Validation of multiple linked administrative datasets to identify people with dementia

Background Accurately measuring dementia prevalence and incidence (including Alzheimer's disease) is important for service planning and chronic condition management. Linked administrative data present an accessible and inexpensive method to estimate dementia prevalence and incidence. Objective To examine the accuracy of identifying people with dementia in linked administrative data through comparison with adjudicated cases of dementia from a community-based clinical trial. Methods We created a comprehensive linked administrative dataset containing people with dementia in Australia and used it to develop algorithms that identified people with dementia. These algorithms were tested against adjudicated cases of dementia (both confirmed and not confirmed) in the ASPirin in Reducing Events in the Elderly (ASPREE) clinical trial (Number: ISRCTN83772183, Registration date: 14/07/2005) to help determine optimal sensitivity and specificity. Results The algorithm that provided the highest sensitivity and positive predictive value estimates on balance without excluding ASPREE trial participants with suspected (but not confirmed) dementia was having at least one record for dementia in a minimum of two sources or two records with dementia in one source based on truncating the timeframe of randomization in the ASPREE trial to one year from the date that the first participant was randomized. This algorithm produced a sensitivity of 70.7%, specificity of 98.2%, positive predictive value of 72.6% and negative predictive value of 98.1%. Conclusions Linked administrative data can identify people with dementia with moderate accuracy. This is promising but requires enhancement before it can be used for estimating dementia prevalence and incidence. Clinical trial name and URL ASPirin in Reducing Events in the Elderly; ISRCTN - ISRCTN83772183: ASPirin in Reducing Events in the Elderly ( https://www.isrctn.com/ISRCTN83772183 )