Author's response to reviews Title:Systematic review and meta-analysis of the effect of increased vegetable and fruit consumption on body weight and energy intake Authors: Oliver T Mytton (otm21@medschl.cam.ac.uk) Kelechi Nnoaham (Kelechi.nnoaham@bristol.gov.uk) Helen Eyles (h.eyles@nihi.auckland.ac.nz) Peter Scarborough (Peter.scarborough@dph.ox.ac.uk) Cliona Ni Mhurchu (c.nimhurchu@nihi.auckland.ac.nz) Version:3Date:18 July 2014 Author's response to reviews: see over
Response to reviewers Reviewer One (Charlotte Evans) 1. Please include dates searched in the full paper, not just in the supplementary material. We have added the following sentence to the methods: Only papers published in the last 25 years (i.e. on or after the 1 January 1988) were considered for inclusion. 2. In the sections on study selection and data extraction, studies would also be included if they reported mean difference between control and intervention group together with standard error of mean difference but this isn t stated. This is less common than reporting change between baseline and follow up for each arm but is preferable. This result could be adjusted for baseline covariates which is the gold standard or may not be, depending on the study. This is correct. The methods sentence now reads: Studies were only included if they reported, or such measures could be derived from other reported measures, the difference in body weight or energy intake between the control and intervention adjusted for baseline measures. 3. Related to this, under data extraction it is stated that the main outcome is change between baseline and follow up but the main outcome is the mean difference between control and intervention groups. This can be calculated using t tests and results for change between baseline and follow up which you have mentioned. If change between baseline and follow up is not reported (only the results at baseline and the results at follow up) then the calculation using a t test to obtain standard error needs to be repeated or the results for only follow up are used. Sometimes if only a p value is reported with no other measure of variation the p value is used to work backwards and calculate the standard error so also useful to say whether studies were excluded if they had no measure of variation and only a p value. Please clarify exactly what methods and reported results were used in each case. Thank you this is helpful comment. We have now used this method, where we can. There are four studies that fall into this category: Singh, 1992 for which we do not have the necessary data and have not been able to contact the author, so we have now excluded from the meta-analysis. Basu, 2010b - for which we do not have the necessary data and have not been able to contact the author, so we have now excluded from the meta-analysis (more-over after reviewing this study given that whole fruit was not used, we have decided to exclude completely from the study) Peterson, 2011 used estimate of the mean difference adjusted for baseline and p-value Smith-Warner, 2000 - used estimate of the mean difference adjusted for baseline and p-value 4. Weighting by study is not a common method. If it is included there needs to be a stronger rationale and a reference provided. In my view it would be better not to include and mention the relevant issues under limitations in the discussion. An analysis by sample size cannot rectify the problem as the degree to which standard error is decreased when sample size is doubled is incrementally larger. Given the improvement above we have now dropped this part of the analysis. 5. There is no mention of how many trials are in children and adults in the results. Is this because no trials increased fruit and veg by more than 0.5 portions? It may be worth citing Evans et al AJCN which is a review of school based interventions to increase f and v. The mean increase was 0.3 portions which was
mostly fruit. We did not identify any trials in children. We have now amended the following sentence under description of studies: All studies were conducted in adults, with no trials in children. Having briefly reviewed the Evans et al paper, the studies would have been excluded for the following reasons: a) they are school based either non-randomised or cluster randomised; b) they don t measure all food consumption (i.e. considering only consumption in one particular environment); c) do not have body weight or energy intake as an outcome; or other fail to meet other inclusion criteria However we have added the following comment under the limitations section: No studies were identified in children so one should be cautious about extrapolating the findings to children. While trials have been undertaken in children,[32] our review did not identify any trials in children, in part because the trials in children have been cluster randomised. Children might be better treated in a separate meta-analysis given that they are growing and so naturally experiencing an increase in body mass, for which it might be necessary to relax the criteria around individual-randomisation. 6. Brackets are used very liberally in this paper (although the authors may disagree!). The brackets used in the abstract and conclusion should be removed and commas used instead. Another example is in the methods section where it would be better to have a separate sentence to state that the protocol is provided as supplementary material. There are other examples in the text such as in the discussion/comparison with other studies section where brackets could be removed from or reduced weight gain Whilst some brackets have come out, we are happy with our use of brackets and think it helps keep sentences shorter and improves readability. What would the editor prefer? Reviewer 2 Abstract Population health strategies moved from 5-a-day to More Matters several years ago so the text in Background should be updated accordingly. The reference to 5-A-Day was for the UK and New Zealand, where 5-a-day is still in existence. We understand Norway and Germany also run similar (5-a-day) campaigns. We have changed the abstract to read 5-A-Day or similar campaigns. In the full text we have made reference to 5-A-Day and More Matters In the results within the abstract the authors should include the mean duration of the RCT included as well as state the mean change by group followed by the mean difference in change between groups. We have added the following line to the abstract (as well as including the information in the description of studies: The mean study duration was 14.7 weeks (range four to 52 weeks). The mean difference in vegetable and fruit consumption between arms was 133g. The reference to dose response in the abstract perhaps could be removed as in the text the authors suggest dose could not be adequately studied given limitations of information provided within individual publications. We have removed the sentence that includes the results. different types of interventions should either be defined in abstract or this sentence removed. This sentence has been removed. Conclusion should reflect the SHORT-TERM nature of the trials. This now reads as follows: Promoting increased fruit and vegetable consumption, in the absence of specific advice to decrease consumption of other foods, appears unlikely to lead to weight gain in the short-term and
