🔗 Share this article

Artificial intelligence can predict medical conditions over a decade ahead, according to researchers.

This system has learned to spot patterns in people's medical records to calculate their risk of over a thousand conditions.

Scientists say it is like a weather forecast that predicts a 70% chance of rain – though focused on individual medical outcomes.

The objective is to implement this technology to identify vulnerable individuals to stop health issues and to support healthcare centers anticipate needs in specific regions, years ahead of time.

Underlying Technology

This system – referred to as this predictive tool – employs comparable methods to well-known AI chatbots such as language processors.

Language models are trained to understand patterns of language so they can anticipate the order of verbal elements.

The predictive system has been trained to find patterns in anonymous medical records so it can predict what comes next and at what point.

It avoids estimating precise timelines, like a heart attack on October 1, but instead determines chances of multiple health issues.

"Comparable to climate forecasting, wherein there might be a significant likelihood of showers, we can implement similar methods for wellness management," commented one lead researcher.

"This approach allows not just for one disease but all diseases at the same time - we've never been able to achieve this previously."

Development and Confirmation

This system was initially developed using anonymous UK data - covering medical admissions, doctor's notes and personal behaviors like nicotine consumption - obtained from more than 400,000 people.

The algorithm was then examined to verify if the forecasts proved accurate using information from further subjects, and then with a vast population's health data obtained internationally.

"Results are promising, it's really good across different populations," stated the principal investigator.

"Whenever the algorithm estimates a specific likelihood, the data confirms that it manifests to be the predicted rate."

The algorithm is most accurate with conditions such as adult-onset diabetes, myocardial infarctions and systemic inflammation that have a defined development pattern, as opposed to more random events including viral conditions.

Implementation Scenarios

Patients sometimes get cardiovascular drugs through probability estimation of their probability of cardiac events or cerebral incidents.

The AI tool is not yet approved for healthcare implementation, but the intention involves to use it in a similar way, to detect at-risk cases while there is a window for prevention early and prevent disease.

This could include pharmaceutical interventions or specific lifestyle advice - including individuals likely to develop some liver disorders gaining advantage through moderating drinking habits beyond standard recommendations.

This technology could also contribute to planning disease-screening programmes and analyse all the healthcare records within a region to forecast requirements - including the number of cardiac events a year there will be across defined regions in 2030, to help plan resources.

"This marks the commencement of a new way to comprehend wellness and health deterioration," observed an authority figure in AI and oncology.

"Forecasting algorithms including these approaches could one day help personalise care and predict medical requirements across populations."

Next Steps

This technology requires improvement and verification before it is used clinically.

Furthermore present potential biases as it was developed using information primarily obtained from specific age groups, instead of comprehensive demographics.

The model is now receiving enhancements to account for more medical data like radiographic studies, DNA information and blood analysis.

"We must highlight that this is research – everything needs to be verified and appropriately supervised and thought about ahead of application," explained the main investigator.

Scientists project it will develop analogously to genetic testing implementation in healthcare where it took a decade to go from scientists being confident toward clinical application to implement it standardly.

Another expert stated: "This research looks to be an important advancement towards scalable, interpretable, and – critically – ethically responsible form of predictive modelling in healthcare."