Background

Low vegetable and legume consumption is a leading modifiable risk factor for non-communicable diseases globally [1, 2], accounting for over 2% of global deaths in 2017 [1]. International guidelines for vegetable intake recommend at least 3 serves/day (≥ 240 g/day) [3]. However, nationally representative survey data from 162 countries found that, in 2020, an average of 88% of the populations of these countries had an inadequate vegetable intake [4].

Interventions designed to address low vegetable intake often target low fruit intake simultaneously [5]; however, this is more likely to increase fruit intake than vegetable intake [6]. This is largely attributable to interventions not addressing barriers to vegetable intake, which are distinct from those of fruit intake, including lower palatability, lack of cooking confidence, and perceived higher cost and time to purchase, prepare and cook vegetable-rich meals [6,7,8,9,10,11]. Interventions that specifically focus on vegetables show promise, but are often setting-specific and delivered face-to-face, such as a workplace interventions [12]. While setting-specificity may be an important component of some personalisation approaches, more scalable approaches are needed to ensure interventions can serve large populations across a wide range of settings [13,14,15].

As an estimated 66% of people globally have access to the internet [16], digital interventions provide an accessible delivery model for increasing vegetable intake in adults [10, 11]. Furthermore, digital interventions are well aligned with the global drive to utilise digital technologies to improve health [17]. For example, 55% of European citizens aged 16–74 reported that they had sought online health information [18], and 88% of Australians reported wanting to access their health information digitally [19]. However, while there is some evidence that digital interventions increase fruit and vegetable intake [20], the effectiveness of digital interventions to increase vegetable intake alone is unclear.

Digital interventions offer the ability to personalise content and delivery to the needs and preferences of the user. Although evidence from randomised controlled trials (RCTs) suggest that personalised dietary advice motivates greater improvement in dietary intake than generalised dietary advice [21], personalisation of digital interventions alone may not be sufficient to increase vegetable intake. To help ensure dietary interventions meet the needs of the user, interventions are increasingly being designed with stakeholders, i.e., using co-design practices [22].

Co-design practices involve the lived experiences of the users, and individuals with technical expertise or service providers in the design process [23]. Research suggests that the use of co-design may help improve consumer engagement and satisfaction with a digital intervention by ensuring it meets their needs [23,24,25]. However, there is limited understanding of whether existing digital interventions to increase vegetable intake have used co-design methods or whether the use of co-design contributes to effectiveness.

Mediators of behaviour change, including knowledge of, attitudes towards, and skills in using vegetables, can be targeted in digital interventions to meet the needs of the user [26, 27]. However, achieving higher vegetable intake is also dependent on complex interactions between individual- and environmental-level influences, such as self-efficacy or access to affordable and healthy foods, which require specific policy actions [7, 8]. The NOURISHING framework [28], which maps interventions according to their alignment with policy actions related to behaviour change communications, the food environment or the food system, is a useful framework for considering such approaches. By mapping across each of these domains, gaps, and opportunities for policy actions for achieving behaviour change can be identified and targeted by digital interventions. Therefore, we aimed to systematically review digital interventions to increase vegetable intake in adults to: 1) describe the effectiveness of the interventions in terms of increased consumption; 2) examine links between effectiveness and use of co-design, personalisation, behavioural theories, and/or a policy framework; and 3) identify other features that contribute to effectiveness.

Methods

The protocol for this systematic review is registered with the international prospective register of systematic reviews (PROSPERO; CRD42022290926). The design and reporting of this review were guided by the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) statement (Additional file 1) and the synthesis without meta-analysis (SWiM) in systematic reviews reporting guidelines [29].

Eligibility criteria

The population, intervention, comparison, outcome (PICO) framework was used to develop the inclusion and exclusion criteria for study selection. Study designs included RCT, pseudo-RCTs, and pre-post interventions. The population included community-dwelling adults (18 years and older). Studies were excluded if they included pregnant and/or lactating women and/or institutionalised adults. Studies on populations for primary and secondary prevention were included. Interventions were included if they were a digital intervention targeting knowledge of, attitudes towards, and skills in using vegetables. In this review, “digital interventions” were interventions that included any of the following digital components: applications (apps; native, web, progressive and hybrid), websites, computer programs, mobile games, Short Message Services (SMS), Social Networking Services (SNS) and wearable devices [10]. Multi-modal interventions with non-digital components (e.g., face-to-face consultations) were included if digital features represented the primary focus of the intervention. The focus of this review was on vegetable intake, so the primary outcome was change in vegetable intake (i.e., measured as serves, portions, or grams/day). Secondary outcomes considered included changes in attitudes, knowledge, skills, self-efficacy, access and/or intentions related to vegetable intake. Studies were excluded if vegetable intake could not be examined separately. Only peer-reviewed original research articles published in English were included.

Search strategy

The search was developed in consultation with a librarian and undertaken in November 2021 and updated in August 2022. Published literature from January 2000 to August 2022 was searched. The year 2000 was selected as this coincided with an increase in the use of digital technologies in nutrition research and is in alignment with similar reviews of digital interventions [30]. The following databases were searched: MEDLINE (Complete), Embase, PsycINFO, Scopus (only extra searching), CINAHL (EbscoHost), Cochrane Library (Wiley), Rural and Remote Health database (INFORMIT), Health and society database (INFORMIT), IEEE Xplore, ClinicalTrials.gov and the Australian New Zealand Clinical Trial Registry. The full search strategy can be found in Additional file 2. Briefly, search terms were combined using the AND/OR operators for digital (‘digital, ‘smartphone’, ‘website’, ‘app’), intervention (‘intervention’, ‘randomized controlled trial’) and outcomes (‘vegetables’). Reference lists from systematic reviews identified in the search and included records were hand-searched to identify any additional studies. Where relevant protocol papers were identified during the search, an attempt was made to find the accompanying trial papers.

Data extraction

Studies were screened using Covidence software by two members of the team (KML, LA), first by title and abstract and then by full text. Discrepancies were resolved by discussion. Duplicates were removed in Covidence. Data were extracted by one reviewer (KML) and checked by a second reviewer (LA). A data extraction template was developed and piloted in Excel specifically for this review. The following information was extracted from each study: study design (setting, intervention and control conditions, duration), intervention features (digital tools used, co-design methods, behaviour change framework and taxonomies used, personalisation, NOURISHING framework policy domains and areas), population (country, age, sex, rurality, primary or secondary prevention); outcome measures (primary or secondary outcome, change in intake, behaviour, attitude, knowledge, skills, self-efficacy, intention and/or access); results for vegetable intake and effectiveness (yes/no determined based on statistically significant results for vegetable intake).

Data synthesis

A descriptive synthesis of the findings from the included studies was conducted. No meta-analysis was undertaken due to the heterogeneous nature of the digital tools used, characteristics of the populations in the included studies and the indicator of vegetable intake reported. The effectiveness and features of all interventions were summarised to better understand the characteristics that may increase likeliness of effectiveness. Features investigated included the population and study design, such as age, sex, rurality, use of co-design practices, behaviour change theory and personalisation methods. Studies were also mapped against the World Cancer Research Fund International’s NOURISHING framework [28]. This framework comprises three broad domains of policy actions (food environment, food system and behaviour change communication), 10 key policy areas within these domains, and the specific policy actions, which should be identified and implemented by policymakers to fit their national contexts and populations [28]. Examples of policy areas for these three domains included using economic tools to address food affordability (food environment domain), supply chain actions (food systems domain) and nutrition education and skills (behaviour change communication domain). We mapped whether the three broad domains and underlying 10 key policy areas were employed in the design of the intervention.

