A variety of terms and concepts exist that describe the use of technology in health including Health IT, eHealth, Telemedicine, and Health Informatics. In the early 2000s [1], the term Connected Health (CH) began to appear. It did so in the context of research that investigated how information and communication technologies could advance health by connecting people, knowledge, technical artifacts, and organizations. Some years later, Poon and Zhang introduced a new CH information system based on a “four-layered architecture: personal, home, community, and hospital” [2]. One of the most cited definitions for CH was introduced in 2013 [3]:

The European Network for the Joint Evaluation of Connected Health Technologies defines the CH vision as “a paradigm shift looking after the individual and community health in a process that speaks to the health journey of the person, through the entire lifespan, leveraging a variety of technologies to do so” [4]. Achieving this vision will require attention to be paid to policy and regulation, technology and interoperability, training and education, business and revenue models, as well as citizen and clinician engagement [5]. Successful implementation of the CH vision needs time and effort from all health care stakeholders.

Given the recent growth in CH research, we set out to map its evolution up to the present day. We use a systematic mapping study to chart the research landscape combining scientific contributions from the research disciplines of medicine, business, computer science, and engineering. Although previous CH reviews do exist, they are confined to descriptions of how CH solutions are being used in specific conditions such as cancer [6,7] or in specific measurement technologies such as measurement of vital signs [8] or weight [9]. Past efforts also concentrated on presenting CH-related literature in systematic ways [6,10,11]. Our goal is to provide a comprehensive, interdisciplinary overview of existing CH research. This will help researchers understand how the field has developed since its earliest studies in the late 1990s. It will also identify those topics that have drawn the attention of the research community. This knowledge will add value by identifying gaps or interdisciplinary opportunities in the study of CH. Perhaps most importantly, it may also underpin future work to develop an integrated and interdisciplinary research agenda for CH that will answer efficacy, design, policy, and sustainability questions for patients, clinicians, technology developers, and businesses.

This paper followed a systematic mapping study method [12]. Systematic mapping studies aim primarily to present an overview of a research area to report the quantity and type of literature and results that are published within it. The systematic mapping process comprises 3 steps: (1) the identification of relevant literature, (2) the composition of a classification scheme, and (3) the mapping of literature [12]. Figure 1 presents the mapping process including the search for relevant literature, the definition of a scheme, and the mapping of relevant publications.

The method was used to examine the body of existing research conducted by researchers in medicine, business, computer science, and engineering to understand the nature of research conducted in the area of CH. A systematic mapping study was found to be suitable for this task as it provided a high-level framework for combining interdisciplinary research efforts as well as an analytical framework that spanned disciplinary boundaries.

The aim of this study was to present an overview of the available publications pertinent to CH. Following the systematic mapping study method, the study was guided by a set of mapping questions. Table 1 presents the 6 mapping questions (MQs) and the rationale for conducting this study. More detail as to the logic supporting the selection of suitable MQs is included in the Data Extraction Strategy section below. The study search strategy as well as the inclusion and exclusion criteria (EC) were based on these 6 MQs.

As CH is inherently interdisciplinary, our goal was to use the systematic mapping method to study research contributions on the topic across disciplines. We, therefore, searched papers from the most recognized scientific literature databases in each of the chosen disciplines.

The papers of the study were retrieved from 7 databases of scientific literature, namely, (1) Institute of Electrical and Electronics Engineers Xplore Digital Library, (2) Association for Computing Machinery Digital Library, (3) ScienceDirect, (4) SpringerLink, (5) MEDLINE and PubMed, (6) Business Source Complete (EBSCO), and (7) ABI and INFORM Collection (ProQuest), with the help of the corresponding search engines. The search was performed in October 2018. Different search strings were proposed and discussed over the course of joint meetings to arrive at a set of primary keywords. Two of the authors tested different strings of potential keywords such as “Connect Health,” “Connecting Health,” “Connect-Health,” or “Connecting-Health.” After evaluating the search results, the authors agreed to proceed using the following search strings: (“Connected” AND “Health”) OR (“Connected” AND “-” AND “Health”).

The search was applied to the title, abstract, and keywords to include relevant papers. On the basis of our methodology, we included a wide selection of papers on the first iteration and thereafter relied upon the inclusion criteria and EC to identify the relevant literature [13].

The search results that were retrieved from the 7 chosen search engines were merged in a single list and saved in a spreadsheet document. Duplicate entries were removed based on the digital object identifier (DOI) and the intertitle lexical distance (the Levenshtein metric was used to measure the lexical distance) with the help of a custom Python script [14]. A total of 7 papers had to be manually reviewed as potential duplicates.

The final list of papers was distributed to all authors for analysis. On the basis of the area of expertise, pairs of authors were assigned to analyze papers retrieved from each database. A total of 3 areas of expertise were identified as follows: technical (computer science and engineering), medical, and business. Each pair of authors reviewed the title, abstract, and keywords and made a recommendation as to whether that paper should be included or excluded. Discrepancies across coding teams were resolved through further scrutiny of the paper.

