Lectures and seminars Biostatistics networking event: Applying Bayesian Statistics in Medical Research

The event is free and open to everyone working in, or interested in, biostatistics, bioinformatics, data science, epidemiology, and related fields. We welcome colleagues from all branches (academia/industry/government) and geographic locations. The general theme of this meeting is "Applying Bayesian Statistics in Medical Research".

Program
The program consists of a scientific part with 3 invited keynote speakers and a panel discussion. The scientific presentations will be followed by an informal networking session.  

Program outline

15.00 - 15.30: Guests will arrive and take a seat in Aula Medica 
15.30 – 15.40: Welcome and introduction
15.40 - 16.10: Probability: a somewhat mysterious mathematical object at the heart of “Bayesian statistics”: Professor emeritus Marcel Zwahlen, PhD, University of Bern, Switzerland
16.10 - 16.40: Bayesian methods for the analysis of large-scale, multi-modal, and high dimensional data to inform healthcare policy: Professor Rhiannon Owen, Swansea University Medical School, UK
16.40 - 16.50: Short break
16.50 - 17.20: Development and implementation of advanced Bayesian survival models: Dr Michael Crowther, Red Door Analytics, Stockholm, Sweden
17.20 - 17.50: Panel discussion - see below for details on topic and  discussants
17.50 - 18.00: Closing of scientific part
18:00 - 20:30 Mingle with food and drinks (Foyer, Aula Medica)
20:30 Close

About the speakers

Marcel Zwahlen is professor emeritus of epidemiology and biostatistics at the institute of social and preventive medicine (ISPM) at the University of Bern with a first degree in theoretical physics from the University of Bern and a PhD in epidemiology from John Hopkins University, Baltimore, USA. Before joining ISPM in 2003, he was head of the scientific office at the Swiss Cancer League (a charity organization) and before that head of the section of viral diseases at the Swiss Federal Office of Public Health in Bern (government office). 

He is a methodologist with a long-standing interest and experience in the analysis of health related observational and longitudinal data. He promotes the use of probabilistic and – ideally - deterministic record linkage methods to enrich existing data, and the use of Bayesian / fully probabilistic approaches for extracting or discussing the available information from existing data (hopefully of good quality).

Rhiannon Owen is Professor of Statistics at Swansea University Medical School. Her main research interests include the development and application of Bayesian methods in Health Technology Assessment, Population Health, and Health Service Evaluation. In particular, her research interests include evidence synthesis methods, analysis of large scale linked electronic health records, simulation-based methods, clinical trial evaluation, economic decision modelling, and value of information.

Michael Crowther obtained his PhD in medical statistics at the University of Leicester where he also spent many years as an academic biostatistician, rising to Associate Professor of Biostatistics, before relocating to Stockholm in 2021 where he founded Red Door Analytics. He is an expert in survival analysis and joint longitudinal-survival models, having made numerous contributions to the fields, and widely respected as a statistical software developer. He has developed and taught many training courses on his research, and is a Fellow of the UK Higher Education Academy.

Panel discussion

The panel discussion will dive into how the statistical tools we use shape scientific inference. The panel will discuss whether and how a universal, mechanistically applied method for scientific inference may derail the research process. The alternative proposition emphasises the need for tailoring our statistical models to our data and not vice versa – irrespective of whether those models are Frequentist or Bayesian. 

Discussants will be:

Rhiannon Owen (Swansea University Medical School, UK) , 
Marcel Zwahlen (University of Bern, Switzerland) ,
Michael Crowther (Red Door Analytics, Stockholm, Sweden), 
Martin Modrák (Institute of microbiology, The Czech Academy of Sciences, Czech Republic),
Vianey Leos-Barajas (Department of Statistical Sciences, University of Toronto, Canada) and 
Michael Betancourt (Freelance statistician, New York, USA). 
The session will be moderated by Simon Steiger (Department of medicine Solna, Karolinska Institutet)

Abstracts

Probability: a somewhat mysterious mathematical object at the heart of “Bayesian statistics” (Prof Zwahlen)

I will cover fundamental aspects of “probability”, discuss whether it is a concept, a mathematical object, or part of logical reasoning. I will present examples to illustrate how and why I arrived at looking very carefully at the Bayesian/fully probabilistic way of extracting information from data. Finally, I will propose to replace the terminology “Bayesian statistics” by “fully probabilistic approach”.

Development and implementation of advanced Bayesian survival models (Dr Crowther) 

In this talk I will take a brief look back (and critique) at my own use of Bayesian methods, with a focus on survival analysis. Starting from a Master’s project using multilevel Poisson models in WinBUGS, to my latest work, bringing advanced survival analysis tools to Stata, I will touch on how the lack of user-friendly (and validated) implementations are often the limiting factor to widespread adoption of new techniques. Using an example in breast cancer, I will then show how to estimate competing risks and illness-death multi-state models seamlessly within a Bayesian framework, incorporating prior information for covariate effects, and baseline hazard parameters. Previously, colleagues and I have developed likelihood-based methods for flexible parametric survival analysis, competing risks analyses and general multi-state modelling. Thanks to Stata’s -bayesmh- command, it requires relatively little amounts of developer time to sync up a user-written likelihood function, with a general MCMC engine, providing us with a user-friendly, powerful modelling framework which brings together cutting-edge modelling techniques in survival analysis, with the Bayesian paradigm, all with a single line of code. Importantly, we have also developed the predict functionality to obtain a wide range of easily interpretable predictions, such as cumulative incidence functions and (restricted) life-expectancy, along with their credible intervals.

Bayesian methods for the analysis of large-scale, multi-modal, and high dimensional data to inform healthcare policy (Prof Owen)

I will discuss the use of Bayesian methods to address the most pressing public health questions using large-scale linked electronic health records. In the context of large-scale data (such as population-scale, routinely collected, electronic health records), traditional Bayesian approaches such as Markov Chain Monte Carlo (MCMC) methods are intractable, even in the presence of supercomputers or computer clusters. This complexity is further exacerbated when fitting complex shared-parameter and/or hierarchical models. In this talk, I will discuss the complexities of fitting Bayesian models to large-scale, multi-modal, and high dimensional data using MCMC-based methods, and discuss the development and application of alternative Bayesian approaches to population-scale data. I will consider the different properties of Bayesian methods, which motivated their use, in three case studies: i) multiple long-term conditions, ii) treatment pathways and iii) environmental health. 

Registration
To register for the event, please click here to fill out a short online form. There is no cost for participation in the network event.