Risk of bias assessment

Two authors (KML, SP) performed an independent assessment of the risk of bias on the included studies, with any discrepancies resolved by consensus. Three Cochrane Risk of Bias tools were used: for randomized trials (RoB 2), for cluster RCTs (CRCT; RoB 2 CRCT) and for non-randomized studies of interventions (ROBINS-I) [31, 32]. The RoB 2 and RoB 2 CRCT domains for risk of bias assessment included randomization process, deviations from the intended interventions, missing outcome data, measurement of the outcome and selection of the reported result. The judgement within each domain was assessed to carry forward to an overall risk of bias judgement as low risk, some concerns or high risk of bias. The ROBINS-I domains for risk of bias assessment include confounding, selection of participants, classifications of interventions, deviations from intended interventions, missing data, measurements of outcomes and selection of reported results. The judgement within each domain was used to inform an overall risk of bias judgement as either low-risk, moderate-risk, serious risk, critical risk or no information reported.

Results

The search strategy retrieved 1,347 records (Fig. 1). After the removal of duplicates, 1,049 articles were screened for inclusion based on their title and abstract. Of these, the full texts of 97 articles were screened. This review included 30 studies [33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62] (Table 1).

Fig. 1
figure 1

PRISMA flow diagram of study selection

Full size image
Table 1 Characteristics of included studies (n = 30)
Full size table

Study characteristics

The 30 included studies comprised of RCTs (n = 22) [33,34,35,36,37,38, 40, 41, 43,44,45,46,47, 49,50,51,52,53, 55, 57, 61, 62], a CRCT (n = 1) [48] and non-randomized trials (n = 7) [39, 42, 54, 56, 58,59,60]. Intervention duration ranged from 3 days [35] to 2 years [43, 51]; more than half (n = 17; 57%) of studies had a follow-up period less than 6 months. Most studies were conducted in Australia [41, 42, 45, 48, 52, 58, 61], followed by the United States [35, 37, 38, 49, 53], Spain [34, 43, 50, 56, 59], the Netherlands [40, 51], the United Kingdom [47, 62], Belgium [60], France [44], pan-European [57], Israel [33], Iran [36], Brazil [39], Bangladesh [46], China [54] and Mongolia [55]. The studies included sample sizes ranging from 16 [56] to 5,055 [51], with 16 studies (53%) including a sample of 150 or more participants. The mean age of participants ranged from 18 years [54] to 70 years [38], with many (n = 19) conducted in mid-aged and older adult populations (> 40 years). Two studies delivered the digital interventions exclusively in rural areas [46, 48]. Eleven (37%) interventions recruited populations with health conditions, including hypertension [36, 39, 46], type 2 diabetes mellitus [33, 34, 44, 55], heart disease [43] prostate cancer [38] and overweight or obesity [45, 53]. The remaining studies were conducted in generally healthy populations and were designed to improve diet and/or lifestyle (n = 18) or weight management (n = 1). Over half of the studies (n = 17) were published since 2019.

Risk of bias

Risk of bias within 25 (83%) studies was high or serious because of missing outcome data for RCTs or bias due to confounding in non-RCTs (Additional file 3). Most RCTs (n = 17) and the CRCT adequately generated and concealed allocation resulting in no imbalances apparent between groups. Participant blinding was not possible because of the nature of digital health interventions and was not considered to increase risk of bias. The measure of assessment of vegetable intake was considered appropriate in most RCTs and the CRCT except for three studies where insufficient information was provided. Assessors were blinded to the intervention received by participants in 11 studies. Assessment of the outcome could have been influenced by knowledge of intervention received. However, this was deemed unlikely due to the dietary assessment methods and protocols used to assess vegetable intake, where it is unlikely that dietary coders were aware of the intervention allocation. Finally, seven studies did not reference a protocol or trial registration with a pre-specified analysis plan that was finalized before unblinded outcome data were available for analysis, which may be due to publication preceding the development of reporting guidelines.

Characteristics of digital tools

The most common digital tools used in the included studies were apps (n = 19; 63%), followed by SMS messaging (n = 10; 33%) and websites (n = 9; 30%). Some studies also used phone coaching and emails, and some interventions included a ‘dashboard’ feature to summarise resources and goals [39, 47]. Just under half (n = 13; 43%) used a combination of digital tools (Table 2).

