CatBoost machine learning achieved an AUC of 0.836 (95% CI 0.758-0.915) for identifying frailty in AF patients, which was comparable to a simplified logistic regression model (AUC 0.784, P=0.125).
Cross-Sectional (n=501)
Do machine learning algorithms improve the identification of frailty compared to logistic regression in middle-aged and elderly patients with atrial fibrillation?
Machine learning algorithms like CatBoost provide robust identification of frailty in middle-aged and elderly AF patients, although simplified regression frameworks offer comparable accuracy with lower clinical complexity.
Absolute Event Rate: 0.836% vs 0.784%
p-value: p=0.125
Background: Frailty significantly increases the risk of adverse outcomes in patients with atrial fibrillation (AF), particularly among middle-aged and elderly individuals. Despite its clinical importance, there is a lack of efficient, multidimensional identification tools specifically tailored for this population. This study aimed to identify clinical and psychological factors associated with frailty and evaluate machine learning-based identification frameworks for middle-aged and elderly AF patients. Methods: In this cross-sectional study of 501 AF patients, the dataset was randomly partitioned into a training set (80%) and an independent test set (20%). Within the training set, five-fold cross-validation was implemented for hyperparameter tuning and feature selection via LASSO penalized regression (λ 1se). Seven machine learning algorithms were compared against the logistic regression model. SHapley Additive exPlanations (SHAP) analysis was applied to identify key frailty-related factors and provide model interpretability. Performance was explicitly assessed on the independent test set using the Area Under the Curve (AUC), Brier score, and Decision Curve Analysis (DCA). Results: Frailty prevalence was 36.73%. Smoking (OR=3.36), mild cognitive impairment (OR=2.04), valvular heart disease (OR=2.08), and depressive/anxiety symptoms were independently associated with frailty. On the independent test set, CatBoost achieved the highest AUC (0.836, 95% CI: 0.758– 0.915), and Brier score (0.169), with a sensitivity of 84.4% and a specificity of 67.6% at the 0.5 threshold. In comparison, the simplified logistic model demonstrated a sensitivity of 85.9% and a specificity of 35.1% (AUC=0.784, P =0.125). SHAP and DCA validated model interpretability and net clinical benefit. Conclusion: Frailty is prevalent among middle-aged and elderly AF patients and is associated with clinical and psychological determinants. While machine learning algorithms provide robust identification, a simplified regression framework offers comparable accuracy with lower clinical complexity. Given the cross-sectional design, external validation in prospective cohorts is essential before clinical application can be considered. Keywords: frailty, middle-aged and older adults, atrial Fibrillation, prevalence, machine learning
Liao et al. (2026) conducted a cross-sectional in Atrial fibrillation (n=501). Machine learning algorithms (CatBoost) vs. Logistic regression model was evaluated on Area Under the Curve (AUC) for frailty identification (95% CI 0.758-0.915, p=0.125). CatBoost machine learning achieved an AUC of 0.836 (95% CI 0.758-0.915) for identifying frailty in AF patients, which was comparable to a simplified logistic regression model (AUC 0.784, P=0.125).