Application of data mining for predicting hemodynamics instability during pheochromocytoma surgery

Area of Science:

• Medical Informatics
• Surgical Oncology
• Data Mining

Background:

• Pheochromocytoma surgery carries significant risks of intraoperative hemodynamic instability (IHD).
• IHD during these procedures can be life-threatening.
• Predicting and mitigating IHD is crucial for patient safety.

Purpose of the Study:

• To investigate risk factors for IHD during pheochromocytoma surgery.
• To apply data mining techniques for predictive modeling.
• To identify key attributes associated with IHD risk.

Main Methods:

• Evaluated seven data mining models including Random Forest, Naive Bayes, and logistic regression.
• Utilized Relief-F for feature selection and cross-validation for model validation.
• Assessed model performance using accuracy, Area Under the Curve (AUC), specificity, precision, recall, and F1-score.

Main Results:

• The Random Forest algorithm demonstrated high accuracy (0.8509) and AUC (0.8636).
• An improved Random Forest model achieved the highest specificity and precision, with the highest F1-score.
• Key predictors identified include body mass index, age, 24h urine VMA levels, tumor size, and enhanced CT findings.

Conclusions:

• An improved Random Forest algorithm shows promise for predicting IHD risk factors in pheochromocytoma surgery.
• Data mining offers valuable support for clinical decision-making in diagnosis, treatment, and prevention.
• This approach can enhance patient management and outcomes in high-risk surgical cases.