Methods
Population Characteristics
We included patients enrolled in the OMNI and TRENDS clinical trials. In brief, the inclusion criteria for the OMNI trial were the presence of a specific model of Medtronic (Minneapolis, Minnesota) device (InSync Sentry [CRT-D], EnTrust [ICD-VR and DR systems], Instrinsic [ICD-DR], and EnRhythm [IPG-DR]) in patients 18 years of age or older. Inclusion criteria for the TRENDS study were an established Class I/II indication for an implantable cardiac rhythm device capable of long-term trending of atrial tachycardia or AF burden and at least 1 of the following risk factors for stroke: congestive heart failure, hypertension, 65 years of age or older, diabetes mellitus, or prior stroke or transient ischemic attack. In the OMNI trial, single chamber devices and devices that did not have an atrial lead were excluded because of their inability to detect AF. Patients from the TRENDS trial were excluded from this analysis if they had an attempted cardioversion or AF ablation anytime during follow-up, underwent device replacements, already had permanent atrial tachycardia/AF, had known re-entrant supraventricular tachycardia, or had a terminal illness.
From the initial population of the OMNI (n = 737) and TRENDS (n = 598) trials and for the purposes of the present analysis, we excluded 60 patients with AF specific treatments (medical/electrical cardioversion or catheter ablation), 27 patients with single chamber devices, and 7 patients in whom no atrial lead was implanted. The total population (n = 1,195) included patients with at least 180 days of documented rhythm history from the device trending data (Cardiac Compass, Medtronic Inc., Minneapolis, Minnesota) and the analyzed follow up duration was limited to 365 days in order to avoid having progression of AF as a confounding factor.
Clinical AF classification was performed according to AHA guidelines just prior to device implantation. The OMNI and TRENDS trials studied the magnitude of AF on clinical outcomes and collected data on patients' clinical management, and careful attention was paid to the clinical classification of the patients' AF according to the AHA guidelines.
Additionally, we sought to compare the degree of agreement between the clinical AF classifications with a device-derived AF classification on the basis of objective, device-derived criteria. For the device-derived AF classification, we used the following definitions: no AF: no day with >5 min of AF;paroxysmal AF: at least 1 day with >5 min of AF but <7 consecutive days with >23 h of AF; persistent AF: at least 7 consecutive days with >23 h of AF;permanent AF: All days with >23 h of AF (or >95% AF burden). Although these device-based definitions may seem somewhat arbitrary, they were designed to align with published guidelines and have been used in several AF trials. Device-derived definitions have the advantage of being consistent and reproducible, and are based on objective temporal AF indices.
AF burden was defined as the proportion of the monitored time that a patient was in AF. AF density, as described previously, characterized the temporal aggregation of the AF burden. In short, AF density is a quantitative measure of the temporal aggregation of AF burden and was calculated as an index consisting of values between 0 (AF burden evenly spread over the observation time) and 1 (maximal possible AF burden aggregation; i.e., "one continuous episode of AF"). A thorough presentation of the AF density has been reported previously. The AF detection algorithms utilized in the study devices have been evaluated extensively and have been shown to quantify AF burden with 99% accuracy.
Statistical Analyses
Simple statistical tests (such as the t test, chi-square test, Mann-Whitney U test, analysis of variance, and Kruskal-Wallis tests) were employed where appropriate to identify differences in the demographics of the patient population subgroups. The agreement between clinical and device AF classifications was evaluated using Cohen's kappa. Logistic and multinomial logistic regression was used to investigate the influence of patient demographics on the AF classification. The temporal persistence of AF as measured by the AF burden was significantly associated with the clinical AF classification and was included in subsequent models investigating the additional effect of the following variables on the clinical AF classification: age, sex, presence of coronary artery disease, presence of cardiomyopathy, functional status (New York Heart Association functional class), history of ablation for AF, history of heart surgery, AF density, and left ventricular ejection fraction (LVEF). Receiver-operating characteristic analyses were used to evaluate the performance of AF burden as a discriminator of the clinical AF classification. The p values of 2-sided tests at a significance level of 0.05 are reported.
All statistical analyses were performed with R version 3.0.1 (R Development Core Team 2013, Vienna, Austria).