infection was associated with a reduced incidence of stomach cancer. They concluded that the benefit of eradication was dependent on the stomach cancer risk at baseline. In another meta-analysis,37 Ford and colleagues limited their study to randomised controlled trials and showed limited, moderate-quality evidence for reduced incidence of stomach cancer associated with screening for and eradicating H pylori in healthy asymptomatic infected individuals from Asia. This study concluded that these results might not necessarily be generalisable to other populations.

ised controlled trials and showed limited, moderate-quality evidence for reduced incidence of stomach cancer associated with screening for and eradicating H pylori in healthy asymptomatic infected individuals from Asia. This study concluded that these results might not necessarily be generalisable to other populations. We found that reductions in age-standardised incidence rates did not necessarily parallel those of age-standardised death and DALY rates, meaning that while age-standardised death rates fell considerably in many locations, age-standardised incidence rates decreased more slowly. For example, east Asia, particularly China, witnessed a relatively small decrease in age-standardised incidence rates over the study period, whereas the decreases in age-standardised death and attributable DALY rates were much larger. A study of cancer registries in China showed that the disparity in cancer mortality rates was far greater than cancer incidence when rural and urban areas were compared.38 The authors suggested that this disparity was due to limited medical resources, lower levels of cancer care, and a larger proportion of patients diagnosed with cancer at a late stage in rural and underdeveloped areas. China has taken steps to reduce the cancer care disparities between rural and urban areas,39 and we think these efforts might explain the sharp decrease in death rates due to stomach cancer. Another successful strategy to reduce deaths due to stomach cancer has been used in Japan and South Korea, two of the high-income Asia Pacific countries. These countries have implemented population screening programmes leading to early detection of cancer cases40 and better survival rates. This has led some investigators to believe that China should adopt a similar strategy to further reduce the burden of stomach cancer.41 But population screening for stomach cancer includes invasive methods, and its feasibility and cost-effectiveness outside high-risk regions have never been investigated.2 A serum pepsinogen test combined with H pylori testing has been suggested as a potential method to triage suitable candidates for more invasive screening methods, but evidence for their clinical application is still mainly limited to Japanese populations.42Although regional patterns convey a lot of useful information about the distribution of stomach cancer burden and its trends and correlates, significant variations exist within each region.

for more invasive screening methods, but evidence for their clinical application is still mainly limited to Japanese populations.42Although regional patterns convey a lot of useful information about the distribution of stomach cancer burden and its trends and correlates, significant variations exist within each region. For example, stomach cancer incidence rates in Canada are almost double those in the USA, and stomach cancer contributes to more age-standardised DALYs per 100 000 population in Portugal, Chile, Guatemala, Bolivia, Haiti, Zimbabwe, and Mali than in other countries in the same regions. In addition, the two countries with the highest age-standardised incidence, death, and DALY rates globally (Mongolia and Afghanistan) have much higher rates than those of their respective regions (central Asia, and north Africa and Middle East, respectively). While making such comparisons, it is important to note that data quality differs across individual countries, and GBD uses a rating system (from zero to five stars) to assess the quality of the available data from each country.43 In addition to data quality, differences in data collection and coding systems are other challenges facing such comparisons.

is important to note that data quality differs across individual countries, and GBD uses a rating system (from zero to five stars) to assess the quality of the available data from each country.43 In addition to data quality, differences in data collection and coding systems are other challenges facing such comparisons. A major limitation of our study was the inability to distinguish cardia and non-cardia forms of stomach cancer. Non-cardia stomach cancer is predominantly associated with H pylori infection and is probably the main reason for the changing rates of stomach cancer across the world.4 Cardia tumours are estimated to account for about 12% of stomach cancers globally, but are responsible for a higher proportion of stomach cancer burden in some low-risk populations.44 Cardia cancer in the USA is more common among the non-Hispanic white population and is not strongly linked to socio-economic status.45 Comparing different populations and studying the secular trends of cardia versus non-cardia tumours over time is complex, because the definition of cardia cancer has evolved over time, and some cardia tumours might have been classified as lower oesophageal adenocarcinomas, and vice versa.8 We could not determine the burden of stomach cancer directly attributable to H pylori, as described above, and the lack of data on some of the other risk factors limited our risk factor analysis. We also did not have information on the molecular subtypes of stomach tumours.46 As a general limitation of GBD, we relied on estimates from the modelling process for locations where data had low levels of completeness. But, by providing comprehensive measures of uncertainty, the degree of error in the estimates resulting from data scarcity is clarified.

the molecular subtypes of stomach tumours.46 As a general limitation of GBD, we relied on estimates from the modelling process for locations where data had low levels of completeness. But, by providing comprehensive measures of uncertainty, the degree of error in the estimates resulting from data scarcity is clarified. Beyond general improvements in socio-economic status leading to improved health and lower H pylori infection rates, specific local strategies are needed to further reduce the number of incident cases and deaths due to stomach cancer, and these should be tailored to each country's risk factor profile. Targeting the risk factors that affect stomach cancer incidence and mortality (such as smoking and diet), in addition to country-specific feasible and cost-effective interventions aimed at lowering H pylori infection rates, early detection of suspected cases, and improved access to standard treatment facilities, can be among such strategies. By providing annual updates to regional and country-level stomach cancer estimates, future iterations of GBD will be useful for monitoring the success of such strategies. This online publication has been corrected. The corrected version first appeared at thelancet.com/gastrohep on Feb 12, 2020 Supplementary Material Supplementary appendix

