Abstract
Background/Objectives: Preeclampsia (PE) and fetal growth restriction (FGR) are placenta-mediated pregnancy complications that may coexist but are not metabolically equivalent. We sought a compact, interpretable urinary metabolic signature associated with these phenotypes. Methods: Urinary ¹H NMR spectra from 42 pregnant women (20 Healthy, 3 isolated FGR, 7 PE, 12 PE+FGR) were aligned, total-area normalized, screened by PLS-LDA/VIP, and reduced to ten locked resonance windows for quantitative testing. Reduced panels were evaluated by repeated nested cross-validation and 999 conditional label permutations; representation and same-series sensitivity analyses were also performed. Results: Citrate, alanine, and glycine formed Core3 and showed the strongest resonance-level effects after FDR correction (q≤0.0025). For Healthy versus PE_ANY, Core3 achieved median balanced accuracy 0.820 [IQR 0.795–0.820] and median AUC 0.887 [0.862–0.896]. Because repeated-CV IQRs quantify repartition stability rather than patient-level sampling uncertainty, a fixed out-of-fold reference realization was also reported: AUC 0.868, patient-level bootstrap 95% CI 0.734–0.971, and conditional permutation p=0.001. Algorithm-matched Core3 PLS-LDA yielded median AUC 0.914 versus 0.864 for the full-resolution spectrum. Expanded5 reached AUC 0.949 but remained exploratory. A 0.02-ppm representation yielded AUC 0.895 and retained stable citrate, alanine, and glycine bins. Conclusions: Core3 is a compact candidate urinary signature for targeted validation. Complete-cohort performance may reflect both phenotype and acquisition series because these factors were non-orthogonal; therefore, the reported metrics are internal discriminatory estimates rather than diagnostic accuracy.

This publication has 15 references indexed in Scilit: