Of the cardiometabolic interventions the wiki holds, BP-lowering is the one whose hard-outcome benefit is proven to reach primary prevention — the partial exception to Cardiometabolic Interventions and Hard CV Outcomes in Low-Risk People. The largest randomised evidence base (BPLTTC 2021, «344 716 participants from 48 randomised clinical trials») settles two long- contested questions: the benefit does not require pre-existing heart disease, and it does not vanish at lower baseline blood pressure. (Blood Pressure Lowering Treatment Trialists Collaboration, 2021)
The effect, and why its independence from baseline is the point
«a 5 mm Hg reduction of systolic blood pressure reduced the risk of major cardiovascular events by about 10%, irrespective of previous diagnoses of cardiovascular disease, and even at normal or high–normal blood pressure values».
Per 5 mmHg SBP reduction, the hazard ratio for a major CV event was 0.91 (0.89-0.94) in people without prior CVD and 0.89 (0.86-0.92) in those with it (Blood Pressure Lowering Treatment Trialists Collaboration, 2021). Across seven baseline-SBP strata from <120 to >=170 mmHg there was «no reliable evidence of heterogeneity», and «in no subgroup was the HR above 1·0». Component reductions per 5 mmHg: stroke 13%, heart failure 13%, ischaemic heart disease 8%, CV death 5%.
This is what distinguishes BP-lowering from the other cardiometabolic levers. Lifestyle weight loss was null on hard events even where it was achieved (Does Weight Loss Reduce Cardiovascular Events); an obesity drug’s CV benefit is unproven outside established disease (Semaglutide for Cardiovascular Risk in Obesity); a statin’s is real but decays to below the action threshold at low risk (Statins for Primary Prevention and the Power of Zero CAC). BP-lowering alone has a directly proven relative benefit in primary prevention, at an IPD-meta-analytic warrant.
But the ceiling’s mechanism survives — treat on absolute risk, not a BP number
The paper does not conclude lower everyone’s BP. Its explicit decision rule is the Baseline Risk and the Relative-Absolute Split applied to blood pressure:
physicians should «emphasise its importance on reducing cardiovascular risk rather than focusing on blood pressure reduction itself»; the finding «calls for consideration of blood pressure-lowering treatment for any individual who has a sufficiently high absolute risk of cardiovascular disease».
And the guard against over-reading, in the authors’ own words: the uniform relative effect «does not necessarily mean that it is worthwhile treating every patient» — decisions «will require consideration of… an individual’s overall risk of future cardiovascular events, potential risk of adverse effects, the cost of treatment, and patient preferences». They even «caution against using the reported absolute risk differences from clinical trials for making policy decisions» — the trials’ populations were higher-risk (absolute rates «31·9… in the comparator group and 25·9… in the intervention group» per 1000 person-years without prior CVD, i.e. a ~3.2%/yr comparator-group baseline, far above a low-risk primary-prevention person). (Blood Pressure Lowering Treatment Trialists Collaboration, 2021)
So BP-lowering refines, not refutes, the ceiling. Its relative benefit is proven where weight-loss and GLP-1 CV benefit are not — but the absolute benefit still scales with baseline risk, exactly the ceiling’s mechanism, and the paper legislates the same rule (treat on absolute risk). A constant ~10% per 5 mmHg on a low absolute risk is a small absolute gain; on a high one it is large.
What it does not settle
- Not the target. BPLTTC standardises to a fixed 5 mmHg reduction and shows a consistent relative effect; it «is unable to directly address» the optimal magnitude or a target BP. How low to go is a separate question — now partly held via SPRINT below.
- Not the drug/adverse-effect trade-off for an individual — adverse effects, cost and preferences are named as decision inputs but not quantified — SPRINT supplies the harm side, below.
(Blood Pressure Lowering Treatment Trialists Collaboration, 2021)
SPRINT — the intensive-target trial that operationalises treat on absolute risk
BPLTTC is the meta-analysis and carries the per-mmHg claim; SPRINT is one landmark RCT, read inside that framing as the worked target-trial. It randomised 9,361 high-CV-risk, non-diabetic hypertensives to a systolic target «less than 120 mm Hg (intensive)» vs «less than 140 mm Hg (standard)» (achieved ~121 vs ~136), and stopped early (median 3.26 y):
- Intensive won on hard outcomes: primary composite (MI/ACS/stroke/HF/CV-death) «hazard ratio … 0.75; 95% CI 0.64 to 0.89»; and all-cause mortality «hazard ratio, 0.73; 95% CI 0.60 to 0.90» — a 25% and 27% relative reduction, «number needed to treat … to prevent one primary outcome event was 61» over 3.26 years. (SPRINT Research Group, 2015)
- It confirms BPLTTC’s own logic rather than competing with it. SPRINT selected a high-risk population — exactly where BPLTTC says the absolute benefit of a given relative effect is largest — and the ~15 mmHg achieved gap, at BPLTTC’s ~0.91 per 5 mmHg, predicts a composite HR near 0.75, which is what SPRINT observed.
