
Identification and validation of plasma biomarkers for predicting postpartum pain risk using machine learning-integrated proteomics
Pak J Med Sci. 2026 Oct;42(10):2809-2820. doi: 10.12669/pjms.42.10.20485.
ABSTRACT
BACKGROUND AND OBJECTIVE: Postpartum pain (PP) remains a significant clinical challenge with poorly understood molecular mechanisms. This study aimed to identify and validate plasma biomarkers for PP through integrated proteomics and machine learning approaches.
METHODOLOGY: The study included a discovery cohort of 16 and a validation cohort of 267 women with singleton deliveries at Changxing County People's Hospital between 2022 to 2024, comprising 110 PP cases and 157 controls; the machine-learning training and internal holdout test sets included 214 and 53 participants, respectively. Plasma samples from a discovery cohort were analyzed using tandem mass spectrometry (TMT) LC-MS/MS. Differentially expressed proteins (DEPs) were functionally characterized and validated in a validation cohort by parallel reaction monitoring (PRM). Clinical risk factors were analyzed, and a Random Forest (RF) model was developed using top-performing biomarkers.
RESULTS: The analysis identified 27 DEPs (19 up-, eight down-regulated), enriched in platelet degranulation (FDR=1.17×10-6) and complement cascades (P = 1.08×10-7). PRM successfully quantified 20 of the 27 DEPs, showing strong data-dependent acquisition (DDA) -PRM correlation (R² = 0.71). TIMP1, complement components (C1QA/B/C), and platelet factor 4 (PF4) demonstrated the strongest individual predictive performance (AUC > 0.7). The RF model achieved promising internal discriminatory performance (training AUC = 0.941; internal holdout test AUC = 0.935), with TIMP1 showing the highest relative variable importance (67.8%). Within the cesarean subgroup, gestational age showed an inverse univariable association with PP risk (OR: 0.510, 95% CI 0.317, 0.822, P=0.006).
CONCLUSIONS: This study identified a preliminary plasma biomarker panel for PP, highlighting TIMP1 and complement pathways as key contributors. The RF model showed promising internal discriminatory performance for PP risk stratification, warranting external validation.
PMID:43042102 | PMC:PMC13772145 | DOI:10.12669/pjms.42.10.20485
