meta分析范文展示

The effect of fructose consumption on plasma cholesterol in adults: a meta-analysis of controlled feeding trials1,2,3Tao An4,5, Rong Cheng Zhang4,5, Yu Hui Zhang4, Qiong Zhou4, Yan Huang4, Jian Zhang4, *.4 State Key Laboratory of Cardiovascular Disease, Fuwai Hospital, National Center for Cardiovascular Diseases, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China5Tao An and Rong Cheng Zhang contributed equally to this study.3 Supplemental Table 1 and supplemental Figures 1-4 are available as Online Supporting Material with the online posting of this paper atRUNNING TITLE: Fructose and cholesterolWORD COUNT: 5618; NUMBER OF FIGUREA: 3; NUMBER OF TABLES: 2 SUPPLEMENTARY MATERIAL: Online Supporting Materials: 5AUTHOR LIST FOR INDEXING: An, Zhang, Zhang, Zhou, Huang, Zhang1 The study was supported by the Ministry of Science and Technology of China with grant of the National High-tech Research and Development Program of China to Dr Jian Zhang.2 Author disclosures: T. An, R.C. Zhang, Y.H. Zhang, Q. Zhou, Y. Hung, J. Zhanghave no conflicts of interest.* To whom correspondence should be addressed. Mailing address: Heart FailureCenter, Cardiovascular Institute and Fuwai Hospital, Chinese Academy of MedicalSciences and Peking Union Medical College, 167 Beilishilu, Beijing, China; Zip code: 100000; Telephone number: 86-10-88396180; Fax number: 86-10-; E-mail:PROSPERO REGISTRATION NUMBERS: CRD420120033511ABSTRACT2Fructose is widely used as a sweetener in production of many foods, yet the relation 3between fructose intake and cholesterol remains uncertain. We performed a systematic 4review and meta-analysis of human controlled feeding trials of isocaloric fructose 5exchange for other carbohydrates to quantify the effects of fructose on total 6cholesterol (TC), LDL cholesterol (LDL-C), and HDL cholesterol (HDL-C) in adult 7humans. Weighted mean differences were calculated for changes from baseline 8cholesterol concentrations by using generic inverse variance random-effects models. 9The Heyland Methodological Quality was used to assess study quality. Subgroup 10analyses and meta-regression were conducted to explore possible influence of study 11characteristics. Twenty-four trials (with a total of 474 subjects) were included in our 12meta-analysis. In an overall pooled estimate, fructose exerted no effect on TC, LDL-C 13and HDL-C. Meta-regression analysis indicated that fructose dose was positively 14correlated with the effect sizes of TC and LDL-C. Subgroup analyses showed that 15isocaloric fructose exchange for carbohydrates could significantly increase TC by 1612.97 mg/dL (95%CI: 4.66, 21.29; P = 0.002) and LDL-C by 11.59 mg/dL (95%CI: 174.39, 18.78; P = 0.002) at >100g fructose/d but had no effect on TC and LDL-C when 18fructose intake was ≤100g/d. In conclusion, very high fructose intake (>100g/d) 19could lead to significantly increase in serum LDL-C and TC. Larger, longer and 20higher-quality human controlled feeding trials are needed to confirm these results.21Key words: fructose, cholesterol, meta-analysis2223INTRODUCTION24Hyperlipidemia is a common risk factor for coronary heart disease (CHD), with 2544.4% of adults in the United States having abnormal TC values and 32% having 26elevated LDL-C levels (1). Compared to subjects with normal blood lipid, those with 27hyperlipidemia have a 3-fold risk of heart attacks (2). Lifestyle modification should be 28initiated in conjunction both primary and secondary prevention of CHD. More 29consideration exists as to what constitutes healthy eating.30Fructose is the most naturally occurring monosaccharide, and has become a major 31constituent of our modern diet. Fruit, vegetables, and other natural sources provide 32nearly one-third of dietary fructose, and two-thirds come from beverages and foods in 33the diets (eg, candies, jam, syrups, etc) (3). Fructose is preferred by many people, 34especially those with diabetes mellitus because of its low glycemic index (23% versus 35glucose 100%) (4). After intestinal uptake, fructose is mainly removed from the blood 36stream by the liver in an insulin-independent manner, and is used for intrahepatic 37production of glucose, fatty acids or lactate. Cross-sectional studies in human suggest 38that excessive fructose consumption can lead to adverse metabolic effects, such as 39dyslipidemia and increased visceral adiposity (5-7). The Dietary Guidelines for 40Americans, 2010, point out that it is lack of sufficient evidence to set a tolerable upper 41intake of carbohydrates for adults (8). Although The Candian Diabetes Association 42suggests consumption of no more than 60g of added fructose per day by people with 43diabetes for its triglyceride-raising effect (9), the threshold dose of fructose at which 44the adverse influence on cholesterol is controversial.45To determine the effect of fructose on cholesterol, a substantial number of clinical 46trials have been performed on adult humans with different health status (diabetic, 47obese, overweight, hyperinsulinemic, impaired glucose-tolerant and healthy). These 48trials used various intake levels of fructose and different protocols. Thus, it is difficult 49to reach a consistent conclusion across these studies. Therefore, we conducted a 50systematic review of the scientific literature and meta-analysis of controlled feeding 51trials to evaluate the effect of isocaloric oral fructose exchange for carbohydrates on 52cholesterol and to clarify the active factors of fructose.53Materials and Methods54This meta-analysis followed the Preferred Reporting Items for Systematic Reviews 55and Meta-analyses (PRISMA) criteria (10).56Search strategy.We searched PubMed (http://; from 1966 to December 2012), 57Embase (http://; from 1966 to December 2012) and the Cochrane Library database () 58by using the following search terms: fructose and (lipemia or lipaemia or lipids or 59cholesterol or “total cholesterol”or “LDL cholesterol” or “HDL cholesterol”) in 60English. We also searched China National Knowledge Infrastructure () and Wangfang 61database () in Chinese according to the search strategy. The search was restricted to 62reports of trials on humans.63Study selection.All clinical trials using fructose and indexed within the above 64databases were collected. Two independent reviewers (T.A., R.C.Z) screened the 65abstracts and titles for initial inclusion. If this was not sufficient, full texts articles 66were obtained and reviewed by at least two independent reviewers (T.A., R.C.Z, Q.Z., 67Y.H.). The reference lists of retrieved articles also used to supplement the database.68Any disagreements were resolved through discussion. We included controlled feeding 69trials investigating the chronic effect of fructose on blood cholesterol, from both 70randomized and nonrandomized studies, if they met the following criteria: subjects 71must have been administered fructose for at least 2 weeks; studies investigated the 72effect of oral free (unbound, monosaccharide) fructose when compared with isocaloric 73control diet with another carbohydrate in place of fructose; studies were performed in 74human adults with either a parallel or crossover design; subjects in both experimental 75groups and control groups were instructed to consume isocaloric diets. If the study 76reported any comparisons, we included all such comparisons in the meta-analysis.77Data extraction and quality assessment.Two reviewers (T.A., R.C.Z) independently 78extracted relevant data from eligible studies. Disagreements were resolved by one of 79the two authors (Y.H.Z., J.Z.). These data included information on study features 80(author, year of publication, study design, randomization, blinding, sample size, 81comparator, fructose form, dose, follow-up and macronutrient pro the background 82diet), participant characteristics (gender, age and healthy status) and baseline and final 83concentrations or net changes of total cholesterol, LDL-C and HDL-C. Data initially 84extracted were converted to system international unit (eg, TC: 1 mmol/L converted to 8538.6 mg/dL). For multi-arm studies, only intervention groups that met inclusion 86criteria were used in this analysis. If blood lipid concentrations were measured several 87times at different stages of trials, only final records of lipid concentrations at the end 88of the trials were extracted for this meta-analysis.89The quality of each study was assessed with the Heyland Methodological Quality 90Score (MQS) (11), generalized as follows: randomization; analysis; blinding; patient 91selection; comparability of groups at baseline; extent of follow up; treatment protocol;92co-intervention; outcomes. The highest score for each area was two points. Higher 93numbers represented a better quality (MQS≥8).94Data synthesis.Statistical analyses were