We chose to use the Web of Science (WoS) core collection and the Scopus database as basis for the study. The search query aimed to retrieve actually implemented behavior change systems with a technical component. Therefore, the following terms were used: (“persuas*” OR “ehealth” OR “mhealth”) AND (“ICT” OR “prototype” OR “techno*” OR “system”) AND “behavio* change.” The search was limited to the articles publicated during January 2000 and December 2018. The process of merging the collected data is discussed below. The query search resulted in a set of 317 and 325 articles from WoS and Scopus, respectively. Among the 325 items from Scopus, 11 appeared empty or incomplete at further inspection, so 314 were left.

For evaluating the existing scientific output through citations count, keywords, geographical data, authors collaborations, and discipline-wise interactions, the method of bibliometric analysis was used. The basic statistics included yearly publication output, publishing countries, the field of study, citation count, keyword co-occurrences, coauthorships, and collaboration networks between countries and institutions.

Several software packages are available to support a bibliometric analysis, each with different capabilities and limitations. Some of the most popular tools include HistCite [21], CiteSpace [22], BibExcel [23], and Science of Science (Sci2). In this study, the Sci2 tool was used, which is based on the Cyberinfrastructure Shell toolset, for the study of science [24].

In addition to tools for bibliometric analysis, we also used network visualization and analysis tools for the co-occurrence networks of keywords and research fields and the collaboration networks between countries and institutions. We used Gephi and VOSviewer that use a 3-dimensional render engine to render illustrations of large networks [25,26]. For the network analyses, the 394 extracted articles and their references were converted into graphs. For example, for the coauthor network, 394 articles resulted in 1777 nodes, where each node represented an author, and 5583 edges, where each edge represented coauthored studies.

A scoping review is a useful methodology to determine the coverage of a body of literature on a given topic and identify and analyze knowledge gaps [27]. A number of research questions in the introduction section were defined that will help investigate the use of behavior change theories and BCTs in actual BCSS.

Owing to the broad scope of the review, any conventional systematic review or meta-analysis framework was not followed strictly. However, the Preferred Reporting Items for Systematic Reviews and Meta-Analyses for Scoping Reviews guidelines (PRISMA-Scr) were applied [28] for our scoping review.

To be considered for review, the following inclusion criteria were formulated: (1) publication in a peer-reviewed academic journal; (2) publication in the English language; (3) the research should have the primary purpose of changing behavior, either increasing or decreasing the behavior or stopping the behavior altogether; and (4) the article should discuss a technological solution that is used for behavior change.

In this section, the key categories used in this study to classify the different interventions described in our dataset have been defined. The first category, technology, refers to the type of computer-based tool that is used to change a behavior, that is, a digital device, hardware, or a software solution. The target domain describes the behaviors that are targeted for change—for example, diet, sedentary behavior, mental health, or physical activity. The term behavior change theory/model is used to describe the theories and models about behavior change from behavioral sciences that are used as the basis for the persuasive systems. The category development frameworks/models covers a slightly different aspect; it describes the frameworks and models that are used to guide the design and development of the intervention. To further clarify the distinction between the two categories discussed above, the behavior change theory/model is used as the key behavior change principle for approaching the problem. For example, in a study by Lyons et al [29], the authors approached narrative transportation theory as a basis for increasing autonomous motivation about health behavior. In contrast, elements of the development frameworks category provide a systematic approach to select the behavioral principles and technological features during the design and development of the system. Finally, the category BCT describes the active ingredients of an intervention, for example, goal setting, reminder, and education. For this category, the taxonomy found in the study by Michie et al [7] was used.

Table 1 provides an overview of the categories and some examples of values. The categorization of the systems has been performed by the first author and finalized with the approval of other authors.

We applied our search query on 2 major literature databases, that is, WoS and Scopus. The WoS and Scopus databases returned 317 and 314 articles, respectively. There was an overlap of 179 articles between both datasets; thus, we identified 452 unique articles.

To perform our bibliometric analysis, we ideally would have merged the data from WoS with the data from Scopus. However, there are some technical constraints when performing a bibliometric analysis on different datasets: first, each database uses different sources for indexing the articles and, second, each database starts to index different journals at a different moment in time. For example, WoS started indexing the Journal of Medical Internet Research and its sister journal articles in 2015, whereas Scopus started in 2018 and is still indexing only a few of the journals published by JMIR Publications. Another issue is that the citation count for each article is different in the different databases, apparently, because different sources are used.

Owing to its quality and completeness of data, we decided to base our bibliometric analyses on data extracted from the WoS database [30,31]. To include as many relevant papers as possible, a manual search was performed in the WoS database for the 135 (314 minus 179 overlap) Scopus results that were not returned by the search query in WoS. Of those 135 articles, 77 were found in the WoS database, and 58 were not found. This resulted in a dataset of 394 items (317 as results to a query and 77 manually added items) for the bibliometric analysis.

Our scoping review started from the full set of 452 unique articles. A 2-phase screening was performed to determine their relevance for our scoping review. The first author initially screened the articles by reading the title and abstract and removing those that clearly did not match the inclusion criteria. As a result, 149 articles were excluded. The remaining 303 articles were thoroughly studied by the main author and compared with the inclusion criteria, which led to the exclusion of another 175 articles. The excluded articles were characterized in 3 main categories through group discussion. Furthermore, 10 articles were found as duplicates, as they were discussing the same intervention. Eventually, 118 articles were considered for the scoping review. The detailed flow of the selection process is presented in Figure 1.

