Development and Validation of a Diabetes Risk Prediction Model With Individualized Preventive Intervention Effects

Type 2 diabetes risk prediction models lack the option to predict risk conditional on initiating different preventive interventions. Our objective was to develop and validate a diabetes risk prediction model with individualized preventive intervention effects among racially diverse populations.

The derivation cohort included participants in the Diabetes Prevention Program (DPP) trial randomized to placebo, metformin, or intensive lifestyle intervention (n = 2640). A risk prediction model for incident diabetes was developed using Cox proportional hazards regression using clinically available predictors: sex, glycated hemoglobin, fasting plasma glucose (FPG), body mass index (BMI), triglycerides, and intervention. To create individualized intervention effects, pairwise interactions between intervention and age, FPG, and BMI were included. The discrimination, calibration, and net benefit of the model's 3-year predictions for incident diabetes were internally validated within the DPP and externally validated among participants with prediabetes in the Multi-Ethnic Study of Atherosclerosis (MESA; n = 2104).

In DPP and MESA, mean (SD) age was 51 years (11) and 64 (10), and 67% and 50% of participants were women, respectively. The mean C-statistic was 0.71 [95% confidence interval (CI): 0.68, 0.74] in DPP and 0.86 (95% CI: 0.83, 0.88) in MESA. The optimal preventive intervention (lowest 3-year risk) was lifestyle for 86% and 97% of DPP and MESA participants, respectively, and metformin for the remaining. Model performance was similar across race/ethnicity groups.

This is the first study to develop and validate a diabetes risk prediction model with individualized preventive intervention effects that may improve clinical decision-making and diabetes prevention.