Development and Validation of a Risk Prediction Model for Atrial Fibrillation in Patients with Chronic Obstructive Pulmonary Disease

Chronic obstructive pulmonary disease (COPD) is associated with an increased risk of incident atrial fibrillation (AF). Existing AF prediction scores were developed primarily in general populations, and their transportability to patients with COPD remains insufficiently established. This study aimed to develop and temporally validate a novel, disease-specific AF prediction model for patients with COPD.

A total of 8385 COPD patients without prior AF from the UK Biobank (UKB) were included. Participants recruited in 2006-2009 formed the development set and those recruited in 2010 formed a temporal validation set. Cox proportional hazards, Elastic Net, random survival forest, and DeepSurv models were developed using clinical, biochemical, treatment-related, and genetic predictors. Performance was assessed using C-index, time-dependent AUC, Brier score, calibration, and decision-curve analysis, with 95% confidence intervals from 1000 participant-level bootstrap samples. Models were compared with CHARGE-AF and HARMS₂-AF.

During a median follow-up of 14.17 years, 1402 participants developed AF. Internal validation C-indices were 0.748 for Cox, 0.750 for Elastic Net, 0.738 for random survival forest, and 0.745 for DeepSurv; temporal validation values were 0.705, 0.706, 0.696, and 0.712, respectively. In temporal validation, Elastic Net showed higher discrimination than CHARGE-AF (C-index difference, 0.047; 95% CI, 0.022-0.073) and HARMS₂-AF (0.061; 95% CI, 0.032-0.090). Decision-curve analysis suggested potential clinical utility across selected 5-year thresholds. The Elastic Net model was implemented in an interactive web application.

AF risk prediction in COPD achieved moderate discrimination across several modeling approaches. The developed models showed higher discrimination than published AF risk scores in several analyses, particularly in temporal validation, supporting further evaluation of individualized AF risk stratification in patients with COPD.