Machine learning-based exceptional response prediction of nivolumab monotherapy with circulating microRNAs in non-small cell lung cancer

Abstract

Immune checkpoint inhibitors (ICIs) have significantly improved the survival of advanced non-small cell lung cancer (NSCLC). Detecting NSCLC patients with exceptional response to ICIs is necessary to improve the treatment. This case control study profiled circulating microRNA expressions of 213 NSCLC patients treated with nivolumab monotherapy to identify patients with exceptional response. Based on the response and progression-free survival, patients were divided into 3 groups: Exceptional-responder (n = 27), Resistance (n = 161), and Others (n = 25). Resistance group was further randomly partitioned into six non-overlapping sets (n = 26 or 27), while each partition was combined with Exceptional-responder and Others to make balanced datasets. We built machine learning models optimized for identifying Exceptional-responder via 3-group classification and constructed a panel of 45 microRNAs and 3 fields of clinical information. Machine learning models based on the selected panel achieved 0.81-0.89 (median 0.85) sensitivity and 0.52-0.71 (median 0.59) precision for Exceptional-responder in 3-group classification with 5-fold cross validation in all six datasets constructed, while conventional method relying on tumor PD-L1 immunohistochemistry achieved 0.44-0.44 sensitivity and 0.55-0.67 (median 0.62) precision. This study demonstrated the machine learning models achieved much higher sensitivity and accuracy in identifying Exceptional-responder to nivolumab monotherapy when comparing to conventional method only using companion PD-L1 testing.