Introduction

Non-alcoholic fatty liver disease (NAFLD) is the most prevalent chronic liver disease and a cause of raised liver enzymes in developed countries,1 affecting 10–30% of people in developed countries.2, 3 The increasing prevalence of NAFLD, which is closely linked with the increasing prevalence of obesity and type 2 diabetes mellitus (T2DM),2 has been associated with increased cardiovascular morbidity and mortality.1

Dietary factors that influence NAFLD have become a focus of attention. In particular, recent concerns have been raised regarding the role of dietary fructose in inducing NAFLD.4, 5, 6, 7 Animal models featuring extreme levels of fructose exposure8, 9, 10, 11 and poor-quality observational studies12, 13, 14 have been used to underpin this hypothesis. In the absence of consistent clinical evidence, it is unclear whether fructose at typical levels of exposure induces NAFLD. To determine the effect of fructose on markers of NAFLD in humans, we conducted a systematic review and meta-analysis of available controlled feeding trials.

Subjects and methods

Design

We followed the Cochrane Handbook for Systematic Reviews of Interventions15 and the Preferred Reporting Items for Systematic Reviews and Meta-Analysis guidelines.16 The review protocol is available at ClinicalTrials.gov (registration number: NCT01363791).

Study selection

We searched MEDLINE, EMBASE, CINAHL and the Cochrane Library databases through 3 September 2013 for relevant articles. Supplementary Appendix Table 1 shows the full search term used in this study. Manual searches supplemented the electronic search strategy. No restrictions were placed on language. We included controlled trials investigating the effect of oral fructose on markers of NAFLD. A comparison was considered isocaloric when the carbohydrate comparator was exchanged for an equal amount of fructose. If the trial involved overfeeding of fructose so that the fructose provided excess energy resulting in a positive energy balance, then the comparison was still considered isocaloric as long as the carbohydrate comparator was matched for the excess energy resulting in the same positive energy balance. A comparison was considered hypercaloric when a control diet was supplemented with excess energy from fructose compared with the same control diet alone without the excess energy. Trials that involved a follow-up of <7 days follow-up, administered intravenous fructose, lacked a control diet or did not provide suitable endpoint data were excluded.

Data extraction

Two reviewers (SC, AIC) independently reviewed and extracted relevant data from each report. The quality of each study was assessed using the Heyland methodological quality score (MQS).17 Disagreements were reconciled by consensus. Mean±s.d. differences between fructose and control arms were extracted as the main end points. In those trials where the data were included in figures and not provided numerically, we used software program Plot Digitizer (http://plotdigitizer.sourceforge.net/) to extract the data. Additional information was requested from the authors of all included trials.

Access to study

All authors had access to the study data and reviewed and approved the final manuscript.

Statistical analysis

Data analyses were conducted using Review Manager (RevMan) version 5.1.6 (Copenhagen, Denmark) for primary analyses and Stata (version 12, College Station, TX, USA) for subgroup analyses. Separate analyses were conducted for the isocaloric and hypercaloric trials using the generic inverse variance method with random effects weighting. Data were expressed as standardized mean differences with 95% confidence intervals (CIs) for intrahepatocellular lipid (IHCL) and mean differences (MD) with 95% CIs for alanine aminotransferase (ALT).

Trials that did not report SE values had these computed from the available statistics using standard formulae.15, 18 To generate SE for included crossover trials, we assumed a paired analyses as described by Elbourne.18 If insufficient data were available for computations in crossover trials, SE values were imputed using the pooled correlation coefficient between baseline and end-of-study values derived from a meta-analysis of trials reporting sufficient data or assuming a conservative correlation coefficient of 0.5 with sensitivity analyses at 0.25 and 0.75.

Inter-trial heterogeneity was assessed by the Cochran Q statistic with α<0.10 considered significant, and quantified by the I2 statistic, where I250% indicates substantial heterogeneity.15 Sources of heterogeneity were investigated by sensitivity analyses in which each individual study was removed from the analysis and through a priori subgroup analyses by comparator, baseline values, fructose form, follow-up, MQS, randomization, design and energy balance. Meta-regression analyses assessed the significance of subgroup effects. Publication bias was evaluated via visual inspection of funnel plots and Egger and Begg tests.

