Supplements · Metabolic & Cardiometabolic
garlic decreases blood pressure
In plain terms: Does garlic lower blood pressure?
Strong support Supplements 💰 Industry COI noted
Part of: 🧪 Garlic
RefutedContestedStrong support
consensus score 0.94
Yes, modestly — and mainly if your blood pressure is already high. Several meta-analyses show garlic drops systolic pressure by roughly 5–9 mmHg in people with hypertension, but little in those with normal pressure. It's a mild helper, not a replacement for medication.
📅 Last reviewed: 2026-07-15 ⓘ
Evidence ladder
How far up the ladder this claim has climbed. A high consensus on a low rung means "consistent so far," not "proven in people."
Top evidence so far: All trials, pooled (Meta-analysis)
MechanismIn-vitroAnimalObservationalRCTMeta-analysis
How the studies fall
14 support 0 contradict 0 tested null 2 mixed · 16 sources, 12 independent groups · 2 superseded — pooled inside a review, counted once
What the evidence shows
Garlic (esp. aged garlic extract) modestly lowers blood pressure, with the effect concentrated in people who already have hypertension; normotensive subgroups show little to no change. Three independent meta-analyses agree on direction.
The evidence (18)
| Source | Grade | Stance | Quality | Finding |
|---|---|---|---|---|
| Ma X et al 2025 · Asian Biomed (Res Rev News) | meta-analysis | supports | moderate | Full-text re-read 2026-08-25 (PMC JATS XML, fulltext-cache/xml/40735665.xml, 87,107 bytes; the four article figures from the 2026-08-16 paper drop). Ma X, Zhang H, Jia J. Asian Biomedicine (Res Rev News) 2025;19(3):131-140, doi 10.2478/abm-2025-0016. STANCE supports and grade meta-analysis both STAND; ROLE stays counting; quality lowered high -> moderate on the six measured grounds below. THE RESULT AND ITS DENOMINATOR. 'Twelve reports, comprising 405 patients treated with garlic derivatives and 333 receiving placebo, were included in this investigation.' Random-effects pooling of change-from-baseline differences (pre and post means in both arms were the extracted quantities): SBP 'difference in mean: -8.121, 95% CI: -10.95 to -5.28, P = 0.000'; DBP '-4.256, 95% CI: -5.99 to -2.52, P = 0.000'. Both estimates rest on the FULL 12-trial pool - no subset is involved, so the prior extract's denominator was correctly attached. But the headline denominator does not reconcile with the paper's own table: summing the per-arm counts printed in Table 1 gives 358 garlic and 286 placebo, so the reported 405/333 exceeds the tabulated total by exactly 47 in EACH arm, and no thirteenth trial appears in Table 1 or in the Figure 2B forest plot (which lists the same 12 studies). Where the extra 47+47 come from is cannot-tell from the text. GROUND 1 - the paper contradicts itself on publication bias, unreconciled. Results: 'the Eggers regression analysis indicated the absence of publication bias in the included reports (SBP: intercept = -1.22, 95% confidence interval [CI] = -3.69-1.23, P = 0.29; DBP: intercept = -1.38, 95% CI = -3.99-1.21, P = 0.26).' Limitations, four paragraphs later: 'Third, there was significant publication bias in both the SBP and DBP analysis groups.' A clean Egger test is the single strongest 'high' signal this design can offer (rubric axis 5, the Ding-2014 precedent), and here the authors themselves contradict it; a high cannot rest on it. GROUND 2 - GRADE and RoB 2 are asserted but never reported. The abstract states 'the GRADEing of evidence also supports the robustness of the observations', yet the paper's entire GRADE section is one sentence naming the software - 'The GRADE Pro GDT software was utilized to evaluate the strength of the evidence derived from the meta-analysis of RCTs, with the aim of determining the significance of garlic extracts in reducing blood pressure among patients with hypertension.' - with no certainty rating and no summary-of-findings table. RoB 2 is described at length in Methods with no per-domain and no overall judgement anywhere. This is a FULL-TEXT observation (the four supplied figures are funnel/forest/sensitivity/TSA only and PMC records no supplement), not an abstract-level cannot-tell. GROUND 3 - heterogeneity, and a model rule the paper did not follow. 'The investigation revealed heterogeneity in SBP ( Q = 21.22, P = 0.03, I 2 = 48.17) and DBP ( Q = 24.56, P = 0.01, I 2 = 55.22).' Random effects was used for both, although their own pre-stated rule requires P < 0.05 AND I2 > 50 and SBP's 48.17 does not meet it. The authors list heterogeneity as a limitation (different populations, extract types, active-compound quantities, ethnicities and BP measurement methods). GROUND 4 - search. Three databases only ('First, the search was limited to 3 databases, meaning that articles from other databases may have been excluded.') - PubMed, ScienceDirect, Scopus, no CENTRAL and no Embase - English-only, and 'The last database search was performed on November 25, 2023.' for a 2025 paper titled 'updated'. GROUND 5 - what the pool actually is. Inclusion criterion (i) is '(i) only placebo-controlled trials were considered; (ii) all subjects had hypertension;' - placebo-controlled, NOT randomised - while the abstract and GRADE section both describe the pool as RCTs. Four of the 12 predate 1995 and two sit in grey venues (De Santos & Gruenwald, Br J Clin Res 1993; Kandziora, Arztl Forsch 1988, carried in the reference list as '[in German]', which the paper's own English-only exclusion should have removed). GROUND 6 - the effect sizes are large for this literature (-8.1/-4.3 mmHg against roughly -5 to -8 SBP in the sibling syntheses) and the two heaviest single contributors on the SBP forest plot are the two oldest trials (Kandziora 1988, 12.15% weight) and Nakasone 2013 (13.93%), a preHYPERTENSIVE-to-mildly-hypertensive Japanese trial. INDEPENDENCE - FLAGGED, NOT FIXED (rubric axis 7; a supersedes or role edit moves this claim's denominator mid-corpus). Included-trial list, read off Table 1 and confirmed against the Figure 2B forest plot: Kandziora 1988, Auer 1990, Vorberg & Schneider 1990, De Santos & Gruenwald 1993, Sobenin 2008, Sobenin 2009, Ried 2010, Ried 2013, Nakasone 2013, Ried 2016 (AGE at Heart), Ried 2018 (GarGIC), Serrano 2023. SIX of the 12 come from TWO groups - four Ried/NIIM Australian trials and two Sobenin/Moscow trials - holding 33.55% and 10.70% of the SBP relative weight respectively (44.25% combined, figure-derived). FOUR of the 12 also vote separately in this vault as counting RCT rows: garlic-ried-2010, garlic-ried-2012 (= Ried 2013 Eur J Clin Nutr, epub Nov 2012), garlic-sobenin-bp-2009 and garlic-serrano-2023. The source note's declares the first three and NOT garlic-serrano-2023 - which CONFIRMS, from the included-trial table, the flag raised at sibling reread #645. Sibling #644's question - whether these 12 are a superset of Rohner 2015's 9 - is cannot-tell from this text: Rohner 2015 appears here only as a cited prior meta-analysis (ref [10]) and its trial list is not reproduced; note that Rohner's author line carries both Ried K and Sobenin IA, the same two networks that supply half this pool. Claim-level: 18 rows, of which ~10 are meta-analyses re-pooling one shared garlic-trial literature. STANCE re-examined against mixed and HELD: every reported result points the same way on the claim's object, and the TSA conclusion is directional - 'a sufficient number of trials with sufficient subjects were included to draw a definitive conclusion on the role of garlic in lowering blood pressure in hypertensive patients'. (Note that Serrano 2023, which this vault grades mixed on its own, contributes -3.000 at P = 0.437 here - a null inside a positive pool.) ROLE re-examined: the units pooled are PRIMARY TRIALS, not syntheses, so this is not an umbrella review and stays counting (rubric axis 2). Moderate lands this row level with every sibling synthesis on this claim already re-read: Silagy 1994, Reinhart 2008, Ried 2008, Kwak 2014, Wang 2015, Rohner 2015, Xiong 2015 and Ried 2016 all sit at moderate on their own measured grounds. FUNDING/COI, stated: 'This work was supported by the Research Fund Project of Hebei Provincial Health and the Health Commission Youth Science and Technology Project.' and 'All authors state no conflict of interest.' |
