Endometrial CD8+ T cells predict reproductive outcomes in recurrent implantation failure patients.

The recurrent implantation failure (RIF) remains a major clinical challenge in assisted reproduction. While endometrial immune dysregulation is implicated, its specific role and interaction with clinical factors are poorly defined. The lack of integrated, multimodal predictive models that combine clinical history with immune profiling limits personalized management. This study conducted a retrospective cohort study of 110 RIF patients, collecting comprehensive clinical and immune parameters. Traditional statistics and machine learning were employed to identify key predictors and build predictive models. Model interpretability was assessed using SHAP analysis, and causal pathways were explored via mediation analysis and restricted cubic splines. Previous implantation failure number was the strongest negative predictor (aOR = 0.74, 95% CI 0.60–0.91, P = 0.004). Endometrial CD8+ T cell proportion exhibited a positive, threshold−dependent effect: above 2.0%, each 1% increase raised the odds of success by 25% (aOR = 1.25, 95% CI 1.03–1.52, P = 0.025). Embryo quality was an independent positive predictor (aOR = 1.62, 95% CI 1.04–2.53, P = 0.033). Machine−learning modeling (XGBoost) achieved an AUC of 0.762 (95% CI 0.734–0.790). Mediation analysis revealed that 22.8% of the total effect of immune dysregulation on outcome was mediated through CD8+ T cells. Furthermore, the protective effect of CD8+ T cells was significantly enhanced in patients with severe immune disorder (interaction P = 0.034). This integrated clinical-immune signature underscores the pivotal, threshold-dependent role of endometrial CD8+ T cells and the cumulative burden of previous failures in RIF. The internally validated machine-learning model offers a prognostic tool, and the elucidated CD8+ T cell-mediated pathway suggests a target for immunomodulation, advancing the precision management of RIF.