Comments on “Machine learning and SHAP value interpretation for predicting comorbidity of cardiovascular disease and cancer with dietary antioxidants”

Cardiovascular disease (CVD) and cancer are leading causes of death worldwide, and the comorbid risk between these conditions has become an important area of public health research. Recently, a study published in Redox Biology (Volume 79, February 2025, 103470) applied machine learning models to explore the predictive value of dietary antioxidants for CVD and cancer comorbidity, making a significant contribution to this field. However, upon further examination, we found that the SHAP analysis in the paper did not reveal a significant contribution from baseline features, which raised concerns about the rationale for variable inclusion. Through independent replication analysis, we found that the SHAP results might be based on a model that omitted baseline features. A further comparison of SHAP results with and without baseline features indicated that baseline characteristics were far more important for disease prediction than dietary antioxidants. Additionally, the SHAP plots presented by the authors closely aligned with those generated from our replication of the "model without baseline features."