A risk prediction model for overall and Grade C anastomotic leakage after rectal cancer surgery

Objective: This study aimed to identify predictors of anastomotic leakage (AL), including Grade C AL, after rectal cancer surgery and to establish a risk prediction model for clinical risk stratification. Methods: A retrospective study was conducted on rectal cancer patients who underwent anterior resection (AR) at Tianjin Medical University Cancer Institute and Hospital between November 2020 and November 2024. Clinicopathological variables were analyzed, and multivariable logistic regression was applied to construct predictive models for overall and Grade C anastomotic leakage. Results: A total of 901 rectal cancer patients were included, with an AL incidence of 8.9% (80/901) and Grade C AL occurring in 4.7% (42/901). Multivariable analysis identified postoperative numerical rating scale (NRS) pain score (OR = 9.556; 95% CI, 6.014-15.184; p < 0.001), neoadjuvant chemoradiotherapy (NACRT) (OR = 3.070; 95% CI, 1.525-6.182; p = 0.002), intersphincteric resection (ISR) (OR = 4.928; 95% CI, 1.340-18.126; p = 0.016), intestinal obstruction (OR = 2.926; 95% CI, 1.105-7.748; p = 0.031), tumor size (OR = 2.238; 95% CI, 1.239-4.042; p = 0.008), operative time (OR = 2.416; 95% CI, 1.092-5.349; p = 0.030), diverting stoma (OR = 0.124; 95% CI, 0.031-0.491; p = 0.003), and gender (female vs. male) (OR = 0.410; 95% CI, 0.220-0.765; p = 0.005) as independent predictors of overall AL. For Grade C AL, NRS pain score (OR = 6.563; 95% CI, 2.565-16.791; p < 0.001) and NACRT (OR = 7.534; 95% CI, 2.012-28.216; p = 0.003) were significant predictors. The nomogram demonstrated strong discrimination, with C-statistics of 0.872 for overall AL and 0.817 for Grade C AL. NRS pain score achieved the highest individual predictive performance (AUC = 0.812 for overall AL; 0.759 for Grade C AL). Combined models integrating NRS with other variables further improved accuracy (AUC = 0.856 for overall AL; 0.817 for Grade C AL). Calibration curves showed excellent agreement between predicted and observed outcomes. Conclusion: We developed a risk prediction model for AL after rectal cancer surgery using preoperative, intraoperative, and early postoperative variables. The NRS pain score was the strongest predictor, and any unexplained rise in pain should raise suspicion of impending AL. This model offers a practical tool for early postoperative risk stratification and enhanced monitoring in high-risk patients.