Results

Search results

Figure 1 shows the trial selection process. We identified 1437 eligible reports. A total of eight reports (providing data for 13 trials) were selected for analyses.19, 20, 21, 22, 23, 24, 25, 26

Figure 1
figure 1

Flow of the literature.

Trial characteristics

Table 1 shows the trial characteristics. Only two markers of NAFLD were identified: IHCL and ALT. None of the available trials assessed NAFLD histologically from liver biopsies. There were a total of seven isocaloric trials (four for IHCL and six for ALT) in 184 healthy participants, and six hypercaloric trials (five for IHCL and four for ALT) in 76 participants (n=60 healthy and n=16 offspring of type 2 diabetes). The majority of both sets of trials were conducted in European countries in an outpatient setting and tended to be small (median (interquartile range (IQR)) sample size, 29.0 (24.5–31.5) and 13.5 (10.5–15.8), in isocaloric and hypercaloric trials, respectively).

Table 1 Characteristics of controlled feeding trials investigating the effect of fructose on markers of NAFLD

Participants tended to be healthy, young (median (IQR) age=30.5 years (26.3–33.9 years) and 27.6 years (24.7–33.9 years)), male (median (IQR) percent male:female ratio=100% (56–100) and 100% (77.5–100)), and overweight (median (IQR) body mass index=25.9 kg/m2 (22.4–29.4 kg/m2) and 24.3 kg/m2 (22.2–28.5 kg/m2)) in isocaloric and hypercaloric trials, respectively. Median (IQR) baseline ALAT values (in U/l) were 26.0 (23–28.9) in isocaloric trials and 20.58 (18.5–25.8) in hypercaloric trials. Median values for baseline IHCL (in %) could not be computed from the data reported.

Crossover designs were used in 29% of isocaloric trials and all hypercaloric trials. The majority of isocaloric trials (86%) and 17% of hypercaloric trials were randomized. Glucose was the comparator in all isocaloric trials except in the two trials of Aeberli et al.,21 where glucose and sucrose were the comparators in the high dose trial and glucose and starch were the comparators in the low dose trial. The control diet alone without the added energy from fructose was the comparator in all hypercaloric trials. Although comparisons in all isocaloric trials were matched for energy, 86% of the isocaloric trials provided fructose and the carbohydrate comparator under conditions of positive energy balance (that is, both arms provided excess energy), whereas 14% provided fructose and the carbohydrate comparator under conditions of neutral energy balance (that is, both arms provided energy to maintain weight). All isocaloric and hypercaloric trials administered fructose in fluid form at a median (IQR) dose of 182 g/day (115–204 g/day) for isocaloric trials and +193 g/day (+158–211 g/day) for hypercaloric trials. The median (IQR) excess energy provided by the hypercaloric trials was +25% (+23–33%). All isocaloric and hypercaloric trials featured high-carbohydrate and low-fat diets with similar macronutrient profiles: 50–55% energy carbohydrate, 30–35% energy fat and 13–15% energy protein. Metabolic feeding control was used in 14% of isocaloric trials and 17% of hypercaloric trials; partial-metabolic feeding control was used in 43 and 17% and the remainder provided fructose as a supplement. The median (IQR) dietary follow-up was 4 weeks (3–8 weeks) for isocaloric trials and 3 weeks (1.25–4 weeks) for hypercaloric trials.

The Heyland MQS was considered high (MQS8) in 71% of isocaloric and 33% of hypercaloric trials. Lack of or poor description of randomization, nonconsecutive or poorly described patient selection and absence of double-blinding contributed to lower scores. Funding of all trials was from a combination of agency alone (69%) or agency–industry sources (31%). None were funded by industry alone. None reported a potential conflict of interest.

