SEARCH

SEARCH BY CITATION

Keywords:

  • self-assessments of health;
  • HIV/AIDS;
  • mortality;
  • South Africa;
  • demographic surveillance
  • auto-évaluation de la santé;
  • VIH/SIDA;
  • mortalité;
  • Afrique du Sud;
  • surveillance démographique
  • autoevaluación de salud;
  • VIH/SIDA;
  • Mortalidad;
  • Sudáfrica;
  • vigilancia demográfica

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References

Objectives  While self-assessments of health (SAH) are widely employed in epidemiological studies, most of the evidence on the power of SAH to predict future mortality originates in the developed world. With the HIV pandemic affecting largely prime age individuals, the strong association between SAH and mortality derived from previous work might not be relevant for the younger at-risk groups in countries with high HIV prevalence in the era of antiretroviral treatment. We investigate the power of SAH to predict mortality in a community with high HIV prevalence and antiretroviral treatment (ART) coverage using linked data from three sources: a longitudinal demographic surveillance, one of Africa’s largest, longitudinal, population-based HIV surveillances, and a decentralised rural HIV treatment and care programme.

Methods  We used a Cox proportional hazards specification to examine whether SAH significantly predicts mortality hazard in a sample composed of 9217 adults aged 15–54, who were followed up for mortality for 8 years.

Results  Self-assessments of health strongly predicted mortality (within 4 years of follow-up), with a clear gradient of the adjusted hazard ratios (aHRs), relative to the baseline of ‘excellent’ self-assessed health status and controlling for age, gender, marital status, the socio-economic status (SES), variables education, employment, household expenditures and household assets, and HIV status and ART uptake: 1.40 (95% CI 0.99–1.96) for ‘very good’ self-assessed health status (SAHS); 2.10 (95% CI 1.52–2.90) for ‘good’ SAHS; 3.12 (95% CI 2.18–4.45) for ‘fair’ SAHS; and 4.64 (95% CI 2.93–7.35) for ‘poor’ SAHS. While a similar association remained in the unadjusted analysis of long-term mortality (within 4–8 years of follow-up) the hazard ratios capturing SAH are jointly insignificant in predicting of mortality once HIV status, ART uptake and gender, age, marital status and SES were controlled for. HIV status and ART programme participation were large and highly significant predictors of long-term mortality.

Conclusions  Our findings validate SAH as a variable that significantly predicts short-term mortality in a community in sub-Saharan Africa with high HIV prevalence, morbidity and mortality. When predicting long-term mortality, however, it is much more important to know a person's HIV status and ART programme participation than SAH.

Objectifs:  Bien que les auto-évaluations de la santé (AES) soient largement utilisées dans les études épidémiologiques, la plupart des données sur le pouvoir des AES à prédire la mortalité future proviennent des pays développés. Avec la pandémie du VIH affectant largement les individus à la fleur de l’âge, la forte association entre les AES et la mortalité provenant de travaux antérieurs pourrait ne pas être relevante pour les groupes à risque plus jeunes dans les pays à forte prévalence du VIH à l’ère du traitement ARV. Nous avons étudié le pouvoir des AES à prédire la mortalité dans une communauté avec une prévalence élevée du VIH et une couverture étendue des traitements antirétroviraux (ART), en utilisant les données d’une surveillance démographique longitudinale, une des plus vastes surveillances longitudinales du VIH basées sur la population en Afrique, et avons relié les données avec celles d’un système décentralisé d’un programme rural de traitement et des soins du VIH.

Méthodes:  Nous avons utilisé la spécification des risques proportionnels de Cox pour examiner si l’AES prédit de façon significative le risque de mortalité dans un échantillon composé de 9217 adultes âgés de 15 à 54 ans qui ont été suivis pour la mortalité pendant 8 ans.

Résultats:  L’AES prédit fortement la mortalité dans les 4 ans de suivi (RR ajusté: 4,6; IC95%: 2,09 - 7,3), même après que des ajustements pour le statut socioéconomique (SSE), le statut VIH et la participation à un programme d’ART ont été inclus dans le modèle. Alors qu’une association similaire demeure dans l’analyse non ajustée 8 ans après les données de référence, cette association disparaît une fois que des ajustements pour le statut VIH, la participation au programme ART et le SSE sont effectués dans ce modèle (RR ajusté: 1,7; IC95%: 0,9 - 3,0). Les résultats restent robustes à la ventilation selon le statut VIH et le sexe.