Beyond general improvements in socio-economic status leading to improved health and lower H pylori infection rates, specific local strategies are needed to further reduce the number of incident cases and deaths due to stomach cancer, and these should be tailored to each country's risk factor profile. Targeting the risk factors that affect stomach cancer incidence and mortality (such as smoking and diet), in addition to country-specific feasible and cost-effective interventions aimed at lowering H pylori infection rates, early detection of suspected cases, and improved access to standard treatment facilities, can be among such strategies. By providing annual updates to regional and country-level stomach cancer estimates, future iterations of GBD will be useful for monitoring the success of such strategies. This online publication has been corrected. The corrected version first appeared at thelancet.com/gastrohep on Feb 12, 2020 Supplementary Material Supplementary appendix Acknowledgments This study was funded by the Bill & Melinda Gates Foundation. AA is supported by Department of Science and Technology, Government of India, New Delhi through INSPIRE Faculty programme. FC and JF acknowledge support with funding from Fundacao para a Ciencia e a Tecnologia/Ministerio da Ciencia, Tecnologia e Ensino Superior (FCT/MCTES) through Portuguese national funds, through UID/MULTI/04378/2019 (FC), UID/QUI/50006/2019 (FC), and UID/Multi/50016/2019 (JF) grants. VMC acknowledges her grant (SFRH/BHD/110001/2015), received by Portuguese national funds through Fundação para a Ciência e Tecnologia (FCT), IP, under the Norma Transitória DL57/2016/CP1334/CT0006. APK is supported by the South African Medical Research Council. YJK is supported by Xiamen University Malaysia Research Fund (Grant No. XMUMRF/2018-C2/ITCM/0001). IMV is supported by the Sistema Nacional de Investigación (SNI), Panama.

Ciência e Tecnologia (FCT), IP, under the Norma Transitória DL57/2016/CP1334/CT0006. APK is supported by the South African Medical Research Council. YJK is supported by Xiamen University Malaysia Research Fund (Grant No. XMUMRF/2018-C2/ITCM/0001). IMV is supported by the Sistema Nacional de Investigación (SNI), Panama. GBD 2017 Stomach Cancer Collaborators Arash Etemadi, Saeid Safiri, Sadaf G Sepanlou, Kevin Ikuta, Catherine Bisignano, Ramin Shakeri, Mohammad Amani, Christina Fitzmaurice, Molly R Nixon, Nooshin Abbasi, Hassan Abolhassani, Shailesh M Advani, Mohsen Afarideh, Tomi Akinyemiju, Tahiya Alam, Mahtab Alikhani, Vahid Alipour, Christine A Allen, Amir Almasi-Hashiani, Jalal Arabloo, Reza Assadi, Suleman Atique, Ashish Awasthi, Ahad Bakhtiari, Masoud Behzadifar, Kidanemaryam Berhe, Neeraj Bhala, Ali Bijani, Muhammad Shahdaat Bin Sayeed, Tone Bjørge, Antonio M Borzì, Dejana Braithwaite, Hermann Brenner, Giulia Carreras, Félix Carvalho, Carlos A Castañeda-Orjuela, Franz Castro, Dinh-Toi Chu, Vera M Costa, Ahmad Daryani, Dragos Virgil Davitoiu, Gebre T Demoz, Asmamaw Bizuneh Demis, Edgar Denova-Gutiérrez, Subhojit Dey, Mostafa Dianati Nasab, Shirin Djalalinia, Mohammad Hassan Emamian, Mohammad Farahmand, João C Fernandes, Florian Fischer, Masoud Foroutan, Mohamed M Gad, Silvano Gallus, Gebreamlak Gebremedhn Gebremeskel, Getnet Azeze Gedefew, Fatemeh Ghaseni-Kebria, Giuseppe Gorini, Nima Hafezi-Nejad, Arvin Haj-Mirzaian, Josep M Haro, James D Harvey, Amir Hasanzadeh, Maryam Hashemian, Hamid Y Hassen, Simon I Hay, Hagos D Hidru, Mihaela Hostiuc, Mowafa Househ, Olayinka Stephen Ilesanmi, Milena D Ilic, Kaire Innos, Farhad Islami, Spencer L James, Ensiyeh Jenabi, Rohollah Kalhor, Farin Kamangar, Amir Kasaeian, Andre Pascal Kengne, Yousef Saleh Khader, Rovshan Khalilov, Ejaz Ahmad Khan, Gulfaraz Khan, Maryam Khayamzadeh, Maryam Khazaee-Pool, Salman Khazaei, Abdullah T Khoja, Fatemah Khosravi Shadmani, Yun Jin Kim, Jonathan M Kocarnik, Hamidreza Komaki, Ai Koyanagi, Vivek Kumar, Carlo La Vecchia, Alan D Lopez, Raimundas Lunevicius, Navid Manafi, Ana-Laura Manda, Birhanu Geta, Hailemariam Meheretu, Getnet Mengistu, Bartosz Miazgowski, Seyed Mostafa Mir, Karzan Abdulmuhsin Mohammad, Naser Mohammad Gholi Mezerji, Mahdi Mohammadian, Abdollah Mohammadian-Hafshejani, Reza Mohammadpourhodki, Shafiu Mohammed, Farnam Mohebi, Ali H Mokdad, Lorenzo Monasta, Mahmood Moosazadeh, Maryam Moossavi, Ghobad Moradi, Farhad Moradpour, Rahmatollah Moradzadeh, Ilais Moreno Velasquez, Abbas Mosapour, M

