Diagnostic accuracy of linked administrative data for dementia diagnosis in community-dwelling older men in Australia

BACKGROUND: Routinely collected health administrative data can be used to estimate the prevalence or incidence of dementia at a population level but can be inaccurate. This study aimed to examine the accuracy of hospital and death data for diagnosing dementia compared with a clinical diagnosis in community dwelling older men in Australia. METHODS: We performed a retrospective analysis of the Concord Health and Ageing in Men Project (CHAMP) in Sydney, Australia. Of the 1705 men aged ≥70 years in the CHAMP study, 1400 had available linked administrative data records from 1 year prior to 1 year post the date of clinical dementia diagnosis. The primary outcome was the accuracy of dementia diagnosis using linked administrative data records compared to clinical dementia diagnosis. The linked data diagnosis was based on hospital and death records for the 1 year pre and post the clinical diagnosis. Clinical dementia diagnosis was a two-stage process with initial screening, followed by clinical assessment for those meeting a validated cut-off. A final clinical diagnosis of dementia based on the Diagnostic and Statistical Manual of Mental Disorders (4th edition) criteria was reached by a consensus panel. RESULTS: Administrative data identified 28 participants as having dementia, compared to 88 identified through clinical assessment. Administrative data had a sensitivity of 20% (95% CI: 13–30%, 18/88), specificity of 99% (95% CI: 99–100%, 1301/1312), positive predictive value (PPV) of 62% (95% CI: 44–77%), negative predictive value of 95% (95% CI: 94–95%), positive likelihood ratio of 24.4 (95% CI: 11.9–50.0) and negative likelihood ratio of 0.80 (0.72–0.89). CONCLUSIONS: Administrative hospital and death data has limited accuracy for dementia diagnosis with poor sensitivity and PPV. The prevalence of dementia is likely underestimated using hospital and deaths data. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12877-022-03579-2.