Self-reported clues to future falls: Chinese study tests a low-cost community pre-screening tool

Many existing tools for assessing a person’s risk of falling depend on objective performance tests or clinical diagnoses, which makes them difficult to deploy at scale in community settings. A new longitudinal study drawing on the China Health and Retirement Longitudinal Study, known as CHARLS, asked a different question: can simple, easily obtainable self-reported indicators predict falls that occur later, and can they be combined into a practical pre-screening aid for community health workers? The research, published in Aging Clinical and Experimental Research, analyzed four CHARLS waves collected between 2011 and 2018, identifying 46,405 eligible adjacent-wave pairs from 20,364 participants aged 45 years or older.

The investigators took a two-step analytical approach. First, they examined descriptive same-wave associations, asking whether five self-reported indicators — recorded prior falls, impairment in instrumental activities of daily living, self-rated health, knee pain, and sleep duration — were linked to falls reported for the preceding 12 months at the same visit. All five indicators showed such descriptive associations. Second, and more demanding, they fitted lagged generalized estimating equation models that related indicators and covariates measured at one wave to falls reported in the 12 months before the next visit, roughly two years later. Association estimates were pooled across 20 multiply imputed datasets using Rubin’s rules to account for missing data.

The lagged results were more selective than the same-wave findings. Recorded prior falls carried the strongest association with the next-wave fall report, with an odds ratio of 2.540 (95% confidence interval 2.402–2.687). IADL impairment was also associated with subsequent falls (OR = 1.050, 95% CI 1.018–1.082), as was self-rated health (OR = 0.923, 95% CI 0.894–0.954). In contrast, knee pain (P = 0.056) and sleep duration (P = 0.235) did not reach statistical significance in the primary lagged model. This pattern underscores a recurring theme in fall-risk research: cross-sectional correlations can overstate what self-reported information actually predicts prospectively.

Exploratory subgroup analyses added nuance. The same-wave association between IADL impairment and falls was stronger among participants without an earlier recorded fall (false discovery rate-adjusted interaction P = 0.002), suggesting that functional impairment may be a particularly informative signal in people whose fall history appears clean. The corresponding lagged contrast pointed in the same direction but was less conclusive. These interactions were explicitly framed as exploratory, with false discovery rate correction applied across candidate subgroups and recorded fall-history groups.

To translate the associations into something usable, the team developed a parsimonious seven-variable nomogram and evaluated it in a separate prediction cohort containing one eligible interval per participant. In this held-out sample, the model achieved modest discrimination, with an area under the receiver operating characteristic curve of 0.668 (95% bootstrap CI 0.649–0.687). Its Brier score was 0.132, compared with 0.140 for a prevalence-only prediction that ignores individual characteristics, yielding a scaled Brier score of 0.057. For context, a full 21-variable model reached an AUC of 0.674 in the same held-out sample; the difference between the two models was 0.006 (95% paired-bootstrap CI −0.002 to 0.015), indicating that the extra variables added little predictive value.

Those numbers matter for interpretation. An AUC of 0.668 reflects discrimination that is better than chance but far from the accuracy needed for confident individual-level classification. The authors are explicit about this limitation: given the modest predictive accuracy, the 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. Its intended role is to help identify individuals who warrant further comprehensive assessment, not to replace clinical evaluation. External validation and implementation research are described as essential before any routine adoption.

The study’s design choices strengthen its internal credibility. Using lagged models means the predictors were measured before the falls they were used to predict, reducing the reverse-causation ambiguity that affects same-wave analyses. Pooling estimates across 20 imputed datasets addresses the missing data that are inevitable in large population surveys. Restricting the prediction cohort to one eligible interval per participant and evaluating performance in a held-out sample guards against the optimism that inflates apparent accuracy when models are judged on the data used to build them. Bootstrap confidence intervals quantify the uncertainty around the reported AUCs.

Several caveats remain. The findings come from middle-aged and older Chinese adults surveyed in CHARLS between 2011 and 2018, and generalization to other populations or health systems has not been demonstrated. The nomogram’s discrimination was modest even within the development context, and the exploratory interaction findings require confirmation. The article itself was shared early as a peer-reviewed, accepted manuscript carrying a permanent DOI, subject to further edits before the final Version of Record appears. All of these constraints are acknowledged by the authors rather than obscured.

What the study ultimately offers is a calibrated, modest claim. Simple self-reported indicators — particularly a recorded history of previous falls, difficulty with instrumental activities of daily living, and poorer self-rated health — are associated with falls reported roughly two years later in a large Chinese cohort, and a compact seven-variable model built from such information performs slightly better than knowing the population’s baseline fall rate alone. That performance level is consistent with a screening aid that flags people for closer attention, not a tool that determines who will fall. The authors’ recommendation that external validation and implementation research precede routine use reflects the gap between statistical association in survey data and reliable performance in everyday community practice.

Subject of Research: Longitudinal analysis of self-reported indicators for predicting falls among middle-aged and older Chinese adults in the CHARLS cohort

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

Article References: Chen, W., Geng, Y., Zhu, Y., Han, C., Gong, C., Liu, S., Lu, X., & Wang, Y. (2026). Self-reported indicators as a community pre-screening aid for fall risk in middle-aged and older Chinese adults: a longitudinal CHARLS study. Aging Clinical and Experimental Research. https://doi.org/10.1007/s40520-026-03557-3

Image Credits: AI Generated

DOI: 10.1007/s40520-026-03557-3

Keywords: fall risk, CHARLS, self-reported indicators, nomogram, community pre-screening, older adults, IADL impairment, self-rated health, longitudinal study, predictive accuracy, China, geriatrics