A Hybrid Clinical–Machine Learning Framework for Fall Risk Assessment Using Standardized Geriatric Measures

Abstract

Fall-related injuries, disabilities, imbalance and mortality in elderly adults are serious global public health concerns. Despite the clinical utility of widely used fall risk assessment tools like STRATIFY, the Morse Fall Scale, and the Hendrich II Fall Risk Model, their predictive performance is frequently limited by static scoring mechanisms, limited environmental sensitivity, and decreased compliance across heterogeneous care environments.

The design, validation, and assessment of a Hybrid Fall Risk Assessment and Prediction (HF-RAP) model that combines machine learning-based prediction with standardised clinical scoring are presented in this paper. Five clinically proven and significant predictorssuch as falls in the past, support of ambulatory aids, transfer capacity, Use of potentially inappropriate medicines (PIMs), and mental and emotional factors form the foundation of the suggested evaluation instrument. Over the course of 15 months, information was gathered from 687 older persons in old age home, pain and palliative care homes, hospitals and at homes.
ANOVA, multivariate logistic regression, and chi-square analysis were used to select features.
A variety of supervised ML methods were assessed, such as Random Forest (RF), XGBoost, Logistic Regression(LR), K Nearest Neighbours, Naïve Bayes, LightGBM, and Deep Neural Networks(DNN). With classification accuracy ranging from 86% to 90% and ROC-AUC values ranging from 96% to 98% among models, the HF-RAP model showed significant discriminative capabilities, with ensemble approaches consistently outperforming others.
Optimal risk stratification while balancing sensitivity and specificity was made possible by a theoretically driven threshold (CRS > 8) that was determined using ROC curve analysis. The suggested technique overcomes significant shortcomings of current tools by fusing clinically interpretable scoring with sophisticated predictive analytics. Solution for fall risk assessment in geriatric care that is scalable, comprehensible, and applicable worldwide.