performed with Stata software (version 11.0;95StataCorporation, TX, USA) and REVMAN software (version 5.2; Cochrane 96Collaboration, Oxford, United Kingdom). Separate pooled analyses were conducted 97by using the generic inverse variance random-effects models even where there was no 98evidence of between-study heterogeneity because these models give more 99conservative summary effect estimates in the presence of undetected residual 100heterogeneity than fixed-effects models. The different changes from baseline between 101fructose and carbohydrate comparators for total cholesterol, LDL cholesterol and 102HDL cholesterol were used to estimate the principle effect. We applied paired 103analyses to all crossover trials according to the methods of Elbourne and colleagues 104(12). Weighted mean differences of fructose consumption on cholesterol 105concentrations and corresponding 95% CIs were calculated. A 2-sided P value <0.05 106was set as the level of significance for an effect. The variances for net changes in 107serum cholesterol were only reported directly in two trials (29, 31). We calculate net 108changes for other studies by using the means±SDs cholesterol concentrations at 109baseline and at the end of intervention period (13). SDs were calculated from SEs 110when they were not directly given. If these data were unavailable, we extrapolated 111missing SDs by borrowing SDs derived from other trials in this meta-analysis (14). In 112addition, we assumed a conservative degree of correlation of 0.5 to impute the 113change-from-baseline SDs, with sensitivity analyses performed across a range of 114possible correlation coefficients (0.25 and 0.75) (13). For crossover trials in which 115only final measurements were included, the differences in mean final measurements 116were assumed on average to be the same as the differences in mean change scores 117(13). Inter-study heterogeneity was tested by the Cochrane’s Q-test (P < 0.1), and was 118quantified by the I2statistic, where I2 ≥ 50% was evidence of substantial heterogeneity. 119To explore the potential effects of factors on the primary outcomes and investigate the 120possible sources of heterogeneity, we performed meta-regressions and predefined 121subgroup analyses stratified by comparator, dose, study duration, randomization, 122health status, study design and study quality. As for studies used a range of fructose 123doses, the average doses calculated on the basis of the average reported energy intake 124or weight of participants (28.5 calories per kilogram of body weight). Sensitivity 125analyses were also performed according to the Cochrane Handbook for Systemic 126Review. Funnel plots and Egger’s linear regression test were conducted to detect 127publication bias.128RESULTS129Based on our search criteria, 1602 eligible studies were identified, and 1565 130studies were excluded on review of the titles and abstracts. The remaining 37 studies 131were retrieved and fully reviewed. Fifteen of these did not meet the inclusion criteria 132and were excluded in the final analysis. A total of 22 studies (providing data for 24 133trials) involving 474 subjects (15-36) were included in the meta-analysis 134(Supplemental Fig. 1, Table 1).135The reports of Koh and Reiser (22, 23) included two trials (bringing the total 136number of trials to 24). Eleven trials were randomized (17, 18, 20, 21, 25, 27-29, 31, 13734, 36). Nineteen trials used crossover (15-19, 21-32), and five used parallel designs 138(20, 33-36). As for the 19 cross-over trials, 10 trials have reported the washout period 139(16, 18, 22, 25, 27-31), 9 trials did not have washout period (15, 17, 19, 21, 23, 24, 26, 14032). The trials varied in size, from 8 to131 subjects. The mean age of trial participants 141ranged from 26.7 to 64.4 years. Seventeen trials (15, 17-23, 25, 27, 28, 30, 31, 34, 36) 142were performed in outpatient settings, 3 trials (26, 29, 32) in inpatient settings, and 4 143trials in both outpatient and inpatient settings (16, 24, 33, 35). Nine trials were 144conducted on diabetic subjects (19-21, 24-27, 29, 30), 8 trials in healthy subjects (17, 14518, 22, 23, 28, 31, 34, 35), 3 trials in overweight/obese subjects (32, 33, 36), 2 trials in 146hyperinsulinemic subjects (16, 23), 1 trial in those who were impaired 147glucose-tolerant (22), and 1 trial in subjects with type IV hyperlipoproteinaemia (HLP) 148(15). Background diets were 42-55% carbohydrate, 25-38% fat, and 13-20% protein. 