During the COVID-19 crisis, telemedicine gained global prominence and became more accessible and applicable than ever before [1]. Social distancing measures necessitated the use of telemedicine as a substitute for in-person health care interactions in many countries. Health care users and providers, who had previously been reluctant to adopt telemedicine, were compelled to embrace it [2]. A substantial disruption such as a pandemic can act as a catalyst for the acceptance of telemedicine services, potentially leading to a shift in users’ perceptions and attitudes [3]. Consequently, the current expansion of telehealth has provided an opportunity to explore and reevaluate the concepts of technology acceptance and risk.

Telemedicine is an umbrella term defined as the use of technology to deliver health care and health information to individuals who are located at a distance from health care providers [4]. While the concept of telemedicine is not new, its wide-scale adoption began in the 1990s. However, telehealth is currently operating within a more information-rich environment. Due to the rapid development of telecommunication technologies and the high cost of devices, its adoption continues to face various challenges, particularly in certain regions where the full benefits of telemedicine services have not yet been realized [5,6].

According to Chen et al [7], the impact of technology on the development of telemedicine in Taiwan began with the National Information Infrastructure project, which used various network bandwidths for telemedicine starting in 1995. Before the legalization of telemedicine through an amendment to the Physician Act in 2018, Taiwan’s medical laws restricted its practice to remote areas, mountainous regions, and outlying islands [8]. Following this amendment, the Taiwan Ministry of Health and Welfare expanded the nationalized health insurance program to include coverage for telemedicine outpatient consultations across different specialties in remote areas. Despite the technological advancements, use of telemedicine remains limited in Taiwan. Reports indicate a disappointingly low rate of use, with only approximately 5.83% of individuals quarantined at home in Hsinchu City using telemedicine consultations from February to May 2020 [8]. With the onset of the COVID-19 pandemic, the government further promoted telemedicine services, resulting in a larger pool of virtual visits. This expansion provides researchers with a valuable opportunity to better understand user concerns. During the COVID-19 pandemic, telemedicine in Taiwan was primarily used for treating confirmed COVID-19 cases and chronic diseases care. Due to quarantine and social distancing requirements, patients used mobile phones, tablets, or computers to engage in telemedicine consultations with health care professionals from the comfort of their homes.

Telemedicine service is not only a technological innovation but also a sociocultural phenomenon [5]. It transforms the traditional physician-patient relationship, altering both its complexity and essence [5]. A systematic review of telemedicine emphasizes the need for more patient-oriented studies, recognizing it as a complex collaborative process [9]. In addition, Bashshur et al [5] highlight the lack of research on the social, cultural, and psychological dimensions of telemedicine.

This study aims to approach the impact of risk from psychological and social perspectives, comprehending its influence on telemedicine, and proposing strategies for development, implementation, and acceptance [5].

Perceived risk is a substantial barrier to the adoption of telemedicine services, encompassing social and psychological aspects [10]. Bakshi and Tandon [11] introduced and validated a theory-driven framework that uncovers the facets of perceived risk as barriers to telemedicine acceptance in their study involving physicians in north India. Harst et al [12] argued that user-centered design and theory-guided approaches are crucial in telemedicine development. The impact of risk on the adoption of a technology is undeniable. However, with the increasing stream of literature, many different dimensions of risk have been proposed. Some studies adopt a limited number of risk dimensions, while others introduce risk dimensions from other fields; some risks share similar concepts or have a high degree of overlap. First, there is a shortage of effective categorization and definition among the various risks. Second, there remains a lack of research that comprehensively assesses the multiple dimensions of risk in telemedicine. Therefore, this study addresses substantial gaps in the telemedicine literature related to perceived risk facets and provides solutions. We aim to reconstruct the effective dimensions of perceived risk and explore what primarily influences users’ acceptance of telemedicine services. We seek to evaluate the integration of the technology acceptance model (TAM) into the perceived risk dimensions, providing a solid theoretical foundation for telemedicine acceptance during a crisis.

