A Tool That Supports the Psychologically Based Design of Health-Related Interventions Anthony Jameson anthonyjameson@chusable.com Abstract Taking this complex context into account is challenging. It re- quires, among other things, (a) a systematic way of dealing with This paper offers a holistic and psychologically oriented per- multiple choosers and choices and (b) an understanding of the many spective on health recommender systems and introduces a tool— different phenomena that occur when people make choices and the C HUSAPEDIA—that is designed to support researchers and practi- many different ways in which these choice processes can be sup- tioners in putting this perspective into practice. One component of ported. C HUSAPEDIA is a semantic wiki in which a wide range of knowl- I have provided a foundation for the understanding described in edge is formalized about how people make choices and how these the second point in the form of the A SPECT model of choice and the choices can be supported. The second component is a web appli- A RCADE model of choice support ([3, 4]). But work on providing cation that accesses the knowledge in the semantic wiki to help computational support for the application of these models so that designers to create choice support interventions of various types. In they can be used by designers of recommender systems and other the paper, a starting point is a consideration of the strengths and choice supporting technology for the design of complex interven- limitations of the most advanced existing comparable resource, the tions began only recently; the present paper is the first publication Behavior Change Technique Taxonomy of Michie et al. An example on this work in progress. I start by discussing the most advanced of an analysis performed with C HUSAPEDIA then briefly illustrates existing approach that attempts to achieve similar goals, noting its the benefits offered by the tool, which will be demonstrated at the strengths and limitations. I then introduce the C HUSAPEDIA system workshop and made available to all workshop participants. with reference, for concreteness, to an example of a recent health- related intervention program that includes both recommendation technology and other forms of choice support. CCS Concepts •Applied computing →Consumer health; Health care informa- 2 Related Work tion systems; •Information systems →Wikis; Online analytical processing; The Behavior Change Technique Taxonomy (BCTT) In this short paper, instead of attempting even a modestly compre- Keywords hensive literature review, I will focus on a single well-known line of work which itself aims to integrate a great deal of previous work: Choice, choice support, health recommender systems the program of the group of Susan Michie, whose best-known result ACM Reference format: is the Behavior Change Technique Taxonomy (BCTT, [5, 6]). In this Anthony Jameson. 2017. A Tool That Supports the Psychologically Based research program,1 the authors synthesize insights from dozens of Design of Health-Related Interventions. In Proceedings of the Second Inter- models of how people make behavioral choices ([7]) and how these national Workshop on Health Recommender Systems co-located with ACM choices can be influenced and supported. Some of these models can RecSys 2017, Como, Italy, August 2017, 4 pages. be seen as general models of human behavior, while others include concepts that are specific to the domain of healthcare, which is the 1 Introduction application domain in the main focus of the authors’ program. This knowledge about how people make health-related choices is syn- Encouraging and helping people to live healthier lives is an enter- thesized in the COM-B model, which identifies six general factors prise that involves a great deal besides recommender systems. The influencing people’s health-related behavior, several of which in turn health-related behavior of the target person depends on a number comprise subfactors. of choices made by that person as well as others, such as family Likewise, the new system C HUSAPEDIA aims to achieve gen- members and healthcare providers. The choices made by these peo- erality through the inclusion of high-level concepts whose value ple will almost inevitably be influenced not just by recommendation has been demonstrated in many different domains and also more technology but also by other forms of advice and support offered domain-specific models such as those concerning the domain of either by interactive computing technology or via other channels. health-related behavior.2 1 I summarize this approach using terms and concepts that are employed in the rest of HealthRecSys’17, Como, Italy the present paper as opposed to adopting the authors’ original terminology. In particular, © 2017 Copyright for the individual papers remains with the authors. Copying permitted whereas Michie et al. speak of “changing people’s behavior”, I speak of “helping people for private and academic purposes. This volume is published and copyrighted by its make better choices”, for the reason explained below. editors. 