Artificial intelligence (AI) that helps emergency call centres identify the early signs of a cardiac arrest may become mainstream in Europe.
The time it takes for an ambulance to reach the victim of a cardiac arrest can be the difference between life and death, literally.
However, so many variables impact this situation from the time the ambulance is called, to the traffic conditions en route, to the positive interventions of a bystander. All can be deciding factors on whether the patient will live or die.
The importance of early detection
According to statistics from a 2017 report, entitled Resuscitation to Recovery, compiled by the British Heart Foundation, just 8% of patients survive a cardiac arrest. In the majority of these cases, the primary cause of the arrest is an abnormality of the heart.
The speed of effective action is essential in the early phase of a cardiac arrest to maximise the chance of functional recovery, both for the heart and for the brain.
Beyond the immediate resuscitation period, a coordinated response between ambulance services and hospital specialists will increase the chances of a return to independent living.
What is a cardiac arrest?
A cardiac arrest is an electrical fault causing the heart to stop beating. It is fundamentally different from a heart attack, which is a blocked artery, limiting the circulation of blood around the body.
As the chances of surviving a cardiac arrest are very time-dependent, the longer the attempted resuscitation is delayed, the worse the outcome. In patients with a shockable heart rhythm, there is approximately a 10% reduction in survival for every minute’s delay in providing defibrillation.
To compound the problem, less than half (30-40%) of bystanders feel able to intervene when they witness someone collapse.
AI detection: Help could be with us
But great leaps have been made towards positive change. Danish firm Corti has developed artificial intelligence-led (AI) healthcare to assist emergency call centres in identifying the early signs of cardiac arrest and will begin trials across Europe this summer.
The startup company says its AI algorithms can recognise the signs of cardiac arrest in patients suffering cardiac arrests in their homes or in public, far quicker, and with more accuracy than humans.
The software will begin with four new pilots in as-yet-unnamed European cities. And although already deployed with success in Copenhagen, these additional pilots will further endorse its effectiveness.
The scheme is also in partnership with the European Emergency Number Association (EENA). EENA is a non-governmental organisation based in Brussels, with the mission to contribute to improving the safety and security of people.
How it works
Early recognition of the signs of cardiac arrest is crucial. Every minute of inaction will reduce an individual’s chances of survival by up to 10%.
Corti’s software connects with emergency calls and can detect particular verbal and non-verbal patterns of communication. These include the caller’s tone of voice and whether the subject is breathing or not.
It acts as a personal assistant for the dispatcher during the call, which can then prompt the dispatcher to ask specific and relevant questions. The software can then make a recommendation on whether it’s likely that the victim is suffering a cardiac arrest. If necessary, an ambulance can be sent, or instructions on administering CPR can be given.
Results to date
When conducting its tests, the startup company (Corti) reported pretty impressive results. In one particular study, when using a historical database of 161,650 emergency calls, the AI software identified 93.1% of out-of-hospital cardiac arrests compared to 72.9% recognised by the human dispatchers taking the original calls.
It was also far quicker than its human counterparts, spotting signs of a cardiac arrest in 48 seconds, compared to humans who averaged 79 seconds. Although these were tests on historical calls, similar results have been reported in the ongoing real-time tests in Copenhagen.
Although not as yet, it’s possible that medical professionals could well raise questions about how Corti’s software is unable to explain exactly how it makes its decisions. Machine learning software algorithms learn by combining large datasets and looking for patterns that match specific outcomes.
In an interview with The Verge, Corti’s CTO Lars Maaløe said the development team knows that certain words “have a higher impact on the final output than others”, but he says this analysis is necessarily “imprecise”. He also explained that what patterns the software spots and how it weights them is not part of the software’s design.
Corti is also yet to publish its study in full, which means that certain key figures (like the software’s false positive rate or the number of times it incorrectly identifies cardiac arrests) may still be unknown. Speaking to The Verge, Corti’s CTO Lars Maaløe said these statistics were being processed, but that the rate was “comparable to that of humans”.
Software is intended for guidance, not diagnosis
Corti is more than confident that its software makes the right decisions, and the test scenarios seem to back that up. But a concern may be that the AI could miss specific nuances or, when faced with unknown situations, make inaccurate judgments.
Maaløe gives the example of someone calling to a report that a loved one is suffering a cardiac arrest. In these incidents, the callers are likely to be more confident the person is breathing. Not because they are, but because they want it to be so. What is unsure is whether AI can pick up on these human subtleties?
It’s for this reason that Corti’s software doesn’t make the decision itself; it only offers guidance to a trained dispatcher.
Can we trust a machine?
As diagnostic tools such as these increase in use, questions among the medical community will no doubt continue as to what level we are willing to trust a machine with our lives. But for now, let’s watch and hope the trials are successful and do indeed save lives.