NMR-based serum metabolite and lipoprotein profiling for endometriosis across clinically relevant and physiological comparator settings: assessment of diagnostic utility and exploratory biological signals

BACKGROUND: Reliable non-invasive biomarkers for endometriosis remain unavailable in routine practice, and their translational value depends on performance in symptomatic referral populations rather than only against healthy controls. We evaluated Nuclear Magnetic Resonance (NMR)-based serum metabolite and lipoprotein profiling for endometriosis across clinically relevant and physiological comparator settings, alongside exploratory analyses of systemic biological variation. METHODS: H-NMR-based metabolite and lipoprotein profiling. A subset also underwent cytokine profiling. Two diagnostic settings were prespecified: endometriosis versus symptomatic controls (primary) and endometriosis versus healthy volunteers (secondary). Baseline models included age and body mass index, while full models incorporated the IVDr metabolite-lipoprotein panel using elastic net regularization. Performance was assessed using fully nested repeated cross-validation and an independently processed temporal cohort. Exploratory analyses included covariate-adjusted group comparisons, weighted correlation network analysis, cytokine correlations, and paired pre-/post-operative comparisons. RESULTS: In the primary symptomatic-control comparison, the IVDr panel did not improve diagnostic performance beyond age and body mass index (AUC 0.620 vs. 0.637 for baseline). Discrimination was substantially higher in the healthy-volunteer comparison (AUC 0.994 for the full model vs. 0.882 for baseline), but this pattern was not reproduced in the temporal cohort, where performance was poor in both comparator settings. Exploratory analyses showed that the clearest biological differences were concentrated in healthy-based contrasts, with lower amino acids, creatinine, lactic acid, and selected low-density lipoprotein (LDL) measures in endometriosis. Part of the amino-acid pattern was also present in symptomatic controls, whereas particularly LDL6 lipoprotein subfractions, appeared more restricted and were supported by lipoprotein-enriched network structure. Cytokine-cytokine correlations showed reproducible within-panel immune covariance, but no cross-domain correlations remained significant after false discovery rate correction. CONCLUSIONS: While NMR-based serum metabolite and lipoprotein profiling showed strong apparent discrimination against healthy volunteers, performance was limited in the clinically relevant symptomatic-control setting, underscoring the importance of comparator spectrum for translational biomarker evaluation. Exploratory analyses identified biologically informative serum patterns, particularly a more restricted lipoprotein-subclass LDL6 signal that warrants targeted replication in clinically representative and analytically harmonized studies.