To be or not to be digital? A bibliometric analysis of adoption of eHealth services

Purpose – The purpose of this study is to introduce new tools to develop a more precise and focused bibliometric analysis on the field of digitalization in healthcare management. Furthermore, this study aims to provide an overview of the existing resources in healthcare management and education and other developing interdisciplinary fields. Design/methodology/approach – Thisworkusesbibliometric analysistoconductacomprehensivereview to map the use of the unified theory of acceptance and use of technology (UTAUT) and the unified theory of acceptance and use of technology 2 (UTAUT2) research models in healthcare academic studies. Bibliometric studies are considered an important tool to evaluate research studies and to gain a comprehensive view of the state of the art. Findings – Although UTAUT dates to 2003, our bibliometric analysis reveals that only since 2016 has the model, together with UTAUT2 (2012), had relevant application in the literature. Nonetheless, studies have shown that UTAUT and UTAUT2 are particularly suitable for understanding the reasons that underlie the adoption and non-adoption choices of eHealth services. Further, this study highlights the lack of a multidisciplinary approachin the implementation ofeHealth services.Equallysignificant isthe factthatmany studies have focused on the acceptance and the adoption of eHealth services by end users, whereas very few have focused on the level of acceptance of healthcare professionals. Originality/value – To the best of the authors ’ knowledge, this is the first study to conduct a bibliometric analysis of technology acceptance and adoption by using advanced tools that were conceived specifically for this purpose. In addition, the examination was not limited to a certain era and aimed to give a worldwide overview of eHealth service acceptance and adoption.

1. Introduction 1.1 Digital innovation and healthcare industry Many studies have described that, owing to digital tools, we are seeing a shift in the way that healthcare services are delivered (Carboni et al., 2022;Gaddi et al., 2013;George et al., 2012).In this shift, information technology (IT) and information and communication technology (ICT) have played a critical role and have the potential ability to make healthcare more accessible, minimize adverse occurrences and lower operational costs (Thuemmler and Bai, 2017, pp. 2168-2194).The creation of technologies (i.e.apps, programs, software) has used IT and ICT tools to improve existing procedures inside healthcare facilities (Haluza and Jungwirth, 2018).The adoption of these technologies can result in a variety of benefits, including quicker access to personal health data and increased sustainability, efficacy, efficiency and quality of delivered services (Scheibner et al., 2021).
In this sense, there are two major digital innovations that the healthcare industry can enjoy, namely electronic Health (eHealth) and Artificial Intelligence (AI) (Impedovo and Pirlo, 2019;Shaikh et al., 2023).In the following paragraphs, these two categories of innovation will be reviewed, specifying that our study focuses on the adoption of one of the two, namely eHealth.

AI for the healthcare industry
The role played by AI in the digitalization process of various industry sectors, including healthcare, is substantial.The increasing utilization of EHRs and digital imaging offers AI the opportunity to support patients and providers (Shaikh et al., 2023).AI applications include not only the utilization of machine learning (ML) algorithms for real-time data but also IoT eHealth ecosystems to support informal caregivers and vulnerable populations (Blasioli and Hassini, 2022).The proper management, interpretation and use of the data about the user or patient and their condition generated by the novel person-centered service model pose one of the major challenges in the relationship between eHealth and IoT.The techniques and ML methodologies finding application in healthcare depend on both the data and the IoT infrastructure under consideration (Cabestany et al., 2018).Below, we offer a succinct description of the most common AI branches with eHealth applications.
Natural language processing (NLP).Computers' ability to understand and process the natural language.Systems can make sense of a written text and perform different tasks, including topic classification, translation, synthesis and spell checking.
ML (algorithms trained with datasets capable of producing outputs from given inputs) techniques: (1) Supervised learning.Labeled datasets are used to trained ML models to learn and increase accuracy.The trained datasets include inputs and outputs: the accuracy in learning is measured by the algorithm by means of the loss function, minimizing the error until a threshold is met.Commonly used learning methods in this category include linear regression, logistic regression, neural networks, random forest and support vector machines (SVMs), (2) Unsupervised learning.In this case, the datasets used are unlabeled.This technique allows us to find patterns or trends when there is no knowledge a priori available about the structure of the data.In this case, possible functions that can define the hidden structure from the unlabeled data are deduced as the learning algorithms do not contain any labels to supervise the learning/training (Dike et al., 2018).Unsupervised ML algorithms typically involve clustering, anomaly detection and neural networks (NN), and (3) Semi-supervised learning.By combining the two methods described above, algorithms attempt to enhance the performance in one of the tasks utilizing the information associated with the other.To tackle a classification problem, for example, additional data points for which the label is unclear may be used.On the other hand, for clustering techniques, the learning process might benefit by being aware of the fact that specific data points belong to the same class (Van Engelen and Hoos, 2020).

