Background: PD-L1 expression and TMB, as a proxy for neoantigen burden, have been correlated with response to IO in advanced NSCLC (aNSCLC) clinical trials, but their combined utility is unclear. We assessed TMB and PD-L1 as predictors of response in aNSCLC patients (pts) after IO monotherapy in a real-world setting. Methods: Pts had a diagnosis of aNSCLC, comprehensive genomic profiling of 186-315 genes/1.1 megabase (Mb), PD-L1 testing of pre-IO specimens, and were treated in the Flatiron Health network (1/2011 - 6/2018). Clinical characteristics and real-world tumor response (rwTR) were obtained via technology-enabled abstraction of clinician notes and radiology/pathology reports, and linked to genomic data in the Flatiron Health-Foundation Medicine Clinico-Genomic Database. A general additive model examined the predictive value of TMB (as continuous measure) and PD-L1 level on rwTR. A reduced PD-L1-only model was compared to the full model using Akaike Information Criterion (AIC). rwTR predictions at representative TMB and PD-L1 levels were calculated. Results: Of 426 pts, PD-L1 expression was high (≥50%) in 140, low (1-49%) in 123, and negative (<1%) in 163. Median TMB was 9.6 mut/Mb (IQR 4.4 - 14.8) overall, 11.3 in responders and 8.7 in non-responders. TMB did not correlate with PD-L1 level (Kruskal-Wallis p=0.29). The TMB + PD-L1 model had superior prediction of rwTR than the PD-L1 model, as assessed by lower AIC score. In the combined model, higher TMB and PD-L1 levels were each associated with higher rwTR likelihood (Table). Predicted rwTR probability, % (95% CI), by TMB and PD-L1 in line 1. Conclusions: TMB and PD-L1 expression are independent markers that, when combined, have increased predictive power for response to IO. High TMB + low/neg PD-L1 behaved similarly to low TMB + high PD-L1, and high TMB + high PD-L1 predicted the highest rwTR. Investigation of these biomarkers as complementary predictors of progression and overall survival is ongoing.