Dietary sugars and body weight: systematic review and meta-analyses of randomised controlled trials and cohort studiesBMJ 2013; 346 doi: https://doi.org/10.1136/bmj.e7492 (Published 15 January 2013) Cite this as: BMJ 2013;346:e7492
- 1Departments of Human Nutrition and Medicine, University of Otago, PO Box 56, Dunedin 9054, New Zealand
- 2Riddet Institute, University of Otago
- 3Edgar National Centre for Diabetes and Obesity Research, University of Otago
- Correspondence to: J Mann
- Accepted 28 October 2012
Objective To summarise evidence on the association between intake of dietary sugars and body weight in adults and children.
Design Systematic review and meta-analysis of randomised controlled trials and prospective cohort studies.
Data sources OVID Medline, Embase, PubMed, Cumulative Index to Nursing and Allied Health Literature, Scopus, and Web of Science (up to December 2011).
Review methods Eligible studies reported the intake of total sugars, intake of a component of total sugars, or intake of sugar containing foods or beverages; and at least one measure of body fatness. Minimum duration was two weeks for trials and one year for cohort studies. Trials of weight loss or confounded by additional medical or lifestyle interventions were excluded. Study selection, assessment, validity, data extraction, and analysis were undertaken as specified by the Cochrane Collaboration and the GRADE working group. For trials, we pooled data for weight change using inverse variance models with random effects. We pooled cohort study data where possible to estimate effect sizes, expressed as odds ratios for risk of obesity or β coefficients for change in adiposity per unit of intake.
Results 30 of 7895 trials and 38 of 9445 cohort studies were eligible. In trials of adults with ad libitum diets (that is, with no strict control of food intake), reduced intake of dietary sugars was associated with a decrease in body weight (0.80 kg, 95% confidence interval 0.39 to 1.21; P<0.001); increased sugars intake was associated with a comparable weight increase (0.75 kg, 0.30 to 1.19; P=0.001). Isoenergetic exchange of dietary sugars with other carbohydrates showed no change in body weight (0.04 kg, −0.04 to 0.13). Trials in children, which involved recommendations to reduce intake of sugar sweetened foods and beverages, had low participant compliance to dietary advice; these trials showed no overall change in body weight. However, in relation to intakes of sugar sweetened beverages after one year follow-up in prospective studies, the odds ratio for being overweight or obese increased was 1.55 (1.32 to 1.82) among groups with the highest intake compared with those with the lowest intake. Despite significant heterogeneity in one meta-analysis and potential bias in some trials, sensitivity analyses showed that the trends were consistent and associations remained after these studies were excluded.
Conclusions Among free living people involving ad libitum diets, intake of free sugars or sugar sweetened beverages is a determinant of body weight. The change in body fatness that occurs with modifying intakes seems to be mediated via changes in energy intakes, since isoenergetic exchange of sugars with other carbohydrates was not associated with weight change.
Sugar has been a component of human diets since ancient times, with earliest reports of consumption coming from China and India, and much later from Europe after the Crusades in the 11th century.1 The suggestion that sugar might have adverse health effects has been a recurring theme for decades, with claims that high intake may be associated with an increased risk of conditions as diverse as dental caries, obesity, cardiovascular disease, diabetes, gout, fatty liver disease, some cancers, and hyperactivity.2 3 4 5 6 However, inadequate study design, differences in assessing dietary intake, inconsistent findings, and varying definitions of “sugars” have precluded definitive conclusions regarding these associations.
The most consistent association has been between a high intake of sugar sweetened beverages and the development of obesity,7 8 9 10 11 12 but not all published meta-analyses have reported a statistically significant link.7 11 The expert consultations organised by the World Health Organization and the Food and Agriculture Organization of the United Nations and the scientific updates undertaken by WHO13 14 15 have adopted a classification of carbohydrates and clarified definitions of various groups of sugars including the category of “free sugars” (table 1⇓). This classification enables a more standardised approach to examining potential adverse health effects.
To update the recommendations through the guideline’s development process that was launched in January 2009, WHO commissioned a systematic literature review to answer a series of questions16 relating to the effects of sugars on excess adiposity. These questions asked whether reducing or increasing intake of dietary sugars influences measures of body fatness in adults and children, and whether the existing evidence provides support for the recommendation to reduce intake of free sugars to less than 10% total energy (box).15 Body fatness was selected as an outcome in view of the extent to which comorbidities of obesity contribute to the global burden of non-communicable disease.
Questions posed by the WHO Nutrition Guidance Expert Advisory Group-Subgroup on Diet and Health, to develop recommendations regarding sugars intakes
What is the effect of a reduction in free sugars intakes in adults?
What is the effect of an increase in free sugars intakes in adults?
