Director Janssen Pharmaceuticals, Inc. Titusville, United States
Background: Misclassification is a serious threat to study validity. This threat would be exacerbated if the misclassification differed between exposures in pharmacoepidemiologic studies. The assumption is usually made that misclassification is non-differential between exposures and any bias due to misclassification would move the effect estimate toward the null.
Objectives: To evaluate the performance characteristics, i.e., sensitivity, specificity, and positive (PPV) and negative predictive value (NPV), between different drug exposures for the outcomes of acute myocardial infarction and angioedema.
Methods: This retrospective cohort study included 4 claims databases, IBM MarketScan® Commercial (CCAE), IBM MarketScan® Medicare Supplemental (MDCR), IBM MarketScan® Multi-State Medicaid (MDCD), and Optum's ClinformaticsTM Data Mart (Optum SES) and 1 electronic health record system, Optum'sTM longitudinal Electronic Health Records (Optum EHR). We examined the performance characteristics for acute myocardial infarction (AMI) within 180 days following an initial exposure to either an angiotensin converting enzyme inhibitor (ACEi), an angiotensin receptor blocker (ARB), or a calcium channel blocker (CCB). We also examined the performance characteristics for angioedema within 60 days following an initial exposure to either ACEi or ARB. Performance characteristics were estimated using the Observational Health Data Sciences and Informatics (OHDSI) open-source software package PheValuator. PheValuator uses a machine learning approach to estimate performance characteristics of phenotype algorithms and has shown good agreement with results from prior traditional validation studies. Data was examined from 2010 to 2021.
Results: For AMI, the number of subjects examined ranged from 13,308 for ARB in MDCD to 563,565 for ACEi in Optum EHR. Sensitivity was lower in ARB compared to ACEi by a median value of 10.5 (Interquartile Range (IQR) 9.0) across the 5 databases. Sensitivity was lower in CCB compared to ACEi by a median value of 23.6% (IQR 14.5) Differences for specificity, PPV, and NPV were lower than 5% for both comparisons. For angioedema, the number of subjects ranged from 88,338 for ARB in MDCR to 1,903,651 for ACEi in Optum EHR. Sensitivity was lower in ACEi compared to ARB by a median value of 21.6 (IQR 36.5) across the 5 databases. Differences for specificity, PPV, and NPV were lower than 5% for the comparison.
Conclusions: For AMI and angioedema, we found that misclassification was differential between the drugs studied. With the capability of PheValuator to provide misclassification estimates between exposures, this potential bias should be examined as part of a study validation package.
