Joint Modeling of Primary and Secondary Non-Adherence Outcomes.

Abstract

Medication non-adherence to chronic therapies may severely impact effectiveness of treatment. Non-adherence may occur at different stages in a patient’s treatment journey. It may occur at the very beginning of therapy if a patient receives the initial prescription but does not redeem it at a pharmacy (primary non-adherence), or it may happen after the patient fills a prescription at a pharmacy but fails to follow the instructions or fails to refill the prescription (secondary non-adherence). The purpose of this study is to demonstrate that both primary and secondary non-adherences can be jointly described by a hurdle model, which has the interpretation as a two-part model. The first part is a binary outcome model, and the second part is a truncated count model (Poisson or negative binomial). The hurdle model is an example of the finite mixture models which can be fitted by SAS’s new procedure PROC FMM. Data in this retrospective cohort study of medication non-adherence was obtained from blind computerized pharmacy records of a national retail pharmacy chain. Primary non-adherence was defined as a binary outcome representing failure to fill a new prescription within 30 days after the medication was prescribed to the patient and secondary non-adherence was defined as a number of refills obtained by a patient within a 12 month follow up period. Various measured patient, prescription, and prescribing physician characteristics were included in the model. Hurdle model results indicate that important predictors are missing from the single-component models, but exist in the joint model of primary and secondary non-adherence. The authors conclude that a hurdle modeling approach enables the taking of simple, well-understood models primary (logistic regression) and secondary non-adherence (count regression) and combine them in a way that provides a better description of the data than a single-component models provide separately.