Henrik Støvring,
Department of Public Health Aarhus University will defend his Dr.Med.Sci. dissertation titled
Using prescription data as indicators of disease or drug use, new analytic methods based on renewal processes
on Friday, 22 March 2019 at 14:00 at Aarhus University in Lille Anatomisk Auditorium (building 1231, room 424), Wilhelm Meyers Alle 3, DK-8000 Aarhus C.
Opponents:
Associate Professor Fredrik Granath, Clinical Epidemiology Division, Dept. Medicine Solna, Karolinska Institutet
Professor Robert J Glynn, Division of Pharmacoepidemiology & Pharmacoeconomics, Harvard Medical School
Professor Lars Pedersen (chairperson), Klinisk Epidemiologisk Afdeling, Aarhus University
The dissertation can be acquired by contacting Henrik Støvring (stovring@ph.au.dk).
Abstract
Electronic prescription registries are the fundamental data sources in modern pharmacoepidemiology. Information on the single prescriptions of each patient is detailed, but the duration of the prescription is often not known, i.e. it is not recorded how long a patient remains treated after the redemption. The first part of the dissertation shows how it is possible to study the epidemiology of diabetes in Denmark using only information on redeemed prescriptions. The studies develop a book-keeping approach in which the number of treated patients (prevalence) at the end of the year is given by the prevalence at the beginning of the year plus the number of patients initiating treatment during the year (incidence) minus the number of patients stopping treatment during the year (stopping).
Stopping is further subdivided into treatment cessation and death. The two studies in this part use a simple classification to determine who are considered treated at the beginning of the year based on whether the patient had redeemed a prescription in the preceding year or not. The studies showed that changes in diabetes incidence had little effect on the contemporary change in diabetes prevalence in the period 1992-2003 in Funen County, Denmark. The second part consists of six studies that develop and apply new statistical methods to estimate prevalence, incidence, stopping of treatment, prescription durations and the probability of an individual being treated on a given day. Existing methods in pharmacoepidemiology are largely based on decision rules for determining treatment status at a given point in time, and so are not based on a statistical model as such. The Waiting Time Distribution (WTD) is the distribution of time from the beginning of a time window until the first prescription redemption of each patient within the window. The WTD was first suggested as a tool for graphical analysis without an underlying formal statistical model. This part of the dissertation shows how the WTD approach can be formalized using theory on renewal processes, which allows establishing a parametric model for likelihood analysis. The model is developed for both closed and open cohorts, for the first 2 redemption in a time window (ordinary WTD) and the last redemption (reverse WTD), and for inclusion of covariates. Properties of the models are studied both in simulation studies as well as in pharmacoepidemiological applications. The theoretical foundation provided by formulating a likelihood-based model allows for further development of new statistical approaches in pharmacoepidemiology.