Bioinformatics for Multimodal Precision Pregnancy: A Systematic Review of Causal, Foundation, and Privacy-Preserving Models for Predicting Pregnancy Complications and Fetal Health Outcomes
Wiku Andonotopo, Mochammad Besari Adi Pramono, Julian Dewantiningrum, Muhammad Adrianes Bachnas, Wisnu Prabowo, I Nyoman Hariyasa Sanjaya, Anak Agung Gede Putra Wiradnyana, Anak Agung Ngurah Jaya Kusuma, Khanisyah Erza Gumilar, Muhammad Ilham Aldika Akbar, Ernawati Darmawan, Aloysius Suryawan, Ridwan Abdullah Putra, Theresia Monica Rahardjo, Anita Deborah Anwar, Dudy Aldiansyah, Laksmana Adi Krista Nugraha, Waskita Ekamaheswara Kasumba Andanaputra, Wibisana Andika Krista DharmaAbstract
Pregnancy is increasingly described as a data-rich physiological state, yet the translation of bioinformatics advances into clinically reliable prediction remains uneven. We conducted a systematic review, following Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 guidance, to examine how causal modeling, foundation architectures, and privacy-preserving strategies are being applied to predict pregnancy complications and fetal health outcomes. Comprehensive searches of major databases and trial registers identified 1558 records, of which 30 studies met predefined eligibility criteria after duplicate removal, screening, and full-text assessment. Included studies spanned preeclampsia, preterm birth, fetal growth restriction, macrosomia, stillbirth, and maternal morbidity, using approaches ranging from conventional machine learning on clinical variables to multiomics integration, deep learning for ultrasound, and federated learning frameworks. Methodological heterogeneity was substantial. Although several large cohort models reported encouraging discrimination, calibration was inconsistently addressed, and external validation—particularly across health systems and demographic subgroups—remained limited. Multiomics and benchmarking studies suggested biological signal reproducibility, yet translational gaps persist between mechanistic insight and bedside utility. Emerging foundation models and federated architectures demonstrate technical feasibility for scalable and privacy-aware deployment, but few studies have integrated fairness auditing or real-world performance monitoring. Risk of bias was generally moderate, often reflecting retrospective designs, limited transparency, and potential spectrum effects. Overall, bioinformatics-driven prediction in pregnancy is advancing from proof-of-concept toward early translational readiness, but remains constrained by validation rigor, equity considerations, and governance challenges. Future work must prioritize prospective evaluation, calibration stability, causal inference frameworks, and adaptive model oversight to achieve truly precision-based, equitable maternal–fetal care.