The Daily Pennsylvanian is a student-run nonprofit.

Please support us by disabling your ad blocker on our site.


Researchers from Penn Medicine have created a method that flags at-risk patients who should begin to have a conversation about their end-of-life wishes. (Photo by Austinmurphy | CC BY-SA 3.0)

A Penn Medicine team has developed a machine learning algorithm that flags patients with cancer who are likely to die within the next six months to encourage patients to have end-of-life conversations with their families and doctors, Penn Medicine News reported.

Most patients with cancer do not get the opportunity to have conversations with their families and physicians about their preferences for advanced care, according to Penn Med News. This study, published in JAMA Network Open, is one of the first to examine the application of machine learning algorithms for cancer patient care and has the potential to be integrated into routine clinical practice, Penn Med News reported.

These conversations typically involve which treatments patients would like to pursue despite having incurable illnesses, and how and where they would like to die, according to the Hospice Foundation of America.

In an effort to open this dialogue earlier, researchers from Penn Medicine have created a method that flags at-risk patients who should begin to have a conversation about their end-of-life wishes. The algorithm involves analyzing demographic data, such as gender and age, as well as health information like blood pressure and electrocardiogram data. 

“Patients oftentimes don’t bring up their wishes and goals unless they are prompted, and doctors may not have the time to do so in a busy clinic," lead author, and Penn professor of Medical Ethics and Health Policy Ravi Parikh told Penn Med News. "Having an algorithm like this may make doctors in clinic stop and think, ‘Is this the right time to talk about this patient’s preferences?’"

Out of the patients that the algorithm flagged as “high priority” for these conversations, researchers found that 51% died within six months of the evaluation. Less than four percent in the “lower priority” group died in the same time period. Although this algorithm cannot directly predict how long a cancer patient has to live, the algorithms are thought to accurately identify patients who would benefit most from end-of-life conversations. 

This study was conducted with 26,525 adults being treated at two large academic cancer centers within the University of Pennsylvania Health System. Due to the accuracy of the algorithm, it is being implemented at a medical center that was not part of the original plan for the study and researchers are working on a randomized controlled trial that will last between three and six months.

The study was presented at the American Society of Clinical Oncology Supportive Care in Oncology Symposium in San Francisco.