Cardiac Remodeling in Preeclampsia: A Large Language Model-Assisted Meta-Analysis and Meta-Regression

Preeclampsia is a hypertensive disorder of pregnancy associated with substantial maternal morbidity and long-term cardiovascular risk, but the consistency of echocardiographic remodeling remains unclear. We conducted a mega-meta-analysis of left ventricular function and geometry, enabled by Synthesa AI, a large language model-based suite of tools. A PROSPERO-registered review (CRD420251109103) searched PubMed, Scopus, and Embase without date limits. Synthesa AI screened more than 138,000 abstracts, extracted data, assessed risk of bias, and generated Bayesian analytic code, with all outputs validated by human reviewers. Seventy-five studies which met the eligibility criteria were included. Preeclampsia was associated with a small but statistically significant reduction in ejection fraction (mean difference -0.87%, 95% CrI, -1.58 to -0.16) and a clinically meaningful impairment in global longitudinal strain (-3.08%, 95% CrI, -4.13 to -2.06). Left ventricular mass index was substantially higher in the preeclampsia group (+13.10 g/m 2 , 95% CrI, 10.06-16.21), as was relative wall thickness (+0.062, 95% CrI, 0.042-0.081), whereas fractional shortening showed no significant difference (-0.60%, 95% CrI, -2.15 to +0.86). Moderator analyses revealed that BMI and parity significantly influenced strain, whereas gestational age at diagnosis accounted for nearly all variance in ventricular mass. This mega-meta-analysis defines a remodeling phenotype of preserved ejection fraction with hypertrophic adaptation and impaired strain consistent with subclinical systolic dysfunction. Equally, it demonstrates the transformative role of LLM-based tools, showing that evidence syntheses of this magnitude can be automated, scaled, and standardized in ways previously unattainable.