R10 Predicting direction and magnitude of bias in clinical trials

The MRC Hubs for Trials Methodology Research (HTMR) Network has one non-clinical studentship available for early 2018 entry. Below is a description of one project which can be applied for.
Please contact the named supervisor in advance of submitting your application.
Please see the HTMR website for further details, guidance and application form. https://www.methodologyhubs.mrc.ac.uk/about/phd-studentships/

Location: University of Bristol, Hub: ConDuCT II

Background: Although teams conducting randomised controlled trials (RCTs) aim to minimise bias as much as possible, some desirable features can be challenging to ensure in practice: e.g. successful blinding of all relevant parties and minimal attrition. If one of these features is either infeasible or implementation of it has failed, it is highly desirable to understand the likely effect of this on the estimated intervention effect. Such an understanding would considerably aid decision-making about healthcare policy - either based on the RCT directly or via its inclusion in a meta-analysis. ‘Meta-epidemiological’ studies offer potential to inform us about the likely magnitude and direction of bias associated with particular undesirable study characteristics. These are collections of meta-analyses in which the association of characteristics with estimated intervention effects has been studied across a large number of RCTs. However, it is uncertain how well the results of such studies are applicable to any individual trial. Ultimately, in predicting the direction and magnitude of likely biases, it would be desirable to produce bias-adjusted effect estimates. Again, these could be based on a single RCT or on a meta-analysis. Two main methods have been proposed for this in a meta-analysis context. These differ in that one makes the adjustments based on expert opinion about likely biases in individual RCTs (Turner et al, 1) while the other uses empirical evidence from meta-epidemiological studies (Welton et al, 2).

The MRC-funded COMBAT study (a collaboration between members of the MRC Biostatistics Unit in Cambridge and the ConDuCT II Hub in Bristol), has recently developed an integrated method for that combines the most attractive features of these two approaches: bias-adjustment is based on a combination of meta-epidemiological evidence and on the user’s own anticipation of both the direction and likely magnitude of bias. However, users find it difficult to predict these two things and require more guidance.

What the studentship will encompass: The primary purpose of the project is to develop methods for predicting the magnitude and direction of bias in a clinical trial. A key area of work will be the development of guidance for users of the newly developed (HTMR funded) tool for assessing risk of bias in RCTs (RoB 2.0) [available at www.riskofbias.info], which includes an optional facility of assessing the likely direction of any bias. An extensive library of examples will be collated from among our existing meta-epidemiological studies to gather insight into how large apparent biases might have arisen. The work will start by considering each individual domain of bias. The interactions between biases arising from these domains are critical, however, and further re-analyses of the meta-epidemiological data are expected to lead to further insight into the extent to which the biases are additive.

A second main area of work is to further develop methods for bias adjustment, of either an individual trial result or results from multiple trials combined in a meta-analysis. The student will focus on integrated methods, based on a combination of expert opinion and empirical evidence. Of particular interest is to use quantitative estimates of bias as the opinion component of the integrated method for bias adjustment. To achieve this the student will develop a standardised way of converting qualitative RoB judgements (e.g. “high risk of bias in the direction of overestimation of treatment benefit”) into a numerical interquartile range estimate of bias on the ratio of odds ratio scale (e.g. “0.71-0.99”) and use this IQR as the opinion component for the COMBAT method of bias adjustment. We plan for the student to write software (e.g. a Stata command) for bias adjustments based on an integrated approach.

Details of supervision: Savović will supervise aspects of the risk of bias tool development, meta-epidemiology and epidemiology. Higgins and Jones will supervise statistical, meta-analytical methods and bias adjustment. In addition, Jonathan Sterne, Nicky Welton and Rebecca Turner (MRC BSU Hub in Cambridge) will have an advisory role

Deadline: 8 January 2018 at 4pm (GMT)

General enquiries [Email Address Removed]
Supervisor Dr Jelena Savovic, [Email Address Removed]