Embodied Conversational Agents for Chronic Diseases: A Scoping Review

Background: Embodied conversational agents (ECAs) are computer-generated characters, which interact with users face to face through verbal and nonverbal behavior to establish a harmonious partnership. Therefore, ECAs are expected to achieve long-term use of the tool in chronic diseases. Objective: This scoping review aimed to review the current practice of reporting the design and use of ECAs in the field of chronic diseases. Methods: Arksey and O’Malley framework guided the conduct of this review. Five English databases(PubMed, Embase, Cochrane Library, CINAHL, and Web of Science) and three Chinese databases(CNKI, WAN-FANG, and SinoMed)were searched with the search terms ECA and associated synonyms in October 2022. Two independent reviewers selected studies and extracted the data. Results: Literature search found 9815 articles, 35 of which met the inclusion criteria. Among these studies, 15 studies originated from the United States, and the largest number of articles were published between 2020 and 2022 (n=20). The reported ECAs covered a wide range of chronic diseases, focusing on cancers, type 2 diabetes, substance use disorders, and psychological disorders, mainly for self-management (n=14) and promoting screening (n=7). Since most of the included studies were still in the development and piloting stages, the design features of ECAs were not comprehensive and evaluation outcomes were inconsistent. We generally concluded that the gender of ECAs was mainly female, and communication modalities were text, voice, and nonverbal, and most studies reported acceptability. Conclusions: Owing to their ability to build and maintain an empathic relationship, ECAs are increasingly being considered for chronic disease management. However, the literature on ECAs for chronic conditions is scarce, and the


Preprint Settings
1) Would you like to publish your submitted manuscript as preprint?Please make my preprint PDF available to anyone at any time (recommended).Please make my preprint PDF available only to logged-in users; I understand that my title and abstract will remain visible to all users.
Only make the preprint title and abstract visible.
No, I do not wish to publish my submitted manuscript as a preprint.2) If accepted for publication in a JMIR journal, would you like the PDF to be visible to the public?
Yes, please make my accepted manuscript PDF available to anyone at any time (Recommended).
Yes, but please make my accepted manuscript PDF available only to logged-in users; I understand that the title and abstract will remain v Yes, but only make the title and abstract visible (see Important note, above).I understand that if I later pay to participate in <a href="http

