Reports from the medical literature during the past 20 years have suggested that oral garlic supplementation may be effective in decreasing serum cholesterol levels by as much as 15% to 20%. These reports, however, have varied widely in terms of treatment effect, study design, types of preparations, dosage, and target populations. The three earliest reviews [2-4] suggested the need for further research to evaluate garlic's potential as a preventive agent in atherosclerotic cardiovascular disease. Subsequent reviews [5, 6] called the evidence for garlic's hypolipidemic effects "preliminary," criticized the methods of dietary garlic trials, and called the evidence "inadequate" to justify garlic supplementation. Since then, blinded, controlled trials have been reported, and the most recent review [7] claimed a favorable effect of garlic on blood lipids.

Although all previous reviews have suggested a beneficial effect, no review has summarized the size and consistency of garlic's effect on serum cholesterol levels. Given the potential benefit of oral garlic supplementation and the failure of reviews to provide a quantitative synthesis of the evidence, we conducted a meta-analysis of all randomized, placebo-controlled trials that tested the effectiveness of oral garlic preparations on study groups in which more than 75% of their participants had initial cholesterol levels exceeding 5.17 mmol/L (200 mg/dL). Our primary objective was to produce an overall estimate of treatment effect and to examine the variability of treatment effect across the combined studies.

Methods

Top Methods Results Discussion Author & Article Info References

Study Identification

The medical literature was searched to obtain all human studies that examined the effect of garlic on serum cholesterol. Using MEDLINE (1966 through 1991), we searched key words for the combinations "Garlic and Cholesterol" and "Garlic and Blood." Fourteen articles and six reviews were identified. Thirteen of 14 articles and all six reviews [2-7] were obtained. An additional 17 articles were identified from the bibliographies of the individual articles and from the reviews. Of these 17 articles, 15 were obtained. Of the initial 31 articles identified, 28 human studies that measured the effect of garlic on serum cholesterol were obtained. The three unobtained studies [8-10] were published but were not on record at the National Library in Washington, D.C.

Study Selection

One author reviewed all 28 studies [11-38] by extracting data and placing it in a tabular format by author, year, study design, target population, interventions, duration of therapy, and initial and final cholesterol measurements per group. These tables were reviewed by two authors, and inclusion and exclusion criteria were developed based on the similarity and comparability of study design, target population, initial cholesterol measurements, and research objectives.

A study was included in this meta-analysis if it was published, was randomized, was placebo-controlled, reported garlic's effect on cholesterol as a previous hypothesis, and had control and experimental groups whose initial total serum cholesterol levels exceeded 5.17 mmol/L (200 mg/dL) in 75% of participants. A study was excluded if it did not provide enough data to calculate an effect size, if its control medicine had lipid-altering effects, or if its design was to measure garlic's effect on alimentary lipemia (fat loading). Fourteen uncontrolled studies [25-38] were excluded. Of the remaining 14 studies that provided a control group [11-24], 9 [11-19] were excluded Table 1, leaving 5 studies [20-24] that fulfilled our criteria. Disagreements regarding criteria for admission to the trial were resolved by discussion and consensus.

Quality Score

All evidence of authorship, journal title, abstract, and references was removed from the five articles, and they were then photocopied in their revised form. A questionnaire was developed that evaluated the following criteria: patient selection, objective of study, randomization, blinding, description of therapeutic regimen, unintended intervention and ancillary therapy, accuracy and precision of cholesterol measurement, compliance, biologic equivalence, and adequacy of statistical analysis (see the Appendix). Using the questionnaire, a pair of resident physicians with at least 1.5 years of participation in our internal medicine journal club reviewed each article. Questions answered differently among the pairs were discussed, and answers were revised if an agreement was reached. Answers to questions that were in categoric format (Reported/Not Reported/Ambiguous) were later collapsed into two categories. A +1 value was assigned to the question if the information was clearly reported or assessed, and a 0 value was assigned if not. Each design item received a pair of scores, one from each reviewer, computed as the number of positive values divided by the total number of questions answered. The average of the pair, rounded to the nearest quarter point, served as the quality score assigned to each design item (1-point maximum per design item). A total score for each study was computed as the sum (10-point maximum per study). Agreement between reviewers was measured using a statistic [39].

