NIH Research Festival
Although data from population-based studies have been used to estimate diabetes prevalence, it has been challenging to use these data to estimate the prevalence of diabetes type. This also hampers our ability to discern heterogeneity in undiagnosed populations. We examined data from 3 cohorts of the National Health and Nutrition Examination Survey (NHANES) and the 2010 Coronary Artery Risk Development in Young Adults survey (CARDIA), each with over 3,000 respondents providing biological and survey data. Diabetes prevalence, defined as reporting being diagnosed with diabetes or Hemoglobin A1C > 6.5, was between 9-13%. Six indicator variables (HOMA-IR, HOMA- %B, HOMA- %S, BMI, glucose to insulin ratio, and fasting insulin) were entered into a latent class analysis to discriminate diabetes type among individuals with diabetes. Three latent classes: likely type-1 diabetes, likely type-2 diabetes, and atypical diabetes, were confidently discerned in each of the 4 datasets. Using low C-peptide as a marker of likely type-1 diabetes in each latent class identified revealed that 97.4%, among those in the likely type-2 group, did not have low C-peptide. Additionally, after excluding likely type-1 and atypical diabetes classes, known risk factors of type 2 diabetes (e.g., race/ethnicity, waist circumference) accounted for an additional 3-5% of variance in diabetes cases compared to models including all diabetes classes. Here, we describe a novel tool for classifying diabetes type from large population-based datasets, which will improve how we use these vast datasets to examine the behavioral and environment factors associated with each type of diabetes.
Scientific Focus Area: Epidemiology
This page was last updated on Friday, March 26, 2021