Health care services worldwide face significant challenges from increasing demand on the one hand and an increasing lack of availability and accessibility on the other hand, accompanied by rising health care costs [1]. The current COVID-19 pandemic has also affected health care delivery and has highlighted the need for alternative approaches that can overcome geographic, temporal, and organizational barriers to providing comprehensive high-quality care [2].

A promising way to overcome these barriers is technological progress, which is driven in particular by increasing digitization and advances in the field of artificial intelligence (AI). One promising technology is conversational agents (CAs), also known as chatbots [3,4]. On the basis of previous definitions, we define CAs as digital dialog systems that enable people to have text-based, speech-based, or nonverbal conversations with a computer or another machine based on natural language. The related concepts and variants of CAs are provided in Multimedia Appendix 1 [5-38].

The use of CAs offers new opportunities and various benefits for health care. Current research points to their ability to improve the accessibility of health care services and medical knowledge and to foster patient-centered care while reducing health care costs. Furthermore, their ability to communicate in multiple languages has been discussed [1,39]. This technology can support health care professionals in their daily work and thus reduce their burden [29]. Numerous studies have demonstrated the effectiveness and efficiency of using CAs in health care, such as in supporting diagnostic decision-making [40] and cognitive behavioral therapy for psychiatric and somatic disorders [41-43]. In this regard, CAs support effective, acceptable, and practical health care comparable with that provided by human physicians [44,45]. Owing to the nonjudgmental nature and impartiality of CAs, studies postulate that the systems may even be better suited than health care professionals to meet the needs of patients in some areas [29].

Achieving acceptability, acceptance, and adoption is challenging for new technologies, as the user’s journey to technology acceptability, acceptance, and adoption is complex and nonlinear [46,47]. The success of an innovation depends on its use by end users, that is, its acceptability, acceptance, and adoption. Acceptability is understood as a person’s perception of a technology before its use. Acceptance, by contrast, is a person’s perception of a technology after its initial use [46]. Adoption refers to a multistage process that explains a person’s choice to use an innovation. It involves a decision-making process that begins with the perception of the technology and ends with the confirmation of the adoption decision or achievement of permanent use [46-48]. The users of the technology must, therefore, be at the center of the digitization process because without including their values and interests in the acceptability, acceptance, and adoption processes, an innovation cannot be successful [46,48-52]. Therefore, it is necessary to determine the factors that affect these processes. Such a broad knowledge base of influencing factors will enable the development of effective strategies for the implementation of new technologies and will serve as a starting point for tailoring new technologies in a user-centric manner, which is crucial for sustainable use [46]. Research regarding the acceptability, acceptance, and adoption of CAs in health care has gained interest tremendously in recent years and has become a significant field.

It is not only private users and patients who are crucial stakeholders within the innovation process of CAs but also staff in health care organizations. In particular, the attitudes and beliefs of staff are crucial for the introduction of CAs to medical institutions because the establishment of new technologies often fails not because of the nature of the systems but because of the employees [50]. One of the most common reasons for the failure of innovations is insufficient knowledge about the acceptability, acceptance, and adoption processes at the time of introduction [53].

Several studies have indicated that despite the benefits of CAs, there is insufficient acceptability, acceptance, and adoption among those who can most benefit from this technology, namely people with health issues [54]. In addition, this technology is usually associated with poor adoption by physicians [39]. To date, several studies have investigated the factors influencing the user acceptability, acceptance, and adoption of CAs in health care. Some factors, such as performance expectations, effort expectations, trust, and facilitating conditions, have already been determined [55-57]. However, a complete overview of the factors influencing the acceptability, acceptance, and adoption of CAs in health care does not yet exist.

This study presents an overview of the facilitating and hindering factors that influence the acceptability, acceptance, and adoption of CAs from the perspectives of patients and health care professionals. Both groups are considered separately to assess whether the influencing factors differ and to derive recommendations for how CAs in the health care system must be designed so that they are used by both patients and health care professionals. Furthermore, CAs can be sustainably integrated into care only if health care professionals are convinced of their benefits and prescribe or recommend them to patients. From the perspective of health care professionals, it is crucial to differentiate between providers’ perception of the use of CAs for patients and their perception of the use of CAs for supporting their own work. On the basis of the identified influencing factors, which were derived from previous technology acceptance and adoption research, a comprehensive thematic map was developed, providing a visualization of the factors that determine the acceptability, acceptance, and adoption of CAs in health care. This up-to-date literature review that shows the factors influencing the acceptability, acceptance, and adoption of health care CAs will enable the design of effective strategies for the implementation and establishment of the technology and user-centered design of the systems, which will lead to sustainable use. Furthermore, the review can serve as a guide for developers, as it shows how the technology should be designed so that it is accepted and adopted by the target group.