Background
With the rapid aging of the population and lifestyle changes, chronic diseases have become a significant global public health problem, arousing great concern from all walks of life.In 2018, 51.8% of American adults suffered from at least one chronic disease, and 27.2% had multiple chronic diseases [1].In China, chronic diseases accounted for 86.6% of total deaths and approximately 70% of the total burden of diseases [2].Given the prolonged duration and severe health damage of chronic diseases, patients often require assistance in long-term care [3].To relieve this growing burden, particularly in healthcare services and related costs, advancements in network communication technology have shown promise in improving the availability and quality of support services.The electronic health (eHealth) applications allow for remote patient monitoring and provide patient-tailored support in their home settings.However, many eHealth applications faced the problem of their actual use decreasing after several weeks [4].This decline may be attributed to the fact that the majority of existing eHealth applications provided such support in the form of plain text or via a text-based question-answer module, while person-to-person interaction remains one of the best ways to communicate health information [5].Face-to-face consultations, with the use of verbal and nonverbal behaviors such as empathy and immediacy, can foster trust and satisfaction among patients, leading to better health communication and understanding [6].
An embodied conversational agent (ECA) is a computer-based dialogue system with a virtual embodiment that simulates a face-to-face conversation with human-like physical properties, including verbal and nonverbal behavioral cues (eg, speech, facial expressions, and gestures) [7].Compared with a static character image or a text-only display, the interactive, conversational modes of communication used by ECAs may potentially improve engagement by providing additional motivational and emotional support [8,9].In healthcare, ECAs have been designed to assist with various tasks such as providing diabetes self-management education [10], promoting cancer screening [11], and delivering cognitive behavioral therapy for depression [12].Despite the exciting potential for using ECAs for health purposes, the use of ECAs could be ineffective or even have unintended negative consequences if the design, including visual appearance and intervention content, did not meet the user's expectations [13].Research has shown that design decisions related to the look and feel of ECAs significantly influence users' psychological and emotional responses and engagement with applications [14].However, how ECAs should be designed and used to maximize effectiveness in the context of chronic diseases is still unknown.Therefore, it is crucial to systematically review the development and evaluation of ECAs in a specific context to optimize the design of ECAs to provide a positive user experience and promote engagement.
Currently, there is a lack of comprehensive reviews on the development and evaluation of ECAs specifically in the context of chronic diseases.While there have been literature reviews on conversational agents in eHealth, they often focus on the impact of ECAs rather than the design processes involved.For example, Kramer et al [15] conducted a scoping review of ECAs in a healthy lifestyle and pointed out that the design of an ECA could have a major effect on both impact and uptake.However, reporting on the design activities and their results was generally incomplete or missing.Similarly, two other reviews [16,17] identified that the most common design features of ECAs and their impact on user perception but ignored the design activities of ECAs.Another scoping review by Provoost et al [18] aimed to provide an overview of the the technological and clinical possibilities of ECAs but only for patients with mental disorders.Therefore, a comprehensive literature review focusing on the design processes of ECAs in the context of chronic diseases is needed.
The existing studies have primarily focused on design features rather than design processes and have examined a broader context beyond health or a specific subarea of health such as clinical psychology.In this study, we aim to review relevant studies to understand how ECAs have been designed and evaluated specifically in the context of chronic diseases.After a preliminary exploration of the relevant literature on ECAs to determine the review method, it was found that the traditional systematic review or meta-analysis method seemed unsuitable due to the variability in populations, study designs, and measured outcomes.Compared with the traditional systematic reviews, the scoping review covers a broader range of topics, in which many different study designs may be applicable, and the quality evaluation of the included research is not emphasized [19].Therefore, we adopted the scoping review method, which provided a clear and systematic means to outline this large and diverse body of literature, using rigorous methods to minimize bias [20].

Objectives
In this study, we seek to undertake a scoping review focused on the development and evaluation of ECAs in the context of chronic diseases.In particular, the aims of our scoping review are as follows: (1) provide an overview of all the studies about developmental practices of ECAs for chronic diseases; (2) summarize design and design processes of ECAs in chronic diseases; (3) identify evaluation and outcomes reported in studies.Conducting this scoping review will benefit both technology developers and healthcare professionals.For technology developers, the review will provide a comprehensive understanding of the different approaches and techniques that have been utilized in the development of ECAs for chronic diseases, enabling them to develop a more intelligent ECA that provides a natural experience for users.For healthcare professionals, the review will offer actionable advice that help them better manage and provide medical services using ECAs, ultimately improving patient outcomes.

Study Design
The Arksey and O'Malley framework for scoping reviews was adopted [19].The main five stages were as follows: (1) identifying the research questions, (2) identifying relevant studies, (3) study selection, (4) charting the data, and (5) collating, summarizing, and reporting the results.We followed the process outlined in the published protocol and followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) (see Multimedia Appendix 1) [21].

Identifying the Research Question
This study mainly discussed the following questions: (1) What are the basic characteristics of the included studies?(2) How to design ECAs to guide self-care of chronic diseases?(3) How to evaluate the impact of ECA interventions for chronic diseases?

Identifying Relevant Studies
The databases used to locate the relevant studies were as follows: PubMed, EMBASE (Excerpta Medica Database), CINAHL (Cumulative Index of Nursing and Allied Health Literature), Web of Science, IEEE Xplore Digital Library, and ACM (Association for Computing Machinery) Digital Library.These databases were chosen as they cover relevant aspects in the fields of health and information technology and have been used in other systematic literature reviews covering similar topics [22,23].The search terms were identified based on a preliminary literature scan and the opinions of a research librarian to ensure a comprehensive search for relevant studies.The final search terms included an extensive list of items describing the constructs "embodied conversational agents" and "health".The complete overview of the search terms for each construct and inclusion criteria implemented by selecting different options and limits during the search are available (Multimedia Appendix 2).An exemplary search strategy is shown for PubMed in Table 1.In addition, we applied the snowball method.The search was limited to English papers published prior to the search date of October 1, 2023.

