The study was a 3-armed randomized controlled trial consisting of a no-treatment control group and 2 experimental conditions (eHealth and mHealth). We recruited participants from a Dutch online research panel and randomly assigned them to 1 of 3 conditions (eHealth, mHealth, or control). Participants were excluded from the study in case of (1) physical conditions hindering engagement in physical activity, (2) pregnancy at the time of recruitment, (3) having a holiday scheduled for more than 5 working days during the study period, and (4) participation in another intervention during the study period.

The baseline measurement took place in April 2014 and the follow-up measurement took place 6 months after baseline (in October 2014). All participants (control, eHealth, and mHealth) were informed about the study by email and asked to complete online questionnaires at 5 points in time: at baseline (T0), 7 days after baseline (T1), 7 days after T1 (T2), 7 days after T2 (T3), and 6 months after baseline (T4: follow-up questionnaire). When a questionnaire had not been completed within 7 days after the invitation email, a reminder was sent. The reminder was sent to prevent dropout and stimulate participants to continue with the intervention. It was not possible for participants to skip sessions, and the next session could only be accessed when the previous one was completed. So when participants received a reminder and accessed the intervention, they continued with the session that followed their last completed session; for example, after session 3, participants could not continue with session 5 until they had completed session 4. Participants received 2 bonus points amounting to €2.50 as an incentive for completing the intervention (the first bonus point after T3, the second one after T4). The 2 intervention groups (eHealth and mHealth) received, additionally to the questionnaires, feedback messages and advice based on their answers to the questionnaires at T0, T1, and T2. Participants allocated to the control condition were also asked to complete all questionnaires but did not receive any feedback or information.

To determine the sample size, we conducted a power analysis using G*Power (version 3.1; Heinrich-Heine University Dusseldorf, Germany) [38,39] taking into account an effect size of 0.20, a power of 0.80, and an alpha of 5%. Based on this calculation, a minimum total sample size of 423 (141 participants per condition) was required.

Both the eHealth and the mHealth versions of the intervention were developed using the TailorBuilder software (OverNite Software Europe, Geleen, the Netherlands). Both interventions had exactly the same content. The mHealth intervention was specifically developed for use with a mobile phone, while the eHealth version was developed for use with a computer. Therefore, the intervention within the eHealth condition was delivered via email, whereas in the mHealth condition advice was delivered via short text messages (short message service; SMS). Questionnaires for both groups were sent via email; however, participants allocated to the mHealth group were requested to complete this questionnaire via their mobile phone. Participants in the control condition received an email to inform and remind them that they could assess a questionnaire.

Before starting, participants were clearly instructed that they should use the intervention only via the medium that belonged to their study condition. Participants in the eHealth condition were asked to use the intervention only via the computer and participants in the mHealth version were asked to use the intervention only via their mobile phone or tablet.

We assessed this adherence (use of the intervention) by means of a question in the follow-up questionnaire that asked participants which medium they had used for the intervention. It should be noted that this adherence is correspondingly based on self-reports. It unfortunately was not possible to use the logs of the intervention to assess the medium of use. Hence, we cannot 100% guarantee that the self-reported answers are actually in line with the medium of use. The visual format of the feedback messages was the same in the eHealth and mHealth interventions. In both interventions the feedback messages were merely provided by means of text, without any additional visual content.

The intervention (named SmartMobiel) was specifically focused on physical activity as a healthy lifestyle behavior. It was built on an existing eHealth intervention [10] and framed by the I-Change model [40,41] and the health action process approach [42,43]. The main goal of SmartMobiel was to stimulate participants’ awareness, ability factors (ie, action plans and goal action), and self-efficacy (see Table 1) to engage in more physical activity. The intervention consisted of 5 successive rounds.

All statistical analyses were performed using IBM SPSS Statistics version 20 (IBM Corporation). We used multiple imputation with 25 iterations to replace missing values on sociocognitive and outcome variables at T0. Additionally, we replaced missing values on BMI and physical activity at T4.

Descriptive statistics and frequencies described the characteristics of the study population. We analyzed differences at baseline using analyses of variance (ANOVAs) with Tukey post hoc tests for continuous variables and chi-square tests with Bonferroni correction for categorical variables.