What is the effect of a reduction in free sugars intakes in children?
What is the effect of an increase in free sugars intakes in children?
(Where “free sugars” are defined as all monosaccharides and disaccharides added to foods by the manufacturer, cook, or consumer; plus sugars naturally present in honey, syrups, and fruit juices.)
Since the answers to the questions posed (box) were designed to inform population based dietary guidelines rather than advise individual patients, it was deemed appropriate to include cohort studies and randomised controlled trials of free living people consuming ad libitum diets (that is, with no strict control of food intake). The interventions mainly involved advice to increase or decrease intake of sugars, or of sugar containing foods or beverages, without emphasising the need to achieve weight loss.
We also examined randomised controlled trials comparing higher and lower intakes of sugars, but where energy intake was strictly controlled. Trials specifically designed to achieve weight loss were excluded. We acknowledged that the studies identified by this approach would inevitably be heterogeneous, that it would be difficult to disentangle the effects of a range of dietary changes that might occur after altering the intake of sugars, and that it might be difficult to identify a dose response. However, the findings from such an approach were expected to provide an indication of what might be achieved by population changes in intake of dietary sugars.
In accordance with the WHO guideline’s development process,17 systematic reviews and meta-analyses were conducted according to the methods of the Cochrane Collaboration.18 We prepared tables summarising quality assessment, effect size, and importance of findings, from which recommendations may be derived, in the format required by the Grading of Recommendations Assessment, Development and Evaluation (GRADE) working group. Ethical approval was not required for this research.
Two separate electronic searches were conducted to identify randomised trials and prospective cohort studies relating intake of dietary sugars to measures or changes of body fatness (web appendix 1). OVID Medline, Embase, PubMed, Cumulative Index to Nursing and Allied Health Literature, Scopus, and Web of Science electronic databases were searched for clinical trials and cohort studies, published up to December 2011, which met the inclusion criteria. In OVID Medline, we used the highly sensitive Cochrane search strategy to limit the first search to clinical trials, meta-analyses, and randomised controlled trials. We hand searched meta-analyses and reviews to identify studies that might have been missed.
Two reviewers assessed titles and abstracts of all identified English language studies. Discrepancies in opinion as to whether studies should be selected for full review were resolved by discussion. A similar approach was used to determine which of these studies should be included in the formal analysis. Animal studies, cross sectional studies, and case-control studies were excluded. Studies were required to report intake of total sugars, intake of a component of total sugars (expressed in absolute amounts or as a percentage of total energy), or intake of sugar containing foods or beverages, assessed by continuous or categorical variables; and at least one measure of body fatness.
Participants were adults and children free from acute illness, but those with diabetes or other non-communicable diseases in whom conditions were regarded as stable could be included. Randomised trials were required to be of at least two weeks’ duration, and prospective cohort studies were required to be of at least one year’s duration. We included trials comparing diets differing in sugars intakes and in which the effect of sugars could be separated from the effects of other lifestyle or medical interventions.
Two groups of trials were identified. One group included studies in which participants in the intervention arm were advised to decrease or increase sugars, or foods and drinks containing sugars. Although such advice was generally accompanied by the recommendation to increase or decrease other forms of carbohydrate, there was no strict attempt at weight control. These trials are referred to as ad libitum studies. The other group of trials attempted to achieve isoenergetic replacement of sugars with other forms of carbohydrate. Interventions designed to achieve weight loss were excluded because the ultimate aim of the review was to facilitate the development of population based recommendations rather than nutritional recommendations for the management of obesity.
Data extraction and quality assessment
Data extraction and validity assessment were carried out independently by two reviewers, and any discrepancies resolved by discussion. For both randomised trials and cohort studies, outcomes, data relating to participants, exposure or interventions, potential effect modifiers, and study quality were extracted by use of piloted data extraction forms. In the cohort studies, we aimed to extract the least and most adjusted relative risk, odds ratio, or mean difference when comparing the most exposed group of participants with the least exposed group, or a β coefficient for the continuous effect of a one unit change in sugars intake. We extracted these statistics separately for sugars exposures reported as baseline values or as values for change over time.
Cochrane criteria18 were used to examine validity of each randomised trial, including sequence generation, allocation concealment, blinding of participants, personnel and outcome assessors, incomplete outcome data, and selective outcome reporting. Additional review specific criteria included similarity, or not, of type and intensity of intervention in both arms, and whether the studies were funded by industries with potentially vested interests. We examined the effect of bias on the pooled effect estimates by excluding studies that had a high risk of bias for two or more validity criteria in sensitivity analyses.
Studies were grouped to answer the major questions that had been posed (box). We considered data for adults and children separately. Studies of isoenergetic exchange of sugars with other carbohydrates were examined to help explain possible mechanisms through which sugars might exert their effects.
