4-year PhD Studentship: Developing, evaluating and applying weighting methods for handling selection, dropout and collider bias

Most studies in medical research suffer bias due to non-random selection into the study (participants are different to non-participants), or dropout (those who leave the study are different to those who remain). For example, participants in UK Biobank (UKBB, one of the largest and most-used studies in medical research) are healthier and wealthier than the general population of the UK (1). Similarly, for ALSPAC, those who remain in the study differ in health, education and access to GP care from those who do not (2). Another important area of application is index event bias, a particular form of selection bias that occurs when examining risk factors for prognosis of a given disease (3). Recent examples of index-event bias can be found in the epidemiology of COVID-19 (4).

The most common approach to address these biases is multiple imputation – but an alternative approach, inverse probability weighting (IPW), may often be more appropriate, in particular, where there is information about the marginal distribution of covariates in the wider population (e.g. we know the age, sex distribution in the UK from census data) . However, IPW has been less studied than imputation, and there remain challenges in its application.

Aims and Objectives

The main aim of the project is to develop IPW methods that can be applied to a wide range of studies, especially much-used resources such as UK Biobank. The main objectives are: (i) Investigate the best functional form of IPW’s weighting model. (ii) Evaluate strategies for selection of variables for the weighting model. (iii) Develop diagnostics for assessing the fit of the weighting model. (iv) Develop sensitivity analysis methods to examine robustness of results. (v) Evaluate the utility of the developed methods in real-life applications. (vi) Provide guidelines to medical researchers in the form of tutorial-style papers.

Methodology

Using real-life applications and simulation studies, the student will evaluate and compare their newly developed methods with existing methods. The student will explore ways to simulate data that captures the complexities of real-life data. High performance computing will enable the study of large, simulated datasets. All analyses will be conducted using statistical software such as Stata or R.

Alongside this methodological work, the student will apply the methods in an applied area which can be tailored to their interests. The supervisory team have experience of selection bias/index event bias in a range of areas, including selection in UKBB and ALSPAC, and index-event bias in COVID-19 and other disease prognosis studies.

Technical skills that will be developed are: cross-cutting quantitative skills including statistical methods such as IPW and multiple imputation; computer programming; design and execution of a simulation study; cleaning and manipulation of real-life data; design and execution of an analysis plan to answer real-world questions; effective communication of scientific methods and findings to different audiences (e.g., methodologists and medical researchers) via journal articles in high impact journals, and presentations at internal meetings within the Department and externally (such as national and international conferences).

Keywords

selection bias, index event bias, inverse probability weighting, dropout

How to apply for this project

This project will be based in Bristol Medical School - Population Health Sciences in the Faculty of Health Sciences at the University of Bristol.

Please visit the Faculty of Health Sciences website for details of how to apply