The XGBoost machine learning model demonstrated superior discriminative performance for predicting early recurrence compared to logistic regression (AUC 0.840 vs 0.759, P = 0.004) in patients with triple-negative breast cancer and comorbid hypertension.
Cohort (n=549)
No
Does an XGBoost machine learning model accurately predict early recurrence in TNBC patients with comorbid hypertension compared to logistic regression?
An interpretable XGBoost machine learning model accurately predicts early recurrence in TNBC patients, revealing that systolic blood pressure >140 mmHg and widened pulse pressure synergistically exacerbate relapse risk.
Absolute Event Rate: 0.84% vs 0.759%
p-value: p=0.004
Abstract Triple-negative breast cancer (TNBC) patients often ential hypertension, yet how systemic vascular stress synergizes with tumor aggressiveness to drive early recurrence remains poorly understood. A cohort of 549 TNBC patients with comorbid hypertension was analyzed. Fifteen multidimensional clinical and hemodynamic features were extracted. We constructed and compared four machine learning models, employing SHapley Additive exPlanations (SHAP) on the optimal model to decode non-linear interactions. XGBoost demonstrated superior discriminative performance for predicting early recurrence (AUC = 0.840), significantly outperforming logistic regression ( P = 0.004). SHAP dependence analysis revealed profound non-linear thresholds, with relapse risk escalating exponentially when systolic blood pressure exceeded 140 mmHg. Furthermore, SHAP interaction tensors identified a robust synergistic effect between widened pulse pressure and tumor invasiveness (e.g., tumor size and lymphovascular invasion), indicating that arterial stiffness critically exacerbates malignant dissemination. These findings suggest that systemic hemodynamic stress and endothelial injury synergistically amplify the inherent aggressiveness of TNBC. The interpretable XGBoost framework provides a reliable, multidimensional risk stratification tool to guide personalized cardio-oncology management and safely triage low-risk patients.
Du et al. (Fri,) conducted a cohort in Triple-negative breast cancer with comorbid essential hypertension (n=549). XGBoost machine learning model vs. Logistic regression was evaluated on Discriminative performance (AUC) for predicting early recurrence (p=0.004). The XGBoost machine learning model demonstrated superior discriminative performance for predicting early recurrence compared to logistic regression (AUC 0.840 vs 0.759, P = 0.004) in patients with triple-negative breast cancer and comorbid hypertension.