Developing methods for identifying clinical phenotypes from routinely collected health data with applications to stroke genetics.

点击翻译按钮获取中文摘要

Genetic, and other risk factor associations with stroke are known to be to a large extent type and subtype specific. Systematic review data and UK Biobank pilot work data has shown that stroke type (ischaemic versus haemorrhagic) is not specified for around 40%, and further stroke subtype (TOAST, OCSP, haemorrhage location etc) is not specified for around 70% of stroke cases ascertained from routinely collected coded health data. With expert adjudication it is possible to assign a type and subtype for around 80% of these cases, but this is not a scalable method suitable for very large studies (e.g., UK Biobank). I propose to develop scalable, automated methods that will allow further stroke typing and subtyping from routinely collected health data by investigating the use of various algorithmic code combinations and of natural language processing methods that could be applied to free text medical records and imaging reports. I would then propose to validate these methods directly, as well as indirectly by comparing them with other phenotyping approaches in genetic studies.Phenome-association studies can be used to systematically examine the impact of one or many genetic variants across a broad range of human phenotypes, and have the potential to reveal novel insights to underlying disease mechanisms, as well as hold great potential for the identification of novel drug targets and drug repurposing opportunities. UK Biobank with its vast and varied phenotypic data is a dataset that is highly suitable for these studies. However, to date there is a relative lack of sophisticated phenotypic methods to select and identify outcomes of interest. I propose to apply existing phenome-wide association study methods to investigate hypothesis-based associations with stroke as a model disease, and to develop these methods further for wider use. During the past decade, findings of genome-wide association studies have improved our knowledge and understanding of complex disease genetics. Statistical analysis typically looks for association between a phenotype and single genetic variants taken individually via single-variant tests. However, this is an oversimplified approach to tackle the complexity of underlying biological mechanisms. The next steps would be to also consider the interactions between genetic variants, or epistasis. Epistasis detection gives rise to new analytic challenges since analysing every single nucleotide polymorphism combination is at present impractical at a genome-wide scale. I propose to apply existing methods and develop these further for wider use, starting with a hypothesis-driven approach to investigate epistatic associations between selected stroke genes.