We conducted the systematic literature search between April and August 2015 using the following electronic databases: ScienceDirect, MEDLINE, PyscINFO, Scopus, the Cochrane Library, CINAHL, Embase, and Web of Science. The last search was conducted on August 17, 2015, with no new relevant papers found. We excluded studies published before 1990, as email was not widely used before this period [38]. After hand searching reference lists of key reviews and included studies, as well as conducting a manual search of JMIR journals, we included other relevant studies.

We conducted 2 searches. The first used combinations, synonyms, and truncations of “online intervention,” “computer-assisted therapy,” “electronic mail,” “Internet,” “website,” “cell phones,” “young adult” or “adult,” “fruits,” and “vegetables.” While we were searching largely for eHealth and mHealth interventions, we used other relevant MEDLINE MeSH, such as “telemedicine,” to encompass the terms “mHealth,” “eHealth,” “telehealth,” and “mobile health.” Furthermore, although we were mainly interested in the efficacy of vegetable interventions, we extended the search terms to include “fruit,” as studies typically report on fruit and vegetables concurrently. Additionally, we used the term “adult” alongside “young adult” to broaden the search from 18- to 24-year-olds (the typical database definition of young adults) to 18- to 35-year-olds (based on the US National Institutes of Health cut-off for young adults) [39]. Table 1 shows the first search strategy used in the MEDLINE. The full search strategy is presented in Multimedia Appendix 1 (Tables S1 and S2).

We conducted separate database and Google searches to locate programs that used social marketing and mass media to increase fruit and vegetable intake in young adults. Search terms were “young adult,” “adults,” “fruits,” “vegetables,” “social marketing,” “social media,” and “mass media.” These studies were not limited by publication type and included gray literature, such as nonpublished evaluations of programs by organizations. Table 2 presents the second search strategy used in MEDLINE.

Criteria for inclusion of eHealth and mHealth interventions were as follows: (1) randomized controlled trials (RCTs) with a primary or secondary aim of increasing fruit and vegetable intake in young adults that (2) were targeted at young adults aged 18–35 years inclusive, (3) reported fruit and vegetable intake at baseline and follow-up, (4) involved healthy participants with no disease or illness that would affect the primary outcome or ability to modify fruit and vegetable intake, (5) were written in English, (6) were published after 1990, and (7) were limited to eHealth- and mHealth-based interventions, defined as studies using texting, email, mobile phone apps, phone calls, or websites to deliver the intervention.

Criteria for inclusion of social marketing and mass media interventions were identical to points (1) to (6) above, but were not limited by study design. Social marketing and mass media interventions were defined as those that used media advertising through the Internet, television, billboards, radio, or social media platforms such as Facebook.

We downloaded titles and abstracts of all retrieved studies to EndNote X6 citation management software (Thomson Reuters). Duplicates were removed, then titles and abstracts were reviewed by grouping papers into (1) those meeting selection criteria or (2) requiring further examination; or (3) they were excluded. Papers determined to be potentially relevant to the review were downloaded as full text and reviewed for eligibility by two assessors (MN, JC) and further categorized (Figure 1). We resolved discrepancies in assessors’ results by discussion.

We created a data extraction table according to the principles of the PRISMA statement for reporting systematic reviews [35], with some additional elements included for completing the Cochrane Collaboration’s risk of bias tool [40]. Once we had piloted the process on a random selection of 4 of the included studies, 2 independent reviewers extracted the following data in duplicate: study details (authors, year, country of publication, funding, and affiliations); participants (characteristics, setting, inclusion and exclusion criteria, attrition, and blinding); intervention and comparator details; duration; and the summary outcome measure (change in fruit and vegetable intake between baseline and follow-up for the intervention and control arms). We also extracted the name of the tool used to assess changes in fruit and vegetable intake, as well as citations of available validation studies.

