AI algorithm can rule out heart attacks with 99.6% accuracy

Heart attacks could soon be diagnosed faster and more accurately thanks to a new test developed with artificial intelligence.

It’s hoped the UK-led breakthrough could reduce the pressure on accident and emergency departments and help end inequalities in diagnosis, which previous research has shown sees women 50% more likely to get a wrong initial prognosis.

People who are initially misdiagnosed have a 70% higher risk of dying after 30 days, according to statistics from the British Heart Foundation.

But researchers at the University of Edinburgh, who developed the algorithm it’s hoped will soon be used by doctors, say it is an opportunity to end this clinical bias.

A trial of 10,286 patients in six countries found that compared to current testing methods this algorithm, named CoDE-ACS, was able to rule out a heart attack in more than double the number of patients with an accuracy of 99.6%.

Clinical trials are now underway in Scotland with support from Wellcome Leap – which focuses on discovery and innovation to improve human health – to assess whether the tool can help doctors reduce the pressure on overcrowded emergency departments.

Professor Nicholas Mills, professor of cardiology at the Centre for Cardiovascular Science at the University of Edinburgh, who led the research, said: “For patients with acute chest pain due to a heart attack, early diagnosis and treatment saves lives.

“Unfortunately, many conditions cause these common symptoms, and the diagnosis is not always straight forward. Harnessing data and artificial intelligence to support clinical decisions has enormous potential to improve care for patients and efficiency in our busy emergency departments.”

The current gold standard for diagnosing a heart attack is measuring levels of the protein troponin in the blood.

But the same threshold is used for every patient. This means that factors like age, sex and other health problems which affect troponin levels are not considered, affecting how accurate heart attack diagnoses are.

This can lead to inequalities in diagnosis.

CoDE-ACS – which stands for Collaboration for the Diagnosis and Evaluation of Acute Coronary Syndrome – was developed using data from more than 10,000 patients in Scotland who had arrived at hospital with a suspected heart attack.

It uses routinely collected patient information, such as age, sex, ECG findings and medical history, as well as troponin levels, to predict the probability that an individual has had a heart attack.

The result is a probability score from 0 to 100 for each patient.

The AI tool performed well regardless of age, sex, or pre-existing health conditions, showing its potential for reducing misdiagnosis and inequalities across the population.

The researchers say CoDE-ACS has the potential to make emergency care more efficient and effective, by rapidly identifying patients that are safe to go home, and by highlighting to doctors all those that need to stay in hospital for further tests.

The work, funded by the British Heart Foundation and the National Institute for Health and Care Research, has been published in the journal Nature Medicine.

Cardiovascular diseases are the leading cause of death globally with an estimated 17.9 million lives lost. More than four out of five CVD deaths are due to heart attacks and strokes.

The average age of people at the time of their first heart attack is 65.5 years for men and 72 years for women.

According to the British Heart Foundation, as many as 100,000 hospital admissions every year in the UK are due to heart attacks – that’s one every five minutes.

Professor Sir Nilesh Samani, Medical Director of the British Heart Foundation, said: “Chest pain is one of the most common reasons that people present to emergency departments. Every day, doctors around the world face the challenge of separating patients whose pain is due to a heart attack from those whose pain is due to something less serious.

“CoDE-ACS, developed using cutting edge data science and AI, has the potential to rule-in or rule-out a heart attack more accurately than current approaches. It could be transformational for emergency departments, shortening the time needed to make a diagnosis, and much better for patients.”

This new test is the second medical AI-related announcement to be made within a matter of days.

Research led by investigators at Harvard Medical School and the University of Copenhagen in collaboration with VA Boston Healthcare System, Dana-Farber Cancer Institute, and the Harvard TH Chan School of Public Health, demonstrated that AI was able to determine a person’s risk of developing pancreatic cancer with astounding accuracy up to three years prior to their actual diagnoses, based solely on their medical records.

In March this year researchers in Canada announced that artificial intelligence had developed a treatment for the most common type of liver cancer, hepatocellular carcinoma, in just 30 days and could predict a patient’s survival rate.

And AI is already being used to help develop new drugs, in surgery and for personalising treatment.

Of this latest study, Steve Goodacre, professor of emergency medicine at the University of Sheffield, said: “This intriguing study shows how AI can use complex analysis, rather than a simple rule, to improve diagnosis.

This doesn’t (yet) show that we can replace doctors with computers. Experienced clinicians know that diagnosis is a complex business. Indeed, the ‘ground truth’ used to judge whether the AI algorithm was accurate was a judgement made by clinicians.

“It will be interesting to see how clinicians in the emergency department use this algorithm. What will they do if they think the algorithm has got it wrong? The next stage of the research will hopefully answer that question.”