Accuracy of death certifications of diabetes, dementia and cancer in Australia: a population-based cohort study

Background National mortality statistics are only based on the underlying cause of death, which may considerably underestimate the effects of some chronic conditions. Methods The sensitivity, specificity, and positive and negative predictive values for diabetes (a common precursor to multimorbidity), dementia (a potential accelerant of death) and cancer (expected to be well-recorded) were calculated from death certificates for 9 056 women from the 1921-26 cohort of the Australian Longitudinal Study on Women's Health. Log binomial regression models were fitted to examine factors associated with the sensitivity of death certificates with these conditions as underlying or contributing causes of death. Results Among women who had a record of each of these conditions in their lifetime, the sensitivity was 12.3% (95% confidence interval, 11.0%, 13.7%), 25.2% (23.7%, 26.7%) and 57.7% (55.9%, 59.5%) for diabetes, dementia and cancer, respectively, as the underlying cause of death, and 40.9% (38.8%, 42.9%), 52.3% (50.6%, 54.0%) and 67.1% (65.4%, 68.7%), respectively, if contributing causes of death were also taken into account. In all cases specificity (> 97%) and positive predictive value (> 91%) were high, and negative predictive value ranged from 69.6% to 84.6%. Sensitivity varied with age (in different directions for different conditions) but not consistently with the other sociodemographic factors. Conclusions Death rates associated with common conditions that occur in multimorbidity clusters in the elderly are underestimated in national mortality statistics, but would be improved if the multiple causes of death listed on a death certificate were taken into account in the statistics.