Study Selection
The results of the search query were uploaded into the NoteExpressX9 reference manager and independently assessed by two reviewers (ZLJ and XTH) to decide on their inclusion based on title, abstract, and full text.Following an initial screening of titles and abstracts, full texts were obtained and screened by two reviewers.If the eligibility of the full text was unclear, any discrepancies were reviewed by an additional author (YL) and resolved in a consensus meeting.
For this review, we included chronic diseases identified by the Public Health Agency of Canada (PHAC), including cancer, heart disease, hypertension, stroke, chronic respiratory diseases (asthma, chronic obstructive pulmonary disease, sleep apnea), diabetes, inflammatory bowel diseases, neurological conditions (eg, Alzheimer's disease and other dementia, Parkinson's disease, traumatic brain injury, and traumatic spinal cord injury), arthritis, and osteoporosis [24].Mental illness was excluded from the list given that support interventions of ECAs for this group may have particularly unique features not generalizable to other chronic diseases.In addition, we decided to include other diseases that need self-care outside the list of PHAC, such as obesity and chronic pain.We included full articles that met the following criteria: (1) adults aged 18 years and older, (2) published in English, and (3) ECAs were made available to the general public (general population or patient).The following exclusion criteria were applied: (1) reviews, editorials, opinions, thesis and conference abstracts were excluded, (2) full texts that were unavailable, (3) ECAs were used for training or educating medical professionals or not used in the context of chronic diseases, and (4) articles did not involve ECAs (computergenerated virtual individuals with an animated appearance to enable face-to-face interaction between the user and the system) [25].

Charting the Data and Collating and Summarizing the Results
Data extraction was conducted independently by two reviewers (ZLJ and ZQW) using an Excel (Microsoft) spreadsheet.Any discrepancies in the extracted data were discussed between the authors and resolved through discussion and consensus.Extracted data included: (1) article information, (2) study information, (3) details about the ECAs (eg, identity, communication modality, and personalized content), and (4) evaluation outcomes.The content of concepts could be predefined based on the study by Kramer et al [15] (see Multimedia Appendix 3).In cases where an article included multiple studies, data extraction was carried out only for the studies that met the eligibility criteria.If multiple eligible studies were included in one article, the data were extracted separately.Once all the study data had been collected, we conducted a thematic analysis and categorized them into three main topics.The first topic describes the identities (including ECA's names, roles, and appearances), communication modalities, and personalization in intervention content and delivery.
The second topic focuses on the technologies and theories or principles used.The third topic describes the evaluation measures and outcomes.

Articles retrieved
The initial search identified 6332 references in October 2023.After the removal of duplications, there were 4341 references left.The titles and abstracts of these references were screened by both reviewers, resulting in the exclusion of 4066 references.After further evaluation, 245 articles were excluded and the last 30 articles were considered eligible for a comprehensive review.Additionally, 4 more studies were found through snowballing [26][27][28][29].This resulted in a total of 36 studies as 2 papers [30,31] included 2 studies each.We described the search process and outcome in Figure 1.
Regarding the communication modalities of ECAs, 14 agents were able to communicate with users through verbal and non-verbal behavior [10,27,30,32,33,35,36,38,39,41,45,54,55,57].Non-verbal behavior included facial displays of emotions, gaze shifts, eyebrow raises, head nods, body posture shifts, and hand gestures [30,37,53].The specific presentation of these conversational modalities was detailed in 12 studies.For example, one study reported that an agent with 2D animation would blink her eyes every 10 seconds and move her mouth for a fixed period after a new sentence appeared on the screen [40].Another study mentioned that the virtual human had an idle breathing animation in a sitting pose and featured high-fidelity voices recorded by professional voice talents of the same race and gender [31].To address communication shortcomings, such as lower eyesight accuracy or hearing impairment, the development of the interface took into account the needs of patients.This included customized speech speed [33] and text captions of the audio with the recommended font size [31].For safety concerns, user contributions to the dialogue were fully constrained and made by selecting an utterance from a menu [30,43,53].Users had the option to respond to the agent by speaking, inputting text, or touching an option on the screen in 3 studies [10,26,37].Only one ECA was designed to portray a listener responsive to the respondent's nonverbal speaking behavior [54].