| Ojo GB et al 2025 · bioRxiv | animal | supports | low | Preprint (bioRxiv v1, not peer-reviewed), animal. 'Forty-eight Wistar rats were divided into groups: control, hypertensive, and hypertensive rats treated with different doses of garlic extract or a standard antihypertensive drug' - the 48 spans at least five arms and the abstract gives no per-arm n, so no comparison here rests on a denominator of 48. 'Garlic treatment significantly lowered blood pressure compared to the hypertensive control group' - the reported contrast is against hypertensive controls, NOT head-to-head against the standard drug, and the abstract reports no blood-pressure values, effect sizes or p-values. Kidney tissue analysis revealed 'potential renoprotective effects' (the authors' own hedge). The model chemical is internally inconsistent: the title says cadmium chloride while the abstract says 'Hypertension was induced by CaCl2 administration'. |
| Tang Y et al 2025 · Front Nutr | meta-analysis | supports | moderate | Full-text re-read 2026-08-25 (PMC JATS XML PMC12698422, fulltext-cache/xml/41393949.xml, 120,767 bytes, plus the publisher supplement Data_Sheet_1.docx from the 2026-08-16 paper drop; the source note said provenance: abstract). Tang Y, Lei Y, Xu M, Tu Q, Mao X, Huang N, Jiang Q. Front Nutr 2025;12:1656809, doi 10.3389/fnut.2025.1656809. PROSPERO CRD420251055848, PRISMA, ROB 2.0. 'A total of 1,877 articles were retrieved, with 10 RCTs ultimately included in the meta-analysis' - but ten is the inclusion count, not the denominator of either blood-pressure estimate. SBP: 'For SBP, nine studies with 541 participants were included' (I2 = 48.4%, fixed effects), 'The effect size was −4.21 (95% CI: −5.74 to −2.69).' DBP: 'For DBP, nine studies with 509 participants were included' (I2 = 11.9%), 'The effect size was −3.13 (95% CI: −4.42 to −1.84).' The tenth trial (Sobenin 2009) enters only the safety pool and the meta-regression. Direction supports the claim; the authors' own reading is restrictive - 'Overall, our study suggests that the impact of garlic intervention on hypertension may be limited' - because the effect concentrates in short trials and younger patients: SBP −5.69 (−8.73 to −2.66) in the 8-week band (three trials, 131 participants) against −1.39 (−3.85 to 1.07), non-significant, in the 12-week band (seven trials, 377), and −2.19 (−4.26 to −0.11) in adults over 60 (five trials, 303) against −5.93 (−9.52 to −2.33) at 50-60 (three trials, 151). These are between-trial subgroups with no interaction test reported. Secondary outcomes ride their own much smaller subsets, not the ten-RCT headline: HDL SMD 0.31 (0.03-0.59) from 'three studies with 201 participants' and TNF-alpha SMD −0.38 (−0.72 to −0.04) from 'two studies with 137 participants', the latter both from one group (Ried/NIIM). Central SBP, central DBP, pulse pressure, MAP, CRP and triglycerides were all null. Reliability, in the authors' own words: ROB 2.0 gave 'around 60% of the studies were deemed to have low risk, with 30% showing some concerns and 10% classified as high risk'; the limitations open with 'Insufficient studies: The number of high-quality RCTs available is quite limited, which somewhat diminishes the robustness of the outcome measures'; and while the body says 'The analysis found no publication bias for both diastolic and SBP indicators', the supplement's own SBP Egger test is p = 0.065 with 'Although there was a potential trend for publication bias, it did not reach statistical significance at the 0.05 level.' Three internal contradictions found on full-text read: the abstract calls the adverse-event excess 'not statistically significant' while Results give RR 1.79 (95% CI 1.18 to 2.72); the abstract's meta-regression directions are the exact reverse of section 3.8's; and total cholesterol and HDL are reported with identical estimates from the identical three trials. Also note the title claims 'long-term' consumption while the included trials run 6-12 weeks and the authors concede 'Absence of long-term effect assessments'. Independence, flagged not fixed: four of the nine pooled BP trials are separately-counting rows on this claim (Ried 2010, Ried 2013 = vault garlic-ried-2012, Sobenin 2009, Serrano 2023) and the source note's supersedes line names only the first three. Funded by the National Natural Science Foundation of China (82060085). |
| Ziaei S, Hantoshzadeh S, Rezasoltani P, Lamyian M 2001 · Eur J Obstet Gynecol Reprod Biol | RCT | mixed | low | Abstract-grade re-read 2026-08-25 (NCBI efetch, saved as fetched-efetch.txt; both full-text routes probed and both closed - BioC pmcoa returned 'No result can be found' (saved as probe-bioc.txt) and Europe PMC reports isOpenAccess=N / inEPMC=N / pmcid=null (saved as probe-epmc.json) - so effect sizes, confidence intervals, any power calculation, randomisation detail and COI are cannot-tell from available text). Ziaei S, Hantoshzadeh S, Rezasoltani P, Lamyian M; Eur J Obstet Gynecol Reprod Biol 2001;99(2):201-6; doi 10.1016/s0301-2115(01)00384-0. GRADE RCT STANDS - garlic was the randomised variable: 'In a randomized, single-blind, placebo-controlled study, 100 primigravidas with positive roll-over test were treated with daily doses of 800mg Garlet/day (n=50) or 800mg/day placebo (n=50) during the third trimester of pregnancy.' STANCE tested-null -> mixed, because the prior extract recorded only half of what this paper found. The continuous blood-pressure comparison is indeed null: 'There were not any significant difference in the means of HDL, LDL, triglyceride, inhibition of platelet aggregation, the means of systolic and diastolic blood pressure and the mean arterial blood pressure (MAP), between the two groups' - but the SAME sentence continues 'but there was a significant difference in the means of total cholesterol (P=0.038) and hypertension alone (P=0.043)', and the authors' conclusion is directional in the claim's favour: 'The administration of 800mg/day of Garlet during the third trimester of pregnancy was effective in reducing the occurrence of hypertension alone, but it was no effective in preventing of preeclampsia.' Both halves sit on this claim's own object - continuous BP means null, incident hypertension reduced - so the paper genuinely points both ways and mixed is the honest label. THE NULL-ON-DIRECTIONAL QUESTION, answered: reread #645 flagged this row for the contradicts rule (null on a directional claim), with the Waldvogel/Steinert power caveat in reserve. The rule does not fire here, and NOT because the trial is small - because this is not a pure null. The paper's own blood-pressure conclusion is FAVOURABLE. On the power question separately: what this trial could detect is cannot-tell from the abstract - 50 per arm, no power calculation, and no effect estimate or interval printed for SBP, DBP or MAP anywhere - so nothing readable establishes the continuous null as informative rather than merely underpowered; yet the trial was not so insensitive that it missed the hypertension-incidence difference. Note also that the engine scores tested-null and mixed identically (sign 0, full denominator weight), so this edit moves no number; it changes what the row TELLS a reader. QUALITY moderate -> low, on grounds the abstract PRINTS rather than omits: the trial is SINGLE-blind, not double ('randomized, single-blind, placebo-controlled study'); at least a dozen endpoints were tested - 'Serum total cholesterol, LDL-and HDL-cholesterol, triglyceride, and platelet aggregation were measured before and after the treatment. Blood