Scientists train AI model to predict future illnesses

Artificial intelligence can predict people’s health problems over a decade into the future, say scientists.

The technology has learned to spot patterns in people’s medical records to calculate their risk of more than 1,000 diseases.

The researchers say it is like a weather forecast that anticipates a 70% chance of rain – but for human health.

Their vision is to use the AI model to spot high-risk patients to prevent disease and to help hospitals understand demand in their area, years ahead of time.

The model – called Delphi-2M – uses similar technology to well-known AI chatbots like ChatGPT.

AI chatbots are trained to understand patterns of language so they can predict the sequence of words in a sentence.

Delphi-2M has been trained to find patterns in anonymous medical records so it can predict what comes next and when.

It doesn’t predict exact dates, like a heart attack on October 1, but instead estimates the likelihood of 1,231 diseases.

“So, just like weather, where we could have a 70% chance of rain, we can do that for healthcare,” Prof Ewan Birney, the interim executive director of the European Molecular Biology Laboratory, told me.

“And we can do that not just for one disease, but all diseases at the same time – we’ve never been able to do that before. I’m excited,” he said.

The AI model was initially developed using anonymous UK data – including hospital admissions, GP records and lifestyle habits such as smoking – collected from more than 400,000 people as part of the UK Biobank research project.

The model was then tested to see if its predictions stacked up using data from other Biobank participants, and then with 1.9 million people’s medical records in Denmark.

“It’s good, it’s really good in Denmark,” says Prof Birney.

“If our model says it’s a one-in-10 risk for the next year, it really does seem like it turns out to be one in 10.”

The model is best at predicting diseases like type 2 diabetes, heart attacks and sepsis that have a clear disease progression, rather than more random events like infections.