TQM 35,9
Access to recommendations and automated treatments, as well as personalized medicine, are two eHealth areas that could greatly benefit from the introduction of AI approaches (Cabestany et al., 2018).

eHealth and its subcategories
The World Health Organization (WHO) defines eHealth as the cost-effective and secure application of ICT in support of health and health-related disciplines, such as healthcare services, health surveillance, health publications, and health education, knowledge, and research (World Health Organization [WHO], 2005, pp. 121-123).There is clear evidence that eHealth is having an increasing influence on the delivery of healthcare throughout the world, and it is making health systems more efficient and responsive to people's needs and expectations.
Increasingly, the term eHealth is placed alongside its subcategories mHealth, uHealth, telehealth and telemedicine (Bai et al., 2021;Lee and Yoon, 2021).Often, these terms are used interchangeably although they have specific meanings that differentiate one from the other.
Specifically, mHealth is a subgroup of eHealth and refers to the use of mobile devices in healthcare (Hamberger et al., 2022;Zhao et al., 2018;Park, 2016).The growing number of mobile devices has increased interest in developing and creating applications that could be installed and used to monitor one's health.
However, uHealth refers to healthcare systems that are particularly useful in the management of chronic diseases or those that require long-term care (Hamberger et al., 2022).It has the potential to facilitate diagnoses, improve the quality of care and reduce medical costs (Kim et al., 2022).
The WHO (1998) defines telemedicine as: The delivery of health care services where distance is a critical factor by all health care professionals using ICT for the exchange of valid information for the diagnosis, treatment, and prevention of disease and injuries, research and evaluation, and continuing education of health care providers, all in the interests of advancing the health of individuals and communities (p.10).
Finally, telehealth differs from telemedicine in that it encompasses a larger range of remote healthcare services than telemedicine.Although telemedicine refers to remote clinical services, telehealth may also refer to remote non-clinical services, such as provider training, administrative meetings and continuing medical education (Krupinski and Bernard, 2014).
In a chiaroscuro of opinions, eHealth services appear essential but there is considerable resistance to their adoption (Asthana et al., 2019;Kim et al., 2022;Kl€ ocker, 2015).Thus, it is certainly appropriate to conduct a bibliometric analysis of the studies that have been conducted so far to take stock of the situation and to understand the ways to possibly act so that eHealth services are understood and the distrust that is encountered is comprehended.
More precisely, our study intends to answer the following research questions (RQs): RQ1.How have studies on the adoption of eHealth developed over time?
RQ2.In which countries have studies been conducted on this topic and who are the authors who conducted these studies?
RQ3.What are the most cited studies that have inspired subsequent research?
RQ4.Based on our findings, what is the developing potential for further research?
We decided to focus our study circumscribing it to UTAUT and UTAUT2 for the following reasons: (1) UTAUT and UTAUT2 are two of the most innovative research models, carrying the constructions of several earlier models (Tamilmani et al., 2021).
(2) UTAUT and UTAUT2 have been effectively employed in other healthcare services studies to assess the rationale for technology adoption or non-adoption (Haikal et al., 2022).
(3) UTAUT and UTAUT2 could anticipate whether the healthcare business would adopt new technology (Duarte and Pinho, 2019).

