Machine Learning Predicts Intraoperative Hemodynamic Instability in Hypertensive PPGLs

Overview

This study developed machine learning models using clinical and laboratory parameters to predict intraoperative hemodynamic instability (HI) in patients with sustained hypertensive pheochromocytomas and paragangliomas (PPGLs). The models demonstrated effective risk stratification, highlighting key predictive factors including inflammatory and coagulation markers.

Background

Pheochromocytomas and paragangliomas (PPGLs) are catecholamine-secreting tumors that can cause severe systemic complications and intraoperative hemodynamic instability (HI) during surgical excision. Patients with sustained hypertension have a higher risk of intraoperative hypertensive crises and hypotension compared to normotensive patients. Identifying predictors of HI is critical to improving perioperative management. Machine learning (ML) offers a promising approach to analyze complex clinical data and enhance predictive accuracy for HI risk in hypertensive PPGL patients.

Data Highlights

A total of 197 patients with sustained hypertensive PPGLs undergoing laparoscopic or robotic surgery were retrospectively analyzed. Patients were split into training (70%) and test (30%) sets. Clinical data included demographics, symptoms, tumor characteristics, blood pressure measurements, catecholamine levels, and laboratory indices such as inflammatory markers (SII, PLR, MLR, WLR, NPR), coagulation parameters (APTT, PT, INR, TT), and erythrocyte indices (HCT, MCV, RDW). Preoperative adrenergic alpha-blocker use and duration were recorded. Intraoperative HI events were documented to train and validate ML models.

Key Findings

• Machine learning models incorporating clinical symptoms, tumor size, blood pressure, catecholamine levels, and laboratory inflammatory and coagulation markers effectively predicted intraoperative HI risk.
• Inflammatory indices such as systemic immune-inflammation index (SII) and neutrophil-to-platelet ratio (NPR) were significant predictors of HI.
• Coagulation parameters including activated partial thromboplastin time (APTT) and prothrombin time (PT) contributed to model performance.
• Preoperative blood pressure levels and duration of alpha-blocker preparation influenced HI risk stratification.
• Models demonstrated good generalization on independent test data, supporting their potential clinical utility.

Clinical Implications

Incorporating routine clinical and laboratory parameters into machine learning models can enhance preoperative risk assessment for intraoperative hemodynamic instability in hypertensive PPGL patients. This approach may guide anesthetic planning and perioperative management to mitigate life-threatening complications. Monitoring inflammatory and coagulation markers preoperatively could provide additional prognostic information beyond traditional clinical factors.

Conclusion

Machine learning models utilizing comprehensive clinical and laboratory data can accurately forecast intraoperative hemodynamic instability in patients with sustained hypertensive PPGLs, offering a valuable tool for personalized perioperative risk management.

References

1. Chinese PLA General Hospital Ethics Committee 2024 -- Study on ML prediction of intraoperative HI in hypertensive PPGLs