Digital prediction of falls and fractures in older adults: a systematic review

Background

Falls cause over 550,000 deaths annually among older adults, leading to fractures, disability, and substantial healthcare costs. Traditional risk assessment tools show limited predictive accuracy. Building on the DARE (Digital Lifelong Prevention) Italian national research initiative, this review evaluated the accuracy of electronic health record (EHR)-based artificial intelligence (AI) models and wearable sensor tools for predicting fall and fracture risk.

Methods

Following PRISMA 2020, we searched MEDLINE, Embase, CENTRAL, HMIC, and Epistemonikos from May 2024 to October 2025 to update and expand the systematic review underpinning NICE guideline NG249 (2025) on falls. We included cohort studies evaluating EHR-based prediction models and wearable devices in adults ≥65 years or 50–64 years at increased fall risk. Risk of bias was assessed using PROBAST+AI and certainty of evidence using a modified GRADE approach. The protocol was preregistered on the Open Science Framework.

Results

Of 5,768 records screened, 6 new studies met the inclusion criteria. Combined with 17 studies from the NICE review, 23 studies were analysed: 15 EHR-based, 8 on wearable devices. EHR-based models showed areas under the curve (AUCs) from 0.69 to 0.82 in community settings, 0.57 to 0.85 in hospitals, and up to 0.74 in externally validated residential care. Wearable devices showed AUCs of 0.60-0.76 in internally validated studies and 0.84 in one small externally validated cohort. Evidence on fractures was scarce. Meta-analysis was precluded by heterogeneity. The certainty of the evidence was low, mainly due to risk of bias, inconsistency, and inadequate calibration reporting.

Conclusions

EHR-based tools showed more consistent performance than wearable devices, but limited external validation and poor calibration reporting constrain implementation. Findings will inform evidence-based recommendations within the DARE initiative for integration into fall prevention pathways.

Key messages

• Digital prediction tools, particularly EHR-based tools, showed moderate discrimination for fall prediction across care settings, but calibration was rarely reported.

• Wider external validation and standardised calibration reporting are needed before integration into fall prevention and healthy ageing strategies.