The second objective of this study was a scoping review. The contents of the 118 articles that resulted from our selection process were thoroughly studied. This revealed some interesting insights and trends. The findings about the trends regarding technological choices and the theoretical basis of digital health interventions are presented in the following sections.

The field of PT is still quite young, and the literature regarding persuasive technologies for health and well-being started to appear in early 2000 [46]. The literature has grown exponentially after 2009/2010, which seems to be because of the introduction of smartphones in this era. Given the development around ubiquitous technologies such as sensors and networks, it is reasonable to expect that the growth of the literature will continue. As the field begins to mature, we see the number of publications increasing and researchers from different fields (eg, psychology and behavior sciences) contributing to the field. The geographic distribution of the publications shows that the United States plays a leading role, with a decent contribution from Western European countries and Australia. Other regions such as Asia and Africa almost have no contributions. To be able to include different behaviors and cultures, it would be useful if the collaboration networks are extended to these regions.

Given the fact that technology for behavior change requires expertise from different scientific areas, we expected quite some collaborations between technological and behavioral scientists. This study shows that interdisciplinary collaboration was not as widespread as expected. As human-centered disciplines such as psychology and other behavior sciences are quite mature and can provide essential knowledge about human behavior, technological researches cannot develop effective digital behavior change interventions without their contribution. Similarly, behavioral scientists require knowledge and insights from technological areas to apply their knowledge with modern means.

The keyword network illustrates the most important knowledge structures and thematic evolution in the field of digital behavior change systems. The keyword “mhealth” was strongly connected with the word “physical activity” and “ehealth.” This finding was not so different from the findings of our scoping review. The author collaboration network can be useful for expanding the collaboration network. The network shows a very strong intracollaboration among 3 groups of authors, where one is specifically working on behavior change, another on mobile health interventions, and the last one on persuasive system or BCSS. There is an opportunity to increase the intergroup collaboration, which could add valuable knowledge to the field. For example, with the help of remote sensors and Internet of Things (IoT) devices, health practitioners can monitor and tune systems for better adherence for their target group.

The translation of theories and theoretical frameworks/models into practice is essential for the development of any intervention. Davis et al [4] already concluded that only a few theories were used frequently. We also identified that the theoretical grounding for most of the systems was weak or, at best, not well described. An often-discussed reason for ignoring the majority of the theories is that intervention developers usually choose the most used theories and consider them as an easy and good choice. Another possible explanation is that the PT designer often lacks the skill to translate the theoretical determinants into technology design artifacts that are as effective as originally intended [4]. Conversely, a psychologist could study how digital technology enables new ways to personalize interventions and deliver just-in-time behavior change, with the potential to develop behavioral theories specific to the digital age.

The creation of a taxonomy of BCTs has been an important development in behavioral science, which is visible in the topmost cited articles in this review. However, these BCTs are still underreported in publications that describe persuasive systems [18]. Drawing on [47], the systematic review on mobile apps found that only 10 out of 93 BCTs were mentioned (mean of 2.42 BCTs were present in each app). Our finding in this review is not so different, only 6 articles explicitly used Michie coded taxonomy of BCTs. The most recent addition at the theoretical level is the link between the mechanism of actions and BCTs; this will probably pace up the development and evaluation of effective behavioral change interventions [48]. Owing to nonstandardized reporting, it is very hard to establish any relationship of effectiveness between 1 strategy and the success of persuasive systems. The result shows that each behavior domain is targeted with several techniques and strategies. Therefore, it is quite hard to determine the effectiveness of a certain technique or theory and the recommended process for design and evaluation of a behavior change support system.

On the basis our study, we can formulate a number of suggestions for future directions of the research in this domain.

Our study found a large gap in the process of designing digital technologies for health behavior change. They are, usually, weak in their theoretical grounding, and the papers describing them do not clearly report the different components, for example, persuasive strategies, theories, and BCTs. The main reason for this is the lack of design guidelines for these components. For better utilization and reporting of behavioral theories, the development frameworks also need to be updated to the most recent technological advances, for example, the IoT, that is, technologies capable of collecting a large amount of data, such as sensors and mobiles. Furthermore, computational models based on different theories can be designed that could be used by digital intervention developers [13].

Moreover, the relation between BCTs and behavior change theories/mechanisms requires more elaboration. For example, both for the SCT and theory of planned behavior, the use of self-efficacy construct is important. Self-efficacy is the strongest predictor of intention [49]. Furthermore, the correlations between determinants at different ecological levels for different behaviors need to be established. For example, the sitting time in cars has already been associated with urban design [50]. This requires more cross-sectional studies and controlled trials. These studies could help system developers make early decisions about the approach and techniques to follow for their digital interventions.

A clear framework or mechanism for reporting the components of health behavior change systems will not only advance the evaluation and its research design (eg, assess engagement, acceptability, and effectiveness) but also could enhance the costly process of development.

There are 2 limitations to our study worth mentioning. First, owing to technical reasons, we only considered the WoS database for bibliometric analysis; 58 papers that were relevant according to Scopus were not included in our analysis. A second limitation is a possible subjectivity in our scoping review. The categorizing of the reviewed papers has been done in a thorough manner but might still be influenced by subjective interpretations. Unfortunately, this problem cannot be avoided in this type of studies.