pressure, weight, and edema were also examined during the entire study period' plus preeclampsia and hypertension incidence - with no multiplicity adjustment mentioned and the only two significant results landing at P=0.038 and P=0.043; and the two blood-pressure results are internally awkward, no between-group difference in mean SBP, DBP or MAP yet a significant difference in the occurrence of hypertension. Randomisation method, allocation concealment, dropouts, funding and COI are absent from the abstract, which is cannot-tell and was NOT used as further grounds. POPULATION, and whether it fits the claim: 100 nulliparous primigravidas with a positive roll-over test, treated through the third trimester only. That is gestational-hypertension/preeclampsia physiology, not the essential hypertension the claim's meta-analyses pool. The row stays ON-SCOPE - the claim scopes no population and the measured outcome IS blood pressure, and this vault's own garlic-kwak-bp-2014 re-read counts this trial inside a blood-pressure pool - but the population is the single largest caveat on this row and belongs in front of any reader. DOSE: 800 mg/day of a garlic tablet (Garlet); that the preparation is garlic POWDER comes from Kwak 2014's classification, not from this abstract. INDEPENDENCE, flagged not fixed (rubric axis 7, and answering reread #638): this trial sits INSIDE garlic-kwak-bp-2014's pool - that row's own re-read records it as 'ref 36, scored N for both SBP and DBP in Table 1' - while also counting here as a separate weight-5 RCT row, so these 100 women vote at least twice on this claim, and probably more, since Ried 2016 searched 1955-December 2013 and Wang 2015 to November 2013. No duplicate-population role is proposed: the sibling precedent on this claim is to FLAG primary trials contained in later pools (garlic-sobenin-bp-2009 inside Rohner 2015; garlic-ried-2010/-2012 probably inside Ried 2016) rather than zero them, and fixing one containment mid-corpus would move this claim's denominator inconsistently. NOT APPRAISED HERE, deliberately: the lipid, platelet-aggregation and preeclampsia results are this trial's PRIMARY endpoints and are recorded as candidates, not as rows on this claim. |
| Wang HP et al. 2015 · J Clin Hypertens | meta-analysis | supports | moderate | Meta-analysis of 17 RCTs (Ovid Medline, Cochrane Library, PubMed, 1946 to November 2013); the abstract reports no participant total. Pooled: 'garlic intake caused a 3.75-mm Hg reduction (95% confidence interval [CI], -5.04 to -2.45, I(2) =30.7%; P<.001) in systolic BP and a 3.39-mm Hg reduction (95% CI, -4.14 to -2.65, I(2) =67%; P<.001) in diastolic BP compared with controls.' The DBP half is the unstable one. In the subgroup analyses there was 'a significant reduction in systolic BP in hypertensive (-4.4 mm Hg; 95% CI, -7.37 to -1.42, I(2) =0.0%; P=.004) but not normotensive patients. No significant reduction in diastolic BP was seen.' DBP reached significance only after trials were removed: 'After sensitivity analysis, heterogeneity disappeared and significant diastolic BP reduction (-2.68 mm Hg, 95% CI, -4.93 to -0.42, I(2) =0.0%; P=.020) was shown in hypertensive patients' - the abstract does not say which or how many trials were excluded. Denominator warning: -4.4 mmHg is a hypertensive-SUBGROUP estimate and the abstract gives no trial or participant count for that subgroup, so it must not be read against the 17-trial pool; the normotensive subgroup has no reported estimate at all, only non-significance. Quality moderate on measured grounds: DBP I(2) =67% resolved by post-hoc exclusion and a subgroup result that flips with the included set. Publication-bias testing, risk-of-bias assessment and registration are not mentioned in the abstract - cannot-tell, not absent. Abstract-grade (PMC8031974 is outside the OA subset; BioC and Europe PMC full text both unavailable 2026-08-25). |
| Sobenin IA et al. 2009 · Hypertens Res | RCT | supports | moderate | Abstract-grade re-read 2026-08-25 (NCBI efetch, saved as fetched-efetch.txt; PMC BioC probed and returned 'No result can be found' - Hypertens Res 2009 is not in the PMC OA subset, so no full text exists to read). Sobenin IA, Andrianova IV, Fomchenkov IV, Gorchakova TV, Orekhov AN, Hypertens Res 2009;32(6):433-7, doi 10.1038/hr.2009.36. STANCE supports, GRADE RCT and QUALITY moderate all STAND - the claim's own exposure was the randomised variable ('patients were randomized either to 600 mg Allicor (n=30) or to placebo (n=20) daily for 8 weeks') and a double-blind placebo-controlled core with an 8-week placebo run-in is a real design strength, set against 50 men in that core, men only, one centre, and no allocation, dropout or COI detail available. What did NOT stand is the DENOMINATOR. The old extract read 'RCT n=84' and hung every number on it; 84 is the whole study and no reported estimate is over 84. The headline result belongs to the randomised 30-vs-20 arm: 'Allicor treatment (600 mg daily) resulted in a reduction of both systolic and diastolic blood pressures by 7.0 mm Hg (95% confidence interval (95% CI): 5.3-8.7) and 3.8 mm Hg (95% CI: 2.7-4.8), respectively.' The other 34 men are an OPEN-LABEL branch with no placebo comparator - 2400 mg Allicor (n=18: 'Increasing the Allicor dosage to 2400 mg daily did not provide any additional benefit') and 900 mg Kwai (n=16: SBP -5.4 mm Hg, 95% CI 1.9-8.8, 'but no decrease in diastolic blood pressure was observed with Kwai'). So the Kwai figure is an unblinded uncontrolled n=16, not an RCT result, and the Allicor-beats-Kwai conclusion in the title compares a blinded arm with an open-label one. Population is men only with mild or moderate hypertension over 8 weeks. Whether the reductions are placebo-adjusted between-group differences or within-arm changes is cannot-tell from the abstract, which never says. INDEPENDENCE, flagged not fixed: first author Sobenin is also third author of garlic-rohner-2015, the meta-analysis of nine double-blind placebo-controlled garlic trials in hypertensives that votes separately on this claim - this trial fits its inclusion shape exactly, so these 50 randomised men may be voting twice, once here and again inside Rohner and the nine other synthesis rows on this claim. |
| Kwak JS, Kim JY, Paek JE, Lee YJ, Kim HR, Park DS, Kwon O 2014 · Nutr Res Pract | meta-analysis | supports | moderate | Full-text re-read 2026-08-25 (publisher PDF via unpaywall + JATS XML, both saved in the package; the source note said provenance: abstract). Kwak JS, Kim JY, Paek JE, Lee YJ, Kim HR, Park DS, Kwon O, Nutr Res Pract 2014;8(6):644-654, doi 10.4162/nrp.2014.8.6.644. STANCE supports and grade meta-analysis both STAND; quality deliberately HELD at moderate. DENOMINATOR, now named: the BP result does not rest on the review's headline 27 trials - 'A total of nine (n = 470) or ten (n = 550) studies were appropriate for the meta-analysis evaluation on whether garlic powder intake reduced SBP or DBP', so SBP is 9 trials / 470 subjects and DBP is 10 trials / 550, and the intervals the prior extract dropped nearly touch zero: 'SBP was reduced by -4.34 mmHg (95% CI, -8.38, -0.29, P = 0.036) with significant heterogeneity ( I 2 = 79.3%)' and 'The overall pooled estimate of the effect of garlic powder on DBP was -2.36 mmHg (95% CI, -4.56, -0.15, P = 0.036), and the heterogeneity across studies was significant ( I 2 = 65.4%)'. POPULATION, sharper than 'mixed': Table 1 shows only ONE of the nine/ten BP trials was run in a hypertensive population (Auer W 1990, n = 47); the others were healthy volunteers, hypercholesterolaemic or hyperlipoproteinaemic patients, peripheral arterial occlusive disease, and preeclampsia-risk pregnancy (Ziaei 2001). This row is therefore evidence for BP-lowering in a largely NON-hypertensive pool - the subgroup the claim body itself says shows little change - and it still found a pooled signal. The paper's own strata point the