Total cholesterol (TC) measurements are comprised of three different types of lipoproteins: LDL, VLDL and HDL. LDL and VLDL are apoB100-containing particles and are both positively associated with atherosclerosis while HDL is thought to be cardioprotective in higher concentrations. In addition to promoting atherosclerosis, markedly elevated TAG increases the risk of developing pancreatitis and lipomas. While several pharmacological agents are available to reduce LDL (statins, bile acid sequestrants, niacin, ezetimibe) and TAG (statins, fibric acid derivatives, niacin), not all patients can achieve current National Cholesterol Education Program Adult Treatment Panel III goals(Reference Grundy, Cleeman and Merz1) with available pharmacotherapy while others have trouble tolerating therapy or may seek more natural therapies. As such, patients and clinicians are interested in nutraceuticals that can potentially modulate plasma lipids.
Garlic (Allium sativum) is a nutraceutical touted to lower cholesterol. Previous meta-analyses have demonstrated a decrease in TC although newer meta-analyses have showed smaller reductions with the incorporation of newer studies(Reference Stevinson, Pittler and Ernst2–Reference Warshafsky, Kamer and Sivak5). As such, with additional studies evaluating the lipid-modulatory effects of garlic being published, it will be important to discern whether the literature still supports garlic's ability to reduce TC. Since it is the individual constituents of TC, more so than TC itself, which is more prognostic for the risk of CHD, we will also determine the impact of garlic on LDL, HDL and TAG.
Methods
Data sources
We conducted a systematic literature search of MEDLINE, CINAHL and the Cochrane Database from the earliest possible date through to November 2007. We used the following medical subject headings (MeSH) and keywords: ‘garlic’, ‘Allium sativum’ or ‘allicin’. Results were limited to randomised clinical trials in human subjects. A manual search of references from reports of clinical trials, review articles and previous meta-analyses was performed to identify additional relevant studies.
Study selection
According to pre-specified criteria, trials must have utilised garlic, reported TC, LDL, HDL or TAG endpoints and have been published in English or German to be evaluated for exclusion criteria. Studies were excluded if they were not randomised, placebo-controlled and double-blind, combined garlic with other nutraceuticals or products with lipid-altering effects, or lacked sufficient data to allow for statistical pooling. Both parallel and cross-over trials were eligible for inclusion. If cross-over trials did not include a washout period, they were still included in the analysis if the follow-up after the intervention and control was each at least 3 weeks, at which time lipid values would be considered stable(Reference Kris-Etherton and Dietschy6). Two authors (K. M. R. and R. T.) independently reviewed abstracts or full publications of all identified citations to evaluate them for inclusion and exclusion criteria with disagreement resolved by discussion or evaluation by other investigators (C. M. W. and C. I. C.).
Data abstraction
Two investigators (K. M. R., R. T.) utilised a standardised data abstraction tool and independently abstracted data with disagreement resolved by discussion. The following information was sought from each trial: author identification, year of publication, study design, study population, sample size, type of garlic used, brand of garlic used, duration of treatment, required TC and TAG levels at baseline, Jadad score for quality assessment(Reference Jadad, Moore and Carroll7) and whether or not dietary counselling was given or if patients were required to maintain a low-fat–low-cholesterol diet. Study endpoint data extracted included baseline and final TC, LDL, HDL and TAG concentrations and measures of variance for these variables at these time points. Any discrepancies in extracted data between reviewers were resolved through discussion. For TC, LDL and HDL values reported in mg/dl, the corresponding mmol/l value was calculated by multiplying the value by 0·0259. Any TAG levels given in mg/dl were multiplied by 0·0113 to derive the mmol/l value.
Data synthesis and analysis
The mean change in lipid parameters from baseline was treated as a continuous variable and the weighted mean difference was calculated as the difference between the means in the garlic and control groups. A random-effects model was used in calculating the weighted mean difference and its 95 % CI. For parallel trials, net changes in each study parameter were calculated as the difference (garlic minus control) of the changes (follow-up minus baseline) in the mean values. For cross-over trials, net changes were calculated as the mean difference in values at the end of the garlic and control periods. If variances for net changes were not reported directly, they were calculated from CI, P values or individual variances for intervention and placebo groups/periods. For parallel trials in which variance for paired differences were reported separately for each group, we calculated a pooled variance for net change by standard methods(Reference Follmann, Elliott and Suh8). When the variance for paired differences was not reported, we calculated it from variances at baseline and at the end of follow-up using a correlation coefficient of 0·4 as suggested by a previous garlic meta-analysis(Reference Stevinson, Pittler and Ernst2). Weighted mean differences and 95 % CI were calculated using StatsDirect (version 2.6.2; StatsDirect Ltd, Sale, Cheshire, UK) using a random-effects model.
