Online health education, often referred to as electronic health (eHealth) and now mobile health (mHealth) education, has experienced tremendous growth over the last several years, primarily due to its cost-effectiveness [1]. The rapid growth of mobile broadband technology has expanded access to online health education among individuals with lower socioeconomic status whom may have had less access to fixed devices, such as a personal computer or laptop [2,3]. Not only has mobile broadband technology significantly extended the reach of the Internet, it has become the primary access method worldwide [4].

A recent systematic literature review of mobile nutrition apps concluded that effectiveness of mobile phone and tablet apps for online health education need additional research, as mobile platforms now allow consumers to access information on the go [5,6]. Mobile access to online health education represents a major shift in how users interact with information, resulting in differences in usage patterns and levels of engagement [7]. Many online health sites, including the wichealth website (.org), an online nutrition education and parent-child feeding behavior change system currently being implemented for use by the US Department of Agriculture’s (USDA) Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) programs in 28 states and the platform of focus in this paper, began development when most access was fixed via device locations, such as personal computers at home, work, clinics, or libraries. Wichealth applies the eHealth Behavior Management Model in triaging a client through a series of response-dependent dialogue between the user and a virtual educator toward an initial stage of readiness to change a specific parent-child feeding behavior [8]. It is at this point where intervention content and resources are presented to the user. The client is then presented with the opportunity to engage in further stage-based resources based on intent to move toward active change. Further discussions of the model and features inherent within the wichealth system have been thoroughly described elsewhere [8,9]. However, mobile technology has made impressive gains in just the last several years. Ownership of cellular phones by US adults increased to 92% and mobile phones with app capabilities from 35% to 68% between 2011 and 2015 [10]. Even though mobile phone owners are more likely to be young, affluent, and highly educated, access is not restricted, as 52% of US adults earning less than US $30,000 per year own a mobile phone and mobile technology interest is high among WIC clients [10,11].

It is undeniable that the Internet has become a widely used resource for people seeking health information [12]. Online health education strategies can provide users with more flexibility and an opportunity to become more involved in the management of their health [13]. This increase in consumer knowledge may then lead to improved health outcomes, as online resources offer users greater interactivity and potential for engagement, which should enhance their learning and understanding [14-18]. Mobile access to health education, in particular, may at first appear beneficial for improving user engagement because people tend to be more proximally associated with their mobile devices, frequently keeping them close at hand. However, the quality of the engagement may actually suffer even though the impedance to engage is reduced and the frequency of engagement is increased. The manner in which mobile device users interact with Internet content is sufficiently different from those that access the media from a fixed device [19]. As a result, some evaluation researchers have called for alternative models to assess user impact as, despite the increase in usefulness that accompanies mobile access, this benefit has come at the expense of their usability in some contexts [20].

Over a decade ago, Zhang and Adipat [21] highlighted a number of usability challenges that occurred as a result of the advent of mobile devices, including the ability of users to access the Internet in multiple locations while doing multiple activities, smaller screen size, variable screen resolutions, more restricted user input, and limited processing power. More recent studies have demonstrated these challenges, especially with regard to user interaction. Harrison and colleagues [20] indicated that most usability research does not consider the impact of the mobile transition and its consequences. Their research addressed how mobile devices increase an individual’s cognitive load capacity. The increased ability for users to “multi-task” through mobile device use may come at a cost of user engagement with the content of interest, which is often a critical factor to the success or failure of an application [22]. In addition, mobile access changes the quality of user interaction with health and personal information. A study of data quality in Web surveys found mobile access generally produced lower item completion rates and shorter length of answers [23]. Similarly, a study comparing the differences in survey response completion quality and time found data collection via a mobile device was associated with longer user input time and lower quality and quantity responses. This may suggest that mobile device users either find it more difficult to input data into the online health education system or they may be accessing the system while performing multiple tasks, giving less than their full attention to the task at hand [24]. Furthermore, the effect of primacy is stronger in a mobile setting. Users are significantly more likely to select the top sorted link on a mobile device compared to a computer [25]. Even though it has been clear that Internet interaction differs depending on access device, a systematic review of 8 mobile technologies including mobile phones, personal digital assistants (PDA), PDA phones, enterprise digital assistants (EDA), portable media players, handheld video gaming consoles, and tablets screened from over 26,000 possible studies for inclusion concluded that the overall impact and effectiveness of the applications did not address mobile access as a potential barrier and critical design consideration for online health interventions [26].

