A Linear Discriminant Analysis Model Based on the Changes of 7 Proteins in Plasma Predicts Response to Anlotinib Therapy in Advanced Non-Small Cell Lung Cancer Patients.

Abstract

BackgroundAnlotinib is a multi-targeted tyrosine kinase inhibitor mainly targeting angiogenesis signaling. The predictive marker of anlotinib’s efficacy remains elusive. This study was designed to explore the predictive marker of anlotinib in non-small cell lung cancer (NSCLC).MethodsWe prospectively enrolled 52 advanced NSCLC patients who underwent at least one line of targeted therapy or chemotherapy between August 2018 and March 2020. Patients were divided into durable responders (DR) and non-durable responders (NDR) based on the median progression-free survival (PFS, 176 days). The Olink Immuno-Oncology panel (92 proteins) was used to explore the predictive protein biomarkers in plasma samples before treatment (baseline) and on the first treatment evaluation (paired).ResultsAt baseline, the response to anlotinib was not significantly associated with age, gender, smoke history, histology, oligo-metastases, EGFR mutations, and other clinical characteristics. The results of PFS-related protein biomarkers at baseline were all not satisfying. Then we assessed the changes of 92 proteins levels in plasma on the first treatment evaluation. We obtained a Linear discriminant analysis (LDA) model based on 7 proteins, with an accuracy of 100% in the original data and an accuracy of 89.2% in cross validation. The 7 proteins were CD70, MIC-A/B, LAG3, CAIX, PDCD1, MMP12, and PD-L2. Multivariate Cox analysis further showed that the changes of CD70 (HR 25.48; 95% CI, 4.90–132.41, P=0.000) and MIC-A/B (HR 15.04; 95% CI, 3.81–59.36, P=0.000) in plasma were the most significant prognostic factors for PFS.ConclusionWe reported herein a LDA model based on the changes of 7 proteins levels in plasma before and after treatment, which could predict anlotinib responders among advanced NSCLC patients with an accuracy of 100%. Further studies are warranted to verify the prediction performance of the LDA model.