Anastomotic Leakage Risk Prediction System

Laparoscopic TME · Low Rectal Cancer · Multi-center RCT Data (PP Population, n = 573)

AUC = 0.722  |  NPV = 96.3%
Threshold = 8.5%  |  PP population n = 573
About This Model
This tool is built on multi-center RCT data (n = 573). Six core features were selected by LASSO regression and fitted into a logistic regression model (cross-validated AUC = 0.722, Brier = 0.075). The Youden-optimal threshold is 8.5%; predicted probability ≥ 8.5% is classified as high risk.
Patient Data Entry
Strongest predictor (SHAP = 0.577, OR = 1.87). Typically 1–3 drains placed; ≥2 drains significantly increases risk.
Male sex confers ~1.33× higher AL risk due to narrow pelvic anatomy (OR = 1.33)
ICG is the only modifiable protective factor in the model (OR = 0.79); its use reduces AL risk
Continuous variable (OR = 1.23 per SD); reflects operative difficulty and tissue trauma
Continuous variable (OR = 1.59 per SD); larger tumours increase operative complexity and anastomotic tension
Continuous variable (OR = 1.29 per SD); longer operative time reflects procedural complexity
⚠ Disclaimer: This tool is intended for clinical decision support only and does not replace the judgment of a qualified clinician. All clinical decisions should be made in conjunction with the patient's individual circumstances, physician expertise, and other relevant clinical information.
📊
Enter patient data and click Calculate
The system will compute the predicted probability of anastomotic leakage within 30 days using a logistic regression model (cross-validated AUC = 0.722) and provide individualized clinical management recommendations.
6 Core Features Selected by LASSO:
▲No. of Pelvic DrainsStrongest risk factor SHAP = 0.577 OR = 1.87
▲Tumour Size (cm)Continuous, SHAP = 0.369 OR = 1.59
▲Male SexOR = 1.33, anatomical inherent risk
▼ICG UsedOnly modifiable protective factor OR = 0.79
▲Operation Time (min)Continuous, SHAP = 0.202 OR = 1.29
▲Intraop. Blood Loss (mL)Continuous, SHAP = 0.112 OR = 1.23
Study Background

This model was developed using data from a multi-center randomized controlled trial (RCT). Participants were patients undergoing laparoscopic TME for low rectal cancer (n = 573; ICG : control = 2:1). The primary endpoint was anastomotic leakage (AL) within 30 days postoperatively.

Modeling Approach

LASSO regression was applied to select 6 core features from 29 candidate variables (all 6 features retained with non-zero coefficients). Ten machine learning algorithms were compared; logistic regression achieved a cross-validated AUC of 0.722 (Brier score = 0.075). The Youden index was used to determine the optimal binary risk stratification threshold of 8.5%.