Evaluating variation between states in algorithms used for identifying abortions in Medicaid claims data

To evaluate the variability in abortion identification across four published algorithms using Medicaid claims data in states where abortion is covered by Medicaid for all indications.

We analyzed 2020 Medicaid Transformed Medicaid Statistical Information System Analytic Files from 14 states with Medicaid abortion coverage. Female recipients aged 15 to 44 were included. Four previously published algorithms, each using different combinations of diagnosis, procedure (Current Procedural Terminology/Healthcare Common Procedure Coding System), and medication (National Drug Classification) codes, were applied to identify abortion-related claims. For each algorithm, we calculated the number of identified abortions by state and examined variability in identification patterns and code types.

Among 9.67 million Medicaid enrollees, the number of identified abortions varied substantially by algorithm and state. The max-to-min ratio across algorithms was lowest in Hawaii (2.09) and highest in New Jersey (138.59). Algorithms differed in their use of diagnosis-only, procedure-only, or both code types, with the proportion of claims containing both codes ranging from 0.4% (New Jersey) to 86.2% (Vermont). Abortions identified solely by Healthcare Common Procedure Coding System codes for mifepristone or misoprostol varied from 0.1% to 32.6% by state. No algorithm consistently performed well across all states.

Substantial heterogeneity exists in the performance of abortion-identifying algorithms across states. These differences likely reflect variation in billing practices, Medicaid data reporting, and algorithm construction.

Researchers should exercise caution when using claims data to estimate abortion rates, particularly across multiple states. Validation of an algorithm using health record data is needed.