If you have a patient with that gene mutation sitting in front of you, you need to be able to have the most current information to prescribe the most effective therapy. That’s the kind of thing Watson can do.
—Mark G. Kris, MD
While IBM’s Watson supercomputer may have defeated two former champions on the TV game show Jeopardy! 2 years ago, it is now facing its greatest challenge yet: deciphering huge amounts of scientific data and interpreting clinical information to help oncologists make personalized evidence-based treatment decisions for their patients—all in the blink of an eye.
Named after Thomas J. Watson, Sr, the founder of IBM, Watson is an artificial intelligence computer system that uses natural language processing software to understand written language. Using probabilistic algorithms, Watson reads and understands millions of pages of text to find information relevant to clinicians’ queries and then generates multiple possibilities they can use in making treatment decisions. Watson is considered knowledge-driven support, which means the knowledge exists but needs to be found and identified, as compared to data-driven decision support, which generates evidence used for decisions from structured data that is stored in places, such as electronic medical records or other databases.
“There are so many articles being published all the time, no one person can keep up with the volume,” said Martin S. Kohn, MD, Chief Medical Scientist, Care Delivery Systems at IBM Research-Watson. “The National Library of Medicine catalogued 700,000 new articles in 2010. Watson understands language well enough to infer the nature of a question posed, whether it is about making a diagnosis, choosing a medication, or doing a test or intervention, and then looking at all the literature and finding the appropriate information that is relevant to the decision the physician has to make.”
Oncology Training Underway
According to Dr. Kohn, when Watson appeared on Jeopardy! it was able to read and understand 200 million pages of text in 3 seconds. Now, Watson is undergoing basic training in oncology at Memorial Sloan-Kettering Cancer Center (MSKCC) in New York. The first applications for Watson are in lung and breast cancers, and other cancers will follow.
Memorial Sloan-Kettering is using its own de-identified historical records to “teach” Watson relevant data elements to consider, such as results from blood tests; pathology and imaging reports detailing type, size, and location of a patient’s tumor; the presence of genetic mutations; whether the cancer has metastasized; and other specific details about medical histories, including comorbidities.
Watson’s task then will be to offer physicians a list of treatment options, including clinical trials, based on the expertise of MSKCC physicians and evidence-based guidelines, as well as a confidence rating for each option. Initially, treatment options will focus on appropriate chemotherapeutics for a patient, and will not include suggestions on radiation therapy or surgery.
Testing for the lung cancer prototype should be finalized by midyear. Once testing is completed at MSKCC, beta versions of the system will be distributed to other physicians and hospitals with an initial commercial offering before the end of the year.
Computing with a Personal Touch
Leading the Watson health-care system development effort at MSKCC is Mark G. Kris, MD, Chief of the Thoracic Oncology Service. According to Dr. Kris, the clinical decision support tool will be useful to oncologists in several areas, including providing patients with the assurance that their treatment plan is based on the most up-to-date, relevant laboratory and clinical trial data published in medical journals and determined to be the most effective for their specific cancer type and clinical status.
“Watson can provide a sort of second-opinion for patients because their oncologists are making treatment decisions that are benchmarked against a very broad and expansive network that is looking at all the medical data,” said Dr. Kris. Watson can also customize treatment options based on information in a patient’s electronic health record. “Patients don’t want care that is good for anybody, they want care that is best for them. Watson will ultimately delve into a patient’s medical record and look for special characteristics and clinical nuances and benchmark them against other patients with similar records. It is counterintuitive to think that a computer, a machine, would actually facilitate more personalized care, but it can,” said Dr. Kris.
Another benefit, said Dr. Kris, is that Watson should also help speed up the health insurance preauthorization-approval process for treatments and test procedures, or even render it obsolete. “From the payer’s standpoint, if Watson says a particular treatment choice is best, then the whole preauthorization process disappears because the insurer has the assurance that the physician’s decisions are evidence-based,” noted Dr. Kris.
While knowledge gained in molecular biology and genetics over the last 2 decades have provided physicians with potential strategies for targeting specific molecular alterations in cancerous tumors, it has also added to the complexity of diagnosing and treating individual patients. With Watson, said Dr. Kris, oncologists and physicians who do not specialize in specific cancer subtypes will be able to easily access current information and integrate it into their patients’ treatment plans.
“Most non–small cell lung cancer patients have one genomic alteration, EGFR, that drives their cancer, and we know that EGFR is sensitive to erlotinib (Tarceva). But at ASCO’s Annual Meeting last June, an abstract was presented showing that if a patient has an EGFR mutation with exon 20 insertion, erlotinib won’t work and chemotherapy is a better choice.1 That is a very specific piece of information, and you can’t expect every oncologist to know it. But if you have a patient with that gene mutation sitting in front of you, you need to be able to have the most current information to prescribe the most effective therapy. That’s the kind of thing Watson can do. It can look for those very rare mutations and those exceptions to the rule and then match them up,” said Dr. Kris.
Although Watson’s initial development phase in cancer care is centered on making it an innovative and crucial decision-support tool for oncologists, as its functionality evolves, it may also become an important support tool for patients.
“One of the thrusts of the future of health care is the idea that all the people involved in the decision-making process—the physician, physician assistants, nurses, the patient, and the patient’s family—can benefit from this kind of support,” said Dr. Kohn. “In the future, collaborative access to Watson for the patient and the patient’s care team could enable the patient to ask questions and get information that could be shared with the patient’s medical team. That could really help move us along the direction of the empowered, knowledgeable patient who is an active participant in the decision-making for his or her health care.”
Because Watson would read information from patients’ medical records, it could also provide patients with an explanation of why a specific treatment is being recommended and details of potential side effects and how to manage those effects. Ideally, the information could then be printed or sent electronically to the patient’s mobile device, added Dr. Kris.
ASCO’s Rapid Learning System
Last November, ASCO unveiled its own rapid learning system, CancerLinQ. This system will have the ability to assemble and analyze data from the electronic health records of millions of cancer patients; provide real-time, standardized, clinical decision support integration within a demonstration electronic health record; demonstrate value-added tools, including the ability to measure a clinician’s performance against a subset of Quality Oncology Practice Initiative (QOPI) measures in real time; and create new ways of exploring clinical data and hypothesis generation. The testing of a breast cancer–specific prototype for CancerLinQ is underway. (See “Building CancerLinQ,” an interview with Clifford A. Hudis, MD, in this issue of The ASCO Post.)
“Systems like Watson and CancerLinQ provide the kind of technological competition that will accelerate advancements in cancer care,” said Dr. Kris. “However, these tools are physician advisors, they are not decision-makers. Our vision is that these systems provide information that the physician and the patient can use together to arrive at the best treatment, keeping in mind both the goals of care and the patient’s wishes.” ■
Disclosure: Dr. Kris currently holds full-time employment for an entity having an investment, licensing, or other commercial interest in the subject matter under consideration.
1. Costa DB, Yasuda H, Sng NJ, et al: Sensitivity to EGFR inhibitors based on location of EGFR exon 20 insertion mutations within the tyrosine kinase domain of EGFR. 2012 ASCO Annual Meeting. Abstract 7523. Presented June 5, 2012.