Building a risk prediction model for anastomotic leakage postoperative low rectal cancer based on Lasso-Logistic...

To build a nomogram model for predicting the risk of anastomotic leakage (AL) postoperative low rectal cancer based on Lasso-Logistic regression.

A total of 482 patients with rectal cancer who underwent low rectal cancer surgery in our hospital from June 2017 to May 2023 were selected as the training set, and 127 patients with rectal cancer who underwent low rectal cancer surgery in our hospital from June 2023 to April 2025 were selected as the validation set. According to whether AL occurred postoperative, the patients in the training set were divided into AL group (n = 54) and N-AL group (n = 428). The data of each group were collected, and the influencing factors of AL in patients postoperative with rectal cancer in the training set were analyzed by Lasso-Logistic regression model. H-L goodness-of-fit test, ROC curve and calibration curve were used to analyze the discrimination and consistency of the model. The nomogram model was validated using the validation set. The DCA curve was used to evaluate the clinical utility of the model.

In the training set, the AL group had a higher proportion of patients with tumor stage ≥ T3 and longer operation times compared to the N-AL group; additionally, fewer AL patients had a protective stoma, and the tumor was located a shorter distance from the tumor to the anal verge than in the N-AL group. (P < 0.05). Lasso-Logistic regression analysis showed that when the penalty coefficient λ = 0.02735463, the model demonstrated good performance, gender (OR = 3.107), NRS2002 score (OR = 8.619), protective stoma (OR = 0.297), distance from tumor to anal verge (OR = 0.284), operation time (OR = 1.033) were the influencing factors of postoperative AL in low rectal cancer (P < 0.05). The 5 influencing factors were introduced into R software to establish a nomogram model for the risk of postoperative AL in low rectal cancer. The area under the ROC curve was 0.940. H-L goodness of fit test showed that there was no significant difference between the predicted value of the model and the actual observed value (χ² = 6.438, P = 0.598). The slope of the calibration curve was close to 1. The validation set showed that the nomogram model had good discrimination and consistency. The DCA curve showed that the model had high clinical utility and net benefit when the risk threshold was between 0.08 and 0.85.

Gender, NRS2002 rating, diverting ostomy, distance from tumor to anal margin, and operation time are all influencing factors of postoperative AL in low rectal cancer. The nomogram prediction model based on Lasso-Logistic regression has high consistency, discrimination and clinical application value.