Does lowering blood pressure prevent dementia? What the trials actually show

By Marcus Renshaw ·

Every list of modifiable dementia risk factors puts high blood pressure near the top, and that ranking rests almost entirely on observational data: people with higher pressure in midlife get more dementia decades later. It is a correlation with a believable mechanism, repeated often enough that it now reads like settled fact. The randomized question is a different and much harder one. If you take someone whose pressure is high and actually lower it, do they get less dementia? For twenty years the trials that asked kept returning ambiguous answers. That changed in the last eighteen months, and the strongest result was presented at the European Society of Cardiology congress in Munich on 30 August 2026.

Yes, on the best randomized evidence available. A 33,995 person trial in rural China cut all cause dementia by 15% over four years, and the effect held at seven. SPRINT MIND, the US trial, missed significance on dementia alone but cut cognitive impairment. Blood pressure lowering has the strongest randomized evidence of any modifiable target here.

What did SPRINT MIND actually find?

SPRINT enrolled 9,361 adults aged 50 and over with a systolic pressure of 130 mmHg or higher and raised cardiovascular risk, excluding anyone with diabetes or a prior stroke, and randomized them to a systolic target below 120 mmHg or below 140 mmHg. At one year the two groups sat at 121.4 and 136.2 mmHg, a separation of just under 15 mmHg. The trial was stopped early, after a median 3.26 years, because intensive control was clearly cutting cardiovascular events and all cause death.

That early stop is the whole reason the cognitive answer stayed murky. SPRINT MIND, the prespecified cognitive arm, reported in 2019 that adjudicated probable dementia occurred in 149 participants on intensive treatment against 176 on standard, which is 7.2 against 8.6 cases per 1000 person-years, a hazard ratio of 0.83 with a 95% confidence interval of 0.67 to 1.04. The interval crosses 1, so it missed. Mild cognitive impairment did not miss, at a hazard ratio of 0.81 (0.69 to 0.95), and neither did the combined outcome of mild cognitive impairment or probable dementia, at 0.85 (0.74 to 0.97). The authors said plainly why: because of early termination and fewer dementia cases than expected, the study may have been underpowered for that endpoint.

The investigators went back and followed people for longer. The extended analysis published in Neurology in 2025 tracked cognitive status by telephone out to a median seven years and found probable dementia in 248 participants on intensive treatment (8.5 per 1000 person-years) against 293 on standard (10.2 per 1000 person-years), a hazard ratio of 0.86 (0.72 to 1.02). Missed again, and for the same reason: they accrued 216 new dementia cases against a target of 326. Mild cognitive impairment came in at 0.87 (0.76 to 1.00) and the composite at 0.89 (0.79 to 0.99).

Read those two papers together and the shape is consistent. Every point estimate for dementia sits between 0.83 and 0.86, in the direction of benefit, and never quite reaches significance. Every measure of milder cognitive impairment does reach it. That is not a null result, and it is not proof either. It is an underpowered miss, which is a genuinely different thing and gets flattened in both directions by whoever is summarizing it.

What did the trial in 33,995 people find?

A trial big enough to answer the question ran in rural China. CRHCP cluster randomized 326 villages, 163 to intervention and 163 to usual care, covering 33,995 adults aged 40 and over with uncontrolled hypertension. The intervention was not a drug so much as a delivery system: trained non-physician community health providers started and titrated antihypertensive medication on a simple stepped care protocol toward a target below 130/80 mmHg, supervised by primary care physicians.

It moved the number hard. Over 48 months the net reduction against usual care was 22.0 mmHg systolic (95% CI 20.6 to 23.4) and 9.3 mmHg diastolic (8.7 to 10.0). All cause dementia, the primary outcome, came in at a risk ratio of 0.85 (0.76 to 0.95), P = 0.0035. Serious adverse events were less frequent in the intervention group, not more, at a risk ratio of 0.94 (0.91 to 0.98).