The effects of decreasing or increasing dietary sugars in adults were examined principally by meta-analysing the randomised trials in which participants were required to consume different amounts of sugar (sucrose) or other sugars (which would now be classified as “free sugars”). Terminology varied among trials. The term “free sugars” refers to all monosaccharides and disaccharides added to foods by the manufacturer, cook, or consumer, plus sugars naturally present in honey, syrups, and fruit juices (table 1).14 The term “added sugar” is sometimes used interchangeably with “free sugar” but is considered to include sugars and syrups added to foods during processing, food preparation, or at the table—but does not include honey, syrups, or fruit juice.19 “Sugar” is generally assumed to be purified sucrose.14
Data for each group of studies were pooled using Review Manager 5.1 software.20 In trials involving adult participants, we used generic inverse variance of analysis for mean differences in body weight between intervention and control groups to compare the parallel and crossover experimental designs reporting change in body weight. In the studies involving children and adolescents, we used standardised mean differences because studies reported differences in either body mass index (BMI) or standardised BMI units.
Heterogeneity was assessed with the I2 test and Q statistics. We considered an I2 value greater than 50% and P<0.05 as indicative of heterogeneity.18 We used random effects models because significant heterogeneity was associated with trial design and duration in some analyses.
Estimates for the standard error of the difference in means for treatment groups in crossover studies were derived from reported P values when the standard error of the mean difference was not reported.18 If P values for the differences were reported simply as non-significant, then P=0.2 was assumed.18
We did sensitivity analyses to explore the differences between studies in the short term (<eight weeks) and longer term (>eight weeks). We also tested the effects of removing those studies that achieved a difference in sugars intakes of less than 5% of total energy intake between intervention and control groups. Metaregression (using Stata/IC 11.2 software for Mac (StataCorp)) was used to test for a dose-response effect of sugars on weight change, and for associations between weight change and study duration, study design (that is, crossover or parallel), and whether sugars intake changed in the intervention arm.
Publication bias among the randomised controlled trials of adults was examined by visual inspection of a funnel plot and Egger’s test for bias.21 Publication bias is suspected when the funnel plot is asymmetrical. We combined the 15 ad libitum studies for this analysis because it is generally accepted that asymmetry cannot readily be assessed with 10 or fewer studies.18 Sensitivity analyses examined the influence of small study effects, by comparing the estimates derived from random and fixed effects models22 and by using the Duval and Tweedie23 “trim and fill” method in Stata 12 (Metatrim). There were insufficient studies in children to conduct a meaningful examination of publication bias.
Prospective cohort studies
Cohort studies in adults provided limited additional information. Data from cohort studies in children were necessary to determine the effect of increasing sugars intake on adiposity, owing to a lack of suitable randomised trials. We grouped individual studies for meta-analysis on the basis of the methods used for reporting adiposity outcomes and sugars exposure variables.
We used four main methods of reporting outcomes:
β coefficients for the continuous association between sugars exposure at baseline and adiposity outcome.
Odds ratios for the risk of overweight or obesity comparing participants who had the highest intakes of sugars with those who had the lowest intakes of sugars (groups or frequency of servings).
Mean differences in change in measures of adiposity over time between participants with the highest intakes of sugars and those with the lowest intakes (groups or frequency of servings).
β coefficients for the continuous association between increases in sugars exposure over time and adiposity outcome.
Sugars exposures included sugar sweetened beverages, fruit juice, sweets (including jams, syrups, cakes, and desserts), sucrose, or total sugars. Exposures were reported as servings per time period and were converted to servings per day, volume of beverage consumed per day, percentage of energy intake, frequency of consumption, or grams per day. Where possible, we scaled exposures to comparable units to allow data to be pooled. We assumed that one serving of sugar sweetened beverage was equivalent to 240 mL or 8 fluid ounces, and contained 26 g of sucrose.24 This portion equated to about 5% of daily total energy intake in adults.
Measures of body fatness included weight change, change in BMI or BMI z score, waist circumference, body fat (%), fat mass, and trunk fat (%). If studies reported more than one measure of sugars intake, we derived an average effect size. We ranked adiposity outcomes in terms of importance for pooling, from highest to lowest: BMI z score, BMI, body weight, waist circumference, percentage body fat, fat mass, and percentage trunk fat. If studies reported outcomes for more than one measure of adiposity, we used the highest ranked adiposity outcome. We generated pooled estimates for the various subgroups using metan commands with random effects in Stata. Two sided P<0.05 was considered significant for all analyses.
