Adjustment for ‘Prescriber Type’ in Pharmacoepidemiological Analyses

Abstract

The type of prescriber typically fulfils the criteria for confounding, as it is associated both with the exposure (e.g., prescriber types may differ in their choice of first-line treatment) and with the outcome (as different types of prescribers often treat patients with different disease severity). Additionally, the type of prescriber may correlate with other factors such as treatment adherence, surveillance or coding practices. Although information on the type of prescriber is often available in healthcare registries, it is very rarely employed to control for confounding in pharmacoepidemiological analyses. Here, we argue the potential value in adjusting for the prescriber type in pharmacoepidemiological studies. In an applied example, we conducted a cohort study using Danish healthcare registers of the risk of ischemic stroke associated with the use of direct oral anticoagulants (DOACs) compared to warfarin. We found a hazard ratio (HR) of 0.95 (95% CI: 0.90–1.01) for DOACs versus warfarin when adjusting only for age and sex. Further adjustment for prescriber type showed an effect of similar magnitude (HR 0.92; 95% CI: 0.87–0.98). However, interaction testing and stratified analyses confirmed prescriber type as an effect modifier. Future studies are needed to clarify the role of adjusting for prescriber type across other use cases and healthcare settings.