Interpretable machine learning-based predictive model for fall risk in older adults receiving maintenance...

Area of Science:

• Gerontology
• Nephrology
• Artificial Intelligence in Healthcare

Background:

• Falls are a significant cause of disability in older adults undergoing maintenance hemodialysis (MHD).
• Existing fall risk assessment tools have limited accuracy in this specific patient population.
• There is a need for improved predictive models to identify individuals at high risk of falls.

Purpose of the Study:

• To develop and validate an interpretable machine learning (ML) model for predicting fall risk in older MHD patients.
• To compare the performance of nine different ML algorithms for fall risk prediction.
• To identify key clinical and functional factors contributing to fall risk in this population.

Main Methods:

• A prospective study followed 1,248 older adults receiving MHD for six months.
• A dual-algorithm strategy (LASSO and Boruta) was used for feature selection and multicollinearity reduction.
• Nine ML algorithms were constructed and compared using AUC, calibration plots, and DCA; interpretability was assessed via SHAP analysis.

Main Results:

• 39.0% of participants experienced at least one fall during the follow-up period.
• The Categorical Boosting (CAT) model demonstrated superior performance with an AUC of 0.865.
• SHAP analysis identified frailty, use of walking aids, and older age as the primary predictors of falls.

Conclusions:

• A robust and interpretable CAT-based ML model was developed and validated for predicting falls in older MHD patients.
• The model effectively identifies major clinical and functional risk factors.
• This tool can aid in early risk stratification and personalized fall prevention strategies in clinical settings.