Machine Learning for Early Detection of Preeclampsia and Gestational Diabetes from Routine Clinical Data and Paternal Immunological Proxies: A Systematic Literature Review

Abstract

Systematic literature review (PROSPERO CRD420251061201) of machine-learning prediction models for preeclampsia (PE) and gestational diabetes (GDM) using routinely collected clinical data, with paternal and partner variables as a primary analytic axis. 446 studies were screened across five databases (PubMed/MEDLINE, Scopus, Web of Science, Embase, IEEE Xplore; January 2021 to March 2026) and appraised with a dual-reviewer eight-criterion checklist (C1 to C8, threshold 4.5/8.0). Key findings: only 2 of 446 models (0.4 percent) used a paternal variable as a model input; the random-effects pooled AUC-ROC was 0.836 (95 percent CI 0.760 to 0.892); external validation was reported by 25.3 percent of studies and calibration by 12.3 percent. This deposit contains the manuscript, the LaTeX sources, and the manuscript, the LaTeX sources, the analysis code and protocol, and the complete per-study extraction dataset (extraction_446_full.csv and the source quality-assessment workbook: all 446 studies with dual-reviewer C1 to C8 scores, decisions, and Golden Rule outcomes), enabling independent reproduction of every reported quantity.