Supervisor: Prof. Alfons Hoekstra
There are many potential clinical uses for mathematical models, one of which is in silico clinical trials. Traditional clinical trials are becoming increasingly expensive–—several difficulties relate to their reliance on large cohorts to provide sufficient statistical power for regulatory approval. In silico clinical trials are an emerging concept in which a cohort of virtual patients are used to reduce the required number of real participants or refine the inclusion criteria for patients in a trial.
I am part of an EU Horizon 2020 project, INSIST, developing in silico clinical trials for the treatment of acute ischemic stroke (the occlusion of an artery in the brain). My research focuses on, firstly, how we combine the diverse range of models required to model stroke injury and treatment and, secondly, how we prove the credibility of the trials such that they can be used as evidence in clinical trial submissions to regulators. In order to model stroke injury and treatment it is necessary to incorporate models for: the generation of virtual patients, arterial blood flow, blood perfusion in the brain, brain tissue death, thrombolysis (IV drug) and thrombectomy (surgical removal) treatments, and predicted functional outcome. The spatiotemporal scales and the methodologies used for these models are diverse, and include empirical, statistical, and finite element approaches. Consequently, integrating them is a challenge—one which is assisted by the expertise of clinicians who can provide physiological insight. In order to run these diverse models within a single modelling framework we use an event-based modelling approach: we model three events—pre-stroke, stroke, and treatment—and assume that the system instantaneously reaches a new steady state at each event. This allows models to be run consecutively, for each event, and avoids the need to integrate models solved at second time scales over the hour time scales between stroke onset, treatment, and final patient functional outcome. Further details on this approach can be found in this paper.
The second aspect of my research focuses on proving credibility of the trial such that it can be used clinically, which involves a high level of collaboration with clinical researchers. Assessing the credibility and uncertainty of the models we use is critical for translating research to a clinical setting, which is a long-term goal of my research. There is currently no regulatory standard for proving trial credibility, and development of an appropriate framework is still an open area of research and discussion in the in silico clinical trial community, and within regulatory organisations. In addition to performing in-depth validation studies of both individual models and the full trial, there are several other aspects to proving credibility such as analysing context of use, identifying and quantifying epistemic and aleatoric errors, and evaluating the allowable level of uncertainty in model outcome. To this end, I am performing uncertainty analyses on the trial, in which the effect of uncertainty in model inputs and parameters is determined; their effect on the uncertainty surrounding the predicted outcome is analysed; and the parameters with greatest effect identified. I perform these studies on a subset of the models included in the trial, using a quasi-Monte Carlo approach, and a sensitivity analysis, using Sobol indices, to determine the effect of unknown parameters on the outcome of interest. Results of this work for uncertainty in the blood flow model, and its effect on the predicted volume of dead tissue, can be found here.