familiar way: the SBP net change was significant only for trials 'conducted in Europe or America (P = 0.028), administered a daily dose <= 800 mg (P = 0.019) and involved populations at high-risk for hypertension (P = 0.002)', and that last figure, -8.08 mmHg, is a 4-trial / n = 180 stratum (Table 2), not the pooled estimate; the healthy stratum was 2 trials / n = 98 at -5.17 (95% CI -23.50 to 13.15), non-significant. FRAGILITY the prior extract hid: 'Contrary to the pooled results, the sensitivity test revealed no significant effect on FBG or SBP in all restricted categories' - SBP lost significance in every one of the seven restriction analyses, including placebo-controlled-only (Table 3: 8 trials, n = 372, -4.28, 95% CI -8.77 to 0.21). DBP survived in the placebo-controlled (9 trials, n = 452, -2.61, 95% CI -5.00 to -0.23) and high-quality (3 trials, n = 190, -2.34, 95% CI -4.27 to -0.41) restrictions. EXPOSURE SCOPE: 'Only garlic powder intake studies were included in this analysis' - dehydrated powder only, no aged garlic extract, garlic oil or raw clove, so this row does not speak to the AGE preparations used by the Ried trials that also vote on this claim. QUALITY held at moderate and moved in neither direction, on purpose. Upward: PRISMA, duplicate independent extraction, Cochrane risk-of-bias, random-effects, study-removal sensitivity, meta-regression, and 'Egger's regression test suggested no evidence of substantial publication bias'. Downward: the authors' own meta-regression reports a small-study relationship inside this very result - 'However, participant number was inversely associated with the effect size of LDL-C (slope = 0.004, P = 0.020) and DBP (slope = 0.093, P = 0.006)' - the heterogeneity is high and unexplained ('no factors, such as dose, duration, subject type, location and study quality, significantly changed heterogeneity'), and they state 'Due to the high heterogeneity in meta-analyses, the resulting effect sizes should be interpreted with some degree of caution despite the lack of publication bias'. Those cancel at moderate; neither a high nor a low is earned. FUNDING: 'This study was performed with the support of the Cooperative Research Program for Agriculture Science & Technology Development funded by the Rural Development Administration (Project No. PJ008450022014), Republic of Korea', and two co-authors (Kim HR, Park DS) are employed by that funder's Department of Agrofood Resources. No competing-interest statement appears in the paper - cannot-tell, not a declaration of none. INDEPENDENCE, flagged not fixed: the -16.33 / -9.28 mmHg hypertensive figures quoted in this paper's discussion are Reinhart 2008's (its ref 49), which is already a separate counting row in this vault; Ziaei 2001 sits INSIDE this pool (ref 36, scored N for both SBP and DBP in Table 1) while also counting separately as its own tested-null RCT row; Silagy 1994's Kwai-powder pool is very likely a subset of this powder-only pool; and three Sobenin/Orekhov trials are inside this pool while that group votes again through a separate 2009 RCT row. |
| Ried K 2016 · J Nutr | meta-analysis | supports | moderate | Abstract-grade re-read 2026-08-25 (NCBI efetch, saved as fetched-efetch.txt). ACCESS GAP, not an absent paper: Europe PMC reports isOpenAccess=N / pmcid=null so no PMC or BioC route exists, while Unpaywall lists a FREE publisher PDF at academic.oup.com/jn/article-pdf/146/2/389S/30021378/jn202192.pdf that returns a Cloudflare interstitial (saved as fetch-oup-pdf-BLOCKED.html). The trial list, funding/COI, heterogeneity and any bias testing all sit behind that one gate and are cannot-tell here. Ried K (sole author), J Nutr 2016 Feb;146(2):389S-396S, doi 10.3945/jn.114.202192. STANCE supports and grade meta-analysis both STAND; quality lowered high -> moderate. WHAT THIS PAPER IS: a hybrid. Only the blood-pressure section is a meta-analysis; the cholesterol and immunity sections pool nothing of their own, and this vault correctly draws only the BP row from it. DENOMINATORS, now named. The prior extract set 'hypertensive subgroup -8.7 / -6.1 mmHg' directly under '20 RCTs (n=970)'; neither subgroup figure rests on 20 trials or 970 participants. 'Subgroup analysis of trials in hypertensive subjects (SBP/DBP >=140/90 mm Hg) at baseline revealed a larger significant reduction in SBP of 8.7 +/- 2.2 mm Hg (P < 0.001; n = 10) and in DBP of 6.1 +/- 1.3 mm Hg (P < 0.001; n = 6).' So SBP rests on 10 units and DBP on 6 - the two subgroups are asymmetric with no stated reason, the unit is never labelled (read as trials, consistent with Ried 2008's 4-trial/3-trial hypertensive subgroups), and NO participant count is given for either. 'Hypertensive' is a baseline threshold, not a diagnosis - the same convention as Ried 2008 and Reinhart 2008. THE POOLED RESULT, in full: 'Our updated meta-analysis on the effect of garlic on blood pressure, which included 20 trials with 970 participants, showed a mean +/- SE decrease in systolic blood pressure (SBP) of 5.1 +/- 2.2 mm Hg (P < 0.001) and a mean +/- SE decrease in diastolic blood pressure (DBP) of 2.5 +/- 1.6 mm Hg (P < 0.002) compared with placebo.' Read as printed these two do not cohere: with +/- as SE, 5.1 +/- 2.2 gives t ~ 2.3 and 2.5 +/- 1.6 gives t ~ 1.6, neither of which reaches P < 0.001 or P < 0.002 - while the two subgroup figures in the very next sentence (t ~ 4.0 and ~ 4.7) do cohere with theirs. The defect is confined to the pooled estimates; whether the +/- are mislabelled or the P values misstated is cannot-tell. EXPOSURE SCOPE: 'garlic preparations' with no split by powder, aged garlic extract or oil and no dose or duration anywhere in the abstract, so this row cannot be read as evidence for AGE despite the claim body's 'esp. aged garlic extract'. QUALITY high -> moderate, on four grounds. (1) The search is ONE database with a cutoff two years before publication: 'We searched the Medline database for randomized controlled trials (RCTs) published between 1955 and December 2013' - no CENTRAL, no Embase. (2) The abstract establishes NOTHING about the pool's own reliability: no publication-bias test, no risk-of-bias assessment, no heterogeneity statistic, no registration. That is cannot-tell, and absence settles nothing in either direction - but a 'high' has to be EARNED by what a paper shows about itself, and nothing readable earns it. (3) The headline numbers do not cohere with their own P values, above. (4) The deciding hypertensive estimates are 10- and 6-unit subgroups with unreported denominators. Moderate also lands this level with every sibling synthesis on this claim that has been re-read - Silagy 1994, Reinhart 2008, Ried 2008, Kwak 2014, Rohner 2015 and Xiong 2015 all sit at moderate on their own grounds. STANCE re-examined against mixed and HELD: every result in the paper points the same way on the claim's object, and the author's own conclusion is directional - 'Our review suggests that garlic supplements have the potential to lower blood pressure in hypertensive individuals'. Note what she does NOT claim: 'Future long-term trials are needed to elucidate the effect of garlic on cardiovascular morbidity and mortality.' INDEPENDENCE, FLAGGED NOT FIXED (rubric axis 7 - no role change proposed, because an independence fix moves this claim's denominator mid-corpus). The group_id 'ried-niim' is already correct and already shared with garlic-ried-2008, -2010, -2012 and rohner-2015, so the label is not the gap. The gap is POOL CONTAINMENT, which group shrinkage does not model, and the paper states it itself: 'We updated a previous meta-analysis on the effect of garlic on blood pressure.' The earlier Ried pool (garlic-ried-2008, 11 RCTs) is a SEPARATE counting meta-analysis row on this same claim, so those 11 trials already vote elsewhere - containment, not partial overlap. Whether garlic-ried-2010 (Maturitas 2010, n=50) and garlic-ried-2012 (Eur J Clin