Machine Learning Algorithms for Predicting Mortality in Locally Advanced Colorectal Cancer Patients following Tumor Resection: An Innovative Analysis of Post-Surgery Outcomes

Background and aims: Colorectal cancer is a leading cause of cancer-related deaths worldwide,with locally advanced cases presenting a higher risk of mortality. Accurate prediction of death riskin this population can inform treatment decisions and improve clinical outcomes. Machine learningalgorithms have shown great promise in predicting patient outcomes in various medical fields.Method: In this study, we aim to utilize a machine learning algorithm-based approach to analyzepost-tumor resection outcomes in ۲۹۰ locally advanced colorectal cancer patients and predictdeath risk. We employed six machine learning models – Random Forrest, SVM (Support VectorMachine), Decision Tree, Neural Net, Naive Bays, and XG Boost. Training and testing sampleswere randomly divided in an ۸:۲ ratio.Results: The highest accuracy was in the Decision Tree model with an accuracy of ۰.۹۴۴, followedby the Random Forest model with an accuracy of ۰.۹۳۰. The XGBoost and Naive Bayesmodels showed an accuracy of ۰.۹۰۲ and ۰.۸۷۵, respectively. Finally, the lowest accuracy wasshown by the Artificial Neural Network model with an accuracy of ۰.۶۹۴. The included factorswere as follows: sex, age, tumor location, recurrence status, recurrence-free survival, surgicaltechniques, tumor stage, circumferential resection margin, distal resection margin, recurrencetype (none, local, systemic, local + systemic), complications (stenosis, obstruction, anastomosisleak, collection, rectovaginal fistula, bleeding, and impotency), and surgery duration. Risk factoranalysis was performed by the Random Forest and XGBoost models. The five most influentialrisk factors for tumor-related mortality based on the Random Forest model were recurrence-freesurvival, recurrence type, recurrence status, surgical techniques, and circumferential resectionmargin, respectively. Additionally, according to the XG Boost model, the five most influentialrisk factors were recurrence-free survival, recurrence status, age, surgery duration, and recurrencetype, respectively.Conclusion: The implications of our findings can provide valuable insights that can improveprognostication and personalized treatment strategies for this patient population. Ultimately, theintegration of these insights into clinical practice has the potential to improve clinical outcomesand extend the patients’ overall survival.

کلیدواژه ها:

machine learning ، feature analysis ، prediction ، colorectal cancer