On March 10, the van der Schaar Lab held a roundtable for clinicians on the topic of machine learning for early detection and diagnosis (ED&D).
During the roundtable, Mihaela van der Schaar asked the three clinician panelists to offer guidance and recommendations for members of the machine learning community who hope to develop solutions for healthcare alongside stakeholders in healthcare. The answers provided by our panelists were particularly inspiring and thought-provoking, so we have provided them below.
Our panelists for this session were:
- Alberto Vargas, MD (Chief Attending, Department of Radiology, Memorial Sloan Kettering Cancer Center)
- Prof. Henk van Weert (Professor, General Practice & Family Medicine, Amsterdam UMC; and Research programs in oncology and cardiovascular diseases)
- Prof. Stephen Friend (Visiting Professor of Connected Medicine, University of Oxford; Chairman and co-founder, Sage Bionetworks, 4YouandMe)
Alberto Vargas, MD
Chief Attending, Department of Radiology, Memorial Sloan Kettering Cancer Center
My general advice is to have an inquisitive mind and proactively seek at problems. Look for trouble! Attend multidisciplinary discussions. See where the actual problems are, and only then think of machine learning methods to solve those problems.
I often worry that if we do things just because we can, that leads to us doing things that may be excellent but may not solve a problem—or may attempt to solve a problem that doesn’t exist or isn’t clinically meaningful. This approach carries less possibility of having an impact in medicine.
Prof. Henk van Weert
Professor, General Practice & Family Medicine, Amsterdam UMC, and research programs in oncology and cardiovascular diseases
As a general practitioner, I think we and all clinicians very much aim to do something with the information we are given—to translate it directly to what we can use in healthcare for the patients.
If you want to be useful for clinicians, you have to produce outcomes that can be of direct use in the assessment of the risk that a patient has for a certain disease or outcome. You have to be able to translate the things you discover to helpful information.
If you find unexpected things, translate them! If it’s not translatable, you should ask yourself if it’s worthwhile, or if it should be combined with something else.
Prof. Stephen Friend
Visiting Professor of Connected Medicine, University of Oxford Chairman and co-founder, Sage Bionetworks, 4YouandMe
First, invest in learning the concepts and the nuances that are being used. If you are working with a clinician, it may take 2 months or a year to realize that you, as a machine learner, have a different understanding of words like “prediction” and “risk,” and different ways of thinking things. All too often, I see teams knit together and—despite often using the same words—they’re speaking different languages. Connotations and nuances are important.
As machine learning experts, you need to demand a seat at the table with regard to the design of cohort studies. Very often I see machine learners assume the clinician knows their world, and they come in afterwards and realize that the study won’t answer their question. Some clinicians know how to check on that, but if this doesn’t happen it can be very awkward to have to work around a cohort or dataset. Be sure to demand a role in looking at the design of the study.
Full recording of our March 10 Revolutionizing Healthcare session
Join our community!
If you’re a machine learning researcher and would like to share ideas about how machine learning can revolutionize healthcare, please sign up for our Inspiration Exchange engagement sessions.
If you’re a practicing clinician would like to be part of our ongoing discussions exploring the future of machine learning for healthcare, please sign up for our Revolutionizing Healthcare roundtable series.