Conclusions:  Nos résultats valident l’AES comme une variable permettant de prédire à court terme la mortalité future dans une communauté en Afrique subsaharienne, avec une prévalence, une morbidité et une mortalité du VIH élevées. L’infection par le VIH n’affecte pas l’association entre l’AES et la mortalitéà court terme, bien que le statut VIH prédise mieux la mortalitéà long terme que l’AES.

Objetivos:  Mientras que la autoevaluación de salud (AES) es ampliamente utilizada en epidemiología, la mayor parte de la evidencia sobre el poder de la AES a la hora de predecir la mortalidad futura proviene del mundo desarrollado. Con la pandemia del VIH afectando principalmente a individuos en edad productiva, la fuerte asociación entre AES y mortalidad derivada de trabajos anteriores podría no ser relevante para los grupos de riesgo más jóvenes en países con una alta prevalencia de VIH en la era del tratamiento antirretroviral. Hemos investigado el poder de la AES para predecir la mortalidad en una comunidad con una alta prevalencia de VIH y cobertura de tratamiento antirretroviral (TAR), utilizando datos de un estudio de vigilancia demográfica longitudinal, uno de los estudios para VIH de vigilancia demográfica longitudinal y basado en la población más grande de África, y datos vinculados de un programa rural descentralizado para el tratamiento y cuidados del VIH.

Métodos:  Utilizamos un modelo de Cox de riesgos proporcionales para examinar si la AES predecía de forma significativa el riesgo de mortalidad en una muestra compuesta por 9217 adultos con edades comprendidas entre los 15–54 a los que se les realizó un seguimiento durante 8 años.

Resultados:  La AES predice la mortalidad tras 4 años de seguimiento (HR ajustado 4.6; 95% IC 2.9–7.3), incluso después de introducir en el modelo controles para el estatus socioeconómico (ESE), el estatus de VIH y la participación en programas de TAR. Mientras que 8 años después del inicio se mantiene una asociación similar en el análisis sin ajustar, desaparece en este modelo cuando se controla para estatus de VIH, la participación en un programa de TAR y el ESE (HR ajustado 1.7; 95% CI 0.9–3.0). Los resultados son robustos a disgregación por estatus de VIH y género.

Conclusiones:  Nuestros hallazgos validan el AES como una variable que predice la mortalidad futura a corto plazo en una comunidad africana con una alta prevalencia, morbilidad y mortalidad por VIH. La infección con VIH no afecta la relación entre AES y la mortalidad futura a corto plazo, aunque el estatus de VIH es un mejor vaticinador de la mortalidad futura a largo plazo que la AES.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References

Epidemiological studies conducted over a span of 30 years have consistently reported a negative association between self-assessments of health (SAH) and all-cause mortality. A global assessment to rate one’s health at a given time on a four or five-point scale that typically runs from excellent to poor has been shown to be a robust predictor of the hazards of death in most studies, even after controlling for current objective measures of health (Bowling 2005; Idler & Benyamini 1997; Idler & Kasl 1995).

Most of the evidence on the power of SAH to predict mortality originates from studies in the developed world. Idler and Benyamini (1997, 1999) review community studies on the association between SAH and mortality published in English up to 1999. They find only one article investigating this association in a developing country, for the case of China, where people reporting ‘poor’ or ‘fair’ self-assessed health status (SAHS) were about twice as likely to die in the next 5 years as those reporting ‘good’ or ‘excellent’ SAHS (Yu et al. 1998).

In our own search of more recent literature we identified one other study from a developing country. This 2004 study using the Indonesia Family Life Survey finds that individuals who perceived their health to be ‘poor’ were significantly more likely to die in subsequent periods than their counterparts who viewed their health as ‘good’ (Frankenberg & Jones 2004).

The robust association between SAH and other health-related events observed in developed countries may not be transferable to the developing world. First, anchor points or expectations of health may differ systematically between developed and developing countries (Salomon et al. 2004). Second, less access to information about health in developing countries might lead to less accurate SAH. Third, causes of mortality, availability of effective health services, and age of death differ substantially between developed and developing countries, and these differences could plausibly affect the way people evaluate their own health status. Particularly important examples of this last point are the strong effect of the HIV epidemic on mortality and the impact of the recent antiretroviral treatment (ART) scale-up on HIV-related mortality in many countries in sub-Saharan Africa. This paper provides evidence on the relationship between SAH, HIV-status, ART uptake and mortality for the case of South Africa, a country with low life expectancy, where 18% of the 15–49 year-old population is infected with HIV (UNAIDS 2008) and ART coverage is high (Cleary & McIntyre 2010).

