Artificial intelligence methods in health technology assessment (HTA): efficient decision-making for allocation of pharmacological treatment strategies in subpopulations of cancer patients

Recent advances have led to the discovery of a multitude of pharmacological therapies in cancer. Many of them are targeted to small subsets of the population, for example those patients who are positive for a particular genetic biomarker. When clinical trials of targeted therapies are based on small samples of patients, their long-term effectiveness, measured for example by overall survival (OS), may be obtained with large uncertainty, in particular when the effect of a new treatment under investigation is measured relatively early before sufficiently mature data are collected.

To deliver new therapies to patients early, regulatory agencies (such as the European Medicines Agency in the EU or Food and Drug Administration in the US) have introduced flexible licensing pathways, by allowing conditional licensing based on treatment effects measured on a short-term surrogate endpoint (for example progression free survival (PFS)). Surrogate endpoints can be used to measure the effect of a new treatment earlier and with higher precision compared to a final clinical outcome, such as OS. Nevertheless, use of surrogate endpoints may bring another level of uncertainty if the surrogate relationship between the treatment effects on the surrogate and final outcomes is not properly evaluated, as decisions will rely on predictions based on such surrogate relationships.

This project will bring together a range of artificial intelligence methods to inform a complex decision-making process at the licencing and reimbursement stages by the regulatory bodies and health technology assessment (HTA) agencies (such as NICE in England and Wales). A complex decision-modelling framework informed by effectiveness parameters obtained using advanced evidence synthesis techniques will be developed to combine information from diverse sources of evidence that are needed to make robust decisions under uncertainty.

The PhD project will explore many aspects of Bayesian modelling that include quantifying uncertainty, probabilistic reasoning, decision networks and value of information in HTA decision-making, in the framework of artificial intelligence (AI) methodology. Supervised machine learning technics will support the decision modelling framework by using a range of sources of data to learn about the relationships between key parameters in the decision model. A case study in advanced colorectal cancer, a complex disease with a range of genetic variants, will be used to develop the decision-making framework.

Applications are invited from candidates with MSc in Medical Statistics, Biostatistics, Statistics, Health Economics or related discipline.