The list is on a lot of fridges by now. Hearing aids. Blood pressure. Keep moving, keep talking to people, treat the low mood, check the cholesterol, mind the air you breathe. The 2024 report of the Lancet standing Commission on dementia prevention, intervention and care, published on July 31, 2024, named 14 modifiable risk factors and estimated that addressing them could prevent or delay around 45 percent of dementia cases worldwide. It is the most quoted number in prevention. What it has rarely been asked to do is prove itself against a blood test.

That is what a team led by Carol Franz and William Kremen at the University of California San Diego set out to do, using one of the better-characterised ageing cohorts in the United States. They took 936 men from the Vietnam Era Twin Study of Aging, scored all 14 Lancet factors at an average age of 62 (education, hearing, vision, depressive symptoms, social engagement, exercise, traumatic brain injury, fasting glucose, LDL cholesterol, body mass index, blood pressure, smoking, alcohol and fine particulate air pollution), then looked six years later, at an average age of 68, at four plasma markers of Alzheimer's biology (the amyloid 42/40 ratio, pTau231, neurofilament light chain and GFAP) plus a diagnosis of mild cognitive impairment. Because the factors tangle with one another, they used Random Forest machine learning with nested cross-validation rather than a tidy row of separate regressions, with and without adjustment for age and APOE status. These are preliminary findings, presented at the Alzheimer's Association International Conference and published as a conference abstract in the Public Health supplement of Alzheimer's & Dementia on December 23, 2025, not as a full peer-reviewed paper.

The result is a flat line where many people expected a slope. Taken together, the 14 factors accounted for 0 percent of the variance in plasma pTau231 and in the amyloid 42/40 ratio, 7 percent of the variance in neurofilament light and 8 percent in GFAP. For predicting mild cognitive impairment by age 68, the area under the curve was 0.62, which is better than a coin toss and a long way from a test you would act on. The strongest contributors to neurofilament light were fasting glucose, education and body mass index; for GFAP, body mass index, education and air pollution. For MCI, blood pressure led, followed by air pollution, education, LDL, BMI and glucose, with smoking last. When APOE genotype was added, it was the least important contributor of all to NfL, GFAP and MCI, a finding that says more about how Random Forest ranks correlated inputs in this particular sample than about the gene's weight in anyone's life.

It is worth being careful about what just failed here, because it is not the advice. Amyloid, tau, axonal damage and astrocyte activation measured in plasma are not dementia; they are a partial, noisy readout of brain biology, and three of the Lancet factors (education, hearing, head injury) are thought to work partly through cognitive resilience, which by definition would not show up in a blood tube. Six years is a short window at an age when Alzheimer's pathology has often been accumulating for two decades. And the 45 percent estimate was never a promise about one man's bloodwork: it is a population attributable fraction, an arithmetic of how much dementia might disappear from a whole society if a risk factor were eliminated, which is a different question from how much pTau231 sits in the plasma of a 68 year old.

The sample matters too, and the authors would be the first to say so. The Vietnam Era Twin Study of Aging is men only, recruited from a registry of twins who served in the military between 1965 and 1975; a 2024 analysis of the same cohort described 1,476 participants who were 97.6 percent non-Hispanic and 92.5 percent White. Nothing here speaks directly to women, to people who were not veterans, or to the communities carrying the heaviest dementia burden. This is also observational work: the risk factors were measured, not assigned, so nothing in it establishes that changing one would change the other.

So what does a family do with a study that found so little? Mostly, keep going. The factors that did rise to the top of the list, blood pressure, air pollution, education, cholesterol, weight and glucose, are the same ones your clinician is already watching, and they are watched because of decades of outcome data, not because of a six-year plasma signal. The more useful message is for the people designing the next prevention trial: if blood biomarkers are to be the scorecard for midlife risk reduction, this cohort suggests that over six years the scorecard may be close to blank, and that trials built on it should say so before they start measuring.