Effect of fructose on IHCL

Figure 2a shows the effect of fructose on IHCL in isocaloric trials. Primary pooled analyses showed no effect of fructose on IHCL (standardized mean difference=−0.09 (95% CI: −0.36–0.18), P=0.51), with no evidence of inter-study heterogeneity (I2=0%, P=0.95). Sensitivity analyses did not alter the results. Meta-regression analyses showed no statistically significant subgroup effects (Supplementary Appendix Figure 1).

Figure 2
figure 2

Forest plots of the effect of fructose on intrahepatocellular lipid (IHCL) in healthy participants in (a) isocaloric and (b) hypercaloric feeding trials. Pooled effect estimates shown as diamonds. Data are expressed as weighted MD with 95% CI using generic inverse variance random effects models. Inter-study heterogeneity was tested by Cochrane’s Q statistic (χ2-test) at a significance level of P<0.10 and quantified by I2, where I250% is considered to be evidence of substantial heterogeneity and 75%, considerable heterogeneity. Any CHO denotes any carbohydrate comparator; E neutral, neutral energy balance; E positive, positive energy balance; and Off-T2DM, offspring of T2DM.

Figure 2b shows the effect of fructose on IHCL in hypercaloric trials. Primary pooled analyses showed that fructose raised IHCL (standardized mean difference=0.45 (95% CI: 0.18–0.72), P=0.001), though there was significant inter-study heterogeneity (I2=55%, P=0.07). Sensitivity analyses did not alter the results, but identified that the removal of Lê et al.,23 a study conducted in offspring of individuals with T2DM, eliminated evidence of inter-study heterogeneity (I2=0%, P=0.87). Meta-regression analyses showed no statistically significant subgroup effects (Supplementary Appendix Figure 2), and inter-study heterogeneity remained largely unexplained.

Effect of fructose on liver enzymes

Figure 3a shows the effect of fructose on ALT in isocaloric trials. To approximate paired analyses for crossover trials, we used a conservative correlation coefficient of 0.5. Primary pooled analyses showed no effect of fructose on ALT (MD=0.15 (95% CI: −1.51 to 1.82), P=0.86), with no significant evidence of inter-study heterogeneity (I2=0%, P=0.97). Neither sensitivity analyses nor the use of more (0.75) or less (0.25) conservative correlation coefficients altered the results. Meta-regression analyses revealed no statistically significant subgroup effects (Supplementary Appendix Figure 3).

Figure 3
figure 3

Forest plots of the effect of fructose on ALT in healthy participants in (a) isocaloric and (b) hypercaloric feeding trials. Pooled effect estimates shown as diamonds. Data are expressed as weighted MD with 95% CI using generic inverse variance random effects models. Inter-study heterogeneity was tested by Cochrane’s Q statistic (χ2-test) at a significance level of P<0.10 and quantified by I2, where I250% is considered to be evidence of substantial heterogeneity and 75%, considerable heterogeneity. Any CHO denotes any carbohydrate comparator; E neutral, neutral energy balance; E positive, positive energy balance; HD, high dose; LD, low dose; and Off-T2DM, offspring of T2DM.

Figure 3b shows the effect of fructose on ALT in hypercaloric trials. To approximate paired analyses for crossover trials, we needed to use a conservative correlation coefficient of 0.5. Primary analyses showed a significant ALT-increasing effect (MD=4.94 (95% CI: 0.03–9.85), P=0.05), with significant evidence of inter-study heterogeneity (I2=78%, P=0.003). Sensitivity analyses revealed the removal of either Lê et al.,23 Sobrecases et al.24 or Johnston et al.26 led to a loss of significance (MD=2.97 (95% CI: −1.40, 7.35); MD=5.40 (95% CI: −1.97, 12.78); and MD=4.84 (95% CI: −1.51, 11.19)), respectively. Removal of Cox et al.,25 significantly reduced evidence of inter-study heterogeneity (I2=56%, P=0.11). Sensitivity analysis using higher (0.75) or lower (0.25) conservative correlation coefficients did not alter the results. Meta-regression analyses did not show any statistically significant subgroup effects (Supplementary Appendix Figure 4).

