CHAPTER 2: LITERATURE REVIEW
2.6 Agreement of Causes of Death between Hospital Records and Verbal Autopsy
The planning of public health systems should be based on reliable and timely data on the leading causes of death and disability. Globally, civil registration systemsare important tools commonly used to ascertain population-based mortality; however, in low- and middle-income countries,problems with completeness and accuracy of causes of death persist(Mahapatra et al., 2007; Setel et al., 2007).
In many low- and middle-income countries, especially those in sub-Saharan Africa, cause of death data are inadequate or not readily available (Mahapatra et al., 2007;
Setel et al., 2007; Ye et al., 2012, Phillips et al., 2013). In South Africa, the quality of cause of death data capturesin civil registration systems remains deficient, with many causes of death recorded as “ill-defined” and/or “undetermined” (Burger et al., 2007; Nojilana et al., 2009; Pillay-van Wyk et al., 2011;Birnbaum et al., 2011).Therefore, statistics on registered cause of death require careful evaluation if they are to be useful for epidemiological research and health service planning
Comparison of cause of death reported on death certificates with clinical records or verbal autopsies areusually assessed by using the concepts of sensitivity, specificity or kappa statistics (Rao et al., 2007; Wang et al., 2007; Burger et al., 2012;
Misganaw et al., 2012a; Khalili et al., 2012).However, sensitivity and specificity are the most frequently used measure to quantify interrater agreement (Rao et al., 2007;
Wang et al., 2007; Khalili et al., 2012).
Rao et al compared diagnoses recorded on medical records and those recorded on death registration systems for 2917 deaths that occurred in secondary and tertiary hospitals in China. The study reported the sensitivity of all causes of deaths identified by death notification forms to be on average 50%-75% (Rao et al., 2007).
The study found an overall misclassification rate of 21.6%, with 12% of reported injury deaths (mostly from falling) attributed to natural causes on the medical records.
A comparative study between deaths recorded through the use of verbal autopsy reports and those recorded in death registry records for 2482 deaths in China found that injuries, such as fall and poisoning, ischaemic heart disease, chronic obstructive pulmonary disease and tuberculosis had sensitivity lower than 50% (Wang et al., 2007). A cause of death misclassification rate of 40.4% was reported in this study.
Another study in China where 3290 cause of death records were analysed reported a higher sensitivity for strokes, cancers and transport accidents, while low sensitivity was reported for ischeamic heart disease, chronic obstructive pulmonary disease, diabetes and tuberculosis (Yang et al., 2006).
In an Indonesia study, of the 446 deaths analysed, a greater proportion (66%)of deaths was attributed to cardiovascular disease recorded on death certificateswhen compared to 49%of deaths by verbal autopsy reporting (Pane et al., 2013). In a recent review Rampatige et al revealed massive misclassification of causes of death in hospitals in China, Iran, Mexico, Sri Lanka and Thailand (Rampatige et al., 2014).
In these studies most deaths from ischaemic heart disease were attributed to strokes, while HIV/AIDS deaths were attributed to septicaemia.Various forms of heart disease, diabetes and chronic constructive pulmonary diseases were often coded as ill-defined causes (Pane et al., 2013; Rampatige et al., 2014).
A cohort study in Tehran, Iran that compared deaths diagnosed by a panel of specialists and deaths recorded on death certification forms for 367 deaths found a sensitivity of greater than 60% in all natural causes of deaths, except for the disease of nervous system and genito-urinary system (Khalili et al., 2012). Another study in Iran found that VA reporting had a sensitivity of 80% or more and Kappa of 75% in determining the general cause of death (Khademi et al., 2010).
In a Tanzanian study that compared deaths diagnosed by verbal autopsy reporting and deaths recorded on medical records for 3123 deaths found a sensitivity of higher than 50% for HIV/AIDS, malaria, cerebrovascular disease, injuries and malignant neoplasm of the gastro-intestinal tract (Setel et al., 2006). In Uganda, Mpimbaza et al carried out a study of 719 deaths of under5 year old children with the aim of assessing the diagnostic accuracy of VA procedures compared to hospital medical
procedures for determining death from malaria was 50% or more, specificity 80% or more and the positive predictive value was between 30% and 80%.
In Ethiopia, a validation study amongst 335 adult deaths indicated that for communicable diseases the values of sensitivity, specificity and positive predictive value of VA diagnoses were 79%, 78%, and 68%, respectively, while for non- communicable diseases, sensitivity was 69%, specificity 78% and positive predictive value 79%. Injury sensitivity of the verbal autopsy diagnosis was 70%, specificity 98% and positive predictive value 83% (Misganaw et al., 2012a). The study showed a misclassification pattern of 40% or more for cardiovascular diseases and infectious diseases such as tuberculosis and HIV.
Burger et al (2003-2004), in their study, compared causes of deaths on the death notification forms (DNF) with diagnoses recorded in hospital records for 703 deaths in Cape Town, South Africa. This study found the sensitivity of DNF for identifying all natural causes of death studied was greater than 50%, except for infectious diseases, diabetes and cardiovascular disease (Burger et al., 2012). The overall disagreement was 45% between the underlying causes of death as coded in the medical records and on the death certificate (Burger et al., 2012).