Artificial intelligence at a US health centre can predict a person’s chances of dying from heart test results, including those that look normal to doctors, but how it works remains a mystery.

Algorithms developed by researchers at the health care provider Geisinger in Pennsylvania can calculate a patient’s survival rate within a year by analysing echocardiogram (ECG) results, according to an article published by New Scientist earlier this month.

The AI examined 1.77 million ECG results from 400,000 patients before concluding whether the patients would survive for the next year. Researchers trained the algorithm using two models. One was based on raw historical ECG data, measuring voltage over time, while the other was fed the ECG data in combination with the age and sex of the patients.

When comparing the two groups of patients, those that died within a year and those that survived, the AI scored above 0.85, where 1.0 was a perfect score, while traditional methods by doctors scored between 0.65 and 0.8.

“No matter what, the voltage-based model was always better than any model you could build out of things that we already measure from an ECG,” Brandon Fornwalt of Geisinger told New Scientist.

A parallel algorithm used traditional readings of ECGs that doctors would use to detect conditions such as heart attacks and atrial fibrillation.

The AI accurately detected the risk of death for some patients that were deemed to have a normal ECG by three separate cardiologists, according to the study.

“That finding suggests that the model is seeing things that humans probably can’t see, or at least that we just ignore and think are normal,” said Fornwalt.

The study comes as human doctors are using AI to not only increase efficiency but also improve diagnoses. The technology at the core of the so-called fourth industrial revolution is already being put to use to speed up drug discovery, predict flu outbreaks and detect cases of cancer.

Last year, a study from Denmark showed that AI can exceed the ability of human doctors in spotting signs of cardiac arrest. Researchers from both China and the US have developed facial recognition software that is able to screen newborns for genetic disorders.

The US Food and Drug Administration expects AI technologies to transform health care by “deriving new and important insights” from the vast amount of data generated everyday. It predicted “high value applications” would be found in various areas, including early disease detection, identification of new patterns of human physiology, and in the development of personalised therapeutics.

For more insights into China tech, sign up for our tech newsletters, subscribe to our award-winning Inside China Tech podcast, and download the comprehensive 2019 China Internet Report. Also roam China Tech City, an award-winning interactive digital map at our sister site Abacus.