Clinical Review · Population Genetics · Risk Stratification
APOE ε4 Risk Is Not the Same in Every Population: How Ancestry Changes the Numbers
By Brian Paquette, DO, MPH
The figure most carriers are quoted — that one copy of ε4 raises Alzheimer’s risk roughly three to fourfold — comes overwhelmingly from cohorts of European ancestry. It does not transfer cleanly to everyone. This review sets out what actually varies, what explains it, and what remains genuinely unsettled.
The clinical problem
A patient brings you a direct-to-consumer genetic report. It says ε3/ε4. She wants to know what that means for her. The number you have in your head — a three- to fourfold increase over ε3/ε3 — was derived largely from people of European descent. If your patient is Nigerian, Japanese, Puerto Rican, or African American, that number may be substantially wrong, in either direction.
This is not a peripheral concern. It is one of the clearest examples in clinical genetics of a risk estimate that does not generalise, and it has direct consequences for how aggressively a carrier is counselled, monitored, and worried.
What varies, and by how much
The foundational evidence remains the APOE and Alzheimer Disease Meta Analysis Consortium’s pooled analysis of 5,930 patients and 8,607 controls across forty research teams. Relative to ε3/ε3, it reported (Farrer et al., JAMA, 1997, DOI):
| Population | ε3/ε4 — OR (95% CI) | ε4/ε4 — OR (95% CI) |
| Japanese | 5.6 (3.9–8.0) | 33.1 (13.6–80.5) |
| Caucasian / European ancestry | 3.2 (2.8–3.8) | 14.9 (10.8–20.6) |
| African American and Hispanic | Association reported as weaker; substantial heterogeneity between African American studies, and the authors called explicitly for further work on the attenuated effect in Hispanic cohorts | |
The headline: the same allele, in the same genotype, carried roughly a twofold difference in effect size between Japanese and European-ancestry cohorts — and a further attenuation in African American and Hispanic cohorts. Effect size is not a property of the allele alone.
The Nigerian observation, and its correction
The most striking single finding came from Ibadan. In a community-based study of elderly Yoruba Nigerians — a population with a high ε4 allele frequency — investigators reported that ε4 was not associated with Alzheimer’s disease at all (Gureje et al., Annals of Neurology, 2006, DOI). This became widely known as the Nigerian paradox.
That framing was too strong, and the same research group later refined it. The Indianapolis–Ibadan study is an unusually powerful design: two cohorts, two continents, one harmonised protocol. With longitudinal follow-up, ε4 was a significant risk factor in Yoruba — for incident Alzheimer’s disease (p=0.0489) and cognitive decline (p=0.0425) — but the effect was clearly weaker than in African Americans, in whom one or two copies predicted incident disease at p<0.0001 (Hendrie et al., International Psychogeriatrics, 2014;26(6):977–985, DOI).
So the accurate statement is not “ε4 does not matter in Nigerians.” It is that the same allele carries a substantially smaller effect in Yoruba than in African Americans studied under the same protocol — two populations who share recent African ancestry but differ in how much European ancestry they carry. That observation is what pointed toward the mechanism.
The mechanism: it is the neighbourhood, not the person
The decisive study examined admixed populations — Puerto Rican and African American cohorts, in whom any individual’s chromosomes are a mosaic of African and European segments. This allows a question that cannot be asked in a homogeneous population: within the same person, does an ε4 allele sitting on an African-ancestry chromosomal segment behave differently from one sitting on a European-ancestry segment?
It does (Rajabli et al., PLOS Genetics, 2018;14(12):e1007791, DOI):
| Cohort | ε4 on African local ancestry | ε4 on European local ancestry |
| Puerto Rican | OR 1.26 | OR 4.49 |
| African American | OR 2.34 | OR 3.05 |
The local-ancestry × ε4 interaction was significant in both cohorts (Puerto Rican p=0.019; African American p=0.005). Critically, global ancestry showed no such interaction. What matters is not what proportion of a person’s genome is of African origin — it is which ancestral haplotype the ε4 allele itself is sitting on.