Personalization in Intervention Content and Delivery
Personalization was a common feature in the ECAs used in the studies to customize the content and delivery to suit individual users.This involved addressing users by their names and appropriate time contexts [34,52].In addition, reminders, warnings, or alerts were provided based on individually reported data [55], and feedback on current progress toward set goals was given [36].In one study, users could respond to her queries using forced-choice text options that would trigger different responses, allowing the system to interact responsively in a personalized manner with the users [57].The personalization in other studies was achieved based on various channels of information.For example, 3 studies personalized the script logic based on the clinical targets provided by users' healthcare professionals and users' responses during the interactions [10,26,37].To tailor the therapy for each patient, the ECA utterances, the patient's responses, custom goals, and overall objective metrics such as time in the simulation were stored in a Structured Query Language (SQL) database [47].Only two studies provided dynamic adaptive genetic counseling for breast cancer based on the user model's current state and the discourse context [30,53].

ECA Technology and Theories or Principles
From a technological perspective, the physical appearances of ECAs were primarily created using 3D character modeling and animation software, such as the Unity3D game engine [30,35,38,39,47,53], I-Clone [28], and Adobe Fuse [31].Only one study used 2D animation implemented with scalable vector graphics and Hyper Text Markup Language (HTML) animations [40].In terms of communication modalities, speech recognition technology was used in 6 studies to allow users to answer the agent's questions orally [10,26,37,38,41,42], while speech synthesis technology was used to generate the agent's spoken responses [30,34,35,39,41,42,52,53,56].For example, a text-to-speech software, such as Speech2Go, was used to convert written dialogues into audio files [39].Nonverbal behaviors of the ECAs were generated by the Behavior Expression Animation Toolkit (BEAT) text-to-embodied speech engine [30,35,47], and the LipSync Generator was used to synchronize the agent's lip movements with the spoken words [39].Motion capture technology was used in 2 studies to record the voice and gestures of a real person, adding a level of realism to the ECA's behavior [28,38].Regarding the dialogue management, various approaches were used.These included using a rules engine to determine the agent's responses based on contextual information [39], a hierarchical transition network-based dialogue engine [35,47,53], and a scenario manager based on decision trees [38].In addition, 3 studies used voice recognition with pre-scripted conversational elements and a sophisticated script logic [10,26,37].

