Machine Learning Predicts In-Hospital Falls Effectively
This retrospective multicenter study (N=83,917 inpatients aged ≥65 from two Nippon Medical School hospitals) developed and validated machine learning models to predict in-hospital falls, which occurred in 2.6% (2,173) of patients. CatBoost achieved the highest F1-score (0.189) and AUPRC (0.112), while LightGBM showed the best calibration slope (0.964). Authors recommend EHR integration for real-time risk scoring to target toileting and mobility interventions.
Key Points
- 1Identified CatBoost and LightGBM as top-performing models on 83,917 elderly inpatient records
- 2Showed low albumin, impaired transfer ability, sedative-hypnotics, and diabetes drugs strongly contribute to fall risk
- 3Enable integration into EHRs for real-time scoring to trigger toileting assistance and mobility support
Scoring Rationale
Large multicenter dataset and rigorous calibration deliver practical insights, but modest discrimination and low F1 limit immediate clinical adoption.
Sources
Public references used for this report.
Practice with real Health & Insurance data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Health & Insurance problems
