Algorithm Predicts Which Heart Failure Patients Will Live Or Die Better Than We Can
Just in case you needed more convincing that artificial intelligence (AI) and machine learning are more life-saver than world-ender, a team at the University of California Los Angeles (UCLA) have been showing off their new toy: A computer program that’s able to predict whether people will survive heart failure, and for how long.
Heart failure is part of a suite of cardiovascular diseases that kill 17.7 million people every single year. According to the World Health Organization (WHO), that’s 31 percent of global deaths.
Heart failure describes a situation wherein the heart is unable to pump blood around the body properly, due to stiffness or weakness of some kind. It’s a long-term condition, and it tends to worsen over time. Sometimes, people require transplants to survive if it gets bad enough, but this depends on how likely the patient is to survive if they have one – a call that’s not exactly easy to make.
Making these literal life-or-death judgments is difficult even for medical professionals, which is where UCLA's algorithm, improved over an older version, comes into play.
The team from UCLA explain that, aside from more conventional methods of assessment of heart failure risk and cardiac transplantation, machine learning – which uses statistical techniques to allow software to act autonomously – has also been tested out in this regard before.
Writing in the journal PLOS One, the team explain that “existing clinical risk-scoring methods have suboptimal performance.” To wit, they’ve launched their Trees of Predictors (ToPs), an algorithm that uses 53 data points to predict how long people with heart failure will live, with or without a heart transplant, which you can play around with here.
[youtube https://www.youtube.com/watch?v=HYWmYJNg5Jw]
Most of these points are associated with the potential recipients of a new heart; 14 apply to the donors, and six are linked to the compatibility between the two. Using machine learning, the algorithm was trained and tested on a database of patients who were registered for cardiac transplantation in the United States between 1985 and 2015. The more it learns, the more accurate it gets.