Table 2 Summary of the features of digital interventions grouped according to effectiveness
Full size table

Vegetable intake

As shown in Table 1, vegetable intake was a primary outcome in 63% of studies (n = 19). Of these, some studies reported vegetable intake as a component of a Mediterranean diet score (n = 4), International Diet Quality Index (n = 1), m-Alternate Healthy Eating Index [62] or an overall diet quality index for Dominican adults [56]. Vegetable intake was assessed in most studies using brief diet questions [35, 36, 41, 42, 45, 46, 54,55,56, 58, 60], followed by a food frequency questionnaire [33, 37,38,39,40, 48, 51, 57, 61, 62], 24 h recall [57, 59], Mediterranean diet adherence screener [34, 43, 50], and an image-based dietary assessment tool [52].

Co-design practices

As shown in Table 2 and Fig. 2, 40% of studies (n = 12) reported some level of stakeholder input into the intervention design. Only one study, by Lara et al., referred to co-design specifically; a seven-stage, sequential, iterative series of workshops were used for designing, prototyping, testing and optimising the intervention, which was undertaken with researchers, older adults (the target population) and health and social care professionals [47]. This study was designated as using true co-design. Of the studies that reported stakeholder input, health care professionals, such as dietitians and general practitioners, were the most commonly reported stakeholders involved in the design, followed by software engineers. Only five studies reported involving consumers with lived experiences, including young adults (aged 18–30 years) in the Connecting Health and Technology (CHAT) study [52], adults aged over 55 years in the Living, Eating, Activity and Planning through retirement (LEAP) study [47] and Arab adults in a trial of ethnic minority adults with type 2 diabetes mellitis [33].

Fig. 2
figure 2

Summary of features of digital interventions to increase vegetable intake

Full size image

Personalisation methods

Twenty-three studies (77%) included some level of personalised intervention feedback (Table 2 and Fig. 2). The degrees of personalisation ranged from low (e.g., feedback based on assessment of current diet [52]), to moderate (e.g., personalisation of menus and shopping lists [44]), to high (e.g., individual coaching from a dietitian [38]); only one study reported offering participants the opportunity to customise their personalisation, based on preferred frequency and timing of text messaging [41]. Seven studies provided access to diet or physical activity coaching by a health professional via an app [33, 38, 41, 43, 45, 48, 54], phone calls [38, 41, 48], video calls [43], and SMS messages, emails and online forums [45]. One study personalised content to specifically address barriers to vegetable intake based on participant responses [40], another study used a digital program to design a personalised daily or weekly menu based on user preferences such as taste in foods, season and price range [44], while another study created a personalised video to promote healthy lifestyle behaviours based on age, gender and individual type 2 diabetes risk factors [61]. SMS-based interventions often used the participants’ name within the content [34, 36, 39]. Four studies provided personalised feedback and/or action plans based on demographic characteristics (such as age, sex, ethnicity and culture) and/or participant preferences [34, 36, 39] although limited information was provided on how this personalisation was designed or delivered, or whether personalisation was applied to the dietary component of the intervention. Other studies included some aspects of individualised support, although access to advice and support from dietitians was not provided [42, 62].

Theoretical underpinning and framework

Twenty-one studies (70%) reported embedding behaviour change theories into intervention design and delivery. Social cognitive theory and the trans-theoretical model were the two theories/models used most to underpin the interventions, with behaviour change techniques such as goal setting, motivational interviewing or action planning most frequently used (Table 2 and Fig. 2). When mapping against the NOURISHING framework, all studies aligned with the behaviour change communication domain, with the two policy areas of “nutrition education and skills”, and “nutrition advice and counselling in health care settings” identified. One study also mapped to the food environment domain, with the policy area of “economic tools to address affordability and purchase incentives” identified [42]. In this study, participants accumulated points and received a monetary reward at the end of the intervention relative to the number of healthy dietary choices logged. No studies aligned with the food system domain.

