 |
 |
Applications are invited for a PhD studentship at the University of East Anglia, Norwich. Supervisors: Professor Lee Shepstone, Professor of Medical Statistics and Dr Allan Clark, Lecturer in Medical Statistics Outline of project: A large number of diverse statistics are commonly used to quantify the effect of interventions in clinical trials. These vary, most usually, according to the nature of the outcome measure being examined. For example, a difference between groups with respect to Normally distributed variables is often captured using a difference in means or perhaps Cohen's effect Size. A difference with respect to binary variables is often expressed as an Odds Ratio or Relative Risk or Number-Needed-to-Treat. Rate ratios are used for Poisson variables. Using different statistics creates difficulties with respect to comparability between trials. How, for example, does an effect size of 0.35 compare with a NNT of 12? A general measure of effect size based upon the Mann-Whitney statistic has been proposed by Newcombe, defined as ? = P(Y>X) + ½ P(Y=X), where X and Y are the outcomes from two individuals selected at random from the intervention and control groups of an RCT. A similar approach has been suggested by Shepstone ?= P(Y>X) - P(X<Y) which can be shown to unite several existing measures of effect size. The pivotal quantity in each case is the probability P(X<Y) which has been used in reliability theory. The first strand of this project will investigate further the properties of these, and related measures of effect size, in different contexts. A second strand of this project involves the use of Responder Analysis. This idea has developed rapidly in the field of rheumatology and increasingly other fields. The approach involves categorising individuals as 'responders' or 'non-responders' based upon one or more outcome measures. Typically, this involves a process of dichotomisation and can be criticised on the potential loss of statistical power this may entail and the need for a pre-determined threshold upon which to base the dichotomisation. A proposed alternative is the use of finite mixture distributions. In this case it is assumed that the intervention arm consists of two different sub-populations, those that have responded to treatment and those that do not. A mixture model can then be constructed to estimate the parameters of the two distributions and the proportion of individuals within each group, i.e. the proportion which are responders and non-responders. The project aims to explore and develop this approach for practical use in clinical trials. Entry requirements: Applicants should have a good first degree in mathematics or statistics and, ideally, an MSc in a relevant discipline (eg medical statistics or epidemiology). An interest and experience in clinical trials would be an advantage. Those applicants whose first language is not English must demonstrate evidence of appropriate English language proficiency, normally defined as a minimum IELTS score of 7.5 (Overall Band Score) with 7.5 in all elements or equivalent. Funding Funding includes UK/EU tuition fees, maintenance expenses of £12,940 and some appropriate training costs. Further information Please contact [log in to unmask] or [log in to unmask] for further informaiton.