- The harm side BPLTTC did not quantify: intensive treatment raised serious «hypotension, syncope, electrolyte abnormalities, and acute kidney injury or failure, but not … injurious falls». (SPRINT Research Group, 2015) So a lower target buys real events in the high-risk but at a real adverse-event cost — the target decision is a net-effect judgement, not lower is always better.
A clean refutation of the observational J-curve
The intro records the standing controversy: registries show a «J-shaped association» between BP and events (lowest risk ~130/75, implying harm below), which had been read as a reason not to lower BP far. The randomised evidence shows benefit «down to less than 120 mm Hg» with no rising HR in any low-BP stratum — so the observational J-curve is not a treatment effect. (Blood Pressure Lowering Treatment Trialists Collaboration, 2021) A worked instance of The U-Shaped Association Artifact: the protective-looking lower arm of an observational curve does not survive the randomised (here, interventional) check.
A dietary BP-lowering route that reached hard outcomes — SSaSS [2026-08-04, Neal]
The page’s caution below is that a lifestyle-induced BP drop transmitting to events is «an assumption, not the same evidence» as the pharmacological trials here. SSaSS (Neal 2021) is the closest test the corpus holds: a potassium-enriched salt substitute (a dietary intervention) cut stroke (RR 0.86), major CV events (RR 0.87) and death (RR 0.88) in 20,995 high-risk adults, with an achieved SBP fall of −3.34 mmHg. (Neal et al., 2021) So a diet-route BP reduction can reduce hard events — but two caveats keep it from being a clean confirmation of the sodium→events chain: (1) the salt substitute moves sodium and potassium together (a joint lever, not sodium alone — Sodium Intake and Blood Pressure); (2) the event benefit outruns the BP-channel prediction — this page’s ~13% stroke reduction per 5 mmHg predicts RR ~0.91 from −3.34 mmHg, while SSaSS observed 0.86, consistent with a BP-independent electrolyte effect or the measured BP fall understating the sustained one (adherence dilution). The BPLTTC transmission model is corroborated in direction; SSaSS does not let the diet-route magnitude be read straight off the BP fall. (inferred from Neal et al., 2021)
Another lifestyle BP route — alcohol reduction, and the modelled-events gap made explicit [2026-08-20, Roerecke]
Roerecke’s 2017 SR+MA of 36 alcohol-reduction RCTs (2865 participants), GRADE-high, is a second behavioural BP lever — and it makes this page’s lifestyle-BP-drop-to-events-is-an-assumption caveat concrete. The BP effect is real and dose-dependent but shown only in heavier drinkers: no significant BP change at 2 drinks/day or fewer, then dose-dependent above (SBP -1.18 at 3 drinks/day rising to -5.50 [-6.70 to -4.30] at >=6 drinks/day; overall -3.13 [-3.93 to -2.32]). (Roerecke et al., 2017) Qualified 2026-10-22: the <=2 drinks/day band (SBP -0.18, -1.02 to 0.66) shows no fall, not a zero; it pools five trials, by weight mostly Mori 2015 (a 4-week crossover). (Roerecke et al., 2017) A dose-response meta-analysis of seven cohorts finds SBP rose 1.25 mmHg (0.49 to 2.01) more over 4-12 years of follow-up at a baseline intake of 12 g/day than with no alcohol. (Di Federico et al., 2023) The trials measure BP level after weeks to months of changed intake, the cohorts the difference in BP change over 4-12 years by baseline intake; matched on quantity the two do not clash, unless the cohort difference is read as a level effect that stopping would reverse within weeks, which neither source shows; even then the SBP intervals overlap and a clash appears only on overall DBP -> Does Light Drinking Raise Blood Pressure. The magnitude is «similar to that of other health behaviour changes, such as physical activity» or weight-loss diets. (Roerecke et al., 2017)
Crucially, Roerecke did not measure CV events — it modelled them: its «more than 7000 inpatient hospitalisations and 678 cardiovascular deaths» projected for the UK rest on «standard methods for comparative risk assessment analyses (eg, Global Burden of Disease studies)», run «assuming persistence of lower alcohol consumption within 1 year, and no lag time of eff ects on CVD outcomes». (Roerecke et al., 2017) So the alcohol-reduction lever is exactly the case this page flags — a lifestyle BP drop whose event payoff is assumed through a risk model, not observed — unlike SSaSS, the one lifestyle route that reached measured hard endpoints. Read against BPLTTC’s ~0.91 per 5 mmHg, a -5.50 mmHg fall in a heavy drinker predicts a ~10% CV-event reduction if the pharmacological transmission transfers. Full dose-response -> Alcohol and Mortality and Vascular Disease. (inferred from Roerecke et al., 2017)