Publication bias

We examined funnel plots for evidence of publication bias (Supplementary Appendix Figure 5). There was some evidence of slight asymmetry in the hypercaloric trials for ALT on visual inspection (P=0.056 by Egger test; P=0.089 by Begg test), but no small study effects were detected among the isocaloric and hypercaloric trials for either IHCL or ALT by Egger and Begg tests (P>0.05).

Discussion

The present aggregate analyses of 13 trials in 260 predominantly young, male participants, who were overweight/obese or otherwise healthy, investigated the effect of fructose on markers of NAFLD under two different types of trial conditions: one where fructose in beverage form was isocalorically exchanged for other carbohydrates and the other where fructose in beverage form supplemented control diets with excess energy (+21–35% energy) at extreme doses (104–220 g/day) relative to the same control diets without the excess energy. These two types of trial conditions produced different results. Although there was no effect of fructose in isocaloric trials, fructose increased both IHCL and ALT in hypercaloric trials.

Relation of findings to other lines of evidence

Our finding of a lack of effect of fructose on NAFLD markers in isocaloric trials contradicts evidence from animal models and observational studies. The ability of fructose to induce a metabolic syndrome phenotype and NAFLD is thought to lie in its ability to act as an unregulated substrate for de novo lipogenesis, bypassing the major rate-limiting step of glycolysis at phosphofructokinase.7, 27 This mechanism contributes significantly to de novo lipogenesis in rodent models, in which fructose fed at supraphysiological doses under isocaloric (60% energy) or hypercaloric (+30% excess energy) conditions induces steatosis and steatohepatitis.8, 9, 10, 11 Small cross-sectional and retrospective case–control studies have also shown an association between fructose-containing sugar intake and NAFLD.12, 13, 14 Clinical translation of these data, however, has several limitations. Rodent models are complicated by supraphysiological doses and excess energy,28 and marked differences exist in the metabolic fate of fructose between animals and humans. Although de novo lipogenesis from fructose accounts for 60–70% of fatty acids in rodents,28 its contribution in humans is quantitatively insignificant.29, 30 Two carefully conducted reviews of the available isotopic tracer studies showed that de novo lipogenesis from fructose contributes <1% of fatty acids, whereas glucose (50%), lactate (25%) and glycogenesis (>15%) synthesis remain the major pathways of hepatic fructose disposal in humans.29, 30 Cross-sectional and retrospective case–control studies do not provide evidence of causation and have found positive associations with many other factors that might be equal or better predictors of NAFLD, such as increased intake of energy, total fat, total carbohydrate, animal protein, cholesterol and the n-6:n-3 ratio of polyunsaturated fatty acids and decreased intake of dietary fiber.6 No large prospective observational studies have evaluated the relationship between fructose and NAFLD.

Energy represents an important confounding factor in the effect of fructose. Overfeeding of a ‘fast food’ diet has been shown to relate to an increase in ALT in healthy paticipants.31 Randomized trials of energy-restricted diets focusing on total energy reduction and exercise to promote weight loss have also shown reversal of NAFLD markers in people with NAFLD.6, 32 In the present analyses, we observed increases in IHCL and ALT only in hypercaloric trials. The lack of effect in the isocaloric trials was seen even under conditions of positive energy balance. Six of the isocaloric trials (three of four trials assessing IHCL19, 20, 26 and five of six trials assessing ALT19, 25, 26) used excess energy diets in both the fructose and comparator arms, so permitting the effect of fructose to be isolated from that of energy under matched, yet excess energy-feeding conditions. Restricting our analyses to these trials did not show an effect of fructose on NAFLD markers. We made similar observations for the lack of effect of fructose on both body weight33 and uric acid34 in two earlier systematic reviews and meta-analyses. These data suggest that the effect of fructose on NAFLD markers may not be different from that of other carbohydrates as long as energy remains matched.