149The carbohydrate comparators choose starch in 13 trials (15, 16, 21, 23-25, 27-30, 32, 15036), glucose in 6 trials (22, 31, 33-35), sucrose in 3 trials (17, 18, 26), and mixed 151carbohydrates in two trials (19, 20). Four trials used fructose in crystalline (16, 18, 20, 15221), 5 trials in liquid (19, 32-35), and 15 trials in mixed form (15, 17, 22-31). The 153reported mean baseline serum TC ranged from 170 to 230.8 mg/dl, LDL-C ranged 154from 90.7 to 157 mg/dl, and HDL-C ranged from 35.1 to 57.1 mg/dl. Nineteen trials 155reported the fructose intake among background diet was not different between the 156fructose and control groups, in which 15 trials reported the background fructose intake 157account for ≤3% of total energy (9 to 24g) (15-23, 29, 32, 33, 35), while 4 trials did 158not report the proportion of it (24, 25, 26, 34). Four trials used background fructose ≤3% 159(3.2 to 18g) of total energy in the control groups, but put total fructose into 160consideration in the fructose group (27, 28, 30-31). Only o ne trial reported less than 16120g (4.3 % of total energy) fructose was consumed among basal diet (36). The baseline 162values were not provided in 5 trials (19, 22, 23). The median fructose dose in the 163available trials included in our meta-analysis was 79.25 g/d (range: 30-182 g/d), and 164the duration varied from 2 to 26 weeks.165The quality scores of each study ranged from 6 to 9. Fifteen trials were classified 166as high quality (MQS≥8),and 8 trials were of low quality (17, 19, 26, 30, 32-35). 167Only three trials were blinded, one single-blinded (34) and 2 double-blinded (29, 35). 168Eight trials (19, 21, 24, 26-30) received industry funding. Three studies with four 169trials (15, 16, 22) did not report any information about financial conflicts of interest. 170Effect of fructose on cholesterol171Total cholesterol.Twenty-two trials (16-34, 36) reported the value of TC, and the 172pooled estimate was 2.47 mg/dL (95% CI: -3.04, 7.98; P = 0.38) without statistically 173heterogeneity (heterogeneity Chi2 = 28.14, I2= 25%, P = 0.14) (Fig. 1). The residual 174sources of heterogeneity were investigated by meta-regression models. Univariate 175meta-regression showed that the fructose dose was positively related to TC, even after 176adjusted for study duration and health status(regression coefficient = 0.18; 95% CI: 1770.06, 0.31, P = 0.008)(Table 2). The dose-response relation between fructose 178consumption and TC largely explained the residual heterogeneity of the effect. 179Subsequently, we stratified fructose dose ≤60, >60 to 100, and >100 as moderate, 180high, and very high, respectively, according to Candian Diabetes Association and 181reference ranges for fructose (9, 37, 38). Fructose could significantly increase TC by 18212.97 mg/dL (95%CI: 4.66, 21.29; P= 0.002) when fructose intakes were >100g/d 183but had no effect on TC if fructose was given lower than 100g. Predefined subgroup 184analyses were conducted by study characteristics (Supplemental Table 1). Sensitivity 185analyses according to possible correlation coefficients (0.25 and 0.75) and 186systematically removal of each individual trial did not alter the overall analysis and 187analyses stratified by dose.188LDL cholesterol.The mean change for LDL cholesterol in nineteen trials (15, 16, 18, 18920, 22, 23, 25-35) was 3.76 mg/dL (95% CI: -1.07, 8.6; P = 0.13) without statistically 190heterogeneity (heterogeneity Chi2 = 19.85, I2= 9%, P = 0.34) (Fig. 2). The residual 191sources of heterogeneity were investigated by meta-regression models. Univariate 192meta-regression showed that the fructose dose was positively related to LDL-C, even 193after adjusted for comparators, study duration and health status(regression coefficient 194= 0.15; 95% CI: 0.03, 0.28, P = 0.02)(Table 2). The dose-response relation between 195fructose consumption and LDL-C largely explained the residual heterogeneity of the 196effect. We stratified fructose dose according to CDA and reference ranges for fructose 197(9, 37, 38). Fructose intake >100g/d could significantly increase LDL-C by 11.59 198mg/dL (95%CI: 4.39, 18.78; P= 0.002). Predefined subgroup analyses were 199conducted by other study characteristics (Supplemental Table 1). Sensitivity analyses 200across possible correlation coefficients (0.25 and 0.75) did not alter the overall 201analysis and analyses stratified by dose. The removal of Cybulska et al resulted in a 202significant LDL-C-raising effect in the overall analysis (P = 0.03).203HDL cholesterol.The result of HDL cholesterol was calculated based on 24 trials 204(15-36), the mean difference was -0.56 mg/dL (95% CI: -2.05, 0.93; P = 0.46) without 205heterogeneity (heterogeneity Chi2 = 21.85, I2= 0%, P= 0.53) (Fig. 3). 