Director, Pharmacoepidemiology Eli Lilly and Company Indianapolis, United States
Background: FDA often requires post-approval pregnancy safety studies for medications approved for use in a population that includes women of childbearing age (FDA 2019). One commonly requested approach utilizes secondary data, often claims or electronic health records. Identification of appropriate data to execute these studies, which rely on data linking maternal and infant records, is critical.
Objectives: To implement the structured process to identify fit-for-purpose data (SPIFD) (Gatto et al. 2022) for executing a pregnancy database study.
Methods: We applied the SPIFD framework to identify data source(s) that were optimal for addressing an FDA post-marketing commitment to assess maternal and infant outcomes, including major congenital malformations, in women exposed to an Eli Lilly medication during pregnancy compared to an unexposed control population. Five blinded, commercially available data sources for pregnancy research were assessed on the following broad categories: dataset characteristics (e.g. number of pregnancies and births, mother-to-infant linkage performance), logistical considerations (e.g. data access, contracting), maternal data characteristics (e.g. medication exposure, baseline confounding characteristics, outcomes), and infant data characteristics (e.g. outcomes). Data elements were ranked a priori according to priority. The highest rank was given for data elements related to validated mother-infant linkage, capturing dispensed medication, outcomes (in mother and infant records), ability to follow infants after birth, and potentially confounding diagnoses. Results were summarized into a heatmap to visualize the suitability of each data source.
Results: Data elements related to maternal and infant characteristics were generally consistently met across the 5 data sources. Differentiation for the data sources was most frequently due to dataset characteristics (size, performance of mother-infant linkage) and logistics. Based on visual inspection of the heatmap and output from detailed cohort counts from 2 of the data sources, a single data source was deemed best suited.
Conclusions: Application of the SPIFD framework to assess data for use in a pregnancy database study allowed for a systematic and rapid identification of a fit-for-purpose data source to develop a protocol per milestones required by the FDA, thus informing a critical step towards generating real-world evidence on the safety of medication exposure during pregnancy. A limitation of applying such a framework before marketing authorization is the difficulty in specifying sample size feasibility, as this may be impacted by access to and utilization of medication among women of childbearing age.