The inclusion criteria were limited to studies that discussed CH. In total, 240 papers were identified after the removal of duplicates, whereas 32 papers were excluded after meeting at least one of the following exclusion criteria: EC1—papers that focus on CH comorbidities, that is, the paper discusses the connection between health conditions, not CH as a concept; EC2—papers that focus on law; EC3—papers that focus on medical procedures only, without connection to CH; and EC4—papers that focus on evaluating climate but not in relation to CH. Figure 2 presents the process of the study selection.

The selected studies were analyzed to collect information that would give answers to the MQs according to the data extraction strategy outlined in Table 2 below.

MQ6 comprised 2 parameters: the topic of the paper and the medical condition examined within it. With regard to the topic of the paper, we did not have a predefined list of topics but relied on an open coding process where the researchers conducting the analysis selected a descriptive word for the topic of the paper. To identify the main topics of the included papers, the authors relied on the title, abstract, and keywords. After compiling all topic words, we curated the topic list to come up with a consistent list (eg, using the same word for body-worn sensors and wearables). Similarly, with regard to the medical condition examined, no predefined classification was used that allowed coders to assign the medical condition, if any, of each CH paper. Again, the resulting condition list was curated to develop a more consistent list (eg, using the same word for ageing and elderly).

The synthesis method used was based on the following steps: (1) enumerating the number of papers per publication channel and the number of papers per bibliographic source per year; (2) enumerating the primary studies that are classified in each MQ’s response; (3) presenting visualizations for the classification results, which have been used in the analysis; and (4) presenting a narrative summary to discuss the principal findings.

This section describes the results related to the systematic MQs presented in Table 1. Multimedia Appendix 1 gives an overview of the classification results for all the included papers [6-11,17-217]. Custom Python scripts have been developed to process the classification data and generate the tables (Multimedia Appendix 1) and figures (Figures 3 and 4) of this section. The Pandas and the Matplotlib Python [218,219] libraries were used to manipulate the tabular input data and plot the results, respectively.

The majority of the CH papers were published in scientific journals (139/208, 66.8%), whereas 32.9% (68/208) were published in scientific conferences. Table 3 lists the publication forums that have published at least two CH papers.

Figure 3 presents the publication trend per year from 2000 to 2018. It must be noted that the analysis does not include the full body of CH work published in 2018 as our search was performed before the end of that year. We have added a projection for year 2018, estimating the total number of papers. Our projection assumes that publication frequency would remain consistent till the year end. As Figure 3 shows, there is a gap in publications from 2001 to 2004. Less than 5 papers per year were published until 2008, whereas from 2013 to 2017, the number of publications rises steadily.

Figure 4 shows the research types of the included studies. The majority of the papers are solution proposal studies (77/208, 37.0%), 23.5% (49/208) of the selected studies evaluated CH approaches, 20.6% (43/208) reviewed literature, 16.8% (35/208) reported the opinions of authors about CH, and 1.9% (4/208) of the papers analyzed were classified as other.

Figure 4 presents our findings as to whether or not the included studies were empirically validated and, if so, the empirical validation approaches used. The majority of the studies were, in fact, empirically validated. More specifically, 28.3% (59/208) were validated in case studies, 26.4% (55/208) in controlled experiments, and 13.4% (28/208) with surveys. Nonempirical papers included, for example, those focused on theory (25/208, 12.0%) and those using history-based evaluation (11/208, 5.2%).

Figure 4 shows the wide range of approaches taken in the included CH studies. The approaches most frequently reported belonged to other categories (66/208, 31.7%), followed by methods (56/208, 26.9%) and tool-based techniques (43/208, 20.6%). Only 10.0% (21/208) of the papers were classified as presenting frameworks, with 8.1% (17/208) suggesting models.

Tables 4 and 5 present the results for the main topic and condition of the included CH studies. Topics with fewer than two occurrences have been omitted from Table 4. The most common research topics of the included CH papers were health care (×16) and disease management (×13), followed by telemedicine/telehealth (×11) and electronic health (eHealth)/mobile health (×9). Table 4 lists all research topics and their frequencies, whereas Table 5 lists the conditions and their frequencies. As for the main conditions reported in our study, diabetes (×12), aging (×10), cardiovascular diseases (×10), cancer (×7), chronic diseases (x7), and dementia (x4) were among the prevalent conditions found in the selected studies.

