Optimization of a preeclampsia early warning model driven by psychosocial factors: a five-year cohort study and its clinical impact

Abstract

Objective: This study aimed to optimize a preeclampsia early warning model by integrating psychosocial factors (such as prenatal psychological stress and social support) with traditional clinical indicators to construct a more precise warning tool and analyze its impact on maternal disease burden. Methods: This retrospective case–control study included maternal data from obstetric inpatients between January 2018 and December 2023 at a single institution. Controls were selected via propensity score matching (PSM) at a 1:2 ratio. A preeclampsia early warning model incorporating multiple clinical indicators, including psychosocial factors (Modified Pregnancy Stress Scale-Revised [MPSS-R] scores and social support), was constructed through multivariable logistic regression analysis, and its predictive performance was evaluated via receiver operating characteristic (ROC) curves. Results: A total of 173 preeclampsia patients and 346 controls were included. The model achieved an AUC of 0.823 (95% CI: 0.817–0.829). MPSS-R scores were significantly correlated with disease severity (p < 0.001), and notably, 42.6% of severe preeclampsia patients had significantly elevated psychological health scores (increase ≥2 points) 2–4 weeks before diagnosis. The risk stratification results revealed that the preeclampsia incidence rate in the high-risk group was 38.5%, which was significantly higher than that in the low-risk (8.5%) and medium-risk (24.6%) groups. Conclusion: This study successfully constructed and validated a preeclampsia early warning model that integrates psychosocial factors, significantly improving the accuracy of early identification and providing a new theoretical basis for personalized screening and intervention in high-risk pregnant women. Psychological health status and social support levels have important predictive value for preeclampsia occurrence. Future studies should further optimize this model and validate its generalizability in multicenter settings.