Therefore, we posed the following research questions: (1) What are the evidence-based effective risk dimensions for telemedicine services adoption? and (2) Which risks effectively influence users’ intention to use telemedicine during the COVID-19 pandemic?

To address these questions, we will conduct a comprehensive literature review to identify potential risk types in telemedicine from existing literature, providing a strong theoretical foundation. We will then empirically reconstruct these risk dimensions through exploratory factor analysis (EFA) [13]. By applying confirmatory factor analysis (CFA), we will validate and redefine the risk factors identified in the literature. Subsequently, we will integrate these redefined risk dimensions with the TAM and use CFA to explore the relationship between the 2, aiming to provide robust and credible results to strengthen the theoretical framework.

The target population in our study were users and potential users of telemedicine services in Taiwan. The questionnaire for the survey was adopted from previous telemedicine studies or studies on fields related to health care. Both constructs suggested by the TAM model and perceived risk dimensions were included in the questionnaire [21]. To improve its quality and clarity, the questionnaire’s preliminary version was tested to a pilot group of 60 citizens through the web-based survey and 3 individuals holding master’s degree suggested modifications in the wording of items for conciseness. Finally, some questions were deleted owing to their low α reliability. The measurement scales were assessed on a 5-point Likert scale—from 1 for “strongly disagree” to 5 for “strongly agree.” Then, it was determined whether the value of Cronbach α in all constructs was >0.65, which meant that the internal consistency and reliability of summated rating scales were acceptable [61].

We conducted a noninterventional questionnaire-based study, using social science methods to gain a deeper understanding of participants’ perceptions. According to the Taiawn Ministry of Health and Welfare's Human Research Act, Article 5, research conducted in public settings involving nonspecific individuals, which is noninterventional, anonymous, and where the collected information cannot be used to identify specific individuals, is exempt from review [62]. At the initial stage of questionnaire distribution, participants were informed of the study’s purpose and made aware that their participation was voluntary and anonymous, with no personal or private data being collected. All procedures were conducted in accordance with the ethical guidelines of the Declaration of Helsinki.

The study uses SurveyCake, a widely used and professional questionnaire distribution platform in Taiwan. Our target population includes individuals who have previously used telemedicine services or those who may potentially use them in the future. Recruitment was restricted to adults aged ≥18 years, as they possess medical decision-making autonomy. We primarily used public social media platforms, such as Facebook (Meta Platforms, Inc), to reach a broad audience. To ensure the representativeness of our sample, we extended our outreach to relevant online communities across various geographical regions, aiming to achieve balance in terms of age and urban-rural distribution. In addition, for populations less familiar with the internet, we used local community networks to raise awareness about recruitment. By engaging participants through a diverse range of online and offline platforms, we aimed to ensure that the sample accurately reflected Taiwan’s current population, particularly in terms of sex, age, education level, and regional distribution.

After the questionnaire collection is complete, we conducted preliminary processing of the results, including the deletion of incomplete responses. In addition, some questionnaires deemed to have low reliability were removed, including those with reverse order questions and extreme outlier values (such as very short response times and selecting the same option for all questions). Next, the collected questionnaires were randomly divided into 2 groups: group A and group B, using simple random sampling in SPSS (version 20; IBM Corp). Subsequently, normality tests were conducted to determine whether the datasets approximated a normal distribution by testing the normality of the data using the Kolmogorov-Smirnov residual method with the appropriate criteria.

The exported data were managed by SPSS and the relationship between variables was evaluated with SPSS Amos (version 26). SEM is the preferred technique among many standard methods in software and enables the evaluation of complex, multiple latent constructs and relationships [63]. Moreover, because the study analyzes the cause and different facets of perceived risk and its effect on technology acceptance, SEM has been preferred over regression. SEM comprises 2 parts, EFA and CFA. The questionnaires from group A, used for EFA, were used to ensure that the variables adequately represented the domains of the factors and to exclude variables from unrelated domains, thereby reconstructing several evidence-based risk dimensions. Meanwhile, the questionnaires from group B were used for CFA to validate the hypothesized model and its paths [64].