GRADE assessment25 was carried out to assess the totality of the evidence by the authors and then refined by the WHO Nutrition Guidance Expert Advisory Group (NUGAG) Subgroup on Diet and Health (www.who.int/nutrition/topics/advisory_group/en/index.html) to fulfil the required process for developing WHO guidelines.17 GRADE assessment took into account study design limitations, consistency of results across the available studies, precision of the results, directness, and likelihood of publication bias when assessing the quality of the evidence from the randomised trials.17 25 Further criteria were considered for the cohort studies. These criteria included magnitude of the effect, evidence of a dose-response gradient, and the direction of plausible biases. The quality of the evidence was categorised as high, moderate, low, or very low. Web appendix 2 shows the relevant GRADE tables.
Figures 1⇓ and 2⇓ show the process by which the included studies were identified. We identified 7895 potential randomised trials from the electronic search and a further 10 studies through hand searches of relevant review articles and on recommendation from NUGAG panel members. Removing duplicates left 6634 articles, of which 6557 were assessed to be irrelevant. Abstracts and full text articles for the remaining 77 studies were judged as requiring full review and were reviewed by three independent reviewers. Of these remaining studies, 19 met the inclusion criteria for ad libitum studies 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 and 11 were identified for the comparative analysis of isoenergetic studies.48 49 50 51 52 53 54 55 56 57 58 For cohort studies, we identified 9445 potential studies from the electronic search and an additional 10 studies through hand searches of relevant review articles. Of 69 studies selected for full review, 38 were considered to meet the inclusion criteria.59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 The 47 excluded randomised trials and 31 excluded cohort studies are described in web appendices 3 and 4.
Assessment of study quality
Risk of bias varied among the randomised trials (web figs 1 and 2, web appendix 5). Failure to conceal treatment allocation (almost impossible to achieve in dietary trials involving free living participants) was the major potential source of bias (performance bias). In many trials, it was unclear as to whether outcome measures had been assessed by observers unaware of treatment allocation (detection bias) and whether there had been selection bias. Three trials, in which there was evidence of differences between dropouts and completers, reported data only for those who completed the intervention.28 34 39
Our analysis included 38 prospective studies lasting at least 12 months, and in which data relating to an association between sugars and a measure of adiposity could be extracted; none was excluded on the basis of study quality. Of these 38 studies, 15 used self reported estimates of adiposity outcomes59 64 65 66 67 68 70 71 73 74 75 76 77 78 79 80; seven collected exposure data from questionnaires where the validity for assessing sugars intake was not stated or not assessed60 61 67 79 81 82; 19 involved convenience sampling59 61 62 67 71 73 78 83 84 85 86 87 88 89 90 91 92 93; and 18 provided estimates that were adjusted for total energy intake.59 60 64 66 69 72 75 76 86 88 90 91 92 93 94 95 96 97 There was a lack of consistency in the covariates used to adjust analyses and a wide range of methods of assessing sugars exposures and adiposity outcomes, which made pooling studies difficult.
Effect of reducing dietary sugars on measures of body fatness in adults
Table 2⇓ describes the five studies identified for this analysis,28 30 31 33 41 49 and figure 3⇓ shows the quantitative meta-analysis (forest plot). Reduction in dietary sugars intake was associated with significantly reduced weight (−0.80 kg (95% confidence interval −1.21 to −0.39); P<0.001) at the end of the intervention period by comparison with no reduction or an increase in sugars intake. The trials all involved a reduction in intake of sugars (classified as free sugars) in the intervention arm compared with the control arm.28 31 33 39 41 Study durations ranged from 10 weeks to eight months. In four studies, participants were advised to limit sugar containing foods,31 33 39 41 and in one study, participants were asked to substitute usual sugar rich foods with low sugar alternatives.28 Three of the five trials reported data for completers only.28 39 41 However, only two of these studies considered this to be a potential source of bias.28 41 Exclusion of these two studies from the meta-analysis slightly attenuated the effect, although the effect estimate remained significant (−0.81 kg, −1.41 to −0.21). After excluding three studies28 39 41 that had a high risk of bias for two or more validity criteria, the effect estimate was no longer significant although the difference in weight was similar (−0.81 kg, −1.69 to 0.07).
Differences in sugar intakes between intervention and control groups ranged from less than 1%33 to 14% of total energy intake.39 Two studies achieved a difference in reported sugars intake of less than 5% of total energy intake at the end of the intervention.28 33 Paineau and colleagues33 reported a difference in sugars intake between groups of 2.2 g/day, and Gatenby and colleagues28 reported a difference of about 3% of energy intake (15 g/day). Exclusion of these studies from the meta-analysis strengthened the overall effect of lowered sugar intakes on body weight change (−1.22 kg, 95% confidence interval −1.81 to −0.63). We saw no evidence of heterogeneity (I2=17%, P=0.3), and the test for overall effect showing an association between sugar reduction and increased weight loss was highly significant.
