Watson for Oncology
Watson for Oncology is a decision support tool that is trained by top oncologists at Memorial Sloan Kettering. Watson for Oncology ranks the treatment options, linking to peer reviewed studies that have been curated by MSK. Watson for Oncology also provides a large corpus of medical literature for a physician to consider, drawing on more than 300 medical journals, more than 200 textbooks, and nearly 15 million pages of text to provide insights about different treatment options.
Watson for Oncology now allows client institutions to add localized treatments and dosing to their system, and to identify treatments that are unavailable in their geography. Watson for Oncology intelligently finds the most appropriate articles from PubMed and medical journals using patient cohort analysis, drug class matching, and document quality models. These are matched to the patient and treatment options in seconds.
The Annals of Oncology published a full study led by oncology leaders at Manipal Hospitals in India. Their tumor board found Watson for Oncology was concordant with their own tumor board’s treatment decisions in 93% of breast cancer cases. And it was interesting in this study that when the blinded study initially compared Watson’s recommendations against retrospective treatment decisions from 2-3 years prior, the concordance was only 73%. Manipal’s tumor board re-reviewed the same patient cases manually, and the concordance rose to 93%. This is an indication that the corpus of data on which Watson for Oncology relies is staying up-to-date with the latest science. Once treatment options are provided by Watson for Oncology, we provide the following in support of the options; supportive care guidelines, drug information, Truven cost data, literature references, and dosing details.
At ASCO 2018, physicians from the Affiliated Hospital of Academy of Military Medical Sciences in China in reported a study (n=1,997), which found that disclosure of Watson for Oncology recommendations resulted in treatment decision changes in 5% of cases. Attending physicians were less likely to alter their decisions (3%) than chief physicians (6%; p < 0.001) and fellows (7%; p < 0.001). Importantly, in those 106 cases where clinicians changed their treatment decision based on Watson for Oncology, adherence to professional treatment guidelines improved from from 89% to 97% (p < 0.01).