uhsin Mohammad, Naser Mohammad Gholi Mezerji, Mahdi Mohammadian, Abdollah Mohammadian-Hafshejani, Reza Mohammadpourhodki, Shafiu Mohammed, Farnam Mohebi, Ali H Mokdad, Lorenzo Monasta, Mahmood Moosazadeh, Maryam Moossavi, Ghobad Moradi, Farhad Moradpour, Rahmatollah Moradzadeh, Ilais Moreno Velasquez, Abbas Mosapour, M ehdi Naderi, Gurudatta Naik, Farid Najafi, Azin Nahvijou, Ionut Negoi, Rajan Nikbakhsh, Marzieh Nojomi, Andrew T Olagunju, Tinuke O Olagunju, Eyal Oren, Hadi Parsian, Cristiano Piccinelli, Akram Pourshams, Hossein Poustchi, Navid Rabiee, Amir Radfar, Alireza Rafiei, Mahdi Rahimi, Marveh Rahmati, Andre M N Renzaho, Nima Rezaei, Ana Isabel Ribeiro, Gholamreza Roshandel, Anas M Saad, Seyedmohammad Saadatagah, Hamideh Salimzadeh, Abdallah M Samy, Juan Sanabria, Milena M Santric Milicevic, Arash Sarveazad, Monika Sawhney, Faramarz Shaahmadi, Mario Sekerija, Masood Ali Shaikh, Amir Shamshirian, Sudeep K Siddappa Malleshappa, Jasvinder A. Singh, Catalin-Gabriel Smarandache, Moslem Soofi, Takahiro Tabuchi, Degena Bahrey Bahrey Tadesse, Leili Tapak, Berhe Etsay Tesfay, Eugenio Traini, Bach Tran, Khanh Bao Tran, Marco Vacante, Amir Vahedian-Azimi, Yousef Veisani, Kia Vosoughi, Isidora S Vujcic, Ronny Westerman, Adam Belay Wondmieneh, Rixing Xu, Sanni Yaya, Vahid Yazdi-Feyzabadi, Zabihollah Yousefi, Bhaman Yousefi, Telma Zahirian Moghadam, Leila Zaki, Mohammad Zamani, Maryam Zamanian, Hamed Zandian, Afshin Zarghi, Zhi-Jiang Zhang, Mohsen Naghavi*, and Reza Malekzadeh*. *These authors jointly supervised the study.

ny Westerman, Adam Belay Wondmieneh, Rixing Xu, Sanni Yaya, Vahid Yazdi-Feyzabadi, Zabihollah Yousefi, Bhaman Yousefi, Telma Zahirian Moghadam, Leila Zaki, Mohammad Zamani, Maryam Zamanian, Hamed Zandian, Afshin Zarghi, Zhi-Jiang Zhang, Mohsen Naghavi*, and Reza Malekzadeh*. *These authors jointly supervised the study. Affiliations Division of Cancer Epidemiology and Genetics (A Etemadi PhD, M Hashemian PhD), National Cancer Institute, Bethesda, MD, USA; Cancer Biology Research Center (M Rahmati PhD), Cancer Research Center (A Nahvijou PhD), Department of Cardiology (S Saadatagah MD), Department of Microbiology (A Hasanzadeh PhD), Department of Pharmacology (A Haj-Mirzaian MD), Digestive Diseases Research Institute (S G Sepanlou MD, R Shakeri PhD, M Amani PhD, Prof A Pourshams MD, H Poustchi PhD, G Roshandel PhD, H Salimzadeh PhD, Prof R Malekzadeh MD), Digestive Oncology Research Center, Digestive Disease Research Institute (Prof F Kamangar MD), Endocrinology and Metabolism Research Center (M Afarideh MD), Health Policy and Economic and Management Department (A Bakhtiari PhD), Hematologic Malignancies Research Center (A Kasaeian PhD), Hematology-Oncology and Stem Cell Transplantation Research Cente (A Kasaeian PhD), Iran National Institute of Health Research (F Mohebi MD), Non-communicable Diseases Research Center (F Mohebi MD), Research Center for Immunodeficiencies (Prof N Rezaei PhD), Research center for Immunodeficiencies (H Abolhassani PhD), School of Medicine (N Hafezi-Nejad MD), School of Public Health (M Farahmand PhD), Tehran University of Medical Sciences, Iran (A Etemadi PhD); Aging Research Institute (S Safiri PhD), Clinical Biochemistry Department (Prof B Yousefi PhD), Department of Community Medicine (S Safiri PhD), Drug applied Research center (M Rahimi PhD, Prof B Yousefi PhD), Tabriz University of Medical Sciences, Tabriz, Iran; Department of Epidemiology (M Dianati Nasab MSc), Non-Communicable Diseases Research Center (S G Sepanlou MD, Prof R Malekzadeh MD), Shiraz University of Medical Sciences, Shiraz, Iran; Department of Health Metrics Sciences (Prof M Naghavi MD), Department of Health Metrics Sciences, School of Medicine (Prof S I Hay FMedSci, H D Hidru MPH, Prof A H Mokdad PhD), Division of Allergy and Infectious Diseases (K Ikuta MD), Division of Hematology (C Fitzmaurice MD), Institute for Health Metrics and Evaluation (K Ikuta MD, M R Nixon PhD, C Bisignano MPH, C Fitzmaurice MD, T Alam MPH, C A Allen BA, J D Harvey BS, Prof S I Hay FMedSci, S L Jame

