Evaluation of gestational age by pregnancy outcomes and distribution of pregnancy-related codes in Korean claims data

TL;DR

This study evaluates how well gestational age can be estimated from claims data in Korea, finding it works well for live births but less so for non-live outcomes.

Contribution

The study introduces a fixed-duration algorithm for gestational age estimation and evaluates its accuracy across different pregnancy outcomes in Korean claims data.

Findings

Algorithm-based GA estimation was highly accurate for live births (92.2% within ±2 weeks).

Accuracy was significantly lower for non-live birth outcomes like stillbirth and spontaneous abortion.

Incorporating prenatal tests and complications could improve GA estimation accuracy.

Abstract

This study aimed to evaluate a fixed-duration algorithm for gestational age (GA) estimation according to pregnancy outcomes and to describe the GA distribution of pregnancy-related codes in Korea.

We included 351,055 pregnancy episodes (2019–2022) from linked data between the National Health Insurance Service and the Korea Immunization Registry Information System (KIRIS). GA from claims data was estimated by subtracting fixed durations from the delivery date (algorithm-based GA), and GA derived from KIRIS was defined as the gold standard. Accuracy was evaluated as the proportion of episodes in which the difference between the estimated GA and the reference standard fell within ±2 weeks. We described the distributions of the GA at which each prenatal test, pregnancy complication, and diagnostic code was recorded.

Algorithm-based GA estimation showed high accuracy for live births (92.2% within ±2 weeks) but markedly lower accuracy for non-live birth outcomes, including stillbirth (3.3%), termination (7.2%), spontaneous abortion (45.2%), and ectopic pregnancy (20.0%). In additional analyses aimed at identifying potential indicators for improving GA estimation, most events occurred within clinically expected timeframes, although some individual codes exhibited poor temporal alignment.

Algorithm-based GA estimation using claims data performed well for live births but demonstrated limited accuracy for non-live birth outcomes. Incorporating information from prenatal tests and pregnancy complications may enhance GA estimation.

Figures