Twelve Month Efficacy of Computer-Tailored Communication in Boosting Fruit and Vegetable Consumption Among Adults Aged Forty and over: A Three-Level Meta-Analysis and Systematic Review of Randomized Controlled Trials

Computer-tailored health communication (CTC) can enhance fruit and vegetable (F&V) intake and, consequently, health by providing personalized feedback. However, few studies have examined the long-term effects of such interventions in middle-aged and older adults. This research aimed to assess the 12-mo efficacy of CTC in promoting F&V consumption and potentially identify who among middle-aged and older adults changed their diet after the intervention. The protocol was registered at the International Prospective Register of Systematic Reviews (PROSPERO) on 2021-12-09, code CRD42022330491. The research was performed without external funding. We searched 6 databases (MEDLINE via PubMed, EMBASE, Scopus, Web of Science Core Collection, Cochrane Library, and PsycINFO) for randomized controlled trials (RCTs) comparing CTC interventions for increasing F&V intake with usual care/no intervention control in adults aged ≥40, measured 12 mo after the pretest. The search covered the period from 1 January 1990 to 1 January 2022. We selected 16 RCTs with 25,496 baseline participants for the review systematic literature reviews (SLR) and 11 RCTs with 19 measurements for the meta-analysis (MA). We assessed risk of bias with the JBI Critical Appraisal Checklist. The SLR revealed that at 1-y postCTC intervention, most of the treatment groups increased F&V intake more than the control groups. The overall bias in the data set was not high. The MA model on 11 RCTs revealed a significant effect size for F&V consumption in intervention groups compared with control, standardized mean difference of 0.21 (confidence interval [CI]: 0.12, 0.30), P = 0.0004. The evidence suggests that CTC is a suitable strategy for public interventions aiming to increase F&V intake in adults aged ≥40. The design of CTC for public interventions should consider the process of change and stages of change addressing awareness, attitudes, self-efficacy, and social influence as promising concepts for influencing behavior change.


Introduction
Ensuring adequate consumption of fruits and vegetables (F&V) is crucial for maintaining a healthy diet, as it provides the body with essential nutrients such as vitamins, minerals, dietary fiber, plant sterols, flavonoids, antioxidants, and other beneficial phytochemicals [1].A diet rich in F&V has been shown to have overall positive effects on health, improving every aspect of bodily functioning, from blood pressure to eyesight [1][2][3][4][5][6][7].The evidence suggests that middle-aged and older adults' health suffers when their lifestyle includes a diet with insufficient amounts of F&V [8].Unfortunately, consumption of F&V in many regions of the world is still low [9].
These data indicate that appropriate public health initiatives to increase F&V intake among middle-aged and older adults are needed.Dietary advice has been shown to benefit from personalization [10].Computer-based health information tailoring is a method of assessing individuals (e.g., on sociodemographic, target behavior status, and social-behavioral determinants) and selecting communication content that employs data-driven decision rules that automatically generate personalized feedback from a database of content elements [11].Computer-Tailored Communication (CTC) has shown promise as a method for initiating improvements in people's health behaviors.It might also encourage maintenance of diet change-improvement [12][13][14][15][16][17][18][19].
CTC covers an array of methods that deliver individualized messages to each recipient with the aim of a larger intended communication effect than nontailored messages [12,13].'Tailoring' was first used in the 1990s, and research has shown that it helps messages reach their target more effectively than nontailored [13,20].There are 2 classes of 'computer tailoring' goals: enhancing cognitive preconditions for message processing and enhancing message impact through modifying salient behavioral determinants of goal outcome [13].It uses personalization, feedback, and content matching for message creation [13].In the first stage of tailoring, participants self-report information on their various characteristics.In the second stage, this information is processed by a computer to tailor the message that is then delivered to the participant in the intervention [21].
To find data on the efficacy of CTC in increasing F&V intake in middle-aged and older adults, we performed a preliminary search of meta-analyses (MA) and systematic literature reviews (SLR) in Google Scholar.This search revealed that MA and/or SLR studies have been conducted on CTC, including dietary behaviors, with the most recent published in 2019 [20][21][22][23][24][25].These reviews indicated that CTC is effective in dietary behavioral change in the short term with very small to moderate effect sizes.However, existing reviews have not addressed long-term results of at least 12 mo, and none of them had focused on the somewhat older population targeting adults aged !40, although reviews have included diverse age groups.Thus, this SLR and MA aim to address these research gaps by evaluating the 12-mo efficacy of CTC in increasing F&V intake among adults aged !40, to identify the characteristics of adults who successfully increased their F&V intake after CTC intervention, to examine the measuring instruments used for nutritional intake (e.g., Food Frequency Questionnaire, FFQ), to evaluate the methodological quality of randomized controlled trials (RCTs) conducted on CTC interventions using the JBI Critical Appraisal Checklist, and to provide recommendations for future research.