Bibliometric analysis in contemporary research
This study uses bibliometric analysis to conduct a comprehensive review to map the use of the unified theory of acceptance and use of technology (UTAUT) and the unified theory of acceptance and use of technology 2 (UTAUT2) research models in academic studies on healthcare industry.A growing body of scholarly work has been devoted to defining and to mapping the intellectual structure of diverse research topics.Bibliometric studies are considered an important tool to evaluate research proceedings and to address knowledge gaps (Abramo and D'Angelo, 2011;Moed, 2006).Large-scale bibliometric research is possible because of the creation and development of the Science Citation Index (SCI), which currently incorporates the Web of Science (WoS), a Clarivate Analytics (formerly Thomson Reuters)-maintained platform, in 1963.In addition, within WoS, we find two more indexes that complement the SCI: the Social Science Citation Index (SSCI) and the Arts and Humanities Citation Index (A&HCI; Wouters, 2006).Until the creation of Scopus and Google Scholar in 2004, the scientific community relied uniquely on the WoS for citation analysis (Mongeon and Paul-Hus, 2016).However, the use of Google Scholar for research evaluation has raised doubts about its suitability because of the low quality of data that is found in it, leaving WoS and Scopus as the main sources for citation data (Mongeon and Paul-Hus, 2016).The WoS and Scopus are highly representative of the entire research output in the natural and formal sciences.Their use in bibliometric research makes this methodology a better solution than peer review in terms of robustness, validity, functionality, costs and execution times (Abramo and D'Angelo, 2011).During the past few decades, the WoS has been widely adopted as a source for bibliometric analysis in a variety of scientific fields (Hossain, 2020;Merig o and Yang, 2017;Shukla et al., 2020;Yu and He, 2020;Yu et al., 2017).The decision to adopt the WoS for our study reached a unanimous consensus given the database's features, its reputation and its suitability for the purpose of our work (Zhang et al., 2016).

Research method
It was planned to perform a bibliometric analysis that would focus primarily on two acceptance models: (a) the UTAUT (Venkatesh et al., 2003) and (b) the UTAUT2 (Venkatesh et al., 2012).These models, among many others, were found as a result of the widespread investigation of the variables that influence technology acceptability in the healthcare industry (Rouidi et al., 2022).
As was mentioned, the focus was on the adoption of eHealth services.Bibliometric data were analyzed by using a variety of tools, helping the authors to examine the phenomenon from various perspectives.
Records were initially searched by using the following association of keywords and Boolean operators: "eHealth" OR "telemedicine" OR "mHealth" OR "uhealth" OR "telehealth" AND "adoption" OR "acceptance" AND "utaut" OR "utaut 2" OR "utaut*" OR "Unified Theory of Technology Acceptance" OR "Unified Theory of Technology Acceptance 2" OR "Unified Theory of Technology Acceptance*".
We searched the WoS database, which, as was said, is one of the most significant tools for gathering systematic information on worldwide scientific literature (Zhu and Liu, 2020).The records that emerged were filtered by language and scientific sectors.In particular, the study focused on records in English and that related to the WoS categories "economics," "management," "social psychology," "health literacy and telemedicine," and "health policy." To strengthen the robustness of our study, we double-checked all the documents that were retrieved from this database.These records were compiled by using EndNote 20.2.1 software for further sorting.EndNote's "Find Duplicates" feature was used to eliminate duplicate data.Then, coauthors manually inspected the records to identify any duplicates that were not eliminated by the technology and publications that were not initially published in English.
We did not limit the examination to a certain era, and we included every record that was identified by the search criteria, regardless of the year that it was published.This was done to capture research from diverse periods.Further, we included any study, regardless of the sample's origin, because one of our purposes was to give a worldwide perspective, whereas proceedings, reports and non-peer-reviewed records were excluded.
After a careful reading of the books and book chapters deriving from the data collection, it was decided not to include them in the analyzed sample.In fact, although there were book chapters and books on the topic, these did not present empirical studies.On the contrary, they were mainly conceptual.Instead, we were interested in analyzing systematic contributions that have empirically studied eHealth adoption.
This stage of the sorting procedure resulted in the selection of 105 distinct Englishlanguage scientific articles (see Appendix).