Effects of increasing dietary sugars on measures of body fatness in adults
Table 3⇓ describes the 10 studies identified for this analysis, and figure 4⇓ shows the quantitative meta-analysis (forest plot).26 32 34 36 37 38 43 44 45 47 Because there was statistical evidence for significant heterogeneity among the studies (I2=82%, P<0.001), we used a random effects model to derive the pooled estimates. Increased intake in dietary sugars was associated with significantly greater weight (0.75 kg (95% confidence interval 0.30 to 1.19); P=0.001) at the end of the intervention period by comparison with no increase in sugars intake. The studies involved an increase in dietary sugars; mostly sugar sweetened beverages, in the intervention arm of the randomised trial. Only two studies lasted longer than eight weeks.34 36 Subgroup analysis for these two longer term studies resulted in a significantly greater effect size (2.73 kg, 1.68 to 3.78) than the pooled effect for the shorter term studies (0.52 kg, 0.14 to 0.89). The difference between these subgroups was highly significant (P<0.001).
One trial reported a higher rate of participant dropout in the high sugars group than in the low sugars group and presented results for only participants who completed the whole study.37 Exclusion of this study from the meta-analysis increased the overall effect size slightly (0.83 kg, 95% confidence interval 0.31 to 1.35). The association also remained significant after excluding from the meta-analysis five studies26 32 34 37 43 that had a high risk of bias for two or more validity criteria (0.96 kg, 0.06 to 1.85).
Isoenergetic exchanges of dietary sugars with other carbohydrates or other macronutrient sources
We identified 12 studies that involved isoenergetic exchange of dietary sugars with other macronutrients (table 4⇓).48 49 50 51 52 53 54 55 56 57 58 Interventions ranged from two weeks to six months, and sugars were in the form of either sucrose or fructose used to sweeten foods or liquids. We saw no evidence of difference in weight change as a result of differences in sugars intakes when energy intakes were equivalent (0.04 kg (95% confidence interval −0.04 to 0.13); fig 5⇓).
Findings of cohort studies
Table 5⇓ describes 16 cohort studies in adults that provided analyses of the relation between sugars exposures and measures of adiposity.59 60 61 62 64 65 66 67 68 69 70 71 72 73 74 76 With a vote counting approach, 11 studies reported one or more significantly positive associations between a sugars exposure and a measure of adiposity,59 60 61 62 64 65 68 69 70 71 73 74 and one study reported a significantly negative association.73 Two studies reporting changes in intake of sugar sweetened beverages during follow-up showed a significantly greater increase in weight change among participants with the highest intake than in those with the lowest intake.71 74 Web table 1 summarises pooled estimates for the relation between sugars intakes and various measures of adiposity from all other prospective studies in adults that met the inclusion criteria. Forest plots for these comparisons are provided in web figures 3-5 (web appendix 5).
Effects of reducing dietary sugars on measures of body fatness in children
Table 6⇓ describes the five intervention trials identified for this analysis, and figure 6⇓ shows the forest plot.27 29 33 40 46 Interventions generally included advice to reduce sugar sweetened beverages and other foods containing (free) sugars. We saw no association between such advice to reduce intake of dietary sugars and change in standardised BMI or BMI z score in children (0.09, 95% confidence interval −0.14 to 0.32). The studies included in this meta-analysis involved advice to reduce the intake of sugar sweetened beverages alone,27 29 40 or together with a further reduction in other sugar rich foods and an increase in dietary fibre.33 46 Poor compliance with the intervention advice was reported in three of the five studies,29 33 46 and the effect of the intervention was a reduction of 51 mL/day in another study.40 Significant heterogeneity was observed and a random effects model was used for the meta-analysis. Excluding the study by Davis and colleagues,46 which had a high risk of bias for two or more validity criteria, did not alter the effect estimate.
