Effectiveness of interventions to promote healthy diet in primary care: systematic review and meta-analysis of randomised controlled trials

The review included reports of randomised or cluster controlled trial study designs. Outcome measures included fruit and vegetable intake (servings/day), fat (% of total energy intake), fibre consumption (gram per day) and change in serum cholesterol level (mg/dl or mmol/l). Interventions included any diet promotion intervention in primary care, including dietary counselling, motivational interviews, advice for behaviour change, computer-delivered dietary information, reminder telephone calls and postal newsletters. Primary care in this context refers to interventions delivered through the first point of contact in a health care system, where the service provider acts as the principal source of advice to patients, rather than through specialist referral. Dietary promotion intervention in this context means any methods which are used to promote healthy diet, including healthy eating advice and counselling, telephone calls, group lectures or use of any other dietary education materials including posters, booklets and guidelines. We excluded multifaceted interventions including those with physical activity promotion along with diet promotion and we did not set any threshold amount of physical activity for exclusion. Target populations included the general population of adults aged 16 years or over, including both men and women. We excluded studies in pregnant women, patients at high risk of, or diagnosed with, cardiovascular diseases, type 2 diabetes, cancer or other chronic conditions, as well as studies in first or second degree relatives of affected individuals. We also excluded studies that included participants at high risk of colorectal cancer (because of adenomatous polyps) [20] or breast cancer (with mammographic abnormalities) [21, 22]. In order to focus on a population approach to primary prevention, we excluded trials which included participants who were pregnant, or with existing chronic conditions, or at high risk of diseases such as colorectal or breast cancer, or with participants who were relatives of family members with chronic health problems linked to diet. Such high risk participants or those with established chronic conditions linked with diet, may be more motivated to make dietary behaviour changes which may apparently show higher effectiveness of interventions promoting healthy diet in primary care. Also, there may be possibilities of diet restrictions which may limit the participation in diet promotion intervention and may apparently show lower effectiveness. Comparators included usual care or no intervention. We excluded those trials comparing one type of diet promotion intervention with another only because our aim was to estimate the effect size difference between a diet promotion intervention and the existing usual care or no intervention; we did not aim to compare any two methods of diet promotion interventions. A minimum follow- up period of 12 months after randomisation was required. Only English language publications were included.

We searched Medline, PsycINFO, EMBASE, Centre for Reviews and Dissemination, and the Cochrane Library, with no restrictions in the date and year, using the combined search terms “dietary intervention AND primary care”, “diet promotion intervention AND primary care”, “diet advice AND primary Care”, “counselling AND diet AND primary care”, “diet promotion AND primary care”, “diet advice AND behaviour change”, “advice in primary care AND behaviour AND diet”, “nutritional counselling AND primary care”, “lifestyle counselling AND cardiovascular risk AND primary care”, “fruit AND vegetable AND primary care”, “nutritional counselling AND general practice”, “dietary intervention AND general practice”, “dietary intervention AND primary care practice”, “nutritional advice AND general practice”, “primary care and diet modification”, “fruit and vegetables consumption AND general practice” “primary care AND fiber consumption” and “fruit and vegetable consumption AND diet intervention”. We also reviewed reference lists of relevant articles and previous systematic reviews. The search was carried out initially in September 2012 and again in March 2013. NB carried out initial screening of title and abstracts against inclusion criteria and retrieved those potentially eligible. NB and MCG independently assessed the retrieved full text articles and any differences were reviewed and agreed. NB extracted data concerning participants, interventions, and outcomes in a tabular form designed for this review. AJW and NB extracted data on the nature of each intervention, including total number of contacts with participants, mode(s) of administration and intervention techniques used [23] and coded the extent to which intervention was based on psychological theories of the determinants of behaviour change, using a published coding scheme [24]. NB and MCG cross checked the extracted data.

NB appraised each study for methodological quality using a standard guidance and checklist [25]. We assessed the methodological quality and risk of bias in terms of randomisation, allocation concealment, blinding, loss to follow up and outcome assessment tool validity.

For each trial, we extracted the intervention effects at 12 months. We estimated the intervention effect as the difference in the change in mean outcome values (follow up value minus baseline value) between the intervention group and control group. If the baseline data were not reported in the trials, we used the difference in the mean outcomes between groups at follow up. If not supplied, we followed standard procedure [26] to derive standard error (SE) for each measure. Where fruit and vegetable consumption was expressed in grams, the conversion factor of one serving = 80 gm was used to express them as servings and where the serum cholesterol was expressed in mg/dl, the conversion factor of 38.6598 was used to express them as mmol/L.

We used random effects meta-analysis to pool the estimates from individual studies. We used the I² statistic to describe the variation in effect size attributable to heterogeneity among studies; higher values suggesting greater heterogeneity. We also constructed funnel plots to assess for the publication bias for the studies included in the review. We used the metan command in STATA version 12 for the analysis.

We screened titles and abstracts of 2,932 papers and identified 49 full text articles. We further identified 3 full text articles from cross-checking references of 49 full text articles. We then identified 10 trials [27‐36] for inclusion in this study after excluding trials not meeting the eligibility criteria. The details are shown in the flow diagram (Figure 1).

We present the study and participant characteristics in Table 1. Ten studies included in the systematic review were published between 1988 and 2006. These studies were conducted in samples representing the primary care general population in Japan [27] (1 study), USA [28, 31, 32, 34‐36] (6 studies), Italy [30] (1 study) and UK [29, 33] (2 studies).The randomised study sample size varied among studies and ranged from 213 to 3,179 participants with 12,414 participants randomised in total.