That result was published in Nature Medicine in April 2025, and it is the one that actually answered the question. It is worth being clear about this, because the news coverage in the last few weeks has not been: the 15% figure is not new.

What was new at ESC 2026?

Durability. The intervention ran for four years and then stopped, and the ESC presentation on 30 August 2026 reported what happened three years after that, at the seven year mark. Dementia had occurred in 8.85% of the intervention group against 10.55% of usual care, an adjusted risk ratio of 0.85 (0.78 to 0.91), P < 0.001. Cognitive impairment without dementia was 13% lower. Serious adverse events were 47.15% against 49.78%. Over the full seven years systolic pressure fell 17.6 mmHg in the intervention villages against 2.4 mmHg under usual care.

So the point estimate is identical to the four year result, 0.85 both times, with a tighter confidence interval and a much smaller P value on roughly three more years of accumulated cases. Anyone writing that a new trial has just shown blood pressure control cuts dementia by 15% is reporting the 2025 paper again. What is actually new is that the benefit did not wash out when the program ended, that the milder cognitive endpoint moved too, and that the harms stayed on the favorable side of the ledger at seven years.

One sourcing caveat, and I would rather state it than bury it. As of this writing the seven year result exists as a congress presentation and an ESC press release, not as a peer reviewed paper. The four year result in Nature Medicine is the citable one. I am treating the seven year numbers as strong but provisional until the paper lands, which is the same standard I would want applied to a result I did not like.

Why did the big trial succeed where SPRINT MIND did not?

Four differences, and they matter for deciding how much of this transfers to you.

Power. 33,995 against 9,361, no early stop, and enough accumulated dementia cases to resolve an effect of this size. SPRINT MIND never had that, by its own authors’ account.

Size of the change. CRHCP separated the groups by 22.0 mmHg systolic over four years. SPRINT managed just under 15. A bigger dose of the intervention should produce a bigger effect, and it did.

Where people started. CRHCP recruited people with untreated or uncontrolled hypertension, so there was a lot of room to move. SPRINT recruited treated hypertensives at high cardiovascular risk and pushed an already managed number lower. Those are different experiments, and only the first one tells you what happens when uncontrolled pressure gets controlled.

What got counted, and how. CRHCP’s endpoint was all cause dementia; SPRINT MIND adjudicated probable dementia. CRHCP was open label with blinded endpoint assessment and randomized by village, not double blind and not by person.

That last point carries a caveat worth sitting with. The CRHCP intervention was a whole care program, not a milligram dose: village level health workers, structured follow up, medication titration, adherence support, and education all arrived together. Attributing the entire dementia effect to the millimetres of mercury is an assumption, a reasonable one given the dose response pattern across these trials, but still an assumption. The honest version is that getting uncontrolled hypertension under control, by whatever means actually works in practice, was followed by less dementia.

There is also a harms asymmetry between the two trials that should not be skipped. CRHCP reported fewer serious adverse events in the intervention arm. SPRINT, pushing to below 120 mmHg, reported higher rates of hypotension, syncope, electrolyte abnormalities, and acute kidney injury, though not of injurious falls. Treating to 130/80 and treating to below 120 are not the same trade.

Does any of this apply if your pressure is already normal?

No, and this is where most people reading a headline about brain protection will get it wrong.

Every trial above enrolled people with hypertension. If you are 42 with a home average of 118/74, none of these trials studied you, and there is no randomized evidence that pushing a normal number lower buys brain protection. What the evidence supports is bringing an uncontrolled number under control. The benefit lives in closing a gap, not in optimizing past the point where the gap exists.

There is one wrinkle worth knowing about, with a firm caveat on it. A secondary analysis of SPRINT looked at whether the benefit tracks APOE ε4 status, the strongest common genetic risk factor for dementia, and found the absolute benefit concentrated in carriers: a four year risk difference of 1.7 percentage points in carriers, which works out to a number needed to treat of 59, against 0.2 points in the other direction for non-carriers. The relative effects were similar between the groups and the interaction was significant only on the absolute scale (P = 0.045) and not the relative one (P = 0.35). That analysis is a preprint and has not been peer reviewed. It is interesting, it is not a reason to genotype yourself and change anything, and I am including it mostly because it will get quoted at you eventually without the caveats.

