Telehealth visits are medical consultations where physicians interact remotely (with audio and video or audio only) with patients in real time [1]. They benefit multiple stakeholders by saving travel time and money for patients [2,3], improving access to health care providers [4], allowing easier resource management and reducing facility costs [5], reducing clinical risks [6], and ensuring scalability [7] and continuity of care [8,9] without reducing health outcomes [10,11]. Although the potential benefits of telehealth visits have long been known, and this service has already been introduced in national and international regulations (eg, it was introduced in Italy >10 years ago [12]), its diffusion has been slower than expected [13]. It was only during the COVID-19 pandemic (hereafter pandemic) that the need to maintain social distancing resulted in telehealth visits becoming the primary means of access to care for many nonacute patients [14]. The pandemic forced policy makers, health care providers, and patients to overcome the main barriers to telehealth visit adoption [15]. For example, limited familiarity with telehealth visits and the tools required to access them has often been identified as a barrier to their use [16,17]. However, the pandemic compelled many patients and physicians to become more familiar with IT devices [18], and the act of making a video call has now become a part of the digital skills of many users. As policy makers have witnessed the benefits of telehealth visits, they are now pushing toward its widespread adoption (eg, Italy has planned huge investments in strengthening telehealth visits [19]). The consolidation of telehealth visits is a priority for health care organizations, given the increased familiarity acquired by physicians and patients with telehealth visit platforms and IT devices in general and the substantial investments in organizational transformation, infrastructure upgrades (including the acquisition of new equipment), and staff training. Now that the health risks associated with the pandemic are no longer a concern, the future diffusion of telehealth visits will significantly hinge on whether patients intend to continue using the service [20,21]. Hospitals must now leverage the experience gained to manage and improve this service [22,23]. This requires understanding the factors influencing users’ continuance intention (CINT) toward telehealth visits [24], that is, the users’ intention to continue using telehealth visits in the future after initial adoption [25].

Building on a systematic review of the literature and in-depth interviews with physicians, this work presents the results of a study aimed at identifying the factors influencing users’ CINT toward telehealth visits in a pediatric setting. Specifically, we proposed 2 models and tested them using partial least squares structural equation modeling (PLS-SEM) [26].

The pediatric setting is particularly interesting for conducting this type of study. Delivering telehealth visits in children’s hospitals is both appealing and challenging [1,27]. It is appealing because in-person visits could be inconvenient for at least 2 stakeholders (ie, children missing school and caregivers needing time off work), specialized facilities for children are scant outside major cities [28], and a telehealth visit facilitates continuity of care that is of utmost importance for pediatric patients [29]. It is challenging because surveying patient perceptions in pediatric settings requires the involvement of the patient’s parents or guardians (hereafter informal caregivers). The latter will make judgments based on their perceptions. Because separating the decisions of patients and informal caregivers is impossible in this setting, the term “users” will refer to both in the remainder of this paper.

The study was conducted in a major European children’s hospital, which started providing telehealth visits during the pandemic and was still providing them at the end of this study (February 2023).

We conducted a systematic literature review on the Scopus database following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [30]. The search, conducted in April 2022 and updated in October 2024, focused on journal articles, reviews, and conference proceedings published in English from 2016 to 2025. We used the query string (“telemedicine” OR “telehealth” OR “teleconsultation*” OR “televisit*” OR “mhealth” OR “m-health” OR “ehealth” OR “e-health”) AND (“pls-sem” OR “sem” OR “Structural equation modelling” OR “Structural equation modeling” OR “causal model*” OR “causal inference” OR “path analysis”) to search the “Title,” “Keywords,” and “Abstract” fields. In total, 626 articles were retrieved. After screening titles and abstracts for relevance, 73 articles were retained, with 6 removed for lack of access. These studies were reviewed, and those not focusing on users’ CINT toward telehealth visits or intention to use telehealth visits were discarded, resulting in 46 selected contributions. Two additional studies [31,32] were added through reference tracking. It is important to note that, from the analysis of the selected papers and the examination of the questionnaire items used in these studies, it emerged that authors sometimes use the terms “intention to use” or “behavioral intention” to refer to what is, in fact, a “CINT” (eg, [33-37]). This is particularly apparent in studies where it is specified that the questionnaires were administered to users with prior experience with telehealth services.