These objectives lead to the following research question: what are the factors influencing the acceptability, acceptance, and adoption of CAs in health care from the perspectives of patients and health care professionals?

Numerous studies have demonstrated that technology acceptance and adoption models are suitable for investigating the factors influencing technology acceptance and adoption in the health care sector [58,59]. These models attempt to explain the adoption process and use of new technologies and share a basic conceptual framework. This framework explains how individual attitudes affect the intention to use and, ultimately, actual use of new technologies. Researchers from a wide range of disciplines have developed various user acceptance and adoption models for understanding acceptability, acceptance, and adoption from the perspective of individuals or organizations. Table 1 shows the important models and theories of individual acceptance and adoption and their respective determinants.

Whereas previous models could explain between 17% and 53% of an individual’s intention to use a technology, the Unified Theory of Acceptance and Use of Technology (UTAUT) can explain approximately 70% of the variance in an employee’s behavioral intention and up to 50% of the variance in technology use in the organizational context. Moreover, the Unified Theory of Acceptance and Use of Technology 2 (UTAUT2) can explain approximately 74% of the variance in stated behavioral intentions and approximately 53% of the variance in technology use in the consumer context [51,52]. Furthermore, both models have been successfully used in the health care sector on several occasions [55,56,58,59]. Thus, these 2 models serve as the initial models for this review and are described in more detail below.

The UTAUT emerged from a review and synthesis of the 8 most prominent user acceptance models identified in a literature review by Venkatesh et al [51]. The reformulated model includes 4 core determinants, namely performance expectancy, effort expectancy, social influence, and facilitating conditions, which directly determine behavioral intention and use behavior. Unlike previous models, the UTAUT also includes 4 moderators (gender, age, experience, and voluntariness of use) that have a moderating influence on the 4 core determinants [51]. In 2012, Venkatesh et al [52] modified the UTAUT for the consumer technology acceptance and use context to form the UTAUT2. Whereas the UTAUT focuses on predicting the intention to use and actual use of a technology primarily in the organizational context, additional constructs and relationships were identified for the UTAUT2 to predict the intentions to use and actual use of a technology in the consumer context. In particular, the determinants hedonic motivation, price value, and habit were added, and the moderating factor voluntariness of use was removed. Therefore, in the UTAUT2, 7 determinants are moderated by 3 factors [52].

An integrative review (IR) was chosen to answer the research question. This approach allows the inclusion of studies with diverse methodologies (ie, experimental and nonexperimental research) [67,68] and can precisely represent the state of the current research literature [69]. IRs are the most comprehensive methodological approach to reviews [70] and have many benefits, including identifying gaps in the current research and the need for future studies, evaluating the strength of the scientific evidence, identifying a conceptual or theoretical framework [69], and analyzing methodological issues of a particular topic [71]. Furthermore, the varied sampling frame of IRs in conjunction with the multiplicity of its purpose has the potential to generate a comprehensive understanding of problems related to health care [68].

To ensure methodological rigor, we used Cooper’s [67] 5-stage IR method modified by Whittemore and Knafl [68]. This five-step approach includes (1) problem identification, (2) data collection, (3) data evaluation (quality appraisal), (4) data analysis and interpretation (data extraction), and (5) presentation of results. At the same time, we used the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) checklist as a guide for preparing the IR, which is available in Multimedia Appendix 2 [72].

This review has been registered on PROSPERO (CRD42022343690).

To identify the relevant literature for this review, a broad literature search was conducted in June 2021 with no date limits and using all fields in 5 databases (PubMed, Cochrane Library, Web of Science, LIVIVO, and PsycINFO). To keep the IR current, a second search was conducted in March 2022. The search terms were derived from the guiding research question. The following 3 keyword groups were set: “conversational agent,” “acceptability, acceptance, and adoption,” and “influencing factor.” Various synonyms within these keyword groups were generated from the Medical Subject Headings terms of the 5 databases, and further synonyms were derived through a web-based search and from previously published literature discussing CAs. Finally, we used an extensive list of 43 search terms (Multimedia Appendix 3). The search strategy was cross-checked with the Guideline Statement for Electronic Search Strategies [73]. In addition, a preliminary search was conducted in each database to ensure the appropriateness and relevance of the adopted keywords because different digital databases use search engines with different requirements.

