An increasingly common methodological response to the lack of ‘cause of death’ data for adults in sub-Saharan Africa – itself a product of the general absence of health facilities and the imperfect coverage of vital registration systems [1–4] – is the use of data gleaned from ‘verbal autopsy’ (VA) forms [5–10]. This refers to reports about a deceased individual's symptoms prior to death collected by lay personnel from that individual's close relatives or friends after death. These can then be used to ascertain the likely cause of death.
Prior studies have shown that, notwithstanding concerns about the impact of differential field and analytic procedures, and of limitations in informants’ knowledge or memory of the circumstances and symptoms surrounding death, VA tends to provide reliable data on the distribution of deaths by broad category . In particular, several studies have compared VA of adult mortality with clinical reports or physician's assessments in sub-Saharan Africa settings and have either found that VA data in general tend to have very high levels of specificity, with slightly lower scores on sensitivity indices , or that they score high on both [10,12]. These findings are consistent with the fact that sensitivity and specificity vary across categories of disease (e.g., infectious, parasitic, non-communicable), by specific disease, and across other causes of death (e.g., accident and violence). The typical conclusion of such studies is twofold. First, because VA leads to minimal misclassification of cause of death, such data can be used to identify the most important health problems in a developing country setting. Second, VA may even provide more dependable data on the overall distribution of cause of death than hospital data, since the latter are unlikely to represent all deaths in the community. It is worth noting that this overall reliability of VA data in relation to adult mortality is not fully replicated in VA data on child mortality. For example, several studies have highlighted problems in distinguishing deaths caused by acute lower respiratory infections , malaria [7,14], and diarrhea .
Studies conducted in settings with moderate to high levels of HIV prevalence suggest that the accuracy of VA data extends to the diagnosis of AIDS-related mortality in settings with high levels of AIDS. Indeed, one South African study finds 100% sensitivity with respect to AIDS . This is in spite of concerns that the secondary infections that are the proximate cause of AIDS deaths may lead to a confusing array of verbal reports about individuals’ symptoms, or that the stigma associated with AIDS or local informants’ fear of it, including local constructions in which its symptoms are treated as the outcome of witchcraft, may have caused them to understate what they think are AIDS-related symptoms.
Generalizing from these studies remains problematic for two reasons, however. First, there are issues of sample selectivity. The study conducted in a setting with high levels of AIDS, for example, used South African data [10,16]; yet the South African population, even in its poorest rural sector, is wealthier, healthier, and more educated than equivalents elsewhere in sub-Saharan Africa . Similarly, a Zambian study used only urban data , and a Tanzanian study had relatively low HIV prevalence .
The second problem is related to temporal changes. Most of these studies use data from the early 1990s (the most recent is that of Boerma et al. , who used data up to 1996). There is at least one reason that their claims about the continued accuracy of VA reporting in high-AIDS settings may no longer hold. Specifically, increases in AIDS-related mortality, AIDS-related public health discourses (by government and non- governmental organization officials, in clinics, schools, newspapers, churches, etc.), and AIDS-related conversations between individuals (as indicated in studies of conversational networks) have been ongoing in most areas of east and southern Africa since the mid-1990s. The present study uses data from one such study and the wider data collected in the study have indicated the high frequency of such conversations. A reasonable cultural response to these increases would be for locals to begin to appraise all mortality through a type of AIDS prism. This would directly affect the accuracy of VA reports since it suggests that, even if AIDS is not mentioned as a cause of death, memories of a deceased individual's symptoms would increasingly filter out symptoms seen to be inconsistent with AIDS, leading ultimately to an overestimate of AIDS in the VA data. In short, there could be a danger that the depiction of symptoms in VA reports will become less accurate as AIDS increases to affect informants’ appraisals of mortality. This study examines whether this appears to have occurred in a rural Malawian setting using data from 1998 to 2001.
Research setting and data
Malawi is a relatively small sub-Saharan Africa country in the east and southern African ‘AIDS-belt’ with a population of almost 11 million, 86% of whom reside in rural areas . It is a poor country, even by African standards: its income per capita is US$190, in comparison to a sub-Saharan African mean of US$480 . HIV/AIDS prevalence among adults is on the order of 16% but rising steadily . As elsewhere in the region, the government of Malawi and foreign donors consider AIDS to present a crisis for Malawi . AIDS in Malawi is also an increasingly frequent focus of social research, much of it directed at understanding the micro-level effects on local cultural (i.e., interpretive) and institutional arrangements in rural settings [21–26].
