Simulation and analysis of virtual populations are increasingly being used to guide decisions in oncology drug development. While there are a range of approaches for selecting virtual populations, most rely on sampling parameter values to allow a mechanistic QSP model to match the diverse responses observed in real clinical populations. In this talk we will begin by reviewing these methodologies and how we have adapted them to specific challenges in solid-tumor oncology for matching tumor size and progression free survival data. We will describe a case study focusing on anaplastic lymphoma kinase inhibitors (ALKi), which have shown great promise in circumventing on-target resistance mutations in ALK+ non-small cell lung cancer (NSCLC). In this work our virtual population strategy specifically focuses on capturing the range of both on-target mutations and bypass signaling that can allow for resistance to ALKi therapies to emerge. Thorough sampling of potential resistance mechanisms is essential for understanding clinical response to current therapies and for strategizing about combination partners to target resistant subpopulations.

Despite promising examples like this case study, many challenges remain in selecting and validating virtual populations in oncology. In the second half of this talk we will outline some open problems for improving the quality of virtual populations. Two key questions regarding quality are 1) how many virtual patients are needed to fully capture the potential mechanistic variability implied by the model, and 2) how can we fully leverage all clinical data from multiple potential arms and studies to constrain the virtual population parameter set as much as possible. Answering the first question requires metrics that account for the interaction between prior parameter bounds, non-identifiability, and population variability within the sampling procedure. The second question hinges on properly setting up the score function to account for correlations in response that may exist between patient cohorts receiving different therapies and those correlations that may be implied by the model. We will give some background on these problems and set them up for further work during the workshop.