This is the conceptual heart of the topic. The ε4 coding variant is identical in every population. What differs is the genomic neighbourhood it was inherited with. The authors concluded that protective variant or variants most likely lie within the region surrounding APOE on the African ancestral background — and that this, rather than culture, diet, or measurement artefact, best explains the attenuation.
The contemporary challenge
This is not settled science, and a recent analysis argues against it. Using 40,210 participants from the National Alzheimer’s Coordinating Center database, investigators examined ε4 allele count against cognitive impairment prevalence and age at onset, and reported no evidence supporting the Nigerian paradox across White, Black, Asian and Other groups — with ε4 effects appearing consistent across racial categories (Bobo et al., Journal of Dementia and Alzheimer’s Disease, 2025;2(3):31, DOI).
How should a clinician hold two apparently opposite findings? Not by picking a side, but by noticing that they measure different things:
- Self-identified race is not genetic ancestry. NACC classifies participants by racial category. Rajabli measured ancestry at a specific chromosomal segment. A US cohort labelled “Black” contains individuals with widely varying African and European admixture — precisely the variable that Rajabli found to matter, and that a racial label cannot capture.
- NACC is clinic-based, not community-based. Participants are enrolled through Alzheimer’s Disease Research Centers, largely because of cognitive concern. The Ibadan and Indianapolis cohorts were community samples. Ascertainment differences of this magnitude routinely attenuate or distort genotype–phenotype associations.
- Neither design tests Yoruba Nigerians. The original observation was made in West Africa. A US database cannot confirm or refute it directly.
The honest synthesis: the local ancestry finding is mechanistically specific and has been demonstrated within individuals, which is a strong design. The racial category finding is a useful caution that population labels are poor instruments for genetic risk. Both point the same direction — toward measuring ancestry properly rather than inferring it from identity.
What this changes in clinic
1. Quote the uncertainty, not just the number. For a patient who is not of European ancestry, “three to four times” is a figure imported from a population that may not be theirs. Saying so is more accurate, and in my experience more reassuring, than false precision.
2. Do not reduce risk for a patient of African ancestry on the strength of this literature. The attenuation is real in the aggregate, but it is not a licence to under-monitor an individual. Admixture varies widely between people who share a label, and clinical genotyping does not report local ancestry at the APOE locus.
3. Take East Asian ancestry seriously in the other direction. The Japanese ε4/ε4 point estimate — OR 33 — is more than double the European figure, though the confidence interval is wide. For a Japanese ε4 homozygote, the European-derived number is likely an under-estimate.
4. Modifiable risk is where the leverage stays. Nothing here changes the management plan. Vascular risk, sleep, exercise and metabolic health carry the same weight regardless of ancestry, and structured lifestyle intervention has now been shown to work irrespective of ε4 status.
A note on language
This subject is easy to state badly. The finding is not that some groups of people are biologically more or less vulnerable by virtue of who they are. It is that a specific allele behaves differently depending on the stretch of chromosome it was inherited alongside — a fact about haplotypes, not about identity. Genetic ancestry is continuous, individual, and only loosely correlated with the social categories used in medicine. Where those categories are used in the studies above, it is because the investigators used them, and that is itself one of the field’s limitations.
Limitations
This is a narrative review, not a systematic one. The Farrer meta-analysis is now nearly three decades old and its non-European strata were underpowered. The Rajabli local-ancestry estimates come from case–control samples of modest size and have not been replicated in every admixed population. The NACC analysis uses self-identified race as its stratifier, which is the wrong instrument for a genetic question. No study cited here was designed primarily to produce individualised risk estimates, and none should be used that way.
EDITORIAL NOTE: All effect estimates above are quoted directly from the cited primary sources and were verified against the source record. Where findings conflict, the conflict is presented rather than resolved. Educational content only; not a substitute for individual clinical judgement or genetic counselling.