Thread by @StefanKertesz

1/Listening to @1a "Against the Pain" - @1a is to be commented for covering this issue Hats off.

I wish to share a thought on a form of bias built into NARXCare as it is obvious, correctable, and harms patients.
wamu.org/story/21/11/11…

2/Dr. Nishi Rawat of Bamboo Health (provider of the NARXCare algorithm to assess overdose risk) describes the NARXCare product they put onto the screens of doctors (via Prescription Drug Monitoring Programs) as an "objective summary". About that...

3/Dr. Rawat's definition of NARXCare covers 2 distinct function. One is a graphic summary of the Rx's received by a patient (if the data are accurate, this is helpful).
But the concerning one is an "Overdose Risk Score"

4/The NARXCare score asserts that it communicate risk of overdose. Sadly it is subject to a built-in form of bias against
*patients who use teaching hospitals,
*against Veterans, and
*against persons of color.
It could have been corrected by the way.

5/The bias I get to see daily is NARXCare counts the "# of prescribers" toward OD risk. All things equal, there is an actual association. But the # of prescribers is often high for patients getting care in teaching hospitals (including Veterans hospitals)

6/In a clinical system where academic or non-academic providers cover each others' patients prescriptions, under strict protocol, that counts the patient as having "more prescribers". But that "more" is very different from "more" due to doctor-shopping, for example.

7/NARXCare will designate all such patients as higher risk even when, they may well be uniquely protected from OD. This misclassification is not 'random' but systematic.
It will declare higher risk for patients in these institutions, just because they sought care there.

8/What kinds of patients differentially go to teaching institutions?
People with more severe illness:

Veterans (*Happy Veterans Day*)
To a degree, African-Americans
All these people will be misclassified systematically in ways prejudicial to their care, by NARXCare

9/I want to underscore that this misclassification problem could be addressed by careful analytic and database improvement by the people who developed NARXCare, at any time.
Often the prescribers in one teaching institution share an address, which is known and on screen.

10/To my knowledge, as a regular user of our state's PDMP with NarxCare, I do not see evidence of programming, to date, to address such a problem. If Bamboo wishes to clarify, of course, I will post the clarification their behalf.

11/Based on what I see, I don't think "objective" fairly classifies algorithms that seem to systematically mis-classify risk of overdose based on disability, Veteran status, residence in an urban area, or being African-American. Again, I invite Bamboo to clarify, and I'll post it

12/None of this is to discourage active inclusion of all voices in discussion of what is happening for patients with long-term pain right now. I think it is simply amazing that this episode happened and I thank @1a @ibdgirl76 @maiasz and even Bamboo./fin

13:/PS - I have listened to the account by Dr R that the algorithm can avoid penalizing patients for team based care. I re-reviewed my patients improperly designated as “multiple prescribers”. — continue-

14/I can’t reconcile her statements with each other, with standard analytic approaches to large pharmacy databases, or with the operational definition of multiple prescribers declared explicitly on the Bamboo/Appriss screen I see in our PDMP. That said

15/If there is an unspecified method used to achieve the objective claimed on @1a , it would be a unique accomplishment, but it would also be one that is not consistent with the account of methodology provided on the Narxcare screen

16/I want to vouch that further review or our care panel, and hearing from individual patients I know continues to suggest that the “multiple prescriber designation” without reference to collaborative or team-based care by Narxcare, contrary to what was said on @1a yesterday