Over the past decades, health care systems in most countries around the world have experienced tremendous changes because of the rapid advancement in information and communications technology (ICT). Mobile health (mHealth), as an innovative form of ICT, is one of the most prominent services with remarkable effects on the development of the health care system [1]. According to the Global Observatory for eHealth, mHealth is defined as “medical and public health practice supported by mobile devices, such as mobile phones, patient monitoring devices, personal digital assistants, and other wireless devices” [2]. mHealth has the potential to enhance the quality of health care systems by improving communication, efficiency, treatment adherence, and health/disease management; reducing costs; and increasing access to interventions and health services [3-6].

The popularity of mHealth programs has grown worldwide as evidenced by a Statista report in 2017 regarding the estimated number of mHealth app downloads, which has exponentially increased from 1.7 billion in 2013 to 3.7 billion in 2017 [7]. A global survey undertaken by the World Health Organization in 2011 showed that 114 countries have established mHealth initiatives, with health call centers, emergency toll-free telephone services, management of emergencies and disasters, and mobile telemedicine being the 4 most frequently reported initiatives in many countries [8]. Malaysia in Southeast Asia has started restructuring its health care policies and encouraging more start-ups to use disruptive technology to solve key medical challenges [9]. As Malaysia has a high smartphone usage (78% in 2018 [10]), mHealth adoption in the country can be a new and effective approach to empower people in health management and transform the attitude toward health care from reactive to proactive [11].

The importance and implications of mHealth have inspired researchers to investigate the factors in the adoption of mHealth. A cluster of studies viewed mHealth as a perceived technology–driven behavior and attempted to find the correlates of such behavior using technology adoption theories [12-14]. Another stream of studies attempted to explain mHealth from health-related perspectives and examined health factors as predictors of mHealth adoption [15]. While the former has been widely researched, the latter has received less attention.

Among health-related factors, the belief that health events are caused by one’s own actions is one of the major predictors of health behaviors such as greater engagement in health/disease management, healthier lifestyle, and better physical and mental quality of life [16,17]. Those who believe that they have an active role in one’s own health, also known as internal health locus of control (HLOC), are more likely to take responsibility toward their health and engage in health behaviors [18]. Given that individuals’ health-related dispositional factors are significant antecedents of health behaviors [19], including the use of health technologies [20-22], very limited studies have examined the relationship between internal HLOC and the intent to adopt mHealth [23].

Although the direct relationship between internal HLOC and behavioral intention to use mHealth provides a noteworthy tenet to knowledge, little is known regarding the underlying mechanism of this relationship. Focusing on mediating factors, which facilitate the relationship between internal HLOC and the intent to use mHealth, can provide a better insight to Malaysian health policy makers and health care professionals for embedding mHealth use in daily life and promoting mHealth functions (such as health/disease information seeking, communicating for health-related purposes, and downloading and using any health-related apps) for health management and more importantly active participation in disease prevention, thus reducing the need for health care services and consequently the toll on the Malaysian health care system.

Therefore, this study aims to contribute to the literature by introducing 3 constructs of the Unified Theory of Acceptance and Use of Technology (UTAUT) (ie, performance expectancy, effort expectancy, and social influence) as mediators between internal HLOC and the intention to use mHealth, which, to the best of our knowledge, has not been examined thus far. The UTAUT is one of the most widely accepted technology adoption theories with a wide applicability and a high explanatory power to predict the intent to adopt technology [24]. By integrating the UTAUT constructs as mediators in the association between internal HLOC and the intention to use mHealth, a mediation model has been developed to explain the mechanism that underlies the relationship between internal HLOC and behavioral intent to adopt mHealth. Therefore, the objectives of this study are to investigate (1) the relationship between internal HLOC and behavioral intent to adopt mHealth; (2) the relationship between the constructs of the UTAUT (ie, performance expectancy, effort expectancy, and social influence) and the intention to adopt mHealth; (3) the relationship between internal HLOC and the constructs of the UTAUT; and (4) the mediation effects of performance expectancy, effort expectancy, and social influence in the relationship between internal HLOC and the intent to adopt mHealth.

