EP708 - Transanal multipoint full-layer puncture biopsy for predicting rectal cancer patients response to neoadjuvant therapy

Purpose/Background: Patients who achieve a pathological complete response (pCR) after neoadjuvant chemoradiotherapy (nCRT) generally have a favorable prognosis. Accurately assessing response following nCRT may allow some patients to safely forgo surgery. Based on experience with multiple prostate biopsies, we conducted 16F core needle biopsies at a minimum of 13 locations surrounding the tumor bed or scar tissue under direct visualization to enhance pCR prediction accuracy following nCRT for LARC patients. Preliminary results indicated that transanal multipoint full-layer puncture biopsy (TMFP) significantly improved the accuracy of cCR from 74.4% to 92.6%. However, current models for predicting pCR have limited accuracy, highlighting the need for new assessment methods to enhance predictive capabilities.

Methods/Interventions: In this multicenter prospective study, we included patients with locally advanced rectal cancer (LARC) who underwent total mesorectal excision (TME) following nCRT between April 2020 and June 2023. All patients received transanal multipoint full-layer puncture (TMFP) before TME. We considered 23 factors, including TMFP, to select the model with the highest area under the curve (AUC) and applied the SHapley Additive Explanations (SHAP) method to evaluate the effects of individual characteristics on model output.

Results/Outcomes: A total of 129 LARC patients were eligible for participation, including 79 patients with non-pCR and 50 with pCR. TMFP demonstrated a sensitivity of 100%, specificity of 83.5%, positive predictive value of 79.4%, negative predictive value of 100%, and accuracy of 89.9% in predicting pCR, with no serious complications reported. TMFP showed significantly higher predictive accuracy than clinical complete response (cCR) (89.9% vs. 66.7%, χ2 = 20.527, P < 0.001). Among the ten constructed models, the adaptive boosting (AdaBoost) model performed best, achieving an AUC of 0.924 and accuracy of 0.884. Feature importance analysis revealed that the top five influential features were TMFP, distance from the anal verge, cCR, pre-chemoradiotherapy CEA levels and pre-chemoradiotherapy CA199 levels, with TMFP being the most influential. SHAP force analysis illustrated how the model can visualize personalized predictions for pCR.

Conclusion/Discussion: Our findings suggest that machine learning methods incorporating TMFP significantly enhance the prediction of pCR following nCRT in LARC patients. The proposed modified-cCR concept, which combines TMFP and cCR, could improve the accuracy of cCR in estimating pCR and support the safety of future watch-and-wait (W&W) studies.

Figure1. Photographs showing the TMFP procedure.
Photographs showing the TMFP procedure. A schematic drawing of the 13-point TMFP procedure (A). TMFP was performed under direct vision (B). The puncture tissues are applied to the paper (C). The paper with the specimen is cut off (D), and the puncture tissues are sent out for pathological examination separately (E). Histopathological positive results of TMFP, ×100 (F), ×200 (G)

Figure 2. SHAP summary plot for the top 8 clinical features contributing to the AdaBoost model. SHAP feature importance measured as the mean absolute Shapley values.
This matrix plot depicts the importance of each covariate in the development of the final predictive model (A). The attributes of the features in the model. The position on the y-axis is determined by the feature and on the x-axis by the Shapley value. The color represents the value of the feature from low to high (B).
SHAP, SHapley Additive Explanation; TMFP, transanal multipoint full-layer puncture; cCR, clinical complete response