New test detects dementia risk nine years before onset with over 80% accuracy

The model was able to detect disease indicators up to nine years before diagnosis. The study was published in the journal Nature Mental Health.

The model tracks changes in the “default mode network,” a brain region that becomes active when a person is at rest and not engaged in a specific task, and is thought to contribute to daydreaming and self-reflection. Researchers examined the disconnection among 10 key regions within this network.

The images were sourced from the UK Biobank and included 81 individuals who were not diagnosed with dementia at the time of imaging but developed it within nine years, compared with 1,030 matched control participants. When applied to medical records, the model predicted the disease with accuracy within a two-year window prior to diagnosis.

Researchers also found an association between disruption of this network and risk factors, notably social isolation and genetic predisposition to the disease.

Claire Sexton of the Alzheimer’s Association notes that several studies link Alzheimer’s disease to reduced functional connectivity in this network, consistent with known brain changes that begin years, and possibly decades, before diagnosis.

However, Professor Charles Marshall, the study’s supervisor, emphasizes that most cases are attributable to Alzheimer’s disease alone or in combination with vascular dementia, and calls for expanding research to include rarer forms such as frontotemporal dementia and Lewy body dementia.

Marshall says that some brain regions show decreased activity while others show increased activity, likely as a compensatory response. The model was trained to recognize “dementia-like” patterns.

Sexton identifies three limitations: variability in the definition and assessment of network disruption across studies; reliance on physician coding rather than standardized diagnostic criteria for dementia; and the fact that the UK Biobank sample is predominantly white, healthier, and higher-income than the general population, which limits the generalizability of the findings.

There is no preventive drug. Sexton adds that if the test proves clinically useful, these patients would undergo more precise follow-up with structural MRI, but she notes that no neuroprotective drugs currently exist, even if at-risk individuals can be identified early.