Among 374 participants in this study, there were 145 males and 229 females. The participant age ranged from 18 to 68 years (mean 28.01 years, SD 11.10). Almost 45% (166/374, 44.4%) of the participants were Chinese, and 40.7% (152/374) were Malays. In terms of health status, 47.4% (177/374) of the participants perceived their health status as good, 27.5% (103/374) perceived it as fair, and 18.2% (68/374) perceived it as very good. Participants were also asked whether they had an ongoing or a serious health problem that included heart disease, arthritis, or a mental health condition requiring frequent medical care, such as regular visits to doctors or daily medications. The majority (315/374, 84.3%) of the participants indicated that they did not have any ongoing or serious health problem, while 12.0% (45/374) reported that they did not know of any serious health problems. A small percentage (14/374, 3.7%) of participants had an ongoing disease or serious health problem. Lastly, regarding mobile phone usage experience, 40.4% (151/374) of the participants had 8 to 10 years of experience, 39.0% (146/374) had 4 to 7 years of experience, 15.0% (56/374) had more than 10 years of experience, and 5.6% (21/374) had 1 to 3 years of experience. Table 1 shows the demographic profile of the respondents.

This study used a questionnaire-based cross-sectional design to collect the required data. A total of 400 questionnaires were distributed to Malaysian adults residing in Kuala Lumpur, Malaysia. The subjects for this study were drawn from mHealth users. A screening self-report question was included in the survey to identify mHealth users. Participants were asked if they have ever used their smartphones for any health-related purposes, such as seeking health- and disease-related information online, texting messages for health-related purposes (such as reminders/alerts for appointments, taking medications, and consultations), and downloading and using any health-related apps (such as fitness apps, and apps for health tracking and medication tracking). Participants who reported having used their smartphones at least for one of these purposes were included in the analysis.

The questionnaire comprised an informed consent form, demographic profiles, and questions related to internal HLOC and the modified UTAUT constructs for mHealth use. Data were collected using convenience and snowball sampling methods. A research assistant was recruited for data collection. Participation was voluntary where confidentiality was ensured, and respondent consent was obtained before commencing the survey. Participants were given an absolute right of withdrawal at any time and without giving any reason. The protocol of the study (including the research procedure, the rights and safety of the participants, and the method of data collection) was approved by the review board of Xiamen University Malaysia to ensure compliance with research ethics. The approval number is REC-1911.01.

The following 2 rules of thumb are used for choosing the sample size when partial least square is to be used for model analysis: (1) “10 times the scale with the largest number of formative (ie, causal) indicators (note that scales for constructs designated with reflective indicators can be ignored),” and (2) “10 times the largest number of structural paths directed at a particular construct in the structural model.” [76]. This study did not have any formative indicator. Therefore, the first rule of thumb was not applicable for this study. According to the former rule of thumb, the minimum sample size for this study is 70, as the largest number of structural paths in the research model is 7, which is related to the behavioral intent to adopt mHealth. Further, recent developments suggest that researchers should determine sample size through power analysis [77,78]. Power analysis determines the minimum sample size by taking into account the part of a model with the largest number of predictors [79]. Therefore, we used G*Power to determine the required sample size. Based on G*Power, the adequacy of sample size was computed using the statistical tests recommended. For a mediation model, the input information was as follows: test family: “F tests,” statistical test: “linear multiple regression: fixed model, R2 deviation from zero,” type of power analysis: “a priori: compute required sample size–given α, power, and effect size,” with effect size=0.15, α=.05, power=0.90, and number of predictors=4. G*Power showed that the minimum sample size required for the mediation model is 108, with an actual power of 0.90. Therefore, the number of responses collected for this study was sufficient for analysis.

In this study, partial least squares structural equation modeling (PLS-SEM) was used to examine the proposed conceptual framework using SmartPLS software. By using PLS-SEM, the direct and indirect effects of multiple independent and dependent variables can be tested simultaneously, which provides greater statistical power. PLS-SEM is also able to accommodate a study with a small sample size despite the complexity of the models [79]. After excluding incomplete questionnaires, 374 responses were included in the analysis. The results consisted of a 2-stage analytical procedure, which involved a measurement model and structural model, to validate the model and test the hypotheses developed for the purpose of this study.