Effects of increasing dietary sugars on measures of body fatness in children
There were no randomised trials available in children, thus we used data from 21 cohort studies in children (reported in 22 articles) to assess the effect of increasing sugars intakes on body fatness (table 7⇓). Most studies related to intake of sugar sweetened beverages. A quantitative meta-analysis (fig 7⇓) was based on five cohort studies, with seven comparisons. These studies reported data for the odds of being overweight at follow-up in children consuming about one daily serving of sugar sweetened beverages at baseline compared with children consuming none or very little.80 94 95 96 97 Comparison of the higher intakes with lower intakes suggested a significantly increased risk of being overweight associated with higher intakes (odds ratio 1.55, 95% confidence interval 1.32 to 1.82). We saw no evidence of heterogeneity, and all the studies reported a positive association. When assessing the 23 cohort studies in children using a “vote counting” approach, 15 reported a positive association between increased sugars intake and a measure of adiposity.75 79 80 81 82 86 88 89 91 92 94 95 96 97 98 Fourteen of these 15 studies reported the sugars exposure as a sugar sweetened beverage. By contrast, only four studies reported a negative association,87 90 93 98 of which two reported fruit juice as the sugars exposure.90 98
Web table 2 summarises pooled and unpooled estimates for the association between sugars intakes and measures of adiposity from all other prospective studies in children that met the inclusion criteria. Because of the wide variation in how the study effects were reported, it was not always possible to pool studies reporting similar outcomes, and there was no evidence of association between increased sugars and adiposity. Web figures 6 and 7 (web appendix 5) show forest plots.
The overall meta-regression of randomised trials examining the effect of sugars on adiposity in adults showed no evidence of a dose-response association between sugar as a percentage of total energy intake and body weight (0.02 kg (95% confidence interval −0.03 to 0.08); P=0.392). The difference in weight changes associated with differing intakes of sugars was unrelated to study design (crossover or parallel design trials; 0.30 kg (−0.44 to 1.05); P=0.393), study duration (0.01 kg per week (−0.02 to 0.05); P=0.460), or whether sugars intakes were reduced or increased in the intervention arm relative to the control arm (0.12 kg (−0.73 to 0.96); P=0.817).
The funnel plot of all 15 randomised ad libitum trials conducted in adults was asymmetrical and the Egger’s test for bias was significant (P=0.001), which suggested possible publication bias (fig 8⇓). The pooled effect size for all 15 trials was 0.78 kg (95% confidence interval 0.43 to 1.12), based on a random effects model which accounted for significant heterogeneity (I2=77%, P<0.001) seen between the relatively short term crossover trials with small variances and the longer term parallel trials with larger variances. Use of fixed effects models attenuated the overall effect (0.42 kg, 0.28 to 0.56), but it remained significant. Excluding the studies with the largest study variances34 41 from the analysis had little effect (0.72, 0.37 to 1.06). Trim and fill analysis showed a somewhat attenuated but significant effect size (0.50, 0.18 to 0.21). Visual inspection of the funnel plot and the Egger’s test for bias (P=0.248) did not suggest publication bias among the isoenergetic trials.
The meta-analyses based on controlled trials provide consistent evidence that increasing or decreasing intake of dietary sugars from current levels of intake is associated with corresponding changes in body weight in adults. Although some evidence of potential publication bias existed, this did not seem to have an important effect on the findings. Results from cohort studies were generally comparable with the trial findings. The reviewed studies largely related to the manipulation or observation of intake of sugars which, using current terminology, would be described as “free sugars.” Two six month trials, 99 100 published subsequent to the census date for this systematic review, involved different intakes of sugar sweetened beverages in adults. The trials also showed a trend towards increased body weight in participants with raised intake, but the difference between groups was not significant, perhaps owing to small number of participants.
Poor compliance with dietary advice could explain why the data from trials in children were equivocal. This was confirmed by two controlled trials published after our systematic review’s census date.101 102 De Ruyter and colleagues101 showed a smaller increase in BMI z score after 18 months among trial completers who were provided with sugar free, artificially sweetened beverages, compared with participants who received equal quantities of sugar sweetened beverages. Ebbeling and colleagues102 showed the potential of an intervention designed to decrease the consumption of sugar sweetened beverages in overweight and obese adolescents. BMI and body weight were significantly reduced after one year in the intervention group compared with the control group. However, after a further year’s follow-up with no further intervention, the difference between the groups was no longer significant.
Cohort studies in children confirmed a link between intake of sugar sweetened beverages and the risk of becoming overweight, but showed no consistent associations between other measures of sugars intake and adiposity. Although comparison of groups with the highest versus lowest intakes in cohort studies was compatible with a recommendation to restrict intake to below 10% total energy, currently available data did not allow formal dose-response analysis.
Strengths and limitation
An important strength of this in depth review of the literature lay in the overall quality and consistency of the data, especially those derived from adult populations. Although the trials were published over a long timeframe and used different experimental approaches, the results were consistent. Evidence was derived principally from randomised trials, but data from cohort studies that compared higher and lower groups of intake were also confirmatory. Criteria from both GRADE25 and the World Cancer Research Fund103 for judging strength of evidence of association specify randomised controlled trials as the highest level of evidence, but evidence from another study type is recognised as providing important confirmation.