A third lifestyle BP route — structured exercise, and the drug head-to-head that does not exist [2026-08-20, Naci]
Naci’s 2019 network meta-analysis (Br J Sports Med, gold-tier) is the drug-as-realistic-alternative comparator the telos calls for: it places structured exercise and antihypertensive drugs in one network on SBP-lowering. It pooled «391 RCTs, 197 of which evaluated exercise interventions (10 461 participants) and 194 evaluated antihypertensive medications (29 281 participants)» — but the head-to-head does not exist: «No RCTs compared directly exercise against medications» (Naci et al., 2018). Every exercise-vs-drug number below is INDIRECT (network-inferred across disjoint trials), and the authors call the linking assumption «tenuous» (below).
The naive comparison is confounded by baseline SBP — and the confound is the finding. Across all populations, drugs beat exercise: medications «−8.80 mmHg, 95% CrI −9.58 to −8.02» vs exercise «−4.84, 95% CrI −5.55 to −4.13» from baseline vs control, a «−3.96, 95% CrI −5.02 to −2.91» advantage to drugs (Naci et al., 2018). But the two arms were measured in different populations: «the mean SBP at baseline was 132 mmHg for participants in the RCTs of exercise interventions, it was consistently over 150 mmHg in medication RCTs» (Naci et al., 2018). SBP-lowering scales with baseline SBP, so a drop measured at ~132 mmHg (mostly normotensive) is not the same quantity as one measured at ~157 mmHg (all hypertensive) — the parameter table makes this explicit:
| Naci comparison | Effect vs control (mmHg, 95% CrI) | Population / baseline SBP | Same quantity? |
|---|---|---|---|
| All exercise | −4.84 (−5.55 to −4.13) | exercise trials, mean baseline 132 | reference |
| All medications | −8.80 (−9.58 to −8.02) | drug trials, baseline >150 | NO — different baseline SBP |
| Meds vs exercise (all pops) | −3.96 (−5.02 to −2.91), drugs superior | mixed (normotensive vs hypertensive) | NO — the confounded, indirect contrast |
| Exercise, restricted to hypertensive >=140 | −8.96 (−10.27 to −7.64) | 56 trials, 3508 people, baseline >=140 | YES — baseline now matched to drugs |
| Meds vs exercise, hypertensive >=140 | 0.18 (−1.35 to 1.68), no difference | both hypertensive >=140 | YES — matched; the honest “similar” claim |
Restricting exercise to the hypertensive stratum nearly doubles its estimated effect (−4.84 -> −8.96) and erases the drug advantage: «We did not observe a difference between the SBP-lowering effects of medications and exercise (0.18, 95% CrI −1.35 to 1.68)» (Naci et al., 2018). Naci names the mechanism himself in the limitations (the counter-passage): «Transitivity may be tenuous given the typically different levels of SBP in medication versus exercise trials. However, when we tried to match trial populations more closely, the gap between medication and exercise in efficacy diminished» (Naci et al., 2018). This is a worked instance of Baseline Risk and the Relative-Absolute Split on a continuous outcome — the absolute SBP reduction is a function of baseline SBP, so an unmatched cross-intervention comparison misleads. (inferred from Naci et al., 2018)
The equivalence is real but doubly hedged — do not read it as “exercise = drugs” plainly. The authors state it only «Assuming equally reliable estimates» (Naci et al., 2018), and the estimates are not equally reliable: the exercise evidence is far sparser in the matched stratum (56 trials / 3508 people vs 194 drug trials / 29 281), higher risk of bias (17/20 sampled exercise RCTs at high risk of performance and detection bias from lack of blinding, vs 1/20 drug RCTs), and — since food/behaviour cannot be blinded — this weakness is structural, not fixable. The conclusion carries the hedge: the exercise effect «appear similar to that of commonly used antihypertensive medications, but this is tempered by the observed differences in study population characteristics» (Naci et al., 2018). One bias runs the other way, and Naci names it: the authors say they may have underestimated exercise’s SBP effect, because «We excluded trials of medications and dosages that were considered to be ineffective or suboptimal by the BNF, even though we did not exclude any exercise trials on this basis» (Naci et al., 2018). The drug arm was screened for effective doses and the exercise arm was not, so the matched-stratum parity is not obviously flattering to exercise.
