Overall Survival of Patients with Advanced NSCLC Treated with Nivolumab Correlates with Texture Features on Pre-Immunotherapy CT Imaging and Radiotherapy History

Studies suggest that advanced NSCLC may respond better to radiotherapy (RT) combined with anti-PD-1 immunotherapy (IT) than to IT alone. Texture analysis evaluates diagnostic imaging for patterns of intensity and has been used to construct predictive models for cancers at multiple sites. We hypothesized that for patients with advanced NSCLC treated with Nivolumab monotherapy, texture features of pre-IT CT imaging can be associated with clinical outcomes and that those associations become more significant when RT history is included in a statistical model. From an IRB-approved database of patients with advanced NSCLC treated with Nivolumab, 21 patients with the longest overall survival (OS) and 20 with the shortest were selected for analysis. The last pre-immunotherapy PET CT was used for segmentation. All FDG-avid intrathoracic tumors were delineated on the CT scan per RTOG contouring guidelines. Ninety-two first-, second, and third-order texture features within the largest tumor for each patient were analyzed for association with OS. OS time was dichotomized to < 1 year vs. > 1 year. Univariate logistic regression was used to estimate odds ratio (OR), 95% confidence interval and p-value for each imaging feature. RT history prior to IT was incorporated into the model as a covariate and analyzed for impact on p-value and effect size (OR). Area-under-the-curve (AUC) was compared using RT history alone and RT history plus texture features to assess correlation with OS. Age, sex, ECOG status ≤ 2, pathology and TNM stage were non-significant on univariate analysis between cohorts. Eighteen of the 21 patients with long OS and 10 of 20 with short OS had an RT history prior to receiving IT. RT history alone had border-line p-value (0.08) but large effect (OR > 3.47). Sixteen of 92 texture features showed significant association with dichotomized OS time (p-values ranging from 0.008 to 0.035) and all exhibited large effect (OR < 0.5 or > 1.5). The addition of RT history as a covariate increased OR in 12 of 16 and decreased p-values in 10 of 16 texture features. Area-under-curve (AUC) analysis showed that considering RT history alone resulted in AUC of 0.63 (marginally significant), while the combination of RT history with texture features resulted in AUC of 0.86, a much stronger correlation with clinical outcome. The p-value for comparison of the two AUCs was 0.0017, indicating that the increase of AUC is statistically meaningful. This preliminary study suggests that texture features on pre-IT CT imaging can be associated with OS for patients with advanced NSCLC treated with Nivolumab, and that inclusion of pre-Nivolumab RT history improves model correlation with OS. Future directions include expansion of this study to the full database, correlation of texture features with molecular biomarkers, and assessment of RT timing and fractionation on treatment response.