The self-experimenter layer here is about confounding rather than targets. The clinical floor is straightforward: if you are hypertensive, get to below 130/80 with whoever writes your prescriptions. The optimization question, whether a brain at a steady 115 does better than one at 125, has no trial behind it. And if you are running a stack, several common components push the number the wrong way. Testosterone raises blood pressure across every formulation, which is why it carries a class wide FDA warning, and the rise was roughly four times larger in men already on a blood pressure medication. A brain protection argument for controlling your pressure does not survive quietly undoing that control somewhere else in the protocol.

Which number matters more, systolic or diastolic?

Most of the attention goes to systolic, and for cardiovascular endpoints that is broadly right. For dementia the picture may be less lopsided. A nested analysis inside the same CRHCP cohort modelled four year blood pressure trajectories in 28,135 hypertensive participants and found the high stable diastolic trajectory carried the strongest association with dementia, at an odds ratio of 2.14 (1.74 to 2.63), ahead of the high stable systolic trajectory at 1.75 (1.42 to 2.15).

Two guards on that. It is observational, drawn from the same population but not from the randomized comparison, so it describes association and not effect. And a trajectory is a pattern over years, not a reading, which is exactly the kind of thing a single annual clinic measurement cannot see and a logged daily number can. Both arrows point the same way in practice: log both values, and do not discard the diastolic one because the systolic gets the headlines.

How does this compare to the rest of what gets sold for brain aging?

This is the part that changed how I rank my own interventions, and the timing made it hard to miss. Inside three weeks, three trials reported.

STAREE gave 9,971 healthy adults over 70 a statin for a median 5.9 years. Major cardiovascular events fell 30%. Disability free survival, a composite that counts dementia among its components, did not move at all. The marker moved, the events moved, function did not.

Then on 4 September 2026 Novartis reported that Lp(a)HORIZON missed its primary endpoint. Pelacarsen lowered Lp(a), as designed, in 8,323 patients with established cardiovascular disease, and it did not reduce cardiovascular risk in the overall population. That was the first outcomes trial of any Lp(a) lowering drug, and the marker moved while the events did not.

And blood pressure, the least fashionable number on the panel, delivered a functional outcome in 34,000 people and held it for three years after the intervention stopped.

I do not read that as blood pressure being biologically special. I read it as blood pressure being the one that got the trial, and got it at a size that could resolve the answer.

Two other interventions have been taken seriously enough to be tested against a cognitive endpoint, and they are worth knowing so the claim above stays honest. FINGER randomized 1,260 at-risk Finnish adults to a two year multidomain program of diet, exercise, cognitive training and vascular risk monitoring, and it improved cognitive test scores against control. ACHIEVE randomized 977 older adults with hearing loss to hearing aids or health education, and its primary analysis of three year cognitive change was not significantly different between groups, with a benefit appearing only in the higher risk subgroup recruited from an existing cardiovascular cohort.

Note what both measured: cognitive test scores, not whether people developed dementia. That is the gap CRHCP closed, and it is why blood pressure currently has the strongest randomized case rather than the only one. What almost nothing sold for brain aging has is a trial of either kind, and the intervention with the best evidence is the boring one you can measure at home for the cost of a cuff.

What do you actually track?

The target used by the trial that actually moved dementia is below 130/80 mmHg, and that is a clinic based target. SPRINT pushed to below 120 and did not reach significance on dementia, so the harder target is not the better supported one here. Home readings run lower for the same person, which is why the home threshold sits nearer 135/85 and why technique matters more than the device: most bad home readings come from posture, cuff placement, and timing rather than hardware. If your systolic comes from one hurried reading in a waiting room once a year, no trial result above is being applied to you in any meaningful way.