Effectiveness of digital interventions

Only nine studies (30%) reported statistically significant improvements in vegetable intake (i.e., designated as effective interventions) compared with a control group [38, 44, 55] or compared with baseline. In the latter case, this included pre-post interventions [56, 58], uncontrolled randomised trials [42] and RCTs with no statistically significant increase in the control group (and no statistical comparison for between-group changes reported) [35, 54, 60]. There was heterogeneity in the method of reporting improvements in vegetable intake among effective studies, including serves/day and adherence to guidelines. Three studies reported change in serves/day, with the magnitude of this improvement ranging from 0.29 serves/day [38] to 1 serve/day [42]. One study reported that 87% of participants improved vegetable intake compared to 29% of the control group [55], while another study reported a 7% increase in adherence to ≥ 500 g/day of vegetables compared to baseline (and a non-significant increase in the control group) [54]. One pre-post study reporting a 3.75 points increase in vegetable score (as a component of the Global Diet Quality Index; maximum score 100) compared with baseline [56]. Two studies also reported improvements in vegetable intake, but limited data on the magnitude were provided and no statistical comparisons were reported [36, 37]. Three studies reported a decline in vegetable intake compared with baseline, including a 0.2 portion per day decline [47], a 4% decline in participants consuming ≥ 2 serves/day [50] and a further study did not report any data on the magnitude of change [49]. No studies included in this review reported on attitudes towards, knowledge of, skills in respect of, self-efficacy, access to and/or intentions with respect to vegetables.

Features of effective digital interventions

Of the nine effective interventions, sample sizes ranged from 120 to 171 participants (Table 1). A slightly greater percentage of effective interventions were in healthy populations (n = 6/9; 67%) compared with the ineffective interventions (n = 13/21; 62%). Almost half of effective interventions were in younger adults (< 40y; n = 4, 44%), compared with 19% (n = 4) of ineffective interventions. Neither of the two interventions delivered exclusively in rural communities were effective. Vegetable intake was the primary outcome in 78% (n = 7) of the effective interventions, compared with 57% (n = 12) of the ineffective interventions.

Of the effective interventions, 33% (n = 3) utilised an app [35, 54, 58], 22% (n = 2) used a website [44, 60] and 11% (n = 1) used SMS messages [55] in isolation, while one study used an app and activity tracker [42] and two studies utilised a combination of four or more delivery modalities (including apps, emails, SMS messages, phone calls, videos and websites) [38, 56]. As shown in Table 2, this contrasted with the ineffective interventions, where 29% (n = 6) utilised an app [33, 34, 36, 39, 40, 62], 10% (n = 2) used a website [47, 51], and 10% (n = 2) used SMS messages [41, 46] in isolation, while 52% (n = 11) used a combination of delivery modalities [37, 43, 45, 48,49,50, 52, 53, 57, 59, 61].

The features of effective and ineffective interventions are compared in Fig. 3. Eighty nine percent (n = 8) of the effective studies referenced behavioural theories in their design (Table 2), including the trans-theoretical model theory [55], the social cognitive theory [38] and the health action process [60]. In contrast, 61% (n = 12) of the ineffective interventions referenced theories. Sixty-seven percent (n = 6) effective interventions delivered personalised information, which included personalised dietary advice from a dietitian [34, 54] and personalised menus and food shopping lists based on taste preferences and calorie needs [44]. Of the ineffective interventions, 81% (n = 17) included personalisation methods. Forty-four percent (n = 4) of the effective interventions included some level of input from stakeholders into the design of the intervention, compared with 38% (n = 8) of the ineffective interventions. This included design input from health care professionals, such as dietitians and general practitioners, and software engineers, but rarely involved meaningful consumer involvement. Only one (ineffective) intervention included true co-design, with iterative workshops with researchers, older adults (the target population) and health and social care professionals (Fig. 3).

Fig. 3
figure 3

Heat map summary of features of effective and ineffective interventions to increase vegetable intake

Full size image

Discussion

In this systematic review we identified a paucity of digital interventions that were effective at increasing vegetable intake in adults. Embedding of behaviour change theories and inclusion of stakeholders in the design of the intervention were more common among effective interventions. We also observed that personalisation did not appear to be a feature of effective interventions. However, personalisation methods varied considerably, thus it is possible that the nature or degree of personalisation did not meet the needs of the user. Use of more comprehensive co-design methods may help to ensure that personalisation approaches are informed by the needs of the target population.

