ETD

• dispensation length
• drug use
• pharmacoepidemiology
• secondary data
• waiting time distribution

In pharmacoepidemiology research based on the secondary use of data sources, estimating for how long a patient can be considered treated with a specific medication dispensation is fundamental. Classical assumptions based on fixed time or dose methods are widely used, but they have limitations, as they may not capture real-world variability in drug use. These methods may underestimate or overestimate the true dispensation length, leading to exposure misclassification. When estimating associations in pharmacoepidemiologic studies, such misclassification biases the estimated effects toward the null if non-differential, or either toward or away from the null if differential.
This thesis includes five studies to illustrate the reverse Waiting Time Distribution (rWTD) methodology, as a data-driven approach, methodologically well established, aiming at increase awareness of the method, improve its understanding, and facilitate its application through new implementations.
In the first study of this thesis time windows lengths are explored, when identifying current drug use, including polypharmacy. It is an example of application of standard assumptions for estimating the dispensation length. The study documented that time windows shorter than 120 days were too narrow to identify current drug use, at least in a Scandinavian context. Moreover, it was shown how variable the estimated prevalence of polypharmacy can be depending on the definition used, suggesting the need for considering multiple definitions in a study.
The second and third studies are tutorials based on the R and Stata package to apply the parametric waiting time distribution (pWTD), respectively, that show practically how to apply WTD to estimate dispensation length and interpret results obtained. The R tutorial paper is based on the wtdr R package publicly available on GitHub, which was developed and released as part of this research project.(1)
The forth study investigated methodological aspects of the pWTD that were still unexplored. Being based on a stochastic process, estimates provided are not highly precise. Therefore everything that can be done to improve precision is important, and this is what this study explored. The idea was to understand whether the precision of the estimated dispensation length could improve when saturating the sampling by considering each day of the time window of a pre-specified length as an index date, rather than considering m random index dates. We observed a gain in precision as expected, but it was not substantial compared with the scenario with 50 random index dates.
Finally, the fifth study aimed at predicting the dispensation length of dermatological topical drugs using Danish registry data. We focused on the dermatological and topical field, as one of the standard assumption largely made to estimate the dispensation length, based on the concept of Defined Daily Dose (DDD), could not be used as DDDs are not assigned for topical medications within dermatology. This study is still ongoing and results will be produced and disseminated during the postdoc period.