Working with Watson, IBM’s Jeopardy!-winning computer, oncologists are developing new supercomputing tools for treatment decision support, reported Andrew D. Seidman, MD, senior breast medical oncologist of the Memorial Sloan Kettering Cancer Center (MSKCC)/IBM Watson Oncology Collaboration, and professor of medicine at Weill Cornell Medical College in New York. Dr. Seidman spoke at the 33rd Annual Miami Breast Cancer Conference, held March 10–13 in Miami Beach, Florida.
“Exponentially-growing data and information” represent a central challenge for oncologists’ approach to clinical decision-making, Dr. Seidman said.
Computers are helping tackle that challenge, he noted, with clinical decision-support tools, clinical electronic records, and “back end” coordination of test ordering and insurance authorization, and meaningful-use legislation requires that health IT capture and communicate information for care coordination, to improve quality, safety, and efficiency outcomes—and to provide decision support for high-priority conditions like cancer, he noted.
“There’s growing interest in oncology decision-support tools like ASCO’s CancerLinQ,” he noted. IBM Watson is working with MSKCC and MD Anderson Cancer Center to focus high-power computing on the development of truly sophisticated decision support.
At MSKCC, integrated mutation profiling of actionable cancer targets involves assaying 450 cancer genes, yielding very complex test results reports. These reports and multiple other factors determine choice of treatment for patients with metastatic breast cancer, Dr. Seidman noted.
But guidelines tend to be “vague,” and available options numerous, he said. In the setting of a disease as heterogeneous as breast cancer, disease and patient characteristics must be taken into account and balanced, he said.
“This is where science meets art,” Dr. Seidman said: considering tumor biology and aggressiveness, prior treatments, treatment feasibilities, patient symptoms, comorbidities, and preferences, to “individualize therapy” for metastatic breast cancer.
“Treatment algorithms for metastatic breast cancer take one just so far,” he said. Combination-therapy options further complicate the decisional landscape.
“The therapeutic index of cytotoxic chemotherapy is dynamic, not static,” he explained.
IBM Watson is a massive parallel computing system that uses “cognitive computing” machine learning and natural language processing.
Watson was famously able to beat a human champion on the game show Jeopardy!, Dr. Seidman reminded attendees.
The goals for the Watson Oncology collaboration is to “integrate standards and clinical ‘wisdom’ with latest medical evidence to assist in the care of individual patients, and to facilitate and accelerate research, help test new treatments, and measure outcomes,” Dr. Seidman said.
“We approach training the computer as we train doctors—experientially, in an apprenticeship model,” Dr. Seidman said. “The average oncology specialist sees 200 to 400 new patients during subspecialty training. With case-driven learning, supplemented by medical literature and expert curation, we can follow through actual patient histories and link those to clinical decisions made.”
Watson uses these data to learn “through inference,” identifying key case attributes “and converting them into structured data” for decision-support recommendations. It can help aggregate patient data, build evidence-based radiation and surgical components of patient care plans, and incorporate new treatments and advances in care, while collecting clinicians’ treatment choices and linking those to patients’ outcomes for further machine learning, Dr. Seidman reported.
Interpreting and integrating medical record data has proven to be more of a challenge for natural language processing than parsing relatively simple quiz-show questions, however.
“One of the things we’ve learned along the way, is that you don’t realize how hard this really is,” he said. “Treating cancer is a whole lot harder than winning at Jeopardy!”