Methodological strengths and weakness of cohorts and administrative data for developing population estimates of dementia

Abstract

AbstractBackgroundThere are three main methods of obtaining population data on the incidence and/or prevalence of dementia: cross‐sectional surveys (which may be repeated over time); cohort studies that follow people initially without dementia and count newly diagnosed cases over time; and administrative health records (including linkage of records from multiple sources). The major challenges for all these methods are: how well the study sample represents the target population, the accuracy of diagnoses, and the costs of maintaining the data collection over time.MethodIn a project to improve Australia’s dementia statistics, we conducted a series of studies to compare population estimates of dementia obtained using different methods. Firstly, we used existing general health studies of community‐based cohorts, supplemented by linkage to administrative records of hospital and emergency department admissions, assessments for aged care support, medication prescriptions, and death certificates to estimate the cumulative incidence of dementia. Secondly, we created cohorts based on administrative records for entire populations. Thirdly, we assessed the validity of the identification of people with dementia in the record linkage cohorts in various ways, including linkage with studies that had obtained clinical diagnosis through the standardised assessment of participants.ResultWe will present empirical results illustrating the strengths and limitations of these different approaches. In summary, community‐based cohort studies lack representativeness of national or regional populations due to recruitment biases and differential loss to follow‐up. Cohort studies are also costly to maintain over the long time needed for participants to develop dementia. In contrast, the use of administrative records is relatively inexpensive, but is subject to policy changes that impact on the continuity of data coverage and quality. Population coverage may also be problematic for administrative data if important sources of care for people with dementia are not included; for example, in Australia linkable primary care data are not available. The validation studies showed that accuracy was highly dependent on data sources, and identification of dementia type was unreliable.ConclusionPrevalence and trends data of dementia obtained from multiple sources are needed to provide accurate population estimates, together with detailed contextual knowledge and careful analysis.