Depressive symptoms and low cognitive performance in Chinese older adults: A severity-graded epidemiological analysis and a BARNN-inspired prospective risk-prediction framework with strict participant-level temporal validation

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Background Late-life depression is robustly associated with subsequent cognitive impairment, yet the dose-response gradient and temporally validated, uncertainty-aware prediction tools remain understudied in Chinese older adults. Objective To characterize the severity-graded epidemiological association between depressive symptoms and low cognitive performance, and to develop and validate a prospective risk-prediction framework with explicit uncertainty estimation. Methods We analyzed 31,570 observations from 12,494 participants aged ≥45 years in the China Health and Retirement Longitudinal Study (2011–2020). Depressive symptoms were measured with the 10-item Center for Epidemiologic Studies Depression Scale (CESD-10); low cognitive performance was defined as total cognition score < 10. Layer 1 reported severity-graded prevalence and adjusted odds ratios. Layer 2 implemented a BARNN-inspired prospective state-transition predictor via a diverse tree ensemble, yielding predictive probability and epistemic uncertainty. Both layers used strict participant-level temporal validation. Results Clinically significant depressive symptoms (CESD-10 ≥ 10) were associated with twofold higher odds of low cognitive performance (OR = 2.07, 95% CI 1.97–2.18), with a monotonic dose-response gradient across severity tiers (20.8% to 46.0%, p < 0.001). In prospective validation, the ensemble achieved comparable discrimination (AUC-ROC 0.838 versus 0.839 for logistic regression, P = 0.85) but superior calibration (Brier score 0.099 versus 0.126), plus per-sample uncertainty estimates and an explicit algorithmic deferral tier. The frozen four-tier rule produced a steep risk gradient (Low 5.8%, Moderate 24.9%, High 63.0%; OR high versus low=27.8). Conclusions Depressive symptom severity shows a robust graded association with low cognitive performance in Chinese older adults. The framework supports reliable ordinal risk stratification with algorithmic deferral, providing a transferable template for population-level cognitive monitoring.