A cross-sectional survey was conducted in January and February 2021 in 12 tertiary hospitals across three cities of the Inner Mongolia Autonomous Region. All studied hospitals provide mHealth apps with eHealth codes, appointments with doctors, electronic payment (e-payment), and health record checking.

Clients (ie, patients and caregivers) who were visiting the outpatient department for nonemergency care, were aged 18 years or older, and able to speak Chinese were eligible for the study.

The questionnaire was created by the research team and reviewed by an epidemiologist from the public health school with mHealth research expertise. The IT department of the Affiliated People’s Hospital of Inner Mongolia Medical University was consulted twice on the amended questionnaire.

A study team from the Inner Mongolia Medical University comprised of resident physicians was assembled and trained on the data collection process. The interviewers consecutively contacted clients at a drugstore or outpatient departure areas, explained the study, and asked for their consent to participate in the study. Consenting participants were asked to complete the questionnaire via face-to-face interviews.

This research study was approved by the Office of Human Research Ethics Committee, Faculty of Medicine, Prince of Songkla University, Thailand (REC.63-306-18-1).

Data were double-entered into EpiData (version 3.1; The EpiData Association), and analysis was performed using R (version 4.1.0; The R Foundation). Descriptive statistics were used to summarize the characteristics of the clients, namely frequency with percentage for categorical variables and mean with SD for continuous variables.

CFA was used to analyze the correlation matrix among the domains. A multiple-indicator, multiple-cause model (MIMIC) with structural equation modeling (SEM) [41] was used to examine the association between sociodemographic variables, mHealth use, and time and satisfaction with health care. The “polycor” R package was used for polychoric and polyserial correlations of categorical variables [42], the “psych” R package was use for EFA [43], and the “lavaan” R package was used for CFA and SEM [44]. The sample size per hospital was calculated based on the assumption that 38.4% of Chinese adults have an mHealth app [45], using an infinite population proportion formula as follows:

with a 10% error rate (d) and a 95% CI (α=.05). A 10% nonresponse rate was also assumed. With these parameters, 102 participants were required to be recruited from each hospital. Due to the effect of COVID-19 on mHealth use, we decided to recruit 200 participants from each hospital. Finally, 2366 participants were included.

A total of 2115 clients provided valid responses. Their mean age was 43.34 (SD 15.39) years. Other demographic characteristics are summarized in Table 1. The participants were distributed nearly equally between the two genders. Three-quarters resided in an urban area. Almost half of the participants were educated at the tertiary education level. More than half were employed by the government or a company. Their household incomes were also somewhat evenly distributed, with a median of ¥4000 to ¥6000 (a currency exchange rate of ¥1=US $0.15 is applicable), which was deemed to be middle class in China [46].

Overall, 88.46% of the participants used the internet, 82.74% had access to the internet at home, and 93.90% had a smartphone. In total, 68.37% of participants used a computer, and 56.26% could use email. Three-quarters of the participants could self-install an app, and one-quarter wore smart wearable devices. Around three-quarters of the participants purchased commodities online, used e-payment, used social media, and performed information-searching online. Most used the internet daily and had been using it for more than 5 years.

Overall, 62.55% of the participants had an mHealth app, 67.75% had an eHealth code, 53.76% could make an appointment to see a doctor, 47.90% used e-payment for health care, 33.38% reviewed their health record on an mHealth app, and 11.30% consulted with a doctor online (Table 1).

From the EFA, validation of the classification of digital activities was performed with a Kaiser-Meyer-Olkin test [47], with a sample adequacy of 0.929 [48], and a Bartlett test of sphericity, which was statistically significant (χ²₂₅₃=1260.1, P<.001) [49]. Based on the parallel analysis, four factors were determined [50]. Due to the nonnormally distributed data, principal axis factoring was used as an appropriate extraction method [51], and oblimin rotation was used as an appropriate oblique rotation method [52]. Five items (ie, “have household bandwidth,” “have wearable devices,” “use social media,” “use e-payment,” and “attend consultations with doctors”) had factor loadings of less than 0.4 [53] or high cross-loadings [54] and were, thus, dropped from the analysis.

Finally, we came up with four domains (ie, factors): digital access, digital use, mHealth use, and time and satisfaction with health care. Details of the EFA are shown in Table 2.

The factor loadings were high, ranging from 0.469 to 0.886. Cronbach α values were all above .8. These four factors explained 62% of the variance. All of these values suggested that our construct was adequate.

The second part of the data set was used to assess reliability and validity by CFA. The measurement model of digital divide in mHealth app use was adequately measured by associated indicators based on high factor loadings. The correlations among latent variables by CFA are shown in Table 3. The model fitted the data well, according to the following indices: χ²₁₂₉=528.3, χ²/df=4.095, comparative fit index (CFI)=0.940, Tucker-Lewis index (TLI)=0.929, root mean square error of approximation (RMSEA)=0.054 (90% CI 0.049-0.059), and standardized root mean square residual (SRMR)=0.042. The Cronbach α reliability coefficient was greater than .7, and convergent validity based on average variance extracted was greater than 0.5.