The PRISMA selection process was used to review publications for inclusion [74]. All studies were assessed against a set of predetermined inclusion and exclusion criteria, which were defined and guided by the research question and purpose of the IR. Studies were included in the IR if they met the following criteria: (1) the language was English or German; (2) the papers were primary studies (3) published until December 31, 2021, and (4) described the acceptability, acceptance, and adoption of CAs and their influencing factors (5) in health care; and (6) the studies adopted a quantitative, qualitative, or mixed methods design (Table 2).

Screening of studies for inclusion was independently performed by 2 authors (MW and MH) in 2 stages: title and abstract review and full paper review. There were 5 disagreements between the 2 reviewers. The discrepancies were discussed between the authors and resolved through consensus.

Data analysis was conducted via the 4-phase process described by Whittemore and Knafl [68]. During the initial phase (data reduction), we extracted the following information from the studies: the perspectives (of patients and health care professionals) on acceptability, acceptance, and adoption; the wording used in terms of acceptability, acceptance, and adoption of the technology; methodology; theory; study area; number of participants; year; country; and the influencing factors. In addition, it was noted whether acceptability, acceptance, or adoption was part of the research question. In the second phase (data display), we converted the extracted data from the individual sources into a table matrix. During the third phase (data comparison), the factors were analyzed in more detail, paraphrased, and assigned to superordinate categories based on the relationships between them and their underlying meaning.

The influencing factors were coded through parallel deductive and inductive approaches. In the deductive approach, we searched for statements reflecting the factors proposed by the UTAUT or UTAUT2. During the inductive coding, we developed categories that were not included in the UTAUT or UTAUT2. To verify the identification and classification of the influencing factors by the first reviewer (MW), a second independent reviewer (MH) checked the identification and classification of the influential factors in 10% (8/76) of the included studies that were randomly selected. There was only 1 disagreement between the 2 reviewers. The disagreement was resolved after a short conversation, and the identification and assignment of the first author was followed. The final phase (conclusion drawing and verification) comprised interpreting the information derived from the previous stages [68].

All the retrieved papers were subjected to a quality assessment. The Mixed Methods Assessment Tool version 2018 was used because it is a critical appraisal tool for reviews that include qualitative, quantitative, and mixed methods studies [75].

One researcher (MW) rated all the identified studies, and a second researcher (JK-N) independently rated 10% (8/76) of the identified studies that were randomly selected. For 1 study, the researchers provided slightly different ratings of quality. However, this difference was quickly resolved and was judged to be sufficiently minor to not question the viability of the other ratings.

Figure 1 illustrates the flow diagram of the database searches and study screenings.

The first database search yielded 602 studies, and the second database search yielded 303 studies. After duplicates were removed, the titles and abstracts of 532 studies were screened, of which 195 were included in the full-text screening.

Finally, 72 studies were identified through a systematic search. According to Whittemore and Knafl [68], complementary searches are essential for an IR to identify the maximum number of eligible primary sources. Therefore, we used a snowballing approach by searching the reference lists of the eligible studies. In addition, we conducted a hand search. Through the snowballing approach we were able to identify 1 more study and by hand search another 3 studies. A total of 76 studies were included in the review.

The results of the appraisal are available in Multimedia Appendix 4 [2,10,12-29,35-38,42,54-57,76-122]. Overall, the quality of the included studies was high. Because the aim of this IR is to provide a comprehensive account of the factors influencing the acceptability, acceptance, and adoption of CAs in health care, the authors decided to include all studies.

The critical appraisal of the papers revealed a minor risk of bias in 11 (14%) of the 76 publications.

The characteristics of the included studies are summarized in Multimedia Appendix 5 [2,10,12-29,35-38,42,54-57,76-122].

Of the 76 included studies, 69 (91%) exclusively focused on acceptability, acceptance, or adoption among patients, and 3 (4%) focused solely on acceptability, acceptance, or adoption among health care professionals. The remaining 4 (5%) of the 76 studies (the studies by Dupuy et al [22], Kowatsch et al [76], LeRouge et al [26], and Potts et al [77]) explored and described the influencing factors from both the patient and health care professional perspectives.

