Virtual Tumor Board Faster and Better Than Real Life

Helen Leask

October 11, 2019

A virtual molecular tumor board (VMTB) platform that uses machine learning is blowing the socks off conventional methods of making cancer-treatment decisions. The cloud-based technology was better at identifying biomarker-based treatment options in 80% of patients, and delivery time for reports fell from 14 days to 4 days during the 4-year study, report researchers at Georgetown University Medical Center in Washington, DC.

They believe their platform has the potential to make precision oncology a clinical reality for all US cancer patients, not just those at specialized centers.

A study of the VMTB platform involving 1725 patients was published online October 9 in the Journal of the American Medical Informatics Association.

"The point was, could we bring tumor boards to every patient with cancer?" said senior author Subha Madhavan, PhD, chief data scientist at Georgetown University Medical Center.

"Right now, only the complex cases or the ones that don't have local expertise [receive a tumor board] ― there's a limitation in time and resources. But using this technology, you could scale it and make it available to larger groups of patients," he added.

Even within specialized centers, tumor boards are time consuming and painstaking. Overburdened oncologists can't stay on top of the rapid evolutions in molecular oncology, Madhavan commented. "The physician may have seven or eight data elements they are considering for treatment options, but we could put in 51,000 data points using this methodology."

The VMTB software in the study crunched the patient's molecular profile, medical history, and demographics against 51,165 variables. These were drawn from gene-variant databases, 2064 clinical trials, clinical-trial registers, and drug-gene relationships, among other sources. After a review by experts, the software generated a prioritized list of therapy options for each patient, as well as a list of clinical trials being conducted within a 2-hour drive of the patient.

Many Patients With Pancreatic Cancer

In the study, the VMTB platform was added to usual cancer care for 1725 patients referred to Georgetown University Medical Center by advocacy organizations across the continental United States from 2014 to 2017.

Most of the patients had pancreatic cancer. This is no coincidence, said Madhavan: "[These patients and physicians] are in a dire situation, so they are much more open to using these technologies."

The cloud-based technology allowed more than 200 clinical teams across the United States to integrate data and reach consensus on final reports using email and, later, a purpose-designed chat room. In effect, the VMTB platform allowed community hospitals to take advantage of the expertise of a small group of physicians trained in precision oncology.

During the 4 years of the study, the turnaround time for case review fell from 14 days to 4 days, a statistically significant drop.

The volume of cases reviewed each year increased exponentially, from 46 in 2014 to 622 in 2017.

In approximately 80% of cases, the VMTB was better than commercial laboratory reports at identifying biomarkers amenable to therapy or clinical trials.

An analysis of real-world outcomes of 343 patients found that 47% of the patients were given the top-scoring therapy indicated by the VMTB system by their treating physicians, and 81% received at least one therapy on the list.

The VMTB platform also increased clinical-trial enrollment to 22%; the typical enrollment rate for patients with pancreatic cancer is 5%. Madhavan attributes this increase to the system's ability to highlight trials within a 2-hour drive of the patient.

"For our technical team, that was a pleasant surprise, because we're always focused on complex molecular data and how is that going to play in to the treatment planning, but [information on] simple things like geography...was extremely helpful," she said.

"Augmented Intelligence"

Machine learning is ideal for processing large, complex datasets, but Madhavan is cautious about calling it artificial intelligence. She stresses that her platform does not surrender treatment decisions to a computer.

"There's a lot of hype around artificial intelligence," she said. "The point that we're trying to make in the paper is that you need to integrate technology with human knowledge. So it's more augmented intelligence."

Madhavan also cautions that the small numbers involved are unusual for machine-learning technologies, which typically learn from millions of pieces of data. One way to address this is by tapping into large clinical networks with thousands of investigators and organizations, such as the Global Alliance for Genomics and Health, she said. The immediate next step for the Georgetown team is to scale the technology for use in more common cancers, such as breast and gastrointestinal cancers.

There are several barriers to translating learning systems such as this into real-world care, said Madhavan, including integration with electronic medical records and usability for physicians, who typically have only 15 minutes per patient. Cost is also a factor, especially in community hospitals, which may lack expertise in molecular medicine.

"While the technology provides those telemedicine-type opportunities, it always comes back to the expert humans in the loop," said Madhavan. "How do you make humans collaborate with technology? We're just starting to crack the surface of this ― how to make precision oncology a reality in clinical care."

The study was partly funded by the Lombardi Cancer Center, the Georgetown-Howard Universities Clinical and Translational Science Awards Program, and the Pancreatic Cancer Action Network's Know Your Tumor Program. Madhavan was on the scientific advisory board of Perthera Inc. Several coauthors are or were employees of Perthera.

J Am Med Inform Assoc. Published online October 7, 2019. Full text

Follow Medscape on Facebook, Twitter, Instagram, and YouTube


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.