Nutr, epub Nov 2012, n=79) are also inside is UNRESOLVED: both fall squarely in the stated 1955-December 2013 window, both are RCTs of garlic preparations on blood pressure, both count separately here as RCT rows, and both were run by this paper's sole author - so inclusion is very likely, but the abstract prints no trial list and the PDF is gated, so it cannot be confirmed. Sibling rereads #641/#642 pointed this exposure at this row; it is recorded here as probable and unconfirmed rather than adopted as fact. Claim-level: 18 rows, 10 of them meta-analyses over one shared garlic-trial literature, of which the Ried network holds three syntheses plus two RCTs. NOT APPRAISED, deliberately: this paper's cholesterol figure is someone else's pooling re-described - 'A previously published meta-analysis on the effect of garlic on blood lipids, which included 39 primary RCTs and 2300 adults treated for a minimum of 2 wk, suggested garlic to be effective in reducing total and LDL cholesterol by 10%' - and its immunity section is narrative. Any row drawn from this source for claim-garlic-lowers-ldl-cholesterol or claim-garlic-prevents-common-cold would be evidence_role: review, zero weight, and a literal double count. Correctly absent; keep them absent. FUNDING/COI: not seen - the abstract carries neither statement, which is not a declaration of none. The 'S' pagination (389S-396S) marks a J Nutr supplement issue rather than a regular issue; whether that supplement had a commercial sponsor is cannot-tell and was NOT used as grounds for the downgrade. |
| Ried K, Frank OR, Stocks NP 2010 · Maturitas | RCT | supports | moderate | Abstract-grade re-read 2026-08-25 (NCBI efetch, saved as fetched-efetch.txt; PMC BioC probed and returned no result - Maturitas is not in the PMC OA subset, so no full text exists to read). Ried K, Frank OR, Stocks NP, Maturitas 2010;67(2):144-50, doi 10.1016/j.maturitas.2010.06.001. STANCE supports, grade RCT and quality moderate ALL STAND - the design is 'A double-blind parallel randomised placebo-controlled trial involving 50 patients whose routine clinical records in general practice documented treated but uncontrolled hypertension', garlic itself was the randomised exposure ('four capsules of aged garlic extract (960 mg containing 2.4 mg S-allylcysteine) daily for 12 weeks' vs 'matching placebos'), and it was given as an ADJUNCT to existing antihypertensive medication. DENOMINATOR, now named: the -10.2 mmHg figure is NOT an n=50 effect. It is a baseline-stratum result - 'In patients with uncontrolled hypertension (SBP >= 140 mmHg at baseline), systolic blood pressure was on average 10.2 +/- 4.3 mmHg (p=0.03) lower in the garlic group compared with controls over the 12-week treatment period' - and the abstract states NO n for that stratum and reports NO whole-sample estimate, so the stratum's own denominator is cannot-tell from the available text. THE HALF THAT WAS DROPPED: 'Changes in blood pressure between the groups were not significant in patients with SBP<140 mmHg at baseline.' OBJECT vs MEASURED: the claim's object is blood pressure, and DBP was a stated primary outcome - 'The primary outcome measures were systolic and diastolic blood pressure at baseline, 4, 8 and 12 weeks, and change over time' - yet no DBP number appears anywhere in the abstract. This row therefore carries the SYSTOLIC half only and is silent on diastolic; that silence is cannot-tell from an abstract, not a null. Quality stays moderate rather than dropping because the demerits I can establish (headline is a stratum, p=0.03 is borderline, no registry shown) sit against a genuinely double-blind placebo-controlled parallel design; whether the >=140 split was prespecified is cannot-tell and absence of that statement settles nothing. Author overreach NOT adopted: the conclusion's 'similarly to current first line medications' names a comparator the trial never randomised or measured - it compared garlic against placebo only. INDEPENDENCE (flagged, not fixed): this is an ORIGINAL primary observation and keeps its weight - Ried 2008 predates it and pooled other trials, so 2010 is not inside it. The double count runs the other way: Ried 2016's 20-RCT pool almost certainly re-counts this trial and Ried 2012, so the duplicate-population exposure sits on the ried-2016 META row. The Ried Adelaide group votes five times on this claim (2008 MA, 2010 RCT, 2012 RCT, 2016 MA, plus Silagy 1994 counting separately inside the 2008 pool). ↩ SUPERSEDED — pooled in the review above, counted once |
| Bharatbhai PD et al 2026 · study_type: RCT | observational | supports | low | n=200 hypertensive villagers (100 intervention/100 control), raw garlic clove vs no treatment; SBP fell 152.3→128.5 mmHg and DBP 96.8→82.3 mmHg (p<0.001) vs control. |
| Hesari NG et al 2025 · J Health Popul Nutr | RCT | supports | moderate | Double-blind placebo RCT, 97 Iranian women with PCOS + metabolic syndrome, garlic 500 mg tablet (2-3 mg allicin) twice daily for 8 weeks: 'This study revealed that both SBP and DBP decreased following garlic supplementation.' Table 1: SBP 118.33 -> 110.82 mmHg (baseline-adjusted mean) vs placebo 113.97 -> 116.33, F=16.179, P<0.001; DBP 81.35 -> 75.76 vs 78.06 -> 79.86, F=12.685, P=0.001. Scope caveat for this claim: the cohort was NORMOTENSIVE at baseline (~118/81), i.e. the subgroup in which the pooled literature generally shows little or no effect, and baseline DBP was significantly imbalanced (garlic higher, P=0.04) so the DBP result rests entirely on the ANCOVA adjustment. Blood pressure was measured with a mercury sphygmomanometer by Korotkoff sounds, single-centre, unstated number of readings. |
| Xiong XJ et al. 2015 · Phytomedicine | meta-analysis | supports | moderate | Abstract-grade re-read 2026-08-25 (NCBI efetch, saved as fetched-efetch.xml + fetched.txt; two full-text routes probed and both closed - BioC pmcoa returned 'No result can be found' and Europe PMC reports isOpenAccess=N / pmcid=null - so the trial list, any heterogeneity statistic and the risk-of-bias grid are cannot-tell from available text). Xiong XJ, Wang PQ, Li SJ, Li XK, Zhang YQ, Wang J; Phytomedicine 2015;22(3):352-61; doi 10.1016/j.phymed.2014.12.013. Meta-analysis of placebo-controlled RCTs restricted at inclusion to hypertensive patients ('Randomized, placebo-controlled trials comparing garlic vs. a placebo in patients with hypertension were considered'), searched 'until August 2014'; 'A total of seven randomized, placebo-controlled trials were identified.' Pooled result: 'a significant lowering effect of garlic on both systolic BP (WMD: -6.71 mmHg; 95% CI: -12.44 to -0.99; P = 0.02) and diastolic BP (WMD: -4.79 mmHg; 95% CI: -6.60 to -2.99; P < 0.00001)'. DENOMINATOR: no participant total is reported anywhere in the abstract, so the only denominator this row owns is SEVEN TRIALS and no person-count may be attached to it. Contrary to the DBP-is-fragile pattern elsewhere in this claim family, here SBP is the fragile half - its interval reaches -0.99 mmHg, i.e. the pool is compatible with a clinically trivial 1 mmHg - while the DBP interval (-6.60 to -2.99) is tight. Quality moderate on measured grounds: two independent reviewers and a clean hypertensives-only inclusion rule on the credit side; against them, only seven trials, no reported heterogeneity statistic, no reported publication-bias test (absence of a test, not a negative test), and the authors' own close that 'more rigorously designed randomized controlled trials focusing on primary endpoints with long-term follow-up are still warranted before garlic can be recommended to treat hypertensive patients'. INDEPENDENCE (flagged, not fixed): one of nine weight-8 synthesis rows on this claim; searched to Aug 2014 in hypertensives only, its seven trials must overlap heavily with Rohner 2015 (9 trials, 482 people, same population, searched to Mar 2014) and sit largely inside Wang 2015 (17 trials to Nov 2013) - three passes over one small garlic-powder literature, counted three times. |