More research is needed to determine the extent to which mobile access to Web applications may engage the user differently than fixed access devices and how to design applications to ensure this impact does not affect quality of the intervention. Few studies have been conducted that differentiate fixed eHealth and mHealth education, which is slowly becoming ubiquitous health (uHealth), as devices such as watches, eyeglasses, and home appliances will all soon be tapped into the Internet [2,3,5,13,27,28]. The purpose of this study was to expand the body of research exploring the differences between eHealth and mHealth engagement by exploring the usage pattern differences and impact on key performance indicators (KPI) between fixed and mobile device completion of parent-child feeding lessons associated with the wichealth website.

Mobile access made up 43.66% (267,317/612,201) of all wichealth lessons completed during the study period. Access to the wichealth website by a mobile device was inversely associated with user engagement, in particular the number of educational links viewed within a wichealth lesson and progression of stage of readiness to change. Individuals who accessed wichealth using a mobile device were more than 2 times less likely to visit any educational links that are part of the wichealth lesson. Those who did access a link via a mobile device accessed, on average, fewer links and spent fewer minutes viewing those links than non-mobile device users. With regard to intent to change the parent-child feeding behavior associated with the lesson, mobile device users who began a lesson in an early stage of readiness to change (precontemplation, contemplation, or preparation) were significantly less likely to progress in stage of change than users who accessed wichealth via a personal computer or kiosk (Table 1).

Although these differences in wichealth KPIs appear to be statistically significant (P<.001) between fixed and mobile usage, several user characteristics were also found to be associated with the wichealth KPIs. The paired sample study group of individuals who completed lessons via both fixed and mobile access should control for this confounding; however, this group of users is not defensibly generalizable to the group of individuals that only completed lessons via either fixed or mobile access, but not both. Number of lessons was not used as a measure of KPI because the typical user only completed a single lesson. Within the group that completed lessons only on a fixed or mobile device but not both, 56.56% (158,835/280,845) completed only 1 lesson. Both fixed and mobile access users in this group averaged close to 1.9 lessons per user, which was not significantly different. Users who completed lessons via both fixed and mobile access represented individuals that were likely more engaged, as they completed at least 2 lessons, whereas the typical user completed less than 2 lessons. Further, observed differences between some KPIs of users that completed at least 2 lessons via either fixed or mobile access and users that completed at least 2 lessons via a combination of fixed and mobile access were significantly different (P<.001) indicating that these groups are not representative of each other (Table 2).

Other user characteristics associated with wichealth KPIs included race, Hispanic ethnicity, language, state, pregnancy status, early beginning readiness to change status, and preferred method for receiving nutrition education. Lessons with link views, links viewed per lesson, and link view time demonstrated some significant differences by race (Table 3). Although black users had a similar proportion of completed lessons with link views as white users, those who did have link views, had fewer on average than other users. They were also less likely to advance in stage of change. Users who did not report being either white or black were less likely to view a link during their lesson, but more likely to spend more time on the links that were accessed. Similarly, users of Hispanic ethnicity were less likely to access a link during their lesson compared to other users, but those who did access links also viewed them for longer. Hispanic users were also more likely to advance in stage of change (Table 4). These findings raise the suspicion that users who either did not report race or reported themselves as “other” may have actually considered their race to be Hispanic, as has been found elsewhere [32]. Hispanic engagement in KPI is further demonstrated as users of the Spanish language version of wichealth consistently had more link views and link view time among individuals who used at least one link during their lesson compared to English version users (Table 5).

