Researchers led by Dylan Williams at University College London have calculated that the common APOE gene variants, epsilon3 and epsilon4, statistically account for between 71.5 percent and 92.7 percent of Alzheimer's disease cases across four large study populations, and for roughly 44 to 46 percent of all-cause dementia. The findings, published in npj Dementia on January 9, 2026, use a measure called the population attributable fraction, which estimates what share of disease in a population is linked to a given risk factor, not what share is caused by it in any one person.

The study drew on four separate sources: UK Biobank (171,105 people aged 60 and over), the Finnish registry cohort FinnGen (289,150 people), the ADGC case-control collection of autopsy-confirmed Alzheimer's cases and pathology-negative controls (5,007 people), and the A4 Study, a trial cohort in which 4,415 cognitively unimpaired participants underwent amyloid PET scanning. In each, the researchers compared disease risk among carriers of the epsilon3 and epsilon4 alleles against people with the epsilon2/epsilon2 genotype, then calculated how much of the disease burden that comparison could statistically attribute to APOE. The Alzheimer's fraction was lowest in FinnGen (71.5 percent) and highest in the neuropathology-confirmed ADGC sample (92.7 percent); a separate calculation found 85.4 percent of amyloid PET positivity in the A4 cohort was attributable to the same alleles.

The authors describe this as considerably higher than many previous estimates of APOE's contribution, which had ranged from under 20 percent to about 70 percent in earlier work cited in the paper. Outside experts convened by the UK Science Media Centre cautioned against reading the figures as proof the gene alone drives disease. Oxford Population Health's Najaf Amin noted that unless someone carries two copies of epsilon4, a rare combination, individual risk is still heavily shaped by other factors, and that a population attributable fraction describes a statistical share of burden, not a guarantee that removing the gene's effect would prevent that share of cases outright.