Methods
We developed the SLR/MA protocol following the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement [26].The protocol was registered prospectively at the International Prospective Register of Systematic Reviews (PROSPERO) under the code CRD42022330491 (9th December 2021).

Eligibility
In the SLR, we included RCTs that tested CTC interventions for increasing F&V intake in adults with a mean age of !40.The studies had to compare CTC interventions with a control condition and measure F&V intake at baseline and 12 mo later (F&V intake assessed at least twice and in the same season to account for seasonal variations).We excluded reviews, case studies, case reports, observational studies, management guidelines, commentaries, or opinion papers.We also excluded studies that involved pregnant females, children, teenagers, or adults < 40 y of age and studies that used CTC interventions for behaviors other than F&V intake.

Search Strategy
From January 1, 1990 (the decade when research on computer tailoring started) to January 1, 2022, we searched 6 databases (MEDLINE via PubMed, EMBASE, Scopus, Web of Science Core Collection, Cochrane Library, and PsycINFO) for RCTs on CTC interventions for increasing F&V intake compared with usual care/no intervention control with an adult study population of mean age !40 y with the F&V intake measured at 12 mo after the pretest.
To ensure that all the relevant articles from various sources were discovered, we checked Google Scholar, search alerts in searched databases, referenced literature, and secondary sources (e.g., citations from already identified studies).
We used the following search terms: MeSH: "diet," "tailored communication," "computer-tailored," "behavior change" OR "behaviour change" OR "diet change" AND Keywords: "vegetable intake" OR "fruit intake" AND "weight loss" OR "BMI" AND The Publication Type: "randomized controlled trial" OR "RCT."

Statements of Significance
To our knowledge, this study is the first to provide a comprehensive and robust synthesis of the long-term efficacy of computer-tailored communication (CTC) interventions for increasing F&V intake in adults aged !40 y.Our findings have important implications for public health policy and practice, as they suggest that CTC is a feasible and efficacious way to promote 12-month sustained improvements in healthy eating habits in this population group, particularly given its relative affordability, minimal risks and ease of implementation.
A. Misir et al.Advances in Nutrition 15 (2024) 100150 We initially downloaded the articles into Mendeley software, where we deleted duplicates.Then, we imported articles into Rayyan [27] to conduct independent and blinded study screening.Initially, we screened articles solely based on their title and abstract.We excluded studies if it was clear from their title and abstract that they were not eligible.Following the first screening round, we independently examined full copies of the articles for eligibility based on the inclusion and exclusion criteria described above.The blinding was removed after the first screening round.The final decision on which studies to include was made by consensus.Only the papers that answered the research question were considered.

Quality Assessment for included studies
Two reviewers (AM and IM) separately assessed the methodological quality and potential for bias of the extracted articles, first by using the JBI Critical Appraisal Checklist [28] for RCTs.Disagreements were resolved through dialog.JBI appraisal was transferred into a generic-abbreviated assessment that was performed with the robvis [29] application.Detailed JBI appraisal forms are available upon request.