Many sensitivity and subgroup analyses were conducted. Studies of poorer methodological quality may exhibit inaccurate treatment effect. Limiting the meta-analysis to only higher-quality trials may result in increased internal validity but could reduce its external validity (applicability). As such, we conducted a sensitivity analysis excluding studies of poorer methodological quality (Jadad score less than 3). Since the exact time course of garlic's proposed lipid-lowering effects is not known, studies of longer duration might have yielded a greater change in lipid parameters. As such, we examined studies of 12 weeks or less in duration and studies lasting for longer than 12 weeks separately. Since there are several manufacturers of garlic products and garlic is not regulated by the Food and Drug Administration, there is likely to be variability in the quality of products between companies. Therefore, we performed subgroup analysis whereby we excluded all brands except Kwai® (Lichtwer pharma GmbH, Berlin, Germany), the most studied brand of garlic in the medical literature. Likewise, there exist different preparations of garlic which can potentially deliver differing amounts of potentially active constituents of garlic to the bloodstream and so we performed subgroup analyses based on the type of product including garlic powder, garlic oil and aged extract preparations. Patients with larger TC concentrations might be more amenable to lipid alterations than those with near-normal baseline TC levels. However, subjects with very high TC might have genetic disorders that preclude lipid-modulating therapy from being fully effective. As such, we performed subgroup analysis where we calculated garlic's effects based on the mean baseline TC of the patients being studied. Studies with one or more treatment groups exhibiting a baseline cholesterol < 6·48 mmol/l (250 mg/dl; 2500 mg/l) were analysed separately from the studies where the mean value was ≥ 6·48 mmol/l; studies with one or more treatment groups exhibiting baseline TAG < 1·7 mmol/l (150 mg/dl; 1500 mg/l) were analysed separately from the studies where the mean value was ≥ 1·7 mmol/l. It is possible that the effect of a lipid modulator could be dependent on the dietary intake of fats and cholesterol. Studies with dietary counselling or dietary restrictions in addition to garlic therapy might yield different results from studies without such counselling or restrictions. As such, we performed subgroup analyses whereby we evaluated studies which required dietary modification or counselled patients on low-fat and low-cholesterol diets independently from the studies where this was not part of the methodology. For each main analysis, outlier trials, defined as those whose 95 % CI did not cross the pooled effect, were removed and analysis was performed to evaluate their impact on the observed effects.
Statistical heterogeneity was calculated using the I 2 statistic. An I 2 of greater than 25 % was considered to be representative of a moderate degree of statistical heterogeneity and a value greater than 75 % was considered a high degree of statistical heterogeneity. Funnel plots and Egger's weighted statistic were used to assess for publication bias. Duval and Tweedie's trim and fill plots were used to determine how any publication bias would affect the results. Trim and fill plots were created using MIX software (version 1.54; freely accessible at http://www.mix-for-meta-analysis.info).
Results
The initial search yielded 2958 results and Fig. 1 describes how many studies were excluded, for what reasons and how many studies were ultimately included. In brief, 172 citations represented controlled trials in human subjects published in English or German, and forty-nine trials reported one or more of the lipid endpoints. Twenty studies met exclusion criteria, thus twenty-nine studies were eligible for analysis(Reference Adler and Holub9–Reference Plengvidhya, Sitprija and Chinayon37). Reasons for exclusion included studies that were not randomised(Reference Bordia, Verma and Srivastava38–Reference Brosche, Platt and Dorner40), not placebo-controlled or double blind(Reference Zimmermann and Zimmerman41–Reference Ashraf, Aamir and Shaikh48), garlic not given as a monopreparation(Reference Gardner, Chatterjee and Carlson49), concurrent use of medications known to increase serum cholesterol(Reference Lash, Cardoso and Mesler50) or studies lacking enough data for statistical pooling(Reference Van Doorn, Santo and Meijer51–Reference Jung and Kiesewetter56). One study reported results of selected patients from another trial and was therefore excluded(Reference Byrne, Neil and Vallance57). Of note, in one included study(Reference Zhang, Lowe and Giles30), one investigation centre gave all patients garlic powder in a non-randomised fashion and therefore data for these patients were not included in the meta-analysis. Another study by Rahmani et al. had three active treatment arms of increasing doses of a garlic powder preparation (Garlet; Kowsar Pharmaceutical Co., Tehran, Iran)(Reference Rahmani, Tabari and Niaki22). While this study did not report the allicin content of this preparation, another study using Garlet reported that each tablet contained 1 mg allicin(Reference Williams, Sutherland and McCormick36). Therefore, only the 1200 mg garlic powder treatment arm was included as it provided an amount of allicin closest to the studies using the Kwai® brand garlic powder (3 mg allicin for Garlet, 1·8 mg allicin for Kwai®). Seven studies utilised a cross-over design. In general their design and washout period varied dramatically, ranging from 0 to 4 weeks of washout. We therefore conducted sensitivity analysis to evaluate their impact(Reference Phelps and Harris31–Reference Plengvidhya, Sitprija and Chinayon37). Finally, in three of the included cross-over studies, data were presented in two mutually exclusive groups of patients and therefore each study was broken down into two parts named study (a) and study (b)(Reference Steiner, Khan and Holbert34, Reference Tanamai, Veeramanomai and Indrakosas35, Reference Plengvidhya, Sitprija and Chinayon37).
