Scientists say new technology can predict people’s health problems more than a decade in advance.
The artificial intelligence-based technology has learnt to find patterns in people’s medical history to calculate the risk of more than 1,000 diseases.
Researchers say it’s like a weather forecast that predicts a 70 per cent chance of rain – only for human health.
Their vision is to use the AI model to identify high-risk patients for disease prevention, and to help hospitals understand future demand in their regions years in advance.
The model, called Delphi-2M, uses technology similar to that behind well-known AI chatbots such as ChatGPT.
Chat AI learns to understand patterns can’t predict what’s coming next or when.
It doesn’t predict exact dates – for example, a heart attack on 1 October – but it does estimate the likelihood of developing 1,231 diseases.
“So, as with the weather, where we might have a 70 per cent chance of rain, we can do the same for medicine,” says Professor Ewan Birnie, acting director general of the European Molecular Biology Laboratory.
And we can do it not just for one disease, but for all diseases at once – we’ve never been able to do it before. I’m thrilled,” he adds.
Photo by Jeff Dowling/EMBL-EBI.
The AI model was originally developed using anonymised UK data, including hospital admissions, GP records and habits such as smoking, collected from more than 400,000 people as part of the UK Biobank research project.
The model was then validated using data from other Biobank participants, followed by the medical records of 1.9 million people in Denmark.
“It works well, very well in Denmark,” says Professor Birnie.
“If our model says the risk is one in ten over the next year, that’s what it seems to do,” he adds.
The model is best at predicting diseases such as type 2 diabetes, heart attacks and septicaemia, which have a clear progression, rather than more random events such as infections.
What can be done with the results
Doctors are already offering cholesterol-lowering statins to patients based on a calculated risk of heart attack or stroke.
The AI tool is not yet ready for clinical use, but the plan is to apply it in a similar way – identifying high-risk patients when it is still possible certain liver diseases will develop could be particularly useful to reduce alcohol consumption more than the general population.
Artificial intelligence can also help plan disease screening programmes and analyse all health records in a particular region to predict demand – for example, how many heart attacks there will be per year in Norwich in 2030 for resource planning.
“This is the beginning of a new way of understanding human health and disease progression,” says Professor Moritz Gerstung, head of the AI in Oncology department at the German Cancer Research Centre (DKFZ).
“Generative models like ours may one day help personalise care and predict the needs of the healthcare system on a large scale,” he adds.
The AI model, described in the scientific journal Nature, Nature based on UK Biobank data, which is mainly collected among people aged 40-70, rather than the whole population.
The model is now being refined so that it takes into account more medical data such as radiological diagnostics, genetics and blood tests.
“I want to emphasise that this is research – everything has to be tested, well-tuned and thought through before it can be used, but the technology is already making these predictions,” notes Professor Birnie.
He predicts the journey will be similar to the application of genomics in healthcare, when it took decades to move from scientists’ confidence in the technology to its routine use in medicine.
The research was a collaboration between the European Molecular Biology Laboratory, the German Cancer Research Centre (DKFZ) and;This research looks like a significant step towards a scalable, understandable and – most importantly – ethically responsible form of predictive modelling in medicine.