Prospective Cohort Study of Pheochromocytoma/Paraganglioma

Overview

The study will enroll patients scheduled for PPGL removal surgery at Peking Union Medical College Hospital. Before surgery, researchers will use a 6-variable model to predict the patient's risk of experiencing severe blood pressure swings during the operation. During surgery, a real-time early warning tool will be tested for its ability to accurately predict blood pressure changes 60 seconds in advance. The study will also explore the value of continuous glucose monitoring (CGM) in understanding blood pressure fluctuations and evaluate the performance of an artificial intelligence (AI) agent for preoperative anesthesia assessment, comparing its accuracy, consistency, and efficiency against that of human anesthesiologists. Participation involves no changes to the patient's standard surgical or medical care. It includes collecting clinical data, wearing a CGM sensor from the day before to the day after surgery, and having the preoperative assessment performed by both the AI agent and anesthesiologists.

Background: Resection of pheochromocytoma/paraganglioma (PPGL) carries a high risk of intraoperative hemodynamic instability (HDI) due to catecholamine release. Existing predictive models have limitations, and novel tools require prospective validation. This study aims to address this gap. Objective: The primary objectives are to: 1) Prospectively validate a previously developed 6-variable model for predicting severe HDI subtypes; 2) Assess the accuracy and timeliness of a real-time intraoperative early warning tool for HDI events; 3) Explore the association between continuous glucose monitoring (CGM) metrics and intraoperative HDI; and 4) Evaluate the clinical utility of an AI-based preoperative anesthesia assessment agent against human assessors. Methods: This is a single-center, prospective, observational cohort study at Peking Union Medical College Hospital. Eligible patients (≥18 years) scheduled for elective PPGL resection will be enrolled. Prediction Model Validation: Preoperative data (symptoms, hemoglobin, tumor functional status, epinephrine/norepinephrine elevation multiples, phenoxybenzamine dose) will be used to predict the HDI subtype (mild vs. severe). The predicted subtype will be compared against the actual subtype determined by post-hoc K-means clustering of intraoperative hemodynamic data (24 metrics). Real-time Warning Tool Validation: The tool will be used intraoperatively to predict vital signs (SBP, DBP, MAP, HR) 60 seconds ahead based on the preceding 200 seconds of data. Its predictions will be compared against actual monitored values, and its sensitivity/specificity for predicting hypertensive, hypotensive, and tachycardic events will be calculated. CGM Exploration: Patients will wear a CGM sensor from the day before surgery to one day after. Metrics like mean glucose, glycemic variability, and time in hypoglycemia/hyperglycemia will be analyzed for their association with intraoperative HDI outcomes using multivariable regression. AI Agent Evaluation: Each patient will undergo paired preoperative assessments: one by the AI agent and one by a junior anesthesiologist (≤5 years experience). A senior anesthesiologist (≥10 years experience) will provide the reference standard for risk stratification, tumor functionality, and preparation adequacy. Accuracy, inter-rater agreement (Kappa), and assessment time will be compared between the AI and junior anesthesiologist. Outcomes: Primary outcomes include the Area Under the ROC Curve (AUROC) for the prediction model, the Mean Absolute Percentage Error (MAPE) for the warning tool, the occurrence of intraoperative HDI for the CGM analysis, and the accuracy of the AI agent's risk stratification. Sample sizes have been calculated for each sub-study, with a total target enrollment of approximately 202 participants to meet all objectives.