Fig. 1 Flow diagram of trial identification, inclusion and exclusion.
Of the twenty-nine included studies(Reference Adler and Holub9–Reference Plengvidhya, Sitprija and Chinayon37), twenty-eight studies included TC endpoints (n 1262)(Reference Adler and Holub9–Reference Rahmani, Tabari and Niaki22, Reference Santos and Grünwald24–Reference Plengvidhya, Sitprija and Chinayon37), twenty-one studies included LDL endpoints (n756)(Reference Adler and Holub9, Reference Budoff, Takasu and Flores12–Reference Kannar, Wattanapenpaiboon and Savige15, Reference Macan, Uykimpang and Alconcel17, Reference McCrindle, Helden and Conner19–Reference Rahmani, Tabari and Niaki22, Reference Santos and Grünwald24, Reference Satitvipawee, Rawdaree and Indrabhakti26, Reference Superko and Krauss27, Reference Zhang, Lowe and Giles29–Reference Williams, Sutherland and McCormick36), twenty-four studies included HDL endpoints (n 845)(Reference Adler and Holub9, Reference Bordia11–Reference Kannar, Wattanapenpaiboon and Savige15, Reference Macan, Uykimpang and Alconcel17, Reference McCrindle, Helden and Conner19–Reference Santos and Grünwald24, Reference Satitvipawee, Rawdaree and Indrabhakti26, Reference Superko and Krauss27, Reference Zhang, Lowe and Giles29–Reference Plengvidhya, Sitprija and Chinayon37) and twenty-eight studies included TAG endpoints (n 1220)(Reference Adler and Holub9–Reference Kannar, Wattanapenpaiboon and Savige15, Reference Macan, Uykimpang and Alconcel17–Reference Plengvidhya, Sitprija and Chinayon37). Characteristics of the included trials are given in Table 1.
Table 1 Characteristics of included studies
TC, total cholesterol; GP, garlic powder; GO, garlic oil; AGE, aged garlic extract; SDL, spray-dried garlic; GE, garlic extract.
The results of the meta-analysis of trials evaluating TC are displayed in Fig. 2. Overall, garlic was found to significantly reduce the levels of TC by 0·19 (95 % CI − 0·33, − 0·06) mmol/l (7·35 mg/dl; 73·5 mg/l) compared with placebo. There was a moderate amount of statistical heterogeneity among the studies (I 2 64·3 %) and four studies reported 95 % CI that did not cross the pooled effect. When these outlier trials were excluded, TC was reduced by a more moderate 0·12 (95 % CI − 0·21, − 0·04) mmol/l but still to a significant degree, and statistical heterogeneity was reduced (I 2 2·9 %). The Egger's weighted statistic (P = 0·37) suggested that publication bias was unlikely and the trim and fill plot imputed no additional studies minimizing the possibility that publication bias had an impact on results. When cross-over studies were excluded from the analysis, the decrease in TC was slightly greater at 0·27 (95 % CI − 0·45, − 0·09) mmol/l.
Fig. 2 Forest plot results of total cholesterol meta-analyses. Impact of garlic on total cholesterol as compared with placebo (mmol/l). (—), No effect of garlic on total cholesterol compared with placebo. Trials to the left of this show the further reductions in total cholesterol over and above that seen with placebo while trials to the right show the opposite effect. (- - -), Combined effect. The I 2 value for statistical heterogeneity was 64 %.
