P3‐117: Validity of dementia diagnoses in two Swedish health registers

Swedish health registers are excellent sources for population-based information in epidemiological studies. The Swedish National Patient Register covers all hospital discharges in Sweden since 1987 (with partial coverage since 1964) and the Cause of Death Register covers all deaths since 1961. This study aimed to investigate the validity of register-based diagnoses of dementia and Alzheimer's disease (AD) against independent clinical information. The validity measures specificity, positive predictive value and sensitivity were calculated for register-based dementia and AD diagnoses coded according to the International Classification of Diseases, revisions 7–10, in the Swedish National Patient and Cause of Death Registers. As gold standard, we used data from several population-based studies of dementia in Sweden that screened participants by telephone or in-person testing and used clincal work-ups to identify, in total, more than 1000 prevalent and 600 incident cases of dementia. Preliminary results showed that the specificity of register-based dementia diagnoses was more than 90 percent. Positive predictive value of register-based dementia diagnoses was 82percent in the National Patient Register and 90 percent in the Cause of Death Register. Gold standard cases of dementia other than AD were commonly misclassified as AD in the registers, leading to reduced accuracy of register-based AD diagnoses (67% in the National Patient Register and 63% in the Cause of Death Register). Additionally, about 10 percent of all register-based dementia cases were gold standard cognitively impaired but non-demented at the time of the register-based diagnosis. When combining both registers, 62 percent of gold standard dementia cases were ever detected. Age at onset was inversely associated with risk of detection of dementia in the registers. However, risk of detection seemed to be non-differential with regards to sex and educational attainment of the dementia case. Register-based diagnoses of dementia are overall accurate and specific, but many cases are missed. The Swedish National Patient and Cause of Death Registers are valid and useful as data sources in epidemiological studies of dementia etiology. However, due to the reduced sensitivity the registers are less suitable in studies of dementia incidence or prevalence.