First, the questionnaires from group A were used to justify the application of EFA by accessing the Bartlett test of sphericity with Kaiser-Meyer-Olkin, which was conducted with Varimax rotation and the results of EFA [65,66]. Next, we redesigned the facets of perceived risk based on eigenvalues exceeding 1.000, as determined by CFA. The EFA aims to extract domains with eigenvalues greater than 1.000, and to name the newly generated dimensions based on a combination of data analysis results (such as factor loadings) and theoretical background. In this step, we reconstructed latent variables of perceived risk, and informed the hypothesis according to the abovementioned literature review.

To address common method bias (CMB), we conducted a Harman 1-factor test, which involved performing an EFA on all study variables. The results indicated that the first factor accounted for >50% of the variance among the variables [67]. To mitigate concerns of multicollinearity, we used the variance inflation factor to evaluate the potential bias generated by the measurement design and determine the degree of multicollinearity among the multiple regression variables, with a suggested threshold of <10 [68].

To assess the validity of the model, 2 types of validities were used: convergent and discriminant validities. The convergent validity reflects the extent to which 2 measures capture a common construct [69]. Convergent validity is estimated by using the factor loading of all indicators and the average variance extracted (AVE). If the value of the factor loading of all items is >0.50, it is considered substantial. In addition, if the AVE is >0.5 and composite reliability is not <0.7, it indicates adequate convergence [70]. Moreover, the discriminant validity reflects the degree to which items are discriminated against amid constructs or individual conceptions are assessed for the measurement model. Two criteria are used for the estimation of discriminant validity. The first one is that the AVE itself is greater than correlations in a given construct’s column or row [71]. The second one is that the square root of AVE of the corresponding construct exceeds any correlation with another construct, which means the discriminant validity is considered to be established [72].

To ensure the fit between the data and the measurement model or to assess the goodness of fit, 4 different measures were used, including chi-square—the ratio of the chi-square statistic to the respective df (χ²/df), the goodness of fit index (GFI), root mean square error of approximation, and the comparative fit index. The chi-square is an absolute fit index, with a low chi-square value relative to the df indicating a better model fit. However, the chi-square fit statistic is affected by large samples and a ratio is <5 for the χ²/df is preferred for acceptable model fit evaluation [73-75].

After the measurement model was confirmed, we used the structural model to explain the relationship among the variables and discover the connection between constructs followed by hypothesis testing. Then, bootstrapping resampling technique was used to examine model paths in the structural model—the P values <.05 were considered statistically significant [76].

The web-based survey of our study was open to the public in Taiwan from June 13 to 30, 2022. A total of 1638 responses to the questionnaire were received, of which 38 were deleted owing to incompletion or low reliability caused by short use of answer time. Therefore, 1600 valid responses were collected for the analysis with a valid response rate of 98.2%. Using simple random sampling, all the questionnaires were divided into 2 groups: group A (600 questionnaires, 37.5%) and group B (1000 questionnaires, 62.5%). Table 2 reports the characteristics of all respondents in more detail, including reasonable representations of sex, age, and educational qualifications. Normality was assessed using the 1-sample Kolmogorov-Smirnov test, with significance values for group A and group B recorded as 0.200 and 0.550, respectively. Because both values exceed 0.05, we conclude that the data are normally distributed.

Thus, group A were used to justify the application of EFA by accessing the Bartlett test of sphericity with Kaiser-Meyer-Olkin value was 0.925 and the Bartlett test of sphericity was significant (P<.001) [65,66]. The test was conducted with Varimax rotation and the results of EFA indicated that 8 different factors, with eigenvalues being superior to 1.000, accounted for 69.9% of the cumulative variance. Technology risk 1, technology risk 2, technology risk 4, and perceived COVID-19 infection risk 3 were deleted owing to lower factor loading, wrong concept of contrast, or the crossing of 2 different contrasts during the EFA process.