A MIMIC model with SEM was investigated for the model of mHealth digital divide by using the weighted least square mean and variance adjusted estimator [55], since most of the variables were categorical. The overall indices of the final SEM model fitted the data well: χ²₂₁₆=682.6, χ²/df=3.160, RMSEA=0.045 (90% CI 0.041-0.049), SRMR=0.036, CFI=0.949, and TLI=0.938 [53].

Figure 3 shows the regressions of all the paths. The details of the β coefficients and 95% CIs are shown in Table 4. Age was taken as a continuous variable. Although education and household income were initially ordinal categorical variables, we standardized them as continuous variables to suit the SEM. Their coefficients were interpreted as whether these had a dose-response relationship with the outcome. Age and education were strongly associated with digital access and digital use. Income had a low effect on digital access, and income, employment status, and urban residence were weakly correlated with digital use. No significant gender gap regarding these variables was seen. The cascaded coefficients (95% CI) from digital access to digital use, then to mHealth use, and then to time and satisfaction with health care were 0.28 (95% CI 0.22-0.35), 0.51 (95% CI 0.38-0.64), and 0.14 (95% CI 0.05-0.22), respectively. mHealth use, however, had a weakly significant effect on time and satisfaction with health care. mHealth use was not significantly associated with sociodemographic variables, except for employment status, with which it had a weak correlation.

The potential influence of sociodemographic characteristics reflected the level of digital divide. The number of variables and the magnitude of the coefficients were higher for digital use than for digital access and mHealth use. Thus, use divide in our setting was the most important gap.

We confirmed the framework of digital divide in mHealth app use. Our subjects were mostly educated and middle class, with good experience in internet use and other digital media. Between one-half to two-thirds were using basic mHealth features. Sociodemographic factors had stronger potential influences on digital use than on digital access; they also had the least direct effect on mHealth use and time and satisfaction with health care. However, mHealth use was potentially influenced by digital use. Time and satisfaction with health care, on the other hand, was only weakly associated with mHealth use.

The data from this study identified a cascade of potential successive influences, where digital access influenced digital use, which then influenced mHealth use. mHealth use was determined by digital access and use. Similar to our study, a study by Tirado-Morueta et al [20] found that there was an indirect potential influence pathway from physical access to operative use and expressive informative use of the internet; ignoring this intermediary and simple running regression that predicted mHealth use from sociodemographics would lead to a misinterpretation of the results.

Among the sociodemographic variables examined, age and educational level were the stronger potential influencing variables. Both had direct independent influence on digital access and use, but had no direct effect on mHealth use; in addition, educational level had little effect on time and satisfaction with health care. Thus, their effects occurred in the early part of the digital chain. Based on this potential pathway, assistance for older adults and less educated public clients would need to start with improvements in access and use of the internet in general, as well as mobile facilities, such as email, social media, and online business. Experience with these will make mHealth use easier for them.

Sociodemographic variables had a stronger potential influence on digital use than on digital access. This may reflect that the use divide was due to lifestyle differences more than it was due to inequity problems [56]. The cost of digital access in China was relatively small (around US $15/60 Mbps or more per month [57]) and, hence, did not contribute much to digital access inequalities. On the other hand, certain sociodemographic groups, such as youth and upper-class people, choosing or needing to use digital technology is due to their lifestyle [58]. The small but significant correlation (β=0.28) between digital access and digital use may, in fact, reflect a noncausal relationship.

With the use of SEM, our findings on the effects of sociodemographic variables were different from those of other studies using one-step regression [59,60]. These other studies showed that sociodemographic variables potentially influenced mHealth use, but they missed the fact that the effect passed through digital use. Their findings would indicate the emphasis to improve mHealth use among the underprivileged. Our findings, on the other hand, imply that improved general digital use would be a more natural way to empower these groups of clients. This will make it easier and probably more effective to introduce mHealth to them. To reduce the existing digital divide among hospital clients, the hospital administration should provide special services or appropriate education to assist clients in making better use of mHealth apps.

According to another study in China, mHealth was effective in reducing patient waiting times and increasing patient satisfaction in tertiary hospitals [26]. Another study found that waiting times for consultations and prescription filling reduced, resulting in increased outpatient satisfaction with pharmacy services [27]. Moreover, our study validated the marginally significant effect of mHealth use on shorter waiting times and improved satisfaction. This indicates that mHealth app use cannot adequately explain shorter waiting times and increased satisfaction. This indicates a need for further study. Additionally, since mHealth app use in hospitals is in its infancy, mHealth apps must be improved in terms of design and marketing based on existing digital use to increase their use and provide benefits to clients [33].

This was a cross-sectional study. One may argue that the causation proposed in the SEM was limited by temporal sequence and may not be valid. We argue that sociodemographic variables are long-term values and do not vary much over time, whereas digital divide in these domains only comes after. Mobile apps were developed nearly a decade after the wide use of the internet began, and hospital-based mHealth is the most recent development. Therefore, our proposed pathway may not be flawed in terms of temporal sequence. The current stage of mHealth development in our setting is changing continuously. Thus, further studies may produce different results.

In order to close the mHealth use divide, it is important to improve digital access and digital use among older adults and less educated groups.