Validation of Algorithms to Identify Small or Large for Gestational Age in the Korean Nationwide Healthcare Database
- 1College of Pharmacy, Seoul National University
Abstract
ObjectiveSmall for gestational age (SGA) and large for gestational age (LGA) are critical endpoints in perinatal pharmacoepidemiology, yet gestational age is frequently missing in administrative databases. This study validated algorithms for identifying SGA and LGA in the Korean National Health Information Database (NHID).
MethodsWe linked the NHID with national vaccination and infant health screening registries (2018-2021) to create a validation cohort of 94,159 pregnancies with reference standard gestational age and birth weight; infants with birth weights below the 10th or above the 90th percentile were considered SGA or LGA, respectively. We evaluated four algorithms: ICD-10 diagnosis codes from infant claims (Method A), maternal claims (Method B), either infant or maternal claims (Method C), and birth weight combined with estimated gestational age (Method D).
ResultsICD-10-based algorithms consistently underestimated prevalence, showing high specificity but low sensitivity (<17%). In contrast, Method D demonstrated superior performance: for SGA, sensitivity was 89.3%, specificity 96.8%, and positive predictive value (PPV) 71.6%; for LGA, sensitivity was 77.5%, specificity 98.3%, and PPV 84.7%.
ConclusionSGA and LGA can be identified with reasonable accuracy in the NHID using birth weight and estimated gestational age, whereas diagnosis codes alone can underestimate prevalence.
Related
- Algorithms to Identify Preeclampsia in Routinely Collected Healthcare Data: A Systematic Review.
- Knowledge of gestational diabetes mellitus and its sociodemographic determinants among pregnant women in Bangladesh: a nationwide facility-based cross-sectional study.
- Maternity versus non‐maternity healthcare settings for the diagnosis of gestational diabetes or diabetes in pregnancy
- Healthcare utilization and medical costs among adults with selected psychiatric disorders: a nationwide retrospective cohort study
- Outdoor play and catch-up growth in small-for-gestational-age children: a nationwide prospective birth cohort study from the Japan environment and children's study.
BibTeX
@misc{cho2026validation,
title = {{Validation of Algorithms to Identify Small or Large for Gestational Age in the Korean Nationwide Healthcare Database}},
author = {Y. Cho and H. Lim and B. Suh and Y. Lee and J.-Y. Shin},
howpublished = {medRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.09.26362678},
url = {https://www.medrxiv.org/content/10.64898/2026.09.09.26362678v1},
note = {CC BY-NC},
}
RIS
TY - UNPB
TI - Validation of Algorithms to Identify Small or Large for Gestational Age in the Korean Nationwide Healthcare Database
AU - Y. Cho
AU - H. Lim
AU - B. Suh
AU - Y. Lee
AU - J.-Y. Shin
PY - 2026
DA - 2026/09/09
PB - medRxiv
DO - 10.64898/2026.09.09.26362678
UR - https://www.medrxiv.org/content/10.64898/2026.09.09.26362678v1
AB - ObjectiveSmall for gestational age (SGA) and large for gestational age (LGA) are critical endpoints in perinatal pharmacoepidemiology, yet gestational age is frequently missing in administrative databases. This study validated algorithms for identifying SGA and LGA in the Korean National Health Information Database (NHID). MethodsWe linked the NHID with national vaccination and infant health screening registries (2018-2021) to create a validation cohort of 94,159 pregnancies with reference standard gestational age and birth weight; infants with birth weights below the 10th or above the 90th percentile were considered SGA or LGA, respectively. We evaluated four algorithms: ICD-10 diagnosis codes from infant claims (Method A), maternal claims (Method B), either infant or maternal claims (Method C), and birth weight combined with estimated gestational age (Method D). ResultsICD-10-based algorithms consistently underestimated prevalence, showing high specificity but low sensitivity (<17%). In contrast, Method D demonstrated superior performance: for SGA, sensitivity was 89.3%, specificity 96.8%, and positive predictive value (PPV) 71.6%; for LGA, sensitivity was 77.5%, specificity 98.3%, and PPV 84.7%. ConclusionSGA and LGA can be identified with reasonable accuracy in the NHID using birth weight and estimated gestational age, whereas diagnosis codes alone can underestimate prevalence.
ER -