Associate, Vaccines Real World Evidence Pfizer Inc. New York, United States
Background: In accordance with the Health Insurance Portability and Accountability Act Privacy Rule, month and day of birth are not often available in US real-world databases. However, month of birth is essential when assessing timing and compliance with pediatric procedures, such as immunization schedules, in real-world data. Several algorithms have been used to impute birth month, but these have not been validated in published literature.
Objectives: The objective of this study was to validate commonly used birth proxy algorithms in Medicaid administrative claims, which provide beneficiaries’ month, day, and year of birth, and determine the utility of these algorithms for real-world data studies.
Methods: We developed four cohorts within Centers for Medicaid and Medicare Services Medicaid administrative claims. The validation cohort were members with true birth dates during 2016-2019. The enrollment cohort were members aged 0 when first enrolled during 2016-2019 and enrollment dates were used as proxy birth dates. The well-visit cohort included members with ≥1 newborn well-visit diagnosis/procedure code and aged 0 at first enrollment during 2016-2019. The date of the first well-visit code was the proxy birth date. The comprehensive cohort were members with ≥1 newborn well-visit code or diagnosis code indicative of a birth outcome (i.e., single live birth) and aged 0 at first enrollment during 2016-2019. The date of the first well-visit or diagnosis code was the proxy birth date.
We examined birth year, sex, race/ethnicity, US region, and continuous Medicaid enrollment in all cohorts and calculated the difference in days between proxy birth dates and true birth dates in the enrollment, well-visit, and comprehensive cohorts.
Results: There were 9,460,464 Medicaid members born during 2016-2019. The enrollment date, well-visit, and comprehensive cohorts identified 94%, 49%, and 67% of members born during 2016-2019, respectively. No substantial differences in demographics or enrollment were identified across the cohorts.
In the enrollment cohort, 56% of proxy birth dates were true birth dates, 27% were 1-30 days before birth, and 4% were 1-30 days after birth. In the well-visit cohort, 8% of proxy dates were true birth dates, < 1% were before birth, and 91% were 1-30 days after birth. In the comprehensive cohort, 90% of proxy dates were true birth dates, < 1% were before birth, and 10% were 1-30 days after birth.
Conclusions: Birth proxy algorithms using evidence of newborn well-visits and birth outcomes provide representative and sizeable populations with accurate imputed birth months for real-world data studies assessing compliance with pediatric procedures.