Which type. The single types lower SBP about equally; combined endurance plus resistance training beats dynamic resistance alone (−2.98, 95% CrI −5.04 to −0.93), and in the >=140 mmHg sensitivity analysis it beats endurance, resistance and isometric training alike (Naci et al., 2018), reaching −13.51 (−16.55 to −10.45) vs control (Naci et al., 2018). Naci notes that only a small subset of trials tested the combination and their frequency and duration were not consistently reported (Naci et al., 2018), so the edge may partly be more total training. The type groups differ in baseline SBP (125 mmHg resistance, 135 combination; Table 1), so part of the all-populations edge may be baseline; the edge persisting at >=140 mmHg is the better test. The decision reading and its caveats are on Is the Exercise Modality Doing Any Work.
Decision relevance. For a hypertensive person, structured exercise (endurance or dynamic resistance) is a credible-magnitude SBP lever — indirectly comparable to a first-line drug in the matched stratum — so it is a defensible adjunct, and a candidate alternative where a drug is declined or not yet indicated. Two guards: (1) the equivalence holds only at hypertensive baselines; at a normotensive baseline the exercise SBP effect is smaller (−4.84) and the drug comparison was never the point; (2) SBP here is a surrogate — Naci measured no CV events. The event payoff runs through this page’s BPLTTC channel (~10% per 5 mmHg), which is evidenced for pharmacological lowering; whether an exercise-induced SBP drop transmits at the same rate is the same assumption flagged for the sodium, alcohol and salt-substitute levers above. Exercise also carries large off-target benefits (fitness, glycaemia, weight) that a BP-only comparison ignores -> Physical Activity Dose and Mortality, Cardiorespiratory Fitness and Mortality. The head-to-head RCT that would settle the comparison directly does not exist — a standing gap. (inferred from Naci et al., 2018)
Second outcome — antihypertensive treatment also lowers dementia risk
Hypertension is one of the 14 modifiable dementia risk factors, and the treatment lever carries over -> Dementia Prevention and Modifiable Risk Factors. The Commission holds this observationally: «those with untreated hypertension had a 42% higher risk of dementia than healthy controls (HR 1.42; 1.15-1.76), but this risk was attenuated or lost with treatment» (HR 1.13, 0.99-1.28). (Livingston et al., 2024)
The interventional effect, de-secondhanded first-hand (Peters 2022, F-refinement [2026-09-04]). The
IPD OR the Commission borrows is now held at source: Peters 2022, an IPD MA of five double-blind
placebo-controlled antihypertensive RCTs (HYVET, SYST-EUR, PROGRESS, ADVANCE, SHEP; «28 008 individuals
recruited from 20 countries», 861 dementia cases) (Peters et al., 2022). «Multilevel logistic regression reported an adjusted odds ratio 0.87 (95% confidence
interval: 0.75, 0.99) in favour of antihypertensive treatment reducing risk of incident dementia with a
mean BP lowering of 10/4 mmHg» (Peters et al., 2022) —
«Class I evidence» (Peters et al., 2022). In absolute terms
dementia occurred in «403 (2.9%) and 458 (3.3%)» of active vs placebo (>=2y follow-up)
(Peters et al., 2022). That is an ARD of ~0.4 percentage
points, i.e. an NNT of ~250 over the median follow-up (inferred from Peters et al., 2022); and it is conservative
because the trials stopped early on the CV endpoint (dementia accrues slower). This is NOT an [E-independent] corroboration of
this page’s CV-events evidence: Peters’ five trials sit inside BPLTTC’s 48-trial base and the author lists
overlap (Chalmers, Woodward, Anderson — George Institute), so the shared RCT substrate defeats independence
on the BP-lowering mechanism — what is new is the OUTCOME (dementia), a new-endpoint extension. Full
facets (mediation 53%, no effect-modification, cognitive-decline null) on
Dementia Prevention and Modifiable Risk Factors.
