Accepted for/Published in: Journal of Medical Internet Research
Date Submitted: Sep 26, 2025
Date Accepted: Apr 6, 2026
(closed for review but you can still tweet)
Warning: This is an author submission that is not peer-reviewed or edited. Preprints - unless they show as "accepted" - should not be relied on to guide clinical practice or health-related behavior and should not be reported in news media as established information.
Machine learning and deep learning models for predicting future falls in community-dwelling older adults: a systematic review and meta-analysis of longitudinal evidence
ABSTRACT
Background:
Machine learning (ML) and deep learning (DL) show promise for fall risk prediction, but prior reviews focused mainly on real-time fall detection, in-hospital falls, or conventional statistical models. The performance and applicability of ML-based models for predicting future falls in community-dwelling older adults remain unclear.
Objective:
To systematically review and evaluate ML and DL models developed to predict future fall risk in community-dwelling older adults.
Methods:
Six databases were searched from inception to September 23, 2024 (updated on 31 August 2025)(PROSPERO registration: CRD42024580902). Studies were eligible if they used ML models to predict future falls in community-dwelling individuals aged ≥60 years. Risk of bias was assessed using the Prediction Model Risk of Bias Assessment Tool (PROBAST). A random-effects meta-analysis was performed to pool the discrimination of validated models and explore heterogeneity.
Results:
Twenty-seven studies were included, with 18 studies focusing on general community-dwelling older adults. The primary outcome was fall occurrence, with incidence rates ranging from 1.6% to 46.6%. Most models (n = 22) applied ML techniques, while 5 used DL. The most common approaches were random forest (RF, n = 5) and decision tree (n = 5). Data modality included text (n=17), sensor (n=5), image (n=1), and multimodal sources (n=4). Frequently selected predictors included age (n=14), sex (n=13), fall history (n=12), depression (n=10), and basic activities of daily living (n=8). Eighteen studies conducted cross-validation, but only one model underwent external validation. All models were rated as having a high bias risk, with 16 studies including fewer than 1,000 participants. Meta-analysis of 10 validated models yielded a pooled AUC of 0.79 (95%CI 0.70-0.88), indicating good discrimination but substantial heterogeneity (I²=99.8%). Pooled sensitivity and specificity were 0.71 (95%CI 0.47-0.95) and 0.75 (95%CI 0.59-0.90), respectively.
Conclusions:
ML-based models show considerable potential for predicting future falls among older adults. However, substantial heterogeneity and methodological limitations remain. Future research should emphasize external validation, transparent reporting, and rigorous modeling approaches to enhance applicability.
Citation
