A revolution in the evidence base

Ella Rhodes reports from the launch of the Human Behaviour Change Project.
A ‘revolutionary’ approach to behaviour change, incorporating artificial intelligence (AI) approaches, information and behavioural science, has been launched in London. The interdisciplinary Human Behaviour-Change Project (HBCP) has been funded by the Wellcome Trust for the next four years. 
 
The project will build an AI system to continually scan international research literature on behaviour change, extract key information, and use this to build and update a model of human behaviour. This will be used to answer the question: ‘What behaviour change interventions work, how well, for whom, in what setting, for what behaviours and why?’ 
Computer scientists will build this AI system, trained by behavioural scientists, to apply natural language processing to extract relevant information from scientific reports and to organise that information into an ‘ontology’ of behaviour change interventions using reasoning and machine learning. The information scientists will build and evaluate an online interface to interact with the AI system to enable users to readily access the up-to-date evidence, and get answers to their questions, with explanations of these answers that people can understand and trust. This interface is aimed to be used by policy makers, national public health bodies, NICE, academics, the Cochrane and Campbell collaborations, information scientists and people in local authorities. 
 
Professor of Health Psychology Susan Michie (University College London) explained, at a launch event in London, that this ambitious project seeks to revolutionise the approach to how behaviour change evidence is used and disseminated – to build knowledge as well as using the information to answer real-world questions. She pointed to some of the problems in behaviour change science: Around 200 publications of behaviour interventions are published each day – more than people can conceivably keep abreast of. Each study is reported variably, using different terms and measures, making it difficult to draw useful conclusions from systematic reviews. Interventions suffer as a result. Many of the studies are conducted in the USA, making it questionable how far these can be applied to different local authorities or by governments across the world. And there’s the problem of time – a huge gap between a study being published and it being included in systematic reviews that may inform policy or solve public health problems. 
 
Michie said that over the last 15 years in the field she has reached the conclusion that theoretical progress cannot be made without revolutionising the way she and her colleagues in the area do business. The team’s first test case using this system will be that of smoking cessation. Robert West, Professor of Health Psychology and Director of Tobacco Studies at UCL, explained the development of the ontology for the initial part of this project, taking into account what interventions work, what it contains and how it’s delivered, the reach and engagement of such interventions, the population an intervention is aimed at, and the setting or culture in which it operates. 
 
But how will all these disciplines work together? Professor of Health Psychology Marie Johnston (University of Aberdeen) outlined the role of behavioural scientists in developing the ontology, in effect training machines to extract information from science literature. Explaining why the project will require something beyond a simple search engine, Pól Mac Aonghusa (Senior Manager of Social Mobile and Decision Science Research for IBM Research, based in Dublin) said that the ideal system would be able to overcome ambiguities in natural language, get to the crux of what an author meant to say, interpret what certain scores or confidence intervals really mean, and relate all of this back to the organising ontology. Thankfully, Professor John Shawe-Taylor, Director of the Centre for Computational Statistics and Machine Learning at UCL, was optimistic about this, thanks to a revolution where machine learning and logic-based approaches are used together.
 
The user interface that results from this project will be developed and evaluated by, among others, James Thomas (UCL),  who is also Director of the EPPI-Centre’s Reviews Facility for the Department of Health. He said he is frustrated over the time spent looking at irrelevant evidence during systematic reviews, which increases the time it takes for good evidence to find its way to informing policy. In his role Thomas will be carrying out two evaluations to discover which parts of systematic reviews can be automated, and whether it will transform the nature of evidence synthesis in terms of the types of evidence used and the inferences developed.
 
Finally Professor Mike Kelly, Senior Visiting Fellow in the Department of Public Health (University of Cambridge) noted that the number of policy problems where behaviour change is involved is significant, yet the number of evidence-informed policies are ‘extremely limited… there is a gap between knowledge and action’. Kelly described the HBCP as ‘nothing less than a revolution in the evidence base, on a par with what happened in evidence-based medicine two decades ago.’
 
- Keep up to date with developments in the project and also news of recruitment of staff.
 

BPS Members can discuss this article

Already a member? Or Create an account

Not a member? Find out about becoming a member or subscriber