ci, H D Hidru MPH, Prof A H Mokdad PhD), Division of Allergy and Infectious Diseases (K Ikuta MD), Division of Hematology (C Fitzmaurice MD), Institute for Health Metrics and Evaluation (K Ikuta MD, M R Nixon PhD, C Bisignano MPH, C Fitzmaurice MD, T Alam MPH, C A Allen BA, J D Harvey BS, Prof S I Hay FMedSci, S L Jame s MD, J M Kocarnik PhD, Prof A H Mokdad PhD, R Xu BS, Prof M Naghavi MD), Institute of Health Metrics and Evaluation (Prof A D Lopez PhD), University of Washington, Seattle, WA, USA (Prof E Oren PhD); Montreal Neurological Institute (N Abbasi MD), McGill University, Montreal, QC, Canada; LABMED (H Abolhassani PhD), Karolinska University Hospital, Huddinge, Stockholm, Sweden; Colombian National Health Observatory (C A Castañeda-Orjuela MD), Social Behavioral Research Branch (S M Advani PhD), National Institute of Health, Bethesda, MD, USA; Cancer Prevention and Control (S M Advani PhD), Oncology Department (D Braithwaite PhD), Georgetown University, Washington, DC, USA; Department of Population Health Sciences (T Akinyemiju PhD), Duke Global Health Institute (T Akinyemiju PhD), Duke University, Durham, NC, USA; Department of Health Services Management, School of Health Management and Information Sciences (M Alikhani PhD), University of Medical Sciences, Tehran, Iran; Colorectal Research Center (A Sarveazad PhD), Department of Community and Family Medicine (Prof M Nojomi MD), Health Economics Department (V Alipour PhD), Health Management and Economics Research Center (J Arabloo PhD, T Zahirian Moghadam PhD), Ophthalmology Department (N Manafi MD), Preventive Medicine and Public Health Research Center (Prof M Nojomi MD, K Vosoughi MD), Iran University of Medical Sciences, Iran (M Alikhani PhD); Department of Epidemiology (A Almasi-Hashiani PhD), Health Management and Economics Research Center (V Alipour PhD), Tehran, Iran; Education Development Center (R Assadi PhD), Mashhad University of Medical Sciences, Mashhad, Khorasan Razavi, Iran; University Institute of Public Health (S Atique PhD), The University of Lahore, Lahore, Punjab, Pakistan; College of Public Health (S Atique PhD), University of Hail, Hail, Saudi Arabia; Indian Institute of Public Health, Gandhinagar, Gujarat, India (A Awasthi PhD); Public Health Foundation of India, Gurugram, Haryana, India (A Awasthi PhD); Social Determinants of Health Research Center (M Behzadifar PhD), Lorestan University of Medical Sciences, Khorramabad, Iran; Department of Nutrition and Dietetics (K Berhe MPH), Nursing (G Geb