206Meta-regression analysis did not show significant effect modifier of HDL-C. 207Predefined subgroup analyses were conducted by study characteristics (Supplemental 208Table 1). Sensitivity analyses according to possible correlation coefficients (0.25 and 2090.75) and systematically removal of each individual trial did not alter the overall 210analysis.211Publication bias212Funnel plots and Egger’s test indicated no significant publication bias in the 213meta-analyses of TC, LDL cholesterol, and HDL cholesterol (TC Egger’s test: P= 2140.881; LDL cholesterol Egger’s test: P= 0.815; HDL cholesterol Egger’s test: P= 2150.484) (Supplemental Figs. 2-4).216DISCUSSION217This meta-analysis of 24 controlled feeding trials with 477 subjects found no 218effect on TC, LDL-C and HDL-C when fructose was substituted for other 219carbohydrates. Residual heterogeneity was detected by meta-regression for this 220outcome that fructose dose was positively correlated with the effect sizes of TC and 221LDL-C.222The present meta-analysis is consistent with a prospective 2-year trial on chronic 223effect of fructose from Turku sugar studies XI, which did not report any change in 224cholesterol for those individuals who consumed more than 100g fructose/d (39). 225Aeberli et al reported another prospective, randomized, 3-week controlled crossover 226trial in which healthy young men were fed 80 g/d free fructose, and found a 227significant atherogenic LDL subclass distribution (40). However, there was an average 228of 34g combined fructose consumed among basal foods in this study, which meant 229subjects consumed fructose over 110g/d. The median dose of fructose available in our 230meta-analysis was ≈79.25 g/d, it was higher than 90th percentile (78 g/d) and lower 231than 95th percentile (87 g/d) in the United States, reported by the National and Health 232and Nutrition Examination Survey III (41). As for subjects with diabetic mellitus, 233Sievenpiper et al did not report cholesterol-raising effect if the fructose dose was >60 234g/d (median: 97.5 g/d) in their meta-analysis (42). The result of our study and 235intervention trials may be supported the idea that fructose did not increase cholesterol 236for the subjects with generalizable levels of exposure.237The results of subgroup analyses showed that the effects of fructose intake on TC 238and LDL-C were significant as the fructose dose > 100g/d. An intake of 100g/d is 239approximately equal to 400kcal/d or 20% of energy intake for a sedentary person with 240an energy requirement of 2000 kcal/d. The doses for cholesterol-raising effect account 241for less than 10 percent of intake in males and females aged 19 to 22 years, the group 242with the highest level of exposure in the United States (41). Another study found that 243the upper quintile of Americans consume more than 110g fructose daily as added 244sugar or as high-fructose corn syrup (43). Although a small number of people 245consume fructose at very high dose, it is necessary to advise them to change their 246lifestyle.247The dose-dependent effect on triglyceride was also reported in a recent 248meta-analysis that concluded the same dose threshold of 100g/d for a 249triglyceride-increasing effect of fructose on fasting triglyceride level in adult humans 250(38). For healthy subjects who consumed 150g of fructose/day, endogenous 251cholesterol synthesis and the fat content of viscera and liver have been shown to 252increase (44). All evidences have proved that fructose is proposed to have adverse 253effects at very high or excessive doses. The mechanism of the cholesterol increase by 254fructose might be due to increased levels of advanced glycation end products, which 255cause damage to LDL and make it poorly recognized by lipoprotein receptors and 256scavenger receptors (45). Furthermore, excess exposure to fructose can damage the 257function of adipocytes and may reduce the recycling of cholesterol extracted from 258serum LDL. Studies have shown