Interpretable machine learning models for analyzing predictors of first-aid knowledge among older adults in Chinese communities: a cross-sectional survey study

Abstract

This study aimed to investigate the current status of first aid knowledge and its associated factors among community-dwelling older adults using machine learning (ML) models. This cross-sectional study recruited community-dwelling older adults aged 60 years and above. Face-to-face questionnaire surveys were conducted to collect basic demographic information, first aid knowledge, and eHealth literacy data. Seven ML algorithms (logistic regression, decision tree, random forest, XGBoost, LightGBM, support vector machine, and artificial neural network) were employed to construct prediction models. SHAP (SHapley Additive exPlanations) analysis was performed on the best performing ML model to interpret key predictors. A total of 318 older adults were included, with a mean age of 67.2 (SD 5.8) years. The mean first aid knowledge score was 8.35 (SD 2.84), and the mean eHealth literacy score was 26.71 (SD 7.29). Based on feature importance ranking and AUC curve trends, eight features were selected to construct the final model. Among the seven ML models, logistic regression performed best, with an AUC of 0.705 (95% CI: 0.568–0.825). SHAP analysis revealed that occupation before retirement, education level, living status, eHealth literacy, and gender were the five most important predictors. The findings suggest that community health promotion programs and policymakers should prioritize older adults with low education levels, agricultural backgrounds, those living alone, and male older adults when developing strategies to improve first aid knowledge. By understanding their eHealth literacy levels and usage habits, more concise and comprehensible first aid information should be provided to enhance the accessibility and effectiveness of first aid education.