Stoma-free Survival After Rectal Cancer Resection With Anastomotic Leakage: Development and Validation of a Prediction Model in a Large International Cohort

Abstract

Objective: To develop and validate a prediction model (STOMA score) for 1-year stoma-free survival in patients with rectal cancer (RC) with anastomotic leakage (AL).

Background: AL after RC resection often results in a permanent stoma.

Methods: This international retrospective cohort study (TENTACLE-Rectum) encompassed 216 participating centres and included patients who developed AL after RC surgery between 2014 and 2018. Clinically relevant predictors for 1-year stoma-free survival were included in uni and multivariable logistic regression models. The STOMA score was developed and internally validated in a cohort of patients operated between 2014 and 2017, with subsequent temporal validation in a 2018 cohort. The discriminative power and calibration of the models' performance were evaluated.

Results: This study included 2499 patients with AL, 1954 in the development cohort and 545 in the validation cohort. Baseline characteristics were comparable. One-year stoma-free survival was 45.0% in the development cohort and 43.7% in the validation cohort. The following predictors were included in the STOMA score: sex, age, American Society of Anestesiologist classification, body mass index, clinical M-disease, neoadjuvant therapy, abdominal and transanal approach, primary defunctioning stoma, multivisceral resection, clinical setting in which AL was diagnosed, postoperative day of AL diagnosis, abdominal contamination, anastomotic defect circumference, bowel wall ischemia, anastomotic fistula, retraction, and reactivation leakage. The STOMA score showed good discrimination and calibration (c-index: 0.71, 95% CI: 0.66-0.76).

Conclusions: The STOMA score consists of 18 clinically relevant factors and estimates the individual risk for 1-year stoma-free survival in patients with AL after RC surgery, which may improve patient counseling and give guidance when analyzing the efficacy of different treatment strategies in future studies.

FIGURE 1

Flowchart of patient inclusion.

FIGURE 2

Flexible calibration curves of the internally and temporal-validated model. A, Flexible calibration curve after internal validation. B, Flexible calibration curve after temporal validation. Discrimination represents the ability to distinguish high-risk patients from low-risk patients and is quantified by concordance statistic (c-index), in which a 0.5 represents a noninformative model and a 1 is a perfectly discriminating model. Calibration represents the agreement between the predicted risks and the observed outcome. Calibration is presented with a flexible calibration curve for the prediction of stoma-free survival and by calculating the slope and intercept. The flexible calibration curve allows the examination of calibration across a range of predicted values. A curve close to the diagonal line (ie, perfect calibration) indicates that the predicted (x-axis) and observed probabilities (y-axis) correspond well. The flexible calibration curve shows that predicted probabilities are in line with the observed probabilities across the entire risk range, indicating near-perfect calibration. The slope is ideally equal to 1 and describes the effect of the predictors in the validation sample versus the development sample. The intercept is ideally 0 and measures if the model tends to under or overestimate probability. At the bottom, the broom plot shows the distribution of the predicted probabilities for 1-year stoma-free survival in patients who did (0) and patients who did not (1) have stoma-free survival.