RACS 1 - RACS- 2023 Mark Killingback Prize Winner
Development of an Online risk Predictor for Malignant Colorectal Polyps

Purpose
There is uncertainty when determining the optimal management for malignant colorectal polyps. Whilst treatment with polypectomy alone avoids the associated morbidity of major surgery, patient selection for this conservative approach is paramount. To aid in this critical decision, a validated web-based prognostic nomogram was developed to predict the risk of an adverse outcome in endoscopically resected malignant polyps.

Methodology
All pathological details from 1,646 malignant polyps obtained from a state-based registry were reviewed. Utilising artificial intelligence-based machine learning approach, a generalised linear mixed model with Bonferroni correction was used to evaluate the potential significance of variables upon the risk of an adverse outcome from a malignant polyp. Adverse outcomes were defined as residual disease, lymphatic disease or delayed disease recurrence.

Results
The following variables were included in the final model: age, gender, polyp location (right colon, left colon, rectum), depth of tumour invasion, lymphovascular space invasion, tumour grade, associated polyp type, mismatch repair status by immunohistochemistry, and margin status. Based on receiver operating characteristic analysis, the area under the curve for the final model was 0.76. The mean accuracy of the model was calculated as 0.69 (95% CI: 0.647-0.725).

Conclusion
The nine factors used in the nomogram are readily available in clinical practice (and considered the standard of care in pathology reporting) and the risk calculator provides an easy to understand percentage probability of an adverse outcome after polypectomy including 95% confidence intervals.

It is envisioned that this web-based nomogram will assist in the difficult decision of whether individual patients with malignant colorectal polyps can be treated conservatively. This nomogram is now available at https://malignantpolyp.com/risk-calculator.