Abstract Body: INTRODUCTION: Preeclampsia (PE) is characterized by systemic endothelial dysfunction and distinct molecular phenotypes. Identification of high-precision molecular signatures, and incorporating them into diagnostic models is essential to develop modern diagnostic methods for PE. This study explores the potential for the novel biomarkers angiotensinogen (AGT) and vasorin (VASN) as diagnostic tools for PE in combination with short-FMS-like tyrosine kinase 1 (sFLT-1) and placental growth factor (PlGF) within a machine learning (ML) computational framework.
HYPOTHESIS: We hypothesized that the AGT/VASN ratio is a robust molecular signature of sPE comparable to the sFLT-1/PlGF axis, and that computational integration via ML would superiorly differentiate PE from normotensive pregnancies (NTP).
METHODS: We analyzed 72 pregnant individuals (43 PE [19 early/24 late]; 29 GA-matched NTP) quantifying VASN, AGT, sFLT-1, and PlGF via enzyme-linked immunosorbent assays (ELISA). We evaluated diagnostic precision using receiver operating characteristic (ROC) analysis, area under the curve (AUC), negative and positive predictive values (NPV/PPV), net reclassification index (NRI), and DeLong statistics. We compared conventional multi-variable regression (MLR) against three ML models (LR, RF, SVM) to assess diagnostic accuracy and potential for prospective PE detection.
RESULTS: With the exception of elevated BMI (p<0.05), and the expected difference in blood pressure, the demographics of the PE and NTP groups were not significantly different. VASN was significantly downregulated in sPE, while AGT was significantly elevated on par with PlGF and sFLT-1 in terms of D and p. AGT/VASN achieved an AUC of 0.88, demonstrating biological parity with sFlt-1/PlGF (AUC 0.89); NRI difference (n.s.), NPV and PPV identical. Combining all four biomarkers by MLR provided incremental increased precision, why LR and SVM ML models significantly enhanced predictive precision compared to sFLT-1/PlGF; LR (AUC=0.949), (NRI=0.227), (DeLong p=0.033), SVM (AUC 0.947), (NRI=0.207), (DeLong p=0.028).
CONCLUSION: The diagnostic performance of AGT/VASN is comparable that of sFLT-1/PlGF as a high-precision molecular signature of PE pathophysiology. Multi-biomarker integration with ML offers significant improvement in precision compared to individual biomarkers or their ratios. Future studies will test predictive performance in prospective cohorts of at-risk pregnant cohorts.