A Systematic Review of Multivariate Models for Predicting Fall-Related Injuries in Older Adults

To evaluate the quality, risk of bias, and clinical applicability of prediction models for fall-related injuries in older adults.

Numerous prediction models for fall-related injuries in older adults have been developed, but their quality and applicability in clinical practice and future research remain uncertain.

We systematically searched Medline (via OVID), Embase (via OVID), Cochrane Library, CINAHL (via EBSCO), Web of Science, and Scopus from inception to May 23, 2024, for English-language publications. All observational and experimental studies reporting the development or validation of any multivariable prediction model for fall-related injuries in older adults were included. The risk of bias and applicability was assessed using the PROBAST, and the reporting quality was measured based on the TRIPOD + AI checklist. Data were synthesized using a narrative synthesis approach.

Thirty-one models from 15 studies were included. Twelve studies focused on the development and/or internal validation of a model, two studies dealt with development and external validation using a nonrandom split-sample, and one study externally validated existing models. The reported model discriminative statistics exhibited a broad range, from 0.54 to 0.89, in internal or external validation contexts. The risk of applicability was low for all studies, while the overall risk of bias was high in all studies (100.0%). High bias risk was notably prevalent in the analysis domain (100% of studies) and observed in the predictors (33.3%), participants (26.7%), and outcome (6.7%) domains. Median adherence to TRIPOD + AI reporting items was 56.4%.

The discriminative ability in the prediction models of fall-related injuries in older adults varied widely, with all models exhibiting a high risk of bias according to the PROBAST. Upcoming research should focus on developing high-quality and reproducible models that undergo proper external validation, followed by studies on implementation.

Existing fall-related injuries prediction models exhibit high bias and inconsistent accuracy, limiting clinical utility. Nursing leaders should advocate for future models that undergo thorough internal and external validation, ensure sufficient events per variable, properly handle missing data, and adopt transparent reporting practices. This will underpin data-driven clinical decisions and enable targeted fall prevention strategies in vulnerable older adults.