Evaluation Measures and Outcomes
The studies included in this review reported on both the acceptability and effectiveness.Acceptability refers to the emotional attitude towards new digital health interventions, usage intentions, actual usage, and satisfaction [58].Effectiveness refers to the impact of ECA-based intervention on health-related outcomes.4 studies only described protocols [33,[55][56][57], which were not considered in this section.
Other instruments used included the Acceptability E-scale [38], the Almere model [41,42], the closeness scale [41,42], Technology Acceptance Model [47], Computer Self-Efficacy scale [27], Working Alliance Inventory [32], and ECA Trust Questionnaire [38].For example, one study assessed participants' perceived ease of use of the system using a 24-item scale adapted from the Unified Theory of Acceptance and Use of Technology [27].Another study used the Portuguese version of SUS and calculated an aggregate average score of SUS was 73.75, corresponding to a borderline rating of excellent [39]. 2 studies on dementia illustrated that ECA Anne received a mean score of 66.2 [42] and 67.1 [41], respectively.In addition to questionnaires, customized items were used in 14 studies to assess users' satisfaction with ECAs and the overall systems [10, 27, 29-31, 34, 36, 38, 40, 43, 45-48, 53, 54], while interviews and focus groups were conducted in 18 studies to explore more topics [10, 11, 26, 28-31, 35, 39-44, 49-52].User satisfaction concerns items related to constructs such as liking, trust, ease of use, and desire to continue using the ECA, for example, "How much did you like Tanya?" [30].Objective measures of user engagement with the ECAs were reported in 8 studies [32,34,36,37,[40][41][42]52].These measures included the number of log-ins to the agent application, the time and number of interactions with the ECAs, and the time of program use.For example, one study showed that the time for participants to use relationship agents ranged from 3 to 30 days and the number of log-ins to the relational agent ranged from 4 to 43 [34].Telemetry data were used in 2 studies to detect problems and evaluate the status quo [41,42]. 2 studies measured usage over time, all showing a decrease [37,40], for example, "the program use, including the number of chats and number of blood glucose uploads, reduced over time of the program access".
Participants in the adaptive condition had significantly greater knowledge gain than participants in the non-adaptive condition and the control condition [53].3 studies assessed changes in quality of life using questionnaires [34,37,52]. 2 studies [37,52] found a positive difference in quality of life levels between participants who engaged with the ECA and those who did not.However, there was no significant difference in HbA1c (glycated Hemoglobin) change between participants in the intervention and control groups [37].For motivational outcomes, 3 studies assessed changes in users' motivation [27,32,34].Self-efficacy [27,32] and patient activation [34] were assessed using questionnaires.Symptom improvements were assessed in 3 studies using questionnaires [37,47,54].The results showed that the use of ECAs led to a greater reduction in pain interference and a marginally greater reduction in pain intensity compared with standard interviews [54].Safety was assessed in one study, where participants were asked standardized questions about adverse events such as falls, diseases, injuries, and the use of any medical services [36].