that elevated uric acid might contribute to LDL-C 259increases, and this effect can be reduced by allopurinol (46).260Based on the composition of added sugars in the United States where the fructose: 261glucose ratio is close to 0.43, and the NHANES 1999–2004 estimates (41), the 262increase of fructose consumption is always accompanied with an increase in total 263energy intake. Persons consuming >100g/d of sugars are potentially eating in excess 264of their energy requirement (47), and then overweight and obesity could result. So we 265can not suggest that it is safe to only limit fructose to <100g/d in coronary heart 266disease management and prevention. It may need to take into account the other 267components of foods that accompany the fructose. This dose threshold effects on TG 268and LDL-C can only help better inform nutritional guidance and avoid inappropriate 269marketing of carbohydrates.270Our meta-analysis did not show significant effect of fructose on HDL-C. 271However, Perez-Pozo et al (46) reported a significant HDL-C-lowing effect in 74 272adult men fed with 200g fructose/d in a randomized, 2-week crossover trial, 273suggesting that excessive fructose dose intake can also affect HDL-C. Further trials 274are needed to find the threshold of fructose on HDL-C.275There are several limitations to our work. First, many trials had a relatively small 276sample size, and most of them were funded by industry which can affect the quality of 277studies. Second, the change of fructose in the background diet can affect the practical 278utility of the outcomes of meta-analyses. However, most of trials used the background 279diet with ≤ 3% of total energy derived from fructose (15-23, 27-33, 35), others trials 280did not report the proportion of fructose in the background (24, 25, 26, 34). It was 281hard to make sure the dose of background fructose in every trial. Third, the data 282provided by Reiser et al (23) must be interpreted with caution. Although this study 283met all of our inclusion criteria, they choose a low P:S (polyunsaturated : saturated) 284rate of the fat as the background diet, which might change the metabolism of fructose 285as diets high in saturated fatty acids can enhance intestinal fructose absorption (48). 286Fourth, some of included trials lack test statistics, baseline values and SDs. We 287overcame these problems according to the methods proposed by Cochrane Handbook 288for Systematic Reviews of Interventions. Finally, it is difficult to differentiate effects of 289other modifiers, such as exercise and age, from those included trials. These factors can 290also influence the final result. Fructose can indeed be metabolized during exercise, 291and the rate of metabolism is different between exercise and sedentary lifestyle. Most 292of participants were requested to follow a designed regiment at home, but it is not 293easy to maintain the activity intensity. On the plus size, the age of participants in our 294meta-analysis ranged from 18 to 72 years old. Evidence from animal experiments 295shows that fructose absorption may affected by age, as older rats showed decreased 296fructose absorption (49). However, no human trial has been done to assess the 297difference in fructose effect among different age groups. Therefore, further studies 298should attempt to limit or isolate the degree of heterogeneity present in the study 299population to better assess the effect of age.300In conclusion, our meta-analysis shows that fructose used as a sweetener in 301isocaloric exchange for other carbohydrates has significant increasing effects on TC 302and LDL-C in individuals with very high fructose (>100g). This effect seems not to be 303dose-dependent when fructose is given at moderate or high dose of fructose (<100g). 304Further studies should concentrate on larger, longer and higher-quality human 305controlled feeding trials, which provide a better assessment of the effect of fructose on 306cholesterol.307Acknowledgements308Tao An, Rong Cheng Zhang and Jian Zhang designed the research; Tao An, Rong 309Cheng Zhang, Yu Hui Zhang and Jian Zhang preformed the research; Tao An and 310Rong Cheng Zhang summarized the data and had primary responsibility for the 311accuracy of the analysis; Rong Cheng Zhang wrote the manuscript. All the authors 312had full access to the data. None of the authors declared a conflict of interest.313。