(292) Prediction of Hemolytic Disease of the Fetus and Newborn in neonates from alloimmunized pregnancies

Background: Maternal alloimmunization puts fetuses and neonates at risk of Hemolytic Disease of the Fetus and Newborn (HDFN), with manifestations including anemia, hyperbilirubinemia, hydrops fetalis, and fetal or neonatal demise. Little is known about predictors of HDFN and its severity in early pregnancy and pre-conception.

Objectives: To evaluate the performance of models in insurance claims data to predict HDFN due to Rhesus alloantibodies (RhHDFN).

Methods: Pregnancies with an alloimmunization code ending in live birth and linked to infant records in MarketScan Commercial Claims and Encounters (CCAE) database and Optum’s de-identified Clinformatics® Data Mart Database (Clinformatics®) from 2000 to 2022 were included. Outcomes were: 1) RhHDFN diagnosis within 30 days of birth; 2) Severe RhHDFN (neonatal mortality, hydrops fetalis diagnosis, intrauterine transfusion, neonatal simple or exchange transfusion). Candidate predictors included investigator-specified and data-driven characteristics. The population of interest was pregnancies with alloimmunization diagnosis in trimester 1. Lasso regression models were trained and tested in CCAE in a larger cohort of pregnancies with alloimmunization diagnosis in trimester 2 or 3, and externally validated in CCAE and Clinformatics® in the population of interest.

Results: In CCAE, 64,719 mother-infant pairs had a first alloimmunization diagnosis code in trimester 2 or 3 of pregnancy, with 342 experiencing RhHDFN, and 33 Severe RhHDFN. Among 18 candidate predictors, the strongest predictors of outcomes were related to antibody screening and measurement in trimester 1. Maternal age, region, and diagnosis of recurrent pregnancy loss were additional predictors of RhHDFN, while number of visits in trimester 1 was included for severe RhHDFN. Internal discrimination was estimated by the AUROC (95% CI): 0.58 (0.54 – 0.62) for RhHDFN; 0.68 (0.51 – 0.85) for Severe RhHDFN. Models were externally validated in 1,824 mother-infant pairs in CCAE and 1,619 in Clinformatics® with a trimester 1 alloimmunization code. Of these, 36 (2.0%) and 44 (2.7%) were RhHDFN-affected in CCAE and Clinformatics®, respectively. Discrimination was higher for Severe RhHDFN (AUROC (95% CI): 0.83 (0.77 – 0.89) CCAE; 0.83 (0.74 – 0.92) Clinformatics®) than all RhHDFN (AUROC (95% CI): 0.76 (0.68 – 0.83) CCAE; 0.71 (0.64 – 0.79) Clinformatics®).

Conclusions: Prediction of RhHDFN and severe RhHDFN was improved by model-building in a larger related population and achieved good performance when validated in two databases in the population of interest. Some predictors may reflect monitoring for HDFN; further exploration of obstetric history and sociodemographic characteristics may improve prediction.