We found less consistent findings from the trials conducted in children, which can be attributed to several factors. These trials tended to last longer than adult trials, and where compliance was assessed, it was clear that adherence to dietary advice (typically advice to reduce sugar sweetened beverages) was poor. For example, in a trial by Davis and colleagues,46 children receiving nutrition education to improve carbohydrate quality achieved a reduction in added sugars intake of only 8 g/day, compared with control children. However, in children (as in adults), comparison of the highest intakes with the lowest intakes (usually of sugar sweetened beverages) suggested that those participants consuming the largest quantities had a higher body weight or other measure of adiposity.
The limitations of these findings are those inherent to the primary research on which they are based, notably inadequacy of dietary intake data, and variation in the nature and quality of the dietary intervention. Most cohort studies and some trials reported effects largely or solely related to the consumption of sugar sweetened beverages. Most trials involved different levels of intake of sugar (sucrose) and other monosaccharides and disaccharides in the control and intervention arms. These compounds have been described as “free sugars,” as defined by WHO (all monosaccharides and disaccharides added to foods by the manufacturer, cook, or consumer, plus sugars naturally present in honey, syrups, and fruit juices).14 We had originally intended to report separately on the effects of total sugars as well as the various subcategories of sugars, but presentation of data in the studies precluded such analyses.
Assessment of dietary intake of sugars, whether by some method of recall as used in the trials, or by food frequency questionnaires as in cohort studies, was associated with a considerable degree of measurement error even when using validated methods. This is probably one explanation why a dose-response effect could not be shown between change in dietary intake and magnitude of weight change. Nevertheless, even crude estimates of intake provided assistance in interpreting potentially inconsistent findings. The studies of long term intervention in children27 29 33 40 46 and two studies of interventions reducing dietary sugars in adults28 33 found little difference in intakes between intervention and control groups, and no meaningful change in weight.
The heterogeneity of the studies, especially in terms of the consequences of altering intake of sugars in ad libitum diets, resulted in difficulties in fully explaining the effects of different dietary changes. Nevertheless, the changes in weight observed in studies of adults provided some indication of what might be achieved by the implementation of a dietary guideline relating to sugar, and conversely what might occur if consumption continued to increase.
The potential problem of residual confounding to explain some or all of an effect is inherent to all cohort studies. However, the overall consistency of our findings, regardless of study type, is reassuring. The only potential major source of bias identified in the trials was that four trials in adults reported data for completers. These data could have overestimated the effect, but we saw no meaningful difference in the magnitude of the effect between these trials and the other studies. Both participants and researchers in many of the trials were not blinded to intervention allocation. Studies providing beverages as a means of manipulating sugars intakes were blinded, but blinding was clearly not possible in studies relying on the provision of dietary advice to manipulate sugars intake. However, we do not believe that a lack of blinding altered our findings substantially. Measurement of body weight did not involve judgment that was subject to bias.
The most obvious mechanism by which increasing sugars might promote weight gain is by increasing energy consumption to an extent that exceeds energy output and distorts energy balance. For sugar sweetened beverages, it has been suggested that energy in liquid form could be less satiating than when derived from solid foods, resulting in increased consumption.104 Solid foods containing sugars are typically (although not invariably) energy dense, and frequent and substantial consumption of energy dense foods is associated with excessive weight gain and other measures of excess adiposity. We observed that isoenergetic replacement of dietary sugars with other macronutrients resulted in no change in weight (fig 5). This finding strongly suggested that energy imbalance is a major determinant of the potential for dietary sugars to influence measures of body fatness. However, other less direct mechanisms independent of energy balance have been proposed.
Sugars (particularly table sugar, sucrose, and high fructose corn syrup) contribute to the intake of fructose, which in turn can, at least in some people, increase levels of uric acid and hyperinsulinaemia.105 Hyperuricaemia has been identified as a potentially important and independent predictor of obesity and the metabolic syndrome.2 Sugar sweetened beverages and other sources of dietary fructose have been suggested to promote the deposition of liver, skeletal, and visceral fat and an increase in serum lipids independently of an effect on body weight.106 Although this issue is relevant to any overarching discussion regarding the health consequences of dietary sugars and the extent to which they should be restricted, it is beyond the scope of this review.
Results in the context of existing knowledge
Most of the relevant published studies, reviews, and meta-analyses related to the association between intake of sugar sweetened beverages and body weight, weight gain, or other measures of adiposity. Widely discrepant conclusions have emerged, ranging from strong or convincing evidence for an association8 107 to evidence described as inconclusive or equivocal.3 7 11 108 109 110 This variance is hardly surprising, owing to the poor compliance in most intervention trials, the insensitive instruments used for assessing dietary intakes in cohort studies, and that in such studies, intakes might have changed between initial dietary assessment and measurement of outcome. One meta-analysis combined data for adults and children.11 We found no evidence for an association between intake and weight in children when considering the intervention trials, nor were the data sufficient to examine for a dose-response effect when considering β coefficients for the continuous association between baseline sugars exposure and adiposity outcome. Nevertheless, we were able to show a consistent effect when comparing groups with the highest intakes of sugars with those with the lowest intakes.