Statistical Method for Meta-analysis

The statistical technique used to estimate the magnitude and direction of garlic's effect on serum cholesterol was the computation of effect size [40]. Effect size, in this analysis, was computed as the difference in the change in cholesterol level between the garlic and placebo groups. A negative effect size indicated a greater cholesterol reduction in the garlic-treated group compared with placebo. For each study, a treatment effect size and 95% confidence intervals (CIs) were calculated. Effect sizes determined to be homogeneous were pooled to estimate for a common treatment effect. Each study contributed to the pooled estimate a weight that was inversely proportional to its variance. Variance was defined as the square of the standard error (SE) difference of the cholesterol changes between garlic and placebo groups. The standard deviation (SD) terms for the change in cholesterol in trials 20 to 23 were estimated from the initial and final mean cholesterol SDs reported in each trial and from a computed correlation coefficient between initial and final cholesterol values obtained from the raw data provided in study 24 [41].

Appropriateness of Pooling Trial Results

Effect sizes were analyzed using a chi-square test of homogeneity to determine whether they estimated for a common treatment effect. This test assessed whether the distribution of effect sizes was compatible with the assumption that interstudy differences were attributable to random sampling alone. If all the studies were homogeneous (that is, if garlic's effect was consistent throughout all studies), this chi-square statistic would be nonsignificant.

Sensitivity Analysis

A sensitivity analysis was done by relaxing our inclusion criteria to include studies without adequate control groups that satisfied the other criteria. Eight trials [14, 17, 19, 29, 34-36, 38] satisfied the above criteria and were included in the pooled sensitivity analysis. Effect size estimates were calculated for these eight studies. Also, effect size estimates were calculated for the garlic treatment arms of each study included in the primary analysis. In this analysis, pretreatment mean cholesterol values were used as a proxy for control group performance. We assumed, for purposes of the sensitivity analysis only, that hypercholesterolemic patients would not improve if left untreated. Effect size, in this analysis, was the difference between the final mean and initial mean cholesterol levels. To compute a pooled estimate of treatment effect, studies were weighed by the inverse of the square of their SE difference between initial and final means. Tests of homogeneity and visual displays were done. Meta-analysis was done using only homogeneous trials.

Results

Top Methods Results Discussion Author & Article Info References

Study Characteristics and Methodologic Issues

Study characteristics of the five included trials [20-24] are summarized in Table 2. Four of five trials claimed double blindedness [20-23], and all five used a parallel group design. One study [23] crossed participants over to the opposite group. A total of 410 persons were enrolled in the trials. The total number completing the studies was 365, 184 in the garlic group and 181 in the placebo group. In four of five trials [20, 21, 23, 24], patients were selected from an outpatient setting. In one trial [22], the location of selection was not reported.

Studies varied in terms of oral garlic preparations. Three used Kwai powder tablets [20, 21, 23], one used a spray-dried powder [22], and one used an aqueous extract [24]. Doses varied from 600 to 900 mg/d for the powder preparations and 1000 mg/d "dry weight of the active garlic components" for the aqueous extract. The Kwai powders were reported to contain a standardized amount of alliin per tablet equivalent to 2.7 g of fresh garlic for the 900 mg/d dose [21] and 1.8 g of fresh garlic for the 600 mg/d dose [7]. Equivalence to fresh garlic was not reported for the spray-dried powder or for the aqueous extract. One clove is estimated to represent 3 g of fresh garlic [7].

All trials randomized participants to receive either an oral garlic preparation or placebo. Two trials [20, 24] claimed them to be identical in appearance, one claimed them to be similar in appearance [23], and two [21, 22] did not describe appearance. Four trials [20-23] made no mention of the smell of either the placebo or the garlic preparation. The Lau study [24] claimed that its aqueous extract was "odor modified."

None of the trials placed any dietary restrictions on participants. One trial [20] excluded patients receiving lipid-lowering drugs from statistical analysis; three trials [21-23] excluded patients who were receiving lipid-lowering drugs before the initiation of the trial; and one trial [24] stated that no participant was receiving any medications.

Mean patient age ranged from 49 to 58 years. Mean weight, reported in three of the five studies, averaged 77 kg [20], 75.5 kg [23], and 112% of ideal body weight [22] for the three studies, respectively. Of the 410 total patients enrolled, 185 were men and 255 were women. The proportion of men within each trial ranged from 42% to 56%. All trials measured total serum cholesterol and triglyceride levels. In addition, one trial [22] measured high-density lipoprotein cholesterol, and one trial [24] reported a mean ratio of low-density lipoprotein cholesterol to very-low-density lipoprotein cholesterol in their garlic-treated group.

