AI has promised to have a profound impact on many industries, but perhaps none more so than in healthcare, where the ability to rapidly digest and understand huge quantities of data has tremendous potential.
The latest example of this potential comes via a study published recently in PLOS One. The paper describes an AI-based system that can accurately predict the risk of death in people with heart disease.
“It won’t be long before doctors are routinely using these sorts of tools in the clinic to make better diagnoses and prognoses, which can help them decide the best ways to care for their patients,” the authors say. “Doctors already use computer-based tools to work out whether a patient is at risk of heart disease, and machine-learning will allow more accurate models to be developed for a wider range of conditions.”
Big data
The researchers were able to tap into electronic medical data from over 80,000 patients via the CALIBER platform. The team were hoping to determine whether they could develop a system to predict coronary artery disease, which is the leading cause of death in the United Kingdom.
It’s a condition that develops when the blood vessels that supply blood to the heart become damaged or narrowed. This eventually results in chest pain and shortness of breath, and ultimately a heart attack.
The team began by developing a prognostic model for coronary artery disease based upon the best expert opinion available. The model made predictions across 27 different variables, including age and gender. The AI-based system largely trained itself from a set of 600 or so variables.
Lo and behold, the AI-based system not only performed well, but performed better than the human-designed system in terms of predicting patient mortality. It also highlighted a number of new variables that the doctors hadn’t thought of.
“Along with factors like age and whether or not a patient smoked, our models pulled out a home visit from their GP as a good predictor of patient mortality,” the researchers say. “Home visits are not something a cardiologist might say is important in the biology of heart disease, but perhaps a good indication that the patient is too unwell to make it to the doctor themselves, and a useful variable to help the model make accurate predictions.”
At the moment, the project is purely at a proof-of-concept stage, but the team are confident that it can eventually be scaled up into a commercially available system that can assist doctors in making more accurate prognoses.
“Machine learning is hugely powerful tool in medicine and has the ability to revolutionise how we deliver care to patients over the next few years,” they conclude.