There have been fewer reviews and meta-analyses relating to sugars or sugar rather than sugar sweetened beverages. In a systematic review and meta-analysis, Sievenpiper and colleagues concluded that isoenergetic substitution of fructose for other carbohydrates was not associated with weight gain.110 However, free fructose at high doses that provided excess calories modestly increased body weight to an extent probably due to the extra calories rather than any particular metabolic attributes of fructose. Dolan and colleagues111 drew similar conclusions when reviewing studies in which fructose was fed at “normal levels of intake.” Van Baak and Astrup3 and Ruxton104 recently concluded that there was insufficient evidence to indicate that replacing sugars with other carbohydrates resulted in a reduction in body weight. However, by limiting analyses to ad libitum trials, and considering studies in adults and children separately, our systematic review showed a clear positive association between higher intake of sugars and body fatness in adults, and provided an explanation as to why the findings in children were less conclusive.
This series of meta-analyses provides evidence that intake of sugars is a determinant of body weight in free living people consuming ad libitum diets. The data suggest that the change in body fatness that occurs with modifying intake of sugars results from an alteration in energy balance rather than a physiological or metabolic consequence of monosaccharides or disaccharides. Owing to the multifactorial causes of obesity, it is unsurprising that the effect of reducing intake is relatively small. The extent to which population based advice to reduce sugars might reduce risk of obesity cannot be extrapolated from the present findings, because few data from the studies lasted longer than ten weeks. However, when considering the rapid weight gain that occurs after an increased intake of sugars, it seems reasonable to conclude that advice relating to sugars intake is a relevant component of a strategy to reduce the high risk of overweight and obesity in most countries.
What is already known on this topic
Excessive intakes of dietary sugars have been linked to obesity, and a higher risk of chronic diseases, but the link with obesity is tenuous
The most consistent association has been between a high intake of sugar sweetened beverages and the development of obesity
No upper safe limit of intake has been agreed universally, but WHO has suggested that intakes of free sugars should be less than 10% of the total energy intake
What this study adds
Among free living people, advice to reduce free sugars was associated with an average 0.80 kg reduction in weight; advice to increase intake was associated with a corresponding 0.75 kg increase
This parallel effect seems to be due to an altered energy intake; isoenergetic replacement of sugars with other carbohydrates did not result in any change in body weight
Evidence was less consistent in children than in adults
Cite this as: BMJ 2012;345:e7492
We thank Carolyn Summerbell and Bernard Venn for their help on the initial development of this research; Melissa Butt and Sarah Harvey, who contributed to the data search for the randomised controlled trials; Marcus Du, who contributed to the data search and extraction for the cohort studies; and the members of the WHO NUGAG Subgroup on Diet and Health for their contribution to this work.
WHO agreed to the publication of this systematic review in a scientific journal, because it serves as the background evidence review for updating WHO guidelines on total sugars intake and should therefore, be available widely.
Contributors: The questions for the review were discussed and developed by the WHO NUGAG Subgroup on Diet and Health in February 2010, and the protocol was approved by the NUGAG Subgroup on Diet and Health. LT and SM supervised study searches. LT, SM, and JIM assessed inclusion, extracted data, and assessed validity. LT did the meta-analyses. LT and JM wrote the manuscript. The NUGAG Subgroup on Diet and Health reviewed the first draft of the report and contributed to the GRADE assessment. All authors read and approved the final draft of the report.
Funding: The authors were supported by the University of Otago and the Riddet Institute, a New Zealand National Centre of Research Excellence. The research was supported by the University of Otago, Riddet Institute, and WHO. The authors undertook the submitted work for WHO for the purposes of updating WHO guidelines on sugars intake, and WHO provided some funding to the University of Otago towards the cost of carrying out the review.
Competing interests: All authors have completed the Unified Competing Interest form at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare: support from the University of Otago, Riddet Institute, and WHO; no other financial relationships with any organisations that might have an interest in the submitted work in the previous 3 years; and no other relationships or activities that could appear to have influenced the submitted work.
Ethical approval: Not required.
Data sharing: No additional data available.
This is an open-access article distributed under the terms of the Creative Commons Attribution Non-commercial License, which permits use, distribution, and reproduction in any medium, provided the original work is properly cited, the use is non commercial and is otherwise in compliance with the license. See: http://creativecommons.org/licenses/by-nc/2.0/ and http://creativecommons.org/licenses/by-nc/2.0/legalcode.