The 76 included papers were published between 2005 and 2021, with most (n=62, 82%) published from 2019 to 2021. The studies originated from 19 countries. Most studies were conducted in the United States (29/76, 38%) and the United Kingdom (13/76, 17%). The sample sizes of the studies ranged from 4 to 16.519. Of the 76 studies, concerning the study design, 21 (28%) studies had a qualitative design, 17 (22%) had a quantitative nonrandomized design, 15 (20%) had a quantitative randomized controlled design, 14 (18%) had a mixed methods design, and 9 (12%) had a quantitative descriptive design.

Furthermore, the included studies were mostly pilot studies with short intervention periods (mostly between 2 and 4 weeks). Among the 76 studies, there were only 1 (1%) long-term study conducted for >12 months [19] and 2 (3%) studies with a timeframe of >6 months [78,79]. Moreover, the studies were mostly laboratory studies conducted in a controlled environment; only 14 (18%) of the 76 papers used and tested CAs in real-world conditions [25,28,37,56,76,78-86]. Only the studies by Sillice et al [19], Baptista et al [78], and Fan et al [79] were long-term studies under real-world conditions.

The reviewed studies displayed a wide variation in the wording used in relation to the acceptability, acceptance, and adoption of technology. Of the 76 studies, 36 (47%) used the term “acceptability” exclusively, 12 (16%) used the term “acceptance,” and 11 (14%) used the term “adoption.” In addition, of the 76 studies, 8 (11%) studies used both “acceptance” and “adoption congruently,” 7 (9%) used both “acceptance” and “acceptability,” and 1 (1%) used both “acceptability” and “adoption.” Similarly, 1 (1%) study used all 3 terms, (“acceptability,” “acceptance,” and “adoption,”) congruently in their descriptions. By contrast, 3 (4%) studies used none of the terms explicitly but only described the users’ perceptions [35,87,88]. These studies were assigned to the term “adoption” [47]. It was also noted that none of the included studies defined the terms used. Overall, the included studies described a high level of acceptability, acceptance, and adoption of CAs in health care.

Furthermore, only in 10 (13%) of the 76 studies, “acceptability,” “acceptance,” “adoption,” or a synonym was part of the research question or primary research objective [13,21, 22,27,55,56,76,83,89,90]. In 44 (58%) of the 76 studies, “acceptability,” “acceptance,” “adoption,” or a synonym was part of the secondary objectives. Of the 8 studies with an established model to measure the acceptability, acceptance, and adoption of health CAs, 6 (75%) referred to the Technology Acceptance Model [10,86,90-93], and 2 (25%) referred to the UTAUT2 [55,56].

The CAs of the included studies targeted various health domains, as shown in Textbox 1. The textbox shows which health domain was supported by a CA and where it was used within a common medical care pathway. The most CAs dealt with mental health issues and covered the complete care path.

The objective of this IR was to identify the factors that influence the acceptability, acceptance, and adoption of CAs among patients and health care professionals. We identified 13 factors that influence the acceptability, acceptance, and adoption of CAs among patients and 10 factors that influence the acceptability, acceptance, and adoption of CAs among health care professionals.

We found that performance expectancy and effort expectancy are the most studied factors influencing the acceptability, acceptance, and adoption of CAs in health care. The findings are consistent with the literature on human-computer interaction (HCI). Perceived ease of use and perceived usefulness are described as key factors in predicting the use of technologies in general and CAs in particular [49,127]. Overall, both health care professionals and patients clearly recognize the benefits of CAs in health care, which has already been shown in a number of studies [41,128]. In addition, studies demonstrated that the health care provided by a CA is comparable with that provided by human physicians [44,45]. Thus, the systems represent a cost-effective alternative to the classic therapy option with the same benefits [39,95].

One of the most interesting findings of the analysis was that, in addition to performance expectancy and effort expectancy, anthropomorphism, trust, perceived risk, and working alliance have been identified as having a decisive influence on the acceptability, acceptance, and adoption of CAs in health care and have not previously been considered in UTAUT or UTAUT2.

In accordance with the literature on the theory of anthropomorphism [124], this IR found that patients attribute human-like characteristics to health CAs and try to interact with them as if the systems were human. HCI research has also found that individuals interact with internet-based agents as if they were humans, even when they know that they are computer programs [129]. In addition, previous work has shown that anthropomorphism has a positive effect in terms of continued use and increased satisfaction with the technology [124]. Moreover, previous work on CAs has already indicated that perceived anthropomorphism can influence CA acceptability, acceptance, and adoption [130].