According to the results of EFA, 2 risk facets, namely “technology risk” and “social risk,” were removed from the analysis. In addition, the facet originally labeled as “perceived psychological risk” was renamed “psychological and social risks” because the 2 items related to social risk were classified under the same category as perceived psychological risk by the EFA. As a result, a total of 8 facets related to perceived risk were identified (Figure 1) and details including factor loading are presented in Table 3. To sum up, 8 risk dimensions were reconstructed, including financial risk, performance risk, psychological and social risk, physical risk, privacy risk, provider risk, time risk, and perceived COVID-19 infection risk.

The definitions of the 8 facets of perceived risk and constructs of TAM are provided in Table 4. The 8 risk dimensions included financial risk, performance risk, psychological and social risk, physical risk, privacy risk, provider risk, time risk, and perceived COVID-19 infection risk.

According to the aforementioned literature review, 18 hypotheses were derived (Textbox 1).

The research model (Figure 2) consists of 8 redefined dimensions of risk combined with the TAM, resulting in a total of 18 hypotheses.

In CFA for the reliability and validity test (Table 5), financial risk 1 and perceived COVID-19 infection risk 5 were deleted owing to lower standard loading. PU1 was deleted from the subsequent analysis according to the result of modification indices because of a higher correlation with UI and failure of the test of discriminant validity (the square root of the AVE for PU 0.715 is less than PU of the correlations with UI 0.766). After that, the tests indicate adequate discriminant validity. Two criteria regarding discriminant validity showed that not only all AVEs were greater than the correlations in that construct’s rows and columns but also the square root of AVE (diagonal) was greater than off-diagonal elements in rows and columns. The two tests mentioned earlier support discriminant validity (Table 6).

In addition, the result of the evaluation showed that all Cronbach α values were >0.7 and the internal consistency of items was considered. Moreover, all item loads on their own factors were >0.5 and the internal consistency reliability for each construct was >0.6. Most constructs’ AVE values met the minimum requirement because the constructs exceeded the criterion of 0.50, though the construct—“perceived COVID-19 infection risk”—was 0.394 and “financial risk” was 0.483. However, we could accept AVE<0.5 if the composite reliability was >0.6 because the convergent validity of the construct was still adequate. Therefore, these tests indicate adequate convergent validity [72].

Financial risk, performance risk, psychological and social risk, physical risk, privacy risk, provider risk, time risk, and perceived COVID-19 infection risk had mean scores of 3.33 (SD 0.854), 3.58 (SD 0.783), 2.43 (SD 0.769), 2.49 (SD 0.897), 3.39 (SD 1.00), 3.32 (SD 0.878), 3.37 (SD 0.831), and 3.81 (SD 0.751), respectively. The mean scores of predictors of TAM such as PU, PEOU, and UI, were 4.00 (0.761), 3.81 (SD 0.804), and 3.87 (SD 0.768), respectively.

The values of variance inflation factor for all the latent variables in the model were between 1.107 and 2.241, all of which were lower than the suggested threshold, indicating low concerns about multicollinearity. For addressing CMB, the Harman 1-factor test during the EFA on all study variables showed the first factor accounted for 31% of the variance among the variables, which was lower than the suggested threshold, indicating no CMB issues.

In the following procedure, the measurement model (Figure 3) assessment was done, and a summary of the model fit measurements is presented in Table 7, in which most results surpass the acceptable levels but the value of GFI is 0.840 which is below 0.9. As Doll et al [84] suggest, GFI in the 0.8-0.9 range presents a reasonable fit, particularly for more variance [84]. Overall, the measurement model indicates a satisfactory fit.

