Trends in Diabetes Subgroups and Their Risk for All-Cause, Cardiovascular Disease and Diabetes-Specific Mortality in the US: A Data-Driven Reproducible Machine Learning Approach

Background: Diabetes has been described as a heterogeneous entity which can be studied through data-driven subgroups (obesity related [MOD], severe-insulin deficient [SIID], severe-insulin resistant [SIRD] and age-related diabetes [MARD]). However, trends in prevalence and mortality risk are still unclear. Aims: To analyze diabetes subgroup trends and to evaluate mortality risk in the US. Methods: Data and follow-up causes of mortality (all-cause, cardiovascular disease, and diabetes specific) was collected from NHANES cycles 1999–2018. Subgroup diabetes classification was performed using the self-normalizing neural networks algorithm using clinical parameters (HbA1c, time since diabetes diagnosis, HOMA2-IR, HOMA2-B, and BMI) proposed by Bello-Chavolla et al (https://bit.ly/3jSm1xv). Prevalence was estimated using sample weights. 2-year cycles were used as a continuous variable to evaluate the biannual change (BC) of the overall prevalence of diabetes and subgroups. Trends were stratified according to race. Cox-proportional and Fine-Gray semiparametric hazard regression models were used to evaluate mortality risk. Results: Data from 59,204 adult subjects was extracted for trend analysis. Follow-up information was obtained for 3,980 subjects. Diabetes prevalence in the US increased from 8.2% (95%CI 7.8–8.6) in 1999–2000 to 13.9% (95% CI 13.4–14.4) in 2017–2018 (BC 1.38%, 95% CI 1.20–1.56, p<0.001). Non-Hispanic Blacks had the largest increase in diabetes prevalence (BC: 1.40%, 95%CI 0.71–2.08, p=0.027), followed by Non-Hispanic Whites (BC: 1.36%, 95%CI 1.13–1.58, p<0.001), and Mexican Americans (BC: 1.33%, 95%CI 1.20–1.54, p<0.001). Regarding diabetes subgroups, MARD had the highest prevalence, with a moderate increase over time; however, MOD had the greatest increase over time (1.5%, [95%CI 1.2–1.8] to 4.5% [95%CI 4.0–5.0]; BC: 0.73% [95%CI 0.60–0.86], p<0.01). Both SIRD and SIID had non-significant increases in prevalence during the studied period. Non-Hispanic Blacks had an increase in the prevalence in MOD and SIID, Mexican Americans in MOD and SIRD, and non-Hispanic Whites in MOD and MARD. Compared with MOD, the risk for all-cause mortality was higher for MARD (HR 2.9 95% CI: 2.1–3.9), SIRD (HR 2.0 95% CI: 1.5–2.8), and SIID (HR 1.6 95% CI: 1.1–2.3). For CVD mortality, only MARD (HR 2.8 95% CI: 1.4–5.7) and SIRD (HR 2.5 95% CI: 1.2–5.3) displayed higher risk. For diabetes-specific mortality, only MARD (HR 2.2 95% CI: 1.3–3.7) was associated. Conclusion: There is an overall increase in diabetes prevalence and its subgroups from 1999 to 2018; MORD had the highest increase. The risk for all-cause, CVD and diabetes-specific mortality was different among subgroups. Our results supports the use of diabetes subgroups for a better understanding of diabetes and its complications.