Just over a year ago, Alex Trebek introduced the world to Watson. Borne out of the DeepQA project at IBM, Watson made international headlines after handily defeating “Jeopardy!” legends Ken Jennings and Brad Rutter. While the whole episode was reminiscent of the Deep Blue/Gary Kasparov match-up more than a decade prior, Watson in fact represents something quite different.
Satish Misra, MD, Managing Editor, iMedicalApps Watson was being designed to play in the real world. And soon after the “Jeopardy!” match, IBM and Nuance Communications announced that Watson's first venture would be into healthcare. Through a partnership with the Samuel Oschin Comprehensive Cancer Center at Cedars-Sinai Medical Center, oncology would be one of the first fields Watson would tackle.
Dr. Harlan Levine of Wellpoint, an early IBM partner in this endeavor, explained the rationale to Forbes magazine:
Oncology is a very complex field. The amount of information in medicine in general doubles every 5 years. In oncology, the amount of new information emerging is particularly prolific. It’s a hard area in which to keep up to date.
Many physicians will be familiar with Nuance's medical dictation software. The technology under the hood is a pretty powerful natural language processing and clinical language understanding package. For Watson, what that means, in part, is the ability to understand the world of medicine as it delves into textbooks, clinical trials, case-reports, panel guidelines, and so on.
So what might this look like? You've just finished interviewing and examining a patient who presents with a new diagnosis of some malignancy. After walking out of the room, you pull out your mobile device (let’s call it the iPad 7), describe the clinical situation, and pose a question. Perhaps the question is which neoadjuvant therapy would give this patient the best shot at clinical cure when all is said and done. Watson would then read and synthesize every textbook, clinical trial, retrospective series, and case report, as well as that patient's history, labs, imaging, and other personal data. From the collective knowledge of the medical world, it would deliver an answer.
IBM insists that Watson is a decision-support tool, meant to assist and not supplant physicians. Others feel differently and argue that certain areas of medicine are ripe for takeover. At iMedicalApps, we wrote a piece making the argument that Watson would not replace physicians.
For now, I'll focus on a fairly narrow question: how Watson might impact how oncologists pick chemotherapy regimens. In some ways, it’s a decision based on a fairly limited number of patient factors, such as staging, patient comorbidities, functional status, and so on. There is, nonetheless, an incredible amount of data that can be applied to the decision.
The recommendations that Watson can make are only as good as the data that it bases those decisions on. The publication of any trial is often followed by critiques of study methodology, inclusion/exclusion criteria, and so on. Large repositories where data are input by hundreds or thousands of remote users, can have problems with data accuracy. Arguably, these problems can be overcome by simply “teaching” Watson how to weight different data sources or by looking at a billion instead of a million data points; it certainly won't be a trivial challenge though.
And then for every challenge that is accounted for, there is one that is not anticipated. For proof of that, one need look no further than Watson's response of “What is Toronto?????” in Final Jeopardy—to a clue in the category “US Cities” (“its largest airport is named for a WWII hero, its second largest for a WWII battle”). So errors will be inevitable. Let’s say Watson is wrong. And by wrong, I don't mean as a result of clinical uncertainty—I mean the kind of wrong that a first-year medical student would catch. Perhaps that is part of why IBM engineers insist that Watson will help inform, rather than make, clinical decisions.
Finally, there is the issue of how the medical community will critically evaluate Watson's methodology. A lot of Watson’s inner workings are undoubtedly proprietary and secret, meaning that the details of how it goes from question to answer are unlikely to be open to scrutiny by the medical community. In some ways, this runs counter to one of the most basic tenets of the scientific community.
Advanced decision-support tools like Watson will undoubtedly reshape the practice of medicine in the future. Projects like those underway at Cedars-Sinai, Columbia University, and the University of Maryland will be instrumental to figuring out how to apply this technology in medicine.
