Emerging Methods in Progression Modelling of Alzheimer's Disease: A Comparative Analysis
Authors
Aldous, Jonas Benjamin Rose ; Sørensen, Sandra Helweg
Term
4. term
Education
Publication year
2024
Submitted on
2024-06-03
Pages
73
Abstract
This thesis examines progression models for repeated measures (PMRMs), introduced by Raket (2020), using data supplied by Novo Nordisk A/S from the Critical Path for Alzheimer’s Disease database. Repeated measures means following the same participants over time. We first review mixed models—methods that combine overall trends with individual differences—and explain PMRMs alongside the commonly used constrained longitudinal data analysis (cLDA) model. We then propose simple modifications to PMRMs, such as dropping a correlation assumption in the errors and adding a random effect to better capture variation between individuals. We also show how PMRMs can be applied to study heterogeneity of treatment effects across subgroups. To evaluate performance, we run a simulation study that compares PMRMs (and their extensions) with the cLDA model across different scenarios. The results show trade-offs: cLDA delivers robust performance with well-controlled type I error (few false alarms), while PMRMs can be easier to interpret and offer higher statistical power in certain settings, making them promising for clinical trials. Finally, we outline how PMRMs can be used in health economic modelling by linking them to a Markov model under an assumption of a constant treatment effect over time, and we sketch how this can feed into cost-effectiveness analysis. We conclude by discussing how understandable the model estimates are and where these approaches may support health economic modelling and decision-making.
Dette speciale undersøger progression models for repeated measures (PMRMs), introduceret af Raket (2020), med data leveret af Novo Nordisk A/S fra Critical Path for Alzheimer’s Disease-databasen. Repeated measures betyder, at de samme deltagere følges over tid. Først gennemgår vi blandede modeller (mixed models) – metoder der kombinerer overordnede tendenser med individuelle forskelle – og forklarer PMRMs sammen med den almindeligt anvendte constrained longitudinal data analysis (cLDA)-model. Dernæst foreslår vi simple ændringer af PMRMs, fx at fjerne en antagelse om korrelation i fejl og at tilføje en tilfældig effekt (random effect) for bedre at opfange variation mellem individer. Vi viser også, hvordan PMRMs kan bruges til at undersøge heterogenitet i behandlingseffekter på tværs af undergrupper. For at vurdere ydeevnen gennemfører vi et simulationsstudie, der sammenligner PMRMs (og deres udvidelser) med cLDA-modellen i forskellige scenarier. Resultaterne viser afvejninger: cLDA giver robust ydeevne og velkontrolleret type I-fejl (få falske alarmer), mens PMRMs i visse situationer er lettere at fortolke og har højere statistisk styrke, hvilket gør dem lovende til kliniske forsøg. Til sidst skitserer vi, hvordan PMRMs kan bruges i sundhedsøkonomisk modellering ved at koble dem til en Markov-model under antagelsen om en konstant behandlingseffekt over tid, og vi giver et overblik over, hvordan dette kan indgå i en omkostningseffektivitetsanalyse. Afslutningsvis diskuterer vi, hvor forståelige modelestimaterne er, og hvor metoderne kan understøtte sundhedsøkonomisk modellering og beslutningstagning.
[This apstract has been rewritten with the help of AI based on the project's original abstract]