Previous meta-analyses have identified subgroup effects on related metabolic end points. A dose threshold was observed for a triglyceride-raising effect of fructose: 100 g/day for fasting and 50 g/day for postprandial triglycerides across different participant groups35 and >60 g/day for fasting triglycerides in type 2 diabetes.36 A fasting triglyceride-raising effect of fructose was also seen where starch was the comparator and follow-up was 4 weeks in type 2 diabetes,36 whereas a weight-loss effect of fructose was seen in overweight/obese individuals and where fructose was in fruit form.33 None of these subgroup analyses were significant in the present analysis. Although the number of trials was small, the lack of effect modification across a priori subgroup analyses was consistent with that seen in our earlier meta-analyses for blood pressure37 and uric acid.34

Limitations

Our analyses have several limitations. First, the available trials had small sample sizes and narrow participant demographics. Combining the seven isocaloric and six hypercaloric trials, our median sample size was 16 participants, the majority of whom were young, male, and either overweight/obese (without any comorbidities) or otherwise healthy. Although the baseline IHCL values in the overweight/obese participants were >95th percentile for the general population (>5.56%),38 the data generated from such a generally healthy group may not be truly reflective of the disease physiology in people with or at risk for NAFLD, especially given that in patients with histologically established NAFLD, fructose may be associated with worse disease.39 Second, none of the trials in our meta-analysis had a follow-up period exceeding 10 weeks. The isocaloric and hypercaloric trials had a median follow-up of 4 weeks and 3 weeks, respectively. It is unclear whether the changes in IHCL and ALT seen in hypercaloric trials or the null effects seen in isocaloric trials are sustainable over the longer term. Third, study quality was poor (MQS<8) in 46% of the trials. Most of the low-quality scores were attributable to a lack of or poor description of randomization, nonconsecutive or poorly described patient selection and absence of blinding. However, no effect modification by study quality was seen in subgroup analyses. Fourth, none of the available trials assessed NAFLD by histological analysis of liver biopsies. This analysis remains the gold standard assessment for NAFLD, as ALT is quite insensitive, while IHCL by 1H-magnetic resonance spectroscopy cannot detect inflammation and/or fibrosis.1 The two measurements, however, showed good agreement among the trials. Finally, given the small number of available trials, publication bias remains unclear, although no small study effects were detected.

Implications

Although our results bear on the question of whether fructose-containing sugar-sweetened beverages have a unique role in the development of NAFLD, their translation to ‘real-world’ intake patterns is complicated. The median level of fructose exposure was >95th percentile U.S. intake (87 g/day)40 across all trials: 2.5-fold greater than this threshold (+215 g/day providing +35% excess energy) in the hypercaloric trials, in which there was an effect, and 1.4-fold greater than this threshold (115 g/day) in the isoclaloric trials, in which there was no effect. Also, no trials used non-beverage grain or fruit sources of fructose, which together account for >30% of fructose in the U.S. diet40 and have been linked (as whole grains and fruits) to weight loss and improved metabolic outcomes in large prospective cohort studies41, 42 and randomized trials.43, 44 Dietary trials of more representative sources of fructose at more representative levels of exposures remain a research priority.

Conclusions

In conclusion, our preliminary systematic review and meta-analysis does not support a NAFLD-inducing effect of fructose in isocaloric exchange for other carbohydrates at levels of exposure that are well above that found in Western diets. The evidence does support an IHCL- and ALT-increasing effect of diets supplemented with fructose providing excess energy (+21–35% energy) at extreme doses (104–220 g/day). Confounding from excess energy, however, cannot be excluded in the hypercaloric trials, such that the observed NAFLD-inducing effect is more attributable to the excess energy than the fructose itself. Other sources of uncertainty in our analyses include the small number of available trials (four for IHCL and six for ALT), as well as the relatively small sample sizes (<30 participants/trial) and narrow participant demographics (most participants were young and relatively healthy). The short follow-up (all trials were <12-weeks) using IHCL and ALT as markers of NAFLD may also not be relevant to the natural history of NAFLD over the longer term, especially in people who may be at low risk. It is unclear whether a larger number of trials which address these many issues will show the same findings when analyzed collectively. To understand the role of fructose in the epidemic of NAFLD, there remains a need for larger, longer, high-quality trials of the effect of ‘real-world’ intake patterns of fructose on histopathological changes of NAFLD in at risk populations.