| Ried K, Frank OR, Stocks NP 2012 · Eur J Clin Nutr | RCT | supports | moderate | Full-text re-read 2026-08-25 (PMC JATS XML, PMC3561616, local artifact fulltext-cache/xml/23169470.xml; the source note said provenance: abstract). Ried K, Frank OR, Stocks NP, Eur J Clin Nutr 2013;67(1):64-70 (epub 2012-11-21), doi 10.1038/ejcn.2012.178, registered ACTRN12611000581965. STANCE supports, grade RCT and quality moderate ALL STAND - garlic itself was the randomised variable ('one, two or four capsules daily of Kyolic aged garlic extract ... containing either 240/480/960 mg of aged garlic extract and 0.6/1.2/2.4 mg S-allylcysteine or placebo capsules for 12 weeks'), given as an ADJUNCT to existing antihypertensive medication in treated-but-uncontrolled SYSTOLIC hypertensives (mean age 70). DENOMINATOR, now named: -11.8 mmHg is NOT an n=79 effect. It is the prespecified PLANNED ADJUSTED analysis of 74 patients - 'planned adjusted analysis of 74 patients, which revealed a significant difference in reduction of SBP between the garlic-2 group and placebo at 8 and 12 weeks, and over time compared with baseline (g2; 0-12 weeks: mean diff. SBP+/-s.e. (95% confidence interval)=-11.8+/-5.4 (-22.6;-1.0) mm Hg; P=0.006)' - and at 12 weeks that contrast is garlic-2 n=17 vs placebo n=14 (Table 2). The n=79 ITT figure is smaller and only just significant: 'Intention-to-treat analysis of 79 patients revealed a significant reduction in SBP from baseline in the garlic-2 group compared with placebo over 12 weeks (mean diff. SBP+/-s.e. (95% confidence interval)=-9.7+/-4.8 (-19.3; -0.1) mm Hg; P=0.03)'. THE HALVES THAT WERE DROPPED: (a) DBP was a co-primary outcome and was NULL - 'Intention-to-treat analysis of DBP did not reveal a significant effect of treatment between the groups', with the adjusted garlic-2 estimate -5.3+/-4.2 (-13.7; 3.1) mmHg, 'albeit insignificantly different to placebo'; the authors state this was designed-in, 'Our trial had limited power to detect any significant changes in DBP ... as participants were selected on the basis of systolic hypertension, subsequently including <13% of participants with essential hypertension'. So the diastolic silence is underpowered, not a demonstrated null. (b) The DOSE-RESPONSE the trial was built to find did not appear: the top dose reached only 'borderline significant reduction in the garlic-4-capsule group at 8 weeks (-7.4+/-4.1 mm Hg, P=0.07)' and the 1-capsule arm's ITT change difference numerically favoured PLACEBO (+4.6, 95% CI -7 to 16.2) - 'Although SBP dropped significantly within the garlic-1 group at 12 weeks, the change did not reach statistical significance when compared with placebo.' The authors' explanation for 480mg beating 960mg is conjecture: the smaller g4 effect 'may have been linked to the poorer compliance and lesser tolerability'. (c) Every interval nearly touches zero. QUALITY held at moderate on purpose; the reasons cancel. Upward: registry-listed, computer-generated allocation by an independent statistician, odour-matched placebo (garlic sachets in identical opaque containers), blinding formally TESTED at exit and near chance ('A third of the participants correctly guessed their allocation'), prespecified sample size, ITT primary, calibrated Omron HEM-907 with triplicate seated readings. Downward: arms of ~20, no multiplicity control across 3 doses x 4 timepoints x 2 analysis sets at P<0.05, non-monotone dose-response, the headline drawn from the adjusted set, 14% recruitment response ('Of the 840 patients invited, 14% responded'), and a third of all participants moved <5 mmHg regardless of arm. That lands level with sibling garlic-ried-2010, also moderate. STANCE re-examined against mixed and HELD: the disconfirming halves are same-sign or explicitly underpowered, not results pointing the other way on the claim's object. FUNDING: Royal Adelaide Hospital New Investigator grant 11RAHRC-7360 + Australian Government PHCRED; 'Trial capsules were provided by Vitaco Health Ltd, Sydney, Australia, which was not involved in study design, data collection, analysis or preparation of the manuscript'; 'The authors declare no conflict of interest.' INDEPENDENCE, flagged not fixed: this is a NEW cohort, not a re-report of Ried 2010 - Adelaide, Aug 2011-Mar 2012, n=79, four arms, ACTRN12611000581965, and the paper cites 2010 as prior work ('in line with our previous findings') - so it keeps its weight as an original primary observation. The double count sits on the ried-2016 META row, whose 20-RCT pool almost certainly re-counts this trial and Ried 2010. The Ried Adelaide group votes five times on this claim (2008 MA, 2010 RCT, 2012 RCT, 2016 MA, rohner-2015 MA). ↩ SUPERSEDED — pooled in the review above, counted once |
| Rohner A, Ried K, Sobenin IA, Bucher HC, Nordmann AJ 2015 · Am J Hypertens | meta-analysis | supports | moderate | Abstract-grade re-read 2026-08-25 (NCBI efetch, saved as fetched-efetch.xml + fetched.txt; no full text obtained, provenance stays abstract). Rohner A, Ried K, Sobenin IA, Bucher HC, Nordmann AJ, Am J Hypertens 2015;28(3):414-23, doi 10.1093/ajh/hpu165. STANCE supports, GRADE meta-analysis and QUALITY moderate ALL STAND; the defect was the extract. DENOMINATOR, now named: 'Nine double-blind trials with 482 individuals fulfilled our inclusion criteria', hypertensive patients only, 'Follow-up ranged from 8 to 26 weeks. All trials reported office BP measurements', literature searched through March 2014. The intervals and heterogeneity the prior extract dropped: 'heterogeneity was high (weighted mean difference (WMD) for SBP was -9.1 mm Hg; 95% confidence interval (CI), -12.7 to -5.4; P for heterogeneity = 0.0006; and I2 = 71%; WMD for BP was -3.8 mm Hg; 95% CI, -6.7 to -1.0; P for heterogeneity = 0.00001; I2 = 80%)' - the bare 'BP' in that second arm is a typographical slip for DBP. DBP IS REPORTED HERE and significant, unlike much of this claim's synthesis family. The prior extract's blanket 'effect shrank in higher-quality sensitivity analysis' named no endpoint and was therefore wrong in half: 'When analyses were restricted to higher-quality trials using intention-to-treat analysis or to trials with concealed treatment allocation and standardized and blinded BP measurement, effect sizes for SBP but not for DBP were lower and heterogeneity disappeared' - the attenuation is the SBP result; DBP survived the quality restriction. The attenuated SBP point estimate is not in the abstract (cannot-tell from available text). QUALITY held at moderate on this paper's own measured grounds, not by family consistency: off high because 'Included trials were rather small, and the quality of the majority of included trials was moderate', heterogeneity was severe on both endpoints, and the authors close 'the evidence is not strong. A well-conducted and powered trial of longer duration is needed to confirm these findings'; off low because the paper ran the ITT / allocation-concealment / blinded-measurement sensitivity analysis that produced those findings and published the attenuation against its own headline. No publication-bias or Egger test is described in the abstract - cannot-tell, not an absence. INDEPENDENCE (flagged, never fixed): group_id ried-niim is right - Karin Ried is second author and first author of four other rows on this claim. Two overlaps it does NOT capture. (1) Third author Igor A Sobenin is the author of garlic-sobenin-bp-2009 (group_id sobenin), an RCT voting separately at full weight on this same claim and a likely included trial - the abstract never names the nine, so which-nine is cannot-tell. (2) This 482-subject hypertensive pool is drawn from the same small-trial literature pooled by the other NINE synthesis rows on this claim (kwak-bp-2014, reinhart-2008, ried-2008, ried-2016, silagy-1994, wang-bp-2015, xiong-2015, s40735665, s41393949); s40735665 alone reports 12 placebo-controlled hypertensive trials, a probable superset of all nine here. Ten meta-analysis rows, roughly one literature. |