may have a role in weight maintenance or loss. Longer studies or other methods are needed to understand the long-term effects on weight maintenance and loss. Background Reference #18 concluded no difference in body weight with fruit and vegetable; caution in reporting that 8/12 reported reduced bodyweight. Methods We have amended this sentence to read as follows: Among experimental studies in adults examining the effect of increased fruit or vegetable consumption, the majority (8/12) reported a reduction in body weight, although these positive studies were predominantly conducted in adults with a raised body mass index. Suggest the search terms and databases searched ne included in the text (not as supplementary materials) We have added the terms to the methods under data sources and search strategy. The supplementary material is now redundant and has been removed. Was the criteria increased vegetables and fruit or increased vegetables and/or fruit? Some studies listed seem to have targeted one or other rather than both. And/or. We have amended methods to clarify. Results In methods it states studies of children and adults, in results it states all studies were conducted in adults. Please correct. The search and protocol were not restricted to adults, but we did not identify any studies in children. We have reflected this in the results, by clearly stipulating that no studies in children were identified. Moreover we discuss why this is in the limitations section. Studies ranged in duration from 4 weeks to 3 months physiologically is it probable that weight could change with 50 grams of fruit and/or vegetable in 4 weeks? At minimum the impact of short duration on interpretation should be described as a limitation in discussion. The mean change was 133g (one and half portions) and the mean study duration was 14.7 weeks. It is uncertain how much energy density 133g might displace from the diet (it will likely depend on baseline diet and nature of the 133g of vegetable and fruit) and therefore possible effects on body weight. We discuss the issue of short duration and physiological plausibility in the limitations. We write: The study duration is typically short and the dose of fruit and vegetables relatively modest (mean of just over one and half portions per day). Consequently in some studies it may not have been physiologically likely that changes in body weight would have been observed. This might be particularly likely in individuals of low or normal body weight, in whom weight maintenance (rather than loss) is more likely (and desirable) goal. Consequently despite the absence of a noticeable effect on body weight, one should be cautious about concluding that vegetables and fruit do not have an important role in weight maintenance (or loss). Do you have information on the specific energy intake contributed by the various fruit and vegetable interventions for example a banana would have higher energy density than a kiwi. Again, could include in discussion. We don t have any information on energy content. There might be a variety of differences between fruits and vegetables that could be important (e.g. fibre content, absorbable energy, etc). This theme is reflected within the discussion, under limitations. Within the limitations section we acknowledge a variety of potential sources of heterogeneity including type of vegetables and fruit (e.g. energy density, fibre content, carbohydrate type).
Discussion First paragraph the lack of clarity in terms of energy intake is not surprising as the error in reporting was likely much greater than the kcal content of the fruit and/or vegetable intervention. We agree. This is reflected in the discussion with the following sentence: Dietary surveys might also be a poor means (insufficiently sensitive) to detect the relatively small changes in energy intake sufficient to explain the observed changes in body weight, reflected in the wide confidence intervals. Page 9, paragraph beginning Our findings second to last sentence were should be where Amended. Probably do not need to get into energy absorption it is highly likely energy misreporting drove the lack of associations here. While inaccuracy in energy reporting seems important, it also seems possible that energy intake could go up or stay the same with a negative effect on body weight if the fraction of absorbed energy falls. Overall this is an interesting review and leads to several questions that need to be more fully evaluated in well-designed trials in which the primary objective is to test fruit/vegetable substitution versus fruit and vegetable addition in weight control conclusion should reflect this. Thank you. We have added this sentence: Further research should consider the importance of vegetables and fruit in the long-term; the effects in overweight and normal weight, including their role weight maintenance; as well as considering the effect of directed substitution versus addition of vegetables and/or fruit. Editor's comments: The reviewer have highlight the quality and the interest of the meta analysis. However, there is one major methodological weakness which need to be improve which as been notice by a reviewer as minor essential reviewer. To my point of view it is major. When standard error of the difference were no available the standard error was derived from the paired estimates of standard error. As variability between subject is probably much more important than variability between the beginning and the end of the study this probably a very bad estimate of the standard error of the difference. Weighting on study size despite it is not conventional may potentially correct weighting of the study but not confident interval. Therefore effort must be made to better estimate standard error. For most of the study it may probably be back calculated form p value as suggested by reviewer. If it is impossible effort must be made to contact the authors. and (3) We have followed the editor s recommendation. Please see our response to reviewer one points (2) Minor comment: meta regression should be performed to evaluate moderation of "type a" or "type b" study page 10 We have added included this analysis.