Owing to the interdisciplinary nature of our topic, we used an interdisciplinary team of researchers for analysis. Having researchers with different backgrounds could decrease inter-rater reliability, especially where we did not have a predefined list of values, as with the topic and condition parameters. Nevertheless, for the analysis of results, the authors relied upon the interpretation of descriptive statistics and visualizations, thus decreasing the threat to validity. In the same vein, to alleviate the authors’ influence on the classification process, the development of the classification scheme relied on widely accepted guidelines [13]. The differences between content and style of abstracts in different research disciplines and traditions may have resulted in slight differences in information retrieval process. In some cases, the abstract did not include all information needed to classify the paper, and the researchers had to read parts of the full text to obtain all relevant information. However, the vast majority of the 208 publications identified have been classified purely based on the title, abstract, and keywords. In our view, subtle differences in the publication screening process have had only minor impacts on the main conclusions drawn from the 208 publications identified in our study.

The differences in publication practices between the disciplines probably had an influence on how the results of MQ1 were interpreted, as the role of conference and journal publications differs between disciplines [241]. For example, many highly regarded conferences within the business discipline do not publish conference proceedings (eg, European Group for Organizational Studies). Indeed, even those conferences that do publish proceedings may only do so for a subsection of the best papers, and authors will still be offered the option of removing their papers from those proceedings to protect future publishing opportunities (eg, Academy of Management). These factors combine to mean that verbal discussions may well have commenced within the business academic community in conferences that are not reflected within our review as it deals only with published material.

As the methodology of the systematic mapping study that we used in this research was originally developed in the context of software engineering, it is likely that some of the analysis parameters were less optimal in other disciplines. This is reflected, for example, in the large number of studies classified into the category other for the parameter approach. In the joint analysis meetings, the coding authors shared experiences of their difficulties in classifying the papers within the agreed analysis parameters. Alternative solutions were discussed, but it was difficult to reach a consensus that would have been satisfactory across disciplines. Ultimately, we decided to use the classification parameters proposed by our methodology. The validity of our conclusions is only applicable within the CH context.

To limit the threat related to the identification of primary studies and to include as many relevant papers as possible, 2 of the authors ran several iterations to test different strings of keywords. The adoption of the final set of keywords was used as it returned the largest number of studies. However, the list of studies might be incomplete, as additional or different terms might have an impact on the final selection of papers [242]. Nevertheless, in light of the interdisciplinary endeavor and the scope of the study, we believe that we have included the majority of the relevant literature. For the bibliographic search, no timeframe was defined; hence, the representativeness of the included studies was not affected by this factor. The results of this study should be considered under the prism of the specific search string and classification scheme and offer a baseline for future endeavors.

Our search strategy and inclusion criteria have omitted studies that are referenced in grey literature. However, the literature search was conducted in the world’s most leading and comprehensive databases for scientific knowledge. Furthermore, to alleviate the threat of publications’ nonavailability because of subscription paywalls, we performed the initial screening using a combination of university libraries to improve our access to papers. To address validity threats because of duplication, duplicate entries were removed based on the DOI and the intertitle lexical distance with the help of a custom Python script, which was ultimately manually reviewed to ensure duplicate removal.

On the basis of our results, we can conclude that CH research is an established field of research. The interdisciplinary nature of the field can be seen especially in papers at the intersection of the medical and technical disciplines. The number of papers in the business research publication forums is still smaller. However, business-related themes are visible in topics of papers, such as consumer orientation and innovation research, although at a much smaller scale than the topics of more technical and medical nature. For CH to succeed, money needs to move differently around the health care system. Most developed health care systems continue to reimburse care in a cure rather than a prevention mode. Cutting-edge technologies and redesigned care pathways may fail if they run contrary to the flow of health care finance. Our findings emphasize the need to increase business research in the area of CH or to find the vocabularies and keywords necessary to link existing business research with CH endeavors.

There is a growing need to involve and engage patients in their own care and, by extension, in the design of digital solutions to improve their efficacy. Tailored CH interventions may more effectively reach the intended audience in a meaningful way, but this requires in-depth understanding of the condition’s needs, barriers, and facilitators. The important role that health care professionals play in the health care system is in contrast with their lack of involvement in the design of CH. In the same vein, recent research suggests that health care professionals’ education in Europe is lacking in the area of health care information technologies [243]. Emerging trends such as user-centered design and the inclusion of patient representatives attempt to address these problems by creating CH solutions that are tailored to the characteristics and tasks of the intended users. Adoption of these and similar approaches could further the field of CH research and implementation.

Given that CH research has become far more widespread in the years since 2013, perhaps it is time to devote more research resources to the scalability of CH as reflected in empirical approaches that facilitate the use of larger populations.

We see our findings as the foundations of a research roadmap for CH researchers that challenge current thinking in health care. Such a research agenda would go beyond investigations into the feasibility of individual technical solutions to examine and develop ecosystems of stakeholders, technologies, and infrastructures that together form new kinds of systemic solutions. Such an agenda would require more focus on research that addresses interdisciplinary methodological questions alongside the creation of vocabularies and frameworks for researchers working in different disciplines to effectively collaborate and examine interdisciplinary research questions through joint methodological approaches.