GDF15: The Blood Marker That Warns of Dementia 20 Years Early, and Why It May Not Be Alzheimer's

#1

A large multi-cohort study found that a single blood protein, GDF15, measured as early as middle age, predicts who will develop dementia over the following 15 to 25 years. The signal was strongest for vascular dementia, the type driven by damaged small blood vessels in the brain, and was weak or absent for the amyloid pathology that defines classic Alzheimer’s. Genetic causal-inference analysis and lab experiments on human immune cells suggest GDF15 is not merely a passive flag but may actively push the brain toward disease by dampening the antiviral arm of the immune system and disturbing metabolism in immune cells.

Some of the most useful predictions in medicine come from things that are easy to measure. GDF15 is one such thing: an stress-signaling protein released mainly by organs outside the brain, such as the kidney, that rises steadily as we age. Researchers led by a team at the US National Institute on Aging have now assembled evidence from six independent human populations, spanning the United States, United Kingdom, Iceland, and Japan, showing that people with higher GDF15 in their blood are substantially more likely to develop dementia decades later

The scale of the data is the story. In the American ARIC cohort, more than 11,000 middle-aged adults had blood drawn and were then followed for 20 years. Those in the top half for GDF15 had roughly double the dementia rate of those in the bottom half. The same pattern repeated in the UK Biobank across 35,000 people and in an Icelandic cohort of nearly 5,000 older adults. Higher GDF15 in midlife carried a risk equivalent to being about three years older than your actual age.

The more surprising finding is what kind of dementia GDF15 predicts. Across cohorts, the protein was two to five times more strongly linked to vascular dementia than to Alzheimer’s disease. It tracked with silent strokes, microbleeds, and white-matter damage on brain scans, but it did not track with amyloid plaques, the hallmark of Alzheimer’s. In other words, GDF15 appears to be reading out the health of the brain’s blood supply rather than its amyloid burden.

To move from correlation toward cause, the team used Mendelian randomization, a method that uses inherited genetic variants as natural, lifelong randomizers of GDF15 levels. This preliminary genetic evidence supported a causal contribution to dementia risk. They then exposed cultured human immune cells to GDF15 and watched it suppress interferon and antiviral pathways while rewiring cellular metabolism, the same pathways that independently predicted dementia in the cohorts.

The practical hope is a cheap midlife blood test to flag vascular brain risk long before symptoms, and eventually a drug target. The caution is that GDF15 is a stress alarm for the whole body, so lowering it blindly could do more harm than good. [Confidence: High for the association; Medium for causation]

Actionable Insights

This is an observational and genetic study. The take-home value is risk stratification and mechanism.

Magnitude of the signal. In midlife, people in the upper half of GDF15 had a 20-year dementia rate of 7.5 percent versus 3.9 percent in the lower half. That is an absolute risk difference of about 3.6 percentage points and a relative risk near 1.9, meaning roughly a doubling. In older adults the gap was larger: 18.7 percent versus 9.5 percent over just 7 years (absolute difference about 9 points). Statistically, each doubling of GDF15 raised all-cause dementia risk by 36 to 55 percent across cohorts, and vascular dementia risk by about 100 percent (hazard ratios near 2.0).

What it means for you. GDF15 behaves as a barometer of vascular and metabolic brain aging. Because it rose with poor kidney function, obesity, and smoking, the levers that plausibly move it are the standard cardiovascular ones: blood-pressure control, not smoking, metabolic and renal health, and cardiorespiratory fitness. A practical step is that if you get GDF15 measured (some aging panels include it), a high midlife value is a prompt to aggressively manage vascular risk factors, not a diagnosis. Note the paradox: metformin raises GDF15, yet high GDF15 tracks with harm here, so do not treat the marker as a simple target. [Confidence: Medium]

Context and Source

  • Open Access Paper: Plasma GDF15 affects long-term dementia risk and alters neuroimmune signaling.
  • Lead institution: Laboratory of Behavioral Neuroscience, National Institute on Aging, National Institutes of Health, Baltimore, Maryland, with a large international consortium (Johns Hopkins, University of Exeter, Icelandic Heart Association, University College London, and others). Country: United States (lead), with UK, Iceland, Germany, Japan, and Singapore collaborators.
  • Journal: Science Advances (2026).
  • Impact evaluation: The impact score of this journal is 13.9 (2025 Journal Impact Factor; CiteScore 19.6), evaluated against a typical high-end range of 0 to 60+ for top general-science journals, therefore this is a High impact journal.

Related Reading:

#2

Biomarker and Effect-Size Data (written for non-statisticians)

Because there is no lifespan outcome, the effect sizes below concern dementia risk and brain-imaging changes. Two effect-size languages are used: hazard ratios (HR, how much faster an event happens in a group) and standardized betas or SD units (how many standard deviations a marker moves, which functions like Cohen’s d for a continuous predictor).

Dementia risk (hazard ratios per doubling of GDF15):

  • ARIC late-life, 7-year follow-up: HR 1.61 (95 percent CI 1.36 to 1.90). Plain reading: about 61 percent higher dementia rate per doubling. High-versus-low group event rates were 18.7 percent versus 9.5 percent, close to a two-fold difference.
  • ARIC midlife, 20-year follow-up: HR 1.55 (1.32 to 1.82), about 55 percent higher risk per doubling. Event rates 7.5 percent versus 3.9 percent (high vs low).
  • UK Biobank, 14-year follow-up: HR 1.43 (1.29 to 1.59).
  • AGES-Reykjavik, up to 17 years: HR 1.36 (1.21 to 1.53).
  • Japanese NILS-LSA, N = 340: HR 1.19 (0.90 to 1.57), not statistically significant (the confidence interval crosses 1.0, so no reliable effect was detected in this smaller, East-Asian cohort).
  • Pooled meta-analysis: roughly 44 to 47 percent increased all-cause dementia risk per doubling.

Vascular versus Alzheimer’s split (the key result): per doubling of GDF15, vascular dementia HR was about 2.0 (UKB 2.01; AGES 2.06), meaning roughly a doubling of vascular dementia rate, versus Alzheimer’s HR of only about 1.20 to 1.24. The vascular effect is therefore two to five times larger than the Alzheimer’s effect.

Putting the size in human terms: the authors convert the GDF15 effect into “years of aging.” A one-unit GDF15 effect equaled roughly 2.6 to 3.75 years of additional chronological age across cohorts. So carrying high GDF15 looks like adding about three years to your brain’s clock.

Context against better markers: GDF15 is a weaker predictor than the established neurodegeneration markers. In the same data, neurofilament light (NfL) reached HR about 2.18 and pTau-181 about 2.05, both stronger than GDF15’s roughly 1.6 to 1.8.

Brain imaging (SD-unit effects, ARIC N = 1,345, per doubling of GDF15): 0.17 SD lower total brain volume, 0.27 SD lower cortical thickness, 0.31 SD lower temporal-parietal thickness, 0.18 SD higher white-matter hyperintensity volume, plus 40 percent higher odds of a lacunar infarct and 90 percent higher odds of a cerebral microbleed. These are small-to-modest per-doubling effects (0.1 to 0.3 SD is a small effect in Cohen’s d terms), but they point consistently at small-vessel and diffuse damage.

Fluid biomarkers (standardized beta, ARIC): pTau-181 beta 0.32 and NfL beta 0.29 (both real, small-to-moderate), whereas amyloid ratio Aβ42/40 (beta 0.005) and astrocyte marker GFAP (beta 0.02) showed essentially no association. GDF15 did not predict cortical amyloid on PET. This is the biochemical fingerprint of a vascular and axonal-injury signal, not an amyloid signal.