Medicine (RMH) - Theses
Permanent URI for this collection
Search Results
1 - 1 of 1
-
ItemBioinformatics: the application of multigenic models to predict disease and treatment outcomes( 2011)One of the single most difficult aspects of treating newly diagnosed epilepsy, for patients and clinicians, is managing the uncertainty surrounding whether the anti-epileptic drug (AED) therapy will prevent further seizures. By investigating the pharmacogenomics of a population of Australian newly treated epilepsy patients, this thesis seeks to identify whether it is possible to develop risk prediction models that could improve personalising the treatment care provided to this population of patients by reducing the uncertainty and helping select the “right drug for the right patient”. The thesis explores the premise that developing predictive models for a complex phenotype such as seizure control in newly treated epilepsy, will require the utilization of multiple genomic variants, and also genomic and non-genomic factors need to be integrated. The results of the research demonstrated a proof-of-concept for this premise, showing that using the genetic profiles across five genetic variants improves the ability to significantly predict epilepsy pharmacoresponse in comparison to investigating each of the individual genomic components. This model was developed in an Australian population of newly treated epilepsy patients prospectively followed to determine the outcome of their drug treatment, and then validated in two other Australian cohorts. This model has potentially important clinical implications given that the current standard of care for newly diagnosed epilepsy patients does not provide clinicians with a meaningful tool to determine which of the patients are at a higher risk of not-responding to their initial AED therapy. To understand the broader clinical utility of the model, the Australian derived multigenic model was tested in two newly treated epilepsy cohorts from the United Kingdom. It was found that the Australian population did not significantly predict treatment outcome in the UK newly treated populations overall. However, the combination of five-SNPs identified as being relevant to pharmacoresponse in the Australian population treated with carbamazepine and valproate, were significantly predictive of pharmacoresponse in all the UK patients that were treated with these two drugs. This suggests that these five SNPs have drug specific predictive value. With a focus on developing a more accurate predictive model, non-genomic factors were also investigated and integrated with the genomic predictors into a unified predictive model. This model was found to have high predictive value for seizure control in newly treated epilepsy, and would potentially provide a clinically useful tool to assist in the ability to personalise treatment advice given to newly treated patients. The ability to determine non-responders improved from the currently generic 30% likelihood of not responding to initial carbamazepine or valproate treatment in the Australian newly treated population, to a 75% likelihood of not responding if the integrated, pre-treatment, model predicts the patient to be a “non-responder”, or conversely a 82% likelihood of responding if the model predicts the patient to be a “responder”. As a final task, the model development approach, which resulted in successfully identifying a multigenic predictive model for epilepsy pharmacoresponse based on a limited candidate gene dataset of approximately 4,000 carefully selected genetic markers, was applied to two genome-wide datasets. Based on this attempt to directly upscale the model development from candidate genes to a genome-wide scale, a number of limitations in the marker selection and model development stages of the approach were identified. These limitations emphasize the importance of designing tailored approaches to identifying multigenic models based on the differing contexts of datasets. Here, both in the epilepsy pharmacoresponse and HIV-1 susceptibility genome-wide attempts, the approach did not result in the development of significantly predictive models when applied to independent validation cohorts. However, possible future directions on how to overcome some of these direct upscale limitations are provided. Additionally, potential future directions that could result from this body of work are also explored.