2 C HUSAPEDIA can accordingly be applied to arbitrary other choice domains when equipped with the necessary domain-specific models. HealthRecSys’17, August 2017, Como, Italy Anthony Jameson Michie et al. also take into account the fact that understanding or in parallel, and one choice pattern can invoke another one as a how people make choices does not automatically enable you to help sort of subroutine. them make better choices. Drawing from many lines of previous 3. Like many other authors in the domain of health and well-being, research, they developed the Behavior Change Technique Taxonomy, Michie et al. speak of “changing behavior” rather than of “helping which comprises 93 behavior change techniques ([5, 6]). people to make better choices”. In addition to being more clearly In addition to offering these compact syntheses, the authors offer relevant to the recommender systems field, the latter conception is a practical procedure for putting the synthesized knowledge into more general than the former conception: Because of the mostly practice. voluntary nature of human behavior, changing people’s behavior in general involves inducing them to make particular choices (e.g., what to eat or how to exercise). The “choice support” perspective is more Strengths and Limitations of the BCTT Approach general than the “behavior change” perspective in that it includes The BCTT program has rendered a valuable service by enabling not only inducing people to choose options that a particular behavior intervention designers to deal with compact and coherent taxonomies change agent considers desirable (e.g., following a Mediterranean of choice-related variables and choice support techniques, respec- diet) but also helping people to make satisfying choices where there tively, instead of having to cope with dozens of largely overlapping is no predetermined correct option (e.g., finding out which partic- models that use conflicting conceptual schemes and terminologies. ular foods within the Mediterranean diet the chooser most enjoys Among other visible benefits, hundreds of published behavioral in- preparing or eating). In my keynote talk at Persuasive 2013 ([2]) and terventions have been described in terms of the BCTT,3 making it in my tutorial at Persuasive 2017,5 I argued for the desirability of possible to identify their “active ingredients” and to more rapidly viewing these two types of support for choices—which I have called accumulate knowledge about what interventions work in what con- persuasion and choice support, respectively—as complementary texts. But especially from the point of view of those who aim to approaches that should in general be combined (cf. also Section apply recommender systems and related technologies in the health 1.2.2 of [3]). domain, there are several limitations that remain to be overcome: 4. One of the main uses of the BCTT to help integrate knowledge 1. The authors’ focus on the healthcare domain makes it some- about existing interventions has been the practice of coding an inter- what difficult to apply the concepts to situations that fall even partly vention in terms of the BCTT techniques that it employs. But most outside of that domain. Consider, for example, the many applications interventions involve a number of different concrete “features” (e.g., that help encourage people to engage in sports activities. Although the functions provided by a mobile recommendation app; regular such activities can have a large impact on achievement of the goal of consultations with a doctor), which can take very different forms. healthy living, sports is not entirely a matter of keeping healthy. It Moreover, a single feature can realize two or more BCTT techniques, also involves goals such as having fun, competing with others, and and a single BCTT technique can be realized by two or more features. maintaining social relationships. So it would be desirable to have a So it would be more helpful to know which particular BCTT tech- framework that (a) enables the exploitation of domain-specific in- niques were realized by which particular features of the intervention. sights but also (b) enables the intervention designer to decide which Accordingly, C HUSAPEDIA’s concepts are applied not to an entire domains are relevant for his or her particular problem. intervention but rather to specific features of an intervention.6 2. The COM-B model of behavioral choice ([7]) is what might be called a “variables-based” model: It identifies a number of variables 3 Brief Introduction to Chusapedia that influence people’s behavioral choices, such as “psychological capability” and “reflective motivation”. Designing an intervention C HUSAPEDIA is a web-based system designed to be used by therefore is a matter of taking steps that will lead to more desirable anyone who wants to design a choice support intervention, which values of some of these variables (e.g., improved skills or factual may be anything from a single recommender system application to knowledge). One limitation of this type of model is that it is too a combination of applications and/or human activities (e.g., advice coarse-grained to express many of the insights that have been de- provided by health specialists). C HUSAPEDIA has two main com- veloped in decades of research into how people make (behavioral) ponents: The first is a semantic wiki (realized within the Semantic choices. MediaWiki platform7 ) in which a wide range of knowledge about By contrast, C HUSAPEDIA (like some other approaches) employs how people make choices and how these choices can be supported a “process-based” model of how people make choices. The A SPECT is collaboratively formalized. This knowledge is derived from the model,4 which forms part of the knowledge base of C HUSAPEDIA A SPECT and A RCADE models and from other models such as BCTT distinguishes six choice patterns, two of which in turn have subpat- and more specific models such as those that BCTT has already