Machine Learning Approaches to Identify Physical, Psychological, and Social Predictors of Worsening Frailty Transitions Among Community‐Dwelling Older Adults: A Longitudinal Cohort Study

BACKGROUND: Frailty poses a significant challenge to aging societies, with increasing attention being paid to the concept of frailty transitions. Worsening frailty transitions are associated with higher risks of hospitalization, increased mortality, and substantial healthcare costs. Although frailty is influenced by multiple factors, the specific contributors to worsening transitions remain unclear due to interdependent and nonlinear relationships among physical, psychological, and social factors. Machine learning, which can effectively identify nonlinear patterns in health data, has been underutilized in this context. METHODS: In this longitudinal study, we analyzed data from the Taiwan Longitudinal Study on Aging (TLSA) from 2007 to 2011, including 1670 community-dwelling adults aged 60 years and older. Frailty was assessed at the baseline and follow-up using the frailty phenotype, with modified Fried criteria used to classify participants as nonfrail, prefrail, or frail. Worsening frailty transitions were defined as the outcome variable, including transitions from nonfrailty to prefrailty, nonfrailty to frailty, and prefrailty to frailty. Physical, psychological, and social variables were examined as predictors. Four machine learning algorithms, including logistic regression, random forest, support vector machine, and eXtreme Gradient Boosting, were applied, and model performance was evaluated using the area under the receiver operating characteristic curve (AUROC). RESULTS: During the study period, 30.2% of participants experienced worsening frailty transitions. Among the machine learning models tested, the random forest algorithm yielded the highest predictive performance, with an AUROC of 0.88. The top 10 predictors of worsening frailty included age, marital status, gender, elevated depressive symptoms, educational level, social participation, alcohol consumption, social support, cognitive status, and physical activity. CONCLUSIONS: Worsening frailty transitions result from multifactorial and complex interactions. Machine learning methods can effectively handle multidimensional data and reveal patterns not captured by traditional statistical approaches. These findings enhance our understanding of frailty progression and support the development of targeted interventions for older adults.