Artificial intelligence hides winners in ‘failed’ Alzheimer’s drug trial

After researchers used artificial intelligence to divide patients into fast, slow disease progressers, an allegedly unsuccessful Alzheimer’s drug trial achieved surprising results, which reduced cognitive decline in a group by 46%.

The findings, published today in Nature Communications, demonstrate how AI can greatly improve drug development for dementia by determining which patients are most likely to benefit from treatment. The approach could cut clinical trial costs and accelerate the search for effective therapies in areas plagued by 95% failure rates.

When “fruit” becomes promising

Lanabecestat’s Amaranth trial, a drug designed to clear the brain’s toxic amyloid, was terminated after showing no obvious benefits in all participants. But another story emerged when researchers at the University of Cambridge applied their AI models to reanalyze the data.

An AI system that received brain scans, genetic data and amyloid measurements divided 1,354 trial participants into two groups: slowly developing Alzheimer’s disease and rapidly declining progression in people. Although the drug failed to slow down cognitive decline in fast-progressive people, it turned out to be very effective in the slow group.

“When they don’t have the chance to benefit from it, they don’t benefit from people, and their new drugs fail,” said Professor Zoe Kourtzi of the Cambridge Department of Psychology who led the study. “With our AI model, we can eventually identify patients accurately and match the right patients with the right drugs.”

The right patient at the right time

The key insights rotate around time. Alzheimer’s disease gradually develops through a series of brain changes, starting with amyloid deposits, and eventually triggering widespread damage. AI analysis shows that progress is slow:

• Reduce the burden of amyloid in the brain
• Save better brain tissue in key areas
• Excellent performance of cognitive testing

These patients are essentially at an earlier stage of the disease and interventions may still make a difference. In contrast, fast-progressors have suffered too much brain damage to amyloid-slipping drugs, which can still be helped by symptoms despite successfully reducing protein accumulation.

This finding is consistent with the recently FDA-approved drug for Alzheimer’s disease (such as lecanemab) that shows moderate benefits, but only in carefully selected early stage patients. The Cambridge AI approach can make the choice of such patients more accurate.

Reduced by 90% of the test size

In addition to identifying effective treatments, AI stratification also provides practical solutions to the huge cost and complexity of clinical trials. The researchers calculated that focusing on slow progressers could reduce the required trial participants by 90%, from 762 patients per treatment group to just 82.

“This makes the trials more precise, so they are progressing faster and costly, turbocharged to find the urgently needed precision medicine methods for dementia treatment,” Kurtz explained.

Based on memory tests, MRI scans and blood tests when predicting disease progression, the AI model has three times more accurate than standard clinical evaluations. It analyzes the complex relationship between brain imaging, genetic factors and protein levels to generate individual patient prognostic scores.

From research to reality

Health innovation in the NHS Innovation division East England is now supporting efforts to translate this AI-Spable approach into clinical practice. These implications go beyond the design of the trial, which has the potential to change the way doctors diagnose and treat patients with dementia.

“This AI-enabled approach could have a significant impact on the stress and cost of NHS care by enabling more personalized drug development,” said Joanna Dempsey, chief consultant at Health Innovation East England.

This study addresses the key needs of dementia care, and the global burden is expected to triple by 2050. Despite $43 billion spent in three decades on research and development, effective treatments remain elusive. The disease is currently selling for $1.3 trillion per year worldwide.

For Kourtzi, the work is personal. “Like many people, I watched hopelessly when dementia stole a loved one from me,” she said. “We have to speed up the development of dementia drugs. More than £40 billion of research and development have been spent – we can’t wait for 30 years.”

Researchers envision augmenting their approaches to support more complex trial designs that may test multiple treatments simultaneously and may accelerate precise medical interventions for different patient subgroups.

Although AI models currently require PET scans and MRI imaging, future versions may undergo less invasive blood tests, making the method easier to access and cost-effective for widespread clinical use.

Related