Development of Fall Risk Classification Models for Community-Dwelling Older Adults using Latent Class Analysis and Machine Learning.

Abstract

Introduction: The aim of this study was to identify fall-risk groups among community-dwelling older adults in South Korea and build a classification model to investigate risk-associated factors.

Methods: This cross-sectional study analyzed data of 9,231 older adults from the 2020 Korea Elderly Survey. We used latent class analysis to identify fall-risk groups based on fall indicators. Thereafter, classification models were developed with these identified groups as outcome variables.

Results: Latent class analysis results indicated that a three-class model was more interpretable and fit the data better than other models. Among the models, the XGBoost algorithm displayed superior performance (accuracy = 0.70, precision = 0.69, recall = 0.70, F1-score = 0.68). Key variables associated with fall-risk groups included self-rated health, cognitive function, recent healthcare use, and assistance needed in instrumental activities of daily living.

Conclusion: The study adopted a preventive approach by differentiating among low-, moderate-, and high-fall-risk groups, thus providing valuable insights for healthcare professionals. Identifying these risk factors can support the development of customized fall prevention programs for older adults.