A clinical model to predict incident AF achieved an AUROC of 84.1%, which improved to 86.5% (95% CI 84.5-88.5%; p<0.001) with NT-proBNP, and was cost-effective compared to no screening.
Cohort (n=23,830)
Yes
Does universal or model-guided atrial fibrillation screening improve quality-adjusted life years and reduce costs compared to no screening in people with no prior atrial fibrillation?
A clinical prediction model can effectively guide atrial fibrillation screening, and both universal and model-guided screening strategies are cost-effective compared to no screening.
Effect estimate: AUROC (95% CI 84.5-88.5)
Absolute Event Rate: 86.5% vs 84.1%
p-value: p=<0.001
Abstract Background and aims Atrial fibrillation (AF) causes ischaemic stroke but population AF screening is not recommended. We developed and externally validated a model to guide AF screening and assessed health economic impact. Methods We analysed clinical demographics, NT-proBNP concentrations and artificial intelligence analysis of ECG data from people with no prior AF in the Generation Scotland study, with external validation in UK Biobank. Record linkage identified people with AF during follow-up. A clinical model for incident AF was developed using penalised logistic regression with Lasso modelling (80:20 training/testing). We assessed incorporating NT-proBNP or ECG data by De Long’s test. A decision-analytic model assessed cost-effectiveness of screening everyone or using the clinical model to guide AF screening, compared with no screening. Results We analysed 23,830 people: 14,061 (59%) female and mean (SD) age 47.0 (15.3) years. AF was diagnosed in 901 people (3.8%). The clinical model included age (OR 1.20, 1.13-1.28), age2 (OR 0.99, 0.99-1.00), weight (OR 1.03, 1.02-1.03), ischaemic heart disease (OR 1.63, 1.31-2.02) and heart failure (OR 3.78, 2.03-7.03). AUROC was 84.1% (82.1-86.1%) in test data and 75.4% (75.1-75.6%) in external validation. Model performance improved with logeNT-proBNP (86.5%, 84.5-88.5%; p0.001) but not ECG data (84.5%, 82.2-86.5%; p=0.76). The quality-adjusted life years (QALYs) gained and cost reduction per-person was 0.05 years and £656 for screening everyone; and 0.04 and £548 for the clinical model to guide screening. Conclusions A clinical model could guide AF screening to prevent stroke and can be augmented by NT-proBNP. Screening everyone and using a clinical model is cost-effective. Conflict of interest
Cameron et al. (Fri,) conducted a cohort in Atrial fibrillation (n=23,830). Clinical model augmented with NT-proBNP vs. Clinical model alone was evaluated on Incident AF prediction (AUROC) (AUROC, 95% CI 84.5-88.5, p=<0.001). A clinical model to predict incident AF achieved an AUROC of 84.1%, which improved to 86.5% (95% CI 84.5-88.5%; p<0.001) with NT-proBNP, and was cost-effective compared to no screening.