Joint Modelling of Adverse Events and Drug Response in Electronic Health Record Databases - Biomedicine, PhD (GW4 BioMed MRC DTP)

About
Supervisory Team: Prof William Henley
Location: Exeter, Medical School

Project
Clinical decisions about choice of drug therapy should be based on an understanding of both efficacy and safety. Randomised trials provide the gold standard of evidence for making such decisions but have important limitations, particularly when considering risk of adverse events. These include insufficient sample sizes to detect rare but serious drug reactions, a relatively short duration of follow-up and exclusion of participants at greatest risk of adverse events based on criteria such as co-morbidities or age. Post-marketing surveillance of new drugs, therefore, plays a vital role in ensuring that patients receive safe treatments but brings up new challenges. An increasingly important approach is to study adverse events using longitudinal data from electronic health records (EHR). An issue with using these data, however, is the problem of confounding by indication whereby selection for treatment is based on factors, such as disease severity, that are related to prognosis. In many cases the sources of confounding are unmeasured, preventing adoption of standard adjustment methods to remove bias.

Recent advances in quasi-experimental methods make it possible, under relevant assumptions, to address directly the bias that arises when relevant confounders are omitted. One promising strategy is prior event rate ratio (PERR) adjustment. The Exeter team have developed a statistical framework for PERR adjustment and successfully applied this to EHR data on side-effects to second-line therapies for type 2 diabetes, validating the results against outcomes of randomised trials.

This project will extend the work on drug side-effects in EHRs to incorporate treatment response, by using joint models for longitudinal and time-to-event data. It is hypothesised that there may be an interdependency between longitudinal blood glucose (HbA1c) response and risk of certain side effects but this is difficult to test reliably using standard methods.

The student will develop cutting edge methods for longitudinal modelling and causal inference, using them to explore how risk of adverse events relates to drug response for commonly prescribed diabetes medications. This offers a potential framework for optimising treatment decisions for patients based on a combination of outcome and adverse event profiles. Methods that allow for combinations of repeated biomarkers (eg HbA1c), terminal events (eg death) and recurrent events (eg hypoglycaemia) will be evaluated.

The project is a collaboration between the Universities of Exeter and Bristol drawing on their combined strengths in statistical methodology development and observational data modelling. Tiilling and Hughes will provide expertise on joint modelling of longitudinal and time-to-event data. Henley and Rodgers (Exeter) will provide expertise on methods for addressing unmeasured confounding in EHR data, with further clinical support provided by the University of Exeter’s world leading diabetes research team.

Start date: October 2018

Most studentships will be 3.5 years full time or up to 7 years part-time, and can be longer where additional training is undertaken.


How to apply
APPLICATIONS OPEN ON 25 SEPTEMBER AND CLOSE AT 17:00 ON 24 NOVEMBER 2017.

IMPORTANT: In order to apply for this project, you should apply using the DTP’s online application form. More information on the application process may be found here: http://www.gw4biomed.ac.uk/projects-2/for-students/

You do NOT need to apply to the University of Exeter at this stage – only those applicants who are successful in obtaining an offer of funding from the DTP will be required to submit an application to study at Exeter.