Evidence Review · Diet, Alcohol, and Cannabis · Dementia Risk

Alcohol, Binge Drinking, Ultra-Processed Food, and Cannabis: Where the Dementia Evidence Draws the Line for APOE ε4 Carriers

By Brian Paquette, DO, MPH

The clinical problem

“How much is too much?” is a dose question, and most dementia research answers it badly. Cohort studies compare broad categories of drinkers or eaters, questionnaires are imprecise, and people who abstain often do so because they are already unwell, which makes moderate consumers look healthier than they are. The result, for decades, was a reassuring J-shaped curve: a little alcohol appeared protective. For carriers of APOE ε4, who already face a higher lifetime probability of Alzheimer’s disease and a median symptom onset in the mid-60s if they carry two copies, the stakes of getting that curve wrong are high.[1]

Newer tools have changed the picture. Mendelian randomization uses genetic variants that influence how much people drink as a natural experiment, largely free of the confounding and reverse causation that distort questionnaire studies. Large biobanks now follow hundreds of thousands of people for more than a decade. And dietary research has shifted from single nutrients to the degree of industrial processing, which is what “greasy food” usually means in practice.

Key conclusions

• Among drinkers, the J-curve does not survive causal analysis. In 313,958 UK Biobank current drinkers, conventional modeling showed the familiar J-shape, but Mendelian randomization showed risk rising with every increment of genetically predicted consumption and found no protective dip.[2] That analysis excluded non-drinkers, so it cannot settle whether abstainers do better or worse than light drinkers; observational studies on that question still conflict.[3,4]
• Pattern beats average. Across seven European cohorts (131,415 people), exceeding 21 UK units a week raised dementia risk modestly (HR 1.22), but alcohol-induced loss of consciousness roughly doubled it, in moderate and heavy drinkers alike (HR 2.19 and 2.36).[5] In two analyses of the Finnish Twin Cohort, binge drinking at least monthly in midlife was linked to later dementia (RR 3.2, adjusted for age, sex, and education) and to cognitive impairment (RR 1.98, independent of total intake).[6,7]
• For ε4 carriers, the alcohol signal is stronger. In a Finnish cohort followed 23 years, dementia risk rose with midlife drinking only in carriers.[8] A 2025 meta-analysis of 170 genotype-stratified studies found frequent drinking more strongly linked to dementia in carriers, and the apparent benefit of light-to-moderate drinking stronger in non-carriers.[9] One large UK Biobank analysis disagrees.[4]
• Social contact and alcohol are separate variables. Strong social engagement is associated with about 19% lower dementia risk and loneliness with 42% higher.[10,11] Nothing in these data suggests the benefit depends on, or offsets, ethanol.
• Ultra-processed food is linked to higher dementia risk. In UK Biobank, each 10% increase in the ultra-processed share of diet was associated with 25% higher risk of all-cause dementia, and swapping 10% back to minimally processed food with 19% lower risk.[12] A US replication found higher risk at the highest intake but no clear trend across intake levels for dementia.[13]
• Diet quality lowers risk independent of genetic risk. High Mediterranean-diet adherence was associated with 23% lower dementia risk with no interaction with polygenic risk.[14] In two large US cohorts, its protective association with dementia appeared strongest in ε4 homozygotes.[15]
• Cannabis evidence is split by severity. People who needed emergency or hospital care for cannabis had higher dementia risk within five years (adjusted HR 1.23 vs other acute-care patients).[16] In UK Biobank and the Million Veteran Program, lifetime use and cannabis use disorder were not associated with cognitive decline or dementia, and Mendelian randomization was null.[17]