A new machine learning algorithm can predict whether heart patients are at risk of dying within a year. This is demonstrated by artificial intelligence at a health care center in the United States by predicting the likelihood of a patient's death within a year more accurately than a doctor.

In a report published in the British Telegraph newspaper, Tom Hoggins highlighted the ability of the machine learning algorithm to calculate cardiac survival by analyzing ECG results.

The algorithm analyzed a wide range of ECG results for more than 400,000 patients. This advanced program was able to outperform traditional methods of predicting the chances of survival of these patients used by doctors.

The Geisner Health System team in Pennsylvania trained the algorithm to work on two models. The first is to feed the initial ECG historical data and measure the effort over time. In the second model, ECG data were analyzed, as well as the age and sex of each patient.

The AI ​​program has been tasked with detecting patterns to predict the chances of a patient's death over the next year, or if he is at risk of additional complications, such as heart attack and atrial fibrillation.

Algorithm
A parallel algorithm was used that uses a traditional reading for comparison. Brandon Fornewalt of the Geisner Center told New Scientist that the machine-based model was the best finding.

The author explained that the results of traditional methods may not fully correspond to the individual estimates made by the doctor. However, the AI ​​program was able to accurately predict the deaths of people that doctors believe are healthy based on "normal" ECG results.

In fact, three cardiologists read the ECG results "naturally", but were unable to identify the risk patterns detected by the AI ​​program. Fornualt concluded that artificial intelligence can teach humans what they have failed to explain for decades. Because the test is based on a set of confidential historical data, and researchers have not identified the patterns on which this technology relies, there are still ethical obstacles to using artificial intelligence to treat patients.

Premature death
Geisner's study was not the first AI-based death prediction study.In April, the University of Nottingham published a study of a machine learning algorithm that could predict the early death of Britons aged 40-69. The study claimed that AI performance was better than traditional methods.

Last week, the US Food and Drug Administration made a proposal for a regulatory framework for the use of artificial intelligence in health care. The FDA "expects manufacturers to be transparent and monitor the performance of this technology on the ground," he said.

Although the use of artificial intelligence to change treatment may take some time, the FDA also emphasized that artificial intelligence and machine learning techniques “have the potential to transform health care by extracting new and important ideas from the vast amount of health care data available.”