| Serrano JCE et al. 2023 · Nutrients | RCT | mixed | low | Triple-blind placebo-controlled RCT (NCT04915053), 81 randomized / 77 completed (38 placebo, 39 garlic), Grade-I hypertensives (mean age 64, 15 years diagnosed) ALREADY on antihypertensive drugs - an add-on design. Dose: one 250 mg aged-black-garlic (ABG10+) tablet giving just 0.25 mg S-allyl-cysteine/day for 12 weeks. MIXED because the two blood-pressure routes disagree. OFFICE BP (co-primary, Table 3): no treatment effect - two-way ANOVA p=0.3011 systolic and p=0.9219 diastolic, and the arm with the significant office systolic fall was PLACEBO (144 to 138 mmHg, Sidak p=0.0169), not garlic (146 to 143, p=0.3011). HOME daily BP (Figures 2-3, linear regression of daily readings over 84 days): the between-group slope difference IS significant (p=0.0033 systolic, p=0.0002 diastolic; garlic slope -21.6e-3 mmHg/day, p=0.0021; placebo no slope), but the 12-week magnitude it implies is only -1.8/-1.5 mmHg with a 95% CI SPANNING ZERO - the Results define the bounds as running from a maximum reduction of 4.1 and 3.0 mmHg to an INCREASE of 0.7 and 0.3 mmHg, so the abstract's rendering as '1.8 mmHg (0.7 to 4.1 95% CI)' reads as excluding zero when it does not. The -1.8/-1.5 figure is HOME BP only; it is not an office finding and its denominator is the 39-person garlic arm's daily readings. The trial was powered for a 9 mmHg systolic drop (36/group) taken from Ried 2016 and undershot that by roughly five-fold. Endothelial function null (RHI treatment ANOVA p=0.9753); every secondary the abstract presents as an effect - nitric oxide, FRAP, uric acid, ACE - has a NON-significant treatment ANOVA (p=0.51, 0.70, 0.40, 0.96) and is a post-hoc within-arm change, as the authors state: except for IL-2 there were no significant changes due to treatment in Table 3. Lipids flat (total cholesterol 209 to 209 mg/dL, p=0.9759); LDL not measured. Funded by Pharmactive Biotech, whose employees are three of the nine authors and which produces the extract. |
| Ried K, Frank OR, Stocks NP, Fakler P, Sullivan T 2008 · BMC Cardiovasc Disord | meta-analysis | supports | moderate | Full-text re-read 2026-08-25 (PMC JATS XML, PMC2442048, saved in the package as fulltext.xml/fulltext-plain.txt; the source note said provenance: abstract). Ried K, Frank OR, Stocks NP, Fakler P, Sullivan T, BMC Cardiovasc Disord 2008;8:13, doi 10.1186/1471-2261-8-13. STANCE supports and grade meta-analysis both STAND; quality lowered high -> moderate. DENOMINATORS, now named: '11 RCTs' is the set eligible for meta-analysis, not the pool behind any number stored here. The overall SBP estimate is TEN studies - 'Meta-analysis of ten studies of the effect of garlic on SBP showed a significant difference between garlic and control groups, with garlic having a greater effect in reducing SBP than placebo by 4.56 [95% CI (confidence interval), -7.36, -1.77] mm Hg compared with placebo (p < 0.001)' - on 'A total of 252 individuals allocated to a garlic intervention group and 251 individuals allocated to a control group'. The hypertensive figures rest on FOUR studies (SBP) and THREE (DBP): 'four studies reporting mean SBP in the hypertensive range' and 'three studies reporting mean DBP in the hypertensive range'; SBP -8.38 [95% CI, -11.13, -5.62] mm Hg and DBP -7.27 [95% CI, -8.77, -5.76] mm Hg. THE OMISSION: the prior extract recorded the pooled SBP result and the hypertensive subgroup but dropped the pooled DBP result, which was NULL - 'Meta-analysis of eleven studies of the effect of garlic on DBP did not show a significant difference between garlic and placebo groups (-2.44 [95% CI, -4.97, 0.09] mm Hg, p = 0.06)' - on 283 garlic vs 282 control. Both normotensive subgroups were also non-significant (SBP: 'showed no significant difference between the garlic and placebo groups'; DBP: 'In contrast, subgroup meta-analysis of 'normotensive' individuals was not significant'). STANCE STILL SUPPORTS, not mixed: nothing in this paper points the other way - the null DBP pool has the same sign (-2.44) with an interval grazing zero, and the authors conclude 'garlic preparations are superior to placebo in reducing blood pressure in individuals with hypertension'. EXPOSURE SCOPE, which matters because the claim body says 'esp. aged garlic extract': this pool is almost entirely POWDER - 'Nine studies used garlic powder (mainly 'Kwai', a standardised garlic supplement), one study used aged garlic extract and another used distilled garlic oil', 600-900 mg/d, 12-23 weeks. Two of the eleven are add-on designs ('two studies compared the effect of garlic on blood pressure in addition to a drug compared to drug plus placebo'), where garlic is still the randomised variable, so the meta-analysis grade holds. 'Hypertensive' here is a BASELINE-MEAN definition (mean SBP >= 140 or mean DBP >= 90), not a diagnosis, and most included trials were lipid or peripheral-arterial-disease trials reporting BP. QUALITY high -> moderate. Upward, on the authors' own testing: all eleven trials 'reported adequate randomisation and double blinding', quality was assessed independently by two investigators, 'Funnel plots and Egger regression tests suggested no publication bias in the meta-analyses', and 'regression analysis did not provide any evidence to suggest that receipt of industry funding (n = 3) was associated with blood pressure outcomes'. Downward, and decisive: the authors report bias in their own included trials - '20% loss to follow up in two trials and non-reporting of drop out rates in three trials might have biased the results in those studies'; every pooled variance is IMPUTED, not observed - 'Standard deviations of these differences were estimated applying the equation published in Taubert et al.' with 'a conservative correlation coefficient of R = 0.68'; heterogeneity in the pooled analyses is high ('Heterogeneity was moderate for meta-analysis of SBP of all ten studies (I 2 = 57.1%)' and 'I 2 = 83.2%' for pooled DBP); the search was two databases plus Cochrane, 1955 to Oct 2007, restricted to 'English or German'; and the deciding hypertensive estimate is a 4-trial / 3-trial subgroup where an Egger test has no power. An Egger null on 10-11 studies is a weak reassurance, not a strong one. This also lands Ried 2008 level with its two nearest siblings on this claim, Reinhart 2008 and Kwak 2014, both held at moderate on their own grounds. INDEPENDENCE, flagged not fixed: (a) this paper states it CONTAINS the earlier Silagy & Neil 1994 pool, which counts separately in this vault - 'Whilst data from two studies, included in a previous meta-analysis, were not available, our meta-analysis incorporates data from six additional recent studies'; (b) the Ried network votes repeatedly on this claim - Ried is first author here and of ried-2010, ried-2012 and ried-2016 (the 2016 meta-analysis pools her own 2010 and 2012 trials, which also count as separate rows) and second author of rohner-2015, so one group carries three of the ten meta-analysis rows plus two RCT rows; (c) all eleven included trials were published 1988-2000, i.e. they sit inside the same shared 1979-2010 powder-trial literature that the ~10 garlic-BP syntheses on this claim (~78% of its raw weight) keep re-pooling; overlap with Kwak 2014 is certain on the Kwai-powder trials (Auer 1990 is in both) but Kwak's full list cannot be enumerated from this paper's text. COMPETING INTERESTS: 'The authors declare that they have no competing interests'; funded by the Royal Australian College of General Practitioners 2006 Vicki Kotsirilos Integrative Medicine Grant and the Australian Government PHCRED Program - no garlic-industry sponsor. INTERNAL INCONSISTENCY, flagged: the abstract gives the SBP meta-regression as 'R = 0.057; p = 0.03' while the Results give 'SBP: R = -0.151, p = 0.03'; the sign and magnitude disagree between the two, and which is correct is cannot-tell from the paper. |