(Blood Pressure Lowering Treatment Trialists Collaboration, 2021; inferred from Peters et al., 2022)
The observational U survives no better here than for CV events. Peters’ abstract records the same J/U-curve puzzle — «Observational studies indicate U-shaped associations of blood pressure (BP) and incident dementia in older age, but rando-mized controlled trials of BP-lowering treatment show mixed results on this outcome in hypertensive patients» (Peters et al., 2022) — and randomization erases it: «no evidence of a U-shaped re-lation of the effect at any age», with a monotone «linear relationship between lower risk of dementia and lower BP, down to at least 100 mmHg systolic and 70 mmHg diastolic» (Peters et al., 2022). The 2024 Lancet Commission’s target — «maintain systolic BP of 130 mm Hg or less from age 40 years» — is a dementia-prevention framing of the same midlife-BP lever. As with CV events, MR studies suggesting BP is protective are discounted as survival-bias artifacts, consistent with this page’s treat-on-absolute-risk conclusion -> The U-Shaped Association Artifact.
preDIVA — BP-lowering in the elderly is cognitively SAFE, and the dementia benefit is stratum-specific
[2026-08-07]. preDIVA (Moll van Charante 2016), a 6.7-yr nurse-led multidomain vascular-care cluster-RCT
in 3526 unselected elderly (70-78), lowered systolic BP by -2.06 mm Hg but did not reduce all-cause
dementia (HR 0.92, 0.71-1.19) (van Charante et al., 2016). Two decision-relevant reads: (i) it removes a standing worry — «our results mitigate fears that
antihypertensive treatment in older age groups evokes cognitive decline … they show that such an
intervention is safe, which is in accordance with findings from the recent SPRINT trial»
(van Charante et al., 2016), so BP-lowering in
the over-70s carries no cognitive-harm penalty (an inverse late-life BP-dementia association is a
frailty/reverse-causation artifact -> The U-Shaped Association Artifact); (ii) any dementia benefit is
concentrated where BP is genuinely untreated — the adherent-untreated-hypertension subgroup reached
HR 0.54 (0.32-0.92) (van Charante et al., 2016),
a route-(b) effect-modification hypothesis (post-hoc, adherence-selected), consistent with this page’s
treat-on-absolute-risk logic: the lever pays where the deficit is real, not in an already-well-managed
population. Full multidomain-trial context -> Multidomain Lifestyle Intervention and Cognitive Decline.
A fourth lifestyle BP route — vegan dietary pattern (observational, very-low COE) [2026-08-29, Landry]
Landry’s 2024 umbrella adds a diet-pattern BP lever: pooling 16 observational studies (Benatar), «vegan diets were associated with lower systolic BP [MD (95 % CI): −2.56 mmHg (−4.66, −0.45); I2=83 %]», non-significant for diastolic, and «largely driven by participants from non-Eastern Asian countries» (Landry et al., 2024). This is the weakest-warranted of the four lifestyle routes on this page: observational only (the controlled-trial BP evidence was set aside because «19 of these studies targeted adults with diagnosed diseases»), GRADE very low, I2=83%, and healthy-adherer confounded — vegans differ on many mortality levers at once -> Vegetarian Dietary Patterns and Mortality. Read through this page’s ~10%-per-5-mmHg channel, −2.56 mmHg predicts only ~5% CV-event reduction if the pharmacological transmission transfers — and the vegan diet simultaneously moves LDL, BMI and CRP, so the BP channel is not separable as the cause. A dietary-pattern BP association, not a titratable lever; the SBP number is a surrogate here as for the sodium/alcohol routes above. (inferred from Landry et al., 2024)
A fifth lifestyle BP route — yoga, the least-warranted of them [2026-09-23, Geiger]