Participants were men and women, but three studies [28, 31, 36] included women only, with the age ranging from 18 to 79 years. Participants were generally healthy without established chronic diseases. Participants in one study [28] were postmenopausal women consuming at least 36% energy from fat; participants in another study [31] had serum cholesterol values of 200 mg/dl (5.17 mmol/l) or more but as this is close to the population mean value, this study was not excluded as being directed at high risk individuals.

The diet promotion interventions in the trials varied in number of contacts with participants, mode of delivery and behaviour change techniques employed (Table 2).The number of scheduled contacts for intervention with the participants in the intervention groups ranged between one and twenty. Most involved at least one face to face contact, but two [32, 34] involved only a combination of telephone calls and mailed intervention materials. Of the interventions using face-to-face sessions, only one [28] was solely delivered in a group format, while the others used a combination of group and individual contacts. Many of the interventions also involve printed materials. The interventions involved between two and eight intervention techniques. Interventions that involved a greater number of contacts with participants did not necessarily employ a greater number of techniques.

Four [31, 32, 34, 35] of the ten interventions were explicitly described as being based on at least one psychological theory of behaviour change. While none of the four reports explicitly linked all components of the intervention to all the relevant constructs of the theoretical model(s) upon which they claimed to be based, they all explicitly linked at least one intervention technique to at least one determinant of behaviour specified by relevant psychological theory. In three [31, 32, 34] of the interventions, the intervention was tailored for participants according to how they varied on a psychological construct specified by a theory.

In each study, the control group was not enrolled in any intervention, but four had minimal interventions: dietary guidelines [28], standard health education from leaflets [29], a non-personalised conversation without diet counselling [30] and breast self-examination counselling [31] (Table 2).

In each trial, previously validated self-administered food frequency questionnaires, or modified simpler versions, were used to measure the study outcomes. Diet intakes were estimated using the average of the diet consumption in the previous 24 hours to 1 month, collected using food frequency questionnaires.

The funnel plots and Egger’s test for potential publication bias were not informative as insufficient studies were identified for each outcome. Table 3 presents a summary of the methodological quality of included studies. In general, description of randomisation method, blinding of the outcome assessment and allocation concealment were poor in the trials. None of the studies reported what steps were taken to ensure intervention fidelity [37]. Only one study assessed the psychological constructs targeted by the intervention, and briefly reported a mediation analysis to see if the intervention worked by changing the targeted beliefs. Therefore, the ability of the studies to test the psychological mechanisms of the interventions’ effects on dietary intakes was seriously compromised [24].

Our review has several strengths compared with earlier reviews [14‐17]. We estimated the long-term effectiveness of dietary interventions in primary care using only randomised controlled trials or cluster randomised trials with at least 12 months follow-up, where interventions were delivered mostly by health care professionals. In order to focus on a population approach to primary prevention, we did not include trials which included participants with existing chronic conditions or at high risk of diseases such as colorectal or breast cancer and participants with relatives or family members with chronic health problems linked to diet. We did not include trials implemented in the workplace or faith settings where characteristics of participants may be different from those attending the primary health care. We did not restrict the inclusion of trials on the basis of percentage loss to follow up which limited the evidence to trials with more adherent participants and we included two trials [29, 34] in our review which were not included in the Cochrane review. The restriction of studies with more than 20% loss to follow up may introduce bias in the findings limited to trials with more adherent participants. We also explored the existing interventions in terms of behaviour change techniques used.

We acknowledge several limitations. Most of the included trials used self-reported measures of dietary change, and there are possibilities that intervention effects may have suffered responder bias. Furthermore, we cannot rule out the possibility of contamination in open trials which could have resulted in the exchange of information on diet modifications and may have modified outcomes in controls. Dietary interventions cannot be completely blinded; however, outcome assessors can be blinded to avoid the chance of selection bias in trials. Not blinding outcome assessors and trial personnel may have introduced bias in trials thereby overestimating the effectiveness of intervention. We restricted the inclusion of trials investigating the effects of multi-factorial interventions to avoid the potential confounding effect of other health promoting interventions. However, assessing effectiveness of only diet promotion interventions may have overestimated the effects in practice where diet promotion may run simultaneously with other health promotion activities. Where reported, most of the trial participants were white and had some college education, while few reported socio-economic status. We cannot confirm whether our results can be applied to populations with different ethnic, educational and socio-economic characteristics. We do not know whether it is cost-effective to implement diet promotion interventions with similar intensity as reported in the trials in the primary care general population. Participants with unknown previous exposures to health campaigns and media may increase the selection bias. We may have missed some unpublished or published trials.

A Cochrane systematic review [17] reported that diet promotion intervention in health care setting increased fruit and vegetable intake by 1.88 (95% CI1.07 to 2.70) servings per day; and irrespective of setting fruit intake alone increased by 0.67 (95% CI 0.007 to 1.28) servings per day and vegetable intake alone by 0.92 (95% CI0.34 to 1.49) servings per day, fibre intake increased by 6.51 (95% CI2.20 to 10.82) grams per day, in health care setting total dietary fat intake expressed as percentage of total calories fell by 5.38% (95% CI -7.84 to -2.92) and total blood cholesterol level reduced by 0.11 mmol/l (95% CI – 0.19 to -0.03). Effect sizes were generally considerably smaller in our study compared to the Cochrane review. The Cochrane review included studies with participants with chronic conditions, as well as participants at high risk of colorectal cancer and breast cancer. Studies were carried out in faith and work settings as well as in primary care. These differences in participant characteristics, resulting from differences in inclusion/exclusion criteria between our study and Cochrane review may explain these differences in results. Another review [16] published in 1998, reported 5.5% reduction in total blood cholesterol level; expressed in a different measurement unit than our study and we could not compare the changes in blood cholesterol with our results.