Validating the predictors of future mortality, such as SAH, in resource-limited settings is important for health policy and planning. With HIV surveillance becoming more widespread in Sub-Saharan Africa (Asamoah-Odei et al. 2004), SAH might not be as useful as a screening tool if it does not provide additional information to that contained in HIV-test results. In the South African case, for instance, asymptomatic HIV-infected individuals might see themselves as unhealthy simply because their expectations regarding health status are influenced by the surrounding high risk of disease and death. In this case the true associations between SAH and mortality may be confounded by HIV status and ART uptake. In this study, we test whether SAH predict mortality in rural South Africa after HIV status and ART uptake are controlled for.

Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References

Setting

The data used in this analysis come from the Africa Centre for Health and Population Studies, which has collected information on approximately 11 000 households and 90 000 individuals since 2000 in a demographic surveillance area (DSA) in KwaZulu-Natal, South Africa. This community has high overall levels of HIV prevalence in adults (28% in 2010) and HIV incidence (3.4 per 100 person-years from 2003 through 2007) (Bärnighausen et al. 2009), but substantial geographic variation in HIV prevalence (Tanser et al. 2009) and incidence across sub-communities (Bärnighausen et al. 2010). As of 2009, life expectancy was 59 years for women and 50 for men (Herbst et al. 2011).

The DSA is located in the catchment area of a decentralised, public-sector ART programme launched in August 2004 (Houlihan et al. 2010). By the end of 2010, ART coverage in the DSA had risen to about 75% under CD4 count eligibility threshold of CD4 < 200/μl stipulated by the South African national treatment guidelines at the time (Malaza et al. 2011).

Data

At 6-month intervals, a fieldworker team visited every household in the DSA, and data on vital events for all household members were collected. In each round, four attempts were made to find a suitable informant in the household. If by the fourth visit to a household no household member could be contacted, the case was referred to a tracking team that makes three further attempts at contacting a member of this household. On average, 99.5% of all households participated in the biannual surveillance rounds. This high participation rate and the ongoing nature of the surveillance ensure high quality mortality data and reduce the likelihood that any death in the DSA is missed (Herbst et al. 2009). All deaths that were recorded by the surveillance team were followed up by a verbal autopsy interview conducted on average 6 months after the person’s death (Soleman et al. 2006).

In 2003, a population-based HIV surveillance was launched in the DSA. The HIV surveillance contains questions on SAH, fertility history, sexual behaviour and offers testing for HIV. Tanser et al. (2008) describe the eligibility and participation rates in the HIV surveillance: (See Tanser et al. (2008) for details on the Africa Centre site, eligibility criteria and surveillance protocols. Also refer to the Africa Centre website: http://www.africacentre.com for further information.

Field workers collected blood by finger prick and prepared dried blood spots for HIV testing. Testing followed the Joint United Nations Programme on HIV/AIDS and World Health Organization Guidelines for Using HIV Testing Technologies in Surveillance (UNAIDS/WHO 2001). HIV status was determined by antibody testing with a broad-based HIV-1/HIV-2 ELISA, followed by a confirmatory ELISA.

We linked the population-based surveillance data to patient records from the public-sector ART programme operating in the area. Approximately 38% of HIV-positive individuals in our sample initiated treatment between the end of 2004 (when the ART programme started) and 2010.

All surveillance-based studies in the Africa Centre have approval from the Medical Ethics Board at the University of KwaZulu-Natal.

Mortality

In order to assess the differential effect of HIV with increasing time since infection, we studied short- and long-term mortality. Regardless of the specific date of infection, the disease’s duration and severity will only increase as the window of analysis widens for the average HIV-positive individual in the sample (Vanhems et al. 1998). We defined short-term mortality as deaths that took place between the baseline in 2003/4 and the end of 2006. Deaths that took place between the beginning of 2007 and the end of 2010 were considered long-term mortality.

The first line of Table 1 shows the number of deaths in our sample. There were 739 deaths between 2003/4 and 2010. Of these, 374 deaths (51%) occurred during the short-term mortality period and the remaining 365 between 2007 and 2010.

Table 1.   Rates of short-term (2003/4–2006) and long-term (2007–2010) mortality, by self-assessments of health, HIV status, ART uptake and age
CharacteristicAliveDeadMortality rates, per 1000 person-yearsAliveDeadMortality rates per 1000 person-years
(1)(2)(3)(4)(5)(6)(7)(8)
Follow-Up to 20062006–2010
884337412.46P-value*847836512.65P-value*
  1. ART, antiretroviral treatment. SAH, self-assessments of health.

