APOE is the strongest common genetic influence on dementia risk, and it is now something people discover about themselves almost by accident — a line in a consumer ancestry report, or a secondary finding from a panel ordered for something else entirely. The result arrives without context, usually without a physician in the room, and it provokes the one question genetics cannot answer: am I going to get Alzheimer's disease?
The honest answer is that APOE shifts probability, sometimes considerably, but it does not determine the outcome. Many people who carry the higher-risk allele reach their nineties with intact cognition, and many people who develop dementia carry none of it. The more useful conversation — and the one that too rarely happens — is about the substantial share of dementia risk that is attributable to modifiable factors, most of which are the same factors that govern vascular and metabolic health.
What the APOE gene actually does
APOE encodes apolipoprotein E, a protein central to lipid transport. In the brain it is produced mainly by astrocytes and is responsible for shuttling cholesterol and phospholipids to neurons — the raw material of membranes, synapses and myelin repair. It also participates in the clearance of amyloid-beta from brain tissue and in the maintenance of the blood-brain barrier.
Three common variants circulate in the population: e2, e3 and e4. Everyone carries two copies, one from each parent, giving six possible genotypes. The e3 allele is by far the most common and serves as the reference. The e4 protein is less efficient at clearing amyloid and is associated with earlier and heavier amyloid deposition, more neuroinflammation, and a less robust blood-brain barrier. The e2 allele appears to be modestly protective, though it carries its own effects on triglyceride metabolism.
What an e4 result means — and what it does not
| Genotype | Approximate frequency | Effect on late-onset Alzheimer risk |
|---|---|---|
| e3/e3 | Roughly 55–60% | Reference |
| e3/e4 | Roughly 20–25% | Increased, commonly cited near 2–3× reference |
| e4/e4 | Roughly 2% | Substantially increased, often cited near 10× reference |
| e2/e3 | Roughly 10–12% | Modestly reduced |
| e2/e4 | Roughly 2% | Increased; the e4 effect dominates |
| e2/e2 | Under 1% | Reduced |
Two things about this table matter more than the numbers in it. The first is that the effect is real and, for two copies, large. The second is that it is probabilistic. A single e4 allele also shifts the age of onset earlier rather than guaranteeing an event; a great many carriers never develop dementia at all. The relationship is also not uniform — the risk associated with e4 appears greater in women than in men across parts of the age range, and smaller in some populations of African ancestry than in those of European ancestry.
An APOE result tells you which distribution you were drawn from. It does not tell you where in that distribution you will land — and a considerable amount of what determines that is not genetic at all.
Whether to test at all
APOE genotyping is not a routine test, and it should not be treated as one. Unlike most of what we measure, the result cannot be un-known, and it carries implications for blood relatives who did not consent to learning anything. Distress after an unexpected e4 result is common, and in some jurisdictions predictive genetic information has consequences for insurance underwriting.
The defensible position is to test when the result would plausibly change something — behaviour, screening intensity, or a treatment decision — and always with counselling beforehand, not afterwards. Genotype is genuinely relevant when anti-amyloid therapy is being considered, since the risk of amyloid-related imaging abnormalities is highest in e4 homozygotes. It matters for eligibility in prevention trials. And for some people, knowing does sharpen adherence to prevention in a way that abstract advice does not.
There is an uncomfortable corollary worth stating plainly: nearly everything a physician would recommend to an e4 carrier is worth doing regardless of genotype. If the plan would not change, the test is information without leverage.
The modifiable share of dementia risk
This is the part of the conversation that deserves the most time. Successive Lancet Commissions on dementia prevention have estimated that a large fraction of dementia cases worldwide — roughly forty to forty-five percent — is attributable to risk factors that are, in principle, modifiable across the life course. The factors identified cluster into recognisable groups:
- Vascular and metabolic: hypertension in midlife, diabetes, elevated LDL cholesterol, obesity, smoking.
- Sensory and neurological: untreated hearing loss, untreated vision loss, traumatic brain injury.
- Behavioural and social: physical inactivity, excess alcohol, depression, social isolation.
- Environmental and educational: air pollution exposure, and limited education in early life.
These estimates deserve an honest caveat. Population attributable fractions are derived from observational data, the factors overlap heavily, and for some of them the direction of causation is contested — hearing loss and depression may in part be early manifestations of the disease process rather than causes of it. Single-factor intervention trials have produced mixed results, and multidomain prevention trials such as the Finnish FINGER study have shown modest benefit on composite cognitive measures rather than dramatic ones. The claim that forty-five percent of dementia is preventable in any individual would be an overstatement. The claim that a meaningful share of risk is addressable is not.
