(B-259) Development of an Infant Outcome Tree of Congenital Anomalies across ICD Eras

Background: Tree-based scanning allows simultaneous assessment of multiple outcomes, individually or within clusters. This data mining method is useful for teratogenicity studies to evaluate specific congenital anomalies across all organ systems. Due to differences in coding vocabularies between the ICD-9-CM and ICD-10-CM era, there is no tree available for scanning across both eras.

Objectives: To develop an outcomes tree for congenital anomalies across ICD coding eras.

Methods: We selected all ICD-9-CM codes within (congenital anomalies subsection of the Multilevel Clinical Classification Software, MLCCS; 444 codes) and all ICD-10-CM codes under the equivalent subsection in the Beta Classifications Software (839 codes). We mapped all ICD-9-CM codes to ICD-10-CM codes and vice versa using the general equivalence mappings. We relied on the hierarchical structure in the existing MLCCS and the mapped pairs to construct an initial multilevel tree that included both code types. To focus on alerts of interest and maximize power when applying tree-based scanning within the context of drug safety, we removed codes for minor malformations, chromosomal anomalies, and single-gene conditions.
To assess consistency of mapping across ICD eras, we identified infants in MarketScan® Research Databases and National Medicaid Claim data born in 2011-2013 and 2016-2018 and identified malformations in the year after birth. We calculated prevalence and percentage change across eras at different levels of specificity. A two-tailed Chi-square test and standardized mean difference were used to assess prevalence balance between eras.

Results: Our tree included 1,053 codes at the most granular level of MLCCS, which were subsequently aggregated into 252, 98, and 45 groups at incrementally higher levels. Our birth cohorts included 572,107 infants born in 2011-2013 and 360,167 infants born in 2016-2018 from Marketscan and 362,820 and 3,500,589 in Medicaid, respectively. The most common anomaly clusters were atrial septal defect (27 per 1,000), ventricular septal defect (8 per 1,000), and congenital anomalies of urinary system (8 per 1,000). Prevalences were comparable between ICD-9 and ICD-10 cohorts, with standardized mean differences below 0.1. A small proportion of clusters showed over 50% change in prevalence, likely due to inherent differences in the coding specificity between eras.

Conclusions: Overall, there was good agreement between ICD eras at various aggregation levels of the tree. Differences in coding conventions, especially code specificity in either era, resulted in some differences in congenital anomaly capture. These inconsistencies should be considered to prevent measurement bias.