London School of Hygiene & Tropical Medicine

The three short talks will reflect on the different career trajectories the speakers have taken, drawing examples from electronic health record data, registries and randomised trials, spanning a range of clinical areas, to illustrate the importance and contribution of robust statistical methodology. Methods developed by the speakers will be outlined, focusing on the areas of missing data, time-to-event outcomes and propensity scores.

Programme

17.15 - 17:20: Welcome and introduction of the speakers by Liz Allen

17.20 – 17.50: Ruth Keogh - Time is of the essence: using observational data to estimate causal effect

17.50 – 18.20: Fizz Williamson - P.S. I love you: an ode to propensity scores

18.20 – 18.50: Jonathan Bartlett - Multiple imputation - the solution to every statistical problem?

18:50 – 19:00: Vote of thanks by James Carpenter and Chris Frost

19:00 – 20:00: Reception in the South Courtyard

Time is of the essence: using observational data to estimate causal effects by Ruth Keogh

Observational data available for health research often involve longitudinal measurements of health status and information on treatments received, alongside times at which events such as disease diagnosis and death occur. This talk will focus on methods for tackling questions about causal effects of treatments on time-to-event outcomes using such data. An overarching challenge is in how we make use of the data collected on each individual over time. Ruth will discuss some of her work in this area, including on the ‘sequential trials’ approach in which individuals contribute their data from multiple ‘time zeros’, and on methods for obtaining individualized predictions under hypothetical interventions. Examples will be given from her applied work in cystic fibrosis and organ transplantation. 

P.S. I love you: an ode to propensity scores by Fizz Williamson

Many of the most important questions in health research relate to causality: does this exposure have an effect on this health outcome? Where such questions are tackled using observational data, confounding is an ubiquitous problem, caused by differences between patients who are exposed and those who are unexposed. This talk will focus on one way of mitigating confounding bias: propensity score methods. It will outline some of the work Fizz has done in this area, touching on connections to missing data, randomised trials and applications within large scale electronic health record data.

Multiple imputation - the solution to every statistical problem? by Jonathan Bartlett

Faced with a dataset where some values are missing, multiple imputation comes up with a number of plausible values to use in place of each missing value, resulting in a series of completed imputed datasets. Each of these is then analysed and the results suitably pooled. In his talk, Jonathan will further outline the idea behind multiple imputation, and discuss his work developing and applying multiple imputation methods for tackling problems involving measurement error, missing data and causal inference.

Speakers

Elizabeth (Fizz) Williamson, Professor of Biostatistics and Health Data Science, LSHTM

Fizz Williamson, Fizz studied Mathematics at the University of Cambridge and then completed a MSc in Medical Statistics at the University of Leicester. After a PhD at LSHTM, she moved to Australia, working in positions at the Murdoch Childrens Research Institute, the University of Melbourne and Monash University. She then returned to London to take up a position at LSHTM, where she has since focused on methods to improve the use of electronic health record data in health research.

Ruth Keogh, Professor of Biostatistics and Epidemiology, LSHTM

Ruth studied Mathematics and Statistics at the University of Edinburgh, before completing an MSc in Applied Statistics and a DPhil in Medical Statistics at the University of Oxford. She worked at the Cancer Epidemiology Unit in Oxford and at the MRC Biostatistics Unit in Cambridge before moving to the LSHTM Medical Statistics Department in 2012, where she has focused on research on statistical methods for making best use of observational data, alongside applied research in a number of areas.

Jonathan Bartlett, Professor of Medical Statistics, LSHTM

Jonathan studied Mathematics at the University of Warwick before taking the MSc in Medical Statistics at LSHTM. Jonathan completed his PhD at LSHTM and worked as a Lecturer in the Department of Medical Statistics for a number of years before joining AstraZeneca’s Statistical Innovation Group in Cambridge. After a period as Reader in Statistics at University of Bath, Jonathan rejoined the School in 2022 as Professor in Medical Statistics. He maintains a blog on biostatistics at thestatsgeek.com