Does a garlic supplement lower blood pressure, and if so for whom? Two gold meta-analyses now bear on this — Ried 2008 (11 placebo-controlled RCTs, hypertensive + normotensive) and Ma 2025 (12 RCTs, all hypertensive). Both agree the effect is real but modest, concentrated in hypertensives, on a surrogate endpoint, over the short term. But their agreement is a non-independent update, not two independent confirmations — Ma re-pools the older trials and adds mostly Karin Ried’s own later RCTs (independence note below) — so the confidence gain is limited.
Effect estimate
- effect_measure: absolute reduction vs placebo in mmHg (no baseline-risk-scaled absolute-risk figure — the outcome here is the BP number itself, a surrogate).
- population_and_comparator: adults in 11 RCTs, garlic (mostly standardised dried powder) vs placebo, 12-23 weeks.
- outcome: office SBP / DBP — a surrogate, not a patient-important endpoint (see below).
| Contrast | n | WMD (mmHg) | 95% CI | significance | I2 |
|---|---|---|---|---|---|
| SBP — all | 10 | -4.56 | -7.36, -1.77 | p<0.001 | 57% |
| DBP — all | 11 | -2.44 | -4.97, 0.09 | NS (p=0.06) | 83% |
| SBP — hypertensive | 4 | -8.38 | -11.13, -5.62 | p<0.001 | 0% |
| SBP — normotensive | 6 | -2.28 | -4.61, 0.05 | NS (p=0.06) | — |
| DBP — hypertensive | 3 | -7.27 | -8.77, -5.76 | p<0.001 | 0% |
| DBP — normotensive | 8 | -0.06 | -1.37, 1.25 | NS (p=0.93) | — |
The pooled SBP effect (~4.6 mmHg) is real but modest and heterogeneous; the DBP effect is not significant overall. Split by baseline BP, the heterogeneity vanishes (I2=0) and the picture is stark: the benefit lives in the hypertensive stratum and the normotensive arms are near-null.
Ma 2025 — updated hypertensive-only MA: same quantity as Ried’s hypertensive subgroup
Ma 2025 is an updated meta-analysis + trial sequential analysis of 12 placebo-controlled RCTs in which every subject was hypertensive (405 garlic, 333 placebo; RoB 2, GRADE, Egger, sensitivity analysis, TSA). Because its population is entirely hypertensive, Ma’s pooled estimate is the same quantity as Ried’s hypertensive SUBGROUP, not Ried’s all-populations pool — the load-bearing “same quantity?” check before any comparison:
| Parameter | Ried 2008 | Ma 2025 | Same quantity? |
|---|---|---|---|
| SBP, hypertensive (between-group MD, mmHg) | -8.38 (-11.13, -5.62), 4 trials, I2=0% | -8.121 (-10.95, -5.28), 12 trials, I2=48% | YES -> confirms (~8 mmHg, near-identical) |
| DBP, hypertensive (between-group MD, mmHg) | -7.27 (-8.77, -5.76), 3 trials, I2=0% | -4.256 (-5.99, -2.52), 12 trials, I2=55% | YES quantity; Ma smaller -> attenuation |
| SBP, all/mixed populations | -4.56 (-7.36, -1.77), 10 trials | (no mixed arm) | N/A — do NOT compare Ma to Ried’s -4.56 |
| Normotensive arms | SBP -2.28 NS, DBP -0.06 NS | (no normotensive arm) | N/A — Ma is silent on normotensives |
| k trials / base | 11 total | 12, all hypertensive | overlapping base (below) |
| Preparation mix | 9/11 Kwai dried powder | Kwai + Allicor + Kyolic AGE + Japanese trad | Ma broader/newer |
The trap avoided: Ma’s headline -8.121 mmHg SBP looks like it nearly doubles Ried’s headline -4.56 — but those are different quantities (Ma all-hypertensive vs Ried all-populations). Matched like-for-like, Ma’s hypertensive SBP (-8.12) essentially reproduces Ried’s hypertensive subgroup (-8.38): garlic’s ~8 mmHg SBP effect in hypertensives is stable across 17 years and more trials. DBP attenuates (Ma -4.26 vs Ried’s hypertensive -7.27, landing between Ried’s all-pool -2.44 NS and hypertensive -7.27). Ma adds no normotensive arm, so it neither confirms nor refutes Ried’s near-null normotensive result — the baseline-BP effect-modification claim still rests on Ried alone. «…the SBP in hypertensive patients was significantly reduced compared to that in the placebo group (difference in mean: −8.121, 95% CI: −10.95 to −5.28… DBP… difference in mean: −4.256, 95% CI: −5.99 to −2.52…)»
Not an independent confirmation — a non-independent update (type-F, not type-E)
A naive read — a second gold MA agrees, so confidence rises (type-E) — fails the independence test. The two MAs do not share authors (Ma/Zhang/Jia vs Ried/Frank/Stocks/Fakler/Sullivan), but independent backing requires independent trials, and Ma’s evidence base is not independent of Ried:
- Four of Ma’s 12 trials are Karin Ried’s own RCTs (Ried 2010, Ried 2013, Ried 2016, plus one further Ried trial — the Kyolic aged-garlic-extract series) — i.e. a third of the pooled data is the incumbent MA’s first author’s own primary evidence.
- The older canonical Kwai trials (Auer 1990, Vorberg & Schneider, De Santos & Gruenwald, Kandziora) are the standard shared base of garlic-BP MAs and almost certainly overlap Ried 2008’s 11-trial set.
