First-Trimester Machine Learning to Predict Preeclampsia in Normotensive Pregnancies by American Heart Association Guidelines.

OBJECTIVE: This study aimed to determine whether unsupervised machine learning can identify phenotypically distinct subgroups at increased risk for preeclampsia among pregnant individuals with American Heart Association (AHA)-defined normal blood pressure in the first trimester.

METHODS: This was a secondary analysis of a prospective cohort study of singleton pregnancies enrolled at ≤136/7 weeks' gestation at two academic centers. Participants with prepregnancy chronic hypertension or major fetal/placental abnormalities were excluded. First-trimester blood pressure was categorized using the 2017 AHA guidelines. Among individuals with AHA-defined normal blood pressure (<120/80 mm Hg), unsupervised machine learning (k-means clustering) was applied to systolic, diastolic, and mean arterial pressure to identify distinct hemodynamic phenotypes. The primary outcome was preeclampsia; secondary outcomes included hypertensive disorders of pregnancy (HDP) and small-for-gestational age (SGA) neonates. Associations were assessed using multivariable Cox regression and Kaplan-Meier analyses.

RESULTS: Of 570 participants, 378 (66.3%) had AHA-normal blood pressure. Among these, machine learning identified a high-risk cluster (7.4%) and a low-risk cluster (92.6%). Despite normotensive values, individuals in the high-risk cluster had a significantly higher incidence of preeclampsia (25.0 vs. 3.1%; p < 0.001) and HDP (28.6 vs. 5.7%; p < 0.001) compared to the low-risk cluster. After adjustment, the high-risk normotensive cluster had an eight-fold increased hazard of preeclampsia (adjusted hazard ratio [aHR] = 8.01; 95% CI: 3.09-20.74) and increased risk of SGA (adjusted odds ratio [aOR] = 3.36; 95% CI: 1.36-8.31). Risk within this group exceeded that of individuals with AHA-abnormal blood pressure.

CONCLUSION: Among pregnant individuals with first-trimester AHA-normal blood pressure, unsupervised clustering identified a distinct subgroup at elevated risk for preeclampsia and SGA. These findings suggest that conventional thresholds may overlook early vascular risk and support further investigation into machine learning-based risk stratification in pregnancy.

KEY POINTS: · Machine learning identified a distinct high-risk cluster (7.4%) within normotensive pregnancies.. · This cluster had an eight-fold higher risk of preeclampsia and a three-fold increased risk of SGA neonate.. · Machine learning may enhance early pregnancy risk stratification..