Nevertheless, it was found that the perceived anthropomorphism could also trigger fear and discomfort in some patients. They perceived the CA as creepy, scary, and strange. In the HCI literature, this is known as the “uncanny valley effect.” The uncanny valley theory states that a technology that appears almost human can evoke negative affective reactions in users [131]. The findings obtained are also consistent with the results of other studies on this topic. Although some studies have reported positive effects of anthropomorphic CAs [124], others have shown that anthropomorphism can lead to frustration, confusion, and even a sense of eeriness [132]. Another criticism of anthropomorphism is that users can be deceived into thinking that they are interacting with a real person instead of a system [16]. Therefore, a CA should always be labeled as a machine.

Furthermore, our results support previous literature on trust by showing that trust is not a monolithic concept but must be differentiated into “trust in the provider” and “trust in the technology” from the patient’s perspective. Whereas trust in the provider refers to patients’ beliefs about the provider’s benevolence, integrity, and competence, trust in the technology refers to patients’ beliefs about the system’s benevolence, functionality, helpfulness, and reliability [126,133]. With regard to health care professionals, it was found that they also have little trust in their patients to use the CA correctly and to interpret the given information correctly. Thus, our results show that trust is represented by the categories “trust in the technology,” “trust in the provider,” and “trust in the patient” from the perspective of the health care professionals. To increase the trustworthiness of CAs, it is suggested in relation to research on AI-driven intelligent systems that responses be presented in a meaningful, understandable, and trustworthy format. In addition, users should be provided with a variety of system-related information, including data on the reliability and performance of the system and the source used for the response output. This should enable users to better understand the information displayed and its origin and then decide whether to trust the technology’s recommendation [134]. It is further suggested that credibility can be demonstrated to users through expert vocabulary and appropriate presentation [133]. Nonmedical studies have also demonstrated that credibility in the form of systems’ functionality, capability, reliability, and benevolence can predict the acceptability, acceptance, and adoption of wearable technologies such as CAs [135]. Furthermore, it was shown that for trust building, the user should have a positive impression of the technology. These impressions are influenced by static and dynamic features. Static features include the appearance of the system, and dynamic features include the verbal and nonverbal behaviors of the system [136]. Moreover, in line with the literature on trust, the results show that building trust in automated systems is a major challenge for developers [137]. Furthermore, it is assumed that patients will use CAs only if they trust them. These explanations show that trust is one of the key factors influencing CA acceptability, acceptance, and adoption [36]. Therefore, it is suggested that trust in health CAs should always be systematically assessed before deployment. However, standardized and validated scales to measure trust are lacking, especially in medicine [138].

Another key barrier to the acceptability, acceptance, and adoption of health CAs is the perceived risk of the technology, stemming from the uncertainty around the protection of personal data and the risk to users’ lives and well-being. Concern about data privacy and the fear of misuse of sensitive information are key barriers to the acceptability, acceptance, and adoption as well as to the widespread use of digital health applications [139]. Studies have shown that privacy concerns can be addressed by the automatic transfer of data from an electronic health record and the regular addition of information by health care professionals. In addition, concerns may be addressed by explaining the measures succinctly and presenting them in layperson’s terms [140].

The literature on digital health applications also frequently discussed whether patient safety is compromised [141] and who is responsible if the CA misdiagnoses someone [4]. The results show that health care professionals fear a safety risk not only for patients but also for themselves. They fear that CAs will play such an important role in the future that they could replace human workers and compromise the quality of health care. We believe that this fear is one of the key barriers to the acceptability, acceptance, and adoption of CAs by health care professionals. In line with the literature, our results clearly show that the development of CAs is still in the early stages, is rudimentary, and thus does not jeopardize jobs [139]. Patients are more willing to share confidential information with a CA than with a health care professional because of its anonymous and nonjudgmental nature of the interactions. However, the preferred use of the systems is for minor illnesses. For more serious conditions, patients prefer to seek advice and treatment from a physician [44,45]. Thus, the use of CAs is purely supportive and does not jeopardize employment. This should be clearly communicated to increase the acceptability, acceptance, and adoption of the technology among health care professionals.