What is worth logging is an average, not a reading. A single number carries too much noise to act on, a fortnight of morning readings does not. The pattern worth watching beyond the average is the gap between your morning and overnight numbers, which consumer apps hold the data for and almost never compute. And the device caveat still stands: cuffless and smartwatch blood pressure is not ready to carry this, so a wrist estimate is not the input you want feeding a decision this size.

I am on a blood pressure medication, dosed at bedtime, and I started logging daily when I was sitting in Stage 2 and running a stack I expected to move the number. The log is how I watched it come down over months and how I knew when a medication change was safe to discuss. The BP Tracker computes the AHA category from both values and averages the mornings, which is the part I actually use, because the category and the trend are the two things a single reading cannot give you.

The seven year CRHCP result still needs to appear as a paper, and the question nobody has run yet is whether any of this holds for people whose pressure is already controlled, where the room to improve is small and the trial would have to be enormous. Until someone funds that, the defensible position is the unglamorous one: if your number is high, the evidence that getting it down protects your brain is now better than the evidence for anything else you could buy for the same purpose.

FAQ

Does lowering blood pressure prevent dementia?

On the best randomized evidence, yes, in people whose blood pressure is uncontrolled. A cluster randomized trial of 33,995 rural Chinese adults reduced all cause dementia by 15% over four years with a community led program targeting below 130/80, and the effect persisted at seven years. The US SPRINT MIND trial pointed the same direction but was underpowered for dementia alone.

What did SPRINT MIND find about dementia?

Probable dementia occurred in 149 participants on intensive blood pressure control against 176 on standard control, a hazard ratio of 0.83 with a confidence interval of 0.67 to 1.04, which missed significance. Mild cognitive impairment was significantly reduced, at 0.81. An extended follow up to seven years found a hazard ratio of 0.86, again missing significance on dementia alone. The trial was stopped early and accrued fewer dementia cases than planned.

How much does blood pressure control reduce dementia risk?

In the trial large enough to measure it, dementia occurred in 8.85% of the intervention group against 10.55% under usual care over seven years, a relative reduction of about 15% and an absolute reduction of 1.7 percentage points. Dividing into that 1.7 point difference gives roughly 59 people needing the intervention over seven years to prevent one case, which is a good number for a preventive measure. That arithmetic is mine rather than the trial’s, and it is a coincidence that the APOE analysis mentioned above lands on a similar figure over four years from a different trial.

What blood pressure target was used in the trials that reduced dementia?

Below 130 mmHg systolic and below 80 mmHg diastolic, achieved with a stepped care medication protocol. That is a clinic based target. Home readings typically run lower for the same person, so a home threshold sits nearer 135/85.

Does blood pressure medication protect the brain if my blood pressure is already normal?

There is no randomized evidence for that. Every trial in this area enrolled people with hypertension, most of them uncontrolled. The benefit shown is from bringing a high number under control, not from pushing a normal number lower, and the trial that would answer the second question has not been run.

Does APOE ε4 status change the benefit?

A secondary analysis of SPRINT found the absolute benefit concentrated in APOE ε4 carriers, with a four year risk difference of 1.7 percentage points against 0.2 points in the opposite direction for non-carriers, while relative effects were similar. That analysis is a preprint and has not been peer reviewed, so it is not a basis for changing anything.

Is high diastolic pressure bad for the brain too?

Possibly more than is generally assumed. An observational analysis within the same Chinese cohort found that a persistently high diastolic trajectory carried a stronger association with dementia than a persistently high systolic one, at odds ratios of 2.14 against 1.75. That is association rather than effect, and it comes from modelling patterns over four years rather than single readings, but it is a reason to log both numbers.

Is intensive blood pressure control safe?

It depends how intensive. The trial targeting below 130/80 reported fewer serious adverse events in the intervention group than under usual care. SPRINT, pushing below 120, reported higher rates of hypotension, fainting, electrolyte abnormalities and acute kidney injury, though not of injurious falls. Treating to 130/80 and treating to below 120 are different trades, and the decision belongs with whoever manages your prescription.

Sources