| Silagy CA, Neil HA 1994 · J Hypertens | meta-analysis | supports | moderate | Abstract-grade re-read 2026-08-25 (NCBI efetch, saved in the package as fetched-efetch.txt; the MEDLINE abstract carries NO pooled numbers, so magnitude is cannot-tell from available text). Silagy CA, Neil HA, J Hypertens 1994;12(4):463-8, doi 10.1097/00004872-199404000-00017. STANCE supports, GRADE meta-analysis and QUALITY moderate all STAND; extract corrected. DENOMINATORS, now named: the pool is 'Eight trials were identified (all using the same dried garlic powder preparation (Kwai) with data from 415 subjects included in the analyses' - but the placebo comparison rests on SEVEN of those eight, and inside it most single trials were null: 'Of the seven trials that compared the effect of garlic with that of placebo, three showed a significant reduction in systolic blood pressure (SBP) and four in diastolic blood pressure (DBP)'. POPULATION, the material omission from the prior extract: 'Only three of the trials were specifically conducted in hypertensive subjects, and many had other methodological shortcomings' - five of the eight pooled trials are therefore not hypertensive populations and no normotensive/hypertensive subgroup split is reported, which matters because this claim's effect is held to concentrate in hypertensives. MAGNITUDE: this abstract gives direction only - 'The overall pooled mean difference in the absolute change (from baseline to final measurement) of SBP was greater in the subjects who were treated with garlic then in those treated with placebo' and 'For DBP the corresponding reduction in the garlic-treated subjects was slightly smaller' - no point estimate and no CI, so the prior extract's 'modest BP reduction' was not sourced from any number in this text. STANCE tested against mixed and HELD: both endpoints point the same way on the claim's object and the authors' own conclusion is directional - 'The results suggest that this garlic powder preparation may be of some clinical use in subjects with mild hypertension' - while the counterweight the prior extract leaned on, 'there is still insufficient evidence to recommend it as a routine clinical therapy for the treatment of hypertensive subjects', is about clinical recommendation, not effect direction. QUALITY moderate on measured grounds, cutting both ways: real strengths are a search for unpublished data 'through direct contact with garlic manufacturers', dual independent extraction, and a 4-week minimum duration; against that stand the authors' own 'many had other methodological shortcomings', a total of 415 subjects across eight small trials, and no bias test or heterogeneity statistic readable here (cannot-tell, not absent). Moderate is also where every re-read synthesis sibling on this claim sits. SUBJECT NARROWER THAN THE CLAIM: a single-preparation pool (Kwai dried powder only), so this row is not evidence for the aged-garlic-extract reading in the claim body. INDEPENDENCE, FLAGGED NOT FIXED (rubric axis 7 - no role change): this is the earliest meta on the claim and its eight Kwai trials are re-pooled by most later syntheses (Ried 2008's note records incorporating Silagy's studies bar two), so one 1980s-90s powder literature votes through ~10 meta-analysis rows here. What this 1994 pool ADDS that its re-poolings do not carry: the only preparation-scoped estimate, and a manufacturer-contact hunt for unpublished trials the later Medline-only pools never repeated. What it does not add: any independent patient data. Supersedes-candidate for a group/shrinkage pass, not a duplicate-population role. |
| Reinhart KM, Coleman CI, Teevan C, Vachhani P, White CM 2008 · Ann Pharmacother | meta-analysis | supports | moderate | Abstract-grade re-read 2026-08-25 (NCBI efetch, saved as fetched-efetch.txt; no PMC full text - the BioC/pmcoa probe returned 'No result can be found', i.e. not in the OA subset, so risk-of-bias detail, the trial list and the preparations pooled are all cannot-tell from available text). Reinhart KM, Coleman CI, Teevan C, Vachhani P, White CM, Ann Pharmacother 2008;42(12):1766-71, doi 10.1345/aph.1L319. STANCE supports and grade meta-analysis both STAND; quality lowered high -> moderate. DENOMINATOR, now named: the -16.3 / -9.3 mmHg figures are NOT the 10-trial pool. 'Ten trials were included in the analysis; 3 of these had patients with elevated SBP. Garlic reduced SBP by 16.3 mm Hg (95% CI 6.2 to 26.5) and DBP by 9.3 mm Hg (95% CI 5.3 to 13.3) compared with placebo in patients with elevated SBP.' So the deciding estimate rests on a 3-trial subgroup, not on ten trials, and its interval spans 6.2 to 26.5 mmHg - a fourfold range the prior extract dropped. The remaining seven trials produced the null: 'However, the use of garlic did not reduce SBP or DBP in patients without elevated SBP.' The subgroup was defined by baseline, not by diagnosis: 'Studies whose population had a mean baseline SBP greater than 140 mm Hg were evaluated separately from those whose population had lower baseline blood pressures.' Search ran to June 26, 2008 across MEDLINE, CINAHL and CENTRAL with a language restriction - 'studies must have been written in English or German'. QUALITY high -> moderate. Upward, the authors report on themselves: 'There was only a minor degree of heterogeneity in the analyses and publication bias did not appear to influence the results', and the analysis is random-effects with a prespecified baseline-SBP split. Downward, and decisive: that bias statement cannot cover the estimate this row actually votes with - a funnel/asymmetry test on a 3-trial subgroup has no power, and the abstract does not say which analysis was tested, so whether the subgroup was assessed at all is cannot-tell. The paper's own precision statement on its headline number (95% CI 6.2 to 26.5) is a reliability fact, not a size fact. Abstract-only provenance leaves no risk-of-bias table, no trial list, no registration. A high on an abstract would also sit oddly beside sibling Kwak 2014, held at moderate on full text with PRISMA, duplicate extraction, Cochrane RoB and a null Egger visible. INDEPENDENCE, flagged not fixed: this claim carries 18 appraisal rows of which 10 are meta-analyses (Silagy 1994, Reinhart 2008, Ried 2008, Kwak 2014, Rohner 2015, Wang 2015, Xiong 2015, Ried 2016, s40735665, s41393949), taking a raw ~78% of the claim's unshrunk weight (rough tally, no group shrinkage - a flag, not the engine's number) from repeated passes over one shared 1979-2010 garlic-trial literature; Reinhart's ten trials, searched to 2008, are necessarily a near-subset of the later pools. CROSS-CONTAMINATION QUESTION RESOLVED: the '-16.33 / -9.28 mmHg' figures quoted in Kwak 2014's Discussion are genuinely Reinhart's (its ref 49) and match this paper's own abstract to rounding (16.3 / 9.3); no numbers were borrowed onto this row, and Kwak's own -4.34 / -2.36 are correctly stored separately. CONTEXT, not the basis of this grading: every later synthesis of substantially the same literature places the hypertensive effect at -4 to -9 mmHg, so -16.3 is the outlier of the row set. |
Disagree, or know a study we missed?
We grade by evidence, not opinions. The way to weigh in is to point us to a study we haven't cited (check the evidence table above first), or to flag a problem with one we have. Every submission is reviewed; if it holds up, the grade updates and shows in Science Changes Its Mind.
Opens a short form. You'll sign in with Google so submissions are tied to a real account — we don't display your identity, and we only accept a link we can verify (PubMed, DOI, ClinicalTrials.gov).
Educational only, not medical advice. Grades and scores reflect published evidence weighted by study design and quality; see the methodology.