Interpretable depressive symptoms screening via statistical reasoning-augmented large language models using wearable and environmental data.

Although DeLong’s test comparing the combined and clinical-only models did not reach statistical significance on weekdays (Z = 1.76, p = 0.079) or weekends (Z = 1.36, p = 0.174), adding wearable-derived behavioral features significantly improved overall model fit and reclassification, with likelihood ratio test, continuous net reclassification improvement, and integrated discrimination improvement all indicating incremental value on both weekdays (χ 2 = 37.62, df = 6, p < 0.001; NRI = 0.469, 95% CI 0.228–0.703; IDI = 0.037, 95% CI 0.021–0.053) and weekends (χ 2 = 29.22, df = 4, p < 0.001; NRI = 0.372, 95% CI 0.134–0.602; IDI = 0.018, 95% CI 0.005–0.030).

also in 111,027 other papers