Development and internal validation of a nomogram to predict colorectal cancer risk in patients with diverticulitis.

AIM: Diverticulitis, considered a benign condition, has emerging evidence linking it to colorectal cancer (CRC). Current guidelines lack standardised risk stratification tools to guide post-diverticulitis surveillance. This study aims to create and validate a nomogram to predict CRC risk in patients with diverticulitis.

METHOD: This retrospective cohort study included 1546 patients diagnosed with diverticulitis at a UK tertiary hospital between January 2021 and December 2023. Patients aged ≥ 18 years who underwent endoscopic assessment following a diverticulitis episode were included. A logistic regression model with L2 regularisation and balanced class weights (C = 0.1) was developed to predict a binary outcome: high CRC risk (histologically confirmed cancer) versus low/moderate risk (no cancer), designed to guide referral for flexible sigmoidoscopy (FS).

RESULTS

The mean age of the cohort was 69.2 years, with near-equal sex distribution. CRC was identified in 42 patients (2.7%). Significant predictors of high CRC risk included older age, male sex, and CT-detected colonic wall thickening, whereas abdominal pain, PR bleeding, and diverticulosis were associated with lower risk. The model achieved a cross-validated mean AUC of 0.736 (95% CI 0.661-0.811) and full-sample AUC of 0.784. At the Youden-optimal threshold (0.523), sensitivity was 72.1% and specificity 73.8%.

CONCLUSION

We developed and internally validated a nomogram to predict CRC risk in patients with diverticulitis, using a binary classification to guide FS referral. The tool demonstrated robust discriminative ability and could reduce unnecessary colonoscopies while prioritising high-risk patients for timely evaluation. Unlike existing CRC risk models targeting asymptomatic individuals, our nomogram is tailored to diverticulitis patients. External validation in multi-centre cohorts is required before adoption into routine clinical practice.