A comparison of the CHARGE-AF and the CHA2DS2-VASc risk scores for prediction of atrial fibrillation in the Framingham Heart Study

Abstract

Background: Atrial fibrillation (AF) affects more than 33 million individuals worldwide and increases risks of stroke, heart failure, and death. The CHARGE-AF risk score was developed to predict incident AF in three American cohorts and it was validated in two European cohorts. The CHA2DS2-VASc risk score was derived to predict risk of stroke, peripheral embolism, and pulmonary embolism in individuals with AF, but it has been increasingly used for AF risk prediction. We compared CHARGE-AF risk score versus CHA2DS2-VASc risk score for incident AF risk in a community-based cohort.

Methods and results: We studied Framingham Heart Study participants aged 46 to 94 years without prevalent AF and with complete covariates. We predicted AF risk using Fine-Gray proportional sub-distribution hazards regression. We used the Wald χ(2) statistic for model fit, C-statistic for discrimination, and Hosmer-Lemeshow (HL) χ(2) statistic for calibration. We included 9722 observations (mean age 63.9 ± 10.6 years, 56% women) from 4548 unique individuals: 752 (16.5%) developed incident AF and 793 (17.4%) died. The mean CHARGE-AF score was 12.0 ± 1.2 and the sub-distribution hazard ratio (sHR) for AF per unit increment was 2.15 (95% CI, 99-131%; P < .0001). The mean CHA2DS2-VASc score was 2.0 ± 1.5 and the sHR for AF per unit increment was 1.43 (95% CI, 37%-51%; P < .0001). The CHARGE-AF model had better fit than CHA2DS2-VASc (Wald χ(2) = 403 vs 209, both with 1 df), improved discrimination (C-statistic = 0.75, 95% CI, 0.73-0.76 vs C-statistic = 0.71, 95% CI, 0.69-0.73), and better calibration (HL χ(2) = 5.6, P = .69 vs HL χ(2) = 28.5, P < .0001).

Conclusion: The CHARGE-AF risk score performed better than the CHA2DS2-VASc risk score at predicting AF in a community-based cohort.

Copyright © 2016 Elsevier Inc. All rights reserved.

Figures

Figure 1

Study design. Flowchart of inclusion and follow-up of participants from the Original (top) and Offspring (middle) cohorts.

Figure 2

Correspondence between CHARGE-AF risk score and CHA2DS2-VASc risk score groups. Mean (diamond) and median (horizontal line) CHARGE-AF risk score for each group of the CHA2DS2-VASc, including 25% to 75% percentiles (box) and range (whiskers).

Figure 3

Calibration curves for the CHARGE-AF and the CHA2DS2-VASc risk scores. We calculated the predicted cumulative incidence and the empirically estimated (the observed) incidence within each subgroup defined by either score. A, Comparison of the predicted values (red for CHA2DS2-VASc, blue for CHARGE-AF) and the observed cumulative incidence. The black diagonal is the reference. A shorter distance from the dots to the reference line indicates better calibration. B, The tables show the groups used for the Hosmer-Lemeshow calibration, including number of incident AF cases and total number, for each risk score. The lower rows show number of follow-up windows.