After the measurement model was confirmed, we used the structural model (Figure 4) to discover how closely the theory is supported by empirical data and whether the theory is empirically confirmed [85]. The result of the structural model and the summary of the model fit measurement are given in Table 8. The structural model indicates a satisfactory fit. We also tested for second-order factors in multifaceted risk perception. However, the model fit indices were inferior to those of the first-order model. Therefore, we decided to retain the original structural model.

The hypothesis testing showed that 5 out of 8 redefined risks significantly affect the TAM. The result of the hypotheses testing is shown in Table 9. Three kinds of perceived risk—financial risk, performance risk and psychological and social risk—had significant negative effects on PEOU. The psychological and social risk showed a stronger effect (β=−0.377; P<.001) than the financial risk (β=−0.116; P=.003) and performance risk (β=−0.091; P=.006). PU was affected by 2 types of perceived risk. The performance risk (β=−0.133; P<.001) had a significant negative effect, but the perceived COVID-19 infection risk (β=0.120; P=.003) had a significant positive effect on PU. Furthermore, performance risk was the only kind of risk that had a significant direct negative impact on all 3 variables: PEOU, PU, and UI. However, time risk, provider risk, and privacy risk did not significantly impact the 3 contrasts of TAM. PU was significantly influenced by PEOU (β=0.511; P<.001). Moreover, both PU and PEOU were found to significantly affect UI (P<.001). In terms of goodness of fit indicators, the research model explained 66% of the variance in UI of telemedicine service as shown in Figure 5. To further assess the significance of the direct and indirect effects of predictor variables on the UI of telemedicine service, a decomposition of the effects analysis was conducted in Table 10. Overall, among the 8 types of risks, performance risk had the most significant total effect on UI (β=−0.362).

The study developed a theoretical model based on TAM to determine the barriers to UI in telemedicine services. By reconstructing the risk dimensions, 8 facets of perceived risk were extracted through EFA, and 5 facets of perceived risk were considered to substantially affect the TAM of telemedicine services during the COVID-19 pandemic in Taiwan. Three risk dimensions have been adjusted according to the results of EFA. Perceived psychological risk and social risk have been categorized into psychological and social risk, mainly due to their high degree of overlap. Social risk emphasizes the potential for image damage due to lack of recognition by others, while perceived psychological risk highlights the potential for inconsistency with self-identity. Social perceptions may also influence one’s self-image, as well as attitudes toward and anxiety about telemedicine. Technology risk, which is not traditionally categorized under risk perception, was included in previous studies primarily from a technological perspective to assess the impact of risks. This inclusion leads to some overlap with the original risk dimensions. For instance, survey questions asked users whether they were concerned about the poor performance of telemedicine due to various technological limitations (such as network, equipment, and physical examination). Although these are technological limitations, they fundamentally address concerns about performance. This was the purpose of EFA—to reconstruct 8 effective types of risk.

The psychological and social risk seems to be the strongest barrier to perceived risk affecting the ease of using telemedicine. Our study supports the effect of technology anxiety and pressure, which have been shown to affect users’ resistance to using new technology [10]. Technology anxiety and uncertainty can lead to resistance to change owing to unexpected errors in technology [49]. Therefore, when users are faced with a new telemedicine service, they may be reluctant to switch from a face-to-face service to an alternative one. In addition, our results agree with the past finding that psychological and social risk prevents not only providers but also users from adopting technology [11]. Users care about social pressure from their friends, family members, and work groups with regard to telemedicine service.

