Authors
Mierzhati Maimaiti, Alapate Abuduguli, Shenhong Liu, Xianglin Du, Wen Bai
Published in
Medicine. Volume 105. Issue 44. Pages e51381. Oct 30, 2026.
Abstract
This study aimed to investigate clinical correlates of persistent atrial fibrillation (AF) in older adults, compare seven classification models, and interpret the selected gradient boosting machine (GBM) classifier. This single-center observational study included 572 adults reported to be aged ≥ 60 years who received care between June 2021 and June 2023, comprising 385 participants classified as having persistent AF and 187 comparators. Univariable and multivariable logistic regression and least absolute shrinkage and selection operator (LASSO) were used for candidate-variable assessment and feature reduction. Logistic regression and six machine-learning models-support vector machine, GBM, neural network, extreme gradient boosting (XGBoost), AdaBoost, and CatBoost-were compared using a reported 70:30 development/test split and tenfold cross-validation during model development. Model performance and interpretation were examined using the area under the receiver operating characteristic curve (AUC), decision-curve analysis (DCA), and SHapley additive explanations (SHAP). The sequence of feature selection and data partitioning could not be verified. Fourteen candidate variables were significantly associated with persistent AF in univariable analysis, of which 9 remained significant after multivariable adjustment. LASSO retained 5 features: glycated hemoglobin, C2HEST score, heart failure, alanine aminotransferase (ALT), and CO2. GBM achieved a development-set AUC of 0.970 (95% CI 0.957-0.983). In the test set, GBM, CatBoost, and XGBoost achieved AUCs of 0.919 (95% CI 0.874-0.963), 0.920 (95% CI 0.876-0.963), and 0.919 (95% CI 0.876-0.962), respectively. DCA suggested favorable net benefit for GBM across selected threshold probabilities. SHAP analysis identified ALT as the feature with the greatest contribution to GBM predictions, followed by the C2HEST score. GBM achieved the highest development-set AUC, whereas test-set AUCs were similar across GBM, CatBoost, and XGBoost. The five-feature classifier warrants further investigation for distinguishing persistent AF status in older adults. Verification of the analytical workflow, external validation, and calibration assessment are required before clinical application.
PMID:
43012790
Bibliographic data and abstract were imported from PubMed on 11 Nov 2026.
Read full publication at:
Please sign in
to see all details.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 3
- Comments 0