Israeli-led AI breakthrough set to transform diabetes prevention

An artificial intelligence model developed by Israeli researchers can identify people at risk of developing diabetes more than ten years before diagnosis, reveals a new study.

The research, published this week in science journal Nature, was led by teams at Nvidia Israel, the Weizmann Institute of Science and Israeli health-tech firm Pheno.AI, alongside the Mohamed bin Zayed University of Artificial Intelligence.

Researchers say the AI system, known as GluFormer, can forecast the likelihood of type 2 diabetes up to 12 years in advance using patterns identified in continuous glucose monitoring data, significantly outperforming current indicators and raising the prospect of earlier intervention.

The model was trained using data from Israel’s long-running “10K Project”, one of the most comprehensive health datasets of its kind.

The initiative has tracked more than 14,000 participants over several years, combining continuous glucose measurements with genetic testing, blood and stool analysis, microbiome profiling, sleep studies, movement tests and detailed lifestyle and medical histories.

Professor Eran Segal, a computational biologist at the Weizmann Institute and senior author of the study, said the scale and depth of the data enabled the team to identify patterns that conventional tools fail to detect.

“People classified as pre-diabetic are often treated as a single group,” Segal said. “But existing measures do not reliably predict who will actually go on to develop diabetes. The AI model can.”

According to the findings, around 66 per cent of participants who later developed diabetes were flagged as high-risk by GluFormer years earlier, while only a small fraction of those assessed as low-risk went on to develop the disease. Although trained on glucose data, the model also demonstrated a strong ability to predict other diseases including cardiovascular mortality. It identified nearly 70 per cent of individuals who later died from heart-related causes as high-risk.

The researchers found that the system was able to anticipate a range of other health outcomes linked to metabolic health, including indicators associated with cardiovascular disease, kidney and liver function, blood lipid levels, visceral fat, and sleep disorders. In each case, the model outperformed other prediction methods based on glucose monitoring alone.

The technology uses a transformer-based architecture similar to that underpinning large language models, enabling it to extrapolate long-term health outcomes from relatively short sequences of data – sometimes as little as one to two weeks of glucose readings.

Guy Lutsker, an AI researcher at Nvidia and a doctoral student in Segal’s lab, said the system appears to have learned fundamental aspects of metabolic disease that are difficult to express in simple clinical rules.

“Like many powerful AI systems, it functions partly as a black box,” he said. “But it delivers predictions that go well beyond what existing methods can achieve.”

The findings have significant implications for healthcare providers, insurers and health-tech companies. Preventing diabetes is a major priority for health systems, particularly in the United States, where pre-diabetes interventions account for billions of dollars in annual spending.

The global cost of diabetes is projected to reach $2.5 trillion by 2030, according to estimates cited in the study.

By identifying which patients are most likely to progress to full diabetes, the technology could allow preventative resources, from lifestyle programmes to drug treatments, to be targeted far more precisely.

Pheno.AI holds the rights to commercialise the technology and plans to work with healthcare organisations to bring the model into clinical use.

The study highlights Israel’s growing role at the intersection of artificial intelligence, healthcare and big data, alongside the expanding footprint of Nvidia’s research activity in the country.

Writing on LinkedIn, Yuval Mann, Nvidia’s Israel’s head of communications, said GluFormer was another “powerful reminder” of how AI is set to revolutionise healthcare.

“Congratulations to Gal Chechik, Guy Lutsker, Eran Segal, and the entire research team behind this impressive achievement,” wrote Mann.

For Segal, the goal is prevention. “Only a minority of people defined as pre-diabetic will actually develop diabetes,” he has said. “The challenge is knowing who they are  and now, we’re much closer to answering that.”