The primary outcome of interest was the change in vegetable intake postintervention. Where possible, for all study arms we recorded mean or median intakes (as servings, cups, frequency, or percentage consuming) pre- and postintervention. If vegetable intake was not reported separately, we documented the change in fruit and vegetable intake. We also noted the measures of error (SE or SD) and associated P values for change between groups over time. To determine the magnitude of intervention outcomes, we calculated effect sizes (Cohen d and 95% CIs) for studies that reported sufficient data (means, and measure of error or frequencies). Web-based calculators [41] based on Lipsey and Wilson’s formulas [42] assisted with calculations. We assessed the magnitude of the effect sizes according to Cohen’s categories, whereby an effect <0.2 is considered negligible, between 0.2 and 0.49 is small, 0.5-0.8 is medium, and >0.8 is large [43].

We also considered the clinical significance of outcomes. There is no consensus in the literature regarding what change in intake is considered clinically significant. However, several meta-analyses and longitudinal studies suggest a dose-response relationship, whereby an increase in vegetable intake by approximately 1 serving is protective for cardiovascular health (decreased risk of stroke and cardiovascular disease mortality by 11% and 4%, respectively) [44,45]. Furthermore, every 1-serving increase in vegetable intake has been associated with a 0.12 kg reduction in weight (95% CI -0.35 to -0.14) [46]. These studies define a serving of vegetables as approximately 1 cup of leafy vegetables or half a cup of cooked vegetables (frozen, fresh, or canned) in line with previous US and current Australian dietary guidelines [31,47].

To pool the outcomes for the meta-analysis, we grouped studies for which an effect size was calculated. We used STATA version 13 (StataCorp LP) to conduct the analyses using the metan, metabias, and metafunnel commands. A random effects model was applied. Publication bias was determined through Egger’s statistical test for funnel plot asymmetry and visual inspection of the funnel plots of the Cohen d effect size (standardized mean difference), plotted against its standard error. The I² value for heterogeneity was calculated based on the Q statistic: [(Q statistic - df/Q statistic) × 100%]. Cochrane Collaboration guidelines [48] suggest that an I² for heterogeneity below 40% is considered low, and a value above 50% is considered substantial.

We determined the validity of each tool used to measure changes in vegetable intake based on published literature demonstrating its accuracy [49-59]. The checklist of requirements by Nelson et al [60] was also consulted to qualitatively examine the effectiveness of reporting on measurement tools. This checklist assesses factors such as data-collection procedures (objective measure vs self-report), methods of quantifying portions, variety of foods captured, food composition databases used and whether checking procedures were applied.

We assessed the external validity of included studies based on the Green and Glasgow’s criteria [61]. The assessment explored components under 3 sections: (1) reach and representativeness of participants, (2) intervention implementation and adaptation, and (3) program maintenance and institutionalization (sustainability of program implementation). Quantitative and qualitative data pertaining to these external validity components were extracted. We recorded specific data that were not reported as not reported, and if an assessment component did not apply to the particular study we reported it not applicable. Individual participation rate (%) was calculated as the percentage of eligible participants agreeing to participate. Attrition rate (%) was calculated as the percentage of participants who dropped out after randomization. Attrition was further grouped by intervention arm (treatment vs control). Extracted data were used to examine the number of studies adhering to the external validity components. The frequency and adequacy of reporting of these components were also examined and compared between studies.

As the flow diagram in Figure 1 shows, we found 2680 studies through database searching and 3 additional studies through hand searching the references. We screened a total of 2252 papers by title and abstract. Of these, we assessed the full text of 87 studies. A total of 14 studies [62-75] met the selection criteria and were included in the review. See Multimedia Appendix 2 for the complete list of references excluded by full text with corresponding reasons. None of these studies used social marketing strategies or mass media to encourage vegetable consumption in young adults specifically. Therefore, the remaining results report the effectiveness and external validity of eHealth and mHealth interventions aimed at increasing fruit and vegetable intake in young adults. We included 12 studies in the meta-analyses. For the meta-analysis, we combined the reported results in 2 groups for comparability: fruit and vegetable (8 studies) and vegetable only (5 studies); 1 study contributed results for both groups [64].

