Enhancing Prediction for Tumor Pathologic Response to Neoadjuvant Immunochemotherapy in Locally Advanced Esophageal Cancer by Dynamic Parameters from Clinical Assessments

Simple Summary Neoadjuvant immunochemotherapy (NICT) has demonstrated impressive short-term efficacy, with over half of patients experiencing significant tumor shrinkage and achieving major pathological responses (MPR). These findings highlight the pressing need for further investigation into strategies for organ preservation and radiotherapy adjustments in patients who achieve MPR. Our objective was to utilize non-invasive and accessible clinical assessments to predict pathological response before surgery. By employing enhanced CT scans, esophagograms, and esophagoscopy before and after neoadjuvant treatment, we collected objective and quantitative parameters that reflected the dynamic shrinkage of tumors. Subsequently, we constructed prediction models for pathological response using multivariate logistic regression based on those dynamic parameters. These models accurately predicted pathologic complete response (pCR) (AUC 0.879) and MPR (AUC 0.912) of the primary tumor after neoadjuvant immunochemotherapy. This advancement may significantly aid informed decision-making in patient management.Abstract To develop accurate and accessible prediction methods for assessing pathologic response following NICT prior to surgery, we conducted a retrospective study including 137 patients with esophageal squamous cell carcinoma (ESCC) who underwent surgery after two cycles of NICT between January 2019 and March 2022 at our center. We collected clinical parameters to evaluate the dynamic changes in the primary tumor. Univariate and multivariate analyses were performed to determine the correlations between these parameters and the pathologic response of the primary tumor. Subsequently, we constructed prediction models for pCR and MPR using multivariate logistic regression. The MPR prediction Model 2 was internally validated using bootstrapping and externally validated using an independent cohort from our center. The univariate logistic analysis revealed significant differences in clinical parameters reflecting tumor regression among patients with varying pathologic responses. The clinical models based on these assessments demonstrated excellent predictive performance, with the training cohort achieving a C-index of 0.879 for pCR and 0.912 for MPR, while the testing cohort also achieved a C-index of 0.912 for MPR. Notably, the MPR prediction Model 2, with a threshold cut-off of 0.74, exhibited 92.7% specificity and greater than 70% sensitivity, indicating a low rate of underestimating residual tumors. In conclusion, our study demonstrated the high accuracy of clinical assessment-based models in pathologic response prediction, aiding in decision-making regarding organ preservation and radiotherapy adjustments after induction immunochemotherapy.