A new AI tool flags early Alzheimer’s, identifying about four in five people who would otherwise be missed by clinicians.

Trained on UCLA Health patient records and tuned to work more fairly across Black, Latino and Asian patients who are often underdiagnosed, the system aims to find people earlier, when treatment and lifestyle changes can still help.

In tests on more than 97,000 UCLA Health records, the model reached a sensitivity of about 77 to 81 per cent across non-Hispanic white, non-Hispanic African American, Hispanic/Latino and East Asian groups, roughly double that of conventional supervised models, the authors report.

UCLA Health researchers developed the tool and built fairness measures into training before checking the model’s picks against genetic benchmarks.

Patients flagged as likely cases but previously unlabelled showed higher polygenic risk scores and APOE ε4 counts than those the system did not flag.

Polygenic risk scores measure the combined effect of multiple genes on disease risk, while APOE ε4 is a genetic variant linked to increased Alzheimer’s risk.

The AI looks beyond memory-related billing or diagnostic codes and finds patterns that include signals such as decubitus ulcers, commonly known as pressure sores, and palpitations, which could prompt clinicians to take a closer look and consider screening.

“We were able to capture about 80 per cent of the people who actually would have undiagnosed Alzheimer’s disease,” said Dr Timothy Chang.

He said studies estimate up to 40 per cent of Alzheimer’s cases go undiagnosed, a gap that hits Black and Latino communities especially hard.

The framework learns from both labelled Alzheimer’s cases and unlabelled patient records, using race-specific probabilistic labelling and post-processing cutoffs tuned for group benefit equality to reduce bias.

Rather than relying only on diagnostic codes, the model draws on a wide range of electronic health record features, including diagnoses, encounter history and age, helping uncover likely cases that never received a formal label.

Earlier detection matters because disease-modifying treatments and targeted clinical referrals are now an option for people in the earliest symptomatic stages of Alzheimer’s.

Amyloid-targeting therapies such as lecanemab and donanemab are intended for early disease, which makes timely screening and specialist evaluation more consequential for patients and families.

Lifestyle changes and symptom management remain key tools for slowing decline, planning care and helping families prepare.

Researchers stress that the tool is a flagging system, not a diagnosis, and that its output should trigger follow-up evaluation rather than replace clinical judgement.

The team plans prospective validation in partner health systems to test how well the tool generalises and how useful it is in real-world practice before any routine roll-out.

Models can reveal new biases when used outside the environment where they were trained. Clinicians and ethicists will also have to weigh benefits against the risk of false positives, added patient anxiety and uneven access to specialist care.

The coming months will focus on broader testing and conversations with health systems about how to roll out the technology responsibly, if future studies support its use.

Gap is partnering with Spear Bio on Bio-Hermes-002, an Alzheimer’s study comparing blood and digital biomarkers across cognitive conditions.

The observational platform study compares blood-based and digital biomarkers (measurable signs in blood or behaviour) across a broad range of cognitive conditions, alongside MRI and PET brain scans and diverse racial and ethnic groups, to generate data that may help predict, detect and diagnose Alzheimer’s disease and related dementias.

The study gathers data on how each biomarker, or combinations of biomarkers, perform in assessing, diagnosing or predicting pathologies linked to Alzheimer’s and related dementias, including amyloid and tau in the brain.

Amyloid plaques and tau tangles are abnormal protein deposits and are hallmark features of Alzheimer’s disease.

John Dwyer, president of GAP, said: “We’re proud to have Spear Bio as a valued partner in the Bio-Hermes-002 study.

“Their technology platform is consistent with our goal to catalyse and scale early-stage disease diagnostics and monitoring, therby transforming how we treat patients and conduct clinical trials for Alzheimer’s and related dementias.

“By advancing biomarker detection, we hope to accelerate meaningful progress for individuals and families affected by these conditions.”

In Bio-Hermes-002, Spear Bio will provide results of blood-based biomarker analysis using its SPEAR UltraDetect assay platform, which the company says offers attomolar sensitivity from a one microlitre diluted sample, and what it describes as superior specificity in a homogenous wash-free format, aiming to change early-stage diagnostics and monitoring.

This builds on findings from Bio-Hermes-001.

Data from Bio-Hermes-002 will be stored on the AD Workbench from the Gates Ventures Alzheimer’s Disease Data Initiative (ADDI).

The AD Workbench is a global, secure, cloud-based data sharing and analytics environment that enables researchers worldwide to share, access and analyse data across multiple platforms.

Spear Bio joins industry partners in the study, including Biogen, Eli Lilly and Company, IXICO and Roche, along with a growing list of partners providing blood-based biomarker assessments or digital assessments for Bio-Hermes-002.

To date, the partners include AINOSTICS, Alamar Biosciences, Beckman Coulter Diagnostics, Cambridge Cognition Limited, Cognivue, Cumulus Neuroscience Limited, Fujirebio, iLoF, LifeArc, Linus Health, Lucent Diagnostics, a Quanterix brand, Sunbird Bio and ViewMind.