Principal Findings
This scoping review targeted specifically at ECAs applied for chronic diseases in healthcare, which was aimed to inform technology developers and healthcare professionals of the technological possibilities and the evidence base.Our scoping review identified a total of 32 studies and 4 ongoing clinical trials, with the majority of papers published from 2020 onwards.The most commonly reported chronic diseases were cancer, atrial fibrillation, and type 2 diabetes.The review found that ECAs were predominantly defined as female coaches or counselors, interacting with users through voice and nonverbal behavior.In addition, multiple technologies and theories were applied in the design activities of ECA-delivered interventions.A combination of effectiveness and acceptability was typically assessed.Results from the studies that ECAs have the potential to enhance engagement in self-care of chronic diseases, although the evidence on their effectiveness remains inconclusive.
The identified studies were not geographically diverse, with 75% of them conducted in the United States of America and none based in Africa or Asia.This lack of diversity in research locations limits the generalizability of their findings, as they are embedded in Western cultures.Given the global prevalence of chronic conditions and the need for healthcare system-specific solutions, future research should strive to include diverse geographies to ensure the relevance of interventions in different healthcare systems.Among the included studies, 6 studies explored stakeholders' opinions.It has been shown that there were very positive relationships between homecare providers' and patients' perceptions of virtual agents [38], mirroring the findings of Heerink et al [59], which reaffirmed the finding that social influence played an important role in user acceptance of a social agent.In addition, people who were retired, highly educated, and engaged with the app were overrepresented in some studies in the interviewed sample of participants [10,26,39].Health literacy is relevant to the development, accessibility, and successful implementation of eHealth.eHealth interventions focused on health literacy have the potential to reduce disparities in vulnerable populations, where limited health literacy is more prevalent [60].To incorporate interventions into clinical practice effectively and ensure widespread adoption, it is necessary to identify the experiences and needs of stakeholders and users who are less autonomous and less experienced with technology.
Due to the diversity of the design activities reported in the studies, it can be challenging to draw general findings.However, some trends can be observed regarding the identities of ECAs.One common design feature is giving an ECA a name, which may enhance its social presence.
Through the lens of the Technology Acceptance Model, social presence, the general sense of being with another person, is relevant to patients accepting agents because perceptions of social presence can lead to a desire for future interaction [44].Another design feature is portraying an ECA as a coach, indicating that a relaxed and nonjudgmental role may be more successful in building a supportive relationship than an authoritative role.It's important to note that preferences may vary among different patient populations, as a recent systematic review found that minority patients most often prefer a paternalistic model of health decision-making [61].The research found that the agent's role, such as being called a virtual doctor or healthcare assistant, influenced the user's expectations for the agent's appearance [31].Some studies reported that users tested the prototype and commented on character details that informed refinement.For example, in one study, an ECA was regarded as a medical authority, and changes were made to enhance its appearance, such as adding a name badge, updating clothing to include a white medical coat, and adjusting perceptions of the agent's age [44].The majority of ECAs are depicted as middleaged females, which aligns with previous reports [15].This may be related to gender stereotypes associated with health guidance tasks [62].For chronic diseases involving sensitive information, such as cancer, agents with the same race and gender are often preferred.In conclusion, developers of ECAs will have to learn to determine which identities to prioritize for their own ECA but can begin analyzing this by determining their target population and their specific contexts.
Regarding the communication modalities, ECAs reported in the literature mimic human conversation using interactive voice recognition, allowing users to interact with the system through voice rather than just navigating with the touch screen.When applied in the context of health communication, unconstrained speech input with conversational assistants has been found to pose patient safety risks [63].In addition, the quality of automatic speech recognition and synthesis is still a technical problem and has room for improvement [41].Nonverbal behavior has a deep impact on the process and outcome of communication, with approximately 65% of social meaning derived from nonverbal behavior [64].In one study, an ECA was showed by lack of inflection in voice and had a limited number of random body movements, which did not match the context of the conversation.This mismatch between a character's speech and expected facial expressions and body movements can create an unnatural dissonance and affect acceptability, known as the "uncanny valley" [65].Perfecting natural communication via congruence between verbal and nonverbal cues is critical [66].This requires understanding natural behaviors and the biological processes underlying them, so as to develop efficient algorithms to implement a convincing simulation via ECAs.
Our scoping review shows various forms of personalization are used for building content and delivery.The most common approach to personalization is providing more specific feedback based on user responses and health-related data.In this review, examples of conversational adaptation included using an individual's current knowledge state, preferred information processing method, and other user traits such as health literacy and breast cancer risk level according to the user models [30].From a technical point of view, future personalization may involve tailoring interventions to each user individually or even more to their current status at the moment of interaction [67].For example, Sripian et al [68] developed methods to measure and estimate the user's emotions based on biological information.Using these methods, the ECA could react to negative user's emotions and provide assistance.Benkaouar et al [69] used a multisensory environment to detect when the user would like to interact with a device.Promoting health and well-being for people through humancentered technologies requires the partnership of research networks, medical scientists, technology developers, patients, and their formal and informal caregivers.
14 theories and principles were applied in the included studies to create ECAs-led interventions.However, it is difficult to determine which theory or principle is most suitable in the context of chronic diseases, which is consistent with the findings of Kramer et al [15].Different theories and principles may need to be used in different situations.Behavior change theory is usually combined with various behavior change technologies to guide conversational scripts and algorithms.The Modality, Agency, Interactivity, and Navigability (MAIN) Model, used in 2 studies, is an organizational framework for designing a multimedia learning environment.It helps understand how interface features affect the user's psychology through four affordances [50,51].For example, navigability refers to the user's ability to access information and complete tasks, which can aid developers in designing the navigation structure and interface layout of ECAs (eg, including clear menus, navigation options, and visual cues), making it easier for users to find the desired information and functionalities.The cognitive theory of multimedia learning (CTML), which is a more recent iteration of cognitive load theory, is also mentioned.CTML recognizes that the capacity to process and store new information is limited, and presenting multimodal information redundantly may overload cognitive capacity [70].ECAs can be considered as a form of multimodal learning, utilizing multiple sensory channels to enhance learning and information processing.In addition to information processing theories, future research may adopt other interpersonal communication theories to maximize the persuasiveness of the ECA, as tailoring to other constructs beyond information processing modes may improve the ECA's effectiveness in adherence motivation [53].
The evaluation measures used to assess ECAs and their effects on chronic conditions are broad and not unified.This is similar to other reviews on this topic focusing on CA-delivered mental health interventions [22,71].Many researchers develop their own surveys to measure specific issues.The perceptions of the agents are most frequently reported in evaluating ECAs, for example, "How well do the words below describe Laura?" [10].The respondents rate a range of positive and negative traits (eg, friendly, expert, reliable, annoying, and boring) using a Likert-type scale.However, there is variation in opinions and limited evidence on which agent characteristics are especially important.Future ECA design studies should explore both the perception of the characteristics of the agent designed and the perceived importance of these characteristics for an agent in the specific context [40].Research suggested that incorporating human-like characteristics in ECA design did increase users' engagement [26].However, the program use reduced over time of the program access, and there was a dose-response relationship between the number of chats and the change in the quality of life score [37].The dose-response relationship between the level of app use and its effectiveness suggests that more efforts are still required to improve the maintenance of program use over time.In terms of effectiveness-related outcomes, improvements in behavior, knowledge, symptoms, and quality of life were observed in studies, whereas physiological data did not show significant changes.Future studies should aim to measure the efficacy of the relational agent using objective measures rather than relying on self-reporting, which is subject to multiple biases [72].In addition to the evaluation measures mentioned earlier, the studies reviewed in this paper don't measure the cost, efficiency, or productivity improvements associated with using ECAs in healthcare settings.This is a significant limitation as it prevents us from determining whether ECAs are cost-effective compared to alternative approaches or if they can enhance the work of healthcare professionals.