  2. *For variables with ordered categories, the P-values denote the significance of the trend across the categories.

 SAHS
  Excellent3124605.82<0.0003025999.83<0.000
  Very good2764838.91265610812.03
  Good201011316.1219199113.73
  Fair8038730.637495419.96
  Poor1423156.281291327.72
 HIV status and ART uptake
  HIV-negative7157903.76<0.00069951626.90<0.000
  HIV-positive on ART35332.516718031.40
  HIV-positive not on ART133328156.7881212343.58
 Age
  15–245159844.86<0.00050231368.26<0.000
  25–34155512623.4314678817.31
  35–44147510220.0013859017.77
  45–546506227.505995122.86

Self-assessments of health

SAH were elicited by asking the respondents ‘How would you describe your general health at present?’ The possible choices were ‘excellent’, ‘very good’, ‘good’, ‘fair’ or ‘poor’. The proportions in each category monotonically decreased from best to worst SAHS, with 35% reporting ‘excellent’ SAHS and to only 2% reporting ‘poor health’.

Table 1 presents 4-year mortality rates stratified by SAHS category. Those with ‘poor’ SAHS had rates of death of 56.3 per 1000 person-years. These rates monotonically decreased as SAHS improved: they were 30.6, 16.1, 8.9 and 5.8 per 1000 person-years for those reporting ‘fair’, ‘good’, ‘very good’ and ‘excellent’ SAHS, respectively.

HIV status and ART uptake

Approximately 21% of individuals in the sample tested positive for HIV at baseline. Of these, 18% initiated ART between 2004 and 2006, while 38% initiated treatment in 2007–2010. Because the ART programme has been very successful in reducing mortality in the population (Herbst et al. 2009), we disaggregated HIV status by ART uptake.

Of all deaths registered between 2003/4 and 2006 (2007 and 2010), 1% (22%) occurred in HIV-positive individuals on ART, for a 4-year mortality rate of 2.5 (21.9) per 1000 person-years, and 75% (34%) occurred in HIV-positive individuals not on ART, for a 4-year mortality rate of 56.8 (43.6) per 1000 person-years (Table 1).

Although HIV test results could be obtained confidentially in a number of counselling centres, only a small proportion of individuals retrieved their results. Thus, our HIV measure reflects clinical status, but not necessarily individual awareness of serological condition.

Table 2 shows SAH stratified by HIV status and ART uptake. There is some evidence of HIV status and ART uptake affecting SAH. While HIV-negative individuals were more likely to report ‘excellent’ SAHS (36.8%), HIV-positive individuals on ART were more likely to report ‘good’ (29.9%) or ‘fair’ (15.6) and HIV-positive individuals not on ART were more likely to report ‘poor’ (3.9%) SAHS.

Table 2.   Distributions of self-assessments of health by HIV status
 HIV-negativeHIV-positive on ARTHIV-positive not on ARTTotal
  1. ART, antiretroviral treatment. SAH, self-assessment of health.

  2. Note: The numbers in the table are the absolute numbers of people in each SAHS and HIV status/ART uptake category. The numbers in parentheses are the percentages of people within each HIV status/ART uptake category.

SAHS: excellent2673 (36.8)162 (21.4)356 (29.3)3191 (34.6)
SAHS: very good2293 (31.6)231 (30.5)326 (26.8)2850 (30.9)
SAHS: good1588 (21.9)226 (29.9)313 (25.7)2127 (23.0)
SAHS: fair601 (8.3)118 (15.6)174 (14.3)893 (9.7)
SAHS: poor106 (1.5)20 (2.6)47 (3.9)173 (1.9)
Total7261 (100)757 (100)1216 (100)9217 (100)

Socio-economic status (SES)

Information on household and individual socio-economic statuses was collected between January 2003 and June 2004. In addition to gender, age and marital status, we used the SES variables, education, employment, logarithm of total household expenditures and number of household assets as control variables in our analysis. We have 741 participants in the HIV surveillance with missing education and employment information. Deceased individuals are equally represented among those with missing information. To avoid exclusion of participants with missing data on only one or two measures, we created dummy variables to represent missing data on education and employment and replaced the missing values with zero in the analyses reported here. Substituting for missing values with the mean of the original non-missing variable and multiple imputation led to essentially the same results. The sample is 60% standard deviation (SD = 49) female; with 7.2 (SD = 4.0) years of education; 76% (SD = 0.43) unemployed; 86% (SD = 35) single, widowed or divorced; with a total household expenditure of 783 Rand (SD = 1300); and 7.6 (SD = 3.9) assets.

