A three-armed randomized controlled trial was conducted with 2 experimental conditions (video and text intervention) and a waiting list control group that had the opportunity to use one of the interventions after the study. Measurements took place at baseline (T0) and 6 months (T1) after baseline. Criteria for participation were being at least 18 years old, having a paid job (because of initial recruitment procedure), a BMI between 18.5 and 30 kg/m², and sufficient command of the Dutch language. People with a physical condition that severely influenced their dietary or physical activity pattern (eg, diabetes) were not eligible to participate.

It was estimated that 2000 participants were needed to complete the baseline questionnaire in order to be able to detect a medium-sized effect (d=0.5) on BMI and behavior with a power of .90, a significance level of .05, and taking into account a dropout percentage of 50% between baseline and follow-up. This number of participants would also allow testing interaction effects between participants with a low, medium, and high level of education [26].

Participants were recruited from September 2012 until February 2013. Participants were recruited during medical screenings by various occupational health centers, directly through companies, and via advertisements in national and local newspapers. All recruitment materials (ie, brochures, emails, advertisements) included information about the intervention study as well as a hyperlink to the study website where participants could register to participate. After registration and giving online informed consent, participants were randomly assigned to one of the 3 study conditions (ie, video version, text version, and control group) in a computer-determined sequence. After randomization, participants received a username and password by email. Participants were unaware of which study condition they were allocated to until they accessed the baseline questionnaire (T0). Two weeks after completion of this questionnaire, participants in the intervention conditions were given access to the intervention. Participants could use the assigned intervention for a maximum period of 3 months. Six months after baseline, participants were asked by email to fill out the online follow-up questionnaire (T1). To decrease the likelihood of attrition, participants were informed that they could win one of hundred cash prizes of €100 if they completed all questionnaires [35].

The Web-based computer-tailored intervention was developed systematically using the Intervention Mapping protocol [36]. Detailed information about the development process and the content of the intervention can be found elsewhere [26]. The objective of the intervention was to prevent weight gain or achieve modest weight loss by making small changes in dietary intake and/or physical activity. In the video version, about 75% of the educational content was delivered via videos. The remaining 25% consisted of text-based content to give instructions about setting goals and making action and coping plans as well as for the delivery of optional in-depth information. The videos had a news-driven format in which professional actors read aloud the tailored information. This information was exactly the same as the information that could be read in text in the text version of the intervention. In both the video and text versions, the tailored information was based on participants’ answers to online questions about their dietary intake, physical activity level, and sociocognitive beliefs (eg, self-efficacy). The feedback was very specific and, for example, clearly indicated which specific behavior changes participants could make (eg, decrease intake of chocolate with X per day).

The theoretical framework of the intervention consisted of a combination of self-regulation theories [37,38] and the I-Change Model [39]. Self-regulation theories were in particular used as input for the general framework for the intervention. Accordingly, the intervention aimed to create awareness of behavior, identify areas for change, set goals and make plans, and finally start and monitor the behavior change. The I-Change Model has mainly been used to make people aware of their behavior and for indicating which behavior change participants were most motivated to do. In these sessions, participants received feedback about their motivational beliefs (eg, attitude and self-efficacy) and could make action plans. In line with these theories, the following behavior change methods were used: consciousness raising, tailored feedback on behavior and cognitions, goal setting, action and coping planning, and evaluation of goal pursuit.

The intervention consisted of 6 weekly sessions, and each session lasted about 15 minutes. After Session 1, participants could continue to Session 2 directly. Hence, between Sessions 1 and 2 there was no mandatory waiting period (in contrast to the subsequent sessions). Figures 1 and 2 provide an example of the video and text versions, respectively.

At both T0 and T1, multiple imputation was used to replace missing values [50,51]. Descriptive statistics and frequencies were used to describe the characteristics of the study population and the overall flow through the study. Baseline differences between the 3 study conditions were examined using analyses of variance with Tukey post hoc tests for continuous variables and chi-square tests with Bonferroni correction (P=.05/P=.017) for categorical variables. To examine the possible presence of selective attrition between baseline and follow-up, a logistic regression analysis was performed with attrition at follow-up as outcome (completed T1=0, not completed T1=1) and study condition and all baseline variables as predictors.

The effect analyses were conducted for each outcome variable separately (BMI, dietary intake, physical activity) using linear regression analyses with the enter method. The effects of the intervention conditions were compared to the control condition for which the study condition variable was recoded into 2 dummies (ie, video versus control and text versus control). The analyses were adjusted for potential confounders (ie, baseline behavior, predictors of attrition, and baseline differences) by including these variables as covariates. The analyses also included study condition × educational level interaction terms to assess potential educational differences in intervention effects. Cohen’s d effect sizes were calculated for all outcome variables [52]. As secondary analyses, we also compared the effects of the intervention conditions with each other. Moreover, the analyses were performed with both a complete case and multiple imputation dataset.

