Statistical Methods for Pharamacogenetics

Pharmacogenetics is the use of genetic information to improve the prescribing of drugs, either through rapid prediction of the correct dose for a patient according to his/her genetic type, or by avoiding the prescription of drugs to those with unsuitable genetic types. Until recently pharmacogenetics was focussed on a small number of genes known to be involved in drug transport or action, but increasingly genetic data from the whole genome is taken into account. Although so-called personalised medicine has been talked about for many years now, the achievements to date in this area remain limited to a few diseases and drugs. One bottleneck in the process is efficient statistical analysis of datasets that are often relatively small, but complex in terms of both the outcome measures made on patients and the diversity of patients and their treatments. The goal of this proposal is to develop and make available better statistical methods that can help make pharmacogenetics more productive.

Pharmacogenetics (PGx) has the potential to generate improved drug therapies for patients, particularly in more accurate dose-finding and avoidance of adverse events, as well as economic benefits to the pharmaceutical industry (possibility to rescue drugs not currently marketable) and to healthcare providers such as the NHS (reduced prescriptions of ineffective or harmful drugs). Broadly speaking, PGx has until now failed to deliver substantial return on investment, but the potential rewards are so great that continuing effort and investment is warranted. Because of small and heterogeneous datasets, appropriate statistical analyses are crucial to maximising the output of pharmacogenetics research, but to date there appears to have been relatively little investment, at least in the public sector, in the development of statistical methods tailored to the needs of PGx research. In this proposal we seek to make several contributions to statistical methodology to enhance the prospects of PGx, focussing on two broad strands: (1) the prediction of drug response from genome-wide genetic data, and (2) Bayesian modelling of population and other sources of heterogeneity in order to extract more information from complex data sets.