Overall, 7984 healthy people participated in the eHealth and mHealth RCTs (see Multimedia Appendix 3, Table S3). There were, however, large discrepancies in the sample sizes. Only 3 of the 14 studies had recruited >500 participants at baseline [62-64], and 1 study had a sample size of <100 [64]. More than half of the interventions were conducted in the United States [62,64,67,69,71,73,75], 4 in Australia [65,65,67,72], 1 in New Zealand [74], and 1 in Malaysia [70]. The target audience was college or university students for the majority of the studies [62,63,67-75], and 3 studies reported their target audience to be young adults [64-66].

Recruitment methods were reported for 13 of the 14 studies, but limited details were provided. All but 2 studies recruited through the university or college setting [64,66]. Participants were recruited through undergraduate psychology courses in 2 studies [67,68], from random nonnutrition classes in 2 studies [69,70], and through advertisements and flyers posted on university grounds in 4 studies [62,63,65,71,72]. In 1 study a recruitment table was set up on campus [73], and another study invited patients attending the student university health service [74]. In 1 study [64] advertisements with a toll-free phone number were used, and the final study distributed letters of invitation through participating family doctors, along with electronic and print advertisements [66]. Of the included studies, 9 indicated their participation rate, with a mean of 78.0%. The inclusion criteria were detailed by 10 studies, all of which specified age (years) as one of their criteria. Demographic data were provided by most of the studies although not consistently. Baseline age (years) was reported in all but 1 study (Multimedia Appendix 3, Table S3), with a mean age of 20.8 years across the studies. The ethnicity of participants was reported to be >50% Caucasian or white in 7 studies. The percentage of female participants was reported by 13 studies, with women more commonly recruited than men (mean 69.8% female) (Multimedia Appendix 3, Table S3).

Details of the intervention and comparator groups were provided in detail. All studies recruited an intervention and a control group (see Multimedia Appendix 4, Table S4), with 4 studies using multiple intervention and control arms [67,71,72,74]. A total of 6 studies provided no treatment to the control arm [67,68,70,71,74,75], 7 studies gave the comparator group general information not containing the intervention material [62,64-66,69,72,73], and 1 study provided the control group with the intervention material on completion of the follow-up assessment [63]. The duration of interventions and number of sessions were easily extrapolated from each study. The level of contact between researchers and participants ranged from one-off sessions (provision of feedback) to daily contact by email or text message (Multimedia Appendix 4, Table S4). The majority of the interventions used online education through learning platforms, websites, and emailing, with only 2 studies using apps [65,66] and 4 using text messaging [65,66,70,72]. No studies reported the use of social media platforms. The studies predominantly used goal setting for behavior change, with monitoring and feedback also commonly incorporated. For the majority of the interventions, the aim was to offset weight gain in young adulthood. Targeting improvements in fruit and vegetable intake was one such method used to address weight gain. While 1 study was designed to reduce health-risk behaviors in young adults [74], only 5 studies focused specifically on fruit and vegetable intake [64,68,69,72,75], and none targeted vegetables alone.

The reviewed studies varied in the detail provided regarding the behavior theories and techniques considered in the intervention design. The design of 5 studies was based on the transtheoretical model of behavior change, where the participants’ stage of change determined the content received [63-66,75]. A total of 6 studies were theory or education based [62,63,67,70,71,73]. Social-cognitive theory informed 2 interventions [67,69]. Half of the reviewed studies applied the behavioral construct of self-efficacy in their intervention [62,64,69,70,71,73,75]. The study by Kypri and McAnally [74] did not report consideration of theoretical frameworks in their intervention design. The remaining 2 studies [68,72] were informed by the theory of planned behavior and the theory of habit formation (Multimedia Appendix 4, Table S4). All the studies that we reviewed intervened at the individual level. Only 2 studies were implemented outside of the university setting, thus limiting the generalizability of the interventions to the overall young adult population. Of these studies, one [64] was targeted at lower socioeconomic status young adults, while the other mainly captured young adults from higher socioeconomic areas [66].