We analyzed attrition using logistic regression, with attrition at follow-up (T4) as the outcome variable (0=not completed; 1=completed whole intervention), and intervention condition and all baseline variables (ie, age, sex, educational level, BMI, baseline physical activity, and baseline sedentary behavior) as predictors. Process evaluation was analyzed using ANOVA with Tukey post hoc tests to assess the differences between the experimental conditions with regard to usability and appreciation.

Effect analyses were performed using linear regression analyses with the ENTER method. Analyses examined 3 independent effects: (1) intervention (eHealth and mHealth) versus control condition, (2) eHealth versus control condition, and (3) mHealth versus control condition. To analyze the last 2 effects, we recoded the study condition variable into 2 different dummies. We compared each intervention group only with the control group to examine their independent efficacy. All effect analyses were corrected for potential confounders (ie, baseline behavior, baseline differences, and predictors of attrition). We calculated Cohen d to assess the size of the possible effects.

The study was approved by the Ethical Committee Psychology of the Faculty of Psychology and Neuroscience at Maastricht University, the Netherlands (ECP-138 08_03_2014) and registered at the Netherlands Trial Register (NTR4503).

Table 2 shows the characteristics of the total sample and the baseline differences between the 3 study conditions in terms of demographics, total minutes of physical activity per day, and minutes of moderate to vigorous physical activity. Comparison of baseline variables between groups showed no statistically significant differences.

Figure 1 shows the flow of respondents for the overall sample and separately for the 3 study conditions (see Multimedia Appendix 1 [50] for the CONSORT eHealth checklist). Analysis showed that the overall participation rate at follow-up (T4) was 77.5% (289/373). When comparing dropout rates between the 3 conditions, the highest dropout rate was in the control group, in which 71.7% (91/127) of the participants at baseline completed the last follow-up questionnaire. The lowest dropout rate was in the eHealth condition, with a participation rate of 84.8% (117/138).

Attrition analysis (Table 3) showed that respondents were more likely to complete the follow-up assessment when they were in the eHealth condition (compared with the control condition; odds ratio [OR] 2.43, P=.007), they were female (OR 2.16, P=.02), they were younger (OR 0.97, P=.009), and they had lower levels of daily moderate to vigorous physical activity (OR 0.99, P=.049). We included the significant predictors of dropout in all further analyses as potential confounders.

Results of the process analysis indicate that participants in the eHealth condition evaluated the intervention significantly better than did respondents in the mHealth condition for 3 items: receiving messages, reading messages, and the general clarity of the messages (see Table 4). For the other items, we found no significant differences between the 2 groups.

Regression analyses showed statistically significant differences between the intervention conditions and the control group for the total amount of physical activity (see Table 5). After 6 months, participants who used the intervention (ie, mHealth and eHealth together) were significantly more physically active than were participants in the control group (intervention groups: mean 56.35 minutes/day; control group: mean 47.79 minutes/day; B=8.48, df=2, P=.03, Cohen d=0.27). We found a small effect that was borderline significant for the difference between the eHealth group and control condition (eHealth: mean 57.91 minutes/day; control group: mean 47.79 minutes/day; B=6.13, df=2, P=.09, Cohen d=0.21) with regard to total physical activity. We found no effect between the mHealth group and the control group (mHealth: mean 54.78 minutes/day; control group: mean 47.79 minutes/day; B=1.92, df=2; P=.63, Cohen d=0.04) with regard to total physical activity. Secondary analyses with complete cases revealed similar results.

The aim of this study was to evaluate the efficacy, use, usability, and appreciation of 2 different versions (eHealth vs mHealth) of a Web-based computer-tailored physical activity intervention. Contradicting our hypothesis, the eHealth intervention resulted in better usability and appreciation than did the mHealth intervention. Further, we found no significant differences in use and effects between the mHealth and eHealth versions when compared with a control group. These findings imply that mHealth is not necessarily more suitable than eHealth interventions and even suggest that eHealth should still be preferred.

The effect analyses revealed a significant difference in physical activity when comparing the eHealth and mHealth versions against the control condition. Yet we found no differences in effect between the eHealth and mHealth versions. The effect size for the eHealth version suggests a small effect, but the significance level was only borderline significant due to the small sample size. Recent studies also suggested that the use of mobile phone-based interventions may have positive effects on physical activity and weight loss but did not compare the efficacy of mHealth versus that of eHealth [31,47,48]. In line with our findings, it has been suggested that mHealth may be less suitable to achieve behavior change, since participants in an mHealth condition can use the intervention wherever they are at any given time [49]. One explanation may be that mHealth participants may be more prone to distractions than eHealth users. eHealth users may be more committed to take the time to complete their tasks, whereas mHealth users may have been in distracting surroundings and situations such as supermarkets or public transport, which may lead to skipping or misreading messages. However, this explanation needs more research to demonstrate its applicability. The explanation is in line with the assumption of the elaboration likelihood model of Petty and Cacioppo [51]. The model explains that distraction can result in peripheral route processing rather than in more central processing, which is associated with more (enduring) behavior change [51,52].