The first step in the analysis concerning the measurement model was to examine the factor loading. In this study, the factor loadings of the items varied from 0.632 to 0.945 (Table 2). All items were retained as the factor loadings were above the recommended value of 0.6 [80]. Second, composite reliability, which measures the internal consistency of the constructs, was used to measure the reliability of the variables. According to Hair et al [79], the minimum value of composite reliability is 0.7. Table 2 shows that all the constructs yielded good composite reliability ranging from 0.873 to 0.954. Third, convergent validity was assessed using average variance extracted (AVE). Hair et al [79] mentioned that the minimum acceptable value for AVE is 0.5 or higher, which indicated that 50% or more of the items were explained by the construct. In this study, all the constructs exceeded the acceptable value, which indicated that all the constructs obtained good convergent validity (Table 2).

Lastly, discriminant validity was determined using the heterotrait-monotrait (HTMT) ratio of correlation as recommended by Henseler et al [81] because it is more suitable for discriminant validity assessment compared with the Fornell-Larcker criterion and cross-loading assessment. In order to achieve discriminant validity, Henseler et al suggested that a threshold value of 0.90 or below is required [81]. As shown in Table 3, all the study variables met the criterion to establish discriminant validity where the HTMT values were below 0.90.

Multicollinearity is assessed using the variance inflation factor (VIF). In this study, all VIF values were below 5, which indicated no violation of the multicollinearity assumption. The structural model was assessed using R², beta, and t value, which were obtained via a 5000 resampling of the bootstrapping procedure. The direct, total indirect, and specific indirect effects are shown in Table 4.

Based on the structural model, hypothesis 1 was not supported because internal HLOC (β=−0.039, P=.32) was not significantly related to the intention to use mHealth. Performance expectancy (β=0.316, P<.001), effort expectancy (β=0.169, P=.008), and social influence (β=0.307, P<.001) had a significant and positive relationship with the intention to use mHealth, which supported hypotheses 2, 3, and 4 (Figure 1). Hypotheses 5, 6, and 7 were also supported where results showed that internal HLOC had a significant positive relationship with performance expectancy (β=0.330, P<.001), effort expectancy (β=0.329, P<.001), and social influence (β=0.186, P<.001).

Overall, the results supported the model of this study where the total indirect effect was significant (β=0.217, P<.001). All the study variables explained 47.2% of the variance in the intention to use mHealth. Three specific indirect paths were generated to test hypotheses 8 to 10. Hypothesis 8 was developed to test the mediating effect of performance expectancy on the relationship between internal HLOC and the intention to adopt mHealth. The results showed that the direct effect of internal HLOC on performance expectancy and the direct effect of performance expectancy on the intention to use mHealth were significant, while the direct effect of internal HLOC on the intention to use mHealth was not significant. This indicated that performance expectancy fully mediated the relationship between internal HLOC and the intention to use mHealth with a significant indirect effect (β=0.104, P<.001), supporting hypothesis 8 (Figure 1).

For hypothesis 9, effort expectancy as the mediator for the relationship between internal HLOC and the intention to adopt mHealth was tested. The direct effect from internal HLOC to effort expectancy and the direct effect from effort expectancy to the intention to use mHealth were significant, but the direct effect from internal HLOC to the intention of adopting mHealth was not significant. This indicated that effort expectancy fully mediated the relationship between internal HLOC and the intention to use mHealth (β=0.056, P=.02). Thus, hypothesis 9 was supported (Figure 1).

In Figure 1, the results showed support for hypothesis 10, where social influence significantly mediated the relationship between internal HLOC and the intention to use mHealth. The significant indirect effect (β=0.057, P=.002) revealed that internal HLOC impacted social influence, and social influence in turn affected the intention to use mHealth. Based on the findings of the mediation tests, it can be seen that the mediating effect of performance expectancy on the relationship between internal HLOC and the intention to use mHealth was the strongest, followed by social influence and effort expectancy.