Table 1 summarizes the results of the review, highlighting in the “Unit of analysis” column whether the study focuses on telehealth visit services or a broader range of telehealth services (designated as “Multiple” in the table), including telehealth visits. It also specifies in the “Users” column the target groups of the services (eg, children, adults, and older adults), and in the “Experienced users” column, it specifies whether the study’s participants had prior experience with the service before taking part in the survey. Finally, the table lists the latent variables included in the models and the key theories (columns “Variables” and “Theories,” respectively) that were used or adapted in their development. Variables used to represent CINT, regardless of the name used, are shown in quotes in the “Variables” column. Looking at the studies (explicitly or implicitly) focusing on CINT, it can be noticed that several authors propose structural models, using constructs from the technology acceptance model (TAM), such as perceived usefulness (PU) [34,38-41] and perceived ease of use (PEU) [34,39,42]. PU is defined as the degree to which a user believes that using a technology would enhance their job performance or improve their efficiency in completing specific tasks, while PEU refers to the degree to which a user believes that using a technology would be free of effort. The TAM suggests that PEU and PU positively influence the intention to use, and PEU also positively influences PU. Existing literature on telehealth services has proven that PU [34,38,39] and PEU influence [39] CINT as well.

Quality-related constructs, too, are frequently used in studies investigating CINT toward telehealth visits. Quality is either considered as a global aspect of performance [42,71] or split into more fine-grained components [25,33,36,40,43,56]. In this second case, authors usually resort to the characterization proposed in the DeLone and McLean information systems success model (D&M model) [77]. The D&M model considers information quality (INF_Q), systems quality (SYS_Q), and service quality (SERV_Q). INF_Q measures the system’s success in conveying the intended meaning, SYS_Q measures the desired characteristics of the system used to provide the service, and SERV_Q measures the overall support delivered by the service provider.

While studies examining a broader set of telehealth services or multipurpose platforms could provide valuable insights, their findings cannot be generalized to telehealth visits, as the services they consider could significantly differ from telehealth visits (eg, Megawati et al [45] also consider services such as buying medicines, booking laboratory services, and reading articles). Surprisingly, only 4 studies investigating CINT toward telehealth services specifically focus on telehealth visits [42,52,56,71]. However, 3 of these studies [42,52,56] present models designed for specific contexts, limiting their generalizability to other settings. For example, Amin et al [56] and Lu et al [52] focused on CINT in resource-limited countries and included variables such as “price value” and “pandemic-induced anxiety.” These variables are irrelevant in contexts such as Italy, where telehealth visit services are priced as in-person ones at least in public hospitals, and the pandemic is no longer a concern. Wu and Brannon [42] instead focused on chronic patients and their focus was on “patient-centered communication.” This construct, however, is relevant in the context of patients receiving severe diagnoses (eg, cancer), who must endure considerable emotional distress; navigate complex medical information; and make challenging, life-altering treatment decisions, which is not the case in the context under study (we considered pediatric patients with minor health issues; refer to the Methods section). Finally, Grenier Ouimet et al [71] considered the quality of telehealth visits as a formative variable composed of ease of use, service quality, and security and confidentiality and demonstrated that quality influences CINT. However, conceptualizing quality as a formative variable means that the indicators cause the constructs, implying that omitting an indicator potentially alters the nature of the construct and reduces the generalizability of results [78]. Furthermore, it contradicts the literature, which suggests studying quality-related aspects separately [77,79]. In addition, they collected data on patients who had telehealth visits before the pandemic, arguing that their sample is more indicative of CINT in a postpandemic setting because survey responses collected during the crisis are unlikely to be representative of behavior under normal conditions. However, we argue that examining users who were introduced to telehealth visit during the pandemic represents a crucial complementary dimension. These patients, compelled to experience a new model of care, may have discovered the benefits of telehealth visits that they would not have otherwise experienced and may show a higher propensity to reuse it.

Finally, only 1 study [68] has investigated telehealth services in pediatric settings. This study, however, does not focus on CINT, considers a broad set of telehealth services, and applies the Extended Unified Theory of Acceptance and Use of Technology (UTAUT2) model without introducing any new variables or relationships and dismissing the significance of half of the tested causal relationships.

