The technology acceptance model (TAM) was developed to analyze the usage behavior of information technologies in organizational contexts [42]. TAM suggests that actual usage behavior is driven by the behavioral intention to use a system, as is also the case for the subsequent unified theory of acceptance and use of technology (UTAUT) model [43], and this has been empirically confirmed for mHealth usage among older adults [40,41,43]. Therefore, in this study, we selected behavioral intention to use mHealth applications from TAM as an outcome of interest. Building on the work of Askari et al [29], the statements operationalizing intention to use are taken from Venkatesh and Davis [44] and translated into Dutch, adding one new statement to account for linguistic differences.

Perceived risk, defined by Deng et al [35] as “one’s perception of uncertainty in the use of mHealth services and its severity in terms of consequences,” is measured with four constructs: privacy risk, performance risk, legal concern, and trust. In a 2018 empirical study, these authors identified an association between privacy risk, performance risk, legal concern, and trust with the intention to use mHealth in the general Chinese population. In this study, we adopted the statements, operationalizing these constructs from Deng et al [35] and translated them into Dutch to fit the context of independently living Dutch older adults. The corresponding English-language statements used to measure these constructs, as well as the questions to measure intention to use, are presented in Multimedia Appendix 1 [35,44]. Below, we explain the four different constructs that together capture perceived risk.

A cross-sectional study was designed for this research, in which older adults over the age of 65 years were approached from February through June 2020. We developed a questionnaire and administered it digitally during the initial months of the COVID-19 pandemic. Data were collected by four data assistants in cooperation with different organizations, such as living facilities, senior citizen associations, and health service provider organizations, and via different web-based channels and mailing lists. The data has mainly been collected in the regions of Noord-Brabant, Utrecht, and Zuid-Holland. We have not been able to keep track of the number of recipients of the distributed web-based questionnaire as the cooperation organizations reached out to their clients and members for us. The number of completed questionnaires is reported in the Results section. The reporting of the web-based questionnaire follows the Checklist for Reporting Results of Internet E-Surveys (CHERRIES) checklist [54] and is presented in Multimedia Appendix 2.

The inclusion criteria for participants were as follows:

All respondents were asked to sign an informed consent form before participation. Thereafter, they could answer the questionnaire anonymously. The purpose of the project and information about the questionnaire, data management, and privacy of the participant were provided at the start of the questionnaire. Assistance and explanations were provided to participants who needed help filling out the questionnaire via telephone or email when requested. Data assistants entered the completed questionnaires into an SPSS database (IBM Corp) and pseudonymized the data to ensure anonymity.

Internal validity benefits from using validated instruments [35,42]. The questionnaire was validated by 5 experts (3 eHealth experts, 1 geriatric nurse, and 1 geriatrician). The usability and technical functionality of the web-based questionnaire was tested by 3 data assistants. The questionnaire is available on request. The database was checked for completeness and input errors by comparing a sample of paper questionnaires with the corresponding database information. To increase external validity, data collection took place in several different geographical locations within the Netherlands. Finally, Cronbach α was calculated for each factor to test reliability.

The study was approved by the Medical Ethical Committee of Erasmus Medical Center (number MEC-2018-120). The analysis was conducted using SPSS Statistics software (version 25; IBM Corp).

Our sample consisted of 481 respondents. However, 18 respondents were excluded from the total for not filling out the questions regarding the dependent variable. The respondents in this sample had a mean age of 74 (SD 5.77) years. Almost all respondents (456/463, 98.5%) had prior experience with using the internet, but only 127 (26.6%) had ever used medical applications before. Further details on the respondent characteristics can be found in Table 1.

Cronbach α scores, which show the internal consistency of the items within each acceptance factor, are shown in Table 2. The scores were well above the recommended limit of .70, indicating acceptable reliability [57].

The outcome of the correlation analysis is presented in Table 3. Perceived risk factors were found to be significantly and substantially correlated to one another. Therefore, the factors are not jointly included in the multivariate linear regression analysis. The control variables were also combined in a correlation analysis (see Multimedia Appendix 3). This analysis showed that the control variables were not substantially correlated to one another and could, therefore, be jointly included as control variables in the multivariate linear regression analysis.

