Answers are grounded in this paper and cite their sources.
Rural oldest-old adults in China may experience substantial functional care needs alongside limited formal care resources. This study developed and internally evaluated interpretable machine-learning classifiers of current long-term care (LTC) need, defined by activities of daily living (ADL) dependence, in a rural oldest-old CLHLS analytical sample. Secondary data from the Chinese Longitudinal Healthy Longevity Survey were analysed for 4,319 rural respondents aged ≥ 80 years from the 2002, 2005, 2008, 2011, 2014 and 2018 analytical waves. Current LTC need was coded as assistance with at least one of six Katz ADL activities. Four classifiers were evaluated using a stratified 75%/25% development-test split, repeated 10 × 5-fold cross-validation, training-only model specification, bootstrap uncertainty estimation, calibration assessment and sensitivity analyses. ADL-defined current LTC need occurred in 2,332/4,319 respondents (54.0%), an unweighted analytical-sample characteristic rather than a population prevalence estimate. In held-out testing, logistic regression achieved AUC-ROC 0.965, AUC-PR 0.971, Brier score 0.074, accuracy 0.897 and F1 score 0.905. Gradient boosting and XGBoost each achieved AUC-ROC 0.964 and random forest 0.958. Paired DeLong tests showed no evidence of an AUC difference between logistic regression and gradient boosting ( p = 0.445) or XGBoost ( p = 0.555). Interpretable models showed high internal discrimination for contemporaneous ADL-defined care need in this analytical sample. The model is intended as a research benchmark for prioritising structured assessment when complete ADL ascertainment is not yet available, not as a replacement for direct ADL assessment or as a prospective risk model. External validation and prospective impact evaluation are required before implementation. Not applicable.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: