Poster Session 2
Category: Hypertension
Poster Session 2
Yossi Bart, MD
MFM fellow
Division of Maternal-Fetal Medicine, Department of Obstetrics, Gynecology and Reproductive Sciences, McGovern Medical School at UTHealth Houston
Houston, Texas, United States
.jpg "Zakaria Doughan, MD photo")
Zakaria Doughan, MD
Research Assistant
Department of Obstetrics and Gynecology, McGovern Medical School at UT Health, Houston
Houston, Texas, United States
.jpg "Devangi D. Mahtani, MD (she/her/hers) photo")
Devangi D. Mahtani, MD (she/her/hers)
Maternal-Fetal Medicine Fellow
Division of Maternal-Fetal Medicine, Department of Obstetrics, Gynecology and Reproductive Sciences, McGovern Medical School at UTHealth Houston
UT Health Houston, Texas, United States
Joe Haydamous, MD (he/him/his)
PGY1
Department of Obstetrics, Gynecology and Reproductive Sciences, McGovern Medical School at UTHealth Houston
Department of Obstetrics and Gynecology, McGovern Medical School at UT Health, Houston, Texas, United States
Ahmed Zaki Moustafa, MD, MS (he/him/his)
Assistant Professor
Division of Maternal-Fetal Medicine, Department of Obstetrics, Gynecology and Reproductive Sciences, McGovern Medical School at UTHealth Houston
University of Texas - Houston, Texas, United States
Sean C. Blackwell, MBA, MD
Professor
Division of Maternal-Fetal Medicine, Department of Obstetrics, Gynecology and Reproductive Sciences, McGovern Medical School at UTHealth Houston
Houston, Texas, United States
Baha M. Sibai, MD
Professor
Division of Maternal-Fetal Medicine, Department of Obstetrics, Gynecology and Reproductive Sciences, McGovern Medical School at UTHealth Houston
Houston, Texas, United States
During expectant management of preeclampsia with severe features < 34 weeks, delivery is indicated when maternal or fetal condition deteriorates. This study aimed to develop machine learning models using data available at admission to predict short latency (delivery ≤3 days of diagnosis) and major maternal or fetal complications among patients managed expectantly.
Study Design:
A retrospective cohort study at a Level IV referral center, including patients with preeclampsia with severe features between 23w0d and 33w3d (2016–2025). Patients who delivered within 1 day or presented with contraindications to expectant management were excluded. Using 87 baseline clinical, laboratory, and sonographic features available at admission, we compared patients with short vs. longer latency, and those with and without the composite adverse outcome (CAO; Figure 1). Four models (logistic regression, support vector machines, random forest, and eXtreme Gradient Boosting) were trained using five-fold cross-validation, grid search, and 100 repeated randomized train-test splits. Undersampling was used to balance classes and minimize overfitting. The top predictors per outcome were used in the final models.
Results:
Overall, 370 patients met inclusion criteria; 116 (31%) had a short latency, while 52 (14%) ultimately had the CAO. The best-performing prediction model achieved an AUC of 0.70 (95% CI 0.61-0.79; Figure 1) for short latency with sensitivity of 0.72 (95% CI 0.55-0.89) and specificity of 0.69 (95% CI 0.52-0.86). For the CAO development, the AUC was 0.70 (95% CI 0.63-0.77) with 0.67 sensitivity (953% CI 0.50-0.84) and 0.72 specificity (95% CI 0.53-0.91). The predictors with highest mean importance for both models were chronic hypertension, maternal age, and urine protein-creatinine ratio (Figure 2).
Conclusion:
Machine learning models using admission data showed modest ability to predict early delivery and complications in expectantly managed preeclampsia with severe features. Integrating machine learning into clinical decision support systems may enhance early risk stratification and improve resource allocation.