Pre-conference Courses & Plenaries

On Tuesday 16th July we have two pre-conference courses available!

These are available to book with registration of a main conference ticket. Click the titles to read their abstracts.

An Introduction to Diagnostic Tests and Studies of Diagnostic Tests 

Tuesday 16th am - Jon Deeks, Alice Sitch, and Yemisi Takwoingi

This pre-conference course introduces the concepts of early test evaluation, test accuracy and reviews of test accuracy. Members of Biostatistics, Evidence Synthesis, Test Evaluation and prediction Models (BESTEAM) at the University of Birmingham, funded by the NIHR Birmingham BRC, will go through the fundamental processes and analyses from the early development of a test or biomarker to reviewing studies evaluating diagnostic test accuracy.


The first talk will introduce the studies that can be done prior to estimating test accuracy, particularly studies of test variability. The design and analysis of test variability studies will be introduced as well as considering when this evidence is sufficient to stop investment in a test that may not be fit for purpose.


In the second talk, we consider studies evaluating the accuracy of diagnostic tests. Diagnostic test accuracy studies will be introduced with reference to study design, regulation and transparency recommendations made in the Royal Statistical Society Working Group on Diagnostic Tests report.


The third and final talk will provide an overview of the process of conducting diagnostic test accuracy systematic reviews, with a focus on Cochrane reviews. The steps in undertaking a Cochrane review will be considered, from developing the question to interpreting the analysis, with appropriate signposting to resources including the Cochrane Handbook for Systematic Reviews of Diagnostic Test Accuracy.




10:00-10:50 Early Test Evaluation (Alice)

BREAK – 10 minutes

11:00-11:50 Studies of Test Accuracy (Jon)

BREAK – 10 minutes

12:00-12:50 Overview of Diagnostic Test Accuracy Systematic Reviews (Yemisi)

An Introduction to Risk Prediction Models & Sample Size Calculations for Development & Validation

Tuesday 16th pm - Richard Riley

Clinical prediction models are used to estimate an individual’s risk of a health-related outcome to help guide patient counselling and clinical decision making. The models are often developed using regression approaches or methods attributed to artificial intelligence (AI) and machine learning, which map predictor values to outcomes at the individual level. Examples include QRISK, which is widely used in the UK during primary care consultations to estimate a person’s 10-year cardiovascular disease risk.

However, sadly most published models are not fit for purposes, with concerns of overfitting, instability, poor validation performance, and incomplete reporting, amongst other aspects.

In this course, participants will be provided with an introduction to clinical prediction models and how to improve standards. A pathway will be described from model development to model evaluation and impact assessment.  To help improve standards, we then focus on sample size calculations for model development and evaluation. We introduce the theory behind the approaches and showcase the pmsampsize and pmvalsampsize packages in Stata and R that implement them. Two hands-on practicals are included, and participants will be supported by a dedicated and experienced faculty.

*** Rough agenda ***

Welcome - 5 mins (Richard)

Lecture (1) intro to prediction models and why SS is important  (Richard - 25 mins) 

Lecture (2) SS for development (Joie - overview of our approach and pmsampsize - 30 mins)

Practical  (faculty) - pmsampsize (30 mins)

BREAK - 10 mins

Lecture (3) Key performance measures (Lucy 25 mins) 

Lecture (4) SS for validation (Joie - overview of our approach and pmvalsampsize - 25 mins)

Practical (faculty) - pmvalsampsize (30 mins) 

We have four exciting plenaries to attend on the 17th and 18th July!

Richard Riley, Aad van Moorsel, Laura Bonnett, and Nils Braakmann will be speaking - details for each plenary are added below. Click on the title to read their abstracts.

Predicting the Unpredictable
Wednesday 17th - Laura Bonnett

Will it rain today? When will this head cold clear up? Can I cross this road without being hit by that oncoming car? These are all questions we might ask ourselves on a regular basis. If you have a long-term medical condition such as asthma or epilepsy you might also want to know when you might have your next asthma attack or epileptic fit, or indeed how many episodes you might have in a particular timeframe.

Whilst we cannot answer any of these questions with certainty, we can use statistical methodology to make an informed judgement as to each of these outcomes. Within medicine, we use clinical prediction models to evaluate the chance of a particular medical event given characteristics about a patient with an underlying condition.

This talk will take you on a journey through a selection of prediction modelling adventures so you can decide whether it is possible to predict the unpredictable!

Trust versus Trustworthiness in AI-based Systems
Wednesday 17th - Aad van Moorsel

In this presentation we present recent interdisciplinary research in trust and trustworthiness of AI-based services and systems.  

The dilemma we consider is that trusted systems are not always objectively worthy of trust from end users; reversely, systems that are objectively trustworthy may not always be trusted.  Ideally we want trust to be justifiably placed in AI-based systems.  However, modern-day AI-based services challenge existing approaches to design trustworthy computer and information systems, for instance because of probabilistic outcomes generated by black box, non-explainable models. 

The presentation is based on research projects in which researchers from diverse disciplines (law, computer science, ethics, culture, etc) join forces to consider how to protect citizens against the risks of using modern-day AI-based services.

Understanding crime from the street up: Crime maps meet causal inference
Thursday 18th - Nils Braakmann

Crime varies substantially at low spatial levels. The UK is unusual in providing high quality geocoded data on criminal activity, which when combined with modern methods of causal inference, allow us to investigate which factors drive these differences and to answer important public policy questions: 

Does stop and search work as a crime-fighting tool?
Is the location of crime driven by criminals’ expectations on the value of loot?
What are the effects of gentrification on crime?
Do high-street closures indeed cause unrest and decline in city centres? 

The talk will discuss current and ongoing work on the micro geography of crime. 

Perspectives on a career in academia: uncertainty, variability and confidence
Thursday 18th - Richard Riley

Statisticians are often under-valued in academia and medical research, and navigating a rewarding career can be challenging. 

In this talk, I explain the ups and downs of my career, and describe how I have (tried to!) shape my path whilst staying true to my values both personally and professionally. I discuss ways to shape your own career path as a methodologist, and outline the importance of understanding your strengths and weaknesses. I'll give some fun examples of the mistakes I have made and criticisms received, whilst highlighting the positive and successful avenues that teaching and research collaborations can bring. 

Above all, I will champion the role of statisticians and emphasise their critical importance in this modern era of data science, AI and machine learning.