Geiger’s 2025 SR+MA (30 RCTs, 2283 prehypertensive-to-hypertensive adults, gold-tier) adds yoga — a stress-mitigation lever (Stress Management and Cardiometabolic Health) — as a BP route, and it is the weakest-warranted route on this page, below even the vegan association. The headline against waitlist is respectable (SBP «MD = -7.95 mmHg», DBP -4.93) but three things collapse it as a titratable events lever: (1) GRADE very low, I2=90%; (2) against an active control the SBP effect is not significant (-4.16, 95% CI -10.76 to 2.44, p=0.22); (3) under 24h-ABPM the SBP effect disappears (only DBP survives) (Geiger et al., 2025). Read through this page’s ~10%-per-5-mmHg channel, even the waitlist -7.95 would predict ~16% CV-event reduction if the pharmacological transmission transferred — but that transmission is unproven for any lifestyle route (the caveat that governs the sodium/alcohol/vegan routes above), the fair-comparator estimate is null, and the SBP fall is partly a clinic-BP measurement artifact. Layer-1 sizing: yoga is a small, uncertain refinement, not a big rock — its apparent big-rock size (-7.95) is a waitlist/expectancy inflation, and net of the best realistic substitute (any supervised activity, which the active-control arms already supply) the yoga-specific SBP increment is not distinguishable from zero. It earns a place only as a low-cost adjunct for someone who will actually do it, never as a substitute for salt reduction, medication, or structured exercise. (Blood Pressure Lowering Treatment Trialists Collaboration, 2021; inferred from Geiger et al., 2025)
A sixth route — garlic, a big clean surrogate mover with no hard-outcome warrant [2026-09-25, weave: Ried + Ma]
Garlic supplementation is a supplement BP route (Garlic and Blood Pressure), and placing it beside the lifestyle routes above exposes a structure they only imply: among the non-drug BP levers, surrogate magnitude and hard-endpoint transmission run inverse — a big office-SBP mover need not be a well-warranted route. In the hypertensive stratum garlic’s office SBP effect is large and unusually clean — -8.38 mm Hg (-11.13, -5.62), I2=0, placebo-controlled across 4 trials (Ried), replicated at «-8.121, 95% confidence interval [CI]: -10.95 to -5.28» mm Hg (Ma, 12 trials) (Ma et al., 2025; Ried et al., 2008) — the largest held among placebo-controlled supplement levers, and on par with the biggest lifestyle movers. It is not the single largest non-drug lever: structured exercise restricted to the same hypertensive stratum is larger at -8.96 (Naci, above), but exercise is unblindable, so its estimate carries an expectancy component garlic’s placebo-controlled trials remove — the two are not the same quantity (the baseline-and-comparator caution the Naci section already makes). Yet garlic is the least validated of the set: office BP only (no ambulatory pool), no cardiovascular endpoint ever trialled, and publication bias unresolved (Ma’s results and limitations sections contradict each other on it). So a lever with one of the biggest, cleanest surrogate numbers is the one whose transmission to events — the caveat governing every route on this page — is most completely unproven.
| Non-drug BP lever (stratum) | Office SBP (mm Hg) | Same quantity? | ABPM tested? | Hard CV endpoint? | Pub-bias clean? |
|---|---|---|---|---|---|
| Exercise, hypertensive (Naci) | -8.96 (-10.27, -7.64) | office MD, unblindable | — | observational only | unblindable |
| Garlic, hypertensive -> Garlic and Blood Pressure | -8.38 (Ried) / -8.12 (Ma) | office MD, placebo-ctrl | no | never trialled | uncertain |
| Yoga, mixed | -7.95 waitlist; -4.16 NS active | office MD, waitlist-inflated | vanishes on ABPM | no | — |
| Slow breathing | -5.62; -2.14 NS at low RoB | office MD | unreported | no | — |
| Nitrate (beetroot) -> Dietary Nitrate and Blood Pressure | -4.4 (-5.9, -2.8) (Siervo) | office MD | — | no | — |
| Urate lowering -> Uric Acid Lowering and Blood Pressure | ~-3.3 proxy; primary is SDM 0.321 | NOT mm Hg (SDM) | no | no | Egger asym. p=0.032 |
| Salt substitution (SSaSS) | -3.34 (Neal) | office MD | — | yes, measured | — |
| BP drug (BPLTTC) | per-5-mm-Hg -> ~10% events | events metric | — | yes, measured | — |
Nitrate row: Siervo’s pooled systolic change, printed «24.4 mm Hg (95% CI: 25.9, 22.8)» in the chunked text, where each minus sign was extracted as a leading “2” (so -4.4, CI -5.9 to -2.8) (Siervo et al., 2013).
