Poster Session 1
Rachel Kim
University of Chicago
Chicago, Illinois, United States
Karen Huang
University of Chicago
Chicago, Illinois, United States
.jpg "Ashley Zimmermann, MD (she/her/hers) photo")
Ashley Zimmermann, MD (she/her/hers)
Resident
Northwell Lenox Hill Hospital
New York, New York, United States
Natalie Suder, MD
Resident
Lenox Hill Hospital, NY, United States
Maria Teresa Benedetto, MD
Attending Physician
Lenox Hill Hospital, New York, United States
Mio Sawai, MD
Attending Physician
Lenox Hill Hospital, New York, United States
Teresa Cheon, MD
Attending Physician
Lenox Hill Hospital, New York, United States
Yuzuru Anzai, MD
Attending Physician
Lenox Hill Hospital, New York, United States
Low birth weight (LBW) is classified as weight of less than 2,500 grams after birth. Identifying LBW is crucial due to its association with high-risk neonatal outcomes. This study aims to predict the probability of LBW for full term (37 week) neonates using maternal and pregnancy characteristics as well as the expected fetal weight (EFW) at 30 weeks.
Study Design:
CDC natality data from 2022 was used to select maternal, neonatal, and pregnancy factors associated with infant birth weight. Maternal factors included: BMI, age, height, race, weight gain during pregnancy, gestational diabetes, gestational hypertension, plurality, cigarette use, and pre-pregnancy diabetes. Pregnancy and neonate-related characteristics included: infant sex, gestational age, and total birth order. The birth weights for infants delivered at 37 weeks of gestational age from this database were used and unknown parameters were excluded (n = 439,251). The EFWs at 30 weeks for these neonates were computed using Hadlock’s EFW curves at the 10th and 90th percentiles. A least squares logistic regression model was generated using Google Colab to predict the probability of LBW given all the characteristics and EFW at 30 weeks. The model used an 80:20 split for testing and training, respectively, for validation purposes.
Results:
The model performed with an accuracy of 94.3%. The positive predictive value (PPV) was 93.4% and the negative predictive value (NPV) was 95.2%. Sensitivity was 95.3% and specificity was 93.3%. The area under the curve (AUC) score was 98.7% (Graph 1).
Conclusion:
The high accuracy, NPV, PPV, sensitivity, specificity, and AUC score of the model attest to its predictive strength. Moreover, the model was trained using a large, national dataset, demonstrating its applicability. Physicians can incorporate machine-based learning models in quantifying the risk of LBW during the earlier stages of the third trimester. Further model developments can include modeling based on earlier EFWs ( < 30 weeks) to provide specialized prenatal care for those who are at higher risk for LBW.
