Comparing classifications of death in the Mode Selection Trial: Agreement and disagreement among site investigators and a clinical events committee

Clinical events committees (CECs) are the current standard for endpoint adjudication in clinical trials. However, little data exist with which to compare CEC and site investigator determinations or to evaluate internal agreement among CEC members. Using data from the Mode Selection Trial in Sinus Node Dysfunction (MOST), we analyzed classifications of death in order to compare internal agreement among CEC physician reviewers and agreement between the CEC and site investigators. Death was classified at 2 levels: by major cause (cardiac, noncardiac, or unknown) and by minor subclassification of the major classifications. Reviewer agreement was tabulated at the major and minor levels, and standard and weighted K statistics were calculated. Disagreement at both levels was also determined. Individual decision-making was tabulated in terms of frequency in classifying death as unknown. All 404 deaths were classified by the CEC. Site investigators determined major classifications in 382 cases and minor classification in 379 cases. The CEC and the site investigators disagreed in classifying 41 cases (10.7%) at the major level and 117 (30.9%) at the minor level. CEC reviewers disagreed internally at the major level in 64 cases (15.8%), at the minor level in 63 cases (15.6%), and at any level in 127 cases (31.4%) (kappa = 0.60, 95% confidence interval (CI) [0.55, 0.66]; weighted kappa = 0.66, 95% CI [0.62, 0.75]). In resolving internal disagreements, the full CEC agreed with 1 of 2 CEC reviewers in 85.9% of cases. Disagreements occurred between site investigators and CEC reviewers in classifying deaths. Endpoint determination and decision-making varied among individual CEC reviewers, but second-tier reviews by the full CEC resolved all disagreements. These findings support continued use of CECs for endpoint adjudication in clinical trials. (c) 2006 Elsevier Inc. All rights reserved.