terns. For each choice pattern, there are several typical steps that can taken into account. It will be continually expanded to include also be taken to apply it. The choice patterns can be applied alternately knowledge that is specific to particular application domains or types of choice problem. The second component is a web application 3 http://www.bct-taxonomy.com/interventions that is implemented outside of the semantic wiki but that accesses 4 The A SPECT model and other concepts used in C HUSAPEDIA are presented briefly in the readily available handbook chapter [4] and in detail in the book-length monograph 5 http://www.dfki.de/∼jameson/pt17-tutorial-jameson [3]. The C HUSAPEDIA user interface also includes explanations of the concepts. In 6 The concepts are also linked to particular choices made by particular “choosers”, as is the present paper, these concepts are explained only to the minimal extent required for readers to be able to understand the main features of C HUSAPEDIA and how it can be explained below. useful in the area of health recommender systems. 7 https://www.semantic-mediawiki.org/wiki/Semantic MediaWiki . . . Psychologically Based Design of Health-Related Interventions HealthRecSys’17, August 2017, Como, Italy Figure 1: Screen shot of an early version of the Chusapedia web application showing a partial analysis of the TreC-LifeStyle inter- vention the wiki’s knowledge to enable designers to analyze and improve Figure 1 gives a selective overview of the information contained designs for specific choice support interventions. in C HUSAPEDIA after an intervention designer has used C HUSAPE - Both components of C HUSAPEDIA will be demonstrated at the DIA ’s web application to describe the current version of the T RE C- workshop and made available to participants.8 Because of the space L IFE S TYLE intervention and its design rationale. In the left-hand limitation for this paper, I explain C HUSAPEDIA here with reference column, we see that the designer has distinguished three types of to a concrete example of how it can be used to analyze and improve chooser that are addressed by the intervention: the parent who is a specific intervention in the domain of health and well-being. responsible for buying and preparing food for the child; the child The intervention ([1]) is specifically aimed at families in Italy who is supposed to be encouraged to eat and exercise in a healthier that include a child who is overweight because of some combination way; and the doctor who monitors progress and gives advice. of inappropriate nutrition and inadequate physical activity. One In the second column, we see that the designer has listed the aspect of the intervention is that the child is expected to wear a choices made by each chooser that the intervention is intended to motion tracker throughout the day, whose data are transferred to the support.9 T RE C-L IFE S TYLE mobile application. This app has several types Together, these two columns illustrate the fact that a choice sup- of functionality, including those that help the parent of the child to port intervention often needs to address several different choices monitor the child’s physical activity, to select appropriate food to made by different choosers. serve the child, and to keep track of the child’s food consumption. In the third column, the designer has specified, for each choice, The actions of the child and the parent are monitored by a doctor, which of the six choice patterns (Section 2) distinguished by C HUSA - who gives advice about any necessary adaptations to the actions of PEDIA might be applied by the chooser in making that choice. The these participants. inclusion of these choice patterns in C HUSAPEDIA helps to remind the designer of the wide variety of ways in which people make choices, so that the designer does not make the mistake of basing the intervention’s design on assumptions that may not be true. 8 An unpublished document about the system’s structure and design rationale is available 9 Starting with this column, the table shows only a subset of the relevant elements, on request. because of the limited space available for the presentation and discussion of the table. HealthRecSys’17, August 2017, Como, Italy Anthony Jameson Each choice pattern is associated with some typical choice steps: When it comes time to improve the intervention, the designer can mental operations that the chooser can engage in. In the fourth take advantage of another source of knowledge within C HUSAPEDIA: column, the designer has specified the likely choice steps for each the case base of interventions that have already been contributed by choice under consideration. other designers. Since these interventions will have been similarly For each choice step in each choice pattern, C HUSAPEDIA offers described in terms of C HUSAPEDIA’s concepts, it will be fairly information about some choice support tactics that can be applied to straightforward for the designer to retrieve previous interventions support that choice step. Some of these tactics are quite generic and that involve similar application situations, types of choices, and domain-independent; many of these were presented in the original choice processes. He can then copy and adapt parts of the previous exposition of the A RCADE model of choice support ([3]). Other analysis and intervention design, for example adding a feature from tactics may be specific to particular application domains or problem a previous intervention. types. The semantic wiki