Clinical parameters-based machine learning models for predicting intraoperative hemodynamic instability in hypertensive pheochromocytomas and paragangliomas patients

PURPOSE: To create machine learning (ML) models based on inflammatory markers and coagulation parameters for predicting intraoperative hemodynamic Instability (HI) in sustained hypertensive patients with pheochromocytomas and paragangliomas (PPGLs). METHODS: 197 sustained hypertensive PPGLs patients who underwent laparoscopic or robotic-assisted surgeries were included. Univariate and multivariate logistic regression (LR) analyses were conducted to identify the independent risk factors for HI. Various ML methods were employed to construct predictive models, including random forest (RF) and support vector machine (SVM). The receiver operating characteristic (ROC) curves, decision curve analysis (DCA), calibration curve, and Hosmer-Lemeshow test were employed to assess the performance of the ML models. The SHapley Additive explanation (SHAP) method was used to explain the model by prioritizing feature importance based on their contribution to the prediction. RESULTS: The univariate and multivariate analyses revealed that the white blood cell-to-lymphocyte ratio (WLR), neutrophil-to-platelet Ratio (NPR), international normalized ratio (INR), and other clinical parameters were independent risk factors for HI (P < 0.05). The RF model exhibited the best predictive performance, with an AUC of 0.854 on the training set and 0.812 on the test set. The calibration plot and Hosmer-Lemeshow test showed the model had excellent concordance. DCA demonstrated that the predictive model was clinically practical and effective. The SHAP method identified WLR as the most critical factor contributing to the prediction. CONCLUSION: In patients with hypertensive PPGLs, inflammatory, coagulation, and other clinical parameters are correlated with a high risk of intraoperative HI. ML models have a good predictive ability for HI in patients with sustained hypertensive PPGLs.