The higher dropout rate in the mHealth condition can possibly also be explained by the fact that people are more easily distracted when using their mobile phone. A recent study showed that people tend to use their mobile phones during short waiting times (eg, waiting for the bus, waiting in line at a checkout) [53]. This means, on the one hand, that they use the device frequently; on the other hand, it implies that its use can be short and with many interruptions. Previous studies demonstrated that mobile phone use can distract people from other activities such as driving a car [54,55]. However, ongoing activities and the surroundings might also distract the person from the task he or she is doing on the mobile phone. Distraction might lead to worse performance, as well as to forgetting or neglecting the task completely [54,55]. Furthermore, the possibility of distraction might also explain the finding that the eHealth group evaluated the intervention significantly better than did the mHealth group regarding receiving and reading feedback messages, as well as the clarity of the feedback messages. Elaboration likelihood model research has shown that when information is processed via the peripheral route it is less appreciated by the receiver [56,57]. Furthermore, peripheral route processing can lead to lower motivation to engage with the context of the intervention, which would lead to the lower levels of appreciation [57].

Another explanation could be that, while using a mobile phone is often spontaneous and a direct action that is driven by technology, the use of eHealth might be much more user driven. This means that, whereas participants in the mHealth group might have felt obligated to check their message the moment they received it, regardless of time, place, and concentration, eHealth participants consciously chose to start their computer to check their emails. This feeling of autonomously choosing when to engage in the intervention can lead to more intrinsic motivation and appreciation of the intervention [58]. A different explanation for the low usability and appreciation could be the difference in the technology itself. The intervention was message based, which might have led to more misunderstanding of the feedback messages within the mHealth group than within the eHealth group [37].

An important strength is that, to our knowledge, this is the first study that compared an eHealth intervention with an mHealth intervention with regard to efficacy, use, and appreciation.

The first limitation is that all outcome measures were self-reported [59]. Research has shown that self-reported measures, in comparison with objective measures, have both a lower reliability and less validity. However, the IPAQ has been proven to be a reliable and valid measurement of physical activity [46]. Yet replication with other, more objective assessments for measuring physical activity, such as accelerometers, is recommended.

The second limitation is that it was necessary to replace missing values with multiple imputations. Although multiple imputations are often used, there is discussion about how to correctly apply this technique [60]. However, we found the same results regardless of whether we performed the analyses with the multiple imputation or with the completers-only dataset.

The third limitation is that our process analyses were not accompanied by qualitative measurements. For example, by asking participants why they found messages less clear, we could have gained insight into whether the difference between groups was based on technical difficulties only or could be attributed to the other factors.

The fourth limitation is that, because the tested intervention was message based, the results are difficult to generalize to the broader field of mHealth and eHealth.

The fifth limitation is that participants were all recruited from an online panel and were a random sample from the panel. This might make it difficult to generalize the findings from participants who are used to participating in scientific research to the broader population.

Lastly, as pointed out above as well, our sample size was limited. Each condition had approximately 100 participants, and power analyses revealed that we needed at least 141 participants per group to detect standardized effects of 0.20. As the results showed that effect sizes were indeed roughly 0.20, replication of this study with a larger sample is recommended to be able to demonstrate more statistically significant results.

Based on our results, we can conclude that the eHealth version outperformed the mHealth version of a Web-based computer-tailored physical activity intervention with regard to usability and appreciation, but not with regard to effectiveness.

The eHealth intervention excelled with regard to usability and appreciation compared with the mHealth intervention, and there are indications that the eHealth intervention may have been used more often. However, a study by Morrison et al [61] showed the advantages of combining mHealth and eHealth. They reported that, although their mHealth version did not function as an alternative to eHealth, it enhanced the intervention with regard to perceived accessibility, mobility, and on-the-go gadgets.

We recommend performing more research to assess and develop interventions that combine mHealth and eHealth technologies.