Case Studies:

Decision-making in clinical trials stage of drug development(statistical and machine learning methods)

Drug development is an expensive and time-consuming process with no guarantee of success. Pharmaceutical companies must weigh up the potential risks and rewards to make decisions on whether, and how, to proceed with drug candidates. We are developing a platform to assist pharmaceutical companies make such decisions. We focus on decision-making at the clinical trials stage of drug development; for instance, decisions on whether to proceed at the end of Phase I (Safety/Dose finding) or Phase II (preliminary efficacy). We are developing a Bayesian decision-theoretic framework to make decisions that maximize expected utility. The process will be to build models to quantify the probability of success of a trial or series of trials.

Optimal Decision Making in Clinical Trials(computational and modelling)

It is known that the most important type of experiment in clinical trials is Randomised Clinical Trials (RCT). RCTs have been the gold standard and will probably continue to do so as they have a very important property; they maximise power. However, during the past few years, as well as due to the unexpected COVID–19 pandemic, clinicians have started to look for more effective ways of conducting RCTs. We investigated some of the most common types of RCTs with their advantages and limitations. Then, we focused on guiding in the decision-making of pharmaceutical companies. This is usually done in one of three ways; (a) a combination of frequentist and Bayesian methods, (b) Machine Learning methods, and (c) Markov Decision Processes within the Bayesian framework.

Markov Decision Processes (MDP) to aid Phase II trial design and decision making at the end of Phase II (methodological/theoretical & clinical/applied)

It is assumed that both the Phase II and Phase III trials consist of two-arms with binary endpoints, and that interest is in assessing whether the experimental treatment has a higher probability of successful response.

We use a Markov Decision Process model to calculate the optimal decision of the combination of Phase II (binary endpoint) and Phase III (two–armed, randomised 1:1 allocation) designs. We simulate several examples to showcase how our method would perform in practice, and indicate the strengths and limitations of it.