The ordering is the finding, but it must be read by stratum-matched, comparator-matched quantities, not by the bare number — the same discipline the Naci section applies (restricting exercise to hypertensives nearly doubled its effect). Once matched, the inverse pattern holds among the non-drug levers: the levers with the biggest, cleanest office-SBP moves (garlic placebo-controlled; exercise, though unblindable) have the weakest endpoint warrant — none trialled, or observational only — while the one non-drug row validated on measured events (SSaSS) carries a modest surrogate move, -3.34. Corrected 2026-10-07: this sentence earlier counted the drug channel as a second modest-move row, but BPLTTC’s per-5-mm-Hg is a standardising unit, not an achieved drop, and drugs lower SBP a lot (-8.80, -9.58 to -8.02, vs control across Naci’s drug trials, whose class mean baselines were 156-160 mmHg; Table 1) (Naci et al., 2018); drugs are big on the surrogate and validated on events, so they sit outside the pattern, which rests on one measured-endpoint contrast. Big-and-clean on the surrogate and strong-on-the-endpoint are different levers. (Urate lowering is the one non-comparable cell: Qu reports SDM, so its ~3 mm Hg is a design-discounted proxy from Agarwal 2013, not the same quantity.) Layer-1 sizing: garlic ranks as a small rock despite the big surrogate number — net of a mature low-harm BP drug (which is validated on events) the marginal hard-outcome benefit garlic adds is unmeasured and, given the missing endpoint and unresolved pub-bias, cannot be assumed nonzero; Ma’s «further studies are unnecessary» claim cannot hold for CV outcomes that were never trialled. Garlic earns a place only as a low-cost adjunct in the hypertensive stratum for someone who will take it, never as a substitute for a route with a measured endpoint.
(Blood Pressure Lowering Treatment Trialists Collaboration, 2021; Chaddha et al., 2019; Geiger et al., 2025; Ma et al., 2025; Naci et al., 2018; Neal et al., 2021; Norouzzadeh et al., 2025; Qu et al., 2016; inferred from Ried et al., 2008)
Held levers that barely move SBP — fibre, whole grains, sweeteners
Three exposures the wiki holds for other reasons have randomised SBP results too. One is a small real effect, the rest show no clear effect. Reynolds 2019 admitted people with «mild to moderate hypertension» but excluded those on medications affecting the outcomes, and reports no hypertensive-stratum estimate (Reynolds et al., 2019). So the comparison with the hypertensive rows in the table above is not stratum-matched.
| Exposure (trial contrast) | SBP effect (mm Hg) | Source certainty | Home page |
|---|---|---|---|
| Higher vs lower dietary fibre, general population | -1.27 (-2.50 to -0.04), 15 trials | Moderate | Dietary Fibre and Health |
| Higher vs lower fibre, people with diabetes or prediabetes | -1.86 (-4.85 to 1.12), NS, 12 trials | — | Dietary Fibre and Health |
| Higher vs lower whole grains (Reynolds) | -1.01 (-2.46 to 0.44), 8 trials | Moderate | Whole Grains Refined Grains and Pulses |
| Whole-grain vs refined-grain diets (SACN) | 0.2 (-1.6 to 2.0), p=0.85 | Moderate (SACN: no effect) | Whole Grains Refined Grains and Pulses |
| Non-sugar sweeteners (RCTs) | -1.33 (-2.71 to 0.06), 14 trials | Moderate | Non-Sugar Sweeteners |
(Reynolds et al., 2019) (A. N. Reynolds et al., 2020) (Scientific Advisory Committee on Nutrition, 2015) (World Health Organization, 2023)
Fibre lowers SBP by a real but small amount, at moderate certainty. Read through BPLTTC’s per-5-mm-Hg channel, a sustained -1.27 would predict roughly 2.5% fewer CV events in relative terms, and only if a diet-induced drop transmits like years of drug-induced lowering. SACN listed fibre and blood pressure under insufficient RCT evidence (Table 8.2), a table kept separate from inconsistent evidence (Table 8.3) (Scientific Advisory Committee on Nutrition, 2015). Its methods annex defines the category: «A2.21 The evidence was normally considered as insufficient where there were fewer than three studies of the same type. Alternatively, if there were more studies but they were of poor quality, the evidence may have also been considered as insufficient.» (Scientific Advisory Committee on Nutrition, 2015) So SACN’s fibre-BP entry means too few or too weak qualifying trials, not conflicting ones. Superseded 2026-10-07: the earlier wording read 8.158’s combined definition (two or fewer studies, or too inconsistent) as leaving open which applied; the separate tables settle it. It also graded fibre by type and found one effect: «8.79 An effect is demonstrated for higher oat bran and β-glucan consumption on reducing systolic blood pressure (-2.86 mmHg, 95% CI -4.87, -0.85; p<0.01).» (Scientific Advisory Committee on Nutrition, 2015). So the gap with Reynolds may come from SACN splitting fibre by type under different inclusion rules, rather than from reading the same trials differently. The diabetes-fibre estimate is too imprecise to call either way: its interval runs from about -5 to +1. Whole grains and sweeteners show no clear effect. SACN’s whole-grain pool rests mostly on one trial: «8.136 One trial, contributes 92% to the pooled estimate for systolic blood pressure and 97% for diastolic blood pressure (Howard et al., 2006b). This trial elicited a very small increase in whole grains consumption (less than one serving per day) and also resulted in weight loss differences between experimental groups.» (Scientific Advisory Committee on Nutrition, 2015). Legume intake got no BP conclusion at all (insufficient RCT evidence, Table 8.2, beside fibre and breakfast cereals) (Scientific Advisory Committee on Nutrition, 2015) -> Whole Grains Refined Grains and Pulses. None of these effects is large enough to rank as a BP lever; their value, where they have any, runs through other outcomes on their home pages.
