Methods: In order to identify angina patients, six algorithms comprised of multiple ICD-9 diagnostic codes and pharmacy claims were developed based on American Heart Association treatment guidelines for angina and physician input. Subjects were identified in an administrative claims database using the algorithms. Medical records abstraction for the subjects identified in the database was conducted to confirm angina diagnosis. Positive predictive value of each algorithm as well as for grouped algorithms was calculated.
Results: The positive predictive value of each of six algorithms was 0.95 to 0.25. Having an ICD-9 code for angina did not guarantee a high predictive value. Of the two algorithms with the highest predictive values (0.95 and 0.92) the first used multiple angina ICD-9 codes and multiple nitrate codes separated by explicit time periods. However, the second predictive algorithm used multiple coronary artery disease codes combined with multiple chest pain codes in addition to nitrate prescriptions to identify patients. Although nitrate prescriptions were a component of the two most predictive algorithms, the presence of nitrates alone did not predict angina and had the lowest predictive value (0.25).
Conclusions: A typical patient identification method for claims data, using two angina ICD-9 codes and two pharmacy codes separated by time, was predictive only if the pharmacy codes were for nitrates and identified only a small portion of the actual angina patients in the database. In order to capture a greater number of patients with angina, a variety of algorithms need to be used including combinations of various drug (e.g., nitrates, beta blockers, calcium channel blockers) and diagnostic codes (e.g., angina, chest pain, coronary artery disease).