Improving diagnosis and treatment for the quantified patient

There are often many alternative prevention and treatment options for common diseases, with varying effectiveness. Key aims of stratified medicine are to identify subgroups of patients for whom similar interventions are likely to be most effective, or who are at high risk of further complications. This project aims to find ways of defining these subgroups and predicting which subgroup any given patient belongs to based on their genetic data, clinical presentation of their disease over time, and the environment they have been exposed to. Psoriasis, an inflammatory skin disease, will be the primary focus of the research but opportunities will be sought to extend the findings to other diseases.

Firstly, existing clinical and genetic patient data will be used to identify genetic differences between patients that respond well and those that do not respond to a relatively new class of psoriasis drugs called biologics. Statistical methods will be developed to determine whether these genetic differences can be combined with previously established clinical measurements to predict which psoriasis patients will respond to each biologic. A "treatment algorithm" (a set of rules to help dermatologists identify the most effective biologic as quickly as possible) will be recommended.

Secondly, the project will seek genetic and environmental/lifestyle factors that put psoriasis patients at risk of progressing to severe disease or developing complications in the form of other diseases ("comorbidities"). This will involve using data from UK Biobank, which comprises 500,000 volunteers from the general UK population. It is known that participants do not always accurately report their health conditions, so an initial challenge will be to accurately determine which participants have psoriasis so that we can study them further. Genetic differences will be examined between participants that have psoriasis and those that do not to refine our understanding of the genetic basis of psoriasis. Subsequently statistical analyses will be performed to identify genetic factors that correspond to (1) severity of psoriasis and (2) risk of comorbidities, and to examine how a patient's environmental exposures contribute to psoriasis risk after accounting for their underlying genetic risk.

Thirdly, the project will examine ways of using artificial intelligence (AI) to automatically analyse patient data and help dermatologists or other healthcare providers to make informed healthcare decisions. For these advanced methods to be effective much larger datasets are needed than are currently available, including regular data on individual patients over time. Relevant types of data could include skin images or descriptive data. New methods of collecting these data will be explored, such as the development of a mobile phone app(lication) to collect data directly from psoriasis patients, and the data infrastructure that would be required to support collection and subsequent AI analysis. It is also proposed to identify existing but currently untapped data resources that can be used to run initial test AI analyses. This could include analysis of image data (e.g. can photographs of psoriasis skin lesions be distinguished from other types of inflammatory skin disease automatically?) or patient-reported "life-quality" data (e.g. can patients that experience a large drop in perceived life quality be predicted at an early stage?).

Finally, opportunities will be sought to collaborate with other research groups so that results and methodology developed during this work can support stratification of other common long-term diseases.

Stratified approaches to the clinical management of disease promise a substantial public health benefit. The proposed research seeks to identify and validate generalisable stratified approaches, using psoriasis as a model disease, based on unique clinical, genomic and population-scale datasets.

Firstly, using data generated by the Psoriasis Stratification to Optimise Relevant Therapy (PSORT) project, a genome-wide association study will be completed to identify genetic variants robustly associated with response to multiple biologic therapies. A framework will be established to evaluate combined genetic and clinical predictors of response. A treatment algorithm will be recommended and opportunities to validate it locally or via collaboration will be sought.

Secondly, the contribution of genetic and environmental/lifestyle factors to disease course and severity will be investigated. Initial efforts will focus on utilising several established methods to infer and validate psoriasis phenotypes in population-scale datasets such as UK Biobank, including linked electronic health records. Derived data along with PSORT data will be used to refine our understanding of genome-wide psoriasis susceptibility signals via case-control meta-analysis. Genetic and environmental contributions to psoriasis incidence and severity at the population level will be investigated, to include study of gene-environment interaction and comorbidity risk.

Thirdly, artificial intelligence (AI)-based skin disease assessment and management recommendations will be developed. Novel data collection mechanisms will be explored to generate sufficiently large/dense longitudinal datasets; for example, image or text data collected via mobile technology. AI frameworks will be developed and evaluated which use these data to predict when clinical intervention can most effectively reduce disease burden. Pilot analyses will focus on analysis of longitudinal image and patient-reported "life quality" data.