GRACE :: Cancer Basics
Dr West

Will your next oncologist be a computer algorithm?

Share
download as a pdf file Download PDF of this page

Last year, an IBM computer called Watson not only beat but rather decisively spanked the best human competitors on the TV quiz show Jeopardy!, but that was really primarily a high profile way to highlight the high level artificial intelligence capabilities of the most advanced artificial intelligence algorithms at understanding sophisticated questions in a conversational tone (or even deliberately convoluted Jeopardy! style). 

But the real potential is for advanced artificial intelligence platforms to enhance or replace some of our data-heavy thought processes done by humans now, with health care among the higher priorities.  A “Dr. Watson” project is underway trying to create algorithms that can mimic the thought processes of the best diagnosticians and specialists, and Memorial Sloan-Kettering has been working closely with IBM on cancer workup and management algorithms.

I’m excited to think of the many ways that technology can improve healthcare, but I’m skeptical that computer algorithms will replace the judgment of the best doctors.  Here’s my video discussing the topic:

I’m interested in your thoughts as well. Is the best a human oncologist can hope for that a computer may not be able to convey the human connection that is also (ideally) part of the patient/physician relationship? Would you feel far better knowing your doctor was working with the support of a Dr. Watson system helping to guide decisions? Or would you be concerned that any “art of medicine” would be lost with the greater reliance on defined rules?


9 Responses to Will your next oncologist be a computer algorithm?

  • double trouble says:

    The first thing I thought about listening to this webinar was my primary care doc, who sits typing into the computer constantly as I’m telling her about my latest symptoms. She relies on software to aid her in deciding the best approach to my care.

    Also, the pulmonary doctor who has been helping me get my coughing spells under control, which he calls bronchorrhea, has repeatedly pulled up case studies as I’m sitting there, and we discuss the different approaches we could be taking.

    I appreciate a physician who is humble enough to research his or her own inclinations, to check for possible different, possibly newer ways to manage my care. My case has never been a textbook example, so I think the combination of intuition and experience that my doctors bring to the table, together with what can be accessed online, result in more complete care.

    I think rather than dictating the course of care to the physician, these “algorithms” should serve as an additional tool in his or her arsenal, assisting in the decision making process. I would worry about a plan that was based more upon standard of care than on instinct and experience. The idea of being treated by or according to an “algorithm” does not appeal to me. I am a firm believer in the value of the “gut feeling.”

    Debra

  • michaelholmes says:

    Hi Dr. West. This is a great post and video. Thank you very much. I am Michael Holmes, program director of IBM Watson solutions group. We are very much in agreement with you that neither Watson nor any other technology will replace medical professionals. Watson is intended to help medical professionals make more informed, evidence-based decisions. We are working with Memorial Sloan-Kettering Cancer Center now to train Watson to help oncologists in their diagnosis and treatment of cancer patients. While we are in very early stages of what will certainly be a long journey, we have no intention of ever attempting to replace doctors or other medical professionals. Watson is intended to be a resource to assist them, not replace them. Thanks again for the great post.

  • Dr West
    Dr West says:

    Mr. Holmes,

    First, let me say that I very much appreciate your kind comments. As reflected in my comments following the video, discussion of Watson as a replacement for human doctors isn’t necessarily plan that you or anyone from IBM has identified, though of course you’re aware of the contentious statements from Vinod Khosla and some other thoughtful people that has suggested that most doctors could/should be replaced by algorithms:

    http://techcrunch.com/2012/01/10/doctors-or-algorithms/

    Second, Watson and the work you’re doing are marvelous. That we could have a serious debate about whether algorithms should replace some doctors is in itself a reflection of the wonder of how far your work has brought us.

    My leading concern is really the issue that algorithms are only as good as the data and decisions used to shape them. I don’t mean to suggest that you don’t appreciate these issues, but some certainly see medicine as having stark right and wrong answers, as a Jeopardy! question would. While some of medicine has pretty clear, evidence-based answers, it’s humbling to really know how much remains unknown. I don’t like to invoke the “art of medicine” argument every time someone wants to do any hare-brained idea that runs counter to evidence or good judgment, but the reality is that a huge amount of real clinical practice involves adapting the very limited information from evidence-based work on more idealized cases to the patients who have extenuating circumstances and would be ineligible for these studies for one reason or another. You can have a panel of 5 experts learn about a complex case and offer at least 5 different opinions about the best way to proceed, all with merit and acknowledged respectfully by the other panelists as a sensible alternative. I am just wary about Watson being programmed to direct people to a presumed gold standard approach that is really reflective more of the way Mark Kris treats lung cancer than the “best way to treat lung cancer” as if handed down by divine intervention. As I note in the video, this isn’t a slight against Mark Kris, except to the extent that I don’t think that he, as just one knowledgeable person, lays claim to the right answer any more than any one other knowledgeable person. The experts all work from the same limited collection of data, which we then inviolably infuse with our own biases and judgments.

    As I note in the video, the sheer volume of information emerging in medicine has become untenable, and this has become a bottleneck to the best care for patients. Docs generally can’t keep up with the torrent of new information, so that’s where algorithms can be astonishingly helpful — tracking and integrating the literally unfathomable amount of new medical information coming out. And perhaps algorithms can be used to help us reduce our biases and poor estimates: presumably, algorithms could help us overcome issues like our fear of treatment-related toxicities outweighing concern for cancer-related problems, or faith that a new molecular test is necessary in the absence of evidence to support that claim.