Data analysis 4.1 Analysis overview
As was mentioned, we conducted a bibliometric analysis to study the evolution of the research on the topic of eHealth services acceptance and adoption and emerging research trends, evaluating the impact of the publications that were produced and the productivity of the authors, as well as understanding the potential collaboration patterns between countries.
The first part of this section presents some descriptive statistics about the impact of authors, scientific publications and journals.The second part analyzes the distribution of scientific publications by adopting a three-field plot in which the field selected have been, respectively, publications, keywords and journals.The third section focuses on the methodologies to perform network analysis: co-citation analysis, co-occurrence analysis and bibliographic coupling.Network analyses can help to explain the knowledge, the intellectual structure and the evolution of a research area.A methodology such as co-citation analysis allows for the mapping of subject-specific specialties and the identification of themes in research topics among the clusters of publications (Braam et al., 1991).

Published literature: descriptive analytics
We conducted a preliminary analysis of the corpus of publications, as shown in Figure 1.
As can be seen from Figure 1, the number of studies that have used the UTAUT and UTAUT2 models has grown considerably and there was a particularly significant peak from 2019 to 2022 when the total number of publications equaled 75 units (71.4%), of the 105 that constitute the sample.It is even more significant that the greatest peak occurred between 2020 and 2022, which are years that were marked by the COVID-19 pandemic.Subsequently, we used the software BibExcel.This is a bibliometric software (Persson et al., 2009) that was used to conduct a preliminary analysis of the publications and to generate a co-citation network file.Descriptive statistics are useful for capturing some of the major trends in the literature, embracing the distribution of publications studying the acceptance and/or adoption of eHealth.
In a second phase, the network file was opened using Gephi, a network science software (Bastian et al., 2009), to display and to further analyze the co-citation data that was generated by BibExcel, resulting in different network maps and topic clusters of the co-citation network.
Table 1 displays the most cited authors, whereas Table 2 displays the most cited papers within our network.Table 1 highlights how Venkatesh is the most cited author (21 citations in total).Davis (1989) follows in order with 7 citations and the other authors in the table with 5 citations.Table 2 confirms the importance of Venkatesh's work.Precisely Venkatesh and Davis (2000) can be considered as seminal of the two papers Venkatesh et al. (2003) and Venkatesh et al. (2012) in which the UTAUT and UTAUT2 models have been precisely elaborated.

Three-field plot analysis
A three-field plot provides a graphical representation of data that are organized into three columns.To generate this graph, an R package that is known as Bibliometrix was used.Developed by Aria and Cuccurullo (2017), Bibliometrix is equipped with tools for quantitative research in bibliometrics and scientometrics, addressing data collection, data analysis (descriptive analysis, network analysis and normalization) and data visualization (conceptual structure and network mapping).
Three-field plots, using Sankey diagrams, were used to study the patterns, trends and relationships among three selected fields.A Sankey diagram is a visualization that is used to represent a flow from one set of values to another.Connected elements are called nodes and connections are called links.The use of encryption keys is ideal for showing a "many-tomany" mapping between two domains or multiple paths through a set of stages.Sankey diagrams are frequently used in bibliometric analyses (Linnenluecke et al., 2020).Each column represents a dimension of the information.From left to right, the columns report: (1) publications, (2) keywords, and (3) journals (called "sources").
We decided to limit the number of elements for each column to 10 ( Zhang et al., 2016).Each rectangular node's size indicates the frequency of occurrence of a certain publication, keyword or author in the studied data (larger rectangles indicate higher frequencies).The number of connections or linkages between the nodes is indicated by the breadth of the connections between them (bigger nodes indicate stronger connections between fields).These measurements offer a visual depiction of the relative weight or frequency of each data point.The first column, which includes the research papers' titles in more detail, indicates the articles' specific areas of interest.Information about the academic journals is provided in the third column, allowing us to determine the most productive writers in a certain subject and get a sense of the network of collaborative research initiatives.The categories authors- In the WoS, Keyword Plus is a methodology to index scientific articles that allows for the inclusion of broader and more general terms than Author Keywords thus identifying related studies that might be missed if relying solely on the latter method (Zhang et al., 2016).However, Author Keywords provides a more accurate and precise picture of the research topic.Given the complementary nature of the two methods, we display a three-field plot that uses Author Keywords and another that uses Keyword Plus (see Figure 2 Substituting Author Keywords with Keyword Plus (see Figure 3), we can appreciate that the Keyword Plus algorithm resulted in a more diversified list than the list that was generated by Author Keywords in which some fields might overlap (such as eHealth and e-health).This allowed us to enrich the previous analysis.By expanding the coverage from keywords that were explicitly mentioned by authors to the co-occurrence of keywords within and across

