PhD Research
I am completing my PhD at the STOR-i CDT in partnership with Roche. The project is focused on the design and analysis of basket trials. The following is a description of my PhD:
Clinical trials are lengthy and costly procedures that investigate the efficacy and safety of treatments for human health conditions. Prior to their release onto the market, these treatments undergo rigorous testing. However, only a handful of the treatments that start testing are actually deemed safe and effective to treat the targeted condition. Also, patients’ responses to treatment vary based on their intrinsic factors and thus we wish to target treatments to the patients presenting different characteristics/genetic aberrations, this is known as ‘personalized medicine’.
Recent innovation in trial designs to improve study efficiency has led to the development of basket trials in which a single therapeutic treatment is tested on several patient populations, each of which form baskets. Patients across all baskets share a common genetic marker and as such, an assumption can be made that all patients will have a homogeneous response to treatments. Information borrowing procedures utilize this assumption to draw on information regarding the response in one basket when estimating the response rate in others. This can improve power of estimates particularly in the presence of small sample sizes. By using methods such as the Bayesian hierarchical model (BHM), Calibrated Bayesian hierarchical model (CBHM), Exchangeability-nonexchangeability (EXNEX) model and a Bayesian model averaging (BMA) procedure, notable improvements in power can be achieved. However, this can come at a large cost of inflated error rates, bringing into question validity of trial conclusions.
In the first paper of my PhD (doi: 10.1002/sim.9867), I reviewed and compared the performance of the methods mentioned above, whilst also proposing a modification to the EXNEX model (mEXNEX). The standard EXNEX model fixes the borrowing probability prior to the trial; our proposed modification uses a data-driven approach to set these probabilities based on the homogeneity of the response data, measured through Hellinger distances. Through simulation, we show that in the presence of a basket with a heterogeneous response, unlike the other methods discussed, this model can control type I error rates to a nominal level whilst yielding improved power. Under different data scenarios, we also show that this method has the potential to either improve over the standard EXNEX model or perform similarly.
My current research focuses on adaptive design features in these basket trials, including the addition of a basket alongside a novel calibration procedure.
MRes PhD Proposal: Design and Analysis of Basket and Umbrella Trials
My PhD proposal focused on the use of basket and umbrella designs to improve the efficiency of clinical trials and enhance the use of personalized medicine. The definition of a basket trial is as above, whereas, umbrella trials consist of multiple treatments being tested in parallel on patients who share the same disease but present different genetic changes. My work focused on the Bayesian framework on binary endpoints. The report explored the definition of master protocols, their advantages, and where they’ve been used in practice. We then delved into the use of response-adaptive randomization to improve the number of patients benefiting from trials, as well as, information borrowing between treatment baskets in basket trials, where we explored several different Bayesian hierarchical models. The report finished by describing several areas of open research that I will continue on in my PhD studies.
MSci Dissertation: Simultaneous Inference in Clinical Trials – Simultaneous Confidence Intervals for Treatment Effects
For my undergraduate degree, I completed a dissertation in the field of medical statistics supervised by Fang Wan at Lancaster University. A more detailed description of this dissertation can be found in my blog post here. The following is the abstract provided within my report:
It is fairly common within clinical trials to test multiple hypotheses simultaneously. The simultaneous testing has negative impacts on both Type I and Type II error rates, therefore, without correction procedures, we risk drawing misleading conclusions from our analysis. Despite this, many statisticians still do not correct for multiple testing. Within this report, we focus on three correction procedures that may be applied for the construction of simultaneous confidence intervals (SCIs): Bonferroni, Sidak and Tukey pair-wise comparison intervals. Our focus is on trials with binary endpoints and hence our parameter of interest is the proportion of times a treatment was successful. We explore several methods for interval construction for proportion parameters, adapting them to the simultaneous inference case. Ultimately, through a simulation study, we conclude that the use of Bonferroni correction on the Wilson score interval gives optimal coverage and length properties, regardless of the sample size, the number of hypotheses being tested and the value of the parameter estimate. The quantile bootstrap interval would also be recommended if a large sample size is available. Results from this report will aid in improving the accuracy of SCI construction, making conclusions drawn from clinical trial analysis more precise.