Cardiovascular and metabolic prevention is brain prevention
Pathology studies have made one point repeatedly: in older brains, mixed pathology is the norm. Alzheimer-type changes coexist with small-vessel disease, silent infarcts and white matter damage, and it is often the vascular contribution that determines whether a given amyloid burden translates into clinical dementia. Two people with similar plaque loads can differ enormously in function depending on the state of their cerebral vasculature.
Midlife hypertension is among the most consistently replicated risk factors for later cognitive decline. The SPRINT MIND trial found that intensive blood pressure lowering reduced the incidence of mild cognitive impairment; the dementia endpoint alone did not reach statistical significance, which is exactly the sort of nuance that gets lost in summaries but should not be. The same logic extends to ApoB-lowering, glycaemic control, smoking cessation and atrial fibrillation management. Cerebrovascular protection is not a separate programme from cardiovascular protection; it is the same programme, measured in a different organ.
Metabolic health carries particular weight. Insulin resistance and type 2 diabetes are associated with a substantially higher risk of dementia, and the association appears well before diagnostic thresholds are crossed. Visceral adiposity in midlife is a stronger signal than BMI. Cardiorespiratory fitness — among the most robust predictors of all-cause mortality we have — also tracks with better preserved brain volume and cognition in observational work.
Sleep belongs in the same conversation. Clearance of metabolic waste from brain tissue is enhanced during sleep, and chronically short or fragmented sleep in midlife is associated with higher later dementia risk. Causality here is genuinely unsettled, since poor sleep is also an early symptom. What is not unsettled is that obstructive sleep apnoea is common, under-diagnosed, and treatable — and that treating it is worthwhile on cardiovascular grounds alone. Untreated hearing loss deserves similar attention: the ACHIEVE trial found no overall slowing of cognitive decline from hearing aids, but did find a benefit in the higher-risk subgroup, which is a reasonable basis for correcting hearing loss rather than tolerating it.
A cognitive baseline in midlife
Cognitive testing is far more informative when it is compared against yourself than against a normative table. Between-person variation is enormous, and a high-performing individual can lose a great deal of function while still scoring comfortably within the normal range for their age. By the time a standard screening instrument turns abnormal, considerable change has already occurred.
An objective baseline established in the forties or fifties — computerised testing of processing speed, memory and executive function, ideally alongside a structural brain MRI documenting volumes — converts a crude population comparison into a personal trajectory. It also makes it possible to distinguish genuine decline from the far more common causes of subjective cognitive complaints: poor sleep, untreated depression, thyroid disease, B12 deficiency, alcohol, anticholinergic medication, and stress.
Blood biomarkers and the limits of current treatment
Plasma biomarkers of Alzheimer pathology, particularly phosphorylated tau assays such as p-tau217, represent a real advance. In people with cognitive symptoms they now approach the accuracy of spinal fluid analysis and amyloid PET for identifying underlying Alzheimer pathology, at a fraction of the cost and invasiveness.
Their role in people without symptoms is a different and much less settled question. A positive result in an asymptomatic person indicates a higher likelihood of underlying pathology, not a diagnosis, and the interval between biomarker positivity and clinical symptoms varies from years to never. There is no established screening protocol for asymptomatic adults, and testing outside of a considered clinical or research context can generate a great deal of anxiety for very little actionable return.
The same restraint applies to treatment. Anti-amyloid monoclonal antibodies have been approved in some countries and do slow decline on cognitive rating scales in early symptomatic disease. They do not reverse it, they require infusions and serial MRI monitoring, they carry a real risk of brain oedema and microhaemorrhage that is highest in e4 homozygotes, and whether the average effect size is clinically perceptible to patients and families remains actively debated among specialists. This is a field in motion, and it deserves to be described as such rather than oversold.
What to ask your physician
If brain health is a genuine concern — whether or not you know your APOE genotype — the productive conversation is a specific one:
- Before any genetic testing, ask what you would do differently with each possible result. If the answer is nothing, defer it.
- Treat blood pressure seriously in midlife, and know your actual numbers rather than a single clinic reading.
- Ask for ApoB and a metabolic panel including fasting insulin or HOMA-IR, not only glucose and HbA1c — insulin resistance is visible long before diabetes is.
- Have hearing and vision formally tested and corrected rather than accommodated.
- Screen for sleep apnoea if you snore, wake unrefreshed, or have hypertension that resists treatment.
- Establish an objective cognitive baseline while you are well, so that any future change can be measured against you rather than against a population average.
Genetics sets a starting position. It is not a verdict, and the interventions with the best evidence behind them are available to everyone regardless of which alleles they inherited.