Verdict: what is supported, and at what level
Dementia has no single modifiable cause. Risk accumulates across vascular health, sensory loss, education and social circumstances, activity, smoking, alcohol exposure, head injury and other factors. The strongest public-health case is therefore for reducing several common risks over the life course, rather than identifying one intervention that prevents dementia for a given person.
Among commonly discussed factors, blood-pressure treatment has the clearest randomised evidence relevant to later cognitive outcomes. The SPRINT MIND trial found fewer cases of mild cognitive impairment with an intensive systolic blood-pressure target than with a standard target. Its dementia result was not statistically conclusive, in part because the study ended earlier than planned and dementia events were fewer than expected. This is a meaningful signal, not proof that a particular target prevents dementia in every setting.
Multidomain trials, notably the Finnish Geriatric Intervention Study to Prevent Cognitive Impairment and Disability, usually called FINGER, found benefit on composite cognitive testing from a programme combining diet, exercise, cognitive training, vascular-risk monitoring and social activity. That design cannot identify which component mattered most. It also did not establish that the package prevents diagnosed dementia.
Hearing care, physical activity and educational opportunity are well supported as population priorities, but the evidential route differs. Their links with dementia derive largely from prospective cohort studies, plausible mechanisms and synthesis by the Lancet Commission on dementia prevention, intervention and care. Such evidence can be important without being equivalent to a dementia-prevention trial.
A high grade describes the evidence base. It is not a recommendation or a substitute for individual clinical assessment.
The grading scale and evidence map
This review grades the claim being made, not the importance of a condition or behaviour. A claim that a factor is associated with dementia requires less evidence than a claim that changing it lowers dementia incidence. The latter needs random allocation, adequate follow-up and clinically meaningful outcomes. Cognitive-test scores can be useful, but they are not interchangeable with diagnosed dementia or preserved daily functioning.
| Grade | What the grade requires | Finding in this review |
|---|---|---|
| A: trial-supported cognitive benefit | Randomised evidence with relevant cognitive outcomes, replicated or supported by a substantial trial base | Multidomain vascular and lifestyle programmes improve some cognitive outcomes; blood-pressure intensification reduces mild cognitive impairment in one major trial context |
| B: trial signal, dementia outcome uncertain | Randomised evidence suggests benefit, but dementia incidence, replication, duration or generalisability remains limited | Blood-pressure management; selected hearing intervention findings in people at higher risk of cognitive decline |
| C: observational priority | Consistent long-term association and plausible pathway, without decisive randomised evidence that modification prevents dementia | Physical activity, hearing loss, education and several life-course factors |
| D: not demonstrated for dementia prevention | Insufficient direct evidence, or evidence limited to a surrogate that does not establish dementia prevention | Any claim that one isolated behaviour, device or biomarker change prevents dementia |
Grades are not a hierarchy of personal importance. Education is graded C because randomising years of childhood education to test late-life dementia would be neither practical nor appropriate, not because education is unimportant. Conversely, a short randomised trial can earn a stronger grade for a cognitive measure while still leaving long-term dementia incidence unanswered.
The practical reading rule is simple: distinguish a modifiable contributor to population risk from a proven individual prevention treatment. Those are related claims, but they are not the same claim.
Blood pressure: the strongest randomised signal
High blood pressure, particularly from midlife onwards, is associated with later dementia risk through vascular injury, stroke and small-vessel disease. It is also a risk factor with established importance beyond cognition. For this review, the dementia-specific question is narrower: does lowering blood pressure reduce cognitive impairment or dementia?
SPRINT MIND provides the most prominent randomised evidence. Participants were assigned to intensive or standard systolic blood-pressure treatment. The intensive group had a lower incidence of mild cognitive impairment. The trial did not produce a statistically conclusive reduction in probable dementia. The appropriate conclusion is not that dementia prevention was shown, nor that the study found nothing. It found a credible cognitive outcome signal while leaving its effect on diagnosed dementia uncertain.
Trials of antihypertensive treatment differ in participants, starting blood pressure, treatment targets, follow-up and outcome ascertainment. Pooling them can increase precision but cannot remove those differences. A cognitive result in older adults at elevated cardiovascular risk should not automatically be translated into a target for younger adults, people with frailty, people with low baseline pressure or those taking other medicines.
Grade: B, trial signal with dementia outcome uncertain. This grade applies to the claim that blood-pressure management can reduce later cognitive impairment in studied populations. It does not establish a universal cognitive target, identify one medicine as a dementia-prevention treatment, or displace the need to weigh adverse effects and wider cardiovascular considerations with a clinician.
Hearing, physical activity and education: strong associations, different evidence
The Lancet Commission's 2024 report included hearing loss, less education and physical inactivity among modifiable factors associated with dementia at population level. Its framework usefully draws attention to timing: education is largely an early-life exposure, blood pressure has major midlife relevance, and sensory and vascular factors also matter in later life. That life-course view is more credible than treating dementia as a late-life problem alone.
Hearing loss may increase cognitive load, reduce social participation, worsen access to conversation and share causes with neurodegeneration or vascular disease. These mechanisms are plausible, but they do not settle causality. The ACHIEVE randomised trial of a hearing intervention did not show an overall difference in its primary cognitive outcome across all participants. A prespecified subgroup with greater baseline risk showed a favourable result. This supports further study and careful interpretation, not a general claim that hearing intervention prevents dementia.
Physical activity has a broad observational association with lower dementia incidence. Reverse causation is a central concern: early, unrecognised neurodegenerative change may reduce activity years before diagnosis. Randomised exercise studies often report fitness, mood or short-term cognitive measures, but have generally been too short or too small to determine dementia incidence. The conclusion is therefore stronger for general health and functional outcomes than for dementia prevention specifically.
