Prof Paddy Farrington
I joined the Statistics Department in 1998 as a lecturer, and was appointed professor in 2004. My background is in pure maths: I got my PhD in set theory in 1980. After that I went off to Vietnam where I worked at the Foreign Languages Publishing House. There I got to know some doctors specialising in infectious diseases at the Bach Mai Hospital in Hanoi; I had no idea then that I would later work on statistical methods for infectious diseases. Back in the UK I worked as a maths teacher, as circulation manager for Marxism Today, and as assistant transport officier in local government. I got a job as a statistician at the Communicable Disease Surveillance Centre in 1987 and obtained an MSc in statistics at Birkbeck college the following year. The 11 years I spent at CDSC (now part of Public Health England) largely determined my research interests in applied stats: public health, epidemiology, vaccines and infectious diseases.
Self-controlled case series method
This is a method I first published back in 1995, to investigate adverse reactions to vaccines. Its key feature is that it uses data only on cases, which makes it easy to use. It also avoids many sources of confounding, as in effect cases are matched with themselves. The method has been used in many vaccine studies (including MMR and autism), and to investigate drug safety more generally. Heather Whitaker has programmed the method in several stats packages and put together an excellent website at http://statistics.open.ac.uk/sccs. Heather and I are seeking funding to update these resources, and to program the various extensions to the basic method which we have developed over the past decade. Others who have been involved in this work include my postdocs Mounia Hocine and Patrick Musonda, and currently PhD student Yonas Weldeselassie.
Parameter estimation for infectious diseases
This involves estimating contact parameters and other quantities such as reproduction numbers. I have a very specific interest in developing statistical methods for analysing data from serological surveys. Recently, I've been seeking to develop methods for analysing multivariate serological survey data, which has led to a new research interest in time-varying frailty models. Some of the people I've worked with on this include Heather Whitaker at the OU, Richard Pebody at Public Health England (PHE), and postdocs Mona Kanaan and Steffen Unkel. Heather and I have also been involved in a large EU-funded project, POLYMOD, to collect serological and contact data.
I am involved in an MRC-funded project to update the outbreak detection system currently in use at PHE, which I helped design back in the 1990s. I've been gradually getting back into this area over the past few years. The people I've worked with include Paul Garthwaite at the OU, postdocs Steffen Unkel, Doyo Enki, and Angela Noufaily, but also Nick Andrews and Andre Charlett at PHE. Currently we are developing methods to take account of delays in laboratory reporting. Analysis of surveillance data from PHE have revealed interesting patterns which I hope also to explore further.
I am involved in severalprojects with colleagues in France, Canada and across Europe, to develop methods in pharmacoepidemiology. Heather Whitaker and I are involved in an iMi-funded consortium, ADVANCE, to develop Europe-wide methodologies for rapid benefit-risk evaluation of vaccines.
- Statistical methods in epidemiology, especially infectious diseases. Specific areas I’m interested in include estimation of vaccine efficacy, and estimation of transmission and threshold parameters such as the reproduction number. I’m most interested in methods that can be applied using commonly available data, such as surveillance reports and serological surveys.
- Statistical methods for evaluating drug safety, and in particular the self-controlled case series method (now beginning to be used quite widely). In particular, I have worked on issues such as the safety of MMR vaccine and autism.
- Generalised linear models, in particular assessing goodness of fit when data are sparse. One topic I’m interested in is goodness of fit for sparse product multinomial models.
- Statistical models for interval-censored data, including model fitting, regression diagnostics, and informative censoring.
- Profile likelihood methods for calculating approximate confidence regions.
- Frailty models, in particular time-varying frailty models.