IBM Watson for Oncology platform highly concordant with physician recommendations
Click Here to Manage Email Alerts
SAN ANTONIO — The Watson for Oncology artificial intelligence platform appeared highly concordant with breast cancer treatment recommendations of a multidisciplinary tumor board, according to results of a double blind validation study presented at San Antonio Breast Cancer Symposium.
Watson for Oncology, developed by IBM in collaboration with Memorial Sloan Kettering Cancer Center, achieved a higher rate of concordance with physician recommendations for nonmetastatic disease than metastatic cases. Concordance also was higher in triple-negative breast cancers than HER-2–negative cases.
Clinical factors account for about 10% of data available on which oncologists can base their decisions, whereas exogenous and genomic factors account for the other 90%, according to S.P. Somashekhar, MBBS, MS, MCH, FRCS, chairman of Manipal Comprehensive Cancer Center in Bengaluru, India.
The Watson computing system — which Manipal Hospitals adopted to help oncologists make high-quality, evidenced-based decisions — understands natural language and human communication. It can extract and evaluate vast quantities of data — including medical records and clinical trial evidence — to generate evidence-based hypotheses, and it has demonstrated potential to make treatment recommendations multiple cancer types.
Somashekhar and colleagues examined how Watson’s recommendations compared with those of Manipal Hospitals’ 12- to 15-member multidisciplinary tumor board.
“We wanted to know where this fits, and how useful this artificial intelligence is for physicians,” Somashekhar said during a press conference.
Somashekhar and colleagues reviewed cases of 638 patients with breast cancer who had been treated within the past 3 years at Manipal Hospitals. Two groups of oncologists who were not privy to the tumor board’s deliberations entered data from the cases into the Watson for Oncology system.
Researchers then analyzed Watson’s concordance with the tumor board recommendations. They also calculated the time required for Watson and the tumor board to reach their conclusions.
Watson for Oncology’s responses were classified into three categories: recommended therapy, for consideration or not recommended. Ninety percent of Watson’s “recommend standard treatment” or “for consideration” recommendations were concordant with the tumor board’s recommendations, Somashekhar said.
Overall, researchers reported concordance rates of 79% for nonmetastatic disease, 46% for cases of metastatic disease, 68% for triple-negative breast cancer cases and 35% for HER-2–negative cases.
The fact fewer treatment options exist for triple-negative breast cancer than HER-2–negative disease contributed to the discordance between those cases, Somashekhar said.
“Including HER-2–negative cases opens up many more treatments and variables for consideration,” he said in a press release. “This increases the demands on human thinking capacity. More complicated cases lead to more divergent opinions on the recommended treatment.”
The study only was designed to assess concordance.
“The scope was not to determine who was more right or less right,” Somashekhar said during the press conference. “We did not determine superiority or inferiority.”
Researchers determined the tumor board initially required an average of 20 minutes to make its recommendations, but that declined to about 12 minutes once board members became more familiar with the cases. The Watson platform required a median 40 seconds to analyze data and deliver a treatment recommendation.
Artificial intelligence can contribute to personalized medicine but should be considered a complement to — rather than a replacement of — physician care, Somashekhar said. That is because a variety of personal factors contribute to the distinction between what can be done and what should be done, he said.
“Watson for Oncology is a promising cognitive computing tool that warrants further evaluation in a variety of clinical settings and a variety of study designs,” he said. “The role of Watson for Oncology will always be consultative. [It] cannot replace human clinical judgment and the essential patient–doctor relationship.” – by Mark Leiser
Reference:
Somashekhar SP, et al. Abstract S6-07. Presented at: San Antonio Breast Cancer Symposium; Dec. 6-10, 2016; San Antonio.
Disclosure: This investigative study received no external funding. Somashekhar reports no relevant financial disclosures. Please see the abstract for a list of all other researchers’ relevant financial disclosures.