Purpose: To estimate the positive predictive value (PPV) of claims for acute pancreatitis among initiators of antihyperglycemic drugs in commercial health insurance claims data.
Methods: As part of a systematic study of the occurrence of acute pancreatitis among antihyperglycemic drug initiators (N=260,255) within a large US health insurer's claims database, we identified potential cases of acute pancreatitis and confirmed them through medical record review. Potential cases had an International Classification of Diseases, 9th revision diagnosis code for acute pancreatitis (577.0) associated with an inpatient or emergency department claim. We sought 860 medical records to confirm potential cases and received 585 (70%), which were reviewed by a clinical adjudication committee. We estimated the PPV and 95% confidence intervals (CI) of claims for these medical records and a subset that had the diagnosis code listed in the first position of an inpatient claim.
Results: The PPV was 0.50 (95% CI 0.44-0.53) for an acute pancreatitis diagnosis code in any position and 0.60 (95% CI 0.55-0.65) if in the first position of an inpatient claim. The estimated PPV varied across strata defined by patient characteristics and was generally lower within strata where potential risk factors for acute pancreatitis were present.
Conclusions: These data indicate that health insurance claims-based identification of acute pancreatitis might overestimate actual cases and introduce appreciable bias, usually toward the null. Further case confirmation or relative risk correction may be necessary to address potential bias.
Copyright © 2010 John Wiley & Sons, Ltd.