Development of a prognostic multiomic biomarker of progression-free survival in advanced non-small cell lung cancer patients treated with first line immunotherapy

While tumor expression of PD-L1 is useful for predicting NSCLC tumor response to anti-PD-1/PD-L1 therapy (programmed cell death protein 1 (PD-1)/programmed cell death protein ligand 1 (PD-L1)), it is not entirely reliable for therapy response prediction. Here we employ a multiomic phenotypic approach, combining radiomic, clinical and pathologic biomarkers, to develop a radiomic signature for therapy response prediction on conventional pre-therapy imaging. Tumor radiomics were extracted from baseline CT imaging on retrospective cohort (n = 243) of stage 4 NSCLC patients [squamous (n = 33); non-squamous (n = 210)] undergoing first-line anti-PD-1 therapy. Image parameter heterogeneity was mitigated using nested ComBat harmonization. A novel multiomic graph was developed by combining the constituent radiomics and radiological (SUV_max, longest tumor diameter at baseline) and pathological (PD-L1, STK11 and KRAS expression) graphs. The multiomic phenotypes, identified from this graph, were then combined with clinical variables (smoking status, BMI) into a multiomic graph model. The prognostic performance of this model is compared to the combination clinical model, built by concatenation of the various "omics" variables. For the 210 patients' group, the progression-free survival models yielded the following c-statistics- clinical model: 0.58 (95% CI 0.52-0.61); combination clinical model: 0.68 (95% CI 0.58-0.69); multiomic graph clinical model: 0.71 (95% CI 0.61-0.72). The AIC values for the multiomic graph clinical model, combination clinical model and clinical model are 1278.4, 1284.1 and 1289.6 respectively, showing that the multiomic graph clinical model provides the best fit to the data. This study has constructed a novel "multiomic" signature for prognosis of NSCLC patient response to immunotherapy. The multiomic phenotypes identify patient groups based on various descriptors, thus are well-rounded, and enhance precision in progression-free survival prediction, upon combination with established clinical biomarkers.