Network analysis
The network analysis that we present combines techniques.Specifically, we used co-citation analysis, co-occurrence analysis and bibliographic coupling (Boyack and Klavans, 2010).Bibliographic coupling and co-citation analysis originated several decades ago whereas co-citation analysis was adopted in the 1970s.Since then, it has been the preferred approach (Boyack and Klavans, 2010).Although these three approaches are not combined, Small (1997) proposed using them together.We present different results from the deployment of these methods to provide insights from distinct perspectives in the analysis of our network of publications.
A co-citation network analysis was performed by using Gephi, a powerful open-source software that is capable of displaying large networks in real time (Bastian et al., 2009).Tracking pairs of publications that are referenced together in source articles is what co-citation analysis (Small, 1973) is about.When many writers mention the same pair of publications, research clusters emerge.The publications that are co-cited in these clusters usually have a shared subject.Gephi offers a wide range of filtering methods and tools, which allows for high levels of flexibility and performance in the handling of large sets of data.Co-citation studies represent a widely used methodology in quantitative studies of science, in particular, author co-citation analysis and document co-citation analysis.One of the most important insights that can be obtained by analyzing co-citation relationships is the identification of patterns, in addition to the uncovering of the intellectual structure of a field Adoption of eHealth digital services (Pilkington and Meredith, 2009) and the unveiling of the ways that research has evolved (Chen et al., 2010).A preliminary step consisted of manipulating the data set by using BibExcel through which a network file was generated and that was used in Gephi.Once generated, the co-citation network data were elaborated on by using Gephi for network analysis and visualization.For this purpose, we selected the first 200 co-cited documents.The obtained graph was weighted and directed and there was a total of 198 nodes and 3,445 edges.Each node represented a single publication, whereas each edge represented the co-citation relationship between the documents.The directed nature of the graph referred to the relationship between the nodes whereby the nodes interacted in a specific direction (Node A interacted with Node B, not vice versa).The degree of a node referred to the number of links incident on it.The in-degree index referred to the number of incoming links, whereas the outdegree referred to the outgoing links.The term weight referred to the strength of the co-citation relationship.Through the co-citation analysis, it is possible to assess the frequency with which two entities, in this case scientific publications, are cited together within a network of publications, which results in the degree of association between the two entities.
The graph was adjusted by using ForceAtlas 2 to customize some of its parameters, such as the gravity and scaling parameters.ForceAtlas 2 is a force-directed continuous algorithm that computes the forces between nodes and edges whereby nodes repulse and edges attract (Jacomy et al., 2014).The corpus of forces that governs the algorithm ultimately finds a state of equilibrium.
Subsequently, a method for community detection was applied to identify and extract communities from the network.By selecting the Modularity function that is available from the Statistics panel, the software applied the Louvain algorithm (Blondel et al., 2008), which allows the clustering process.We obtained a modularity index of 0.380, which can be considered an index of moderate quality, and that allowed us to identify some communities that are small and less distinct.
The network analysis resulted in six main clusters.Figure 4 shows how the co-citation and cluster analysis methods of scientometrics are used to perform a comprehensive analysis of the papers on the topic that is related to the study of eHealth services.
The complexity of the resulting network can be explained by looking at some key indicators, in particular, the PageRank algorithm that is used to order the documents according to their importance.
Conceived by Brin and Page (1998), this algorithm aimed to rank webpages by importance by making use of the link structure of the web, finding extensive applications in various domains.In the case of a co-citation network, the PageRank algorithm calculates a score for each document according to the PageRank scores of the other documents converging in it, weighting the score according to the importance of the edges.The computational process is iterative and stops once a stable value is found.
We defined G as a directed and weighted network, G ¼ ðV ; EÞ, where V is the set of vertices that represents the documents and E is the set of edges that represents the citations.The indexes i and j belong to the set of vertices V ði; j ∈ V Þ and identify, respectively, document i and document j; the number of citations from document i to document j is represented by one single directed edge ði; jÞ ∈ E (Fiala and Tutoky, 2017).The PageRank score PRðjÞ for document j depends on the PageRank scores of all the documents that cite j.Finally, we have a parameter, d, known as the damping factor, and the out-degree of D out ðiÞ.
The mathematical formulation of the algorithm follows: Therefore, PageRank was used to classify the publications that were extracted for each cluster because it represents a metric that describes the importance of a document in the network.We report the top 10 publications for each cluster (with the exception of Clusters 5 and 6, which had a smaller number of nodes), hierarchically selected according to the PageRank index, in Table 3.
Once the clustering process was complete, the main research topics for each cluster were analyzed.Table 4 presents the primary research themes, characterizing each community and the percentage of records in the sample that were investigated.Clusters 1, 2 and 3 represent the vast majority of our sample for a total of 88.4%.Clusters 4, 5 and 6 occupy a considerably minor space within the network.

