Key result
COPD-specific Elastic Net model outperforms CHARGE-AF for predicting incident AF, improving C-index by ~0.05.
Why the study?
Does a COPD-specific risk prediction model incorporating clinical, biochemical, treatment, and genetic predictors improve prediction of incident atrial fibrillation compared with general-population risk scores (CHARGE-AF and HARMS2-AF) in patients with COPD?
Cohort (n=8,385)
Yes
Does a COPD-specific risk prediction model incorporating clinical, biochemical, treatment, and genetic predictors improve prediction of incident atrial fibrillation compared with general-population risk scores (CHARGE-AF and HARMS2-AF) in patients with COPD?
Disease-specific AF risk prediction models integrating clinical, biochemical, respiratory treatment, and polygenic predictors provide superior discrimination and calibration compared to general-population scores (CHARGE-AF and HARMS2-AF) in patients with COPD.
May support COPD-specific AF risk stratification; extends general scores but leaves open outcome impact and requires prospective validation.
Background: 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. Methods: 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 HARMS 2 -AF. Results: 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 HARMS 2 -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. Conclusion: 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.
No takes yet. Share an insight, caveat, or question.
Xiao et al. (2026) conducted a cohort in Chronic obstructive pulmonary disease (COPD) without prior atrial fibrillation (n=8,385). Disease-specific atrial fibrillation risk prediction models (Elastic Net, Cox proportional hazards, Random Survival Forest, DeepSurv) vs. Published general-population AF risk scores (CHARGE-AF and HARMS2-AF) was evaluated on Time to onset of incident atrial fibrillation (and model discrimination via Harrell's C-index) (C-index difference 0.047, 95% CI 0.022-0.073). The COPD-specific Elastic Net model achieved a temporal validation C-index of 0.706 for incident AF, outperforming CHARGE-AF (C-index difference 0.047, 95% CI 0.022-0.073).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: