Predicting Severe Postoperative Complications after CRS-HIPEC: An Externally Validated Machine-Learning Tool
Amir Ashraf Ganjouei1, Jane Wang1, Christopher Yi2
1Department of Surgery, University of California, San Francisco, California, USA.
World Journal of Surgery
|April 28, 2025
Summary
We developed a machine-learning tool to predict severe postoperative complications after cytoreductive surgery with hyperthermic intraperitoneal chemotherapy (CRS-HIPEC). This model aids in patient selection and anticipating complications, improving surgical outcomes.
Area of Science:
- Oncology
- Surgical Oncology
- Medical Informatics
Background:
- Current decision support tools for predicting postoperative complications after cytoreductive surgery with hyperthermic intraperitoneal chemotherapy (CRS-HIPEC) have limitations due to small sample sizes and lack of external validation.
- Accurate prediction of severe complications (Clavien-Dindo grade ≥3) is crucial for patient management and resource allocation in CRS-HIPEC procedures.
Purpose of the Study:
- To develop and externally validate a machine-learning (ML) tool for predicting severe postoperative complications following CRS-HIPEC.
- To improve patient selection and postoperative care by providing a reliable predictive model.
Main Methods:
- Utilized a large dataset of adult patients who underwent CRS-HIPEC at the University of Pittsburgh Medical Center (UPMC) for training and internal validation (80:20 split).
- Employed the US HIPEC collaborative dataset for external validation.
- Trained and tested three ML models, using SHAP values to determine variable importance and recursive feature elimination for optimal model performance.
Main Results:
- Severe postoperative complications occurred in 37% of the UPMC cohort and 22% of the external validation cohort.
- A random forest model, optimized with 15 variables, achieved the highest area under the ROC curve (AUC) of 0.71 in internal validation and a mean AUC of 0.65 in external validation.
- Key predictors of severe complications included high peritoneal cancer index (PCI), subtotal gastrectomy, low albumin levels, small bowel resection, and high Charlson Comorbidity Index.
Conclusions:
- Developed and externally validated the largest ML model to date for predicting severe complications after CRS-HIPEC in the US.
- The model can assist clinicians in patient selection and anticipating postoperative complications, potentially enhancing patient outcomes.
