Background: Pragmatic clinical trial designs have proposed the use of claims data to ascertain clinical events. However, little is known about the accuracy of medical claims diagnosis in identifying events.
Methods: Administrative claims were obtained for all rehospitalizations occurring within 12 months of index acute myocardial infarction (MI) in those enrolled in the TRANSLATE-ACS study. Outcome events (recurrent myocardial infarction [MI], stroke, and hospitalized bleeding) were identified based on all diagnosis and procedure codes in these claims. These clinical events were also independently adjudicated by physicians based on medical record review. The cumulative 1-year incidences for each outcome were compared when identified by administrative claims vs. physician adjudication. Kappa statistics assessed the accuracy of claims-identified events using physician adjudication as the gold standard.
Results: Among 12,351 MI patients, the cumulative 1-year incidence of events identified by administrative claims were 4.3% for MI, 0.9% for stroke, and 5.0% for bleeding. Incidence rates based on physician adjudication of medical records were 4.7% for MI, 0.9% for stroke, and 5.4% for bleeding. Agreement between claims-identified and physician-adjudicated events was modest for MI and stroke events: kappa=0.76 and 0.55, respectively. In contrast, agreement was poor between claims-identified and physician-adjudicated bleeding events: kappa=0.24 for any hospitalized bleeding event and 0.15 for GUSTO moderate-severe bleeding.
Conclusion: Event rates at 1 year were lower for MI, stroke and bleeding when using administrative claims to identify events in comparison with physician adjudication. Administrative claims diagnoses are modestly accurate in identifying recurrent MI and stroke admissions but have limited accuracy in identifying bleeding events, which suggests that an alternative approach is needed to ensure good safety surveillance in cardiovascular studies.