A comparison of approaches to identify live births using the medicaid analytic extract

Medicaid claims are an important, but underutilized source of data for neonatal health services research in the United States. However, identifying live births in Medicaid claims data is challenging due to variation in coding practices by state and year. Methods of identifying live births in Medicaid claims data have not been validated, and it is not known which methods are most appropriate for different research questions. The objective of this study is to describe and validate five approaches to identifying births using Medicaid Analytic eXtract (MAX) from 45 states (2006-2014). We calculated total number of MAX births by state-year using five definitions: (1) any claim within 30 days of birth date listed in personal summary (PS) file, (2) any claim within 7 days of PS birth date, (3) live birth ICD-9 in inpatient or other therapies file, (4) live birth ICD-9 code in inpatient file, (5) live birth ICD-9 in inpatient file with matching PS birth date. We then compared the number of MAX births by state and year to expected counts using outside data sources. Definition 1 identified the most births (14,189,870) and was closest to total expected count (98.3%). Each definition produced over- and underestimates compared to expected counts for given state-years. Findings suggest that the broadest definition of live births (Definition 1) was closest to expected counts, but that the most appropriate definition depends on research question and state-years of interest.