A hypothesis-free AI screened the medication histories of nearly a million patients and identified varenicline, a smoking cessation drug, as a candidate for dementia prevention. Validation in an independent cohort of 9584 people showed a dementia risk reduction of nearly half compared with nicotine replacement therapy
A hypothesis-free AI screened the medication histories of nearly a million patients and identified varenicline, a smoking cessation drug, as a candidate for dementia prevention. Validation in an independent cohort of 9584 people showed a dementia risk reduction of nearly half compared with nicotine replacement therapy
Researchers at the Washington DC Veterans Affairs Medical Center and George Washington University trained an AI model on data from roughly one million patients to identify, without a prior hypothesis, drugs that reduce the risk of dementia. Varenicline, a smoking cessation drug approved in 2006, produced the second-strongest protective signal. In a preprint dated September 20, the team validated it in an independent cohort of 9584 people: those who took varenicline developed dementia at nearly half the rate.
The researchers trained a transformer model (the architecture behind modern large language models) to distinguish, from medical records (prescriptions, diagnoses, lab results), patients who went on to develop dementia during years of follow-up from those who did not. Interpretable AI methods then revealed which medications were associated with lower risk.
The death of cholinergic neurons and the decline of acetylcholine signaling is an established mechanism in Alzheimer's disease: all currently approved Alzheimer's drugs act through this same system. Varenicline has a direct connection to it: the drug partially activates the α4β2 nicotinic acetylcholine receptor, which is how it reduces the urge to smoke, and it fully activates the α7 receptor. In animal studies, α7 activation has been linked to synaptic protection against the toxic effects of beta-amyloid, the protein that forms the plaques characteristic of Alzheimer's disease.
To validate the signal on independent data, the team compared varenicline with nicotine replacement therapy, the other approved method for quitting smoking. From All of Us, a database of American medical records, they selected 9584 individuals over 50 with no dementia at baseline who had first started one of the two treatments between 2006 and 2023. After matching on age, sex, race, weight, and nearly forty covariates, 1758 individuals remained in each group. After matching, the dementia risk reduction actually increased, from 42.5% to 48.4%. Over 15 years, those who took varenicline developed dementia at nearly half the rate (hazard ratio 0.516; 95% confidence interval 0.353 to 0.754).
An earlier trial of varenicline for Alzheimer's disease in 2014 found no benefit: in 66 patients with mild to moderate Alzheimer's, six weeks of treatment did not improve cognitive scores. The difference in study design explains the discrepancy: participants in the current work had no dementia at all at baseline, and preclinical data indicate that the protective effect of nicotinic receptor activation manifests early, while neurons and synapses are still intact.
The authors themselves identify this as an open question: varenicline is more effective at helping people quit smoking than nicotine replacement therapy, and smoking itself is a risk factor for dementia, so part of the protection may stem from more successful smoking cessation. However, in their own words, the effect size is larger than what smoking cessation alone typically produces, which points to a possible direct neuroprotective effect of the drug.
Varenicline is already approved, inexpensive, and has decades of safety data, so the path to a clinical trial is shorter than for a new compound. The authors propose testing it as a dementia preventive in a dedicated trial and suggest that clinicians may already have reason to favor it over nicotine replacement therapy for smokers at elevated dementia risk.