Skeletal muscle gauge prediction by a machine learning model in patients with colorectal cancer

Abstract

Objectives: Skeletal muscle gauge (SMG) was recently introduced as an imaging indicator of sarcopenia. Computed tomography is essential for measuring SMG; thus, the use of SMG is limited to patients who undergo computed tomography. We aimed to develop a machine learning algorithm using clinical and inflammatory markers to predict SMG in patients with colorectal cancer.

Methods: The least absolute shrinkage and selection operator regression model was applied for variable selection and predictive signature building in the training set. The predictive accuracy of the least absolute shrinkage and selection operator model, defined as linear predictor (LP)-SMG, was compared using the area under the receiver operating characteristic curve and decision curve analysis in the test set.

Results: A total of 1094 patients with colorectal cancer were enrolled and randomly categorized into training (n = 656) and test (n = 438) sets. Low SMG was identified in 142 (21.6%) and 90 (20.5%) patients in the training and test sets, respectively. According to multivariable analysis of the test sets, LP-SMG was identified as an independent predictor of low SMG (odds ratio = 1329.431; 95% CI, 271.684-7667.996; P < .001). Its predictive performance was similar in the training and test sets (area under the receiver operating characteristic curve = 0.846 versus 0.869; P = .427). In the test set, LP-SMG had better outcomes in predicting SMG than single clinical variables, such as sex, height, weight, and hemoglobin.

Conclusions: LP-SMG had superior performance than single variables in predicting low SMG. This machine learning model can be used as a screening tool to detect sarcopenic status without using computed tomography during the treatment period.