Prospective associations of diabetes with 15 cancers in 2.2 million UK and Chinese adults

Abstract

Background: Diabetes has been associated with the risk of numerous cancers, but the causal relevance of many of these associations remains unclear.

Methods: We investigated associations between diabetes and risks of 15 cancers using Cox-regression and individual-level data from 2.2 million adults (334 978 incident cancer cases) in 3 prospective cohorts, UK Biobank, Million Women Study, and China Kadoorie Biobank. The potential impact of residual confounding was assessed by examining changes in diabetes-associated log hazard ratios (HRs) after adjustment for key confounders.

Results: In combined analyses of individual participant data from 3 studies, diabetes was positively associated with the risk of 11 cancers, most notably liver (HR = 2.04, 95% CI = 1.87 to 2.23), pancreas (HR = 1.62, 95% CI = 1.48 to 1.77), and bladder (HR = 1.44, 95% CI = 1.29 to 1.62) cancer. The positive associations of diabetes with cancers of the breast, endometrium, kidney, and esophageal adenocarcinoma were substantially attenuated (>50%) after adjustment for confounders. The risks were similar in UK and Chinese populations except for liver cancer for which the adjusted hazard ratio was greater in UK than Chinese adults (HR = 2.58, 95% CI = 2.28 to 2.92, vs HR = 1.61, 95% CI = 1.43 to 1.83; Phet = 2.5 x 10-6). For liver cancer, the excess risk associated with diabetes increased with increasing body mass index (Ptrend = 2.7 x 10-4) and alcohol intake (Ptrend = .02). Diabetes was inversely associated with incidence of prostate cancer (HR = 0.78, 95% CI = 0.73 to 0.85) but positively associated with mortality (HR = 1.25, 95% CI = 1.00 to 1.55).

Conclusions: Diabetes increases the risk of liver, pancreatic, and bladder cancer in UK and Chinese populations. It may also have a lesser effect on stomach, colorectal cancer, and leukemia, but its associations with other cancers could well be explained by confounding and/or other biases.

Figure 1.

Age-standardized incidence rates of cancers after age 50 years by sex in UK and Chinese studies. Standardized by the 5-year age groups of 2021 world population structure. Abbreviation: NHL = non-Hodgkin lymphoma.

Figure 2.

Associations of diabetes with risks of 15 cancers in UK and Chinese studies. Cox regression models comparing participants with vs without diabetes with attained age as the underlying time variable, stratified by year of birth, region and sex, and adjusted for socioeconomic factors, smoking, alcohol, physical activity, adiposity, hypertension, blood pressure, and female reproductive factors. Each solid square represents the estimated hazard ratio. The horizontal lines indicate 95% confidence intervals. Hazard ratios for UKB and MWS are the combined results of the cohort-specific hazard ratios from UKB and MWS in fixed-effect meta-analysis, and the overall hazard ratios are the combined results of all 3 cohorts, which are represented by the diamonds. Esophageal AC and esophageal SCC indicate adenocarcinoma and squamous cell carcinoma of esophagus, respectively. Number of events are the number of participants with cancer in those with or without prevalent diabetes at baseline. Em dashes for esophageal AC in CKB indicate that this analysis was not conducted in CKB. P_het indicates P values for heterogeneity test after adjustment for FDR at 0.05. Cancers with FDR-adjusted statistically significant P value (<0.05) in the meta-analysis of all 3 studies are marked with an asterisk on the hazard ratio (95% CI). Abbreviations: CI = confidence interval; CKB = China Kadoorie Biobank; DM = diabetes mellitus; FDR = false discovery rate; HR = hazard ratio; MWS = Million Women Study; NHL = non-Hodgkin lymphoma; UKB = UK Biobank.

Figure 3.

Associations of diabetes with risks of 15 cancers in UK and Chinese studies by level of adjustment. Cox regression comparing associations between diabetes and risk of cancers in the basic model with adjustment for socioeconomic factors only (highest education in all 3 cohorts and Townsend Deprivation index and ethnicity in UKB and MWS) with those in the fully adjusted model. Changes in the estimated associations are calculated by taking the proportional difference between the fully adjusted log hazard ratio and the log hazard ratio from the basic model. Abbreviations: AC = adenocarcinoma; CI = confidence interval; HR = hazard ratio; MWS = Million Women Study; NHL = non-Hodgkin lymphoma; SCC = squamous cell carcinoma; UKB = UK Biobank.

Figure 4.

Associations of diabetes with risk of liver cancer in UK and Chinese studies by levels of (A) BMI and (B) alcohol intake. *The average alcohol intake level in the subgroup of CKB with less than 7 units per week was calculated within those who drank alcohol every week. P_het and P_trend indicate P values for heterogeneity test and trend test. Abbreviations: avg = average; BMI = body mass index; CI = confidence interval; CKB = China Kadoorie Biobank; DM = diabetes mellitus; HR = hazard ratio; MWS = Million Women Study; UKB = UK Biobank.