Risk Prediction and Interpretation for Fall Events Using Explainable AI and Large Language Models

Area of Science:

• Artificial Intelligence in Healthcare
• Gerontology and Public Health
• Machine Learning for Predictive Analytics

Background:

• Falls are a major public health issue, especially for older adults, causing millions of injuries annually.
• Existing machine learning models for fall risk prediction often lack accuracy and interpretability, hindering clinical application.
• There is a need for advanced AI tools that provide both precise predictions and clear explanations for effective fall prevention.

Purpose of the Study:

• To develop and validate an explainable machine learning approach for predicting fall risk in older adults.
• To integrate predictive accuracy with interpretable results using SHAP values and large language models.
• To enhance clinical decision-making by generating personalized, natural language reports on fall risk.

Main Methods:

• An integrated machine learning pipeline using an XGBoost classifier was developed to analyze health indicators like age and diagnosis history.
• SHAP (SHapley Additive exPlanations) values were employed to enhance model interpretability and identify key risk factors.
• Large language models and LangChain were utilized to transform complex model outputs into natural language narratives for personalized patient reports.

Main Results:

• The XGBoost model achieved 71% accuracy, 69% precision, 76% recall, and a 0.71 ROC AUC on the test dataset.
• SHAP analysis provided transparent insights into the most critical features contributing to fall risk.
• Automated generation of personalized reports successfully translated risk assessments and feature explanations into comprehensible narratives for healthcare providers and patients.

Conclusions:

• The integrated approach demonstrates the feasibility of combining high-accuracy fall risk prediction with explainable AI and large language models.
• The system's ability to provide interpretable results and generate clear, personalized risk communications is a significant advancement for clinical fall risk assessment.
• This AI-driven tool has the potential to improve the implementation of preventive interventions in healthcare settings, ultimately reducing fall incidents in older adults.