(180) Validation of claims-based algorithms for identifying neonatal urinary tract, genital, gastrointestinal, and musculoskeletal malformations in healthcare utilization databases

Background: Congenital malformations are one of the most commonly evaluated neonatal outcomes in studies on medication safety in pregnancy. Healthcare utilization databases are often used to generate evidence on the risks of malformations following medication use during pregnancy, but the accuracy of algorithms to identify malformation diagnoses in claims data, outside of cardiac defects and oral clefts, is largely unknown.

Objectives: To develop and validate claims-based algorithms for urinary tract, genital, gastrointestinal, and musculoskeletal malformations among neonates using data from a large hospital network in the U.S.

Methods: Using the Mass General Brigham (MGB) Research Patient Data Registry, we identified infants born 2007-2020 with an encounter at two major MGB hospitals in the year of birth. Specific malformations were identified by relevant ICD-9/ICD-10 diagnosis and ICD/CPT procedure codes within 90 days of birth in the data. Specifically, we required (i) ≥2 visits on different dates with a diagnosis code, (ii) ≥1 visit with a diagnosis and procedure code, or (iii) ≥1 visit with a diagnosis code and infant death. For each specific malformation type, 100 cases (i.e., 50 ICD-9 based and 50 ICD-10 based) were randomly sampled for medical chart reviews by pairs of physicians to adjudicate the outcome. Any inconsistencies were resolved by a third, independent reviewer. The positive predictive values (PPV) and 95% confidence intervals (CI) were estimated separately for the ICD-9 and ICD-10 based algorithms.

Results: Out of 52,934 eligible infants in the cohort, there were 253 and 245 urinary tract, 174 and 281 genital, 134 and 101 gastrointestinal, and 103 and 141 musculoskeletal malformation cases identified by the ICD-9 and ICD-10 algorithms, from which 50 cases each were randomly sampled. The PPVs (95% CI) of the ICD-9 and ICD-10 based algorithms, respectively, were 100% (92.9%-100%) and 100% (92.7%-100%) for urinary tract malformations; both 98.0% (89.1%-100%) for genital malformations; 84.0% (70.9%-92.8%) and 100% (92.9%-100%) for gastrointestinal malformations; and 89.8% (77.8%-96.7%) and 90.0% (78.2%-96.6%) for musculoskeletal malformations. Approximately 70% of the false positives were attributed to suspected malformations that were later ruled out.

Conclusions: The PPVs were high for both the ICD-9 (84% to 100%) and ICD-10 (90% to 100%) algorithms for the four malformations. For gastrointestinal malformations, the performance of the ICD-10 algorithm was better than that of the ICD-9 algorithm. Our findings suggest that claims-based algorithms can be utilized to identify these specific congenital malformations among neonates in healthcare utilization databases.