According to the results of the total effect, performance risk is the most substantial risk factor affecting UI when Taiwanese individuals consider the adoption of telemedicine. In addition, performance risk was the only risk factor that had a negative effect both on UI directly and on PEOU and PU at the same time. This indicates that PEOU and PU are the mediators that partially explain the association between performance risk and UI. Concerns about the performance, technology, and quality of service have been expressed by stakeholders. For both health care providers and users, performance always plays a vital role. A large number of medical studies on the effectiveness of telemedicine have compared it with traditional in-person visits with regard to different diseases, such as anxiety disorders [2] and chronic kidney diseases [86]. Variations of quality and effectiveness in telemedicine and in-person care may lead to performance risk. Another factor leading to performance risk is distrust and concern regarding the technology. As shown in the previous research, the result provides more empirically grounded evidence for the importance of technology concerns perhaps caused by inadequate communication effects between physicians and patients and by disruptions in communication owing to an unstable broadband connection, which may alter the accuracy of instructed physical examination results and even cause medical error. Therefore, the “digital divide” means that telemedicine service cannot completely substitute for face-to-face visits. Indeed, a substantial concern shared by both physicians and users is the inability to conduct an in-person physical examination [87]. In the study, the item technology risk 3, “I fear the absence of physical examinations during telemedicine services,” was classified as a factor related to performance risk during the EFA. This indicates that individuals not only have concerns about the limitations of the technology itself but also about the impact it may have and the resulting performance outcomes. As Nguyen et al [42] mentioned in the review, promoting realistic expectations before the telemedicine visit takes place is essential to effectively improve patient satisfaction [42]. Providers should try to increase technology acceptance and a sense of trust by assuring the users of various technological solutions to reduce the risk or by showing evidence for the similarity between technological use and in-person examination components through virtual assessments [88].

We found that the perceived COVID-19 infection risk has a positive effect on PU with P value of <.001. This is because telemedicine was globally used by force of circumstance and because of social distancing as a substitute for face-to-face interactions in enhancing clinical care, health promotion, and disease prevention. The substantial negative effect of financial risk on the PEOU of telemedicine is notable. In this context, financial risk refers to payment models, personal and governance insurance coverage, and online payment systems. One possible reason for the impact of financial risk could be the general public’s understanding of Taiwan’s nationalized health insurance program, which may not be fully aware of the extent and scope of insurance coverage for telemedicine services. In addition, users may have concerns about online transactions due to their perception of financial loss risks. Studies on online payment highlight the importance of website security signs and quality information, as these factors can influence customers’ perception of financial risk [89]. These results may underscore the importance of trust in online trading platforms and payment systems.

Three facets of perceived risk—time risk, provider risk, and privacy risk—do not appear to be statistically significant to TAM in telemedicine service. In the past, internet privacy concerns were one major barrier to the adoption of digital technologies or online services [90]. However, the absence of a substantial effect of privacy risk on UI in telemedicine in our hypotheses agrees with the findings of the study by Kamal et al [10], in which perceived privacy risk was found to be an insignificant determinant of UI to use telemedicine services. Moreover, as shown by studies among 1059 US residents, just like environmental risks similar to the COVID-19 pandemic, privacy concerns do not negatively impact UI. Another cross-sectional study in Germany also revealed that data privacy and security concerns were seen as less important in case of an emergency condition [44]. One possible explanation by Kato-Lin and Thelen [54] is that people may be involved in a different decision process when they are facing immediate and severe risks. Another explanation may be the result of a tradeoff during the crisis between the privacy concern and the perceived risk of COVID-19 infection.

In general, telemedicine services were expected to reduce the need for transportation, save travel time, and provide cost-saving for the public [87,91]. However, it is worth noting that the health care system in Taiwan is characterized by good accessibility, comprehensive population coverage, and short waiting times [92]. Compared to the study by Weibel et al found that patients using telemedicine saved an average of 438 driving miles per visit, the study also highlighted the potential risks related to time and distance [42]. In Taiwan, seeking medical services is extremely convenient, with shorter time and distance required for medical visits. These advantages make it less likely for the public to experience time risk.

