Background: Using claims data to monitor for potential teratogenic drug effects is an appealing concept, but the validity of automated case definitions for many major congenital malformations (MCMs) has not been established in commercial claims.
Objectives: To estimate the positive predictive value (PPV) of algorithms to identify 10 of the most common MCMs from claims data, using medical record review as a gold standard.
Methods: This validation exercise was part of a retrospective cohort study evaluating birth defects. Using administrative data from two of the largest commercial health insurers in the US, we identified mothers aged 15–49 who delivered a live infant from January 2004 to November 2010. We developed algorithms to identify 10 of the most common MCMs in the US based on diagnosis and procedure codes indicative of malformations and their expected process of care. Case definitions included only non-syndromic MCMs that were not caused by a genetic defect. The analysis required a minimum of 20 randomly sampled cases of each MCM. In total, 349 infants had their medical records retrieved and adjudicated by clinical geneticists. Analyses used PPVs to characterize the algorithms’ validity.
Results: Across all defects, the PPV for specific MCMs was 67.9% (95% CI: 62.5–73.1). The overall PPV for ascertaining any of the MCMs of interest as a composite endpoint was 76.9% (95% CI: 71.8–81.5). PPVs for many specific MCMs were reasonably high (pyloric stenosis: 93.3%, ventricular septal defect: 90.9%, oral cleft: 89.3%, hypospadias: 80.0%, congenital hip dislocation: 75.9%). However, hydrocephalus (PPV = 47.4%) and several cardiac defects (conotruncal heart defects: 68.0%, aortic coarctation: 60.0%, pulmonary valve atresia: 44.4%, atrial septal defect: 37.9%) performed less well.
Conclusions: Many claims-based diagnoses of MCMs, especially cardiac anomalies, were not confirmed in clinical records. Other defects were more accurately identified, especially where a specific pattern of care such as corrective surgery was expected. To minimize outcome misclassification in birth defects research, certain claims-identified MCMs require additional confirmation.
