43 - Validating and Calibrating a Claims-Based Algorithm for Estimating Gestational Age Using Electronic Health Record Data

Publication Number: 3040.43

Nahed O. ElHassan, University of Arkansas for Medical Sciences College of Medicine, Little Rock, AR, United States; Cheng Peng, University of Arkansas for Medical Sciences (UAMS), Little Rock, AR, United States; Mahip Acharya, University of Arkansas for Medical Sciences College of Medicine, Little Rock, AR, United States; Rong Rong, Tulane University, New Orleans, LA, United States; Keran Moll, Regeneron Pharmaceuticals, thousand oaks, CA, United States; Lizheng Shi, Tulane University, New Orleans, LA, United States; Bradley C. Martin, University, Little Rock, AR, United States

Background: Accurate identification of pregnancy episodes and gestational age (GA) is essential for maternal research using real-world data. The Moll claims-based algorithm provides a way to establish pregnancy episodes and estimate GA, but its accuracy in estimating GA has not been extensively validated against EHR data.

Objective: 1) Assess the accuracy of the Moll algorithm to estimate GA compared to clinically derived GAs and 2) Develop a modified Moll algorithm by calibrating a key variable to known clinically derived GAs.

Design/Methods: The Moll algorithm was applied to the University of Arkansas for Medical Sciences EHR data (1/1/2018 - 1/3/2025) to establish pregnancy episodes and GAs. Each Moll-identified episode was linked to EHR-derived episodes by matching on delivery dates. For each linked episode, Moll-assigned GA was compared to EHR-recorded GA. Calibration focused on a common formula to estimate the pregnancy start date: Service Date - (GA in weeks*7 + X), where X was incrementally adjusted to minimize the difference between clinically derived and Moll estimated GAs in a 60% training sample, and subsequently validated in a 40% test sample. Descriptive statistics summarized GA estimates, agreement with EHR-recorded GA, and full-term birth (FTB, GA at birth ≥37 weeks) and preterm birth (PTB, GA at birth < 37 weeks) classification.

Results: The original Moll algorithm identified 22,216 linked episodes when matching delivery date within ±14-days (Table 1) and 9,510 linked episodes when requiring an exact match. On average, GA at delivery was underestimated by 3.22 days using ±14-day window and 2.21 days using exact match samples. In the 60% calibration sample, applying a "+1" adjustment yielded the smallest mean absolute difference in GA estimates (0.45 days) (Figure 1). The updated formula, service date - (GA in weeks*7 +1), reduced the mean GA difference in the test sample from 2.41 days to 0.25 days (Table 2a). PTB rates were 27.0% and 24.7% using the original and updated Moll algorithms respectively, compared to a 20.0% PTB derived from the EHR data (Table 2b).

Conclusion(s): Validation against EHR data confirmed that the original Moll algorithm slightly understates GA, and minor adjustments to formulas used in the most common steps can improve the accuracy of estimated GAs and the classification of PTB.

Table 1. Comparison of Stepwise Gestational Age Estimates Derived from the Original Moll Algorithm versus Electronic Health Records Recorded Gestational Age (±14-day Delivery Date Window, N=22,216)
Abbreviations by alphabetical order: CI, confidence interval; EHR, electronic health records; GA, gestational age.

Figure 1. Calibration of the Modified Moll Algorithm for Gestational Age (GA) Determination
Formula: Pregnancy start date = service date − (GA in weeks × 7 + X), where X is a constant ranging from -6 to +6
Overall mean day difference(X-axis):
Average of (GA estimated using the modified Moll's algorithm - GA recorded in the electronic health records)


Table 2. Comparison of Estimates between the Original and Updated Moll Algorithms
Abbreviations by alphabetical order: CI, confidence interval; EHR, electronic health records; FTB, full term birth; GA, gestational age; PTB, preterm birth.