An Error in Modeling and the Selection of Independent Radiomics Variables Need Caution

We have read with great interest the newly published article by Chen et al.,1 in which the authors developed a novel, noninvasive, radiogenomics composite biomarker for predicting programmed death ligand 1 positivity, disease response, and pneumonitis occurrence in NSCLC. Here, we like to draw attention to some important points regarding the optimal choice of parameter and variable in the development of radiomics signatures, including one error in the original article from the perspective of radiologists.