Abstract of Thesis University of Manchester Abstract of thesis submitted by Imola Nemeth For the degree of Doctor of Philosophy (PhD) Using biomarkers to monitor, predict and model the progress of Alzheimerâs disease 2017 Alzheimerâs disease (AD) is the most common type of dementia, associated with senile plaques and neurofibrillary tangles in the brain tissue. With no existing cure for it, current trials are aimed at modifying or slowing the progress of the disease in subjects at the earliest stages of the disease. A major challenge within the AD research consists of identifying subjects within the early phase of mild cognitive decline who will eventually develop AD, distinguishing them from subjects within the same phase who will remain in the stable MCI phase without further cognitive decline. For this purpose increasingly more prediction studies and studies aimed at modelling the progress of the disease are carried out. In order to facilitate the progress within the field a large longitudinal, multi-centre study, the Alzheimerâs Disease Neuroimaging Initiative (http://www.loni.ucla.edu/ADNI) study was carried out giving open access to years of data collected on subjects being at various stages of AD as well as on normally aging healthy controls. In this thesis different methods for predicting conversion to AD as well as modelling the cognitive decline within AD are assessed with the purpose of improving prediction accuracy and better understanding of the complexity of the disease. Thus, logistic regression models seek to find highly accurate models by comparing several possible combinations of potentially influential predictors. The limitation of the method arising from subjects progressing at different time points following different speeds of decline results in directing the investigation towards assessing latent class growth models. The purpose of this second investigation lies in better characterization of different patterns of cognitive decline. After several tested models giving no convincing evidence of the presence of discrete classes, highly influenced by the high variability of subjects assigned to the different classes a new method (SITAR) for modelling progression using mixed effects and hence reducing variance in the data is thoroughly examined. The SITAR (superimposition by translation and rotation) method uses random effects to move individual curves around one mean curve characteristic of the whole spectrum subject, thus significantly reducing the between-subject variability. The results of the SITAR method in modelling cognitive decline using longitudinal measurements of both cognitive test scores and imaging biomarkers had outstanding results, reducing the variability of the data for the different variables between 87 and over 99%. The best performing biomarker with the SITAR method proved to be the hippocampal volume, on which further investigations were carried out, testing the performance of this new method on various sub-groups. All sub-group analyses provided highly significant results (above 95% reduction in variance), indicating that the SITAR method can be applied for reducing variability in AD data. The results indicate that the usefulness of SITAR method for modelling dementia should be further exploited on larger sets of data in order to validate initial findings and further explore its applicability, usefulness and perspectives.