Development and Validation of Gestational Age Estimation Algorithms for Nonlive Births in Administrative Healthcare...

Area of Science:

• Reproductive Health
• Biostatistics
• Public Health

Background:

• Accurate gestational age (GA) estimation is crucial for non-live births.
• Existing algorithms developed in the US may not directly apply to Korean healthcare data.
• This study evaluates the validity of adapted GA estimation algorithms in South Korea.

Purpose of the Study:

• To adapt and validate algorithms for estimating gestational age (GA) in non-live births using Korean healthcare claims data.
• To compare the performance of different algorithmic approaches, including outcome-specific GA, ultrasound-adjusted GA, regression models, and random forest models.
• To determine the most suitable method for GA estimation in non-live births within the Korean National Health Information Database (NHID).

Main Methods:

• Utilized the South Korean National Health Information Database (NHID), linking vaccination registry data for a reference standard.
• Stratified non-live births into spontaneous/induced abortions and stillbirths.
• Tested four algorithms: outcome-specific GA, ultrasound-adjusted GA, regression model, and random forest model, evaluating accuracy using Mean Squared Error (MSE) and proportion within 1-4 weeks of reference standard.

Main Results:

• Random forest models demonstrated superior performance for both spontaneous/induced abortions and stillbirths in the primary analysis.
• In external validation, ultrasound record-based adjustment showed comparable performance to random forests.
• High proportions of predictions were within two weeks of the reference standard for both methods.

Conclusions:

• Deterministic approaches, such as ultrasound-based adjustments, are preferable for estimating GA of non-live births in the NHID due to their simplicity and comparable performance.
• These validated algorithms can significantly support pregnancy research within the Korean population.
• The findings highlight the adaptability of GA estimation methods across different healthcare systems.