Shaun Seaman

Contact

Shaun Seaman – shaun.seaman@mrc-bsu.cam.ac.uk

Potential PhD Project

Missing data are a very common problem in epidemiological studies.  Individuals do not respond to some survey questions, some measurements are not made (or are made but not recorded), and some individuals drop out of studies prematurely.  A commonly used solution to this problem is multiple imputation.  This involves specifying a model for the distribution of the missing values given the observed values.  The missing values are then replaced by values randomly generated from this model, thus creating a complete set of data.  This is done repeatedly, so that several imputed datasets are generated.  The analysis is carried out on each of these imputed datasets in turn and the results obtained are averaged.  This technique of creating multiple imputed datasets enables the uncertainty associated with the results to be correctly quantified.

The validity of multiple imputation relies on correct modelling of the distribution of the missing values given the observed values.  There is thus considerable interest in methods that flexibly model this distribution.  In the last few years, methods have been proposed that use highly flexible machine-learning techniques, e.g. variational auto-encoders.  These imputation methods are promising, but it is unclear how well they work for the types of missing data that arise in practice in epidemiological studies.  This project will involve studying the performance of these methods and improving them.

About the Supervisors

The primary supervisor will be me, Shaun Seaman.  I am a Senior Research Associate in the MRC Biostatistics Unit.  I have worked for many years in the fields of missing data and causal inference for observational data.  Topics of PhD projects that I have previously supervised or co-supervised include missing data arising due to death in longitudinal observation studies, and missing data and informative cluster size.

Co-supervision will be provided by Sach Mukherjee, Director of Research in Machine Learning for Biomedicine.

Here are comments written by one of my previous PhD students, Lan Wen, who is now Assistant Professor in the Department of Statistics and Actuarial Science at the University of Waterloo, Canada, and by a current PhD student, Maria Skoularidou, some of whose work I am currently supervising.

Please do get in touch with me if you are potentially interested in this project.

How to apply

For details of the MRC BSU application process please see How to apply

To be considered for funding applications need to be submitted to the University of Cambridge application system by 23:59 (GMT) on January 5th 2023