Sweeteners carry a cohort-trial split. WHO’s cohorts link sweetener use to hypertension (HR 1.13, 1.09-1.17, low certainty), while its trials show no SBP rise. In its body-weight discussion WHO says «The majority of RCTs assessing NSS lasted 3 months or less, and the small number that lasted more than 3 months gave inconsistent results.» (World Health Organization, 2023); the 14 SBP trials’ durations are not given separately, but if they follow that pattern they cannot test a harm that takes years to show as incident hypertension. WHO names reverse causation as a possible explanation: «Reverse causation suggests that those already at elevated risk of disease initiated or increased use of NSS because of their risk status, rather than NSS leading to increased risk in otherwise healthy or low-risk individuals.» (World Health Organization, 2023). It then limits that reading: «However, efforts taken by the authors to address reverse causation and confounding in most studies suggest that these phenomena are not the sole causes of observed associations and may not even play a significant role in many of the studies.» (World Health Organization, 2023). So the short-term SBP question has a moderate-certainty null, and the long-term hypertension question stays open. (Blood Pressure Lowering Treatment Trialists Collaboration, 2021; inferred from A. Reynolds et al., 2019; World Health Organization, 2023)
Decision relevance
- BP-lowering is the cardiometabolic lever with the best hard-outcome warrant, and it applies in primary prevention — for a hypertensive person it is the one place a proven CV-event reduction is on the table.
- But decide on absolute CV risk, not the BP number. A ~10%-per-5-mmHg relative effect buys little absolute benefit at low risk and a lot at high risk; combine the relative effect with a stratum- specific baseline (SCORE2 Baseline Risk and the ESC Treatment Thresholds) rather than treating to a threshold. This also means the exact BP measurement matters less than the overall risk.
- Lifestyle BP reduction (e.g. sodium, weight, alcohol reduction in >2 drinks/day drinkers) plugs into the same channel. Sodium Intake and Blood Pressure holds BP as a surrogate with hard outcomes graded very low; BPLTTC supplies the missing link — pharmacological BP reduction does reduce hard events — but the transmission of a lifestyle-induced BP drop to events is an assumption, not the same evidence (a different intervention, unmeasured here — Roerecke’s alcohol-reduction CV benefit is modelled, SSaSS’s is measured).
- Where the levers sit in the ranking. BP is a big rock only conditionally, by baseline risk -> Layer 1 - Ranking Interventions for a Stratum; the lever ordering above is a surrogate ordering, read with the transmission criterion in Surrogate Outcomes. That table leaves out sodium reduction, DASH and potassium, which each have a home page: DASH Diet and Blood Pressure, Potassium Intake and Blood Pressure, Dietary Magnesium and Cardiometabolic Outcomes, and, for whether sodium reduction should target the salt-sensitive or everyone, Should Sodium Reduction Be Population-Wide or Targeted. Weight loss as an events route -> Does Weight Loss Reduce Cardiovascular Events.
Limits
- Trials 1972-2013 — older drug regimens and co-treatment; relative (not absolute) effects are the transportable quantity, by the authors’ own caution.
- One (very large) collaboration’s IPD; the target-BP and adverse-effect questions are out of its design.
- Applies to pharmacological lowering; do not silently transfer the magnitude to a lifestyle route.