The art of research cohort construction using administrative datasets

ObjectivesUsing research case studies, I will give an overview as to how researchers can experience problems translating great research ideas, grounded in the relevant literature, into a robust research design. Assuming that the question cannot be reliably researched using small but rich sample surveys, I will present ways in which routine admin data can help in this context, along with the additional challenges posed when creating the correct cohort to address the research question. MethodsThe examples given are from a study which links together routinely collected administrative data for a representative sample of the population. It includes a wealth of information from census, civil registration data, education data, and with further permission health data can also be linked. The study in question is provided deidentified, and accessed within a data enclave. The size and scope of the study make it an unparalleled resource for analysing a range of socio-economic, demographic and health questions. ResultsI will demonstrate how despite the large number of study members, these cohorts must be carefully considered in order to research outcomes (events/results) due to inconsistency in the years that different admin data are available centrally in admin systems (health/education data). Examples include: life-course events for a cohort of women followed up to mid-life, and a complete sub-cohort followed up into later-life. ConclusionsTo showcase exciting ways in which linked admin data can be operationalised to take a longitudinal/life-course approach to address research questions, and to emphasise cohort construction is complex.