(084) A dynamic fall risk prediction and monitoring tool in aged care using routinely collected electronic health data: Development and internal validation of the FRIPAC tool

Background: Falls cause a substantial challenge in residential aged care (RAC) homes, requiring ongoing risk assessment for effective prevention. Existing falls prediction tools are static, perform poorly, and are unable to capture evolving risk factors over time.

Objectives: To develop and internally validate dynamic falls risk prediction models and create point-based scoring systems separately for residents with and without dementia.

Methods: Study design: A longitudinal cohort study design leveraging routinely collected electronic data spanning over eight years (2014-2022). The sample included 5,492 residents, with a 70:30 split for training and validation.
Settings: 27 RAC homes in Sydney, New South Wales, Australia.
Exposures/predictors/data source: A linked database containing demographics, health conditions, and daily medication administration information.
Outcome measures: Incidence of falls. Data sourced from incident reports.
Statistical analysis: We tracked residents for 60 months, using monthly landmarks with 1-month prediction windows. We employed landmarking dynamic prediction for model development, a time-dependent area under the curve (AUC) for model evaluations, and a regression coefficient approach to create point-based scoring systems.

Results: The falls incident rate was 7.76 falls/1000 resident-day. The model identified 15 predictors of falls in dementia (9 medication-related) and 12 in non-dementia cohorts. Falls history was the key predictor of subsequent falls in both dementia (HR 4.75, 95% CI 4.45-5.06) and non-dementia cohorts (HR 4.20, 95% CI 3.87-4.57). The use of drugs for bone disease was the only protective factor against falls, specifically in the dementia cohort (HR 0.79, 95% CI 0.66-0.96). The AUC across landmarks ranged from 0.67-0.87 for dementia and from 0.66-0.86 for non-dementia cohorts but generally remained between 0.75-0.85 for both cohorts, indicating 'good' to 'very good' discrimination. The total point risk score ranged from -2 to 57 for dementia and 0 to 52 for non-dementia cohorts.

Conclusions: Our novel risk prediction models and scoring systems provide timely person-centred information for continuous and easily interpretable monitoring of falls risk in RACFs. If integrated as a decision support tool, it could empower the implementation of targeted proactive interventions to prevent falls.