FDT-BioTech: A digital twin framework for developing and analyzing virtual patient cohorts to enable virtual clinical trials

The current clinical trial system is notoriously inefficient and resource intensive with only 10% of drugs that enter the system eventually being approved for patient use. This is due, at least partially, to how difficult it is to test all the possible interventions (for example, drugs or surgery) in large groups of patients to determine that they are safe and effective. Furthermore, it is not possible to vary the order, dose, and timing of any promising intervention as there are simply too many combinations and not enough resources (or patients) available to test all the options. Thus, the clinical trial system currently attempts to determining the utility of one intervention delivered in a very limited number of schedules. This project will use mathematical models built on the key characteristics of cancer to build virtual patient populations upon which a large range of interventions can be “tested” via simulation. The results of this project could be used to, for example, select the intervention strategy with the highest likelihood of significantly reducing patient mortality, enriching the patient population to those most likely to benefit from the intervention, or eliminate interventions that are unlikely to achieve FDA approval. Success in this project will result in a fundamental shift in the way clinical trials are currently designed and executed. There are three main technical objectives for this project. The first one is to the improve computational efficacy of organ-scale, biology-based mathematical models to enable high throughput screening of novel interventional strategies. Through the use of surrogate models, we will ensure that the timescales of computing the effect of a set of interventions on a virtual patient population are within acceptable timescales—all while maintaining the interpretability of the results. The second one is to develop rigorous mathematical techniques to generate a cohort of stochastic virtual patients with unique patient anatomies and physiologies with uncertainty (i.e., distributions of model parameters to capture inter- and intra-patient heterogeneity) by combining parameter distributions obtained from model calibration to the historical patient data. We will then validate this approach by reproducing the results of historical clinical trials. The third and final technical objective is to perform a virtual clinical trial that systematically tests an array of practical therapeutic interventions that vary the dose and timing of standard-of-care chemotherapies on a virtual breast cancer patient population to determine the safety and efficacy of novel therapeutic interventions. This will provide the method to perform computationally efficient virtual clinical trials at scale. By completing these technical objectives, we will provide the community with a methodology to dramatically improve the efficiency of in-person clinical trials or, even, eliminate them entirely by evaluating in silico a large range of interventions—in parallel—on representative virtual patient cohorts of the target disease. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.