In summary, the literature lacks comprehensive, valid, and reliable models explaining the CINT toward telehealth visit services that can be applied in pediatric settings. Nevertheless, it provides an abundance of theories, constructs, and relationships that can guide the creation of relevant models. This study, thus, aims to answer the following research question: What factors influence CINT toward a telehealth visit in pediatric settings?

To answer the aforementioned research question, this study involved a qualitative and a quantitative phase. The qualitative phase involved conducting 8 in-depth interviews with physicians who were purposefully sampled based on their extensive experience with telehealth visits. The quantitative phase involved developing and evaluating 2 PLS-SEM models on variables impacting users’ CINT toward telehealth visits.

Eight in-depth interviews with physicians with extensive experience with telehealth visits were conducted.

The interviews had 3 key objectives: understanding the criteria physicians apply to assess patient eligibility for telehealth visit (hereafter referred to as “eligibility criteria”), obtaining insights into physicians’ perspectives on the factors influencing CINT, and fostering physicians’ engagement in the study by actively involving them in the subsequent survey administration process.

We created a concise interview guide featuring open-ended questions designed to explore physicians’ experiences with telehealth visits. Verbal informed consent was obtained from all participants at the start of each interview. Each session lasted approximately 30 minutes and occurred between June and July 2022. Physicians were asked about their medical specialty, the duration of their practice in the field, and the eligibility criteria they used. They were also invited to describe the typical process of a telehealth visit, beginning with proposing the option to a patient and concluding with the session’s end, any challenges they have faced during these visits, the feedback they have received from patients, and whether patients have expressed concerns or complaints about the service. Finally, they were asked to give their opinion on the factors that could incentivize users to continue using telehealth visits after having tried it.

This study did not involve clinical trials, the processing of sensitive patient data, medical record consultation, or sample archiving. According to national and regional regulations (Decreto Legislativo n. 211/2003, Legge Regione Toscana n. 40/2005, Legge Regione Toscana n. 84/2015), ethics committee approval is not required for this type of research. The hospital management formally authorized the study. The research was conducted in accordance with the Declaration of Helsinki and followed Good Clinical Practice guidelines. Verbal informed consent was obtained from all interviewees prior to participation. Participants were informed about the voluntary nature of the interview, the purpose of the study, and how their data would be used and stored. For the questionnaire, respondents were informed via email—sent by a physician with whom they had previously interacted during a telehealth visit—that participation was voluntary. Completion of the questionnaire was considered as implied consent to participate, in line with standard ethical practices for anonymous surveys. All interviews were anonymized, and no identifying information was collected or stored. The identity of interviewees is not disclosed at any point in the publication. Questionnaire responses were collected anonymously. The authors had no means to associate responses with individual users, and the design of the questionnaire (closed-ended questions only) further prevented the inclusion of personal information. Data collection, processing, and management complied fully with national and European data protection regulations (EU Regulation No. 679/2016 and Decreto Legislativo No. 101/2018). Participants (both interviewees and survey respondents) received no compensation or incentives for their participation. No identifiable features or personal data of participants are included in this publication. All data were anonymized at the point of collection and handled in accordance with relevant data protection regulations.

This paragraph presents and discusses the variables included in the models considering the results of the interviews with physicians. The dependent latent variable in our models, CINT, assesses users’ willingness to continue using telehealth visits in the future and their likelihood of recommending the service to others.

The results of the qualitative study proved the importance of quality-related aspects in explaining CINT. Indeed, interviewees frequently cited items reflecting them as influencing CINT.

Moreover, the interviews confirmed the appropriateness of the D&M model to study quality-related aspects in the case of telehealth visits in a pediatric setting. The quality types proposed in the D&M model were contextualized by considering the telehealth visit service’s characteristics and the interviews’ results. A telehealth visit requires direct and remote interaction between physicians and users. This interaction happens either via phone or via a video call application. The system used to deliver the service does not convey any additional relevant information other than that exchanged between the user and the physician (users can, at most, upload a medical report or download a prescription). In this context, the D&M model’s quality types can be interpreted as follows:

Interviewees cited benefits regarding time, money, and professional-personal life balance as factors impacting CINT. They are always cited as the most relevant benefits for users [2,3] as using telehealth visits has proven not to negatively impact health outcomes [10,11]. These benefits reflect the PU variable in the TAM (corresponding to the performance expectancy in the Unified Theory of Acceptance and Use of Technology [UTAUT] and UTAUT2 [99,100]). In our models, PU measures how a telehealth visit enables users to save time or money or better integrate the visit into their home and work commitments.