component of C HUSAPEDIA makes it As another useful source of knowledge, external models such possible for domain experts to contribute specific tactics of this sort, as the Behavior Change Technique Taxonomy are represented in which the designer of an intervention can ask to have loaded into the semantic wiki component of C HUSAPEDIA along with links to the web application when she begins her analysis. One of the goals the “core ontology” that has been discussed so far. So if a designer of C HUSAPEDIA is that the library of such specific models should who is already familiar with the BCTT gets the idea of applying grow over time, making the system increasingly useful in a growing technique 5.2, “Use methods specifically designed to emphasise number of domains. the (health) consequences of performing the behavior with the aim In the final column, the designer has specified which particular of making them more memorable”, she can specify this idea and features (cf. Section 2) of the intervention realize each tactic. As the C HUSAPEDIA will tell her which choice steps and tactics from last four rows illustrate, a single feature can realize more than one the core ontology correspond to that particular technique. This tactic, and the designer can ask for all uses of a given feature to be incorporation of external models is intended to ensure that (a) the highlighted. knowledge already incorporated in these models is also present in C HUSAPEDIA and (b) intervention designers who are already accustomed to using other models will find it easy to take advantage of the benefits that C HUSAPEDIA offers. 4 Benefits of a Chusapedia Analysis 5 Acknowledgements Even before considering how a designer can use C HUSAPEDIA to improve an intervention (or to design an intervention from scratch), Thanks are due to Silvia Gabrielli, Mauro Dragoni, and Claudio let’s consider how the designer can benefit by creating this sort of Eccher of FBK in Trento for detailed information about the T RE C- analysis of an existing intervention: L IFE S TYLE intervention and for extensive feedback on an early 1. The analysis yields a holistic view of the intervention. Instead version of C HUSAPEDIA. of seeing the intervention as a set of specific techniques that are being applied under the assumption that they will somehow be helpful, the designer sees the features of his intervention as fitting into a large References picture comprising one or more choosers, the choices that they are [1] Silvia Gabrielli, Lorena Filippi, Marco Dianti, Rosa Maimone, Marta Betta, Monica making, the ways in which they go about making them, and the ways Ghezzi, and Stefano Forti. Design of a mobile app for nutrition education (TreC- in which they can be supported by the intervention. LifeStyle) and formative evaluation with families of overweight children. JMIR mHealth and uHealth, 5(4), 2017. 2. References to general and domain-specific models of choice [2] Anthony Jameson. How can persuasive technology help people choose for them- and choice support, which in turn are based on large amounts of re- selves? In Shlomo Berkovsky and Jill Freyne, editors, Proceedings of Persuasive 2013. Springer, Berlin, 2013. Abstract of a keynote address. search and experience, enable the designer to tap systematically into [3] Anthony Jameson, Bettina Berendt, Silvia Gabrielli, Cristina Gena, Federica Cena, a large body of knowledge that would otherwise be scattered over Fabiana Vernero, and Katharina Reinecke. Choice architecture for human-computer many different publications and other documents, using different interaction. Foundations and Trends in Human-Computer Interaction, 7(1–2):1– 235, 2014. terminology and conceptual schemes. [4] Anthony Jameson, Martijn Willemsen, Alexander Felfernig, Marco de Gemmis, 3. The analysis will in most cases not simply serve to satisfy the Pasquale Lops, Giovanni Semeraro, and Li Chen. Human decision making and designer that she has already come up with an optimal intervention. recommender systems. In Francesco Ricci, Lior Rokach, and Bracha Shapira, editors, Recommender Systems Handbook. Springer, Berlin, 2nd edition, 2015. On the contrary, it is more likely to bring to light the fact that the [5] S. Michie, M. Richardson, M. Johnston, C. Abraham, J. Francis, W. Hardeman, designer, when originally designing the intervention, did not take M. Eccles, J. Cane, and C. Wood. The behavior change technique taxonomy (v1) of 93 hierarchically clustered techniques: Building an international consensus for into account one or more choosers, choices, choice patterns, choice the reporting of behavior change interventions. Annals of Behavioral Medicine, steps, and/or available tactics. This identification of gaps can help 46(1):81–95, 2013. the the designer to proceed to improve the intervention. [6] Susan Michie, Lou Atkins, and Robert West. The Behaviour Change Wheel: A Guide To Designing Interventions. Silverback Publishing, UK, 2014. Improvement of the intervention proceeds in much the same [7] Susan Michie, Robert West, Rona Campbell, Jamie Brown, and Heather. Gainforth. way as the analysis of the original version of the intervention: The ABC of Behaviour Change Theories: An Essential Resource for Researchers, Policy designer specifies additional elements of the analysis, including Makers and Practitioners. Silverback Publishing, UK, 2014. possibly new features of the intervention that ought to be added— and possibly also the elimination of features which, according to the analysis, serve no clear function.