Data analysis
We have narratively synthesized findings on the SLR regarding the theoretical base of the interventions, measurements used, or populations included [30].
We have performed the MA on 11 studies and 19 measurement entries with sufficient uniformity in the available outcome data [17,[31][32][33][34][35][36][37][38][39][40].The analysis included study arms that used only CTC interventions and the corresponding control condition.Studies were required to have data on the mean number of servings of F&V or fruits and vegetables separately, as well as their standard deviation (SD) for 12-mo intake measurement.We performed (2 researchers, AM and IM) data extraction consecutively.For data from Kanera et al., 2017 [37], we contacted the authors, and the original data set was used to extract the required missing information.
For the MA, we used standardized mean differences (SMDs) corrected for their positive bias (i.e., Hedges' g values) as the effect sizes due to the use of different instruments for measuring intake across studies (e.g., FFQs with different underlying food databases, FFQs-short versus FFQs long forms).The SMDs were calculated so that positive values indicate a higher mean F&V intake in the group receiving the CTC intervention compared with the control condition.For 10 studies [17,31,32,[34][35][36][38][39][40], SMD values were computed using the 12-mo intake posttest scores, and for 1 study [37] based on the baseline to 12-mo change scores, as using posttest results only would reveal an erroneous significant group difference, given there was a large baseline imbalance in vegetable intake.We followed the Cochran recommendations for MA on SMDs for combining posttest scores with change scores [41] by utilizing posttest SDs rather than change score SDs for standardizing the SMD for this study, which accurately reflected the nonsignificant group difference.For Alexander et al. (2010) [31], there was also a baseline imbalance in the 16-item FFQ in favor of control, but it did not affect the MA result whether posttest scores or change scores were used.In addition, unlike for Kanera et al. (2017) [37], for Alexander (2010) [31], no full data set was available.
Some studies allowed us to calculate multiple effect sizes: 2 studies had reported intake for F&V with more than one measuring instrument [31,34], 3 studies had reported intake for fruits separate from vegetable intake [17,32,37], and 1 had separate intake data for strata of colorectal cancer survivors and the general population with separate control for each stratum [39].For studies that used multiple instruments, we assumed a correlation of rho¼0.7 for the sampling errors of the corresponding SMD values.For studies that reported F&V intake separately, we assumed a correlation of rho¼0.3 for the sampling errors.For the study that was stratified based on colorectal cancer, the sampling errors are uncorrelated due to the use of separate control groups and, hence, no overlap in participants for calculating the multiple SMD values.Based on these assumptions and the calculated sampling variances of the SMD values, we constructed an approximate variance-covariance matrix of the estimates, which was then used, together with the SMD values, as input to a 3-level meta-analysis model [42][43][44].The model included random effects for studies at level 3 (to account for between-study heterogeneity and to allow the true effects for studies providing multiple SMD values to be correlated) and the individual estimates within studies at level 2 (to account for within-study heterogeneity).
We compared the results from the fitted model with those obtained when using cluster-robust inference methods [45].Standardized residuals and Cook's distances were used to identify potential outlying and/or influential studies, which were then subsequently excluded from the analysis as part of a sensitivity analysis.In addition, one study [38] did not use appropriate randomization methods, so a sensitivity analysis for this study was also conducted.To examine the data for evidence of publication bias, we used a funnel plot [46].

Results
The search via the 6 databases yielded 1,311 publications; 30 additional articles were identified through other sources (e.g., search alerts in searched databases and referenced literature in found articles).After we removed duplicate records, we screened the title and abstract of 1128 studies, resulting in the exclusion of 1061 studies.After we applied inclusion and exclusion criteria, we selected 17 studies (16 RCTs, 2 studies from Van Keulen were part of the same Vitalum project [17,19] for the SLR, of which 11 studies with 19 entries we selected for the MA.The list of excluded studies we recorded together with the reason(s) for exclusion (Supplemental Table 1).