The mortality data used in this paper were collected as part of the Malawi Diffusion and Ideational Change Project (MDIC), an ongoing longitudinal project involving a sample of almost 1700 ever-married women of reproductive age (15–49) and their husbands (1500) in three areas of Malawi. The overall aim of the MDIC is to examine the role of social networks in changing attitudes and behavior regarding family planning and HIV/AIDS in rural Malawi. The project is based at the University of Pennsylvania's Population Studies Center. (Further details are available at http://www.ssc.upenn. edu/Social_Networks/)
The questionnaire itself asked about symptoms associated with the terminal illness, the relative timing and duration of those symptoms, and the types of medical treatment sought and received. Local interviewers were employed, in accordance with standard methodological recommendations [27,28]. The questionnaire used appropriate local terminology (developed by HVD in pre-fieldwork collaboration with the interviewers) and data collection was supervised by HVD. It is assumed that these methods facilitated the accurate communication and recording of symptoms. The possibility that an interviewer effect could impact on these data, specifically that interviewers may have wittingly or unwittingly directed respondents or miscoded their responses, thereby generating some false positives; this was minimized by instituting thorough training. Its effect was evaluated by estimating intracorrelation coefficients (rho) for the 22 relevant VA variables. In seven variables, rho was < 0.02, the standard threshold for a significant interviewer effect . In an additional seven variables, it was in the modest 0.03–0.06 range, and in the remainder of the variables it was ≥ 0.07. All seven AIDS-related variables (identified below) were in the first two groups, suggesting that there was minimal difference between interviewers in their measurement of informants’ responses insofar as these relate to AIDS (a summary table is available from the authors).
Definition of a death from AIDS
Validation of VA data requires comparison of diagnoses derived from the VA reports with some external standard, usually medical doctors’ assessments or clinical diagnoses . As such information was not available, the external standard for the validation of apparent AIDS cases was an aggregate one: the observed number of deaths identified as ‘AIDS deaths’ in the study was compared with the expected number of such cases.
An ‘AIDS death’ was defined as one where the informants reported either that a death was caused directly by AIDS or the existence of at least two ‘major’ signs and at least one ‘minor’ sign of AIDS. Such an identification of ‘AIDS deaths’ lacks sensitivity to the extent that certain illnesses will be missed. It also lacks specificity to the extent that any non-HIV tuberculosis or cancer will also fit the criteria. It was assumed that these likely compensate for each other.
In past studies, few deaths have been identified in the first way (e.g., 7% in one of the Tanzanian studies , and only 3% in the Lusaka study ). The bulk of AIDS diagnoses are, therefore, based on the more inferential second method.
The specification of ‘major’ and ‘minor’ signs of AIDS is based on the World Health Organization (WHO) classification of AIDS-related symptoms . It draws on WHO guidelines for the provisional clinical case definition for AIDS where diagnostic resources are limited  and is used by the Malawi National AIDS Control Program .
Although the lack of specific medical information means that there are a number of conditions associated with AIDS for which data are not available, including those that are sufficient in themselves to diagnose AIDS (generalized Kaposi's sarcoma or cryptococcal meningitis) and minor signs such as recurrent herpes zoster, oropharyngeal candidiasis, chronic herpes simples infection, and generalized lympadenopathy, the definition used is consistent with ones used in previous studies [31,33]. The major signs for which data was collected are chronic diarrhea for more than 1 month, prolonged fever (intermittent or constant) for more than 1 month, and weight loss of more than 10% of body weight. Since few of these rural informants (nor the deceased themselves) would be likely to know about actual changes in weight, the questionnaire asked about the duration of weight loss rather than its extent, assuming that weight loss for more than 1 month is associated with at least 10% of body weight . The minor signs were prolonged cough, prolonged difficulty breathing, prolonged pneumonia, and ‘rash’ (considered to be an indicator of generalized pruritic dermatitis if it occurred in combination with two major signs).
The expected number of AIDS cases was calculated as the ‘excess mortality factor’ derived from a comparison of adult mortality in this sample with the pre-AIDS mortality levels, as measured in the 1987 Malawi census data . Mortality in this adult sample was greater than mortality calculated from 1987 Malawian census data for the same age groups by a factor of 3.0 (95% confidence interval, 2.1–3.8) . This increase is consistent, it should be noted, with those reported in neighboring Zimbabwe over the same period .
Because the census data covering 1978–1987 largely predate the effects of AIDS-related mortality, and there have been no reports of systematic increases in other non-AIDS-related causes of mortality for adults over the same period (i.e., not including secondary infections facilitated by AIDS), it can be reasonably assumed that this increased mortality reflects the impact of AIDS. Translating the 95% confidence interval into an expected distribution of AIDS and non-AIDS mortality indicated that 68.3–79.3% of all adult deaths would be associated with AIDS (mean, 74.9%).
