(153) Evaluating the Accuracy of Infant Weight and Length Measurements from Electronic Health Records for Use in Maternal Vaccination Safety Studies

Background: Insufficient infant growth following prenatal vaccine exposures has been identified as an important vaccine safety outcome. Electronic Health Records (EHR) can be used for pharmacovigilance studies. However, growth measurements in EHR vary by timing of infant health care visits and may contain data errors.

Objectives: To assess accuracy of infant growth measurements in EHR and predicted values.

Methods: Weight and length data from well-child visits (WCV) were extracted from the EHR for infants aged 0-12 months at a health system participating in the Mothers and Infants LinKed for Heathy Growth (MILk) study, 2014-2017 and 2021-2023. We removed biologically implausible values (BIV) from EHR data using the growthcleanr R package. We used weight in kg and length in cm measured at 1-, 3- and 6-month MILk study visits as the reference standard. We predicted weight and length at MILK study visits from EHR data using a generalized additive model with cubic spline and autoregressive correlation structure. We assessed accuracy of the EHR value nearest to the MILk study visit, limiting to those with EHR measurements within 9 days of the 1-month study visit and within 15 days of the 3- and 6-month study visits, and the predicted values. We estimated absolute differences (ADIFF), 95% Confidence Intervals (CI) and Technical Error Measurement (TEM) relative to the study values.

Results: The study comprised 260 infants with WCVs and 1352 growth measurements from the EHR; 15 BIV were removed. There were 67 (26%) infants with data within 9 days of the 1-month study visit with an ADIFF for weight of -0.04 (CI: -0.09, 0.004, TEM: 0.07 for nearest value and -0.06 (CI: -0.11, -0.02, TEM:0.12 ) for predicted value, and for length of 0.05 (CI: -0.32, 0.42, TEM: 0.04) for nearest value and 0.38 (CI: 0.21, 0.55, TEM: 0.04) for predicted value. There were 9 (4%) infants with data within 15 days of the 3-month visit with an ADIFF for weight of 0.01 (CI: -0.20, 0.23, TEM: 0.06) for nearest value and -0.001 (CI: -0.04, 0.03, TEM: 0.06) for predicted value and for length of 0.74 (95% CI: -0.35, 1.82, TEM: 0.03) for nearest value and 0.65 (CI: 0.50, 0.81, 0.03) for predicted value. There were 199 (77%) infants with data within 15 days of the 6-month visit with an ADIFF for weight of -.02 (CI: -0.05, -0.002, TEM: 0.03) for nearest value and 0.02 (CI: -.02, 0.05, TEM 0.05) for predicted value and for length of 0.78 (CI: 0.56, 1.0, TEM: 0.04) for nearest value and 0.70 (CI: 0.54, 0.87, TEM: 0.03) for predicted value.

Conclusions: When available, weight and length measurements from WCV near in time to reference standard measurements had high accuracy. Predicted values also had high accuracy. When weight and length data are not available from EHR at ages of interest, predicted values may be used to evaluate growth in pharmacovigilance studies.