Elucidation of the Application of Blood Test Biomarkers to Predict Immune-Related Adverse Events in Atezolizumab-Treated NSCLC Patients Using Machine Learning Methods

Abstract

BackgroundDevelopment of severe immune-related adverse events (irAEs) is a major predicament to stop treatment with immune checkpoint inhibitors, even though tumor progression is suppressed. However, no effective early phase biomarker has been established to predict irAE until now.MethodThis study retrospectively used the data of four international, multi-center clinical trials to investigate the application of blood test biomarkers to predict irAEs in atezolizumab-treated advanced non-small cell lung cancer (NSCLC) patients. Seven machine learning methods were exploited to dissect the importance score of 21 blood test biomarkers after 1,000 simulations by the training cohort consisting of 80%, 70%, and 60% of the combined cohort with 1,320 eligible patients.ResultsXGBoost and LASSO exhibited the best performance in this study with relatively higher consistency between the training and test cohorts. The best area under the curve (AUC) was obtained by a 10-biomarker panel using the XGBoost method for the 8:2 training:test cohort ratio (training cohort AUC = 0.692, test cohort AUC = 0.681). This panel could be further narrowed down to a three-biomarker panel consisting of C-reactive protein (CRP), platelet-to-lymphocyte ratio (PLR), and thyroid-stimulating hormone (TSH) with a small median AUC difference using the XGBoost method [for the 8:2 training:test cohort ratio, training cohort AUC difference = −0.035 (p < 0.0001), and test cohort AUC difference = 0.001 (p=0.965)].ConclusionBlood test biomarkers currently do not have sufficient predictive power to predict irAE development in atezolizumab-treated advanced NSCLC patients. Nevertheless, biomarkers related to adaptive immunity and liver or thyroid dysfunction warrant further investigation.

Prediction of Effectiveness and Toxicities of Immune Checkpoint Inhibitors Using Real-World Patient Data

“…Most ML-based immunotherapy solutions focus on a single prediction outcome, usually effectiveness, and are restricted to one or few cancer indications and a single ICI drug. 16 , 17 Although studies have been conducted that focus on immunotherapy irAEs, 19 , 20 this research marks the first to address three significant toxicities.…”

Knowledge domains and emerging trends in immune-related adverse events from immune checkpoint inhibitors: A bibliometrics and visualized analysis

“…As of now, while some biomarkers may assist in clinical decision-making, there is no known biomarker that can accurately predict ICI-related irAEs, which may represent another deficiency in this area of study. Researchers have ample opportunities to further contribute in this area, including utilizing the latest findings from artificial intelligence, big data, and machine learning to develop effective toxicity prediction models [ 49 , 50 ]. These outcomes will provide some basis for the early prediction of irAEs and rational clinical use of ICIs.…”

The artificial intelligence and machine learning in lung cancer immunotherapy

“…Currently, there are no biomarkers that can predict the early occurrence of irAEs, and few studies have been conducted in this regard. Therefore, creating a painless, accurate and standardized prediction method is a considerable challenge that requires further research [ 68 ].…”

“…XGBoost and LASSO methods performed the best, and the AUC of XGBboost for 10 markers was 0.692. However, even after narrowing down to the combination of C-reactive protein (CRP), platelet-to-lymphocyte ratio (PLR), and thyroid-stimulating hormone (TSH), the predictive effect was still not satisfactory, despite the high consistency between the training set and the test set [ 68 ].…”