In the presence of multiple predictive biomarkers, treatment benefits of immune-oncology therapies such as pembrolizumab may vary depending on biomarker status. However, it is not clear which patients are most likely to respond to which pembrolizumab-based combinations. Here, we describe a biomarker-based triage for study participants with advanced non-small cell lung cancer (NSCLC). Participants within groups defined by a biomarker-based classifier (gene expression profile [GEP] and tumor mutational burden [TMB]) will be randomized to receive one of three combination treatments, and the utility of each combination treatment in each biomarker-defined group will be assessed using a Bayesian adaptive randomization design. We will describe the operating characteristics of this design, including practical and theoretical challenges in the design and implementation, including the benefit of equal and unequal response-adaptive randomization, the impact of delayed response, potential delays in the introduction of certain treatment arms, the impact of the number of interim analyses performed, and controlling the risk of over-enrollment.