The @TheLancet hydroxychloroquine paper attracting attention. Getting insights from #EHR data take work. Would be great to see the code. And which one of the authors did the coding? And how were non-standardized inputs across disparate venues reconciled & validated? Imp to know.
@TheLancet Top journals are publishing large #EHR-based studies, but important to know that you cannot assume accuracy across non-standardized inputs. So critical to have the transparency of what transpired. Look to @OHDSI as an example of that transparency. @suchard_group
@TheLancet @OHDSI @suchard_group Also, @nejm paper leveraged heavily the problem list in the #EHR… but lots of studies have shown that the problem list is highly inaccurate. Not enough to say that clinicians are encouraged to keep it up to date. We need more sophisticated ways of phenotypic from the #EHR.
@TheLancet @OHDSI @suchard_group @NEJM Digital means of doing research will surely improve the pace and efficiency of research, but the methods matter. And so does transparency of what was done. Not enough to assume accuracy. Especially in these early days, must be evidence of validation and the strength of methods.
@TheLancet @OHDSI @suchard_group @NEJM Just because you have data and can run a stats program doesn’t mean you have handled the inputs well. What I have learned is that the quality of real-world evidence generation depends importantly on the pre-processing and high-quality mapping. Do you know exactly what you have?
@TheLancet @OHDSI @suchard_group @NEJM My thanks to @wade_schulz for educating me about this line of research - & importance of investing time in being sure your data are fit for purpose; and that you have done the work to map the data appropriately, prior to the analysis. And are able to show what you did. @YaleMed
