BUILDING BETTER PHENOTYPES: MEASURING ANTIDEPRESSANT TREATMENT RESPONSE USING ELECTRONIC HEALTH RECORDS

Linkage to electronic health records provides an important opportunity for examining the genetics of depression and treatment on a population scale. While depression diagnoses are identified using clinical coding schemes, symptoms and treatment regimens are often not systematically recorded. This makes tracing treatment response difficult, despite the volume of data available. In the AMBER project, we are enabling large-scale studies of antidepressant treatment response by using electronic health record data to build better phenotypes. Working with clinical and lived experience experts, we have produced several treatment response phenotypes – including maintenance, treatment resistance, and discontinuation – using both structured (clinical coded) data and unstructured (clinical text) data. In structured work, we combine features extracted from routinely-collected prescribing and dispensing records. In unstructured work, we use natural language processing to extract features of interest from primary care clinical notes. We highlight each of the phenotypes, the methods used to produce them and the datasets they have been developed for. We also discuss the contributions of structured and unstructured data to each phenotype, relative to the difficulty of access and use of the data for research. Future work will focus on combining structured and unstructured methods to improve phenotyping, validating phenotypes in other datasets, and examining links with biology.