Fwd: IMPROVEDX: approval required (8C705E69)
David L Meyers
dm0015 at COMCAST.NET
Thu Sep 13 21:14:28 UTC 2018
Below is a submission to the listserv by Jason Maude, the Founder and CEO of Isabel Healthcare which produces a clinical decision support tool for clinicians and a "symptom checker" for use by patients.
David L Meyers, MD FACEP
Listserv Moderator/Board member
Society to Improve Diagnosis in Medicine
Save the Dates: Diagnostic Error in Medicine, November 4-6, 2018; New Orleans, LA
AusDEM2019, April 28-30, 2019; Melbourne, Australia
> Begin forwarded message:
> This is a very useful article which highlights the key point which is that systems where the rules and probabilities are determined by experts are very difficult to scale and able to cope with the complexities of medicine. To illustrate why I have attached a paper (freely available) describing how a model to help diagnosis just one condition – autism – was built. Note the practicalities of getting expert clinicians together to decide on the probabilities of each relationship. Often, they don’t agree. Why do they decide on 0.1 instead of 0.2 for example? The problems are compounded, and errors amplified when the probabilities are combined. Just for this one condition the final version of the model had over 1,500 dependent probabilities! Once done the model then has to be maintained.
> Now imagine the task of building and maintaining a model like this to cover 10,000 diseases!
> Machine learning enables these probabilities to be calculated and maintained automatically from the training data. Yes, it not as transparent but at least its practical for the intended purpose. We call Isabel an enhanced deep learning tool as it combines both machine learning, rules and statistical NLP. From bitter experience we found that machine learning can do a lot of the heavy lifting but does still not have the precision necessary for DDx Generation.
> IBM’s problem was promising far too much too early - it let its PR department run ahead of reality. It’s very important to set expectations with these tools. DDx generators are called that because they are only intended to produce a list for the clinician to consider. They are a tool to augment the clinician’s memory and should not be expected to do any more. For this purpose, they are proving to be useful tools.
> They will never be called a final diagnosis generator because this is not possible or credible.
> Jason Maude
> Founder and CEO Isabel Healthcare
> Tel: +44 1428 644886
> Tel: +1 703 879 1890
> www.isabelhealthcare.com <http://www.isabelhealthcare.com/>
Moderator: David Meyers, Board Member, Society to Improve Diagnosis in Medicine
More information about the Test