Vascular endothelial growth factor receptor (VEGFR)-targeted therapy improved the outcome of metastatic renal cell carcinoma (mRCC) patients. However, a prediction of the response to VEGFR-tyrosine kinase inhibitor (TKI) remains to be elucidated. We aimed to develop a classifier for VEGFR-TKI responsiveness in mRCC patients. Among 101 mRCC patients, ones with complete response, partial response, or ≥24 weeks stable disease in response to VEGFR-TKI treatment were defined as clinical benefit group, whereas patients with <24 weeks stable disease or progressive disease were classified as clinical non-benefit group. Clinicolaboratory-histopathological data, 41 gene mutations, 20 protein expression levels and 1733 miRNA expression levels were compared between clinical benefit and non-benefit groups. The classifier was built using support vector machine (SVM). Seventy-three patients were clinical benefit group, and 28 patients were clinical non-benefit group. Significantly different features between the groups were as follows: age, time from diagnosis to TKI initiation, thrombocytosis, tumor size, pT stage, ISUP grade, sarcomatoid change, necrosis, lymph node metastasis and expression of pAKT, PD-L1, PD-L2, FGFR2, pS6, PDGFRβ, HIF-1α, IL-8, CA9 and miR-421 (all, P < 0.05). A classifier including necrosis, sarcomatoid component and HIF-1α was built with 0.87 accuracy using SVM. When the classifier was checked against all patients, the apparent accuracy was 0.875 (95% CI, 0.782-0.938). The classifier can be presented as a simple decision tree for clinical use. We developed a VEGFR-TKI response classifier based on comprehensive inclusion of clinicolaboratory-histopathological, immunohistochemical, mutation and miRNA features that may help to guide appropriate treatment in mRCC patients.
Keywords: Machine learning; Metastatic renal cell carcinoma; Response classifier; Tyrosine kinase inhibitors; Vascular endothelial growth factor signaling.