The results of the univariate and multivariate regression analyses are summarized in Table 4. All factors were found to be significantly associated with intention to use mHealth. Privacy risk, performance risk, and legal concern were found to be negatively associated with intention to use, and trust was found to be positively associated with intention to use. In the multivariate regression analyses, we controlled for sex, age, education level, and health status. Privacy risk, performance risk, and legal concern continued to have a significantly negative coefficient. Performance risk had a negative coefficient of –0.266. This coefficient is large in comparison to the coefficients for privacy risk (–0.099) and legal concern (–0.124). Trust had the largest significant (positive) coefficient (0.350). The multivariate regression analyses (see Multimedia Appendix 3) also show that health status is significantly and negatively correlated to intention to use when considered a control variable for each of the four risk factors, whereas education was only significant in the model with trust. Age was significantly related to intention to use, except when being a control variable for legal concern. Sex was not found to be significant in any scenario.

This study explored whether perceived risks influence the intention to use mHealth applications among independently living older adults. Our findings showed that privacy risk, performance risk, legal concern, and trust were significantly associated with participants’ intention to use mHealth applications. The perceived risks (ie, privacy risk, performance risk, legal concern) were negatively associated with intention to use, whereas trust was positively associated with intention to use.

Trust had the largest correlation coefficient and explained more than 20% of the variance in intention to use when the controls are added, indicating that it has a considerable positive effect on the behavioral intention to use mHealth. This finding indicates that older adults who perceive doctors accessible via mHealth applications as trustworthy and reliable have a higher intention to use these applications, as hypothesized in this study. These findings broadly confirm previous findings obtained for the general population in China and strengthen existing evidence that trust is an important determinant for the intention to use mHealth applications [35,51-53].

Performance risk was significantly and negatively associated with individuals’ intention to use mHealth. These findings confirm our hypothesis stating that independently living older adults who doubt whether mHealth applications can meet their health care needs have a lower intention to use these applications. This further confirms previously reported findings for the general population and may be more valid for older adults, as they are more likely to fear that technologies will replace health care professionals [35,45].

Legal concern was negatively and significantly associated with intention to use. This confirms our hypothesis, which was based on the argument that older adults who are more likely to worry about inappropriate law enforcement are less likely to have a behavioral intention to use mHealth. The significance of the relationship differs from previous research in which the relationship between legal concern and intention to use was not found to be significant [35]. This may be explained by the fact that Deng et al [35] addressed the general Chinese population, in the hospital context, wherein mHealth use is more limited and legal issues may be less of a concern to respondents because of specific health laws being in place in this context. In addition, discussions on legal aspects of mHealth applications received considerable attention in the Netherlands during the COVID-19 outbreak, especially about a “Corona App” [65,66]. This may have raised concerns among independently living older adults. Legal concerns regarding mHealth use appear to have received little attention in the scientific literature and form a relevant area for further research.

Privacy risk also was significantly and negatively associated with participants’ intention to use mHealth, albeit with a smaller coefficient. This finding confirms our hypothesis and is consistent with previous results reported by Deng et al [35] in the general Chinese population. Privacy risk is not directly linked to the functions of the mHealth app, as it involves the confidentiality of personal health information during the use of mHealth. Hence, compared with performance risk, privacy risk may likely exert less effect on the intention to use mHealth, which explains the smaller coefficient [35].

As the correlation analysis showed, the perceived risk factors are significantly and substantially correlated to one another. These results are not in accordance with the results from Deng et al [35], where a high discriminant validity was shown between the factors. This could be explained by the fact that Deng et al’s 2018 study included very few respondents of 65 years and above, and older adults are more prone to perceive mHealth risks than younger people [67,68]. Another related explanation might be that there are underlying shared determinants of the risks perceived by older adults.

A recent study from the Netherlands showed that the adoption rate of a COVID-19 tracing app was significantly lower for older adults than for young adults [69]. One of their hypotheses for this lower adoption rate was that older adults would feel insufficiently protected by a contract tracing app. Older adults felt insufficiently protected because of the different perceived risk factors as shown in our study, thereby leading to a lower adoption rate.