lth, Gandhinagar, Gujarat, India (A Awasthi PhD); Public Health Foundation of India, Gurugram, Haryana, India (A Awasthi PhD); Social Determinants of Health Research Center (M Behzadifar PhD), Lorestan University of Medical Sciences, Khorramabad, Iran; Department of Nutrition and Dietetics (K Berhe MPH), Nursing (G Geb remeskel MSc), Mekelle University, Mekelle, Tigray, Ethiopia; Institutes of Applied Health Research and Translational Medicine (N Bhala DPhil), Queen Elizabeth Hospital Birmingham, Birmingham, UK; University of Birmingham, Birmingham, UK (N Bhala DPhil); Department of Clinical Biochemistry (S Mir MSc, A Mosapour PhD, H Parsian PhD), Social Determinants of Health Research Center (A Bijani PhD, P Sajadi PhD), Student Research Committee (M Zamani MD), Babol University of Medical Sciences, Babol, Mazandaran, Iran; National Centre for Epidemiology and Population Health (M Bin Sayeed MSPS), Australian National University, Canberra, ACT, Australia; Department of Clinical Pharmacy and Pharmacology (M Bin Sayeed MSPS), University of Dhaka, Ramna, Dhaka, Bangladesh; Department of Global Public Health and Primary Care (Prof T Bjørge PhD), University of Bergen, Bergen, Norway; Cancer Registry of Norway, Oslo, Norway (Prof T Bjørge PhD); Department of General Surgery and Medical-Surgical Specialties (M Vacante PhD, A M Borzì MD), University of Catania, Catania, Italy; Department of Environmental Health Science (S Gallus DSc), Mario Negri Institute for Pharmacological Research, Milan, Italy; Division of Clinical Epidemiology and Aging Research (Prof H Brenner MD), German Cancer Research Center, Heidelberg, Germany; Institute for Cancer Research, Prevention and Clinical Network, Florence, Italy (G Carreras PhD); Applied Molecular Biosciences Unit (Prof F Carvalho PhD), EPIUnit - Public Health Institute University Porto (ISPUP) (A Ribeiro PhD), Institute of Public Health (Prof F Carvalho PhD), UCIBIO, REQUIMTE, Laboratory of Toxicology, Faculty of Pharmacy (Prof V M Costa PharmD), University of Porto, Porto, Portugal; Epidemiology and Public Health Evaluation Group (C A Castañeda-Orjuela MD), National University of Colombia, Bogota, Colombia; Department of Research and Health Technology Assessment (F Castro MD), Gorgas Memorial Institute for Health Studies, Panama City, Panama (I Moreno Velasquez PhD); Faculty of Biology (D Chu PhD), Hanoi National University of Education, Hanoi, Vietnam; Department of Immunology (Prof A Rafiei PhD), Department of Public Health (M Khazaee-Pool

h and Health Technology Assessment (F Castro MD), Gorgas Memorial Institute for Health Studies, Panama City, Panama (I Moreno Velasquez PhD); Faculty of Biology (D Chu PhD), Hanoi National University of Education, Hanoi, Vietnam; Department of Immunology (Prof A Rafiei PhD), Department of Public Health (M Khazaee-Pool PhD), Environmental Health (Prof Z Yousefi PhD), Health Sciences Research Center (M Moosazadeh PhD), Health Sciences Research Center, Addiction Research Institutes (M Khazaee-Pool PhD), Medical Laboratory Sciences (A Shamshirian BMedSc), Molecular and Cell Biology Research Center (Prof A Rafiei PhD), Toxoplasmosis Research Center (Prof A Daryani PhD), Mazandaran University of Medical Sciences, Sari, Mazandaran, Iran; Department of General Surgery (D V Davitoiu PhD), Carol Davila University of Medicine and Pharmacy, Bucharest, Romania; Departmet of Surgery (D V Davitoiu PhD), Emergency Clinical Hospital Sf.

ei PhD), Toxoplasmosis Research Center (Prof A Daryani PhD), Mazandaran University of Medical Sciences, Sari, Mazandaran, Iran; Department of General Surgery (D V Davitoiu PhD), Carol Davila University of Medicine and Pharmacy, Bucharest, Romania; Departmet of Surgery (D V Davitoiu PhD), Emergency Clinical Hospital Sf. Pantelimon, Bucharest, Romania; Nursing Department College of Health Science (G Gebremeskel MSc), School of Pharmacy (G T Demoz MPharm), Aksum University, Aksum, Ethiopia; Addis Ababa University, Addis Ababa, Ethiopia (G T Demoz MPharm); School of Public Health (H Meheretu MPH), Bahir Dar University, Bahir Dar, Amhara, Ethiopia (G A Gedefew MSc); Nursing Department (A B Demis MSc), Woldia University, Woldia, Ethiopia; Center for Nutrition and Health Research (E Denova-Gutiérrez DSc), Health and Nutrition Research Center (P Sajadi PhD), National Institute of Public Health, Cuernavaca, Morelos, Mexico; Disha Foundation, Gurgaon, Haryana, India (S Dey PhD); Deputy of Research and Technology (S Djalalinia PhD), Ministry of Health and Medical Education, Tehran, Iran; Department of Nursing (R Mohammadpourhodki MSc), Ophthalmic Epidemiology Research Center (M Emamian PhD), Shahroud University of Medical Sciences, Shahroud, Semnan, Iran; Center for Biotechnology and Fine Chemistry (J C Fernandes PhD), Catholic University of Portugal, Porto, Portugal; School of Public Health Medicine (F Fischer PhD), Bielefeld University, Bielefeld, North Rhine-Westphalia, Germany; Abadan Faculty of Medical Sciences (M Foroutan PhD), Abadan School of Medical Sciences, Abadan, Iran; Department of Cardiovascular Medicine (M M Gad MD), Cleveland Clinic, Cleveland, OH, USA; Gillings School of Global Public Health (M M Gad MD), University of North Carolina Chapel Hill, Chapel Hill, NC, USA; School of Pharmacy (G Mengistu MSc), Haramaya University, Dire Dawa, Ethiopia (G A Gedefew MSc); Department of Pharmacy (G Mengistu MSc), Nursing Department (A B Wondmieneh MSc), Wollo University, Dessie, Ethiopia; Golestan Research Center of Gastroenterology and Hepatology (F Ghaseni-Kebria MSc, G Roshandel PhD), Golestan University of Medical Sciences, Gorgan, Golestan, Iran (S Mir MSc); Occupational and Environmental Epidemiology Section (G Gorini MD), Cancer Prevention and Research Institute, Florence, Florence, Italy; Department of Gastroenterology and Hepatology (K Vosoughi MD), Department of Health Policy and Management (A T Khoja MD), Department of Radiology and Radiological Sciences (N Hafezi-Neja

