Adding step counts, PRS to clinical model improves T2D prediction

Daily step counts do not consistently predict the incidence of type 2 diabetes (T2D), as shown by the varying step-count thresholds for reducing T2D risk across genetic risk groups, reports a study.

A team of investigators conducted a prospective cohort study of 4,589 adults in the National Institutes of Health’s All of Us Research Program with valid Fitbit step data and whole-genome-derived PRS. Individuals with T2D before or within a 180-day lead-in period were excluded.

Incident T2D was characterized by HbA1c ≥6.5 percent, plasma glucose ≥126 mg/dL, or an All of Us T2D condition record. Cox and machine-learning survival models were used in the analyses.

Some 265 participants developed T2D (5.77 percent cumulative incidence; 17.27 per 1,000 person-years) over a median of 2.92 years.

Risk-reducing thresholds were approximately 7,000 steps/day (p<0.001), but these differed by Polygenic Risk Score (PRS) group (high vs low: ∼7,800 vs ∼5,800 steps/day; p<0.001).

Each additional 1,000 steps/day appeared to further reduce the risk of T2D (adjusted hazards ratio [aHR], 0.83, 95 percent confidence interval [CI], 0.79‒0.88; p<.005), but every 1-SD increase in PRS tended to elevate such risk (aHR, 2.62, 95 percent CI, 2.32‒2.96; p<0.005).

Adding steps to a clinical model improved the C-index from 0.748 to 0.774, and adding PRS further increased it to 0.867. Moreover, penalized Cox achieved the highest discrimination (C-index 0.859), followed by survival support vector machine (0.85) and classical Cox (0.846). The best calibration was seen with Random Survival Forests.

“Step counts and PRS provided independent, complementary predictive information, and their combination improved prediction of incident T2D,” the investigators said.