Search results
The PRISMA flow diagram (Figure 1) displays the overall search results [26].

Study intervention characteristics
Table 1 summarizes the characteristics of SLR studies' interventions.These studies were published over a 21-y period (the oldest study was published in 2000 [36], and the 2 most recent ones were from 2017 [37] and 2021 [19], with the 2021 publication being related to older research).

Number of observed behaviors and interventions
Of the 16 RCTs, 2 aimed to impact only 'one' behavior-F&V intake [31,60].Fourteen RCTs examined multiple behaviors, including F&V intake (Table 1).Twelve of the RCTs were part of a larger project on multiple behaviors.Four RCTs had more than 1 intervention (other interventions besides CTC, e.g., motivational interviewing) and combined interventions (Table 1).Three studies had more than 1 intervention but without combining interventions (Table 1).
Six RCTs had more than 2 arms, and the number of measurements ranged from 2 with a baseline to 4 with a baseline (Table 1).

Instruments used for measuring nutritional intake
The most used instrument for measuring F&V intake was the FFQ (Table 1).FFQs provide information on the consumption of queried foods and beverages over the specified period.FFQs may assess total dietary intake as well as specific dietary aspects.The specific formats used are shown in Table 1 and they range from short screeners targeting only F&V to longer FFQs.Sometimes, they were combined with a targeted question on intake of F&V ("How many servings per day?" or "How many days a week do you eat at least 200g vegetables/2 pieces of fruit?").FFQs were validated in different countries, and studies used their countryspecific food tables for intake calculations.One study [33] used diet history, and one study [61] measured achieving/not achieving recommended intake with a self-administered questionnaire on health behaviors that encompassed questions on intake of F&V per day (at least 5 servings of vegetables and 2 servings of fruit per day was considered as achieving recommended intake).
The recall guideline for the length of FFQ in Dutch studies was a typical week during the past mo, based on references and additional materials (e.g., Dutch PhD database: https://www.narcis.nl/).The past month was also the most common recall length in the SLR data set (Table 1).Walker et al. (2009) [38] did not specify the recall details, such as a typical week, but required a recall of 6 mo from the last measurement baseline.

Characteristics of meta-analyzed studies
All meta-analyzed studies had a 1-y follow-up.In some cases, there was a slightly different time frame for the 1-y follow-up,  1 mean calculated from the available data in the study, CRC-colorectal cancer, FH-familiar hypercholesterolemia, ⊆ MA-included in meta-analysis e.g., Van Keulen, 2011 [17], who started the measurement of intake in week 47 (around 11 mo after baseline).From the studies included in the meta-analysis, only 1 study [34] also reported on a 24-mo follow-up.

Behavior change approaches
Theoretical models and concepts used for tailoring have been reported in all SLR studies (Table 3).Robroek et al., 2012 [62] only provided information on the measured social-cognitive variables (concepts) without specifying the model.

Intervention effects and sustained outcomes
SRL on 16 RCTs found that after 1 y, the treatment groups in most of these studies had a greater intake of fruits and/or vegetables compared with the control groups, though the degree of improvement varied (Table 4).

Meta-Analysis
We performed a 3-level MA to assess the 12-mo efficacy of CTC when it comes to increasing F&V intake in adults aged !40.The pooled SMD based on the 3-level MA model was SMD ¼ 0.21 (CI: 0.12-0.30),P ¼ 0.0004.The estimated variance components were τ 2 Level 3 ¼ 0.0088 for the between-study heterogeneity and τ 2 Level 2 ¼ 0.0021 for the within-study heterogeneity.This resulted in I 2 Level 3 ¼ 49.09% of the total variation, which can be attributed to between-study heterogeneity, and I 2 Level 2 ¼ 11.88%, which can be attributed to within-study heterogeneity.
Using cluster-robust inference methods did not yield noteworthy differences in results compared to the fitted model.Relative to the rest of the studies, Cook's distance was relatively large for Walker et al. (2009) [38], but a sensitivity analysis excluding this study did not yield any relevant differences in  terms of the pooled effect, confidence interval, or amount of heterogeneity.The MA is summarized in the forest plot (Figure 2), and the heterogeneity split is in Figure 3.The funnel plot (Figure 4) did not show any apparent evidence for publication bias.