Table 3 shows 4-year mortality rates by socio-demographic indicators. The only significant indicators of mortality were schooling, unemployment and number of household assets. None of the other covariates were significantly associated with mortality.

Table 3.   Rates of short-term (2003/4–2006) and long-term (2006–2010) mortality by gender and socio-economic indicators
CharacteristicAliveDeadMortality rates per 1000 person-yearsAliveDeadMortality rates per 1000 person-years
(1)(2)(3)(4)(5)(6)(7)(8)
Follow-Up to 20062006–2010
884337412.46P-value*847836512.65P-value*
  1. *For variables with ordered categories, the P-values denote the significance of the trend across the categories.

Gender
 Female535921411.510.21515520411.450.05
 Male348416014.00332316114.58
Educational attainment
 No school134610221.55<0.00012648218.02<0.000
 Primary or less228811715.14218210613.91
 Some secondary50811538.93490817310.63
 Some tertiary12825.5912449.74
Employment status
 Unemployed672725211.09<0.000647125611.660.01
 Employed211612216.70200710915.79
Marital status
 Single, widowed, divorced761832312.500.91730131712.870.69
 Married12255112.1911774811.35
Quintiles of total household expenditure
 1: (0–270 Rand)16028014.760.0715267614.630.19
 2: (271–492 Rand)19208212.6218417912.65
 3: (493–703 Rand)18538613.6517728113.31
 4: (704–1097 Rand)1806619.9017466010.15
 5: (1098–37 450 Rand)16626511.5115936912.72
Number of household assets
 04732816.570.084442919.52<0.000
 1–10627226812.66600227013.19
 11–1920887710.802022669.63
 20–2510125.311000.00

Analytic models

We used Cox survival analysis, with time measured in days, to calculate adjusted hazard ratios (aHR) with 95% confidence intervals (CI) for mortality for individuals with different SAH, conditional on their gender, age, marital status, SES, HIV status and ART uptake. We used the entire SAH scale, with dummy variables for each SAHS category (‘excellent’ being omitted).

Survival analyses predicting hazards of death were conducted in four steps: (i) SAH as the only independent variable, (ii) SAH, gender, age, marital status and the SES variables education, employment, household expenditures and household assets as independent variables, (iii) SAH, HIV status and ART uptake (HIV-negative omitted) as independent variables and (iv) SAH, gender, age, marital status, the SES variables, and HIV status and ART uptake as independent variables. We repeated step (ii) stratified by gender and HIV-status. Proportionality for all predictors was examined using scaled Schoenfeld residuals (Grambsch & Therneau 1994). The global tests showed no departures from proportionality (P-values were 0.26 and 0.81 for short-term and long-term mortality episodes respectively). The detailed tests showed slight departures for gender and ‘poor’ SAHS. Figure 1 shows Kaplan–Meier survival curves by HIV status where the assumption of proportional hazards can be visually confirmed. Because no HIV-negative individual with ‘poor’ SAHS died within 500 days of follow up, we excluded this group from the stratified analyses. All analyses were conducted in Stata se, version 11 (StataCorp 2005).

image

Figure 1.  Kaplan–Meier survival estimates by self-assessments of health and HIV status. Self-assessments of health: EXL = excellent, VRG = very good, GOO = good, FIR = fair, POO = poor.

Download figure to PowerPoint

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References

Short-term mortality

Column 1 in Table 4 shows that from baseline in 2003/4 to approximately 4 years later, the aHRs for SAH exhibited a clear gradient, as SAHS declined from ‘excellent’ to ‘very good’ (aHR 1.55; 95% CI 1.12–2.16), to ‘good’ (aHR 2.87; 95% CI 2.10–3.91) to ‘fair’ (aHR 5.46; 95% CI 3.93–7.58) to ‘poor’ (aHR 10.40; 95% CI 6.67–16.20). When gender, age, marital status and the SES variables were included in the model (column 2), the sizes of the coefficients decreased by between 9% (‘very good’) and 42% (‘poor’), and the SAH-mortality gradient was diminished. However, the SAH coefficients remained large and highly significant predictors of the hazard of dying.

Table 4.   Hazard ratios for self-assessments of health from models predicting short-term mortality (between 2003/4 and 2006)
Variable(1)(2)(3)(4)
  1. SAHS, self-assessed health status; ART, antiretroviral treatment; Ln, natural logarithm.

  2. 95% Confidence intervals in brackets. All models include a time dummy indicating whether the baseline year is 2003 or 2004.

  3. ***< 0.01, **< 0.05, *< 0.1.