Finally, the process evaluation data were analyzed using linear regression analyses with the enter method. These analyses included study condition × educational level interaction terms to identify potential educational differences in appreciation. When no interaction effects were found, independent samples t tests were conducted to examine differences between the video and text conditions on the process evaluation variables (ie, appreciation).

All statistical analyses were conducted using SPSS 20.0, applying a significance level of .05 for single variables and .10 for interaction terms [53].

The CONSORT-EHEALTH flowchart [34] (Figure 3) shows the number of participants who were randomly assigned to one of the 3 study conditions as well as their flow through the study. In total, 1419 participants completed the baseline questionnaire. At 6 months follow-up, data from 1015 (71.53%) participants were collected. In the video condition, only 328 (70.54%) participants had completed the first session of the intervention, whereas 364 (74.13%) did in the text condition. Overall, the average number of completed sessions was 2.15 (SD 1.94) sessions. In total, 10.88% (104/956) of the participants had completed the intervention fully (ie, use of all 6 sessions).

Table 1 provides a comprehensive overview of all baseline characteristics of the study sample, including baseline differences between the 3 study conditions. Participants’ mean age was 48.13 (SE 0.31) and 58.56% (831/1419) were female. The mean BMI was 26.42 (SE 0.06), and 73.50% (1043/1419) of the study sample was overweight. The majority had a high level of education (769/1419=54.19%), while fewest participants had a low level of education (214/1419=15.08%). The distribution of educational level differed significantly between the 3 study conditions (Pearson χ² ₄=10.380, P=.004). Compared to the text and video conditions, significantly more participants in the control condition had a low educational level. Moreover, the number of participants with a medium level of education was significantly higher in the text and control conditions in comparison to the video condition. Last, participants’ mean score on goal setting in the video condition was significantly higher in comparison to the text and control conditions (F _2,2415=4.740, P=.009). No other baseline differences were observed.

Attrition analysis identified several significant predictors of dropout. Participants in the video (OR 2.11, 95% CI 1.48-3.00, P<.001) and text conditions (OR 3.23, 95% CI 2.29-4.54, P<.001) were significantly more likely to drop out compared to participants in the control condition. Attrition was further significantly higher among participants with a low (OR 2.15, 95% CI 1.46-3.16, P<.001) and medium (OR 1.37, 95% CI 1.02-1.85, P=.037) educational level in comparison to highly educated participants. Older participants were more likely to complete the follow-up questionnaire (OR 0.97, 95% CI 0.96-0.98, P<.001). Finally, participants who had lower levels of self-efficacy to improve their diet (OR 1.36, 95% CI 1.06-1.76, P=.016), intention to improve their diet (OR 1.22, 95% CI 1.02-1.46, P=.031), and coping planning regarding physical activity (OR 1.31, 95% CI 1.04–1.64, P=.022) were significantly more likely to drop out.

There were no significant interaction effects between type of study condition and educational level for any of the outcome measures.

The regression analyses without interaction terms showed several main intervention effects (Table 2). The video intervention had resulted in a significantly lower BMI compared to the control condition (B=-0.25, P=.049), with a small Cohen’s d effect size of 0.10 [52]. No significant difference was found between the text and control condition regarding BMI (B=-0.09, P=.474). Moreover, both the video (B=-175.58, P<.001) and text interventions (B=-163.05, P=.001) resulted in a significantly lower average daily intake of calories from energy-dense food products compared to the control condition, with medium Cohen’s d effect sizes of respectively 0.40 and 0.36 [52]. For physical activity, no intervention effects were found for both the video (B=-1.45, P=.900) and text conditions (B=1.88, P=.863) in comparison to the control condition. In the additional analyses comparing the 2 intervention conditions, no significant differences were found for any of the outcome measures. The complete cases analyses resulted in the same significant findings as the effect analyses with the multiple imputation data described earlier.

Table 3 provides an overview of the results from the process evaluation. In total, 355 participants completed the process evaluation questionnaire. Overall, the mean scores of the process evaluation variables represented neutral to slightly positive scores for both versions of the intervention, without remarkable low scores. The intervention scored best on usefulness, understandability, and autonomy. The mean score for assessment of the intervention as a whole was 6.85 (SD 1.14).

Regression analyses showed that there was no significant interaction effect of educational level regarding the process evaluation variables. Independent sample t tests showed that the information in the video condition was rated as more useful compared to the information provided in the text condition (t ₃₅₄=1.992, P=.047). Feelings of relatedness were also significantly higher among participants in the video condition (t ₃₅₄=2.056, P=.041) as compared to the text condition. Finally, participants in the video condition rated the intervention significantly better than participants in the text condition (t ₃₅₄=2.388, P=.018).

The aim of this study was to examine the effects and appreciation of video and text versions of a Web-based computer-tailored obesity prevention intervention among Dutch adults with low and high levels of education.

