Self-reported indicators as a community pre-screening aid for fall risk in middle-aged and older Chinese adults: a longitudinal CHARLS study

Existing fall risk assessment tools often rely on objective performance tests or clinical diagnoses, limiting their feasibility for large-scale community application. This study examined whether easily obtainable self-reported indicators are independently associated with subsequent falls among middle-aged and older Chinese adults, evaluated heterogeneity across subgroups, and developed a parsimonious nomogram as a community pre-screening aid. Data from four CHARLS waves (2011–2018) were used to identify 46,405 eligible adjacent-wave pairs from 20,364 participants aged ≥ 45 years. We first examined descriptive same-wave associations between visit-level indicators and falls reported for the preceding 12 months, and then fitted lagged GEE models relating indicators and covariates measured at wave T to falls reported in the 12 months before the next visit (T + 1). Association estimates were pooled across 20 multiply imputed datasets using Rubin’s rules. The seven-variable nomogram was developed and internally evaluated in a separate prediction cohort with one eligible interval per participant. Exploratory interactions across candidate subgroups and recorded fall-history groups were assessed with false discovery rate correction. In the descriptive same-wave analysis, all five indicators were associated with falls reported for the preceding 12 months. In the primary lagged analysis, recorded prior falls (OR = 2.540, 95% CI 2.402–2.687), IADL impairment (OR = 1.050, 1.018–1.082), and self-rated health (OR = 0.923, 0.894–0.954) were associated with the next-wave fall report, whereas knee pain ( P = 0.056) and sleep duration ( P = 0.235) were not statistically significant. Exploratory analyses found a stronger same-wave IADL association among participants without an earlier recorded fall (FDR-adjusted interaction P = 0.002); the lagged contrast had the same direction but was less conclusive. In the held-out prediction sample, the seven-variable model showed modest discrimination (AUC = 0.668, 95% bootstrap CI 0.649–0.687). Its Brier score was 0.132 versus 0.140 for a prevalence-only prediction (scaled Brier score = 0.057). The full 21-variable model had an AUC of 0.674 in the same held-out sample; the AUC difference was 0.006 (95% paired-bootstrap CI − 0.002 to 0.015). All five indicators showed descriptive same-wave associations, but only recorded prior falls, IADL impairment, and self-rated health showed associations in the primary lagged model. However, given the modest predictive accuracy, the proposed nomogram is best positioned as a low-cost pre-screening or risk-awareness aid for community health workers, rather than a standalone diagnostic or screening instrument. It may help identify individuals who warrant further comprehensive assessment. External validation and implementation research are essential before routine adoption.