Validity of data collection
Of the base sample of almost 1700 ever-married women of reproductive age (15–49) and their husbands (1500) in three areas of Malawi, 1554 women and 1126 men were interviewed in a first round of data collection in 1998; 104 of these (54 women, 50 men) were reported to have died by the second round in 2001. A good indicator that these reports were accurate is the fact that a related research project drawing on half the sample in 1999 reported 41 deaths and of these all but one were also reported as deceased in 2001. This single ‘resurrection’ may have been related to the project's gifting strategy: that is, to help to maintain the goodwill of respondents across this and future rounds, respondents were given small quantities of sugar and soap.
In 92 of the 104 reported deaths, interviewers were able to identify close relatives or neighbors and administer a VA questionnaire. In 71 of these (77%), the informant was a close relative such as a spouse or sibling; in the remaining 21, the informant was a neighbor. There is no statistical difference in any aspect of the VA report between these two sources. Nor is there any statistically significant difference in sociodemographic characteristics (as measured in 1998) between the 92 individuals with VA data and the 12 without.
Reported VA symptoms are shown in Table 1 (some of the non-AIDS-related symptoms are grouped together). Consistent with studies mentioned above, a few cases (10%) were directly reported as being caused by AIDS. Larger percentages of the deceased were said to have had at least one major or minor sign of AIDS. For example, 54% had prolonged weight loss, 43% a prolonged cough, and more than a quarter prolonged fever or malaria, chronic diarrhea, or prolonged pneumonia. A further 23% had prolonged difficult/rapid breathing and 10% a long-term rash.
Using the WHO algorithm, 75.5% of these cases appear to be AIDS related. This value is almost exactly in the middle of the confidence interval specified by the comparison between observed and expected mortality. There were no significant gender differences in specific symptoms reported and this is reflected in the lack of differences in the overall AIDS-related deaths calculated for men and women: 75.1% of females and 76.3% of males are reported to have died of AIDS. This is equivalent to a death rate of 0.042 for females (20–49 years) and 0.050 for males (20–59 years).
In this study, AIDS adult mortality was estimated based on VA. The study wished to assess whether the increasing scale of the AIDS epidemic in rural areas of sub-Saharan Africa, and its preeminence in local public health discourses and in informal conversations about health, might affect informants’ appraisals of the causes of adult mortality, thereby affecting the reliability of VA as an assessment of AIDS-related mortality. The VA data on deaths collected in 1998–2001 was, therefore, compared with a mortality factor calculated independently to describe the increase in adult mortality in Malawi since the onset of AIDS. The comparison shows that there is no apparent effect of high AIDS-related mortality on the accuracy of the VA in the aggregate. The mortality factor predicted that 75% of the deaths would be AIDS related; 76% of the VA deaths showed consistent signs of AIDS as identified by the WHO criteria, despite the partial lack of sensitivity and specificity in the definition of AIDS-related deaths.
This study, therefore, confirms studies conducted in other areas of Africa, albeit in more educated populations and/or at earlier stages of the epidemic, in asserting that VA remain a reliable data collection tool for differentiating, at the aggregate level, between AIDS and non-AIDS deaths.
1. Feachem RG, Jamison DT, Bos ER. Changing patterns of disease and mortality in sub-Saharan Africa.
In: Disease and Mortality in sub-Saharan Africa.
Edited by Feachem RG, Jamison DT. World Bank: The University Press; 1991:3–27.
2. van de Walle E, Pison G, Mpembele S. Introduction.
In: Mortality and Society in Sub-Saharan Africa.
Edited by van de Walle E, Gilles P, Mpembele S. Oxford: IUSSP/Clarendon Press; 1992: 1–9.
3. Timaeus I. Adult mortality.
In: Demographic Change in Sub-Saharan Africa
. Washington: National Academy Press; 1993:218–255.
4. Kaufman JS, Asuzu MC, Rotimi CN, Johnson OO, Owoaje EE, Cooper RS. The absence of adult mortality for sub-Saharan Africa: a practical solution.Bull WHO
5. Grubb GS, Fortney JA, Saleh A, Gadalla S, el-Baz A, Feldblum P, et al
. A comparison of two cause-of-death classification systems from deaths among women of reproductive age in Menoufia, Egypt.Int J Epidemiol
6. Garenne M, Fontaine O. Assessing probable causes of deaths using a standardized questionnaire. A study in rural Senegal.