A therapeutic relationship is crucial for the success of a treatment [142]. Such a relationship is the result of empathy, care, and trust and can significantly improve the benefits of a health interaction [143]. Empathy is the most important factor in building a working relationship [144]. We were able to identify the factors of empathy, care, and trust as crucial for the acceptability, acceptance, and adoption of CAs in health care. It has already been demonstrated in some studies that a working alliance can be formed between a CA and a user [4,33,34,43]. For establishing and maintaining a relationship between a patient and a CA, memory of past interactions and variability in verbal and nonverbal responses are crucial elements. This finding is consistent with previous research and shows that, for the correct application of relational behaviors, it is necessary to talk about the past and the future [145] and the time spent apart [146]. Bickmore et al [15] suggested designing health CAs such that interactions are initially relatively distant and professional but gradually become more personal, social, and familiar over time. In addition, systems should have a sense of humor as well as empathy and talk to the user about the present relationship to maintain it [34].

Another crucial barrier to the acceptability, acceptance, and adoption of CAs is their lack of comprehensibility and limited communication capabilities. The literature showed that language skills are a major problem and should be urgently improved [3]. Owing to language limitations, CAs currently use predetermined response options because, unlike free-text entry, they can ensure data validity and accuracy and minimize speech recognition errors. This approach is particularly important in a health-related context, as the multiple-choice input modality avoids potentially dangerous effects of misunderstandings due to ambiguous utterances about medical topics in unrestricted text and speech input. At the same time, it clearly communicates to users how they should respond to the system’s output and ensures that the system can understand and process input with high accuracy. It also enables the CA to be more easily accepted and used by people with different computer and language skills [16,147]. However, our analysis shows that many patients did not want a user input restriction while communicating with systems. Instead of choosing between predefined answers, they would like to be able to answer with a free-text entry to describe their health complaints as precisely as possible.

Furthermore, most patients preferred voice-based communication with a CA over text-based communication. The preferred method of communication of CAs was also discussed controversially in the literature. Even within the definition of CAs, there is no consensus on the preferred mode of communication. The advantages of text-based communication are, for example, that text can be indexed, searched, and translated and that it can be easily corrected or improved after completion. Proponents of acoustic communication are of the opinion that speech is more natural and faster than text. In addition, the use of speech can enhance the perceived personality of a CA. Furthermore, systems that allow acoustic communication can also be used by patients with low or no literacy skills [40,148]. Moreover, we found that nonverbal communication also has a decisive influence on the acceptability, acceptance, and adoption of the systems. Nonverbal cues such as facial expressions, gestures, posture, and body movements have a significant impact on guided communication, as they convey empathy, thereby strengthening the therapeutic alliance and trust relationship between patients and CAs [30,33,34,43]. The 55-38-7 rule proposed by Mehrabian and Ferris [31] shows the importance of nonverbal communication and behavior. Communication can be improved only through a combination of verbal and nonverbal behaviors [30]. We believe that all types of communication will be important in health care in the future. Whether written or oral communication is advantageous will depend on the situation in which the CA is used. For example, whereas an oral dialog with a CA may be easier for a human who is severely injured or paraplegic or a human who is illiterate, a written conversation may be beneficial for a prescription transfer or a patient with speech impairment.

One of the main criticisms of CAs in the literature is that they would not be able to develop empathy, recognize users’ emotional states, or tailor their responses to them. A lack of empathy can affect the use of CAs in the health care sector [143]. To increase the acceptability, acceptance, and adoption, as well as effectiveness, of CAs among patients, it is, therefore, important that the systems have the same interpersonal and social characteristics as health care professionals. In addition, empathic responses help create a trusting relationship between the technology and the user, which guarantees continuous and long-term use of the system and increases the benefits for patients [16]. Consistent with the broader literature, our results show that CAs can be empathic [16,33,43]. Some studies even showed that the empathic abilities of CAs can be compared with those of a real person [149]. In health care, empathy as part of anthropomorphism is critical for the success of CAs [38]. It was found that the visualization of a CA in the form of an avatar makes it more credible, comfortable, sympathetic, and useful than a CA without an avatar [33].

The COVID-19 pandemic has had a significant impact on normal health care delivery and has demonstrated the urgent need for alternative approaches that can overcome geographic, temporal, and organizational barriers. The pandemic resulted in limited access to outpatient clinics, and the high rate of infection posed significant challenges to medical facilities, which affected the delivery of health services [2,110]. This situation has clearly demonstrated that the short-term unavailability of health services can occur even when rapid access to services is basically guaranteed. In this regard, technological systems such as CAs are a good alternative for the continued provision of quality care. It has been shown that CAs can improve and facilitate access to health care [150]. In addition, there are concerns about what happens once the internet connection is lost or individuals do not have the necessary resources such as a smartphone or internet access [57]. Services that can be accessed only through technology may lead to a digital divide and inequity in health care. This would limit access to health services, potentially for the very people who need the services most. For example, digital searching for health information is uncommon among older adults and other underserved groups. However, it should be noted that digital technologies expand the availability of health information and resources to many individuals and improve the quality of care [57,151]. Therefore, we propose that health care providers always offer traditional access to health care services alongside technology to provide quality care for everyone and prevent a 2-tier society. Solutions should also be sought to improve the access to digital resources such as the internet that are necessary to access emerging health technologies.

