Objectives: To measure agreement between formal and medical record criteria for the diagnosis of acute coronary syndrome (ACS) among patients undergoing an emergency department evaluation for potential acute coronary symptoms.
Methods: Cases of ACS were determined by both formal (World Health Organization 1984 criteria for acute myocardial infarction [AMI], Braunwald criteria for unstable angina pectoris [UAP]) and medical record criteria. In the latter, a diagnosis was made if providers indicated AMI or UAP anywhere in the medical record. All information included in formal criteria was available to clinicians establishing the medical record diagnosis. The two criteria for diagnosis were compared, and a kappa value was recorded. Two blinded observers adjudicated discordant cases, with a kappa value recorded. Disagreements between these two coinvestigators were resolved by a Delphi technique.
Results: A total of 375 eligible subjects were enrolled, of whom 65 (17%; 45 AMI, 20 UAP) had ACS by both sets of criteria. Formal and medical record criteria disagreed in 32 subjects. This represented 9% (95% confidence interval = 6% to 12%) of the overall study population but 33% (95% confidence interval = 23% to 43%) of subjects with possible ACS. Coinvestigators acting as judges and blinded to each other's determinations agreed that 25 of these subjects had ACS and three did not; they disagreed on four subjects (kappa = 0.54). Among these four subjects, a Delphi consensus technique determined that two subjects had AMI and two had no ACS.
Conclusions: In a single-site study, among subjects who have possible ACS as determined by either or both formal and medical record criteria, these two sets of criteria disagree in almost one third of cases. Among discordant cases, even two expert judges frequently disagreed on the final diagnosis. A modified Delphi technique to address these disagreements is described.