Lead/Presenter: Qiyan Mu, Clement J. Zablocki VA Medical Center
All Authors: Mu Q (Clement J. Zablocki VA Medical Center), Zhang N.(VA San Francisco Healthcare System) Ndakuya-Fitzgerald F.(Clement J. Zablocki VA Medical Center) Whittle J. (Clement J. Zablocki VA Medical Center; Medical College of Wisconsin) Hoggatt KJ (VA San Francisco Healthcare System; UCSF Department of Medicine)
Objectives:
In VA, pregnant Veterans typically receive maternity care through community providers while using VA services for other healthcare needs. This use of multiple healthcare systems creates unique challenges in tracking the care and outcomes for these Veterans. Accurate identification of women Veterans’ pregnancy episodes is critical to VA as a learning healthcare system: Identification is the essential first step in evaluating healthcare access and delivery, patient outcomes, and costs associated with providing pregnancy care. Prior efforts to identify pregnancies using only information on deliveries missed non-viable pregnancies (e.g., those ending in early loss) and thus provide an incomplete picture of women Veterans’ reproductive health.
Methods:
We developed an algorithm for identifying a pregnancy episode using available VA data sources. The six main steps in the algorithm included: (1) Developing comprehensive code lists for potential pregnancy care markers (PCM) and end-of-pregnancy (EOP) events based on literature and adding VA specific indicators; (2) Identifying all women Veterans with a potential pregnancy visit between 2013 to 2019 using the developed PCM list; (3) Establishing a starting “index†date for the first potential pregnancy episode in the study period and then applying a 60-day “lookback†period to ensure the index date was the start of a new pregnancy episode; (4) Creating a pregnancy episode “window†of 280 days from the index date forward; (5) Identifying an EOP event and date within the pregnancy episode window; (6) Reconciling conflicting EOP data for women with more than one EOP event within the pregnancy episode window. Data evaluations were implemented at each step to assess accuracy in data pulling and resolve any discrepancy.
Results:
The final cohort included n = 36,081 unique Veterans and n = 44,969 unique pregnancy episodes with available EOP records. For the 44,969 pregnancy episodes, n = 38757 (87%) resulted in live birth, n = 4,991(10%) had miscarriage/abortion, n = 962 (2%) experienced ectopic pregnancy, n = 259 (1%) had fetal loss. All relevant EOP information was captured, including pregnancy outcome, delivery method, birth complication, and newborn outcome for the pregnancy episode.
Implications:
Using available VA data sources, we developed an algorithm to identify pregnancy episodes using a range of pregnancy indicators. This approach differs from algorithms that rely only on delivery codes to define pregnancy in that it also captures pregnancy episodes with non-viable pregnancy outcomes not reported in previous VA database studies, including miscarriage/abortion, ectopic pregnancy, and fetal loss.
Impacts:
Developing an accurate method to identify pregnancy care episodes is important for VA to monitor and address care quality and disparities in pregnancy care and outcomes. Incorporating VA-specific PCMs enhances our ability to detect pregnancies. Using an algorithm that included a forward step, in addition to ascertaining pregnancies retrospectively based on delivery data, resulted in more complete identification of women Veterans’ pregnancy episodes and outcomes. This algorithm can be adapted and utilized for both operational and research needs.