OVA-LEAK: Prognostic score for colo-rectal anastomotic leakage in patients undergoing ovarian cancer surgery.

Objective: The objective of the present study was to define and validate an anastomotic leak prognostic score based on previously described and reported anastomotic leak risk factors (OVA-LEAK: https://n9.cl/ova-leakscore) and to establish if the use of OVA-LEAK score is better than clinical criteria (surgeon's choice) selecting anastomosis to be protected with a diverting ileostomy.

Material & Methods: This is a retrospective, multicentre cohort study that included patients who underwent cytoreductive surgery for primary advanced or relapsed ovarian cancer with colorectal resection and anastomosis between January 2011 and June 2021. Data from patients already included in the previous predictive model were not considered in the present analysis. To validate the performance of our logistic regression model, we used the OVA-LEAK formula (Annex I: https://n9.cl/ova-leakscore) for estimating leakage probabilities in a new independent cohort. Then, receiver operating characteristic (ROC) analysis was performed and area under the curve (AUC) was used to measure the performance of the model. Additionally, the Brier score was also estimated. 95% confidence intervals (CI) for each of the estimated performance measures were also calculated.

Results: 848 out of 1159 recruited patients were finally included in the multivariable logistic regression model validation. The AUC of the new cohort was 0.63 for predicting anastomotic leak. Considering a cut-off point of 22.1% to be 'positive' (to get a leak) this would provide a sensitivity of 0.45, specificity of 0.80, positive predictive value of 0.09 and negative predictive value of 0.97 for anastomotic leak. If we consider this cut-off point to select patients at risk of leak for bowel diversion, up to 22.5% of the sampled patients would undergo a diverting ileostomy and 47% (18/40) of the anastomotic leaks would be 'protected' with the stoma. Nevertheless, if we consider only the 'clinical criteria' for performing or not a diverting ileostomy, only 12.5% (5/40) of the leaks would be 'protected' with a stoma, with a rate of diverting ileostomy of up to 24.3%.

Conclusions: Compared with subjective clinical criteria, the use of a predictive model for anastomotic leak improves the selection of patients who would benefit from a diverting ileostomy without increasing the rate of stoma use.