Risk of bias
The main study characteristics that we analyzed in assessing study bias were the randomization procedure and its success; blinding of participants, personnel, and outcome assessment; The Netherlands, 45% female, mean age¼57.2ycontrol, CTC letters, motivational calls, previous 2 interventions combined FFQ -16 items and question about F&V intake FFQ-16 items showed that all 3 intervention groups were equally and significantly more effective (with some differences in favor of CTC) than the control group in increasing intake of fruit (svg./d) and of vegetables (g/d) from baseline.Effect sizes (Cohen's d) ranged from 0.15 to 0.18.FFQ-16 items showed that CTC group was more likely to adhere to F&V consumption guideline than control or combined group (CTC vs. control P < 0.001, average for 3 time points).⊆ Walker et al, 2009 [38] US, 100% female, mean age¼57.8ycontrol, CTC intervention FFQ FFQ showed that the CTC group significantly increased F&V svg.(þ0,92 svg.) from baseline to 12 mo, unlike the control that had after initial improvement at 6 mo dropped to baseline at 12-mo measurement.The intervention vs. control difference in the final measurement was even greater þ1.25 svg.
CTC, computer-tailored communication, CI, confidence interval, FFQ, food frequency questionnaire, OR , odds ratio, PTC, "Pathways to Change" intervention, svg, servings, ⊆ MA-included in meta-analysis, 24H, 24 hour dietary recall self-reporting; and other sources of bias (attrition, data analysis).Figure 5 shows a summary of study quality components used for assessing study bias (individual study bias is in Supplemental Figure 1).
Attrition rates were reported in all studies and varied from 5% [32] to 60.6% [64] at 12 mo from baseline with a mean of 23.6% (median ¼ 23.4%).Differential drop-out rates with very low and low overall attrition from > 6% to 17.5% were reported by Demark et al. ( 2007) [33] and Kanera et al. (2017), [37] respectively.Heimendinger et al. (2005) [60] reported that the systematic loss to follow-up did not affect the composition of the experimental conditions at 12 mo with no significant differences for any of the variables or the baseline estimate for F&V consumption.Van Keulen et al. (2011) [17] report differential drop-out between groups (higher in intervention group) and education levels (higher among lower educational participants), whereas Robroek et al. (2012) [62] report more drop-out in the intervention group in the first follow-up, but they accounted for that in the analysis.
Four studies [31,35,36,39] used only a per protocol (PP) analysis.For one study [60], it is unclear whether they used a PP or intention to treat (ITT) analysis.The treatment of missing data in ITT was mostly properly reported and performed.
Measuring diet was based on self-reported tools.In 2 studies, participants were compensated for their participation [31,33], and 1 study paid churches for participating with their participants [35].

