Development and validation of novel data science tools for nationwide prescribing.

Studentship strategic priority area: E-Health Informatics Research
Keywords: Pharmacoepidemiology, drug exposure, pharmacy databases, electronic prescribing

Abstract:
Administrative healthcare databases are increasingly being used as a key data source for pharmaco-epidemiological research, as they provide access to large amounts of data collected during the provision of healthcare. The use of pharmacy records, for example, provides an opportunity to answer research questions related to adherence, post-marketing drug responses and safety and prescribing patterns over a large population without having to go through the costly and time-consuming process of collecting information through other methods such as patient interviews. A number of studies have been performed which validate the use of pharmacy records as a method of determining which drugs a patient has access to, by showing a high level of agreement between pharmacy records, patient interviews, medical records and home medication inventories.

In Scotland, the Prescribing Information System (PIS) is of particular interest for this type of research, as it contains information for over 1.6 billion prescriptions reimbursed in a community setting over the whole country from 1993 onwards, 507 million items prescribed and 344 million items dispensed from 2009 onwards. Additionally, records from 2009 on are linked to individual patients' Community Health Index (CHI) numbers, allowing for linkage to other health records within the Scottish National Health Service. These systems provide an infrastructure for longitudinal pharmaco-epidemiological research which is only partially available elsewhere in the UK, and the use of data from within the Scottish NHS in this type of research is of particular interest as Scotland has a relatively non-mobile population, allowing for long-term follow-up, and high incidences of a number of disease groups including cardiovascular disease and mental health conditions. The complexity of the dataset, however, means that researchers often need to develop new skills and tools to understand the best ways to use it within their research, and there are currently no standardised methods for describing, manipulating or reporting on findings from analysis of prescribing data. Additionally, as these databases were originally designed for reimbursement purposes rather than research, the raw data are not inherently useful for answering certain research questions, and must first be converted into periods of drug exposure, which must currently be done manually.

The aim of this study is to develop a set of generic tools to take individual prescription records and create periods of exposure for medications over a set time period for individual patients - initially modelling exposure to individual drugs, and then investigating the complexity of modelling several drugs over the same period to determine at what point the data becomes too complex to be used within statistical and prognostic modelling. These tools will then be validated by using them to investigate issues relating to polypharmacy within psychiatric prescribing in Scotland.