Regarding provider risk, previous studies found that most patients prefer to use telemedicine with their physician with whom they have an established relationship [58]. One possible explanation for the insignificant hypothesis is that most users in Taiwan primarily used telemedicine services for COVID-19 infection diagnosis and treatment during that period. This is different from the context of general chronic disease management, where the importance of the service provider is comparatively reduced. Therefore, it is the result of tradeoff among provider risk, government policy, and perceived of COVID-19 infection. Consistent with previous research, our study found that PEOU positively influences PU, and both PU and PEOU positively impact UI. However, PU exerts a stronger influence on UI [17].

As >80% of respondents had no prior telemedicine experience, the findings are particularly relevant to nonadopters and may encourage them to consider using telemedicine. Risk perception is a dynamic and continually adjusting process. Nonadopters often perceive increased uncertainty and risks due to their lack of experience with technology, driven by unfamiliarity, fear of the unknown, and privacy concerns [93]. Therefore, these perceived risks have a greater potential for mitigation through awareness campaigns and other adjustments.

In the promotion of telemedicine, addressing psychological and social risk is crucial in influencing user behavior. To alleviate technology anxiety and discomfort, health care providers should prioritize the establishment of emotional bonds and genuine care, particularly when catering to older adult users engaging in mobile health services [94]. Demonstrating empathy and inclusiveness toward users, particularly those experiencing anxiety and discomfort, is essential. Swiftly resolving issues and offering efficient technical support, particularly for new services, can effectively reduce user concerns and mitigate risks associated with social origins. Encouraging users to share positive experiences contributes to diminishing social risk and generating excitement among potential users.

Regarding the main barrier to adoption, performance risk, which may stem from the public’s unfamiliarity and lack of knowledge about telemedicine, as well as the unknowns and misunderstandings surrounding its performance. According to consistent evidence from systematic reviews, the quality of care in telemedicine is comparable to, or even superior to, face-to-face consultations [2]. Governments and health care providers play a critical role in informing individuals about this evidence to alleviate concerns of potential users. Close monitoring of telemedicine service quality and providing additional education tailored to underserved users can substantially enhance adoption rates.

There are some limitations to this study. First, the web-based survey using the platform SurveyCake has a methodological limitation. The sample may produce a bias toward individuals more likely to be on the web and have internet access. Second, we focused on users in Taiwan、 and our result for perceived risk may not be representative of other countries. Furthermore, at the time of the study, approximately 80% of the population in Taiwan still did not have experience with telemedicine services, which may limit the applicability of the study’s findings. Third, this survey would benefit from longitudinal data, which represents only a single point in time and does not assess trends or tradeoffs in patients’ perceived risk over time. In future studies, an examination of user’s perceived risk at other time points will help understand the change of tradeoffs among the risks. Second, to understand the effect of experience or age difference on users’ perceived risk, future studies can recruit subgroups and compare the findings with those of our study.

This study aimed to enhance previous research efforts by reconstructing the facets of perceived risk at a more granular level and further refining the nomological fit of the TAM with the perceived risk dimensions. Eight effective risk dimensions were reconstructed based on a solid review of the background literature through empirical techniques. Upon analyzing the relationships between perceived risk and the TAM, we determined that performance risk was the most substantial risk factor and substantially influenced telemedicine adoption both directly and indirectly. In addition, performance risk, financial risk, and psychological and social risk substantially impacted PEOU. These findings deepen our understanding of user perceived risk and its impact on telemedicine adoption, highlighting the need for strategies to address users’ uncertainties and concerns.

On the basis of the identified factors, including performance risk and psychological and social risk, telemedicine service providers, planners, and policy makers can design better strategies for the successful implementation and adoption of the services in countries that have not yet benefited from telemedicine. To decrease public anxiety and pressure, public awareness about the adoption process should be increased, and the willingness to embrace telemedicine should be instilled in the public as a convenient way to obtain high-quality health care services. Health care providers and governments must also improve the quality of technology and widely disseminate evidence related to the effectiveness and safety of telemedicine, which will decrease performance risk and fundamentally promote the public’s intention to use telemedicine services.