Discussion
This SRL on 16 RCTs and MA on 11 RCTs shows that an improvement in CTC in F&V intake may last for at least 1 y for middle-aged and older adults.Furthermore, in more than half of the meta-analyzed studies, the CTC treatment group outperformed the control.
Out of 7 studies that had either all female [38] or predominantly female (over 60%) population [31,34,35,37,60,61], 6 [31,34,35,38,60,61] successfully sustained improvement in intake at 12 mo (SLR data set) (Alexander et al.,2010 [31] only for 2 item measure).This could indicate that middle-aged and older females are a successful group in sustaining their F&V intake postintervention, which corresponds to previous research on dietary behaviors finding that females are more inclined to be motivated to higher intakes of F&V than males [67,68].Additionally, the limited number of studies [32,33,37,39,63] that involved participants with different underlying diseases did not allow for a clear assessment of how lifestyle behavior interventions might affect these populations.
CTC interventions in this SLR varied, and each research had some unique characteristics (Tables 1-3) in terms of, for instance, the mode of communication (from letters to computer screens) or the country of study (United States, Netherlands, Belgium, Canada, and Australia), or in health status of participants, but the theoretical framework for the CTC method was very similar among studies and should serve as the primary guide for future research.This SRL showed that TTM, I-Change, and SCT were used most often in the past 21 y in theory-based CTC for diet change.As I-Change encompasses elements of both TTM and SCT, it can be concluded that the fundamental theoretical basis for all analyzed studies was comparable.Most analyzed studies used a combination of process of change and stages of change addressing attitudes, self-efficacy (or behavioral control), and social influence (support, pressure, and modeling) as advised by Noar et al. (2007) [20], who found these concepts to result in a larger behavioral impact.
Whereas increasing F&V intake is beneficial for health, it is important to reach the recommended intake levels for F&V.Only 4 studies [19,40,61,62] have reported intervention results for F&V intake related to increasing adherence to nutritional recommendations that were in effect in the country of research at the time of the study.Of these, 3 [19,61,62] showed statistically significant improvement in the intervention group in reaching the recommended intake levels at 12 mo follow-up.Researchers and policymakers are encouraged to monitor adherence to these guidelines in all upcoming research studies and prioritize the implementation of cost-effective interventions that promote adherence to these guidelines.
Although the cost of CTC interventions was not the primary focus of this SLR, it is a crucial consideration for interventions aimed at changing behavior.In our data set, few studies mention the cost-effectiveness [36,39,60,62] or have a separate, related study on this topic [69,70].All except Robroek et al. (2012) [62] find CTC cost-effective and recommendable for use.Given the practical significance of cost-effectiveness for applying interventions in the real world, it is advisable that this topic be covered in any upcoming CTC research.

Meta-analysis
MA on 11 studies resulted in a standardized effect size of 0.21 (P ¼ 0.0004) that, according to Cohen [71], corresponds to a small intervention effect.However, determining whether an effect is small, medium, or large should be based on the findings of previous studies in the relevant field.A 1-y postintervention effect size of 0.21 corresponds to an effect size that is commonly found in psychology (behavioral research), which is the field to which CTC interventions belong.This effect size indicates that the implementation of these interventions can have a significant impact on public health when widely adopted [64,72,73].Its clinical relevance becomes apparent when considering certain critical factors: affordability, minimal associated risks, and broad implementability, all characteristics that align with CTC interventions [72,73].What demands even greater emphasis, however, is the revealed enduring impact of CTC interventions.A small but enduring effect size may possess more profound clinical significance than a larger effect that merely produces short-lived results.
The result from this MA corresponds to or is slightly better than the results of MAs on diverse CTC interventions (exercise, smoking, alcoholism, cancer screening) in various age groups [20,72,73].This research focused on middle-aged and older individuals and found an overall slightly better MA result than previous research performed on more heterogeneous age groups [20,72,73], which could also indicate that middle-aged and older adults in CTC intervention are more likely to increase their F&V intake and consolidate this change.