Trust has been proposed to function as a mediator of five relationships between the three identified perceived risks (ie, privacy risk, performance risk, and legal concern) and the two TAM factors (perceived ease of use and perceived usefulness) with intention to Use [35]. Certainly, these hypotheses have intuitive appeal and, to some extent, theoretical support as well [35,37,38,41]. The empirical results of Deng et al [35] accept two of these five hypotheses and reject the other three.

The theoretical support for the relationship of these variables with trust clearly depends on the definition and operationalization of “trust.” Following the construct definition presented by Deng et al [35], the operationalization adopted in this study focuses on trust in the medical doctors that the mHealth applications connects the user with. Such trust in medical doctors is essentially different from trust in the technology itself, and the intuitive and theoretical arguments cannot be assumed to remain unaffected. In fact, there is literature to support that the technology acceptance factors are influenced by trust in “entities behind the system” [70], suggesting that trust is a determinant of perceived ease of use and perceived usefulness rather than a mediator of their effect on the intention to use.

As shown in Multimedia Appendix 3, the two TAM variables perceived usefulness and perceived ease of use are indeed significantly related to trust in our study. However, in view of the definition and the arguments above, further appropriately designed research on the nature and direction of these relationship is called for.

Similar reflections are in place regarding the relationships between trust and the three perceived risk factors. Multimedia Appendix 3 presents univariate and multivariate regression analyses with trust as the dependent variable for the three risk factors. It shows that performance risk, privacy risk, and legal concern all are negatively and significantly associated with trust. This finding contrasts with previous findings by Deng et al [35] who report that performance risk and privacy risk were significantly and negatively associated with trust, whereas legal concern was not significantly related to trust. This difference in findings might well be explained by a general difference in legal concern between the study populations, which differ in culture, age, and state. In view of the definition of trust, we call for caution to state whether any of the risks are a determinant of trust, and/or the other way around. These relationships deserve further appropriately designed research.

Our study has some limitations. First, this study was conducted during the COVID-19 pandemic, which complicated the data collection, as it became increasingly difficult to approach the targeted population. It became very difficult to recruit respondents who were unable or unwilling to fill out the web-based questionnaire. This may have resulted in a bias towards independently living older adults with better internet literacy. Second, we used a cross-sectional research design. Consequently, the causality of our findings cannot be claimed. Third, we noticed that some of the participants, especially those in the age group above 75 years, struggled to understand the use and utility of medical applications properly. To address this situation, the questionnaires and interviewers provided an additional explanation about medical applications. Finally, although the data are collected from a variety of contexts in the Netherlands, we cannot claim validity in other countries.

The main contribution of this study is to provide the first large-scale quantitative evidence of the validity and significance of the perceived risk factors determining intention to use mHealth applications among older adults in the Netherlands. In our study, we identified trust and performance risk as the most important factors that had a relation with the older adults’ intention to use mHealth services. We suggest involving older adults in the design and development of mHealth applications to ensure that the applications will be tailored to their needs and abilities. Moreover, we suggest involving medical specialists, geriatricians, and other experts in the development of these applications and making this explicit to potential users of the application to increase trust and diminish concerns about the performance of mHealth. Additionally, since privacy risks and legal concerns have a relation with the intention to use mHealth, we suggest that health care professionals, designers of mHealth applications, and the Dutch government use these findings to tailor their mHealth services and campaigns and address the concerns of older adults, to promote better adoption.

As we cannot confirm causality, we recommend studying perceived risk factors using controlled experiments rather than observational studies to confirm or disprove any potential causality of the relationships thus found. To further understand how the perceived risk factors explain behavioral intention to use, possibly via interaction with each other and the variables from models such as TAM, STAM, and UTAUT, qualitative studies are called for. Such qualitative studies can also enable effective solutions to eliminate barriers to medical application adoption. Furthermore, although the intention to use technology has been shown to predict actual usage [41], such experiments may research actual technology adoption, rather than the intention to use mHealth applications.