cupational and Environmental Epidemiology Section (G Gorini MD), Cancer Prevention and Research Institute, Florence, Florence, Italy; Department of Gastroenterology and Hepatology (K Vosoughi MD), Department of Health Policy and Management (A T Khoja MD), Department of Radiology and Radiological Sciences (N Hafezi-Neja d MD), Johns Hopkins University, Baltimore, MD, USA; Cancer Research Center (M Khayamzadeh MD), Department of Medicinal and Pharmaceutical Chemistry (Prof A Zarghi PhD), Department of Pharmacology of Tehran University of Medical Sciences (R Nikbakhsh MD), Obesity Research Center (A Haj-Mirzaian MD), Shahid Beheshti University of Medical Sciences, Tehran, Iran; Biomedical Research Networking Center for Mental Health Network (CiberSAM), Madrid, Spain (Prof J M Haro MD); CIBERSAM (A Koyanagi MD), Research and Development Unit (Prof J M Haro MD), San Juan de Dios Sanitary Park, Sant Boi de Llobregat, Barcelona, Spain; Department of Microbiology (A Hasanzadeh PhD), Maragheh University of Medical Sciences, Maragheh, Iran; Department of Biology (M Hashemian PhD), Other, Utica, NY, USA; Department of Public Health (H Y Hassen MPH), Mizan-Tepi University, Teppi, SNNPR, Ethiopia; Unit of Epidemiology and Social Medicine (H Y Hassen MPH), University Hospital Antwerp, Wilrijk, Antwerp, Belgium; Department of Biostatistics and Epidemiology (H D Hidru MPH), Department of Public Health (B E Tesfay MPH), Adigrat University, Adigrat, Ethiopia; Department of General Surgery (M Hostiuc PhD), Emergency Hospital of Bucharest (I Negoi PhD), General Surgery Department (I Negoi PhD), Surgery 2nd Department-SUUB (C Smarandache MD), Carol Davila University of Medicine and Pharmacy, Bucharest, Romania; Department of Internal Medicine (M Hostiuc PhD), Surgery 2nd Department (C Smarandache MD), Bucharest Emergency Hospital, Bucharest, Bucharest, Romania; Division of Information and Computing Technology, College of Science and Engineering (Prof M Househ PhD), Hamad Bin Khalifa University, Doha, Qatar; Qatar Foundation for Education, Science, and Community Development, Doha, Qatar (Prof M Househ PhD); Department of Community Medicine (O S Ilesanmi PhD), University of Ibadan, Ibadan, Nigeria; Department of Epidemiology (Prof M D Ilic PhD), University of Kragujevac, Kragujevac, Serbia, Serbia; Department of Epidemiology and Biostatistics (K Innos PhD), National Institute for Health Development, Tallinn, Estonia; Surveillance and Health Services Research (F Islami PhD), American Cancer Society,

dan, Nigeria; Department of Epidemiology (Prof M D Ilic PhD), University of Kragujevac, Kragujevac, Serbia, Serbia; Department of Epidemiology and Biostatistics (K Innos PhD), National Institute for Health Development, Tallinn, Estonia; Surveillance and Health Services Research (F Islami PhD), American Cancer Society, Atlanta, GA, USA; Autism Spectrum Disorders Research Center (E Jenabi PhD), Department of Epidemiology (S Khazaei PhD), Neurophysiology Research Center (H Komaki MD), Hamadan University of Medical Sciences, Hamadan, Iran (N Mohammad Gholi Mezerji MSc); Health Services Management Department, School of Health (R Kalhor PhD), Social Determinants of Health Research Center (R Kalhor PhD), Qazvin University of Medical Sciences, Qazvin, Iran; Biology Department (Prof F Kamangar MD), Morgan State University, Baltimore, MD, USA; Non-Communicable Diseases Research Unit (Prof A P Kengne PhD), Medical Research Council South Africa, Cape Town, Western Cape, South Africa; Department of Medicine (Prof A P Kengne PhD), University of Cape Town, Cape Town, South Africa; Department of Public Health and Community Medicine (Prof Y S Khader PhD), Jordan University of Science and Technology, Ramtha, Irbid, Jordan; Department of Physiology (R Khalilov PhD), Baku State University, Baku, Azerbaijan; Epidemiology and Biostatistics Department (E A Khan MPH), Health Services Academy, Islamabad, Islamabad Capital Territory, Pakistan; Department of Medical Microbiology & Immunology (Prof G Khan PhD), United Arab Emirates University, Al Ain, Abu Dhabi, United Arab Emirates; Environmental Health Department (Prof Z Yousefi PhD), Academy of Medical Science, Tehran, Iran (M Khayamzadeh MD); Department of Public Health (A T Khoja MD), Imam Muhammad Ibn Saud Islamic University, Riyadh, Saudi Arabia; Department of Epidemiology (F Khosravi Shadmani PhD), Department of Epidemiology & Biostatistics (Prof F Najafi PhD), Department of Operating Room, School of Paramedical (M Naderi MSc), Social Development and Health Promotion Research Center (M Soofi PhD), Kermanshah University of Medical Sciences, Kermanshah, Iran; School of Medicine (Y Kim PhD), Xiamen University Malaysia, Sepang, Selangor, Malaysia; Public Health Sciences Division (J M Kocarnik PhD), Fred Hutchinson Research Center, Seattle, WA, USA; Brain Engineering Research Center (H Komaki MD), Institute for Research in Fundamental Sciences, Tehran, Iran; Catalan Institution for Research and Advanced Studies (ICREA), Barcelona, Spain (A Koyanagi M