Discussion and conclusion
The bibliometric analysis, conducted with the help of a variety of software to refine the search and to identify the relationships between the papers, leads to some noteworthy considerations.
In regard to the dates of the publications, it can be said that the vast majority of them are after 2015.There are certainly some earlier contributions, but they represent only 10.5% of the sample of identified records.We found that as much as 89.5% of the investigated In fact, the studies that emerged often focused on the application of an eHealth system in relation to a specific service.It should be remembered that the UTAUT and UTAUT2 research models investigate the acceptance and the consequent adoption of an innovation, which, in our case, is eHealth services.Therefore, if there are relatively few studies that have been identified, it seems to be possible to say that much remains to be done to understand the true intention to adopt innovation in healthcare management.
In addition, it is interesting that among the records that were identified in the bibliometric analysis, the vast majority of journals were of a medical nature.If it is true that, by its nature, an issue, such as the adoption of eHealth services, is closely linked to medical practice, many studies have highlighted the ways that the adoption of innovative services, such as eHealth, can only take place in the presence of multidisciplinary skills (De Grood et al., 2016;Van Velsen et al., 2013), including technical and IT, as well as economic and managerial ones.It is interesting to note that medical studies themselves believe in the need for transversal skills in the adoption of innovations such as eHealth services (Razmak et al., 2018;Swinkels et al., 2018).This confirms the importance of creating teams to support healthcare professionals.For example, according to Gaddi and Capello (2014), the slow diffusion of telemedicine and its patchy distribution is closely linked to the fact that telemedicine must not be implemented by doctors alone, but interdisciplinary teams are needed.From the bibliometric analysis that was conducted, it is clear that, at least in terms of scientific journals, there are very few IT, managerial and economic journals that deal with this topic.Referring to some of the most popular rankings for management and economics journals, none of the contributions that were covered by this bibliometric analysis were among the journals that were present in the FT50 rankings or in the broader Academic Journal Guide (AJG) 2021 rankings, produced by the Chartered Association of Business Schools.Yet, it is clear that the implementation of eHealth services has cost reduction among its objectives because the WHO defines eHealth as "the cost-effective and secure use of information and communications technologies in support of health and health-related fields" (WHO, 2005, p. 121).By its very nature, the implementation of eHealth services has economic and managerial implications, which, however, still need to be explored in the scientific literature.

