Diagnostic value of death certificates for case ascertainment of dementia: A population-based assessment of mortality registry data

Background: Use of mortality registry data could facilitate dementia case ascertainment in research, but its diagnostic value is uncertain. Objective: To determine the diagnostic value of death certificates for ascertainment of dementia. Methods: We included 6280 deceased participants of the population-based Rotterdam Study (mean age at death: 83.6 years, 57% women), who had been actively followed during life for the occurrence of dementia through repeated in-person and medical record screening between 1989-2018. We compared lifetime dementia diagnosis to dementia being certified as cause-of-death in the Dutch national mortality registry. We determined concordance of death certificates according to coding method, place of death, and participant characteristics (age at death, dementia duration, and comorbidity). To validate observations, we surveyed 446 certifying physicians. Results: Overall, 1828 persons (29%) had a diagnosis of dementia during life. Specificity of death certificates for dementia diagnosis consistently exceeded 96%. Sensitivity was below 10% using ICD-9 but increased to about 70% in more recent years using ICD-10. Diagnostic accuracy improved with introduction of ICD-10 (odds ratio [95%CI]: 1.94 [1.08-3.48]), and varied substantially with age at death (OR per decade increase: 0.80 [0.66-0.96]), dementia disease duration (OR [95%CI] for >= 6 years versus 0-2 years: 3.95 [2.56-6.09]), place of death, and comorbidity (OR [95%CI] per additional illness: 0.82 [0.72-0.93]), particularly due to stroke or heart failure. Dementia severity and duration were also reported by physicians as important factors for certifying dementia as a cause of death. Conclusions: Death certificates have high specificity but limited sensitivity for diagnosis of dementia, in particular in older patients with cardiovascular comorbidity.