Explaining the Elusive Nature of a Well-Defined Threshold for Blood Transfusion in Advanced Epithelial Ovarian Cancer Cytoreductive Surgery

Abstract

There is no well-defined threshold for intra-operative blood transfusion (BT) in advanced epithelial ovarian cancer (EOC) surgery. To address this, we devised a Machine Learning (ML)-driven prediction algorithm aimed at prompting and elucidating a communication alert for BT based on anticipated peri-operative events independent of existing BT policies. We analyzed data from 403 EOC patients who underwent cytoreductive surgery between 2014 and 2019. The estimated blood volume (EBV), calculated using the formula EBV = weight × 80, served for setting a 10% EBV threshold for individual intervention. Based on known estimated blood loss (EBL), we identified two distinct groups. The Receiver operating characteristic (ROC) curves revealed satisfactory results for predicting events above the established threshold (AUC 0.823, 95% CI 0.76-0.88). Operative time (OT) was the most significant factor influencing predictions. Intra-operative blood loss exceeding 10% EBV was associated with OT > 250 min, primary surgery, serous histology, performance status 0, R2 resection and surgical complexity score > 4. Certain sub-procedures including large bowel resection, stoma formation, ileocecal resection/right hemicolectomy, mesenteric resection, bladder and upper abdominal peritonectomy demonstrated clear associations with an elevated interventional risk. Our findings emphasize the importance of obtaining a rough estimate of OT in advance for precise prediction of blood requirements.

Keywords: blood transfusion; complete cytoreduction; epithelial ovarian cancer; estimated blood loss; estimated blood volume; explainable artificial intelligence; intra-operative mapping; machine learning.

Figure A1

Examples of SHAP value Dependence plots for global explainability features showing the impact of each feature value in the risk prediction of blood transfusion at cytoreductive surgery. (A) Pre-treatment. (B) Grade. (C) PS. (D) Serous vs. non-serous pathology. PS, Performance Status.

Figure A2

Examples of SHAP value Dependence plots for global explainability features showing the impact of each feature value in the risk prediction of blood transfusion at cytoreductive surgery. (A) Operative time. (B) timing of cytoreduction. (C) Intra-operative mapping of ovarian cancer score. (D) Peritoneal Carcinomatosis Index. (E) Surgical Complexity Score. (F) Residual disease.

Figure A3

Examples of SHAP value Dependence plots for global explainability features showing the impact of surgical sub-procedures in the risk prediction of blood transfusion at cytoreductive surgery. The important features include (A) stoma formation; (B) Bladder peritonectomy; (C) Para-aortic lymphadenectomy; (D) Ileo-caecal resection/right hemicolectomy; (E) Mesenteric resection; (F) Upper abdominal peritonectomy; (G) Large bowel resection; (H) Pelvic lymph node dissection. LND, lymph node dissection.

Figure 1

Flowchart of the study cohort.

Figure 2

Performance of the XGBoost model for the risk prediction of blood transfusion at cytoreductive surgery (A) Receiver Operator Characteristic (ROC) curve. (B) Precision Recall (PR) curve.

Figure 3

(A) Summary plot showing a set of feature distribution beeswarm plots for global (threshold) explainability of 10% EBV threshold prediction. (B) Feature importance bar plot of their SHAP values. PCI, Peritoneal Carcinomatosis Index; RD, Residual Disease; PDS, Primary Debulking surgery; IDS, Interval Debulking Surgery; PS, Performance Status.

Figure 4

Kaplan Meier (KM) curve showing overall survivals between the <10% EBV and >10% EBV threshold cohorts of women undergoing cytoreduction for advanced epithelial ovarian cancer. No statistical significance was demonstrated.

Figure 5

Schematic representation of our study. According to our concept, ML-based feature selection identified operative time out of an exhaustive list of patient, disease and operation-specific features as the top feature for the risk prediction of blood transfusion at cytoreductive surgery.