Here, we present the 2 hypothesized models. In model 1, a total of 3 reflective latent variables were included: PU, INF_Q, and CINT. We introduce 3 hypotheses:

We considered 2 control variables: patient’s age (AGE_PAT) and the age of the informal caregiver (AGE_CAR). Moreover, we assessed the model robustness to 2 categorical variables, ie, the delivery channel (CHAN; ie, audio-video or audio only) through which the telehealth visit is provided and the medical specialty (SPEC) for which the telehealth visit was conducted (which depends on the patient’s health issues). The impact of CHAN [101-103] and SPEC [104,105] on stakeholders involved in telehealth visits has been widely cited in the literature, but their impact on CINT has not been studied quantitatively.

In model 2, a fourth latent variable, SYS_Q, was added.

The sample for model 2 was restricted to users who had been visited by a videoconferencing platform (164 in total). As the videoconference visits require using a web-connected device and a videoconferencing platform, we were interested in understanding whether the SYS_Q of this tool affected the other latent variables. Conversely, SYS_Q was ruled out in model 1 because if users had any difficulty interacting via telephone, they would not be considered eligible for a telehealth visit.

Model 2 has 5 hypotheses:

As for model 1, we considered AGE_PAT and AGE_CAR as control variables.

Hypothesis 1.1 and hypothesis 2.1 (INF_Q positively influences PU) and hypothesis 2.5 (SYS_Q positively influences PU) have been proposed and tested by studies on telehealth services based on IS success models [25,59] and by others based on the TAM [31,34,39,47,54,65,67,73]. Indeed, the hypotheses are grounded on both IS success models [77,79] and the TAM [81-83]. The D&M model proposes an association between the 3 types of quality and net benefits (ie, the balance of positive and negative impacts of the system on users, suppliers, employees, organizations, markets, industries, etc). Considering the main benefits a telehealth visit offers users, PU is associated with the net benefits variable in the D&M model. Similarly, in the context of a telehealth visit, the D&M model’s SYS_Q and TAM’s PEU constructs largely overlap.

Hypothesis 1.2 and hypothesis 2.2 (INF_Q positively influences CINT) are coherent with the D&M model [77] and qualitative studies on telehealth visits [106,107]. Moreover, they have been proposed and tested by studies on telehealth services [33,56].

Hypothesis 1.3 and hypothesis 2.3 (PU positively influences CINT) are supported by existing literature on telehealth services [25,34,37,39-41,43,56,60,71] and are coherent with the literature on the IS success model [77,79].

Hypothesis 2.4 (SYS_Q positively influences INF_Q) has been proposed in the literature on telehealth visits [106,107]. However, to the best of the authors’ knowledge, this relationship has not yet been quantitatively tested. While prior studies have investigated the independent effects of quality types, they have largely overlooked the relationships between them. Beyond the telehealth context, studies exploring relationships between quality dimensions indicate that when users perceive a higher quality in the system delivering the service (SYS_Q), their perception of the information quality (INF_Q) also improves [108]. Hypothesis 2.4 is based on the idea that if the system’s quality is not high, the system does not guarantee seamless communication and information sharing with the physician.

This paragraph presents the measurement scales used to test the hypothesized models.

To measure the latent reflective variables in models 1 and 2, we used multi-indicator measurement scales comprising 3 indicators for all the variables (SYS_Q, INF_Q, PU, and CINT). Our questionnaire contains 18 items for users who have received a telehealth visit via videoconferencing and 15 items for those who have received a telehealth visit via telephone. The indicator list and the corresponding variables are listed in Table 2. All the indicators are derived from the measurement scales used in PLS-SEM or structural equation modeling (SEM) studies (refer to the column “Relevant PLS-SEM or SEM studies” in Table 2). However, they have been modified based on feedback from interviewed physicians and insights from relevant studies on telehealth visits [41,109-113].

