AI can predict when you’ll keel over and die clutching at your chest from a heart attack better than doctors can, apparently.
A group of researchers scraped together dataset from over 80,000 patients from CALIBER, a large clinical research project studying mortality in England, and used machine learning methods to find common patterns in the data to predict the likelihood of a patient dying from coronary heart disease. The results have been published in the PLOS One journal.
The disease develops when the coronary arteries that supply the heart muscle with blood are blocked by fatty deposits that build up over time. In the worst case scenarios, one or more coronary arteries are completely obstructed, leading to a heart attack.
First, the researchers built a predictive model by manually choosing 27 variables that they believed were good indicators of heart diseases. These included things such as a patient’s age, gender, and whether they were experiencing chest pains.
Next, they used machine learning algorithms such as random forests and elastic net regression to automatically pick out variables worth studying from the dataset. The model picked out 586 variables automatically.
Both models were tested on how accurate they were at predicting whether a patient would be dead within five years. The machine learning model was slightly more statistically accurate scoring a 0.801 compared to 0.793 to physical doctor diagnoses.
"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," said Andrew Steele, co-author of the paper and a scientist at the Francis Crick Institute.
US watchdog OKs robo-doc AI that spies eye disease all on its ownREAD MORE
The system picked out new variables doctors don’t normally consider. "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," Steele said.
It’s not the first time this sort of grim research has been touted. Another group of researchers also studied mortality from heart attacks from pulmonary hypertension, a symptom describing high blood pressure in the blood vessels connecting the heart to the lungs.
“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,” Steele concluded. ®