Consistent with Ling et al [152], we found that user-related factors influence the acceptability, acceptance, and adoption of CAs. These include the demographic factors age, gender, origin, and education level as well as the UTAUT2 factor user experience with the technology. Regarding the factors age, gender, and education level, we found different results within the analyzed studies as to whether they influence the acceptability, acceptance, and adoption of health CAs among patients. Other studies on this topic also provided different findings. Although some studies demonstrated the presence of these factors, other studies were unable to do so [135,147]. Furthermore, origin was found to influence the acceptability, acceptance, and adoption of CAs. However, overall, this is an understudied area. Little is known about ethical differences in technology acceptability, acceptance, and adoption. However, in line with previous research, our results show that it is crucial to tailor the technology to the target population and its cultural characteristics [83].

Moreover, it was found that the identified influencing factors influence each other and cannot always be clearly separated. At the same time, our results indicate that the importance of an influencing factor also depends on the purpose of the CA used and the health domain concerned. Multimedia Appendix 8 provides a summary of influencing factors by health domains and health categories (ie, aggregated domains). Whereas in the categories “mental health” and “specific diseases,” anthropomorphism is the most important factor in addition to performance and effort expectancy, credibility and the severity and type of health issue are crucial for CAs as general health advisers and promoters. In the category “pregnancy care and healthcare for children,” by contrast, hedonic motivation is the key influencing variable along with performance and effort expectancy. The importance of the individual determinants based on the purpose of a CA application is, therefore, understandable. However, owing to the small number of studies per health domain and category, this can only be generalized to a limited extent. The mutual influence and not-always-clear separation of the influencing factors as well as their variability and importance depending on the health care domain make the research on the acceptability, acceptance, and adoption of CAs in health care so extensive.

As with all studies, this IR has some limitations. One potential limitation is related to the search strategy. It is possible that not all studies on the topic were found despite our comprehensive search strategy, as studies may have discussed the acceptability, acceptance, or adoption of CAs and the influencing factors but used different terms than those we found. In addition, this review included studies published only in English and German, and this approach may have excluded relevant evidence published in other languages. Furthermore, the IR included only primary studies that had already been published, which also excluded relevant studies such as gray literature.

For quality appraisal, we followed the guidelines for rapid reviews [153,154]. A rapid review is a form of knowledge synthesis in which components of the systematic review process are simplified or omitted to produce evidence-based information in a timely manner [155]. As a result, the screening of the studies for the quality assessment of the papers was fully performed by only 1 researcher. A second researcher assessed only 10% (8/76) of the studies. Nevertheless, the expedited process may have introduced biases in quality assessment.

In addition, as the original studies did not consistently define and describe whether they analyzed acceptability, acceptance, or adoption, it was impossible for us to differentiate between these 3 outcomes in our synthesis. Hence, we cannot provide an answer to the question of whether some factors have been researched more frequently or are more influential for one of the outcomes than for the others.

Finally, the findings regarding acceptability, acceptance, and adoption among health care professionals are almost impossible to generalize, as we could only find and evaluate 7 studies on this topic. Owing to the rapid increase in the research literature on this topic, it is possible that new findings already emerged during the preparation and publication of our results and that the review, therefore, no longer reflects the current state of research.

Despite these potential limitations, this IR has several strengths. To our knowledge, this is the first review to provide a comprehensive picture of the acceptability, acceptance, and adoption of CAs and their influencing factors in health care. We described the factors influencing the acceptability, acceptance, and adoption of CAs in health care from the perspectives of patients and health care professionals and created a thematic map that clearly summarizes the findings. Furthermore, the IR follows the same scientific rigor as primary research in that we used Cooper’s [67] 5-step IR method modified by Whittemore and Knafl [68] for its construction. The review was developed, conducted, and reported in accordance with the PRISMA selection process, which allowed us to produce a high-quality review [74]. A total of 5 well-known and frequently used databases in the field of health were searched to retrieve as many studies as possible. The keywords for the search terms used for this purpose were derived from the main research question. Synonyms for the identified keywords were generated using the Medical Subject Headings terms of the 5 databases, a web-based search, and previously published literature on CAs. Freehand searching and forward-backward reference list checks allowed us to identify additional literature missed by the database search and minimize the risk of publication bias. As no restrictions were made with regard to study design, study setting, and country of publication, this review can be considered comprehensive.