With regard to model 1, we verified all the path hypotheses. We verified hypothesis 1.1 (INF_Q positively influences PU), which indicates that when patients perceive the information conveyed through a telehealth visit as high quality, they are more likely to perceive the time savings, cost reduction, and enhanced convenience that the telehealth visit offers. Therefore, it is crucial for users to feel well attended and have their concerns thoroughly addressed; otherwise, the benefits of telehealth visits may not be fully appreciated.

We verified hypothesis 1.2 (INF_Q positively influences CINT): users can assess the clinical outcomes of a telehealth visit only after months of treatments [98], so their CINT depends on whether they feel that they can effectively communicate their needs, doubts, and concerns to the physician and receive attention for sufficient time and that an in-person visit would not have been more effective. It is, therefore, essential to focus not only on the accuracy of the diagnosis or prescription—something users may not always be able to evaluate—but also on the process through which it is delivered. For instance, physicians must not appear hasty even when the diagnosis is straightforward. Similarly, physicians should be careful when squeezing telehealth visits between various daily commitments, as rushing could impact users’ perception of the conveyed information.

We verified hypothesis 1.3 (PU positively influences CINT): as telehealth visits have proven as effective as in-person visits from a clinical standpoint [10,119], CINT is influenced by the time, costs, and inconvenience users can avoid by being visited remotely [15,112,120]. Thus, in evaluating the appropriateness of assigning a user to a telehealth visit, it is essential to consider not just the geographic distance from the study hospital but also variables such as the user’s work schedule, responsibilities for caring for other family members, and potential limitations in mobility as they could influence CINT.

No significant difference in model 1 is found between the 2 groups differing in CHAN, meaning that the delivery channel (teleconference or telephone) does not influence the magnitude of the identified causal relationships. On average, users who receive a telehealth visit by telephone were less inclined to continue using the telehealth visit. The lower CINT may be due to a greater empathy established through an audio-video channel than the telephone [121,122]. Considering this result, audio-video calls should be preferred to phone calls when possible (ie, when users can rely on a reliable device and connection). Surprisingly, on average, users who receive a telehealth visit by telephone perceived less time and cost savings (PU). This could be explained by the fact that when an appointment is made for a telephone visit, physicians are less careful to consider patients’ needs and take their availability for granted. This can result in some inconvenience for users. Therefore, if telehealth visits are assigned, the user’s availability should not be taken for granted to not create inconvenience for users. Finally, no significant differences in INF_Q were observed between patients visited by telephone and those visited by teleconference. This probably implies that before assigning users a telehealth visit by telephone, physicians had ensured that the channel was appropriate to convey the information the telehealth visit was supposed to convey.

No difference in the structural model emerged between the 3 groups created by SPEC. However, on average, users who receive a telehealth visit for dermatological issues are more willing to continue using the telehealth visit and perceive higher time and cost savings than the ones who have received a telehealth visit for another medical specialty. These results could be explained by the fact that most dermatologic pathologies are diagnosed with the naked eye and require frequent consultations [107]. As dermatology obtained a higher CINT, hospitals may consider incentivizing the use of telehealth visit for this specialty.

With regard to model 2, we verified the path hypotheses 2.1 (INF_Q positively influences PU), 2.2 (INF_Q positively influences CINT), and 2.3 (PU positively influences CINT), similarly as for model 1.

We verified hypothesis 2.4 (SYS_Q positively influences INF_Q). Hence, if the videoconferencing platform is user friendly and allows for smooth audio-video interaction, users will more likely perceive the quality of the visit as adequate. This suggests paying close attention to the platform’s user-friendliness. For example, hospitals that want to use their own platforms (eg, to make it easier to report on activities performed remotely) need to ensure that these solutions do not involve significant effort on the user’s part.

We did not verify hypothesis 2.5 (SYS_Q positively influences PU). This means that the platform’s user-friendliness does not influence the PU of telehealth visits.

