Performance of the Framingham Offspring incident diabetes risk model in severe hypercholesterolemia

Background

The performance of risk prediction models for incident type 2 diabetes mellitus (T2DM) in severe hypercholesterolemia populations remains uncertain. A T2DM prediction risk model was developed with the Framingham Offspring cohort and showed good discrimination to predict T2DM risk in a middle-aged1. However its performance in severe hypercholesterlemia stratified by presence or not of genetic variant for familial hypercholesterolemia (FH) was not studied.

Methods

The inclusion criteria were hypercholesterolemic individuals aged 18 years or older who underwent genetic testing for FH. Index cases were selected for genetic testing if they had an FH-compatible phenotype, i.e., LDL-C levels ≥ 190 mg/dL and subsequently ≥ 230 mg/dL. We also evaluated individuals with hypercholesterolemia identified through the cascade screening program who had no causal FH variants detected. This group served as the control group for negative genetic variants.

We used an adapted Framingham Offspring diabetes risk score (excluding parental history of T2DM due to lack of this information in our data) in a retrospective cohort of 851 individuals without diabetes at baseline. Discrimination was assessed using Harrell’s C-index with 95% confidence intervals derived from 1,000 bootstrap resamples. Calibration at 60 months was evaluated using deciles of predicted risk and Kaplan–Meier curves were performed.

Results

See on table 1 characteristics of the population according to Framingham Offspring T2DM risk score tertiles. During a median of 65 months of follow-up (interquartile range 39–90), 86 participants developed diabetes (incidence 1.87 per 100 person-years). In the overall population, the Framingham T2DM score showed good discrimination (C-index 0.77; 95% CI 0.71–0.82). Predictive performance differed markedly by variant status:Variant positive: C-index 0.87 (95% CI 0.82–0.91)Variant negative: C-index 0.69 (95% CI 0.61–0.78)

The absolute difference in discrimination was 0.17 (95% CI 0.09–0.26), indicating significantly superior predictive accuracy among variant carriers. In relation to calibration, the overall cohort, predicted 5-year risks ranged from 1.1% in the lowest decile to 27.4% in the highest decile. Observed Kaplan–Meier–estimated diabetes incidence increased progressively across risk deciles, ranging from 1.5% in the lowest decile to 30.2% in the highest decile. The same was observed when dividing by tertiles (see figure 1). Calibration was generally good across most risk strata.

Conclusions

The Framingham Offspring T2DM risk score demonstrates good overall performance in this cohort of severe hypercholesterolemia; however, discrimination is substantially stronger among those with variant positive for FH. These findings suggest that genetic background may modify predictive accuracy and highlight the importance of subgroup validation in genetically defined hypercholesterolemic populations.Characteristics of the populationKaplan–Meier curves for incident diabetes