Disability risk prediction models in community-dwelling older adults: a systematic review

Comprehensive Summary

This study, presented by Zhang and colleagues, reviews existing prediction models designed to estimate disability risk in community-dwelling older adults. The researchers conducted a systematic review following PRISMA guidelines and searched multiple major databases for studies that developed or validated disability risk prediction models. Two reviewers extracted data and assessed methodological quality and bias using the CHARMS checklist and the updated PROBAST+AI evaluation framework. The review included 19 studies describing 19 different predictive models involving over 280,000 participants. Reported discrimination performance varied, with AUC or C-index values generally ranging from approximately 0.62 to 0.85 during development and 0.65 to 0.80 in validation. Machine learning approaches showed slightly higher predictive accuracy than traditional regression methods. Common predictors across models included age, physical function, cognitive function, cardiovascular disease, and sex. Despite reasonable predictive performance, most models showed substantial methodological weaknesses, including high risk of bias, incomplete reporting of calibration, and limited external validation. In the discussion, the authors emphasize that although many models perform moderately well, their reliability and real-world applicability remain limited due to inconsistent methodology and insufficient validation across populations.

Outcomes and Implications

As population aging increases the prevalence of disability among older adults, placing significant strain on healthcare systems and long-term care services. Early identification of individuals at risk allows clinicians and public health systems to implement preventive interventions that maintain independence and reduce healthcare costs. Clinically, disability prediction models could support screening programs that identify older adults who may benefit from early rehabilitation, lifestyle interventions, or monitoring. However, the study demonstrates that many existing models lack adequate validation and may not generalize across healthcare systems or populations. As a result, they are not yet reliable enough for widespread clinical implementation. The authors emphasize the need for future research that follows standardized reporting guidelines, incorporates broader predictors—including social and environmental factors—and conducts large-scale external validation. With these improvements, prediction tools could eventually become useful components of geriatric assessment and clinical decision-support systems, helping healthcare providers target preventive care strategies more effectively and reduce the long-term burden of disability.