In this study, we adopted a mixed methods approach to propose and test 2 PLS-SEM models that explored the key variables influencing CINT toward a telehealth visit in children’s hospitals. The variables were contextualized through an initial qualitative phase and grounded in established theories, including IS success models [77,79] and the TAM [81,83,99]. We proposed 2 comprehensive yet parsimonious models, avoiding overcontextualized variables and ensuring the validity and reliability of our findings by following state-of-the-art guidelines for constructing and testing measurement and structural models. This study addresses a significant gap in existing literature, which features context-specific models and findings that may not be applicable to children’s hospitals in the postpandemic era. Although several studies have provided valuable insights into telehealth visit adoption, their findings are often confined to health care settings that differ from pediatric care. For instance, some studies focus on developing countries and investigate telehealth visit use during the pandemic [52,56], when emergency circumstances drastically altered health care delivery. These studies do not capture the more stable, long-term integration of telehealth visits into children’s hospitals in developed economies. The study by Wu and Brannon [42] instead focused on adult patients with chronic conditions and patient-centered communication. This study emphasizes the complexities of telehealth visit communication in contexts where patients may require a high level of emotional support in addition to medical advice. This is not the case in pediatric settings where telehealth visit services are only reserved for patients with minor ailments. Finally, some studies rely on prepandemic data [71], which fail to account for the rapid technological advancements, shifts in patient expectations, and changes in regulatory environments that have emerged since the pandemic. This study, therefore, fills in this gap by providing insights specific to the current postpandemic context and to children’s hospitals, where telehealth is increasingly being integrated into routine care rather than being viewed as a stopgap measure.

From a managerial perspective, health care service providers and policy makers could draw insights from this study to answer the wake-up call of taking proactive actions to improve telehealth visits [123]. The findings highlight the critical role of INF_Q in shaping users’ perceptions of the benefits of telehealth visits, such as time savings, cost reduction, and convenience. They highlight the importance of a thoughtful delivery process, urging physicians to avoid appearing rushed, even when providing straightforward information, and to consider users’ unique circumstances, such as work commitments, caregiving responsibilities, and mobility limitation—in addition to the geographic distance from the hospital—when deciding to offer telehealth visit services to patients. The study also suggests that audio-video calls should be preferred over phone calls when feasible and that close attention should be paid to the platform’s user-friendliness. Finally, they suggest that although it may be easier to reach patients by phone rather than scheduling a videoconference, this should not lead to assuming patients’ availability when scheduling phone visits.

Our results are generalizable to other children’s hospitals because the eligibility criteria that guided the choice of variables in our models are widely used and consistent with the literature [4,84-87] and national regulations [12,19,88].

This paper is not without limitations. In the survey, we asked users to recall their last telehealth visit. It could have created a recall bias [90] for some users due to the time gap between filling out the survey and their telehealth visit. Furthermore, the time elapsed since the last telehealth visit could vary from patient to patient. In addition, due to privacy concerns, we could not interview hospital patients directly. Instead, factors influencing CINT were inferred indirectly through interviews with physicians (who, however, had conducted thousands of telehealth visits and had informally gathered substantial feedback over time). Moreover, this study is based only on data from 1 hospital. Administering the survey in different hospitals with different telehealth visit platforms would have allowed for greater external validity of our findings and better investigation and inclusion of platform-related factors in the model.

In addition to healing these problems, future studies could incorporate variables taken from other models into the models explaining users’ CINT toward telehealth visits. For example, users’ expectations, borrowed from the Expectation Confirmation Model, could be included. This would allow for a deeper understanding of how the alignment or misalignment between initial expectations and the actual experience of using a telehealth visit (confirmation) influences satisfaction and, consequently, CINT.

In addition, in this study, we deliberately decided to exclude SERV_Q—one of the quality dimensions included in the D&M model—from our model. This decision was based on the assumption, supported by empirical evidence, that the technical infrastructure underlying modern telehealth visit platforms is consistently reliable. In our case, users could rely on the infrastructure’s availability and functionality; indeed, no instances of system downtime occurred during the period of analysis. However, in our context, the technologies (video call platforms and smartphones) used to deliver telehealth visits were mature, and the patients considered eligible for this service had reliable internet connections and sufficient familiarity with the technology. When these conditions are not met, the quality of the health care provider’s support could play a crucial role in influencing CINT, making it important to include SERV_Q in the model.