Software Can Now Predict If Some Patients Will Die Within a Year

The guesses were 20 percent more accurate than those made by physicians.

by Susan Rinkunas
Jan 17 2017, 7:05pm

British researchers have developed artificial intelligence that can predict when people with a certain heart condition will die, and it could help doctors personalize their treatment.

For a study published in the journal Radiology, researchers with the UK's Medical Research Council wanted to predict the outcomes of patients with pulmonary hypertension, or high blood pressure in the arteries that carry blood to the lungs. Pulmonary hypertension (PH) can lead to hardening and narrowing of these arteries, which forces the heart to work harder to pump blood through and can weaken it over time. About a third of people with PH die within five years of being diagnosed.

So they uploaded MRI scans of 256 patients' beating hearts along with blood test results. The software analyzed more than 300,000 different points on the heart with each beat and, after comparing that information to eight years of health records, it learned which abnormalities predicted when people with PH would die. The tool correctly predicted which people would be alive after one year about 80 percent of the time; physicians have a 60 percent accuracy rate.

The authors say this software could allow doctors to tailor treatment for PH and save the lives of people who need a more aggressive approach, like a lung transplant. 

But the researchers aren't quite ready to roll the tech out to cardiologists just yet: First, they want to test it out in more patients and also see if it can accurately predict death from other forms of heart failure like cardiomyopathy, which can require some patients to get a pacemaker. 

In the meantime, we can all contemplate a grim future in which one click is all that may separate you and the knowledge of exactly how long you've got.