Potential for bias
Out of the 16 RCTs in this review, 12 used individual randomization [17,31,33,34,36,37,39,40,60,61,63,64], 3 [32,35,62] used cluster randomization, and 1 [38] used quasi-cluster randomization.The cluster-randomized trials adjusted their analysis accordingly.Walker et al. (2009) [38] used quasi-cluster randomization, so the study might suffer from confounding and selection bias [74].Nevertheless, the choice of methodology can be justified by the type of research that was performed and their goals-researching diet change in hard-to-reach, older, rural females in the Midwestern United States and controling for spillover between intervention and control that can happen within a village.Thus, they used 2 demographically similar villages randomly sampled for intervention and control to get as much as possible a "representative" sample for the population they set to investigate.In addition, this study was identified as a potential outlier in MA dataset, but a sensitivity analysis did not yield any noteworthy differences in the results.
In the study by Alexander et al. (2010) [31], the tailored behavioral intervention group had lower F&V intake at baseline than the control group, measured by a 16-item FFQ.This could have biased the results in favor of the control group.Kanera et al. (2017) [37] had the opposite situation from Alexander et al. (2010) [31] and in Robroek et al. (2012) [62] the intervention group had more participants who ate enough fruit.All 3 authors reported adjustment for baseline differences.
Blinding participants was not always possible in some of the interventions due to the nature of the study design, which is not uncommon in nutritional interventions [75].When participants cannot be blinded, blinding care providers and assessors is important [76].In this SLR/MA, the CTC interventions had no intervention providers, and assessments were mostly filled out at home without assessors (only self-reporting).
All studies used self-reporting tools for measuring F&V intake, with FFQs used most often.These FFQs were designed to measure intake for the typical week over the course of the past month, which is a good balance between FFQ's aim to rely on a longer recall period (from 1 wk to as long as 1 y) to capture foods that are not consumed every day and recall bias that may increase with longer periods of recall [77,78].Information on psychometric qualities was often limited to claims that an instrument had been validated or was otherwise reliable.
Many of the studies utilized ITT analysis, and all these studies, except for Jones et al. (2003) [63], properly reported and addressed missing data.Attrition was reported in all studies, with the mean attrition rate being 23.6%.Although drop-out > 20% may affect validity, it should be considered that in this SLR/MA studies had a 12-mo follow-up, mostly used ITT to reduce attrition bias, and did power calculations to account for drop-out and maintain power.However, Schultz et al. (2014) [64] had an attrition rate of > 50% at 12 mo, but this is to be expected for a web-based intervention [79].In this SLR, 5 studies [17,33,37,60,62] indicated differential drop-out between control and intervention conditions, but they used ITT and mostly had an acceptable overall drop-out rate (~20%).In addition, Crutzen et al. (2013) [80] claim that there is an indication that slightly higher attrition rates are often seen in intervention groups compared with control, which attenuates the observed effect.
Overall bias in the presented SLR data set was not high, despite that nonblinding of participants produced a high bias score, but this is common in nonpharmacological interventions [81].Most studies were funded by national foundations and governmental agencies, which is expected to reduce risk of conflict of interest that can influence the design of the study, the interpretation of the results, and the publication of the findings.
Analysis of bias found gaps in the reported methodological details of reviewed studies, which could be avoided by adhering to CONSORT and TREND guidelines for randomized and nonrandomized studies, respectively.In this analysis, it has been acknowledged that the absence of information on a certain procedure in a published study does not necessarily imply that the procedure was not done [82][83][84].

Conclusion
This review shows that CTC can help middle-aged and older adults sustain their increased F&V intake 1 y after the intervention.Therefore, CTC is a suitable strategy for public interventions that aim to increase F&V intake in adults aged !40.The design of CTC for public interventions should consider the process of change and stages of change addressing awareness (e.g., discrepancy between current and healthy behavior), attitudes, self-efficacy (or behavioral control), and social influence (support, pressure, and modeling) as promising concepts for influencing behavior change.
To improve the quality of future research on CTC intervention on F&V intake, it is recommended to report F&V intake data separately, as they are distinct behaviors that can exhibit different responses to CTC intervention.Furthermore, reporting on longer term effects (! 12 mo), reaching current recommended guidelines, and tracking and reporting implementation costs would also be advisable.In addition, it would be good to strictly follow CONSORT and TREND recommendations for reporting on randomized and nonrandomized studies.

FIGURE 2 .
FIGURE 2. The three-level meta-analysis forest plot with the explanation table.

FIGURE 3 .
FIGURE 3. Graphic representation of the variance and heterogeneity.

FIGURE 5 .
FIGURE 5. Overall, bias in the SLR data set entailing 16 studies.

Table 1
Study Intervention Characteristics SRL & Meta-Analysis (continued on next page)

Table 4
Key Findings of Analyzed Studies