Abstract
<title>Abstract</title> <p>Background Accurate gestational age (GA) assignment is fundamental to pregnancy epidemiologic research but is not directly available in administrative claims databases. Existing claims-based algorithms were developed primarily for singleton pregnancies; and have not been adequately evaluated for multiple gestations, which differ substantially in obstetric characteristics and healthcare utilization patterns. We aimed to develop and validate claims-based GA estimation algorithms for multiple gestations. Methods We identified multiple pregnancies with delivery-related claim codes in the National Health Insurance Service (NHIS) database between 2015 and 2024. Reference-standard GAs were obtained through linkage with the National Health Screening Program for Infants and Children (NHSPIC). Cases with a neonatal birth date outside the maternal hospitalization period or a reference GA of < 24 or > 40 weeks were excluded. The dataset was randomly divided into development and test sets. Algorithm 1 used preterm birth-related ICD-10 codes, Algorithm 2 used prenatal ultrasonography timing, Algorithm 3 used prenatal screening tests timing, and Algorithm 4 combined these approaches hierarchically. Algorithm performances were evaluated using mean absolute errors (MAE), accuracy within ± 7 and ± 14 days, and missing rate. Results A total of 26,714 eligible multiple gestation pregnancies were included in the final cohort. Algorithm 1, based on preterm birth-related diagnostic codes, showed the lowest estimation accuracy (MAE 10.3 days; accuracy within ± 7 or ± 14 days; of 51.4% and 79.8%, respectively). Algorithm 2, based on first-trimester targeted sonography, showed the highest estimation accuracy (MAE 5.59 days; accuracy within ± 7 or ± 14 days, of 81.8% and 91.3%, respectively) but had a missing rate of 27.5%. Algorithm 3 had an MAE of 6.42 days, an accuracy within ± 7 or ± 14 days of 77.4% and 89.2%, respectively, and a missing rate of 9.6%. Algorithm 4, the hierarchical algorithm, achieved complete GA assignment with no assignment failure while maintaining high estimation performance (MAE 6.14 days; accuracy within ± 7 or ± 14 days of 79.1% and 89.9%, respectively). Conclusions A hierarchical framework that sequentially applied ultrasonography, prenatal screening tests, and diagnostic codes achieved complete GA assignment without missing values while maintaining high estimation performance. This framework provides a potential means of estimating GA in future claims-based studies on multiple gestation pregnancies.</p>