Applying machine learning to predict the risk of cancer cachexia in stage IV colorectal cancer patients

Abstract Background: Cancer cachexia (CC) is a multifactorial syndrome affectingadvanced cancer patients, significantly impacting their survival and quality of life. This study utilized artificial intelligence machine learning (ML) methods to evaluate the risk of cachexia in stage IV colorectal cancer (CRC) patients through clinical data, establishing a cachexia risk prediction model. Methods: We conducted a retrospective collection of data from stage IV CRC patients consecutively visiting Luhe Hospital from January 2015 to December 2022. LASSO regression was employed on this cohort to identify diverse risk variables for CC. Four MLalgorithms, namely, logistic regression model, random forest (RF), Extreme Gradient Boosting, and k-nearest neighbor algorithm, were employed to construct the predictive model. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity. Additionally, Shapley additive prediction (SHAP) values were utilized to elucidate the prediction model. Results: Out of 219 stage IV CRC patients, 119 cases (54.34%) developed malignant fluid. Among the four ML models, the RF model exhibited the most robust predictive performance, achieving the highest AUC (0.768, 95% confidence interval: 0.706–0.831). The RF model demonstrated accuracy, sensitivity, and specificity values of 0.744, 0.790, and 0.690, respectively. We employed SHAP correlation maps to explicate the influence of individual features on the output of RF prediction models. Conclusions: The RF model offers superior predictions, enhancing clinicians' ability to screen for CC, assess patient prognosis, and make informed decisions on targeted CC in stage IV cancer patients.