This review found that researchers used multiple, heterogenous indictors of vegetable intake when reporting outcomes from interventions, which prohibited quantitative synthesis of the magnitude of change in vegetable intake. Nevertheless, in the studies that reported serves/day, vegetable intake increased by between 0.29 to 1 serve/day, which is comparable to evidence from mass media campaigns (0.6 serves/day) [63] and workplace interventions (0.32 serves/day) [64]. Reviews of the effectiveness of interventions to increase vegetable intake specifically are lacking. Our exclusion of studies that did not report intakes of fruit and vegetables separately was critical for discerning how interventions impacted on vegetable intake alone. Given the considerable health and economic benefit at the population level of even a small increase in vegetable intake [65], future research should report these outcomes consistently, and separately from fruit intake. Further, some studies in this review reported vegetable intake as a secondary outcome, or as part of an overall diet quality scores, such as the Mediterranean diet [47, 50]. As a result, interventions targeting more than just vegetable intake may have dedicated less resources to increasing vegetable intake per se and may not have been suitably powered to detect effects on vegetable intake. Although the use of different indicators did not help explain any differences in intervention effectiveness, future interventions should report the magnitude of between-group changes in vegetable intake to ensure that results can be included in a quantitative synthesis.

Degrees of personalisation varied considerably between studies, with no clear difference in the type or level of personalisation between effective and ineffective interventions. Moreover, understanding of personalisation methods used in the included studies was limited because the reporting of the design and delivery of personalisation was often minimal. Nonetheless, while many studies used personalised feedback and/or action plans based on demographic characteristics and/or participant preferences, only one study offered participants the ability to customise the timing and delivery of their personalised content [41]. A recent study of the personalisation of digital health information identified that the preferred approach differed by age group, where young adults were more satisfied with user-driven personalisation as distinct from system-driven personalisation [66]. While system-driven personalisation offers the advantage of lower cognitive load for the user, a user-driven approach offers a greater sense of autonomy. As a result, certain population groups, such as those with higher digital health literacy, may wish to exert more control over their personalisation [67]. This degree of autonomy should be considered when designing more sophisticated approaches to personalisation, such as artificial intelligence algorithms and machine learning [68]. Digital technologies are well suited to delivering large-scale personalised dietary support, because the content, frequency and timing of the intervention can be modified to meet the needs and preferences of the user [15]. Thus, future digital interventions for increasing vegetable intake may be improved by better reporting of the use of personalisation methods, ensuring that the tool has sufficient flexibility for the content and modality to be personalised and by considering the use of more sophisticated digital techniques to achieve personalisation.

Embedded behaviour change theories were common in both the effective and ineffective interventions. There was no clear difference in the application of these theories between effective or ineffective interventions. However, it is worth noting that all interventions, bar one [42], mapped to the behaviour change communication domain of policy actions outlined in the NOURISHING framework and did not map to the food environment or food system domains. This contrasts with a recent review of settings-based and digital interventions, where studies often mapped to the food environment domain, by including strategies such as free provision of fruit and vegetables in workplaces [5]. In addition, in the review by Wolfenden et al., all interventions that mapped to the food environment domain were effective at increasing fruit and vegetable intake. The lack of behaviour change strategies at the food environment level identified in our review requires further attention in future research. For example, food prescription programs that aim to improve the accessibility and affordability of healthy foods have shown promise for improving vegetable intake and reducing food insecurity [69], and could be integrated into digital healthcare interventions via partnerships with relevant stakeholders, such as health care providers, food markets or foodbanks. This is particularly important in the era of the COVID-19 pandemic, which has increased consumer acceptance and use of digital health initiatives [70], as well as stimulated a concerted global investment in building more food secure communities [71, 72].

