(AE) Application of Claims‑Based Algorithms to Ascertain Gestational Age Among Multiple Pregnancy Outcomes

Publication Number: 1029

Background: Accurate estimates of gestational age (GA) and pregnancy outcome from administrative and clinical databases are critical to the validity of pregnancy exposure and outcome studies. Moll et. al. (Drug Safety 2021) recently described development and validation of hierarchical algorithms using HCPCS and ICD-10 diagnostic and procedure codes to ascertain GA and to classify pregnancy outcomes as full-term birth (FTB), pre-term birth (PTB), spontaneous abortion (SA), or stillbirth. The authors found strong agreement between the algorithms and physician adjudication of EMRs.

Objectives: To assess the consistency of the GA algorithms described by Moll across multiple real-world databases.

Methods: Using 4 administrative and electronic health record (EHR) databases (Optum Claims, Optum EHR, Pharmetrics Plus®, and MarketScan® Databases), the GA algorithm was applied to female subjects ages 12 – 55 years (inclusive), with a pregnancy outcome code of interest between 02/01/2017 – 06/30/2021, and continuous enrollment for at least 10 months prior and 30 days following the end of pregnancy. Patients with >1 type of pregnancy outcome within 7 days were excluded.

Results: There were 2,187,148 eligible pregnancies across the 4 databases. Application of the GA algorithm yielded 1,672,334 (76%) FTBs, 269,744 (12%) PTBs, 232,888 (11%) SAs, and 12,182 ( < 1%) stillbirths; GA could not be estimated for 68,065 (3%) pregnancies. The GA algorithm performed consistently across databases. GA estimation was most frequently based on a Z3A code (ICD-10 code for weeks of gestational age) with another type of service (52-69%), a Z3A code with a first trimester ultrasound (16-25%), and a Z3A code with an anatomic ultrasound (10-13%). The distribution of median GA at the end of pregnancy was similar across the datasets: 39 weeks for FTB; 36 for PTB; 23-26 weeks for stillbirths; and 10 weeks for SA.

Conclusions: The GA algorithm performed consistently across 4 databases, according to the step in the hierarchy each GA was estimated, and the age distribution for each outcome type. These results suggest the algorithm can be used in multiple real-world databases to estimate GA and may be a valuable tool for assessing pregnancy exposures and outcomes.