In: Measurement and analysis of mortality: New approaches.
Edited by Vallin J, D'Souza S, Palloni, A. Oxford: Clarendon Press; 1990:123–142.
7. Snow RW, Armstrong JR, Forster D, Winstanley MT, Marsh VM, Newtonet CR, et al
. Childhood deaths in Africa: uses and limitations of verbal autopsies.Lancet
8. Chandramohan D, Maude G, Rodrigues L, Hayes R. Verbal autopsies for adult deaths: issues in their development and validation.Int J Epidemiol
9. Reeves BC, Quigley M. A review of data-derived methods for assigning causes of death from verbal autopsy.Int J Epidemiol
10. Garenne M, Kahn K, Tollman S, Gear J. Causes of death in a rural area of South Africa: an international perspective.J Trop Pediatr
11. Chandramohan D, Rodrigues L, Maude GH, Hayes R. The validity of verbal autopsies for assessing the cause of institutional maternal death.Stud Fam Plann
12. Pickering H, Nunn AJ. A three-year follow-up survey of demographic changes in a Ugandan town on the trans-African highway with high HIV-1 seroprevalence.Health Transit Rev
13. Kalter HD, Gray RH, Black RE, Gultiano SA. Validation of post-mortem interviews to ascertain selected causes of death in children.Int J Epidemiol
14. Todd JE, de Francisco A, O'dempsey TJD, Greenwood BM. The limitations of verbal autopsy in a malaria-endemic region.Ann Trop Pediatr
15. Mobley CC, Boerma JT, Titus S, Lohrke B, Shangula K, Black RE. Validation study of a verbal autopsy method for causes of childhood mortality in Namibia.J Trop Pediatr
16. Kahn K, Tollman SM, Garenne M, Gear JSS. Validation and application of verbal autopsies in a rural area of South Africa.Trop Med Int Health
18. Kelly P, Feldman R, Ndubani P, Baboo KS, Timaeus I, Farthing MJ, et al
. High adult mortality in Lusaka.Lancet
19. Boerma JT, Ngalula J, Isingo R, Urassa M, Senkoro KP, Gabone R, et al
. Levels and causes of adult mortality in rural Tanzania with special reference to HIV/AIDS.Health Transit Rev
20. Joint United Nations Programme on HIV/AIDS
[UNAIDS]. AIDS Epidemic Update, December 2001.http://www.unaids.org
, 2001, retrieved 19 April 2002.
21. Government of Malawi
and World Bank. Malawi AIDS Assessment Study.
Blantyre and Washington, DC: Government of Malawi
and World Bank; 1998.
22. Forster PG. Religion, magic, witchcraft, and AIDS in Malawi.Anthropos
23. Mtika MM. Social and cultural relations in economic action: the embeddedness of food security in rural Malawi amidst the AIDS epidemic.Environ Plann
24. Mtika MM. The AIDS epidemic in Malawi and its threat to household food security.Hum Organiz
25. Weinreb AA. Lateral and vertical intergenerational exchange in rural Malawi.J Cross-Cult Gerontol
26. Weinreb AA. Substitution and substitutability: the effects of kin availability on intergenerational transfers in Malawi.
In: The Distribution of Private and Public Resources Across Generations.
Edited by Gauthier AH, Chu C, Tuljapurkar S. Oxford: Oxford University Press, 2004; in press.
27. Snow B, Marsh K. How useful are verbal autopsies to estimate childhood causes of death?Health Policy Plann
28. Kahn K, Tollman SM, Garenne M, Gear JSS. Who dies from what? Determining cause of death in South Africa's rural north-east.Trop Med Int Health
29. Fowler FJ, Mangione TW. Standardized Survey Interviewing: Minimizing Interviewer-Related Error.
London: Sage; 1990.
30. Kalter H. The validation of interviews for estimating morbidity.Health Policy Plann
31. Todd J, Balira R, Grosskurth H, Mayaud P, Mosha F, ka-Gina G, et al
. HIV-associated adult mortality in a rural Tanzanian population.AIDS
33. Quinn TC, Mann JM, Curran JW, Piot PP. AIDS in Africa: an epidemiological paradigm.Science
:955–963 [reprinted in Bull WHO
Control Programme. Home Based Care Training Manual
. Lilongwe: Ministry of Health and Population, AIDS
35. Doctor HV, Weinreb AA. Mortality among Married Men in Rural Kenya and Malawi.
Philadelphia, PA: University of Pennsylvania Population Studies Center; 2002.
36. Feeney G. The impact of HIV/AIDS on adult mortality in Zimbabwe.Pop Dev Rev