This IR provides the first comprehensive overview of the acceptability, acceptance, and adoption of CAs in health care and their influencing factors from the perspectives of patients and health care professionals. From the results, it is clear that the acceptability, acceptance, and adoption of CAs from the perspective of health care professionals are significantly underresearched. Other reviews have also found that few studies on CAs have focused on health care professionals [40]. Therefore, future research should urgently explore the acceptability, acceptance, and adoption of CAs among this user group. For this purpose, a survey could be designed based on our theoretical model to confirm the identified influencing factors and determine new ones. Furthermore, health care professionals should also test currently available CAs and provide feedback. In particular, it is crucial to explore the acceptability, acceptance, and adoption of CAs and their influencing factors from physicians’ point of view. Moreover, physicians will only recommend or prescribe CAs if they accept and adopt the technology and are convinced of its benefits. In addition, we demonstrated that health care professionals’ opinions about CAs significantly influence patients. With knowledge about acceptability, acceptance, and adoption among health care professionals, CAs could be sustainably established in health care.

We succeeded in creating a comprehensive thematic map of the factors influencing the acceptability, acceptance, and adoption of CAs. However, the influence of the identified factors on acceptability, acceptance, and adoption as well as the interrelationship among them was not quantitatively validated. Future studies could build on the theoretical model and examine the relative influence of the factors on acceptability, acceptance, and adoption and the dynamics among the factors. Furthermore, the results show that the influence of facilitators and barriers depends on the intended use of a CA and the health domain in which it is used. However, nothing about the strength and importance of the identified factors was mentioned in the analyzed studies. Therefore, future research should also investigate the importance of the individual factors and their interactions with each other for individual areas of care.

We found that CAs have been tested almost exclusively in controlled environments that do not simulate the realistic interactions in clinical practice. It has already been demonstrated that the environment in which the interaction occurs influences technological acceptability, acceptance, and adoption [127,156]. The broader literature also criticized the fact that, to date, most studies have examined the use of CAs in controlled environments rather than in real-world contexts [148,157]. Moreover, most studies into the acceptability, acceptance, and adoption of CAs in health care are short-term studies. However, it has been shown that factors such as habit only develop when the technology is used over a longer period. Therefore, we consider it necessary that CAs be increasingly tested in real environments and over the long term in the future.

Consistent with the current review of Camile et al [46], we noted that, in relation to the technology, researchers attach different meanings to the terms “acceptability,” “acceptance,” and “adoption” and often use them synonymously without referring to established models and definitions from the literature. None of the included studies defined the terms used appropriately or distinguished them from each other. The definitions are often misunderstood, or researchers establish their own definitions. The inconsistent use of the terms “acceptability,” “acceptance,” and “adoption” makes it immensely difficult to compare the results of these studies. For future research, we, therefore, consider it necessary to follow the definitions and established models from the literature on acceptability, acceptance, and adoption, which clearly show the differences among the terms, to achieve consistency, which will allow comparisons across studies and the development of targeted implementation strategies.

In this review, we identified 13 factors that influence the acceptability, acceptance, and adoption of CAs among patients and 10 factors that influence the acceptability, acceptance, and adoption of CAs among health care professionals. On the basis of the identified influencing factors shown individually for acceptability, acceptance, and adoption, a comprehensive thematic map that explains the acceptability, acceptance, and adoption of CAs in health care was created. Overall, a high level of acceptability, acceptance, and adoption of CAs in health care was observed. This review shows the variety and complexity of influencing factors. Thus, it presents a comprehensive set of factors that can be implemented, improved, or steered to increase the acceptability, acceptance, and adoption of CAs in health care.

To the best of our knowledge, this IR extends the literature by providing the first overview of the research on the acceptability, acceptance, and adoption of CAs in health care. The findings of this review can, therefore, serve as the groundwork for future implementation studies of CAs in health care. Future research should focus on exploring acceptability, acceptance, and adoption from the perspective of health care professionals. Furthermore, it is crucial to test already developed CAs under real conditions and through long-term studies.