The utility of multiple clinical variables in predicting survival outcome for patients with rectal cancer: An integrated regression and retrospective cohort analysis.

person Jiasheng Zhong HaploX Biotechnology, Shenzhen, China info_outline Jiasheng Zhong, Fuming Lei, Qingkun Gao, Jingyi Shi, Zhuang Sun, Wensheng Huang, Zhichao Zhai, Changming Ding, Ke An, Shufang Wang, Jialin Zhang

Authors person Jiasheng Zhong HaploX Biotechnology, Shenzhen, China info_outline Jiasheng Zhong, Fuming Lei, Qingkun Gao, Jingyi Shi, Zhuang Sun, Wensheng Huang, Zhichao Zhai, Changming Ding, Ke An, Shufang Wang, Jialin Zhang Organizations HaploX Biotechnology, Shenzhen, China, Peking University Shougang Hospital, Beijing, China, Peking University Cancer Hospital & Institute, Beijing, China Abstract Disclosures Research Funding the Ministry of Science and Technology of the People's Republic of China National Natural Science Foundation of China Background: Prior studies have investigated clinical factors impacting the survival rates of rectal cancer patients. However, a robust and precise model for discerning these risks, which can be identified by integrated regressions analysis has yet to be established. The aim of this study is to develop and evaluate the diagnostic value of the constructed prediction model based on selected clinical variables. Methods: Patients diagnosed with stage II-III rectal cancer (April 2015-June 2021) were retrospectively enrolled as the primary cohort. Individual variables were extracted and then subjected to a K-adaptive partitioning algorithm for cut-off point determination and grouping. The Cox and least absolute shrinkage and selection operator regression (LASSO) regression were applied for independent predictors selection over significant overall survival (OS), and a nomogram was subsequently constructed. The performance of the constructed model was evaluated using the receiver operating characteristics (ROC) curves, calibration curves, and decision curve analysis (DCA), and stepwise compared with data cohorts derived from TCGA and SEER databases. Results: The Cox-LASSO analysis results considered the four independent variables as significant predictors, yielding the risk score equation: 1.086(for CEA level≥7.5 ng/ml) -1.7756(if gender is male) +1.829*(pT stages) +0.582*(pN stages). Notably, the distal resection margin (DRM) was excluded from the final model following the analysis (p = 0.060, 95%CI: -0.04-2.17). The constructed model (C-index:0.790, 95%CI: 0.673-0.743, AUC:0.873) outperformed the conventional TNM staging models (C-index:0.539, 95%CI:0.121-0.957, AUC:0.735). It stratified patients into low- or high-risk groups with significant OS differences (93% vs. 37%, p≤0.0001, respectively). The calibration curves for OS probability showed strong homogeneities between predictions and actual observations. DCA revealed positive net benefits at evaluated first, third, and fifth years. Moreover, the external validations confirmed the accurate prediction following the constructed model, with AUC of 0.745 and 0.705 for the TCGA and SEER datasets, respectively. Conclusions: This study resulted in a novel and effective prediction model for rectal cancer patients based on conventional clinical characteristics. It demonstrated solid performance and can serve as a convenient tool for OS prediction, aiding in rectal cancer patients' risk scoring and clinical decision-making.