Development and Validation of an Magnetic Resonance Imaging-Based Nomogram for the Preoperative Prediction of Poor Response to Neoadjuvant Chemoradiotherapy for Locally Advanced Rectal Cancer

Abstract

Background: Neoadjuvant chemoradiotherapy (neoCRT) has been used to promote tumor regression for locally advanced rectal cancers (LARC). However, tumour regression differs from patient to patient and determines the therapeutic strategies. Thus, we aimed to develop a predictive model with pre-treatment magnetic resonance imaging (MRI) findings and clinical parameters for poor response prediction after neoCRT in LARC patients. Methods: Patients with clinicopathologically confirmed LARC (training and validation datasets, n = 100 and 71, respectively) were enrolled. Patients' clinical data were retrospectively collected. All patients underwent pre-neoCRT pelvic MRI. MRI findings were analysed; laboratory examination markers for all patients were collected before treatment. Poor regression was defined as a tumour regression grade (TRG) of 3-5. Univariate logistic regression analysis was performed to select the significant predictive features from the MRI and clinical features for the training set. A lasso regression model was used for further data dimension reduction and feature selection. Based on a multivariable logistic regression analysis, a nomogram was constructed incorporating the MRI signature and selected clinical predictors. Calibration, discrimination, and clinical usefulness of the nomogram were assessed. Findings: Eighty-eight patients (51·5%) showed poor responses (TRG 3-5). The MRI-derived signature was significantly related to tumour regression status and showed predictive performance in both datasets. The nomogram showed good discrimination, with areas under the ROC curves of 0·823 and 0·820 in the training and test sets, respectively, and good calibration in both datasets. The decision curve analysis confirmed the nomogram was clinically useful. Interpretation: A nomogram based on a pre-treatment MRI-derived signature and clinical risk factors has potential for use as a non-invasive tool to preoperatively predict poor responses in LARC patients and guide individualized treatments for an adaptive treatment policy. Funding Statement: The study was funded by the Science and Technology Planning Project of Guangdong Province, China (2016A010101013, 2017B020226004), the Science and Technology Program of Guangzhou, China (201704020060, 201807010057), and grant from the Health and Medical Collaborative Innovation Project of Guangzhou City, China (201803010021). Declaration of Interests: The authors declare that they have no conflicts of interest. Ethics Approval Statement: This study was approved by the institutional ethics committee of our hospital (Sun Yat-sen University Cancer Center) and performed in accordance with the tenets of the Declaration of Helsinki. Written informed consent was provided by each patient before treatment.