Educational attainment is associated with lower dementia risk and may contribute to cognitive reserve, meaning that clinical symptoms emerge later despite brain pathology. It cannot be reduced to an individual late-life cognitive-training prescription. The evidence chiefly concerns social and early-life conditions, and confounding by childhood circumstances remains relevant.
Grade: C for each as a direct dementia-prevention claim, with a B subgroup signal for hearing intervention.
Multidomain programmes and why component claims outrun the trials
FINGER is often cited because it tested a realistic premise: cognition may be influenced by several small, interacting changes rather than a single treatment. Participants received a two-year programme including nutritional guidance, exercise, cognitive training, social activity and monitoring of vascular risk. The intervention improved a composite cognitive score compared with usual health advice in the trial population.
That result supports the feasibility and potential cognitive value of a multidomain approach. It does not show that each component independently prevents dementia. It also does not tell readers whether the result would be the same in people of another age, risk profile, healthcare system or baseline level of activity. Subsequent multidomain studies have varied in design and results, which is expected when interventions, populations and local services differ.
The key failure mode is extracting a favoured component from a package trial. A reader may see exercise included in FINGER and conclude that a specific exercise dose has been proved to prevent dementia. Or they may see vascular monitoring and infer that a named blood-pressure threshold has been established for cognition. Neither inference follows. A factorial trial, in which components are varied separately, would better answer component questions, but such studies are difficult and expensive over the decades required for dementia outcomes.
Grade: A for modest trial-supported benefit on selected cognitive measures from multidomain intervention in studied groups; B for a claim of dementia prevention. The distinction matters because composite test performance, while relevant, is not the same endpoint as fewer dementia diagnoses, better independence or lower care needs.
Population attributable fractions are not personal forecasts
The Lancet Commission estimated that a substantial share of dementia cases at population level could theoretically be associated with a set of modifiable risk factors. This type of estimate is a population attributable fraction, or PAF. It combines the prevalence of an exposure with an estimated relative risk and makes assumptions about causality, overlap between factors and the possibility of changing exposure patterns.
A PAF answers a counterfactual population question: if a causal exposure were removed or reduced across a population under specified assumptions, what proportion of cases might not occur? It does not say that a person with one risk factor will avoid that proportion of their own risk by changing it. Nor do PAFs for several factors add up as separate personal discounts. High blood pressure, diabetes, inactivity, depression, social isolation and hearing loss can cluster, and their pathways can overlap.
| Statement | What it means | What it does not mean |
|---|---|---|
| "This factor has a high PAF" | It is common, associated with dementia, and may have substantial population relevance if causal | Changing it gives a named individual a matching percentage reduction |
| "A trial improved cognition" | Participants had a better result on the trial's defined cognitive outcome | Dementia was prevented, unless dementia incidence was measured and convincingly reduced |
| "The factor is modifiable" | Its level, detection or consequences can potentially be changed | Change is easy, safe, equally accessible or guaranteed to alter risk |
This distinction is not semantic. It prevents population prevention evidence from being repackaged as a personalised promise. Baseline age, vascular history, genetics, education, access to care and competing health risks all affect how a population estimate applies, if it applies at all.
How to read a dementia-risk claim
A useful first question is whether the study measured incident dementia, mild cognitive impairment, a cognitive test, brain imaging or a biomarker. These outcomes sit on a hierarchy but are not interchangeable. A result on an executive-function test may be valuable, yet it cannot by itself establish fewer dementia diagnoses. Likewise, a biomarker associated with neurodegeneration may be a useful research tool without proving that altering it changes clinical outcomes.
Second, inspect the comparison. Random allocation reduces confounding, but it does not cure poor adherence, short duration, missing outcome data or an intervention that is impossible to blind. Cohort studies can track people for longer and capture more dementia cases, but people who are active, socially connected or able to access hearing care often differ from others in ways that statistical adjustment cannot fully measure.
Third, ask when exposure was measured. Dementia pathology can develop long before diagnosis. A late-life fall in activity, weight, social contact or cognitive performance may be an early consequence of disease rather than its cause. Studies that measure exposures in midlife and follow participants for decades are informative, though still vulnerable to residual confounding.
Finally, separate research implications from clinical decisions. A risk-factor review can identify where evidence is promising and where uncertainty remains. It cannot determine an individual's blood-pressure treatment, hearing assessment, exercise capacity or cognitive symptoms. New or worsening cognitive difficulties, sudden neurological symptoms, and concerns about medication effects require appropriate clinical assessment rather than self-directed dementia-prevention protocols.
Limits of this review
This review addresses modifiable factors linked with dementia and cognitive decline in adults. It does not diagnose cognitive impairment, estimate an individual's dementia risk, set treatment targets or provide a prevention protocol. It does not cover genetic counselling, dementia drug treatment, management after a diagnosis, or every potential contributor considered in the dementia literature.
Grades reflect the strength of evidence for a specified claim at the time of review, not the social importance of the factor. They also do not measure feasibility, equity, cost, adverse effects or access. Educational opportunity, hearing assessment and cardiovascular care are shaped by institutions and resources as well as individual choices. Framing them solely as personal lifestyle decisions would misrepresent the evidence and the context in which risk is distributed.
Research gaps remain substantial. Dementia develops over long periods, while trials are often brief. Studies may use composite cognitive outcomes because they are more feasible than waiting for enough dementia diagnoses, but this leaves uncertainty about outcomes that matter most to patients and families. Participants in trials may not represent people with multimorbidity, disability, frailty or limited access to care.
A final limit is that reducing risk is not the same as guaranteeing prevention. Some people with several risk factors will never develop dementia; some people with favourable risk profiles will. The evidence supports risk reduction as a population and clinical-research priority. It does not support certainty about any one person's future.