Adoption of eHealth digital services
From the analysis of the clusters of contributions making up the sample investigated, it emerges how Cluster 1 brings together works that study the healthcare services available through IT technologies.The studies were conducted on the end users to understand their level of acceptance of the new services provided.More specifically these studies focus on mobile health adoption factors, eHealth perception and use, and acceptance of augmented reality in healthcare.Part of these contributions focuses on user acceptance theories.Cluster 2 always focuses on end user acceptance and precisely on the factors that determine adoption in its various forms (i.e. the Barriers, Facilitators toward eHealth services).Cluster 2 focuses on end user acceptance again and precisely on the factors that determine adoption in its various forms (i.e. the barriers and facilitators towards eHealth services).Precisely, it focuses on the factors that determine adoption in its various forms (i.e. the barriers, facilitators towards eHealth services and the impact on healthcare systems of eHealth services utilization).A part of these contributions focuses on aspects of data privacy and security, while the attitude of healthcare professionals is studied only together with that of end users.Cluster 3 is entirely focused on Telemedicine and Telehealth as a tool for maintaining the relationship with their patient in relation to their mental health.A part of these contributions focuses on the mental disorders developed during the COVID-19 pandemic and tried to explain how eHealth services could be of help in the absence of physical presence between the end user and the healthcare professional.Custer 4 represents a small minority of the topics that emerged from the analysis.This appears to be quite surprising, as Cluster 4 collects relevant methodological contributions that try to explain, with an empirical approach, the factors that lead to acceptance or abandonment of eHealth services.It can therefore be said that from our sample it emerges that most studies have focused on a specific eHealth technology, but few have come to theorize or hypothesize the reasons for the greater or lesser acceptance of eHealth services in general.If the studies certainly have their own "internal validity", it cannot be said that there is "external validity".If Cluster 6 is entirely focused on the opportunity to provide eHealth services on the African continent, Cluster 5, absolutely minority compared to the others, brings together studies focusing on the economic impact and how eHealth services and in general the digitization of health services could bring at cost savings.
Finally, from the analysis of the clusters it is clear that most of the articles focus on the acceptance of eHealth services by end users.There were very few studies that focused instead on the propensity to of healthcare professionals to adopt them.Certainly, it is important to know the propensity of the end user but the adoption process can only start with healthcare professionals.Therefore, before listening to the opinion of patients, it would be desirable to pay attention to the opinion of professionals.Their conviction and determination are, in fact, fundamental for innovative services, such as those investigated, to take hold (Veikkolainen et al., 2023).

Limitations and further research opportunities
Like any research work, this study has some limitations.It could be pointed out that the data were collected from the WoS database.Therefore, the limitations of the use of one database may apply to this study.In truth, some scholars encourage the use of the WoS for bibliometric analysis (Hossain, 2020;Merig o and Yang, 2017;Shukla et al., 2020;Yu and He, 2020;Yu et al., 2017;Zhang et al., 2016).
Other limitations of the study should be considered.For example, we are fully aware that this bibliometric analysis focused on two specific research models, UTAUT and UTAUT2.It would come as no surprise that the use of other models would lead to different records and hopefully different journals.
However, the overall goal of this study was to present an overview of the prominent trends according to key bibliometric indices.As a result, journal readers have gained a broad image of the most important records.However, these results are dynamic and subject to change when new mainstream themes emerge, and particular factors increase or decrease their place in the journals.
In regard to research opportunities, it is hoped that in the future, a more central role is given to the healthcare professional, who is believed to be the hub of the adoption of eHealth services.Until there is full awareness of their propensity to use these services, eHealth will remain an excellent project on paper that is only applied in particularly critical and extremely necessary moments (e.g. the COVID-19 pandemic).
The adoption of eHealth services should not happen suddenly or because of the good intuition of a luminary.The adoption of eHealth becomes fruitful when processes are created, and the entire interdisciplinary staff is committed to them.
).The article byVenkatesh et al. (2003; "User acceptance of information technology: Toward a unified view") dominated the network (frequency 5 15.00), comprising seven out of the 10 keywords (Author Keywords).Specifically, the work byVenkatesh et al. (2003) was connected to seven of the research themes that were identified by the keywords in the central column.It was followed by the publication by Venkatesh et al. (2012; "Consumer acceptance and use of information technology: Extending the unified theory of acceptance and use of technology"), which was connected to six out of the 10 keywords in the central column (frequency 5 8.00).The dominant theme that was identified by the Author Keywords was eHealth (frequency 5 21.00), which emerged as a relevant theme in seven out of the 10 papers listed in the left column.Further, it had been treated by three out of the 10 journals listed in the right column (Journal of Medical Internet Research, Canadian Family Physician and Journal of the American College of Clinical Pharmacy).The keyword digital health (frequency 5 4.00) appeared also significantly connected to the journals in the right column and had three connections in total.The most important journals were International Journal of Pharmaceutical and Healthcare Marketing (frequency 5 7.00), which had three connections among the keywords, and JMIR Formative Research (frequency 5 4.00), which had four connections among the keywords.