SAHS: very good1.55 [1.12–2.16]***1.41 [1.00–1.97]**1.47 [1.06–2.05]**1.40 [0.99–1.96]*
SAHS: good2.87 [2.10–3.91]***2.25 [1.62–3.12]***2.32 [1.70–3.17]***2.10 [1.52–2.90]***
SAHS: fair5.46 [3.93–7.58]***3.34 [2.31–4.82]***3.93 [2.82–5.49]***3.12 [2.18–4.45]***
SAHS: poor10.40 [6.67–16.20]***6.01 [3.79–9.53]***6.12 [3.90–9.61]***4.64 [2.93–7.35]***
Indicator: HIV-positive on ART  0.55 [0.17–1.73]0.52 [0.17–1.66]
Indicator: HIV-positive not on ART  13.37 [10.48–17.05]***12.82 [9.99–16.46]***
Indicator: male 1.37 [1.11–1.69]*** 1.69 [1.36–2.10]***
Age 1.06 [1.05–1.07]*** 1.05 [1.03–1.06]***
Some primary education 1.16 [0.83–1.61] 0.89 [0.63–1.26]
Some secondary education 1.13 [0.80–1.61] 0.88 [0.61–1.25]
Some tertiary education 0.46 [0.11–1.93] 0.30 [0.07–1.29]
Indicator: employed 0.85 [0.66–1.10] 0.71 [0.55–0.91]***
Indicator: married 0.42 [0.31–0.58]*** 0.68 [0.49–0.96]**
Ln (household expenditures) 1.11 [1.02–1.20]** 1.07 [0.99–1.15]
Number of household assets 1 [0.96–1.03] 1.01 [0.97–1.04]
Observations9217921792179217

The SAH coefficients also shrink substantially but remain highly significant when we control for HIV status and ART uptake (column 3). The SAH coefficients shrank further but remained highly significant when HIV status, ART uptake and gender, age, marital status and the SES variables were accounted for in the regression (column 4). At the same time, the SAH-mortality gradient was substantially reduced as first HIV status and ART uptake and then HIV status, ART uptake and gender, age, marital status and the SES variables were included in the regressions. Those reporting ‘poor’ rather than ‘excellent’ SAHS had the largest excess mortality risk (aHR 4.6; 95% CI 2.9–7.4).

The hazards of death in HIV-positive individuals receiving ART in the period 2004–2006 were not significantly different from those of HIV-negative individuals. HIV-positive individuals not on ART had an aHR of death of 13.4 (95% CI 10.5–17.1).

Long-term mortality

Table 5 presents the results on long-term mortality (4–8 years past baseline). When gender, age, marital status and the SES variables were added to the regression, the SAH coefficients decreased in size and significance, and the SAH-mortality gradient became flatter (column 2). Adjusting for HIV status and ART uptake produces similar effects (column 3). Once gender, age, marital status, the SES variables, HIV status and ART uptake were accounted for at the same time (column 4) only those reporting ‘poor’ SAHS had a significantly higher risk of death 4 years from baseline (aHR 1.66; 95% CI 0.92–3.00).

Table 5.   Hazard ratios for self-assessments of health from models predicting long-term mortality (between 2007 and 2010)
Variable(1)(2)(3)(4)
  1. SAHS, self-assessed health status; ART, antiretroviral treatment; Ln, natural logarithm.

  2. 95% Confidence intervals in brackets. All models include a time dummy indicating whether the baseline year is 2003 or 2004.

  3. ***< 0.01, **< 0.05, *< 0.1.

SAHS: very good1.24 [0.94–1.62]1.14 [0.86–1.50]1.17 [0.89–1.55]1.12 [0.85–1.47]
SAHS: good1.44 [1.08–1.91]**1.17 [0.86–1.59]1.22 [0.91–1.62]1.11 [0.82–1.51]
SAHS: fair2.16 [1.55–3.00]***1.38 [0.95–2.01]*1.67 [1.20–2.34]***1.28 [0.89–1.84]
SAHS: poor2.97 [1.67–5.28]***1.86 [1.02–3.39]**2.26 [1.28–3.98]***1.71 [0.95–3.08]*
Indicator: HIV-positive on ART  4.58 [3.47–6.03]***4.35 [3.27–5.78]***
Indicator: HIV-positive not on ART  6.08 [4.78–7.73]***6.02 [4.71–7.70]***
Indicator: male 1.34 [1.07–1.66]*** 1.65 [1.33–2.05]***
Age 1.05 [1.04–1.06]*** 1.04 [1.03–1.05]***
Some primary education 1 [0.70–1.42] 0.88 [0.62–1.26]
Some secondary education 1.01 [0.69–1.46] 0.85 [0.59–1.23]
Some tertiary education 0.81 [0.28–2.31] 0.58 [0.20–1.69]
Indicator: employed 0.92 [0.70–1.20] 0.77 [0.59–1.01]*
Indicator: married 0.48 [0.34–0.66]*** 0.68 [0.49–0.95]**
Ln (household expenditures) 0.97 [0.90–1.04] 0.94 [0.88–1.01]*
Number of household assets 0.99 [0.96–1.02] 1.00 [0.96–1.03]
Observations8843884388438843

