AI-Based Model for Early Prediction of Preeclampsia Using First-Trimester Biomarkers-Cureus Study

A recent study on the development and validation of an AI-based framework for first-trimester preeclampsia risk assessment proposed a model that demonstrated superior performance compared to conventional and modern predictive models. It provides interpretable predictions, enhancing clinical transparency and potentially supporting clinician decision-making.

The study also highlights that this framework enables the early identification of pregnancies at a higher risk, which may guide timely monitoring and intervention to improve maternal and foetal outcomes. The design allows scalability and potential integration into existing prenatal screening workflows, with future expansion to incorporate multimodal data sources, such as imaging or genomic information.

This prediction model development and validation study was published in December 2025 in Cureus.

Introduction

Despite significant advances, predicting preeclampsia (PE) in early pregnancy remains inadequate because of low-sensitivity screening tools, inconsistent biomarker performance, and reliance on clinical judgement. Many high-risk women remain undetected until clinical symptoms appear, increasing the need for emergency interventions and the risk of maternal-foetal complications. There is an urgent need to improve the early detection of PE to reduce maternal mortality, perinatal morbidity, and healthcare burden.

AI-enabled diagnostic frameworks can analyse multidimensional biomarker and clinical data, identify high-risk pregnancies earlier, and guide preventive measures. AI and ML techniques excel in identifying nonlinear patterns in high-dimensional data, including maternal characteristics, biochemical markers, Doppler parameters, and inflammatory indices. These models improve risk stratification and predictive accuracy for prenatal care integration. Additionally, interpretability techniques, such as SHapley Additive exPlanations (SHAP) analysis, help clinicians understand biomarker contributions, enhancing trust and clinical decision-making.

Study Overview

The five main stages of an artificial intelligence (AI)-based framework for early prediction of preeclampsia (PE) include data collection, preprocessing, feature engineering, model creation, and performance evaluation

1. Data acquisition included pregnancy-associated plasma protein A (PAPP-A), placental growth factor (PlGF), beta-human chorionic gonadotropin (β-hCG), blood pressure (BP), body mass index (BMI), maternal age, and uterine artery pulsatility index (UtA-PI).
2. Data preprocessing included missing values, outliers, normalization, and class imbalance using the synthetic minority oversampling technique (SMOTE).
3. Feature engineering incorporates multiple median (MoM) conversions, biomarker ratios, and recursive feature elimination (RFE).
4. Model development employs logistic regression (LR), support vector machines (SVMs), random forests (RFs), extreme gradient boosting (XGBoost), and deep neural networks (DNNs).
5. Model evaluation used accuracy, area under the receiver operating characteristic curve (AUC-ROC), and SHapley Additive exPlanations (SHAP), culminating in high-risk versus low-risk PE classification.

Key Findings

• The DNN surpassed all other models, attaining the highest accuracy (95%), area under the receiver operating characteristic curve (AUC-ROC) of 0.97, and balanced sensitivity and specificity, with 92 true positives (TP) and only eight false negatives (FN).
• XGBoost closely followed with an AUC-ROC of 0.95, whereas logistic regression (LR) demonstrated the lowest overall performance.
• SHAP analysis highlighted placental growth factor (PlGF), pregnancy-associated plasma protein-A (PAPP-A), and mean arterial pressure (MAP) as significant contributors to risk classification, with maternal body mass index (BMI) and age also contributing.

The authors highlighted the potential of AI to enhance early PE prediction using first-trimester biomarkers. They also highlighted the connection between placental dysfunction and maternal circulatory adaptation during early gestation. The AI framework achieved high predictive accuracy across the ML models, with the DNN performing best. These findings suggest that DNNs can capture complex interactions among biochemical, biophysical, and clinical features, supporting their utility for early PE detection.

Clinical Implications

This study established a robust foundation for AI-assisted maternal risk assessment during early pregnancy. The predictive framework demonstrated the model's feasibility, clinical relevance, and translational potential. With external validation and integration into health record systems, the AI-based tool could improve the identification of high-risk pregnancies, support preventive interventions, and enhance maternal and fetal outcomes. The broader adoption of AI-driven screening approaches can strengthen precision obstetrics and reduce hypertensive disorders during pregnancy. By identifying risk determinants, the framework guides personalized preventive strategies, such as early aspirin consideration or antenatal surveillance, which reduce the incidence of PE when implemented early in high-risk pregnancies.

Reference: Tabassum S, Kishwar N, Usman Z, et al. Artificial Intelligence-Based Prediction of Preeclampsia Using First-Trimester Biomarker Levels. Cureus 17(12): e100059. Published December 25, 2025. DOI 10.7759/cureus.100059