Skip to main content
SSO 2024 Annual Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.
Presentation Icons
Recorded for On-Demand Viewing
Recorded for On-Demand Viewing
Back

PSM

PSM Parallel Session

72: Predicting Severe Postoperative Complications After CRS-HIPEC: An Externally Validated Machine Learning Tool

Thursday, March 21, 2024
2:32pm – 2:42pm ET
Location: C105-C106
Disclosure(s):
Amir Ashraf Ganjouei, MD, MPH: No financial relationships to disclose
Sean Dineen, MD: Oncoinvent: Consultant (Ongoing)
Introduction: Available decision support tools that predict postoperative complications after cytoreductive surgery with hyperthermic intraperitoneal chemotherapy (CRS-HIPEC) are limited by small sample sizes and lack of external validation. Our objective was to develop an externally validated machine learning tool that can predict severe complications (Clavien-Dindo grade ≥ 3) after CRS-HIPEC using two large datasets.


Methods:
The dataset consisted of adult patients who underwent CRS-HIPEC at the University of Pittsburgh Medical Center (UPMC) between 2001 and 2020. This was split in an 80:20 ratio for the training and internal validation datasets, respectively. For external validation, we used the US HIPEC collaborative dataset. Three models were trained and evaluated: logistic regression, random forest, and XGBoost. Recursive feature elimination and grid search algorithms were used to optimize the models and determine the optimal number of input variables. Lastly, SHAP values were calculated to assess variable importance.



Results: A total of 37% (n=719/1955) of cases in the UPMC cohort experienced severe postoperative complications, compared to 20% (n=373/1880) in the US HIPEC collaborative dataset. Recursive feature elimination algorithm determined that the optimal performance was achieved with 20 variables. After optimization, the random forest model had the highest area under the ROC (AUROC 0.71) and precision-recall curves (AUPRC 0.55) in the internal validation cohort and was chosen as the final model (Fig. 1A). Six centers from the US HIPEC collaborative dataset were selected as the external validation cohorts as they had the most complete data. The random forest model had an AUROC that ranged from 0.60 to 0.73 (mean AUROC 0.65) in these cohorts. Finally, the most important variables associated with severe complications were a higher peritoneal carcinomatosis index, low preoperative albumin level, undergoing subtotal gastrectomy or small bowel resection, and low preoperative platelet count (Fig. 1B).


Conclusions: Using the largest cohort of CRS-HIPEC patients in the US, we developed and externally validated a machine learning model that can aid in the prediction of severe postoperative complications. Importantly, our model may facilitate informed consent, guide patient selection, and help providers anticipate complications in the postoperative setting.

Learning Objectives:

  • Upon completion, participants will be able to identify the limitations of currently available decision support tools for CRS-HIPEC.
  • Upon completion, participants will be able to describe the results concerning the performance of the models, specifically the area under the ROC and precision-recall curves, in predicting postoperative complications.
  • Upon completion, participants will be able to list the potential implications and applications of using the developed machine learning model in clinical practice.