A time dependent adaptive learning process for estimating drug exposure from register data - applied to insulin and its analogues

In register based drug research time dependent drug exposure is evaluated based on information available from the prescription register. In Finland the exact daily dosage information is not available and thus has to be estimated from the total amount purchased and the purchase patterns. The aim of this master thesis project is to develop a new algorithm, named Time Dependent Adaptive Learning Process (TDALP), which is a prospective method and applies the historical records to estimate the current insulin daily dosage. Through testing under dierent situations including both constant daily dosage cases and alterable daily dosage cases, it is demonstrated that the performance of the new algorithm is superior over the more traditional methods. iii Preface I would like to express my sincere gratitude to all the persons and organizations supporting me during the two years master program. My gratitude goes rst to the European Commission who gave me a precious opportunity to join in Erasmus Mundus programme and supports my study with scholarship generously. At the same time, My gratitude goes specically to D.Sc. Fabian J. Hoti who directly guided my thesis work and EPID Research Oy providing this lovely internship position for me. And I also own my great gratitude to my two supervisors: Prof. Erik Aurell and Prof. Torbjörn Gräslund who guided me from macroscopic level and provided me their treasured comments. Finally, I would like to thank for their cozy cares from academic area and daily life during this thesis work.