User state of residence was grouped based on whether mobile access rates in that state were high or low given the relative extent of usage in the state compared to other participating states. Alabama, California, Iowa, Louisiana, Michigan, and South Dakota all had mobile access rates that significantly exceeded the overall average of 43.66%. These states were assigned a high level of mobile access, while the remaining were classified as low. States that tended to have lower mobile access levels were more likely to have users that used at least one link view during their lesson (Table 6).

Pregnancy status was strongly associated with wichealth KPIs, with pregnant users significantly less likely to complete lessons with at least one link view, spend time on links accessed, and progress in stage of change than non-pregnant users (Table 7). These findings may be related to the fact that users completed pregnancy-specific lessons at a greater rate than other lessons. This set of lessons address behaviors that are often more difficult for users to progress along the stage of change continuum, and therefore the cause of the lower level of progression is likely related to the lesson, rather than the user.

User beginning stage status is another characteristic associated with wichealth KPI performance. Specifically, early stage of readiness to change users were more likely to use a link during their lesson, and they accessed about one link more on average than non-early stage of readiness to change users (Table 8). This makes sense because when users progress, they are provided the opportunity to continue their learning with an additional pool of links from which to select in order to help them progress further.

Finally, user preference for wichealth as a means for receiving future nutrition education was assessed for its association with wichealth KPIs in each study group. Users who preferred the wichealth website were more likely to view more links during their lesson and to progress in stage of readiness to change than users who preferred another nutrition education method, such as counseling, group classes, or other onsite learning activities (Table 9).

Given the paired sample study group of individuals having completed a lesson via both fixed and mobile access, control of confounding user characteristics on the association of lower KPIs with mobile access was essentially achieved. Within this group, there was still significant differences between KPIs for lessons completed via fixed compared to mobile access, such as the percent of lessons completed using a link and the link view minutes per lesson; however, the main outcome of stage of change progression was not significant (see Table 1). This suggests that the effect of mobile device on user engagement was still significant in that there was a mobile-specific reason for lower engagement; however, mobile use did not appear to impact progression in stage of readiness to change. Yet as indicated, because this paired sample study group was not representative of the typical user, control for the user characteristics associated with wichealth KPIs is warranted to evaluate the effect on the observed difference in stage progression between typical fixed access and mobile access users. To achieve this, a logistic regression model was developed to include all of the user characteristics previously presented in order to determine if the associations observed of stage progression and lower wichealth KPIs among mobile users was a product of confounding or effect modification. The results of the logistic regression model set up with device access as the dependent outcome variable (fixed or mobile) and the wichealth KPIs as predictors along with all of the associated user characteristics is presented in Table 10. In this manner, the association of the wichealth KPIs and mobile access type could be evaluated controlling for any potential confounding or effect modification of user characteristics found to be related to the KPIs. Table 10 contains the regression coefficients and their standard errors, the z statistic, ORs, and confidence intervals (CIs). Supporting the univariate comparisons made above, all of the model predictors were statistically significant in their association with device type. The logistic regression coefficients can be interpreted as the change in the log odds of whether a mobile device was used for a 1-unit increase in the wichealth KPIs or user characteristic variable.

After controlling for user characteristics associated with mobile device use, users of mobile devices were over 5 times less likely to access any links during their lesson (OR = 0.18, P<.001, 95% CI [.17, .18]). Further, mobile device users were less likely to spend as many minutes viewing links when they did use them (OR = 0.82, P<.001, 95% CI [.81, .82]). Finally, even with all potential confounders accounted for in the model, the stages progressed among early beginning stage of change users was significantly lower among those accessing wichealth via mobile rather than fixed access (OR = 0.82, P<.001, 95% CI [.76, .88]).