Contrary to the findings regarding short-term mortality, both HIV-positive people who received ART and HIV-positive people who did not receive ART had large excess mortality risks that were significantly higher than those for HIV-negative people. ART uptake significantly lowered mortality risk (aHR 4.4; 95% CI 3.3–5.8), but this ‘treatment effect’ was not very large.

Stratified analyses

In Table 6, we show replications of the short-term and long-term mortality analyses after stratifying the data by gender and HIV status. We only show the aHR for SAH and ART uptake, although we control for age, marital status and the SES variables shown in Table 3. Because no HIV-negative individual reporting ‘poor’ SAHS died within 500 days of follow-up, we dropped this group from the subsequent analysis. As with the pooled sample, we observed there were significant SAH gradients in mortality for HIV-negative men and HIV-positive men and women within the short-term mortality period. Despite the differences in the coefficient point estimates across the different strata, we could not reject the null hypothesis of equal coefficients across the sub-samples: In regression with all interactions between SAH, gender, HIV status and ART uptake none of the interaction terms were significant (all P-values >0.2; results not shown in this paper, but available on request from the authors).

Table 6.   Adjusted hazard ratios for self-assessments of health predicting future mortality – Stratified analyses
 2003/4–20062007–2010
(1)(2)(3)(4)
WomenMenWomenMen
  1. SAH, self-assessment of health; ART, antiretroviral treatment.

  2. 95% Confidence intervals in brackets. All models include controls for age, years of education, employment and marital status, as well as household expenditures, number of household assets, a time dummy indicating whether the baseline year is 2003 or 2004.

  3. ***< 0.01, **< 0.05, *< 0.1.

  4. †There were no HIV-negative individuals reporting ‘poor’ SAHS who died within the first 500 days of follow up, and thus the aHR corresponding to this cell become very imprecise to estimate.

  5. ‡No men on ART died between baseline and 2006.

Panel A: HIV-negative
 SAHS: very good2.56 [0.67–9.70]1.15 [0.49–2.66]0.88 [0.47–1.65]1.56 [0.92–2.65]
 SAHS: good5.24 [0.54–50.92]2.08 [0.93–4.65]*0.89 [0.45–1.72]1.68 [0.91–3.09]*
 SAHS: fair3.80 [0.86–16.69]*3.05 [1.25–7.44]**0.51 [0.18–1.47]1.88 [0.89–3.98]*
 SAHS: poor†
 Observations4138310941053052
Panel B: HIV-positive
 SAHS: very good1.45 [0.89–2.37]1.16 [0.62–2.17]0.97 [0.60–1.57]1.00 [0.52–1.93]
 SAHS: good1.90 [1.19–3.02]***1.50 [0.79–2.86]0.84 [0.50–1.40]1.18 [0.57–2.44]
 SAHS: fair2.81 [1.68–4.69]***3.06 [1.61–5.82]***1.45 [0.85–2.47]0.95 [0.36–2.52]
 SAHS: poor3.37 [1.68–6.76]***5.79 [2.76–12.17]***2.24 [1.01–4.96]**1.40 [0.31–6.36]
 = 1 if on ART0.06 [0.02–0.18]***–‡0.63 [0.44–0.89]***1.05 [0.57–1.92]
 Observations14355351254432

Four years from baseline the SAH-mortality gradient flattened, as in the pooled sample. Reporting ‘good’ or ‘fair’ SAHS is to be particularly predictive of mortality in HIV-negative men, but the significance level of this relationship is only 0.1. Reporting ‘poor’ SAHS more than doubled the hazards of long-term mortality for HIV-positive women (aHR 2.2; 95% CI 1.01–5.0).

The stratified analyses also show that ART uptake in women is responsible for the strong ART effect in the pooled analyses. This is expected, as more women than men were receiving ART during the observation periods (Houlihan et al. 2010).

