🔗 Share this article

Artificial intelligence is capable of forecasting people's health problems more than ten years ahead, according to researchers.

The algorithm has acquired the ability to identify sequences in patient clinical histories to determine the likelihood of numerous medical issues.

Experts describe it as a weather forecast that forecasts a 70% chance of precipitation – however applied to human health.

The goal is to use the AI model to spot high-risk patients to stop health issues and to assist medical facilities understand demand in their area, well in advance.

How It Works

The algorithm – called Delphi-2M – utilizes analogous approaches to familiar language models like ChatGPT.

Language models are educated to comprehend linguistic structures so they can forecast the arrangement of verbal elements.

This health model has been educated to detect patterns in anonymous medical records so it can predict what comes next and the timing.

It avoids estimating specific days, such as cardiac events on a particular date, but instead determines chances of 1,231 diseases.

"Similar to meteorological predictions, where we could have a significant likelihood of rain, we can apply this for medical care," explained the main investigator.

"This approach allows not just for one disease but all diseases simultaneously - it's unprecedented to achieve this previously."

Testing and Verification

This system was originally designed using deidentified British information - including hospital admissions, doctor's notes and daily routines like nicotine consumption - obtained from more than 400,000 people.

This system was then tested to see if the forecasts stacked up using information from other participants, and then with almost two million individuals' medical records in Denmark.

"It's good, quite effective across different populations," stated the principal investigator.

"Whenever the algorithm estimates a ten percent probability, it actually appears that it manifests to be that exact probability."

The algorithm is best at predicting conditions such as type 2 diabetes, heart attacks and sepsis that have a established advancement sequence, as opposed to unpredictable occurrences like infections.

Practical Applications

People are already offered cardiovascular drugs based on a calculation of their likelihood of cardiac events or cerebral incidents.

The algorithm is still undergoing testing for clinical use, but the goal includes to use it in a similar way, to detect at-risk cases while there is a chance to take action early and prevent disease.

Potential applications involve medicines or specific lifestyle advice - for example those predisposed to certain hepatic conditions showing improvement with cutting back their alcohol intake exceeding typical guidelines.

The artificial intelligence could also help inform health check initiatives and examine complete health information within a region to forecast requirements - including the number of heart attacks annually expected across defined regions within coming years, to help plan resources.

"This represents the start of an innovative approach to grasp medical outcomes and health deterioration," commented a leading expert in AI and oncology.

"Forecasting algorithms such as ours could potentially support tailor treatments and forecast wellness demands at scale."

Next Steps

This technology demands enhancement and testing before it is used clinically.

There are also potential biases as it was developed using data which is drawn mostly from specific age groups, instead of comprehensive demographics.

The algorithm is now undergoing improvements to incorporate more medical data including scans, DNA information and laboratory results.

"Just to stress that this remains experimental – everything needs to be verified and appropriately supervised and thoroughly evaluated prior to implementation," commented the principal scientist.

The researcher expects it will progress comparably to the use of genomics in healthcare where it needed extensive time to go from scientists being confident to healthcare being able to use it routinely.

A different researcher observed: "This investigation seems to be a major progression towards scalable, understandable, and – critically – principled approach to forecasting in healthcare."