The Human Behaviour-Change Project: Developing a Behaviour Change Intervention Ontology Emma Norris1,* Ailbhe Finnerty1, Marta Marques1, Robert West2, James Thomas3, Pol Mac Aonghusa4, Marie Johnston5, Michael P. Kelly6 and Susan Michie1 1 Centre for Behaviour Change, University College London, UK 2 Research Department of Behavioural Science & Health, University College London, UK 3 Institute of Education, University College London, UK 4 IBM Research, IBM, Dublin 5 Aberdeen Health Psychology Group, University of Aberdeen, UK 6 Department of Public Health & Primary Care, University of Cambridge, UK ABSTRACT years to finish (Bragge et al, 2011), often making their re- Behaviour change is essential to improve population health, the self- sults outdated by the time of publication. management of illness, chronic conditions and health professional prac- tice. Evidence about behaviour change interventions is currently being Advances in organising the fragmented evidence about produced at such a rate that manual systems for evidence review and behaviour change interventions are urgently needed to im- synthesis cannot keep up. Neither can they account for all the relevant prove our understanding of behaviour and how to change it. features of interventions. By accomplishing this we can improve our ability to devel- The Human Behaviour-Change Project (HBCP) aims to bring together behavioural scientists, computer scientists and system architects to ad- op behaviour change interventions to solve real-world prob- vance progress in behaviour change. It aims to answer variants of the ‘big lems, such as the global burden of disease and unsustainable question’ of behaviour change: ‘What works, compared with what, how climate change. well, with what exposure, with what behaviours (for how long), for whom, Recent research has developed a method for specifying in what settings, and why?’ behaviour change interventions in terms of their component The main outputs will be: 1) an ontology of behaviour change interven- tions; 2) an AI system capable of extracting and interpreting evidence from techniques e.g. The Behaviour Change Technique Taxono- published literature and making predictions; 3) an interface allowing users my v1 (BCTTv1; Michie et al., 2013) specifies 93 ‘active (researchers, policy-makers, practitioners) to access the knowledge base ingredients’ of behaviour change interventions. To fully and answer specific questions about behaviour change. understand how interventions have their effects, we need to extend this method of specification to how interventions are delivered, their reach, the target population and intervention 1 INTRODUCTION setting, the target behaviour and the mechanisms of action Behaviour change interventions are policies, activities, of the intervention. (Larsen et al., 2016). Ontologies, which services or products designed to cause people to act differ- are coherent structures for representing knowledge, have ently from how they otherwise would have done (West & been used to unify many areas of science allied to behaviour Michie, 2016). They involve enabling change amongst change, such as for mental disorders and mental functioning members of the target population (e.g. knowledge, skills, (Hastings, 2012; Larsen et al, 2016). The current pro- beliefs, feelings or habits) or their social and/or physical gramme of research seeks to develop an ontology of behav- environment, or both. Typically, the goal is to achieve iour change interventions. change that is sustained over an extended period of time (such as reducing smoking prevalence in the general popula- tion or increasing levels of habitual physical activity). Knowledge of behaviour change interventions tends to be 1.1 Introducing the Human Behaviour-Change fragmented, generated by studies with variable methods and Project (HBCP) from partial, unspecified intervention evaluation reports. The vision of the Human Behaviour-Change Project Evidence about behaviour change interventions is being (HBCP; www.humanbehaviourchange.org) is to synthesise generated at such a high rate that manual systems for evi- evidence about behaviour change interventions and develop dence review, interpretation and synthesis cannot keep up an automated knowledge system to identify patterns in the (Elliot et al., 2014). For example, systematic reviews of published literature and generate new, up-to-date evidence. health interventions currently take an average of almost 6 A collaboration of behavioural scientists, computer scien- tists and system architects to answer variants of the ‘big * To whom correspondence should be addressed: emma.norris@ucl.ac.uk question’ of behaviour change: ‘What works, compared 1 Norris et al. with what, how well, with what exposure, with what behav- iours (for how long), for whom, in what settings, and why?’ This project involves: 1. Developing an ontology of behaviour change interven- tions evaluations: including aspects of intervention (tech- niques and delivery), target population, context, mecha- nisms of action, behaviours, outcomes, study methods and reporting features. 2. Using this ontology to annotate behaviour change in- terventions evaluation reports. These annotations will be used to develop and train an automated system to extract key information from research reports. Fig 1. Top-level of Behaviour Change Intervention Ontology sce- 3. Developing and evaluating Machine Learning and au- nario and their putative interactions tomated reasoning systems to synthesise and interpret the annotated evidence and make predictions. 4. Developing and evaluating an online interface to inter- rogate the knowledge base contained within the system. ACKNOWLEDGEMENTS This project is funded by the Wellcome Trust (UK). Thanks 1.2 Top-level entities of the Behaviour Change to Janna Hastings and Julian Everett for their consultancy Intervention Ontology (BCIO) work on this project. To synthesise the fragmented evidence, an essential ele- ment of HBCP is the development of a Behaviour Change Intervention Ontology (BCIO). At the heart of this ontology is the ‘behaviour change intervention scenario’ whose top- REFERENCES level entities are shown in Figure 1. Each scenario corre- Bragge, P., Clavisi, O., Turner, T., Tavender, E., Collie, A., & Gruen, R.L. sponds to an intervention condition within an evaluation. (2011). The Global Evidence Mapping Initiative: scoping research in  Target Behaviour: behaviour that the intervention broad topic areas. BMC Medical Research Methodology. 11,92. seeks to change (e.g., 6 months of smoking abstinence) Elliot, J.H, Turner, T., Clavisi, O., Thomas, J., Higgins, J.P.T, Maber-  Intervention: set of policies, activities, services or games, C. & Gruen, R.L. (2014). Living Systematic Reviews: An products that is intended to result in a difference in the Emerging Opportunity to Narrow the Evidence-Practice Gap. PLOS target behaviour. This includes content (i.e. techniques Medicine, 11(2), e1001603. used, such as goal-setting or restructuring the physical Hastings, J., Smith, B., Ceusters, W., Jensen, N. & Mulligan, K. (2012). environment; Michie et al. 2013) and delivery (i.e who Representing mental functioning: Ontologies for mental health and and what provides the intervention) disease. ICBO 2012: 3rd International Conference on Biomedical On-  Context: attributes of the target population (e.g aged tology. Larsen, K.R., Michie, S., Hekler, E.B., Gibson, B., Spruijt-Metz, D., 16+) and the intevention setting (e.g GP practices) Ahern, D.,…Moser, R.P. (2016). Behavior change interventions: the  Exposure: the extent and nature of the target popula- potential of ontologies for advancing science and practice. Journal of tion’s access to, receipt of, and engagement with the in- Behavioral Science, 40(6), 1-17. tervention, including reach (e.g proportion of sample Michie, S., Richardson, M., Johnston, M., Abraham, C., Francis, J., Har- that was exposed to intervention) and engagement (e.g deman, W…,. Wood, C. (2013). The Behaviour Change Techniques extent participants interacted with intervention compo- Taxonomy (v1) of 93 hierarchically clustered techniques: building an nents) international consensus for the reporting of behavior change interven-  Mechanisms of action: processes by which interven- tions. Annals of Behavioral Medicine, 46(1), 81-95. tion influences the target behaviour (e.g., by providing West, R. and Michie, S. (2016). A guide to development and evaluation of a cue to action) digital behavior change interventions in healthcare. London: Silver-  Outcome: the property of the target behaviour in the back Publishers. given scenario (e.g., 25%)  Effect: an estimate of the comparison between the out- comes in the evaluated scenarios (i.e. each pair of inter- vention conditions) 2