As a true believer in Freedman’s shoe leather, I share your reservations about all such administrative ‘convenience data’ @ESMD and @f2harrell. Still, I wonder whether these claims databases might support reasonable recognition of episodes of care, and basic analyses of these.
For example, could one identify clear signatures of empiric antibiotic therapy for uncomplicated UTI, and several characteristic alternative branching treatment courses such as (a) clinical resolution, (b) test of cure, (c) change of antibiotic in response to culture sensitivities, (d) hospital admission, etc? Identifying such categorical events would be at the heart of any valid attempt to replicate this EPPC ‘study’.
My impression of the usual practice in the analysis of convenience data is that such categories would typically be defined presumptively ex ante (in a study protocol, say), and then never revisited or revised in response to the actual structure and content of the data. Would anyone here be aware of such efforts in claims databases, where honest-to-goodness shoe leather got spent?