The duration of the interventions (excluding postintervention follow-up) ranged from one-off contact to 6 months of treatment, with a mean of 10 weeks (Multimedia Appendix 4, Table S4). A total of 9 studies allocated a follow-up period [62-64,66,69,71-73], with a mean of 16 weeks. Adherence was most commonly documented as the number of sessions completed or the amount of materials viewed by participants (Multimedia Appendix 4, Table S4), but was not consistently reported across studies. The mean level of compliance among those reporting adherence was 85.4%.

Delivery expertise varied among the studies (Multimedia Appendix 4, Table S4). Research staff were more commonly reported to have conducted the interventions, with little specification of their qualifications and the number of research staff involved. Registered dietitians delivered 5 of the interventions [63,65,66,69,75]. Other expertise included a health promotion officer [71] and outreach educators [64].

The rate of attrition was documented in all reviewed studies. At completion of the interventions the mean attrition rate was 19.6% (see Multimedia Appendix 5, Table S5). All but 4 studies [64,69,71,75] reported attrition for the control and intervention group separately, and 4 did not assess differences in characteristics between completers and noncompleters [65,70,71,73]. Only 2 studies looked at the long-term impacts of the study, by assessing outcomes at least 12 months following treatment [62,63]. Both of these studies found that the changes in fruit and vegetable intake were not maintained at follow-up (Multimedia Appendix 5, Table S5). The sustainability of program implementation was poorly reported, with only 1 study mentioning that results would be used to refine the intervention for trial in a broader young adult population using a larger sample size [66]. Finally, only 2 studies published a process evaluation documenting effective program elements [62,66].

We rated the majority of the studies reviewed as unclear to high risk because they did not perform intention-to-treat analyses, which introduced biases in the outcome data (attrition bias) [62,70-75] (see Multimedia Appendix 6, Table S6). We rated 2 studies high in a second domain (detection bias) [71,73]. The majority of the studies did not clarify their methods of blinding (n=8). Selection bias was mainly unclear within and across studies, with 5 studies not reporting the method of sequence generation in randomization [62,64,69,71,75] and only 2 studies specifying allocation concealment methods [66,74] (Multimedia Appendix 6, Table S6). While all of the studies reported results for prespecified outcomes, we could not completely rule out reporter bias across studies because only 5 RCTs published their original protocol [63,65,66,68,69] or provided details of their trial registration [66]. However, no selective reporting was apparent based on the methods within the reviewed manuscripts (both successful and unsuccessful outcomes recorded). Overall, the combined lack of clarity of the level of bias across studies raises concerns about the plausibility of the studies’ results.

The reviewed interventions had several limitations in study design and did not address the research question directly, resulting in an overall low quality rating (Table 3).

Of the 14 reviewed studies, 9 provided results for fruit and vegetable intake, and we included 8 in the meta-analysis. Of these studies, 7 found positive effects postintervention [62-64,67,68,74,75] (Cohen d 0.14-0.56), 4 of which were statistically significant [62-64,75]. For all but 1 study [75], the magnitude of effect was small. In total, 2 studies also reported clinically significant improvements of ≥1 serving/day [62,75] (see Multimedia Appendix 7, Table S7). The pooled effect size for interventions reporting change in fruit and vegetable intake was 0.22 (95% CI 0.11 to 0.33), indicating a small positive effect of eHealth and mHealth interventions on fruit and vegetable intake. The 4 studies [62-64,75] with significant effects contributed 72.9% weighting (Figure 4). The I² was 68.5%, P=.002, suggesting considerable heterogeneity between these studies, and so findings should be interpreted with caution.

Of the 6 studies that assessed vegetable intake independently of fruit [64-66,70-72], we included 5 in the meta-analysis, 4 of which had positive effects on vegetable intake [64-66,70] (Cohen d 0.11-0.40). Two of these positive effects were statistically significant [64,66]. Increases in intake were <1 serving/day, with the exception of the results reported by Partridge et al [66] (Multimedia Appendix 7, Table S7). The pooled effect size for change in vegetable intake was negligible at 0.15 (95% CI 0.04 to 0.28; I²=31.4%, P=.2) (Figure 5).