Quality Scores

The overall quality scores for each study are summarized in Table 3. These scores ranged from 3.25 to 4.5 of a possible 10 points. Kappa values for individual studies ranged from 0.82 [23] to 0.98 [20, 22]. The overall value was 0.96.

Meta-analysis of Effect Sizes

Effect sizes were determined not to be homogeneous across all five studies (chi-square = 42, P < 0.001). A graphic display of effect sizes identified one outlier in the garlic-treated arm [21], and homogeneity was then shown across the four remaining trials (chi-square = 2.3, P > 0.2). The effect sizes from the four homogeneous trials were then pooled to estimate for a common effect size of treatment.

Study results of included trials are summarized in Table 4. The mean effect size of treatment for individual trials ranged from –0.16 to –1.37 mmol/L (–6 to –53 mg/dL) and were statistically significant in four trials [20, 21, 23, 24]. The pooled effect size estimate for the mean difference in cholesterol change between garlic and placebo groups was estimated at –0.59 mmol/L ( –23 mg/dL) (P < 0.001). The 95% CI was –0.44 to –0.74 mmol/L ( –17 to –29 mg/dL). Treatment effect sizes are shown in Figure 1.

View larger version (17K): [in this window] [in a new window]   Figure 1. Effect sizes with 95% CIs for mean improvement during garlic therapy compared with placebo. Effect size for each trial was computed as the difference between the change of cholesterol level in the garlic and placebo groups. The pooled difference was calculated using data only from homogeneous trials. Any CI that includes zero indicates a nonsignificant result.

Sensitivity Analysis

A total of 13 studies, controlled and uncontrolled, were included in this analysis. Again, effect sizes were determined not to be homogeneous (chi-square = 32.5, P < 0.01). A graphic display identified the same outlier [21] as in the primary analysis, and homogeneity was then shown across the remaining 12 trials (chi-square = 16.0, P > 0.10).

Meta-analysis of the 12 homogeneous trials estimated a significant pooled effect size of –0.86 mmol/L ( –33 mg/dL) after treatment with garlic. The 95% CI of this effect size was –0.74 to –0.99 mmol/L ( –29 to –38 mg/dL). Meta-analysis of the garlic-treated groups of the four homogeneous trials included in the primary analysis estimated a significant pooled effect size of –0.84 mmol/L ( –32 mg/dL) and a 95% CI of –0.63 to –1.05 mmol/L ( –24 to –41 mg/dL).

Discussion

Top Methods Results Discussion Author & Article Info References

Quantitative review of the international literature suggests that garlic decreases total serum cholesterol levels. Effect size in this analysis provided an estimate of the difference between the average improvement seen during garlic therapy compared with placebo.

Compared with placebo, garlic significantly lowered cholesterol levels by about 9% in the groups studied. This estimate is conservative. If we included the group that was excluded from analysis because of heterogeneity [21], the reduction in cholesterol would have been about 12%. The study by Vorberg and Schneider [21] rendered both our primary and sensitivity analyses heterogeneous due to its much larger effect size estimate. This trial's participants had the least variance in initial cholesterol levels, indicating less heterogeneity between participants, and may reflect the inclusion of fewer persons with type IV hyperlipidemia.

Another explanation for the large effect size reported by Vorberg and Schneider [21] may be that this study used the highest daily and cumulative dose of garlic. A dose-response relation, however, is difficult to establish due to the various garlic preparations and study durations. Standardization among dosage has been reported by one manufacturer, Lichtwer Pharma. This manufacturer standardizes its Kwai powder preparation to percentage content of alliin, the main amino acid precursor to the volatile sulfur compounds of garlic. In this manner, one can compare the Kwai preparation with raw garlic [7, 42]. We estimate that the Kwai trials tested the equivalent of approximately one half to one clove of fresh garlic per day. Our analysis would, therefore, suggest that only a small amount of garlic ingestion per day is needed to produce a hypocholesterolemic effect.

These results are consistent with animal studies reporting a hypocholesterolemic effect of garlic [43-47]. Reported mechanisms include reduced hepatic 3-hydroxy-3-methylglutaryl coenzyme A reductase activity [44, 47] and increased bile acid excretion [45].