The results of garlic's effects on TAG are displayed in Fig. 3. Garlic significantly lowered TAG levels by 0·11 (95 % CI − 0·19, − 0·03) mmol/l (9·7 mg/dl; 97 mg/l) compared with placebo. There was a moderate degree of statistical heterogeneity among the studies (I 2 26·7 %). Visual inspection elucidated three studies which had 95 % CI that did not cross the pooled effect. When these outlier trials were excluded from analysis, TAG were reduced by 0·11 (95 % CI − 0·17, − 0·04) mmol/l, the same effect observed with inclusion of the outlier studies. Again statistical heterogeneity was reduced (I 2 0 %). The Egger's weighted statistic (P = 0·87) suggested that publication bias was unlikely and trim and fill plot imputed no additional studies minimising the possibility that publication bias had an impact on results. When cross-over studies were excluded from the analysis, the decrease in TAG was 0·15 (95 % CI − 0·26, − 0·03) mmol/l, similar to the base analysis.
Fig. 3 Forest plot results of TAG meta-analyses. Impact of garlic on TAG as compared with placebo (mmol/l). (—), No effect of garlic on TAG compared with placebo. Trials to the left of this show the further reductions in TAG over and above that seen with placebo while trials to the right show the opposite effect. (- - -), Combined effect. The I 2 value for statistical heterogeneity was 27 %.
Garlic did not exhibit significant effects on either LDL or HDL concentrations, although very modest qualitative reductions in LDL and elevations in HDL were noted (Table 2). A minor degree of statistical heterogeneity was calculated among studies used for these endpoints (I 2 < 24 % for both). A low likelihood of publication bias was noted (Egger's weighted statistic, P>0·18 for both).
Table 2 Main results of the meta-analyses (Effect sizes and 95 % confidence intervals)
Several subgroup analyses found different magnitudes of effect with garlic therapy v. placebo (Table 3). Garlic's effect on TC was found to be greater in studies of 12 weeks or longer and with garlic powder, Kwai®, higher baseline TAG levels, and when diet control was not used. Garlic's effect on TAG was greater in studies of lower Jadad score, and when diet control was not used. No large differences in effects on LDL or HDL were noted for any subgroup analyses.
Table 3 Subgroup and sensitivity analyses results (Effect sizes and 95 % confidence intervals)
TC, total cholesterol.
* The calculation could not be made due to the number of studies in the analysis (n 2).
† Diet control indicates either diet counselling used or diet implemented during study.
Discussion
The 0·19 and 0·11 mmol/l reductions in TC and TAG reductions in the present meta-analysis were modest in size but achieved statistical significance. This is far less than prescription products such as statins (TC reductions of 35–55 % and TAG reductions of up to 28 %)(Reference Rosenson58), niacin (TC reductions of 30 % and TAG reductions of 55 %)(Reference White59) and fibrates (TC reductions of 10 %, TAG reductions of 45 %)(Reference White59). To compare, the effect of garlic observed in this analysis would equate to a 4 % reduction in TC in a patient with a baseline TC of 5 mmol/l. A patient with a baseline TAG of 2 mmol/l would have a 22 % reduction in TAG with garlic. While LDL was qualitatively reduced and HDL elevated with garlic therapy, these effects were very small ( − 0·05 mmol/l and +0·025 mmol/l, respectively) and did not achieve statistical or clinical significance.
While the TC reduction in the present study was less than the 0·41 mmol/l reduction in the previous meta-analysis by Stevinson et al. (Reference Stevinson, Pittler and Ernst2), we allowed inclusion of studies that enrolled subjects without having hyperlipidaemia. We did this because the product is touted for general heart health promotion and including a greater number of trials increases the power of the meta-analysis to show a difference if one truly exists. We, however, did conduct subgroup analysis of studies with a patient population exhibiting hypercholesterolaemia. The reduction in TC in our hypercholesterolaemic subgroup ( − 0·24 mmol/l) was closer to that of Stevinson et al. (Reference Stevinson, Pittler and Ernst2). The smaller effect in our analysis could be due to the inclusion of newer studies exhibiting a more modest effect. Like Stevinson et al. (Reference Stevinson, Pittler and Ernst2), our analysis only included randomised, double-blind and placebo-controlled trials.
An important finding of our meta-analysis was that garlic reduced TC concentrations modestly with little impact on the LDL and HDL component of TC. It is possible that the reductions in the VLDL component of TAG may have provided the majority of the TC reductions seen in our meta-analysis. This is supported by a qualitative analysis that we conducted which was limited to the four trials that provided TC, LDL, HDL and TAG concentration data. In this qualitative analysis, the TC was decreased by 0·05 and TAG was decreased by 0·08 mmol/l.
