Statistical methodology for the analysis of Electronic Health Records (360G-Wellcome-338020_Z_25_Z)

We often want to know how common a disease is (prevalence), how often people get a disease (incidence) and who is most likely to get a disease. We have previous patients’ electronic health records on computers which we can use to answer these questions. This form of data has many benefits as it is quick and cheap to collect whilst covering diverse populations. However, there can be problems as the data was not originally intended for research and contains missing and incorrect data. If we use simple statistical methods that ignore these issues, we can end up with misleading results. I propose to develop new statistical methods that can handle the missing data, combine different data sources, and can take into account inaccurate or uncertain data. I will develop tools which will encourage and enable other researchers to use the methods as well. I expect my work to improve the accuracy of research which uses electronic medical records. In particular, the methods will help doctors know when to invite people for disease screening, help guide doctors and patients on treatment decisions and provide a greater insight on disease progression.