As a practicing oncologist, Ravi Parikh, M.D. believes it's vital that all cancer patients have conversations with their doctors about their goals and wishes for treatment early in the course of their illness.
But oncologists are often so focused on the task at hand—managing symptoms and looking for side effects—that it’s difficult to take a step back and determine which patients should be having conversations about their care goals before it becomes too late, Parikh told FierceHealthcare.
“On any given day, it’s actually pretty difficult to identify which patients in my clinic would benefit most from a proactive advanced care planning conversation,” Parikh, an oncologist at Penn Medicine, said.
Most patients with cancer who die as a result of the illness did not have documented conversation about their treatment goals and end-of-life preferences and without the support of hospice care, according to research.
RELATED: Harvard, Dana-Farber AI challenge uses crowdsourcing to improve cancer care
Parikh and a team of Penn Medicine researchers saw an opportunity to use advanced data analytics to flag cancer patients at risk of short-term mortality to spur doctors to have those discussions sooner rather than later.
The research team developed a machine-learning algorithm tied to oncology patients' outpatient medical records that identifies patients who would benefit most from a timely conversation about their end-of-life goals and wishes, said Parikh, who is also an instructor of Medical Ethics and Health Policy at the University of Pennsylvania and a staff physician at the Corporal Michael J. Crescenz VA Medical Center.
In a study testing the effectiveness of the algorithm, published in JAMA Network Open, the researchers found 51% of the patients designated as “high priority” were deceased six months after being evaluated compared to just 4% of those identified as “low priority.”
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, according to Parikh. "Having an algorithm like this may make doctors in the clinic stop and think, ‘Is this the right time to talk about this patient’s preferences?'," he said.
Parikh notes that the algorithm is intended to inform conversations about goals and wishes between doctors and patients—not to direct treatment decisions.
"Our hope is that the use of this algorithm will encourage conversations that need to be had between patients and clinicians," he said. "We want to make sure that the care patients receive is in line with what they want, rather than just assume that a certain treatment management strategy is the right way to go."
Using advanced analytics in clinical practice
The researchers applied three different predictive models to 26,525 patients receiving outpatient oncology care at two hospitals within the University of Pennsylvania Health System. Each used information commonly available in patients’ electronic health records: demographic characteristics such as gender and age, standard comorbidity data such as whether a patient has high blood pressure and laboratory and electrocardiogram data.
Roughly half of the high-risk patients died within six months and almost 65% of the high-risk patients died approximately a year-and-a-half later, compared to 7.6% of low-risk patients, according to the study.
When 15 oncologists were surveyed, they agreed that 60% of those identified by the algorithm as “high risk” needed to have immediate conversations about their care goals and wishes, Parikh said.
The Penn Medicine researchers previously developed a different algorithm for similar purposes, called Palliative Connect. That system is used to trigger consultations with Palliative Care staff and was recently found to be effective, increasing consultations by 74%.
The latest study is different in that it is targeted at increasing conversations between patients and the oncologists themselves in the outpatient setting, Parikh said.
RELATED: Kaiser Permanente develops machine learning tool to predict HIV risk
Researchers are now testing whether the tool actually does prompt doctors to initiate those end-of-life care discussions. The research team is currently working on a randomized controlled trial involving around 100 clinicians that will last between three and six months.
"This represents low-hanging fruit for where machine learning and other types of advanced data analytics technologies can be applied in practice," Parikh said. "Machine learning and analytics are coming into clinical practice more and more and the question now is how do we do it the right way to support doctors and make sure patients get the care they want."
While the use of big data in healthcare rapidly increases, there is ongoing concern that algorithms may reproduce racial and gender disparities via the people building them or through the data used to train them.
"There is certainly potential for unintentional bias in any algorithm, particularly those that are based on electronic health record data," Parikh said. "We are currently assessing whether there is any differential performance of the algorithm in important subgroups that may be subject to bias."
With adequate input and discretion from oncologists and other clinicians, the Penn Medicine researchers believe the algorithm they developed will not reinforce existing bias in clinical practice, he said.