Studies that were more successful in improving fruit or vegetable intake provided participants with individually tailored advice and feedback based on their stage of change [64,66,75] and incorporated goal setting [62,66,75]. Of the studies producing clinically and statistically significant results for fruit or vegetable intake, or both [62,66,75], 1 used online theory education based on nondiet principles [62]. This intervention was designed according to 2 educational models, Carey and colleague’s system of instructional design [76] and Keller’s instructional motivational model [77]. Fruit and vegetable intake goals were set after completion of each weekly educational lesson, and self-evaluation of progress preceded the next weekly Web-based module. The study by Richards and colleagues [75] used motivational interviewing in combination with Web-based resources and emails. The resources were tailored to the participants’ stage of change, where precontemplators and contemplators were given reasons to and tips on how to eat more fruits and vegetables, as well as a goal-setting framework. Action and maintenance participants received emails with tips for maintaining consumption and trying new fruits and vegetables. Finally, the study by Partridge et al [66] combined multiple eHealth and mHealth strategies to support behavior change, with text messaging found to be the most popular, and the website and discussion boards the least popular, among participants. The text messages contained reminders and tips on how to achieve their individualized goal set during their phone counseling session with a dietitian and were based on the 10 processes of change (transtheoretical model). Participants could monitor their fruit and vegetable intake goals using a personalized app that also provided recipes and tips on how to increase their intake.

Of the reviewed studies, 5 used tools that had not been validated to assess changes in vegetable intake [68,69,71,73,75] (Table 4). While the majority of the tools were validated, only 1 was tested specifically in the young adult population [30]. Of the studies that used validated tools, short screeners were most popular, including the US National Cancer Institute’s fruit and vegetable screener [53], as well as short questions adapted from the Australian and New Zealand national nutrition surveys [52,54,56]. Furthermore, only 2 studies defined what they classified as a serving [65,66], and the outcome measure for intake lacked consistency, with studies reporting change in terms of frequency, servings or cups of vegetables consumed, as well as the percentage meeting recommendations. No studies detailed which food composition databases they used for the analysis, or whether they checked records with respondents as per the requirements specified in the Nelson and colleagues’ checklist [60]. All but 1 study [70] used a self-report measurement tool. The study by Gow and Colleagues [67] did not specify what the outcome measure was (servings vs score).

This is the first systematic review to report on the effect of eHealth and mHealth interventions on vegetable intake specifically and highlights relevant opportunities for future research. We conducted the review protocol in line with the PRISMA guidelines [35] and used a comprehensive search strategy. While we searched several electronic databases and made an effort to include gray literature, we may have missed some studies. The variability across interventions with differences in study designs and measures of vegetable intake, and the overall poor study quality, made it difficult to establish definitive conclusions. Consequently, we were reluctant to rule out any eHealth or mHealth approach as ineffective and rather discussed the outcomes as a means of highlighting gaps in the current literature and opportunities for future research to generate a stronger body of evidence on whether technology-based strategies are effective in this population. Finally, the lack of consistent reporting of external validity components prevented us from making conclusions about the potential for translating interventions to the wider young adult population.

Overall, this review revealed that young adults have been neglected in fruit and vegetable social marketing campaigns, and most interventions target fruit and vegetables concurrently. Very few good-quality eHealth and mHealth interventions using validated dietary assessment tools have been designed to support young adults in improving their vegetable intake. With preliminary evidence suggesting that eHealth and mHealth strategies may be an effective mode of delivering vegetable interventions, continued research using stronger and higher-quality study designs is required to better determine the efficacy of technology-based strategies for improving vegetable consumption in young adults. With previous research suggesting that multiple behavior change strategies should be used for greater improvements, researchers could consider combining promising strategies such as goal setting and tailored feedback in future interventions. The potential impact of using social media platforms to create awareness of the importance of eating enough vegetables also deserves attention. Finally, in light of the lack of reporting of external validity components in the reviewed papers, it is critical that future studies address key factors such as program costs, sustainability, and longer-term impact in order to determine the potential for upscaling interventions to the broader young adult population.