So Ma’s agreement with Ried is expected and largely self-referential, not convergence from a separate line of evidence. This is a type-F update (more/newer trials, plus GRADE + RoB 2 + TSA that Ried 2008 predated, refreshing the hypertensive estimate) — NOT a type-E independent confirmation. The two MAs agree confidence gain is therefore limited.
Quality caveats on Ma 2025 (why confidence: stays low)
- Internal contradiction on publication bias. The results section reports «the Eggers regression analysis indicated the absence of publication bias» (SBP p=0.29, DBP p=0.26), yet the limitations state «there was significant publication bias in both the SBP and DBP analysis groups». — the paper is internally inconsistent; at minimum publication bias is not cleanly excluded.
- Overclaim. Ma concludes garlic «may be recommended by clinicians for the management of hypertension» and, via TSA, that «further studies are unnecessary» — a strong claim for a surrogate-only base with zero cardiovascular-outcome trials and moderate heterogeneity (SBP I2=48%, DBP I2=55%). Its further studies are unnecessary claim cannot hold for CV outcomes never trialled (Surrogate Outcomes).
- Gaps persist. Ma pools varied doses (188-2,400 mg/d) but runs no dose-response meta-regression, and measures only office BP — so both of Ried’s gaps (dose-response, CV endpoint) remain open. Low-profile journal (Asian Biomed / Sciendo).
Baseline BP is the effect-modifier (route a + route b)
Meta-regression makes baseline BP a significant predictor of the reduction (DBP R=-0.316, p=0.02; SBP R=-0.151, p=0.03 in the body — the abstract prints a sign-flipped SBP R=0.057, an internal typo; the direction and both p-values are consistent). This is stronger than a route-(a) baseline-risk story: the relative/absolute reduction itself is larger in hypertensives (~8 vs ~2 mmHg SBP), so garlic shows genuine effect modification (route b) by baseline BP, not merely a constant relative effect applied to a higher-risk group. See Baseline Risk and the Relative-Absolute Split. Practically: for a normotensive person the expected BP effect is small and unproven; for a hypertensive it is meaningful. the decision therefore stratifies on baseline BP.
Dose-response: not established
Dosage and duration were tested as continuous predictors and were not significant. No knee, plateau, or threshold is located over the studied range (600-900 mg/d standardised powder; 12-23 weeks) — the authors explicitly call for a dose-response study. So no minimum effective dose can be stated; the effect is only characterised at the doses trialled, not below or above them.
It is a supplement isolate, not eating garlic
9 of 11 trials used standardised dried garlic powder (mostly Kwai); 1 aged garlic extract, 1 distilled oil. The active agent is credited to allicin (via alliinase) and H2S. Two consequences. First, this evidence is about a dosed, blindable isolate, not the food on a plate — the same isolate-vs-food gap seen elsewhere (Is the Food Category Doing Any Work): the blindable form is what got the clean RCT. Second, low allicin release from commercial supplements is a documented problem (Lawson & Wang), so the labelled dose overstates the delivered active agent — a preparation-specific caveat that blocks transporting the effect to an arbitrary garlic pill.
BP is a surrogate — the transmission to hard outcomes is untested here
Ried measured only office BP. The authors state «larger scale long-term trials are needed to test the effectiveness of garlic on cardiovascular outcomes». So garlic->stroke / CV events / mortality is not shown by this source. The causal transmission BP->CV events is itself an evidenced claim held elsewhere (Blood Pressure Lowering and Cardiovascular Events): a ~5/2-3 mmHg population shift maps onto a material CV-event reduction if the drug-trial BP->outcome relationship transports to a garlic-induced BP drop. That if is the open question — a marker moving is not an outcome moving (Surrogate Outcomes). treat the CV-outcome benefit as plausible-but-unproven, carried by borrowed BP->CV evidence, not by garlic trials.
Layer-1 sizing: net of a mature drug substitute
Ried notes garlic’s SBP effect is «comparable to» commonly-prescribed antihypertensives (beta-blockers ~5 mmHg SBP, ACEI ~8 mmHg SBP, ARBs ~10.3 mmHg DBP) (Ried et al., 2008). Under the substitution rule (Layer 1 - Ranking Interventions for a Stratum), the marginal garlic rock is sized net of these drugs: for a diagnosed hypertensive, mature low-harm antihypertensives already capture most of the BP->CV benefit, so garlic’s additional value on that single channel is small. Garlic’s rock is larger only where the drug is not taken (untreated / drug-averse / mild elevation) or as an adjunct — and even then the durability and delivered-dose caveats apply. This is single-channel substitution; it does not shrink any non-BP effect garlic might have.
Synthesis / status
Two gold MAs (Ried 2008, Ma 2025), confidence: low. The surrogate SBP effect in hypertensives
(~8 mmHg) is reproduced across both and 17 years — the more secure part; everything patient-important
is not. Value types from the Ma differential: F — Ma refreshes/bounds Ried’s dated pre-GRADE
estimate with a larger, newer, GRADE/RoB2/TSA-rated hypertensive pool (composite beats either alone);
E-refuted — the apparent second-MA confirmation is adjudicated NON-independent (no shared MA
authors, but a third of Ma’s trials are Ried’s own + overlapping older Kwai base), so it is not
banked as type-E and the confidence gain is limited; G — persisting gaps: no dose-response, no
CV-outcome endpoint, and (Ma has no normotensive arm) the baseline-BP effect-modification still rests
on Ried alone. No tension filed — Ma confirms rather than contradicts.