    I am confident that we can agree that there are many places in which Watson could be remarkably helpful in medicine, even as we can probably also agree that algorithms shouldn’t replace human doctors. I wonder if IBM and/or you personally have a clear sense of what you’d see as the ideal way in which this work will/should be integrated into medical care in general, and cancer care in particular. Should patients and caregivers have open access to this tool, or is Watson to be for doctors and other health professionals? If really open access, what role would physicians then play? I don’t mean to sound alarmist, but you can’t keep this genie in a bottle, and having a tool as powerful as Dr. Watson for public use would be as disruptive to the medical world as Google and wikipedia have been to Encyclopedia Brittanica, and as Khan Academy and Coursera will likely prove to be in education.

    Above all, thanks for engaging. I’d welcome any additional thoughtful input you can share.

    -Jack West

  • Dr West
    Dr West says:

    Sorry, some additional comments in a good discussion are not above, as I had indicated, but through this link:

    http://cancergrace.org/topic/could-your-next-oncologist-be-a-computer-dr-watson

  • michaelholmes says:

    Dr. West – Thank you for this additional round of equally insightful comments. I think you will find that once again we are in strong agreement. First, yes, Watson, other technology, and indeed, doctors themselves, are only as good as the information they use. All have the same information available to them. But in many cases, Watson evaluates and actually uses far more of this information when considering a patient’s case (the Jeopardy! configuration of Watson analyzed the equivalent of about 200M pages of information in three seconds). As medical fields evolve and new information is added to the body of knowledge, Watson can evolve with the field. By one estimate, it would take some physicians over 144 hours per week to read the necessary journal articles to stay current. (source = Journal of Medical Library Association). Second – I agree completely that in medicine (like many other fields) there is very rarely a single ‘right’ answer. Even seemingly obvious situations like “what is 2+2″, the answer is usually “4” but the question could refer to a car configuration (2 front seats, two back seats), a family unit (2 parents and 2 kids), or even a reference to George Orwell’s Animal Farm. That’s why Watson does not provide “answers”; it provides a confidence-scored panel of responses… and it transparently displays the sources of information used in presenting each response. Furthermore, users can choose to dig deeper into each source to evaluate its validity…AND they can choose to remove individual sources from contributing to a given response. So if a user feels that Dr. Kris and his colleagues have had too much influence in one of Watson’s suggestions, the user could remove the “MSK best practices” reference from being used in a treatment suggestion. Transparency is a primary pillar of Watson’s design. It is meant to help professionals assemble the information they need to make a decision and take action…. it is not meant to make decisions or take actions itself. And finally, per your Q about direct access by individuals, that is not something IBM is pursuing for this use case. Other use cases we’re considering (ie – use in consumer goods contact centers) are more likely to be appropriate for access by individuals.

  • Dr West
    Dr West says:

    Thanks very much for the clarification, which all makes good sense.

    Congratulations on your great successes thus far. I very much look forward to seeing and experiencing how your work evolves. It clearly has the potential to dramatically improve the quality of our care and understanding in medicine. And even as I recognize that it may also prove to be disruptive, I don’t consider that a bad thing.

  • certain spring says:

    Interesting discussion. I appreciate the idea of a confidence rating for the information sources, but I wonder how that’s constructed – how for example do you factor in various kinds of bias (eg selection bias in clinical trials, company-funded trials)?
    I am in two minds about the information retrieval dimension. On the one hand it would be fantastic to have a more intelligent filter than a common-or-garden search engine. On the other, conservative doctors will remain conservative (“We don’t do this”), and will remain more influenced by colleagues they know than by experts from other continents. I also think there is a danger of cultural imperialism, if I may use that term. I sometimes feel that “standard of care” translates as “What we do in the US”, and would be concerned that the responses Watson provides would take no account of countries that don’t have certain drugs and technologies available.

  • michaelholmes says:

    Good points. A couple points of clarification…. Watson displays confidence levels in each of its suggestions in its response panel… not confidence in individual sources of evidence. That said, it weights its contributing sources in many ways in coming up with its suggestions (ie – how recent is the evidence, how many sources of corroborating evidence can it find, how reliable is the publication in which it appears, etc.). Second point – yes we agree that adoption of Watson as well as adoption of the concept of evidence-based medicine in general is not automatic and faces barriers. Third point – our team recognizes that resources such as medical facilities and access to methods of treatment are not universal. This does not mean that the practicing medical professional would not benefit from being made aware of such treatments and the evidence available to support their use. Again, Watson does not present THE answer but rather a panel of possible courses of action based on the body of knowledge available to it. Medical professionals can evaluate these responses and the evidence behind them and then choose what direction to go on their own whether it’s the top-rated option, the bottom rated option, or something that Watson didn’t even suggest. Watson is a resource for the professional and does not make decisions or take actions.

  • certain spring says:

    Thank you for responding. I think that is what would reassure me most – the idea that it would not substitute for clinical decision-making, but rather expand clinical horizons.

Leave a Reply

Ask Us, Q&A
Cancer Basics Expert Content

Archives

Share
download as a pdf file Download PDF of this page

GRACE Cancer Video Library - Lung Cancer Videos

 

2015_Immunotherapy_Forum_Videos

 

2015 Acquired Resistance in Lung Cancer Patient Forum Videos

Share
download as a pdf file Download PDF of this page

Join the GRACE Faculty

Lung/Thoracic Cancer Blog
Breast Cancer Blog
Pancreatic Cancer Blog
Bladder Cancer Blog
Head/Neck Cancer Blog
Kidney Cancer Blog
Share
download as a pdf file Download PDF of this page

Subscribe to the GRACEcast Podcast on iTunes

Share
download as a pdf file Download PDF of this page

Email Newsletter icon, E-mail Newsletter icon, Email List icon, E-mail List icon

Subscribe to
GRACE Notes
   (Free Newsletter)

Other Resources

Share
download as a pdf file Download PDF of this page

ClinicalTrials.gov


Biomedical Learning Institute

peerview_institute_logo_243