We believe that garlic's TAG effects are important because even though they were modest in size, they did not result in the elevation of LDL-cholesterol-like drugs such as gemfibrozil. Garlic has been compared head to head with bezafibrate and similar reductions in TAG as well as other lipid parameters were observed(Reference Holzgartner, Schmidt and Kuhm60). In this study, ninety-eight patients were randomised to garlic (given as 900 mg Kwai®) or 600 mg bezafibrate for 12 weeks and lipid parameters were evaluated. Both bezafibrate and garlic significantly reduced TAG by roughly 0·85 mmol/l (750 mg/l) and 0·81 mmol/l (720 mg/l) respectively (standard deviations not reported) without raising LDL. Bezafibrate did, however, trend towards lowering TAG more than garlic when analysed in patients with baseline TAG greater than 2·8 mmol/l (42 v. 29 % respectively). While we did not include this study in our meta-analysis as it was not placebo controlled, this trial was in a population with hypertriacylglycerolaemia and found much more robust TAG reductions than in our meta-analysis. If garlic can reduce TAG in patients with elevated TAG concentrations without elevating LDL and can work synergistically with fibrates or niacin, it could be an important adjunctive therapy. As such, more study in this area is warranted.
Both the TC and TAG analyses had a moderate amount of statistical heterogeneity, which is not surprising given the variety of garlic preparations used and the diversity of study designs being employed. We were not able to discern noticeable characteristic differences between the outlier studies and the other studies in our analyses. The results of several subgroup and sensitivity analyses warrant discussion. First, the effects of garlic on TC and TAG concentrations appear to be time dependent. Studies lasting greater than 12 weeks tended to have greater effects than those of shorter duration. Garlic powder formulations were superior to other formulations (garlic oil and aged extracts) in terms of reducing TC and TAG but a large constituency of the garlic powder studies used the Kwai® brand. We found that the Kwai® brand provided more robust effects than other brands. We cannot discern from this meta-analysis whether it the garlic powder formulation or some type of quality control or manufacturing feature of Kwai® specifically that is responsible for this difference. Finally, studies employing dietary modification to both the garlic and placebo groups showed markedly attenuated effects with garlic v. studies that utilised garlic or placebo with no such dietary modification. Diet modification was considered to be either dietary counselling or mandatory diet restrictions. The TAG levels were only reduced by 0·03 mmol/l with dietary control relative to placebo (although not significantly) compared with reductions of 0·18 mmol/l in studies without diet control. It is possible that garlic could be reducing TC and TAG by blocking the absorption of dietary cholesterol and fatty acids and the use of dietary modification attenuates the impact of this blockade. This subgroup did, however, have a small but significant increase in HDL, arguing against this theory. Other mechanisms that have been proposed include reducing the activity of 3-hydroxy-3-methylglutaryl coenzyme A reductase and increased bile acid excretion(Reference Warshafsky, Kamer and Sivak5, Reference Isaacsohn, Moser and Stein13). Nonetheless, this discrepancy shows that garlic supplementation may not be appropriate for patients committed to improving their diet, but may be helpful for those who are not compliant with low-cholesterol diets. Randomised, controlled trials could better establish this observation.
There are limitations to the present study. First, while we only included double-blind studies in an attempt to increase the internal validity of the present results, many studies did not describe how patients were blinded. While in many types of studies it may be adequate to assume proper blinding even when not described, this is not possible with a preparation such as garlic as its scent may be difficult to mask. Some studies did report the use of placebo tablets sprayed with a small amount of garlic in order to prevent unblinding but many others did not. There is a possibility that studies of inadequate blinding were included. None of the studies in our analysis used raw, unprocessed garlic, as blinding would be difficult if not impossible. Second, there are inherent limitations in meta-analysis in that the combining of heterogeneous studies limits validity. This may be particularly true for meta-analyses of garlic, as the wide variety of garlic preparations used, the different baseline risks and the different study procedures employed could increase the chances of seeing heterogeneous effects. However, we did employ subgroup analyses to probe for potential sources of heterogeneity.
Conclusion
Garlic reduces TC to a modest extent, an effect driven mostly by the modest reductions in TAG, without appreciable LDL lowering or HDL elevation. Patients taking Kwai® brand garlic, without dietary modification or counselling, appear to have greater lowering of TAG. Future studies should be conducted evaluating the impact of adjunctive garlic therapy with fibrates or statins on TAG concentrations.
Acknowledgements
K. M. R. was responsible for study design, data collection, statistical analysis and writing of the manuscript. R. T. was responsible for data collection. C. M. W. was responsible for preparation of the manuscript. C. I. C. was responsible for study design, statistical analysis and writing of the manuscript.
The authors of this manuscript have no conflicts of interest, financial or other, to disclose.
