Background and objectives: Proteinuria is a common complication after the application of bevacizumab therapy in patients with metastatic colorectal cancer, and severe proteinuria can lead to discontinuation of the drug. There is a lack of sophisticated means to predict bevacizumab-induced proteinuria, so the present study aims to predict bevacizumab-induced proteinuria using peripheral venous blood samples.
Methods: A total of 122 subjects were enrolled and underwent pre-treatment plasma markers, and we followed them for six months with proteinuria as the endpoint event. We then analyzed the clinical features and plasma markers for grade ≥ 2 proteinuria occurrence using machine learning to construct a model with predictive utility.
Results: One hundred sixteen subjects were included in the statistical analysis. We found that high baseline systolic blood pressure, low baseline HGF, high baseline ET1, high baseline MMP2, and high baseline ACE1 were risk factors for the development of grade ≥ 2 proteinuria in patients with metastatic colorectal cancer who received bevacizumab. Then, we constructed a support vector machine model with a sensitivity of 0.889, a specificity of 0.918, a precision of 0.615, and an F1 score of 0.727.
Conclusion: We constructed a machine learning model for predicting grade ≥ 2 bevacizumab-induced proteinuria, which may provide proteinuria risk assessment for applying bevacizumab in patients with metastatic colorectal cancer.
Keywords: Bevacizumab; Colorectal cancer; Machine learning; Proteinuria.
© 2024. The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature.