Mortality levels and trends by HIV serostatus in rural South Africa

Objective: To examine mortality differentials in HIV-infected and uninfected adults by demographic characteristics and the effect of non-testing on the level and pattern of age-sex specific mortality.

Methods: Three annual prospective population-based HIV surveys between 2003 and 2006 provide information regarding individual adult HIV status; households were visited twice a year to collect information about births, deaths, migrations and other demographic, health and socioeconomic data. Deaths and person-years of exposure were aggregated for each calendar year between 2004 and 2006, from which mortality rates were derived. The association between risk factors and mortality was assessed using a Cox proportional hazards model.

Results: The observed rate of mortality in individuals who did not consent to HIV testing was four to seven times higher, and that in HIV-infected adults 11-19 times higher than mortality in HIV-negative individuals. After adjusting for age, sex and socioeconomic status, HIV-infected individuals had a ninefold greater hazard of dying than uninfected individuals. Mortality rates increased with age and peak in the 45-54 years age group, irrespective of HIV status. Multivariably, age and sex were significantly associated with the hazard of dying, but place of residency and socioeconomic status were not. Overall mortality declined from 71 to 48 deaths per 1000 person-years between 2005 and 2006.

Conclusion: The substantial decline in mortality after 2004 is likely to be largely attributable to the increasing availability of antiretroviral therapy. Detailed investigation of the characteristics of the not-tested individuals is needed to understand their impact on mortality patterns.

Introduction

Now in its second decade, the southern Africa HIV epidemic has been rapid and severe. An estimated 5.5 million South Africans are currently infected with HIV ^[1,2], although results from national antenatal surveillance and population-based surveys suggest that HIV prevalence may have started to level off at approximately 18.8% in 2005 (15-49 years) ^[3,4].

In KwaZulu Natal, the province with the highest HIV prevalence in South Africa, the Africa Centre for Health and Population Studies has conducted three population-based HIV surveys. In the first HIV survey in 2003/2004, HIV prevalence was 22% (27% in female and 14% in male residents aged 15-49 years), highest in women aged 25-29 years (51%) and men aged 30-34 years (44%) ^[5]. Between 2003/2004 and 2005, age-adjusted incidence rates per 100 person-years were 2.6 [95% confidence interval (CI) 1.8-3.3] and 4.0 (95% CI 3.3-4.7), respectively in men and women aged 15-49 years ^[6].

Mortality in the study population over the period 1984/1988-2001 has mirrored the spread of the HIV epidemic. After a slow rise in the early 1990s, mortality increased sharply in the second half of the 1990s, with the probability of dying for an adult aged between 30 and 65 years increasing from approximately 35 to 60% for women and from approximately 50 to 78% for men ^[7]. As the epidemic matures, in the absence of effective treatment and care, further increases in mortality are expected.

In this paper we explore mortality patterns in HIV-infected and not-infected individuals by demographic characteristics. In all population-based HIV surveys, HIV status will be unknown for a proportion of eligible individuals either because they could not be contacted or they refused to provide a sample for testing. HIV test refusal has been shown to affect estimates of HIV prevalence with self-selection operating in both directions depending on the study setting and group ^[8,9]. Across the three Africa Centre HIV surveys (2003/2004, 2005, 2006), on average only 5% of the eligible resident population could not be contacted. Among those contacted, however, the proportion refusing to participate in the HIV surveillance ranges from approximately 40% in the first to 58% in the third HIV survey. The strength of our population-based HIV data from the individual surveillance lies in the independent availability of detailed socioeconomic and demographic data from the household surveillance, allowing the comparison of the characteristics and mortality among participants and non-participants in the HIV surveillance. We here examine mortality differentials among adults by availability of test result and HIV status.

Methods

Data sources

The Africa Centre Demographic Information System (ACDIS) is a longitudinal demographic surveillance system conducted in the rural district of Umkhanyakude in northern KwaZulu Natal, South Africa (http://www.africacentre.ac.za). Since January 2000, all resident and non-resident members of approximately 11 000 households have been enumerated (total population 86 000 in 2003); the area is characterized by high levels of migration. Households are visited two times a year to collect information about births, deaths, migrations and other demographic, health and socioeconomic data. Detailed descriptions of ACDIS have been published elsewhere ^[10,11]. In ACDIS, an area with less than 400 residents per square kilometre was defined as rural; any area with more than 400 residents per square kilometre was defined as peri-urban ^[12,13], provided it was not formally defined as an urban area by the registrar general of the municipality. Socioeconomic status was defined using a wealth index measure determined by the number of assets owned by a household (C. Ardington, personal communication, 2007). On the basis of this index, households were divided into three equal categories classified as low, medium and high socioeconomic status.

In addition to the routine household surveillance, three rounds of a prospective population-based HIV survey have been conducted in the same study population (2003/2004, 2005 and 2006) ^[5,14]. All women aged 15-49 years and men aged 15-54 years who were resident household members of the surveillance area were eligible to participate in each HIV survey round. Fieldworkers visited eligible individuals at home. After written informed consent, blood was collected by finger prick and dried blood spots prepared. HIV status was determined by antibody testing with a broad-based HIV-1/HIV-2 enzyme-linked immunosorbent assay (ELISA; Vironostika; Organon Teknika, Boxtel, the Netherlands) followed by a confirmatory ELISA (GAC-ELISA; Abbott, Abbott Park, Illinois, USA). Test results are made available to participants at local counselling points run by the Africa Centre. Ethics approval for the collection and presentation of the demographic, socioeconomic and HIV serostatus data used in this study was granted by the Research Ethics Committee of the University of KwaZulu Natal.

Study sample

Our sample included all individuals who met the following criteria: age eligible for inclusion in the HIV serosurvey during the first, second or third round (women aged 15-49 and men aged 15-54 years), and resident in the surveillance area at the time of drawing up the eligibility list for each round. Data from these three surveys were combined into one sample for this analysis. The median age of the sample was 27.9 years (range 15-54). The sample had individuals surveyed only once (n = 12 662), surveyed twice (n = 16 577) and those in all three surveys (n = 10 538).

Deaths and person-years of follow-up were aggregated for each calendar year between 2004 and 2006. Exposure for individuals taking part in the first HIV survey (June 2003 to December 2004) was left truncated to 1 January 2004 in order to achieve consistency with the later survey rounds, which were conducted within a calendar year. Person-years of follow-up were calculated from 1 January 2004, or a date within this period for individuals joining a household or ageing in to the eligible sample. The exposure period was censored on 31 December 2006, or ended by death or outmigration before this date. We stratify our analyses by HIV status (HIV negative, HIV positive and not tested), because factors influencing mortality are likely to be different among these HIV categories. In the model, interactions between the risk factors on the hazard of dying were assessed and accepted as statistically significant at P = 0.05.

Analytical approach

The association between mortality and age and sex by HIV status was assessed using a Cox proportional hazards model ^[15], appropriate given the longitudinal nature of our data. The possible confounders examined were socioeconomic status, place of residency and HIV survey round. Variables that were not statistically significantly associated with mortality in any of the three HIV categories were not included in the final model. All covariates inputted into the model were categorical variables, including age, which was categorised into 10-year age groups (note that the age group 45-54 years contains more men than women as HIV surveillance stopped at 50 years for women and 55 years for men). The model was adjusted for HIV serosurvey round to account for differences in the population surveyed at each round. The log likelihood ratio test was used to assess the improvement of goodness of fit of the model.

Results

Of the 14 108 resident population in the first HIV survey meeting the inclusion eligibility criteria, 60% (n = 8470) agreed to HIV testing. The majority, 77% (n = 6532), of individuals tested were HIV negative, 21% (n = 1773) tested HIV positive, whereas less than 1% (n = 24) had an indeterminate result. The HIV test result was missing for seven individuals with an HIV dried blood spot sample and for an additional 134 individuals who had initially agreed to the HIV survey, but who could not be tested for HIV because they were too sick or because they moved too far away for tracking. These individuals were excluded from the analysis. The decision to censor this survey round at 1 January 2004 meant that an additional 1290 (596 HIV negative, 136 HIV positive, and 558 not tested) individuals were excluded from the analyses because they were visited before 2004. The number of individuals from the first survey included in the analyses was therefore 12 653 (Table 1).

Table 1

In the second round, 22 349 resident members were eligible for inclusion. Among the eligible residents less than 2% were excluded from the analyses because they were too sick to be tested (n = 67), had outmigrated and could not be tracked (n = 55), or could not be contacted (n = 290). Of the remaining eligible residents in the sample, 41% (n = 8909) agreed to the HIV survey and 59% (n = 13 028) refused. Among the population consenting to the HIV survey, 79% (n = 7055) tested HIV negative; 20% (n = 1818) HIV positive and less than 1% (n = 28) had an indeterminate result. The HIV status result was missing for eight individuals who had consented to the HIV survey. Individuals with indeterminate and missing results were also excluded from the analyses. Therefore, from the eligible individuals in the second survey, 21 901 were included in the analyses (Table 1).

In the third HIV survey, 23 428 resident individuals were eligible for inclusion, out of which 9% (2041) could not be contacted. Of the 21 387 contacted, 38% (n = 8136) agreed to the HIV survey, whereas 56% (n = 13 251) refused to be surveyed. Among those consenting, 78% (n = 6362) tested negative and 20% (n = 1601) tested positive; whereas less than 1% (n = 26) with an indeterminate HIV result and 2% (n = 147) with missing HIV result information after consenting were excluded from the analyses. 21 214 individuals in HIV survey round three were thus included in the analyses (Table 1).

The characteristics of the individuals included in the analyses, counting them only once with their latest HIV statuses, are presented in Table 1. The median age of HIV-positive individuals is considerably higher than for HIV-negative individuals, which is reflective of the dominance of young people among the HIV-negative group. Overall, 56.1% (16 106) of adults contacted in the HIV surveillance were women, but this was as high as 73% in the HIV-infected group, and 54% in the uninfected group were women. Nearly 60% of the non-testers lived in a rural area compared with nearly two-thirds of the uninfected and just over half of the infected individuals. More individuals among the HIV-negative group are of low socioeconomic status, whereas among the HIV-positive group most individuals are in the middle category.

Among the 15-54-year-old men and 15-49-year-old women contacted for the HIV surveillance, 348 died in 2004 (196 women, 152 men); 530 died in 2005 (284 women, 246 men); and 374 died in 2006 (187 women, 187 men). The total person-years of follow-up for the 3-year period 2004-2006 was 69 330 years. Figure 1 presents the observed mortality rates by HIV status over the period 2004-2006 in panel (a) and directly age-standardised mortality rates in panel (b). The population distribution from the three HIV surveys combining individuals consenting and not consenting to HIV testing was used as the standard in this analysis. Age was recorded at the date of drawing up the eligibility list of each sample. Mortality was always highest among HIV-infected individuals, and 11-19 times higher than in HIV-negative individuals. The observed rate of mortality in individuals who did not consent to HIV testing was consistently higher (four to seven times) than in HIV-uninfected individuals. These patterns of mortality remained after adjusting for age distribution (panel b). Mortality also increased with age, peaking at ages 45-54 years in all 3 years. Mortality rates were also generally higher among men than women in all HIV and testing status categories (Table 2).

Equation (Uncited)

Fig. 1

Table 2

Table 3 presents the results of univariable and multivariable analyses stratified by HIV status. Among HIV-negative adults, those aged over 35 years were three to seven times more likely to die than 15-24 year olds. Men were significantly more likely to die than women, both among HIV-negative and HIV-positive individuals, but not in the not-tested group. After adjustment for age and sex, living in a rural rather than an urban area was statistically significantly associated with the risk of mortality only among the not-tested group.

Equation (Uncited)

With the exception of the HIV-negative population, the adjusted hazard of dying differed depending on the year in which an individual was first seen. We explored the potential influence of interactions among age, sex, HIV survey round, socioeconomic status and place of residency. None of the interactions explored were, however, statistically significant and were thus not included in the final model. In multivariable analysis, adults who tested positive for HIV infection were nearly nine times more likely to die than those who tested negative; whereas those not tested were at a fourfold increased risk compared with uninfected individuals (Table 3).

Equation (Uncited)

Discussion

HIV prevalence and incidence rates remain very high in our study area ^[6,14], with a peak incidence in the late twenties. Our results show a generally steady increase in the pattern of mortality with age. HIV-infected individuals were nearly nine times more likely to die than uninfected individuals, after adjusting for age, sex and socioeconomic status. The pattern of mortality with age among HIV-positive individuals we observe is consistent with observations from similar demographic surveillance sites in Africa ^[16], and in line with the average delay of approximately 10 years between infection and death in infected individuals ^[17,18]. An earlier study in northern Malawi on the long-term survival by HIV status also showed a general increase in mortality rates with increasing age ^[19].

Table 3

The declining mortality levels, especially among the infected group and to a lesser extent among the non-tested group, in 2005 and 2006 compared with 2004 are likely to be indicative of a response to the start of the district's public antiretroviral treatment (ART) programme in late 2004. Findings presented elsewhere provide early evidence of an impact of ART on mortality in this population ^[20]. Since the start of the ART programme in late 2004, approximately 1000 individuals had been initiated on treatment by the end of 2005, and by the end of 2006 over 1800 individuals were receiving treatment (G. Cooke, personal communication, 2007). Alternative explanations for the declining trends in mortality since 2004 include differences in the duration of infection in the populations included in each round, respondent bias, and non-AIDS mortality among HIV-positive individuals. Increases in HIV incidence rates are unlikely to explain the declining mortality given the steady high incidence in the surveillance area ^[6]. If ART is already having an impact on mortality levels and trends, it may partly explain the wide disparity in mortality between women and men given that in our area more women than men are receiving treatment. Approximately three-quarters of the individuals enrolled for treatment are women. On the other hand, even before the advent of ART, mortality was reported to be higher among HIV-infected men than HIV-infected women, especially in the younger age groups ^[21,22], and therefore may be more indicative of the differentials in socioeconomic status between men and women and the higher risk of violent deaths for young men in this population.

The individuals not tested in our surveillance are concentrated in the high socioeconomic group. As access to voluntary counselling and testing at government clinics is freely available coupled with this group having the means to test elsewhere, the high refusal rates could indicate that these individuals already knew their HIV status and thus opt not to re-test. Contrary to expectation, however, we did not find a statistically significant difference in the hazard of dying among those in the medium or high socioeconomic group relative to those in the lowest socioeconomic group, except among the not-tested population. The poorest individuals in this population have been shown to have a lower HIV incidence rate than the middle wealth group ^[5], and therefore this may be one possible reason for the absence of a significant mortality-poverty differential in this study. Once infected, poverty can often be an important barrier to accessing and sustaining timely and effective medical treatment, adequate nutrition, and palliative care. These may, therefore, have been the external factors modifying the effect of socioeconomic status on the hazard of dying. Investigating the interactions of HIV acquisition risk, socioeconomic status and access to treatment, although beyond the scope of this paper, will be important in modelling these interrelated risks. The lack of a definitive statistical significance of area of residency on the hazard of dying for both the HIV-negative and HIV-positive population could be explained by the fact that our surveillance area is predominantly rural, and that HIV infection status information is less readily available for the more urban population.

In individuals younger than 35 years, mortality rates for those of unknown HIV status are closest to those who were HIV positive, whereas above the age of 35 years their mortality rates tended to be closer to those in HIV-negative individuals. This is likely to reflect the high testing refusal rate among the higher risk younger age group in the survey. In this study, we do not contrast cause-specific mortality rates and can therefore only infer that much of the higher mortality in untested individuals may be attributable to HIV. It may also be indicative of the shorter duration since infection, and as a result the lower hazard of dying, among young HIV-infected individuals compared with older infected individuals, and the relative high risk of dying of non-HIV causes such as road deaths. Nonetheless, the findings highlight the importance of considering the possible effect of testing bias in mortality estimates derived from population-based surveys. In addition, it gives impetus to efforts to increase contact rates and encourage testing in future rounds of the HIV survey.

Household and HIV surveys in this study area are ongoing, and the ART programme is becoming an established part of health services in the surveillance area. This change in treatment provision will provide further opportunities to monitor the changes in mortality in a late-stage epidemic, as well as changes in the mortality patterns of infected and uninfected adults. The public health message from our analyses is that mortality is significantly higher among HIV-infected individuals than HIV-uninfected individuals; however, as young people form a substantial part of the uninfected population, the impetus remains for improving HIV prevention programmes targeted at young, uninfected individuals.

Acknowledgements

The authors would like to thank the community members in the demographic surveillance area for being ever so willing to provide data to the study teams since 2000. The authors pay tribute to the operations staff for their tireless efforts in gathering the data used in this paper. Special mention should also go to the data quality team for helping to ensure that the data are of high quality.

Sponsorship: The research reported in this paper was supported by the Wellcome Trust through grants GR065377/Z/01/H for the Africa Centre Demographic Information System and GR065377/Z/01/B for HIV surveillance.

Conflicts of interest: None.