Evaluating Maternal and Fetal Risk in Preeclampsia Using Thyroid and Inflammatory Markers: A Naive Bayes and Network Analysis Approach.

BACKGROUND

Preeclampsia is a significant factor in maternal and fetal morbidity globally, highlighting the necessity for early risk assessment and intervention. This study introduces a machine learning (ML) approach that integrates maternal thyroid profiles and C-reactive protein (CRP) levels to predict preeclampsia and potential fetal thyroid abnormalities.

MATERIALS AND

METHODS

A cross-sectional dataset of 174 pregnancies (91 normal, 83 preeclampsia) was analyzed. EN regularization identified maternal CRP, thyroid-stimulating hormone (TSH), and anti-thyroperoxidase antibody as the primary predictors of preeclampsia, with CRP being the most significant contributor.

RESULTS

A Naive Bayes model trained on these markers achieved remarkable predictive performance, with an accuracy of 92.5% ± 7.7%, a sensitivity of 99.9%, and a specificity of 86.1% ± 1.4%. Network analysis identified a substantial maternal-fetal connection via FT3 and cord TSH, establishing the foundation for a secondary NB model predicting fetal hypothyroidism. The model exhibited moderate overall accuracy (63.7% ± 5.3%) but demonstrated significant improvement in the preeclampsia subgroup (73.3% ± 1.7%) while maintaining 100% sensitivity. CRP levels differed significantly between groups, underscoring its role as a systemic inflammatory marker in the pathogenesis of preeclampsia.

CONCLUSIONS

This comprehensive ML architecture links maternal indicators to fetal outcomes, demonstrating the practicality of using interpretable, probabilistic models in prenatal care. The results support early risk classification and targeted monitoring for high-risk pregnancies, emphasizing the clinical value of integrating thyroid and inflammatory markers into maternal screening. Future validation across varied cohorts may facilitate the clinical implementation of these models to enhance maternal-fetal health outcomes.