Reliability of death registration in South Africa : a case study

Abstract

There is under-utilisation of death registration data for mortality estimation in South Africa even though completeness of these data was found to have improved since 1996. Low content quality of these data is gauged by misclassification various causes of death, particularly the under-reporting of deaths due to HIV/AIDS. This thesis sought to investigate overall reliability of these data. To accomplish this, analysis was classified into two parts. The first part focused on the national death registration data for the period 1999-2005. A modified assessment framework that draws from Rao et al. (2005), Franca et al. (2007) and Mahapatra et al. (2008) was used to evaluate quality of these data. This framework is categorized into three categories, namely, generalisability, reliability and content validity. The second part constituted the Emalahleni case study involving record linkage for the age group 15-49 in 2003-2005. Administrative records (include hospital, municipal and police records) believed to be the reference point for death certification were collected in Emalahleni and linked to death registration data for the same population and period. Completeness was found to be above 90%, confirming significant improvement in the registration of deaths in recent years. The usability of death registration data for mortality estimation nationally was evident as far as comparability with other data; shifting sex ratios signifying excess female mortality within the reproductive age group over time and consistency of cause-specific mortality with epidemiological expectations were concerned. Limitations of utilising these data for mortality estimation by sub populations suggest low content validity. Although content validity was found to be low, these data do show what they purport to show. For example, the existence of the bipolar protracted epidemiological transition was confirmed. Also, mortality reversal in the 1990s due to HIV/AIDS was apparent up to 2004. However, consistency of cause-specific mortality at individual disease level was not quite good, suggesting misreporting and misclassification of cause-of-death. Overall, low data quality associated with in-hospital registered deaths gives cause for concern and may as well be one of the effects of the exodus of health personnel nationally. The potential for usability of these administrative data as another data source in verifying accuracy of information recorded in death registration data and for mortality estimation was also confirmed, signifying improvement in data collection systems. Agreement of recorded ages and cause-of-death was linked to grouped data. The kappa statistics provided an overall agreement of 0.1863 (95%CI: 0.1857-0.1870) for cause-of-death when comparing death registration data and hospital records. Hospital of death was found to be significantly associated with the agreement of underlying cause in Emalahleni. The suspicion that deaths due to HIV/AIDS are misclassified to tuberculosis and pneumonia on death certificates was confirmed. The study found that these deaths were also misclassified to other diseases on death certificates other than these two. Again, many other causes of death other than HIV/AIDS were also misclassified on death certificates in Emalahleni. Overall, these findings provide a measure of reproducibility of information from administrative records to death certificates in the study area during 2003-2005 and may not be generalised to other places in the country. There is need to shift efforts from improving death registration completeness towards content quality. This should be accompanied by reviewing current health policies pertaining to service delivery and disease reporting. Also, universal education on the benefits of early detection of diseases is recommended. The approach of the case study may be extended to other municipalities in other provinces for comparison purposes. Show more