Our results showed no significant group × education interaction effects. This implies that both versions of the intervention were equally effective for all educational levels. The video version was the most effective intervention because it resulted in both a lower BMI and lower energy intake (compared to the control condition), while the text version had only a lower energy intake. No intervention effects on physical activity were found. Appreciation of the 2 intervention versions also did not differ per educational level. Yet the video version was appreciated more than the text version on usefulness of messages, feelings of relatedness, and grade given to intervention. Overall, it can be concluded that the video intervention performed better than the text intervention regardless of participants’ educational level.

The fact that we did not find support for our hypothesis that the video version would be more effective for people with a low educational level is not surprising. A recent similar study into a smoking cessation intervention has, for example, also only found a main effect of the video condition and no differential effects per educational level [45]. Furthermore, our hypothesis that video messages may work better for lower educated people [26] was based on indications and assumptions derived from a few previous studies but the evidence for this hypothesis was not compelling. Nevertheless, our study provides preliminary evidence that the use of videos in a Web-based computer-tailored intervention can be effective in the prevention of obesity regardless of people’s educational level [45].

Both the video and text versions had the strongest effects on dietary intake, which is a finding in line with 2 reviews on Web-based computer-tailored interventions [12,16]. The medium effect size indicates that this effect is of clinical relevance [52] and may suggest important public health potential when the intervention is implemented at a large scale. In line with other studies [11,15], only a small effect size was found regarding BMI (of the video version). However, even small intervention effects on BMI can have a large public health impact resulting in a significant reduction of many health problems, an improved quality of life, and cost savings [54-57]. The fact that the text version did not have a significant effect on BMI can possibly be explained by the fact that the effect size for dietary intake (0.36) was somewhat smaller compared to the effect size of the video version (0.40). The fact that no intervention effect was found on physical activity is not surprising. Many reviews have reported mixed findings of Web-based computer-tailored interventions for physical activity [12,19,20]. One explanation for this finding could be that we encountered problems with the measurement of physical activity in our study. The average daily minutes of moderate-to-vigorous intensity physical activity scores were unrealistically high. Consequently, many participants did not receive the advice to increase their physical activity level within the intervention, resulting in little improvement for this behavior. Recently, this problem has also been identified in a similar efficacy study. Hence, future studies should take this problem into account when assessing physical activity and developing algorithms to deliver tailored messages [46].

In line with two recent studies [21,45], it can further be concluded that a Web-based computer-tailored intervention consisting of videos is appreciated better than an identical intervention that consists of merely text. This difference in appreciation can possibly also explain why the video version was more effective than the text version. The Elaboration Likelihood Model [58], for example, suggests that when information is perceived as interesting and attractive, it is more likely that central route processing will occur. Information that is processed via this central path will have a more long-lasting persuasive effect on the receiver [32]. The better appreciation of the video version may therefore have resulted in more central route information processing. The fact that there were no differential effects in appreciation per educational level can also possibly explain the absence of educational differences in effects. This explanation is supported by the Communication Persuasion Matrix [59], which assumes that effective persuasion is the result of, among others, a suitable media channel. Yet our results demonstrate that the video version (ie, delivery format) is not more attractive for people with a low educational level (ie, user), and therefore no differences in outputs (ie, intervention outcomes) per educational level can be expected.

An important strength is that this is one of the first studies that has examined whether the use of videos can improve the effectiveness and attractiveness of Web-based computer-tailored interventions. Another strength is that our intervention met several criteria related to higher effectiveness of weight management interventions, such as the use of self-regulation theories [25,37,38], the small changes approach [54], and the Intervention Mapping protocol [36]. A final strength is that we found exactly the same results with the multiple imputation data and the complete cases data.

A limitation of this study is that all outcome measures were self-reported [60,61]. For example, it would have been better to measure BMI objectively. However, research has shown that self-reported BMI does not affect results when used as a continuous variable in a longitudinal study [62]. In addition, the SQUASH resulted in unrealistically high physical activity scores. As these scores were used to provide tailored feedback, it is likely that participants may have received inadequate feedback about their physical activity level. Future interventions should therefore aim to correct for this overestimation in tailoring algorithms or use objective measurements. Moreover, it should be noted that only a relatively small number of participants had completed the process evaluation questionnaire. Yet it is unclear if the inclusion of dropouts would have led to lower or higher scores. For example, participants may have dropped out because they had achieved their goal or because they did not enjoy the intervention. Finally, the relatively short follow-up period of 6 months can be regarded as a limitation. Research with a longer follow-up period is necessary to examine whether or not the effects will be maintained long term.

The video version of the intervention was more effective and better appreciated than the text version, regardless of participants’ educational level. Hence, our study provides evidence that the effectiveness of future Web-based computer-tailored obesity prevention interventions can possibly be improved by including videos as a delivery format in tailored health information. Our study shows that this is feasible and effective for Dutch adults with a healthy weight and limited overweight. However, more research is needed to study the long-term effects of the video version.