r, Malaysia; Public Health Sciences Division (J M Kocarnik PhD), Fred Hutchinson Research Center, Seattle, WA, USA; Brain Engineering Research Center (H Komaki MD), Institute for Research in Fundamental Sciences, Tehran, Iran; Catalan Institution for Research and Advanced Studies (ICREA), Barcelona, Spain (A Koyanagi M D); Department of Medicine Brigham and Women's Hospital (V Kumar MD), Harvard University, Boston, MA, USA; Clinical Medicine and Community Health, University of Milan, Milano, Italy (Prof C La Vecchia MD); School of Population and Global Health (Prof A D Lopez PhD), University of Melbourne, Melbourne, VC, Australia; Department of General Surgery (R Lunevicius PhD), Aintree University Hospital National Health Service (NHS) Foundation Trust, Liverpool, Merseyside, UK; Department of Surgery (R Lunevicius PhD), University of Liverpool, Liverpool, Merseyside, UK; Ophthalmology Department (N Manafi MD), University of Manitoba, Winnipeg, MB, Canada; Surgery Department (A Manda MD), Emergency University Hospital Bucharest, Bucharest, 5th sector, Romania; Department of Nursing (H Meheretu MPH), Debre Markos University, Debre Markos, Ethiopia; Center for Innovation in Medical Education (B Miazgowski MD), Pomeranian Medical University, Szczecin, Zachodniopomorskie, Poland (B Miazgowski MD); Department of Biology (K A Mohammad PhD), Salahaddin University, Erbil, Iraq; ISHIK University, Erbil, Iraq (K A Mohammad PhD); Department of Epidemiology and Biostatistics (M Mohammadian BA), Bushehr University of Medical Sciences, Bushehr, Iran; Department of Epidemiology and Biostatistics (A Mohammadian-Hafshejani PhD), Shahrekord University of Medical Sciences, Shahrekord, Iran; Department of Clinical Biochemistry (A Mosapour PhD), Department of Parasitology and Entomology (L Zaki PhD), Tarbiat Modares University, Tehran, Iran; Health Systems and Policy Research Unit (S Mohammed PhD), Ahmadu Bello University, Zaria, Kaduna State, Nigeria; Institute of Public Health (S Mohammed PhD), Heidelberg University, Heidelberg, Baden Wuerttemberg, Germany; Clinical Epidemiology and Public Health Research Unit (L Monasta DSc, E Traini MSc), Burlo Garofolo Institute for Maternal and Child Health, Trieste, Italy; Department of Molecular Medicine (M Moossavi PhD), Birjand University of Medical Sciences, Birjand, Iran; Department of Epidemiology and Biostatistics (G Moradi PhD), Social Determinants of Health Research Center (G Moradi PhD, F Moradpour PhD), Kurdistan University of Medical Sciences,

d Child Health, Trieste, Italy; Department of Molecular Medicine (M Moossavi PhD), Birjand University of Medical Sciences, Birjand, Iran; Department of Epidemiology and Biostatistics (G Moradi PhD), Social Determinants of Health Research Center (G Moradi PhD, F Moradpour PhD), Kurdistan University of Medical Sciences, Sanandaj, Kurdistan, Iran; Department of Epidemiology (R Moradzadeh PhD, M Zamanian PhD), Arak University of Medical Sciences, Arak, Iran; Comprehensive Cancer Center (G Naik MPH), Department of Epidemiology (J A Singh MD), Department of Medicine (J A Singh MD), University of Alabama at Birmingham, Birmingham, AL, USA; Heidelberg University Hospital, Germany (R Nikbakhsh MD); Department of Pathology and Molecular Medicine (T O Olagunju MD), Department of Psychiatry and Behavioural Neurosciences (A T Olagunju MD), McMaster University, Hamilton, ON, Canada; Department of Psychiatry (A T Olagunju MD), University of Lagos, Lagos, Nigeria; Graduate School of Public Health (Prof E Oren PhD), San Diego State University, San Diego, CA, USA; CPO Piemonte (C Piccinelli BS), AOU Città della Salute e della Scienza, Torino, Italy; Department of Chemistry (N Rabiee PhD), Sharif University of Technology, Tehran, Iran; College of Graduate Health Sciences (A Radfar MD), A.T.

