Supplementary Material for: Structured magnetic resonance imaging assessment improves diagnosis of pathological complete response in rectal cancer after neoadjuvant chemoradiation

Introduction: Precise prediction of pathological complete response (pCR) following neoadjuvant chemoradiotherapy (nCRT) in rectal cancer may identify candidates for non-operative management. The optimal selection of diagnostic tools is therefore of major clinical importance. Methods: Clinical, laboratory, endoscopic and radiological data of patients with rectal cancer treated with nCRT and surgery at an academic medical center from 2010 to 2020 were retrospectively collected. Pre- and post-nCRT magnetic resonance imaging (MRI) was reviewed with a structured report template and assessed by magnetic resonance imaging tumor regression grade (mrTRG). Two senior radiologists reviewed mrTRG independently to determine the inter-reader agreement. Univariate logistic regression was applied to identify parameters that predict pCR. A multivariate prediction model was developed using L1-penalized logistic regression, with performance assessed by area under the curve (AUC) in the total cohort (apparent AUC) and by cross validation (CV-AUC). Results: A total of 261 patients were identified, of whom 36 achieved pCR. Univariate analysis showed a significant correlation between post-nCRT features with pCR, including radiological T-stage (OR 0.05 [0.02-0.15], p<0.001), mrTRG (OR 0.13 [0.05-0.31], p<0.001) and endoscopic response (OR 0.17 [0.05-0.54], p=0.032). Of those, mrTRG showed the highest AUC of 0.77 with a substantial inter-reader agreement (kappa=0.71, 95% CI: 0.61 to 0.81). The multivariate predictive model selected eight pre- and post-nCRT parameters with apparent AUC of 0.84 and CV-AUC of 0.73. Conclusion: Therapy response assessed by MRI, particularly by mrTRG, strongly predicted pCR. Therefore, mrTRG should be implemented in routine assessment of rectal cancer treated by nCRT.