An explainable online frailty prediction model for community-dwelling older adults based on machine learning algorithms: a cross-sectional study based on retrospective health data

Clinical Snapshot

Key Findings

• P Value: Not specified for model performance

• Effect Size: CatBoost model AUROC 0.831 in testing set

• Primary Outcome: Frailty prediction accuracy using machine learning models

• Nnt Or Sensitivity: 22.3% frailty prevalence in study population

• Confidence Interval: Not reported

Clinical Application

Web-based calculator enhances practical implementation in clinical practice Requires validation in Australian population; could complement RACGP guidelines for comprehensive geriatric assessment and integrate with existing aged care pathways Community-dwelling older adults in primary care settings

Abstract

BACKGROUND: Frailty is a significant health concern associated with diminished physiological reserves and increased healthcare burdens. Currently, effective models for predicting frailty risk are lacking. This study utilizes machine learning-based models to early identify community-dwelling older adults, enhancing risk assessment accuracy and guiding targeted interventions to slow frailty progression. METHODS: A cross-sectional analysis of data from 1,156 older adults across 31 community health centers in Nanjing, conducted between January and October 2024, was performed. Independent predictors of frailty were identified using univariate analysis and the least absolute shrinkage and selection operator. The dataset was divided into 70% training and 30% testing subsets. Six machine learning (ML) models were developed and their performances compared. The SHapley Additive exPlanations (SHAP) method was applied to interpret the models, and a web-based risk calculator was created. RESULTS: Our dataset showed that 22.3% of older adults were frail. Significant predictors of frailty were identified as age, education, medicine, vegetable, cognitive status, number of diseases, hemoglobin, total cholesterol, and neutrophil-to-lymphocyte ratio. Among the six ML models, Categorical Boosting (CatBoost) exhibited the highest performance, attaining an AUROC of 0.886 in the training set and 0.831 in the testing set. CONCLUSIONS: The developed CatBoost model and web calculator can be employed by general practitioners to proactively identify high-risk community-dwelling older adults, thereby enabling timely interventions to mitigate the progression of frailty. The tool's simplicity and replicability effectively facilitate the promotion and management of frailty prevention within the community. An effective model was developed to predict frailty risk in community-dwelling older adults.The integration of biomarkers and self-reported data enhances frailty prediction and optimizes risk-based intervention strategies.The CatBoost model demonstrates superior predictive performance compared to other models.The web-based tool integrates a frailty prediction model for proactive risk management.

References

1. 1.Zhao, S., Zhang, S., Wang, J., Huang, T., Wu, H., Yao, W., Huang, K., Zou, J., & Shen, Y. (2026). An explainable online frailty prediction model for community-dwelling older adults based on machine learning algorithms: a cross-sectional study based on retrospective health data. International Journal of Medical Informatics, 106014. https://doi.org/10.1016/j.ijmedinf.2025.106014