School of Public Health (Prof E Oren PhD), San Diego State University, San Diego, CA, USA; CPO Piemonte (C Piccinelli BS), AOU Città della Salute e della Scienza, Torino, Italy; Department of Chemistry (N Rabiee PhD), Sharif University of Technology, Tehran, Iran; College of Graduate Health Sciences (A Radfar MD), A.T. Still University, Mesa, AZ, USA; Medichem, Barcelona, Spain (A Radfar MD); School of Social Sciences and Psychology (Prof A M N Renzaho PhD), Western Sydney University, Penrith, NSW, Australia; Network of Immunity in Infection, Malignancy and Autoimmunity (NIIMA) (Prof N Rezaei PhD), Universal Scientific Education and Research Network (USERN), Tehran, Iran; Department of Entomology (A M Samy PhD), Faculty of Medicine (A M Saad MBBCh), Ain Shams University, Cairo, Egypt; Department of Surgery (Prof J Sanabria MD), Marshall University, Huntington, WV, USA; Department of Nutrition and Preventive Medicine (Prof J Sanabria MD), Case Western Reserve University, Cleveland, OH, USA; Centre School of Public Health and Health Management (Prof M M Santric Milicevic PhD), Faculty of Medicine Institute of Epidemiology (I S Vujcic PhD), University of Belgrade, Belgrade, Serbia, Serbia; Department of Public Health Sciences (M Sawhney PhD), University of North Carolina at Charlotte, Charlotte, NC, USA; Department of Health Promotion and Education (F Shaahmadi PhD), Alborz University of Medical Sciences, Karaj, Iran; Department of Medical Statistics, Epidemiology and Medical Informatics (M Sekerija PhD), University of Zagreb, Zagreb, Croatia; Division of Epidemiology and Prevention of Chronic Noncommunicable Diseases (M Sekerija PhD), Croatian Institute of Public Health, Zagreb, Croatia; Independent Consultant, Karachi, Sindh, Pakistan (M A Shaikh MD); Department of Hematology-Oncology (S K Siddappa Malleshappa MD), Baystate Medical Center, Springfield, MA, USA; Cancer Control Center (T Tabuchi MD), Osaka International Cancer Institute, Osaka, Osaka, Japan; Nursing Department (D B B Tadesse MSc), Institute of Tropical Medicine, Aksum, Ethiopia; Axum College of Health Science, Mekelle, Tigray, Ethiopia (D B B Tadesse MSc); Biostatistics (L Tapak PhD), Hamedan University of Medical Sciences, Hamadan, Iran; Department of Health Economics (B Tran PhD), Hanoi Medical University, Hanoi, Vietnam; Molecular Medicine and Pathology Department (K B Tran MD), University of Auckland, Auckland, New Zealand; Clinical Hematology and Toxicology (K B Tran MD), Military Medical University, Hanoi,

sity of Medical Sciences, Hamadan, Iran; Department of Health Economics (B Tran PhD), Hanoi Medical University, Hanoi, Vietnam; Molecular Medicine and Pathology Department (K B Tran MD), University of Auckland, Auckland, New Zealand; Clinical Hematology and Toxicology (K B Tran MD), Military Medical University, Hanoi, Vietnam; Trauma Research Center, Nursing Faculty (A Vahedian-Azimi PhD), Baqiyatallah University of Medical Sciences, Tehran, Iran; Psychosocial Injuries Research Center (Y Veisani PhD), Ilam University of Medical Sciences, Ilam, Iran; Competence Center of Mortality-Follow-Up, German National Cohort (R Westerman DSc), Federal Institute for Population Research, Wiesbaden, Hesse, Germany; Master of science in adult health nursing (A B Wondmieneh MSc), Addis Ababa University, Addis Ababa, Ethiopia; School of International Development and Global Studies (Prof S Yaya PhD), University of Ottawa, Ottawa, Canada; Department of Health Management, Policy and Economics (V Yazdi-Feyzabadi PhD), Health Services Management Research Center (V Yazdi-Feyzabadi PhD), Kerman University of Medical Sciences, Kerman, Iran; Department of Community Medicine (H Zandian PhD), Social Determinants of Health Research Center (T Zahirian Moghadam PhD, H Zandian PhD), Ardabil University of Medical Science, Ardabil, Iran; and Department of Preventive Medicine (Z Zhang PhD), Wuhan University, Wuhan, China.

iversity of Medical Sciences, Kerman, Iran; Department of Community Medicine (H Zandian PhD), Social Determinants of Health Research Center (T Zahirian Moghadam PhD, H Zandian PhD), Ardabil University of Medical Science, Ardabil, Iran; and Department of Preventive Medicine (Z Zhang PhD), Wuhan University, Wuhan, China. Contributors AE, SS, RS, and MA prepared the first draft. RM, MN, AE, CF, JK, LF, TA, and CA provided overall guidance. RM, MN, AE, SGS, CF, and TA managed the overall project. AE, SGS, SS, JH, AP, RX, JK, and CA analysed data. RM, MN, AE, and CB finalised the Article on the basis of comments from other authors and reviewer feedback. All other authors provided data, developed models, reviewed results, provided guidance on methods, or reviewed and contributed to the Article. Declaration of interests JMH reports his employer has done a study on stomach cancer under a contract by Eli Lilly, outside the submitted work. SLJ reports grants from Sanofi Pasteur, outside the submitted work. All other authors declare no competing interests.