A paucity of studies in this review included diverse populations. Similar to other reviews of digital interventions [73], most study populations were female-skewed, and of mid or older age (> 40 years). Disadvantaged populations, such as those with lower socio-economic position and who are culturally and linguistically diverse, were under-represented. Thus, there is potential for selection bias and response bias to have limited the generalisability of the findings from these studies. In addition, the “digital divide” persists, where lower income countries, racial/ethnic minorities, older adults, and individuals who live in lower income households and rural areas have less access to the internet and lower digital literacy [74]. However, global internet use has doubled from 33 to 65% in the last decade [16], and there is some evidence that digital inclusion is increasing [10, 11, 75]. Therefore, there is an opportunity to test the effectiveness of digital interventions in diverse populations to help reduce dietary (and health) inequities and improve digital literacy. Moreover, findings from this review confirm recent research highlighting a lack of nutrition research in rural settings, where there is inequitable access to healthcare and fresh produce, such as fruit and vegetables [13]. As a result, future interventions should consider external validity in other less well-represented population groups such as individuals with lower socioeconomic position and those living in rural settings. Digital interventions are well suited to achieve this because of their potential for linguistic and cultural localisation, national scalability at relatively low cost, and the global drive to improve digital health equity in rural and disadvantaged communities.

Fewer than half of included studies reported on interventions that had been developed with some level of design input from stakeholders. In addition, intervention end users were very rarely involved and only one intervention specifically mentioned the use of co-design approaches. Recent reviews on the use of co-design have shown mixed findings, with one review of co-design in health settings showing widespread use [24], and another review of co-design in nutrition and health interventions in community-dwelling adults identifying no interventions implementing a complete co-design process [25]. A more recent review of the use of co-design specifically in nutrition interventions delivered within a healthcare, community or academic setting identified only two studies reporting a partnership with consumers across all stages of research [76]. Taken together, these findings reinforce the need for consistent use of co-design terminology, better reporting of design and development processes and more widespread utilisation of a translational framework for the evaluation of health interventions, such as the NASSS (non-adoption, abandonment, scale-up, spread, sustainability) framework [77]. Future research should include co-design methods at multiple levels (i.e., stakeholders with lived experience as well as technical expertise) and include stakeholders throughout, from project conception to dissemination.

Outcomes from this research have implications for the use of digital tools to improve public health nutrition and provide insights into future research needs. Despite the potential for digital tools to improve access to dietary interventions, the persistent threat that digital technologies can exacerbate social inequities of health remains [78]. As such, the inclusion of diverse populations groups in the design and implementation of digital interventions remains a priority. Without this, there is a risk that some population groups may experience barriers to the use of digital technologies, including individuals experiencing socio-economic disadvantage, individuals with disabilities, individuals who require cultural adaptations, and those with low food and digital literacy and self-efficacy [79]. Countries with diverse geographic settings and the potential for disparities in internet access, such as Australia, should ensure that digital interventions are tested in rural settings, which would otherwise be a missed opportunity for addressing widening health disparities [80]. Further, with a paucity of co-design research and consideration of environmental influences, this research suggests that the design of digital interventions to increase vegetable intake is not yet optimal in maximising effectiveness.

This review has several strengths and limitations. The main strength was the systematic approach used to search, screen, and synthesise the literature, including the PROSPERO registration of the review protocol and the use of Cochrane risk of bias tools. By limiting the search to articles published in English and including experimental study designs only, it is possible that studies that would be informative for the design of future interventions were missed. As most studies included in this review were rated as high risk for bias, findings should be interpreted with caution. Due to the heterogenous study populations and intervention designs, including small sample sizes, no quantitative synthesis could be performed. Further, intervention outcomes for vegetable intake will be subject to misreporting biases due to the self-report nature of dietary assessment tools available, which includes the potential for participants to introduce bias as their food literacy and understanding of dietary reporting improves. Lastly, grey literature and commercial products for dietary behaviour change were excluded, which may have limited our ability to capture evidence of co-design research and the full range of digital tools designed to increase vegetable intake.

Conclusions

Few digital interventions have been effective in increasing vegetable intake among adults. Embedding behaviour change theories and involving stakeholders in intervention design may increase the likelihood of effectiveness. Personalisation was not a distinctive feature of effective digital interventions, however, this feature remains poorly understood due to considerable variation in its design and reporting. There is an unmet opportunity for the use of more comprehensive co-design methods to ensure personalisation approaches meet the needs of target populations. Furthermore, future digital interventions should consider strategies that address both behaviour change and food environment influences.