Evaluating and Refining Claims-Based Algorithms for Pregnancy Outcomes and Gestational Age Estimation in Korea

초록

Background A recently developed claims-based algorithm, originally designed using data from pregnancies with systemic lupus erythematosus, may enhance the identification of pregnancy episodes in Korean observational research. However, its applicability for estimating gestational age (GA) has not yet been validated in the general pregnant population. This study aimed to evaluate and refine the algorithm’s performance. Methods We utilized nationwide claims data from the Korean National Health Insurance Service between February 27, 2021, and December 31, 2022, to identify pregnancy episodes, including deliveries, stillbirths (SBs), and abortive outcomes (ABs). GA for each episode was estimated using both the original and modified algorithms, which prioritized second-to-third trimester target scan (TS) codes and first-trimester TS codes, respectively. Identified episodes with GA outside clinically plausible ranges for each pregnancy outcome were either reclassified into different outcomes or excluded. Algorithm performance was evaluated based on the following three criteria: first, the agreement between the estimated GA and the specified GA range in delivery episodes with a preterm birth (PB) code; second, comparison of estimated preterm birth (ePB) rates with national statistics reported by the Korean Statistical Information Service (KOSIS); third, sensitivity analysis of ultrasound code prioritization on the ePB rate. Results Among 581,740 pregnancy episodes, 456,157 were deliveries, 3,050 were SB, and 122,533 were AB. Deliveries with a GA of less than 37 weeks were classified as ePB, with rates of 39.7% in the original algorithm and 11.9% in the modified algorithm. In delivery episodes with PB codes that included specified GA ranges, the modified algorithm demonstrated higher agreement with the specified GA compared to the original algorithm (96.4% vs. 75.2%). The ePB rate estimated by the modified algorithm was also more consistent with the 9.4% reported by KOSIS, compared to the original algorithm (12.0% vs. 39.8%). Conclusion The modified algorithm, which prioritized first-trimester TS codes, improved the accuracy of GA estimation and reduced the underestimation observed in the original algorithm.

키워드