Comparing interpretable machine learning models for fall risk in middle-aged and older adults with and without pain

Area of Science:

• Gerontology
• Public Health
• Biostatistics

Background:

• Pain is a prevalent issue in middle-aged and older adults.
• Pain is recognized as a risk factor for falls, but the underlying mechanisms remain unclear.
• Falls pose a significant threat to the health and independence of older populations.

Purpose of the Study:

• To develop and validate four-year fall risk prediction models for older adults, separately for those with and without pain.
• To identify key predictors of falls in distinct populations (with and without pain).
• To elucidate the role of pain characteristics in fall risk.

Main Methods:

• Utilized data from 13,074 middle-aged and older adults from the China Health and Retirement Longitudinal Study (2011-2015).
• Employed five machine learning algorithms and 145 candidate features to build prediction models.
• Applied Shapley Additive exPlanations (SHAP) for model interpretability and logistic regression (LR) for analysis.

Main Results:

• Adjusted logistic regression confirmed pain as a significant risk factor for falls (OR 1.40).
• Lower limb pain (OR 1.71), severe pain (OR 1.53), and multisite pain (OR 1.43) showed the highest associated fall risks.
• The LR model demonstrated superior performance (AUC-ROC 0.732 for pain group, 0.692 for non-pain group).
• Key predictors differed: fall history and height were common; pain-specific factors included functional limitation and chronic disease score; non-pain specific factors included age and cognitive function.

Conclusions:

• Pain characteristics are significantly associated with increased fall risk in middle-aged and older adults.
• Machine learning models can effectively identify high-risk individuals, particularly those experiencing pain.
• Distinct predictive features necessitate targeted fall prevention strategies for individuals with and without pain.