Using IBM Watson Electronic System to Guide Patient Management in Oncology
Los Angeles, CA—IBM Watson computer is changing the face of oncology practice through its massive parallel computing system, utilizing natural language processing and cognitive machine computing. Unlike HAL 9000 from the movie 2001: A Space Odyssey, the IBM Watson system does not interpret the results but rather builds on information it receives from previous operations and offers possible solutions to new queries, said Mark G. Kris, MD, William and Joy Ruane Chair, Thoracic Oncology, Memorial Sloan Kettering Cancer Center, New York, at the Fourth Annual Conference of the Association for Value-Based Cancer Care. Dr Kris, lead physician of the Memorial Sloan Kettering Cancer Center–IBM Watson collaboration, discussed how the use of this system is transforming cancer care.
Exponentially growing data (ie, big data) that are increasing in specificity and complexity must be interpreted by oncologists, including medical records, test results, and information from medical publications. Dr Kris provided an example of next-generation sequencing, or integrated mutation profiling of actionable cancer targets, in which probes for 341 genes are examined.
“We do it in all the cancers with a good target; KRAS in colorectal, BRAF in melanoma, etc,” he said. “It’s very important not just to look for a mutation or an amplification, but to look for all those things.”
IBM Watson provides cognitive computing to build possible solutions from previous applications. “It learns, not like a human learns, but approaching that,” said
Dr Kris. “In the course of performing operations and giving answers and solutions, that information informs the operating system the next time that same question is asked or a similar question is asked.”
In addition, IBM Watson uses natural language processing and cues from information provided to obtain possible solutions.
Memorial Sloan Kettering Cancer Center standards and “clinical wisdom” are integrated with up-to-date medical evidence to inform the care of the individual patient. For providers and patients, IBM Watson would be used to facilitate and accelerate research, test new treatments, and measure outcomes.
“We are today trying to aid physicians and patients in making the best choices at the first treatment selection for metastatic solid tumors,” said Dr Kris.
Watson Guiding Treatment Decisions
He presented a Watson Oncology demonstration that analyzes the data to which it has access and prioritizes it when presented to the physician. The system suggests test options based on the information that is available, such as the type and extent of disease evaluation and any pretreatment assessments. In the case of a nonsmoking female Asian patient, the information presented at the top of the web page showed that cancer was found after a biopsy of an adrenal gland. The case information deprioritized the importance of a magnetic resonance imaging (MRI) scan that the patient had after a knee injury 3 years ago.
If a molecular pathology report were available, an MRI scan of the brain would be recommended. “It knows what it needs to make medical decisions, so it looks for those things that are most critical to its making a decision and it puts it on top,” Dr Kris said.
An evidence button would provide access to the medical evidence that supports the decision to perform a particular test (ie, a molecular test) based on the literature, expert panels, guidelines, and medical literature.
Treatment options with evidence and confidence levels of these therapies are provided. Patient preferences are considered (ie, the patient’s lifestyles and desires in the treatment of the illness). Clinical trials relevant to the specific patient are also provided (Figure).
In the example provided, the patient had EGFR-mutation stage IV lung cancer. Guidelines from the National Comprehensive Cancer Network recommend a regimen containing erlotinib (Tarceva), but this therapy is at the bottom of the list according to the Watson demonstration. In this patient, an exon 20 insertion is not a sensitizing mutation. Therefore, the best treatment plan would be a standard 3-drug therapy that includes cisplatin (Platinol), pemetrexed (Alimta), and bevacizumab (Avastin).
IBM Watson evaluated all the available information against current therapy. “I think it’s tough enough to know the EGFR mutation has mandated therapy, but which EGFR mutations?” asked Dr Kris. “Taking this further, there are 10 or so exon 20 insertions. There is 1 of them that probably is still sensitized. You need to know which 1 of the 8 of these rare mutations is the right one.”
As he explained, “we try to train this computer as we train doctors, experientially.”