Use of CTTA to predict treatment response in patients with EGFR T790M+ NSCLC treated with osimertinib

Abstract

Background: Computed tomography textural analysis (CTTA) can be used to quantify intra-tumour heterogeneity, which is linked to adverse tumour biology. We aimed to evaluate if CTTA can predict treatment response in metastatic epidermal growth factor receptor (EGFR) mutant non-small cell lung cancer (mNSCLC) with T790M mutations (T790M+) treated with osimertinib, after prior tyrosine kinase inhibitors. Method(s): We did a retrospective analysis of 39 consecutive patients with T790M+ mNSCLC on osimertinib. Treatment response was independently assessed on CT scans by RECIST 1.1. Routine contrast-enhanced CT images were acquired for each patient pre-osimertinib. Contiguous regions of interest (ROIs) were drawn by a radiologist around all measurable lesions and encompassing the entire axial tumour volume to assess tumour heterogeneity. Acquired image data was evaluated with an in-house program developed for analysing textures using Matlab (MathWorks, Natick, MA). Four textural variables - skewness, kurtosis, entropy and standard deviation (SD) were computed based on voxel histogram statistics, quantifying the distribution and relationship of voxel gray levels in each CT image. Each variable was assessed for association with objective response rates. Patients who had partial or complete response were "responders" (Rs), those with stable disease or disease progression were "nonresponders" (NRs). Result(s): There were 23 Rs and 16 NRs (ORR 59%). Patient characteristics were similar between Rs and NRs. SD (which reflects degree of variability) was significantly higher in NRs: average SD was 35.8 (28.67, 38.91) in NRs vs. 31.2 (23.82, 33.28) in Rs; p=0.04. Except for skewness, there was a trend towards higher entropy (measure of irregularity) and lower kurtosis (peakedness of the histogram) in NRs. Average entropy was 5.93 (5.667, 6.177) in NRs vs. 5.71 (5.434, 5.888) in Rs; p=0.06, and average kurtosis was 3.27 (2.505, 4.189) in NRs vs. 4.14 (3.043, 5.223) in Rs; p=0.08. Findings were consistent with greater tumour heterogeneity in NRs compared to Rs. Conclusion(s): Texture parameters of SD, entropy and kurtosis derived from pretreatment CT images of T790M+mNSCLC may reflect tumour heterogeneity and have potential to predict for response to osimertinib.