I doubt there are many people out there who would volunteer to learn the likelihood of them meeting their maker within the next year.
As a species, we prefer to push the inevitability of death to the back of our minds; with technological advances almost making us feel as if we could live forever.
However, as frightening as it may be to face, Artificial Intelligence (AI) can now predict your chances of shuffling off this mortal coil within the year.
As reported by New Scientist, a team at Pennsylvania based healthcare provider Geisinger Health System used an AI to examine 1.77 million electrocardiogram (ECG) results from a pool of nearly 400,000 people.
ECGs are used to track the electrical activity of the heart, with changes recorded to help diagnose heart conditions such as heart attacks and atrial fibrillation.
This pioneering team, led by chair of Geisinger’s Department of Imaging Science and Innovation, Brandon Fornwalt, trained up two distinct versions of the AI in question, giving them different data sets.
The first AI was fed just the raw ECG data – measuring voltage over time – while the other was given ECG data combined with the age and sex of the patient.
The team measured the performance of the AIs using AUC, a metric which tracks how accurately a model can distinguish between two groups. In this scenario, the groups were divided between patients who died within the year and those who lived.
The AI consistently scored above 0.85, with 1 being a perfect score, and 0.5 indicating no difference between the two groups. The current AUCs used for risk scoring models reportedly vary between 0.65 and 0.8.
In order to have a comparison, researchers built an algorithm based on ECG features currently used by medics, including specific patterns from the recordings.
Fornwalt said:
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.
The AI was able to accurately predict the likelihood of death within the next year even amongst those who appeared to have a normal ECG.
The AI was also able to detect risk patterns that three cardiologists were unable to pick up on after conducting normal-looking ECGs.
Fornwalt continued:
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.
AI can potentially teach us things that we’ve been maybe misinterpreting for decades.
Going forward, this research will be presented before the American Heart Association’s Scientific Sessions in Dallas on November 16.
If you have a story you want to tell send it to UNILAD via story@unilad.com
Jules studied English Literature with Creative Writing at Lancaster University before earning her masters in International Relations at Leiden University in The Netherlands (Hoi!). She then trained as a journalist through News Associates in Manchester. Jules has previously worked as a mental health blogger, copywriter and freelancer for various publications.