Predictors for atrial fibrillation detection after cryptogenic stroke: Results from CRYSTAL AF

Abstract

Objective: We assessed predictors of atrial fibrillation (AF) in cryptogenic stroke (CS) or transient ischemic attack (TIA) patients who received an insertable cardiac monitor (ICM).

Methods: We studied patients with CS/TIA who were randomized to ICM within the CRYSTAL AF study. We assessed whether age, sex, race, body mass index, type and severity of index ischemic event, CHADS2 score, PR interval, and presence of diabetes, hypertension, congestive heart failure, or patent foramen ovale and premature atrial contractions predicted AF development within the initial 12 and 36 months of follow-up using Cox proportional hazards models.

Results: Among 221 patients randomized to ICM (age 61.6 ± 11.4 years, 64% male), AF episodes were detected in 29 patients within 12 months and 42 patients at 36 months. Significant univariate predictors of AF at 12 months included age (hazard ratio [HR] per decade 2.0 [95% confidence interval 1.4-2.8], p = 0.002), CHADS2 score (HR 1.9 per one point [1.3-2.8], p = 0.008), PR interval (HR 1.3 per 10 milliseconds [1.2-1.4], p < 0.0001), premature atrial contractions (HR 3.9 for >123 vs 0 [1.3-12.0], p = 0.009 across quartiles), and diabetes (HR 2.3 [1.0-5.2], p < 0.05). In multivariate analysis, age (HR per decade 1.9 [1.3-2.8], p = 0.0009) and PR interval (HR 1.3 [1.2-1.4], p < 0.0001) remained significant and together yielded an area under the receiver operating characteristic curve of 0.78 (0.70-0.85). The same predictors were found at 36 months.

Conclusion: Increasing age and a prolonged PR interval at enrollment were independently associated with an increased AF incidence in CS patients. However, they offered only moderate predictive ability in determining which CS patients had AF detected by the ICM.

Figure 1. Model predictions of atrial fibrillation (AF) detection rate

Model predictions of AF detection within 12 months (A) and 36 months (B) of insertable cardiac monitor monitoring by age, PR interval, and use of medications that can prolong the PR interval.

Figure 2. Observed vs predicted atrial fibrillation (AF) detection rate

Predicted vs observed AF detection rates are shown for 6 subgroups (sextiles) of insertable cardiac monitor arm patients based on Cox model prediction of AF detection rates. The predicted AF detection rate at the end of the follow-up period based on the mean Cox linear predictor for each subgroup is plotted (x-axis) against the observed rate determined by the Kaplan-Meier estimate (y-axis). The predicted rate for the minimum and maximum Cox linear predictor among patients in the subgroup are also plotted to describe the range of predictions within the subgroup. The predicted vs observed calibration curves are shown for AF detection with 12 months (A) and within 36 months (B).