Although our data support the claim that oral garlic therapy decreases serum cholesterol levels in persons with increased levels, the quality of the included trials was not optimal. None of the studies reported the way in which patients were recruited; enzymatic methods of cholesterol analysis were only reported in two trials [22, 24]; and in only one trial was the smell of the oral garlic preparation reported. In no trial was the similarity in taste or smell of the placebo to the garlic preparation reported. In no trial was dietary intake, weight, or exogenous garlic ingestion assessed within or between groups during the trial period.

This meta-analysis combined persons with both primary and secondary hyperlipidemia as well as those with hyperlipidemia of varying definition. Although the groups studied represented a wide variety of lipoprotein abnormalities, they reflect the broad spectrum of patients with borderline and high cholesterol levels who present to a doctor's office. Despite this heterogeneity, the consistent results suggest that garlic may act on some important step of cholesterol biosynthesis, degradation, transport, or uptake, common to many forms of hyperlipidemia.

Further, our sensitivity analysis showed that the inclusion of methodologically less rigorous trials did not appreciably alter the magnitude of garlic's effect. The sensitivity analysis also showed that garlic's effect was consistent across trials using different methods, preparations, and groups of participants.

Although we detected a significant hypocholesterolemic effect, the overall quality of the included studies weakens the validity of our findings. Improved studies should incorporate placebos that look, taste, and smell like their garlic counterparts. Dietary habits and unintended interventions should be monitored through the study period to detect any inequalities between treatment arms. Further, because the low-density lipoprotein subfraction has been shown to be more precise than total cholesterol as a marker of cardiovascular risk [48], future studies should use it as the primary outcome measure.

Meta-analysis of the controlled trials of garlic to reduce hypercholesterolemia showed a significant reduction in total cholesterol levels. The best available evidence supports the use of garlic as one modality to decrease cholesterol levels in patients with increased levels.

Appendix: Topics Included in the Quality Score Questionnaire

The quality score questionnaire included questions pertaining to each of the following design items. The number of questions about each item is noted in parentheses.

I. Patient selection [19]

Reporting of methods for recruitment [5].

Reporting of eligibility criteria [3].

Reporting of exclusion criteria [5].

Reporting of participants screened for eligibility [3].

Obtainment of informed consent [1].

Reporting of prerandomization visits [2].

II. Objective [3]

Clarity of stated objective [1].

Assessment of cholesterol measurement as a primary objective of the study [2].

III. Randomization [13]

Application of inclusion criteria and exclusion criteria before the allocation of treatment [2].

Reporting of randomization and method [3].

Assessment of randomization blindness [1].

Reporting of dropouts, reasons, and method of handling them for statistical analysis [5].

Comparison of baseline pretreatment variables after randomization [2].

IV. Blinding [10]

Reporting of participant blindness to treatment and method [4].

Assessment of participant blindness to treatment [2].

Reporting of physician blindness to treatment [1].

Assessment of physician blindness to treatment [1].

Assessment of participant and physician blindness to ongoing study results [2].

V. Description of therapy [8]

Description of oral garlic preparation, dosage, timing of daily therapy, and duration of treatment [5].

Similarity of placebo in appearance, taste, and dosage [3].

VI. Unintended intervention and ancillary therapy [6]

Reporting of and measurement of confounders during therapy (smoking, diet, exercise, weight loss, lipid-altering medications, exogenous garlic ingestion) [6].

VII. Accuracy and precision of cholesterol measurement [14]

Reporting of multiple cholesterol measurements per observation period [2].

Reporting of instrumentation, calibration, and enzymatic methods [3].

Reporting of laboratory(s) used [2].

Reporting on method of blood collection and personnel used [3].

Reporting on specific time periods for cholesterol sampling [4].

VIII. Assessment of compliance [2]

Reporting of participant compliance [1].

Method of assessment [1].

IX. Biologic equivalence [1]

Blood or urine measurement of therapeutic agent [1].

X. Statistical analysis [10]

Reported method of statistical analysis for cholesterol measurements [1].

Appropriate use of statistical method [2].

Reported probability or P value [1].

Reported power of study or previous estimation of Beta [1].

Used statistical methods to identify and control for nondrug influences on treatment results [2].

Reported side effects and applied statistical tests to determine their significance [3].