60P Machine learning based on blood biomarkers predicts fast progression in advanced NSCLC patients treated with immunotherapy

Abstract

Fast progression (FP) represents a detrimental situation for advanced non-small cell lung cancer (NSCLC) patients undergoing immune checkpoint inhibitor (ICI) therapy. We here aimed to develop an advanced NSCLC-specific predictive framework for identification of clinical FP using machine learning (ML) methods based on blood test biomarkers. We extracted atezolizumab-treated patients from 4 multicenter clinical trials, namely OAK, BIRCH, POPLAR and FIR. FP was defined as either a ≥50% increase in the sum of largest diameters within 6 weeks or death due to disease progression within 12 weeks. The patients of the OAK trial were taken for for model training, and the patients of the other trials were used for independent validation. By using 21 pretreatment laboratory test variables, seven ML approaches including the LASSO, GLM, SVM, Decision Tree model (DT), the Random Forest model (RF), the eXtreme Gradient Boosting model (XGB), and the Generalized Boosted Regression Models (GBM), a prediction model for FP was developed. The performance was evaluated using the receiver operating characteristic (ROC) curve and the precision-recall curve. The prevalence of FP was 7.6% (118 of 1546) in all atezolizumab treated patients. The most important variables for the prediction model were: C-Reactive Protein (CRP), Neutrophil-Lymphocyte Ratio (NLR), Platelet-Lymphocyte Ratio (PLR), Lymphocyte-Neutrophil-Monocyte Ratio (LMR), Lactate Dehydrogenase (LDH), and Alanine Transaminase (ALT). The XGBoost method with 6 biomarkers demonstrated a robust performance: Area under the ROC Curve (AUC) were 0.8433, 0.6602, and 0.7978 in the training cohort (OAK) and validation cohort 1 (BIRCH) and 2 (merged POPLAR and FIR), respectively. In addition, the absolute difference in median survival between the FP and non-FP groups was significant for both progression-free survival (PFS) and overall survival (OS) (P <0.001). The identified ML method-based, 6 blood biomarkers including, the XGBoost method allow discrimination of FP in advanced NSCLC patients before ICI therapy, which should improve personalized therapeutic decision-making in ICI therapy for NSCLC.