Strengths and Limitations
This scoping review has several strengths as well as some limitations.One of the strengths is that this study offers some important insights into the condition of ECAs in healthcare, with a focus on the design activities.The study was reported according to the PRISMA-ScR guidelines, which enhanced the quality of the review.In addition, the study selection process and data extraction process were conducted by two reviewers independently to reduce selection bias.Agreement between reviewers was very good for the study of both the selection process and the data extraction process.The comprehensive literature search included six databases, but the lack of standardized terminology in this field may have resulted in the omission of some related articles.Additionally, limiting the search to papers published in English may result in the exclusion of ECAs developed for chronic diseases in other languages.Finally, as this is a scoping review without synthesis of the evidence, we present the outcomes as reported by the authors of primary studies, and therefore, no determinations or recommendations are made as to the appropriateness or utility of outcomes.

Recommendations for Future Research and Practice
One important direction for future work is to improve the interaction between ECAs and users to minimize the potential impact of uncontrolled variables (eg, user preferences) and allow for a better study of the impact of agent adaptation.First, researchers can focus on leveraging the unique characteristics of the conversational medium to personalize the interaction.For example, capturing prosodic features in users' speech can be used to automatically detect changes in mood or speech pathologies.This information will allow ECAs to provide adaptive information and services tailored to the user's needs.Second, the behavior data of users can be technically collected and analyzed.Compared with the data collected by the questionnaires, this datadriven quantitative approach is mediated through data objectively describing their behavior, and the domain behind the data is qualitatively understood through interviews.Finally, larger longitudinal studies need to be conducted to measure the effects of ECAs on specified subgroups over time, which will help to identify components of the multifaceted intervention contributing to increased acceptability and positive outcomes.

Conclusions
This scoping review followed a strict methodology to summarize and discuss the development and evaluation of ECAs in chronic diseases.The findings show that the ECAs have been increasingly used in recent years and may have significant potential to deliver effective health interventions to promote self-care.However, there is a need to make technological progress in the embodiment, communication modality, and personalized strategy, as well as a better understanding of user preferences for the appearance, animation, and personalized content.In addition, there is a lack of reliable and comparable evidence in user-centered evaluation approaches.Future studies should incorporate measures of cost, efficiency, and productivity to provide a comprehensive evaluation of the benefits of implementing ECAs in healthcare.

Table 1 .
The search strategy used in PubMed.Search category Search terms

Table 3 .
Theories or principles informing the ECA a -based interventions (N=14) a ECA: Embodied conversational agent