Would you like to know the date of your death? Researchers at Geisinger, a health care provider in Pennsylvania, USA, were able to predict which patients were at risk of dying during the year. To do this, they trained artificial intelligence (AI) to detect signs of potential heart problems in the future, such as heart attacks or atrial fibrillation.
Découvrir notre dernier podcast
The machine examined the results of 1.77 million electrocardiograms (ECGs), or heart activity records of nearly 400,000 patients. Two versions of the AI have been developed. One analysed the raw data, the other also received the age and gender of the participants. The conclusions of the study, relayed by New Scientist, will be presented at the American Heart Association Congress in Dallas (United States) on November 16th.
Things humans don't see
The AI predictions were then examined, using what is called the 'AUC.' This metric measures the performance of a model that distinguishes between two groups of people. In this case, it was deceased on the one hand and those who survived on the other. A score of 0.5 indicates no difference between the groups. A score of 1 is perfect. However, technology has consistently scored above 0.85. The doctors, on the other hand, had a score ranging from 0.65 to 0.8, according to the authors of the research.
According to them, the AI model, therefore, works better than existing methods to detect potential deaths. It even identified heart problems in patients already studied by cardiologists. Three doctors each examined 'normal' ECGs and were unable to detect risk profiles, as the machine did.
'This discovery suggests that the model sees things that humans probably can't see, or at least that we don't know and think are normal,' Brandon Fornwalt, the study's main investigator, told New Scientist. 'Artificial intelligence can teach us things that we may have misinterpreted for decades.'
A functioning that is still unclear
However, scientists still do not know which patterns are detected by the AI, and thus have difficulty explaining how it works. It is this lack of knowledge that worries health professionals, who are reluctant to make decisions based on a misunderstood algorithm.
This study is not the first attempt to predict death. Last year, Google researchers created a predictive model using electronic health records to predict the length of a patient's hospital stay, the time of departure and the time of death. AIs have also been developed to diagnose heart disease and lung cancer. And sometimes with more precision than human doctors.