We also re-estimated the previous models for mortality distinguishing between HIV-related and non-HIV-related causes of death. We find that the SAH-mortality gradient between baseline and 2006 was mostly driven by HIV-related deaths. This would be expected if individuals based their self-assessments on HIV-related symptoms or knowledge of HIV status. Our results were robust to the exclusion of deaths due to accidents.

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References

SAH have been collected and used in epidemiological studies in many developing countries. We examine for the first time the performance of SAH in predicting mortality in a community in sub-Saharan Africa with both high HIV prevalence and high ART coverage.

Within 4 years from baseline a clear SAH gradient in mortality emerged. Individuals reporting ‘poor’ SAHS were most likely to die and individuals reporting ‘excellent’ SAHS were least likely to die, even after age, gender, marital status, SES, HIV status and ART uptake were controlled for. We expected HIV status and ART uptake to substantially influence the power of SAH to predict mortality, as people’s SAH are likely to incorporate their knowledge of HIV status and the effect of HIV on mortality, as well as ART uptake and the survival benefits of treatment. We find that HIV infection (but not ART status) was an important confounder of the SAH-mortality relationship in the short term. However, while both the absolute SAH coefficient sizes and the steepness of the SAH-mortality gradient were reduced when we controlled for HIV status and ART uptake, SAH remained a highly significant and large predictors of morality in the short term.

In contrast, the power of SAH to predict long-term mortality was substantially reduced both in size and significance, once HIV status and ART uptake were controlled for, indicating that the confounding effects of HIV and ART on the SAH-mortality relationship become increasingly more important as the time between SAH observation and mortality measurement increases.

The finding that the long-term power of SAH to predict mortality was substantially diminished when HIV status and ART uptake were taken into account is not surprising. While not all individuals in this population may be aware of their HIV status, all individuals on ART clearly know that they are HIV-positive and, given the widespread availability of HIV testing and counseling in this community, a large proportion of people will in fact know their status (Bärnighausen et al. 2012). This knowledge is likely to have substantially influenced SAH, while the HIV status reflected in the SAH significantly predicted future mortality, in particular in the long term.

The fact that ART uptake was a large and highly significant predictor of mortality in the long term – diminishing some of the survival losses of HIV-positive people –, but not in the short term, is plausible as well. With increasing time since HIV status observation (and thus since HIV infection), ART uptake will become an increasingly important predictor of mortality, as HIV effects on survival will only manifest themselves many years after infection, and the countervailing effects of ART will thus become more and more apparent as time passes.

Our findings for South Africa are complemented by the preliminary results from a recently completed multi-country study, the Study on global AGEing and adult health (SAGE), which was jointly led by the World Health Organization and the International Network for the Demographic Evaluation of Populations and Their Health in developing countries (INDEPTH). This group collected data on SAH and mortality for people aged 50 years and older in eight countries of Asia and Africa, finding strong SAH-mortality associations in initial analyses (INDEPTH WHO-SAGE). However, the SAGE study is confined to older age groups and the current analyses do not control for confounding of the SAH-mortality relationship by HIV status and ART uptake, which we find to be important.

Among potential weaknesses of our study is the relatively short length of follow-up of individuals in the sample. Eight years of follow-up captures, to some large extent, premature mortality in such a young population. Based on our disaggregation between short-term and long-term mortality we can speculate that a longer follow-up would further diminish the SAH-mortality association.

Our results highlight the interactions between SAH, HIV status, ART uptake and mortality in a developing country. Because in places where SAH predict mortality, they usually also predict morbidity, health care use and general health needs (Bowling 2005; Idler & Benyamini 1997; Idler & Kasl 1995), a strong SAH-mortality association validates SAH as an important screening tool for policy purposes. From our findings we conclude that SAH can be used to predict mortality in the short-term, but that it needs to be complemented with HIV status and ART uptake if the goal is to predict long-term mortality.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References
  • Asamoah-Odei E, Garcia Calleja JM & Boerma JT (2004) HIV Prevalence and trends in sub-Saharan Africa: no decline and large subregional differences. The Lancet 364, 3540.
  • Bärnighausen T, Tanser F & Newell M (2009) Lack of a decline in HIV incidence in a rural community with high HIV prevalence in South Africa 2003–2007. AIDS Research and Human Retroviruses 25, 405409.
  • Bärnighausen T, Tanser F, Hallett T et al. (2010) Short communication: prioritizing communities for HIV prevention in sub-Saharan Africa. AIDS Research and Human Retroviruses 26, 401405.
  • Bärnighausen T, Tanser F, Malaza A et al. (2012) HIV status and participation in HIV surveillance in the era of antiretroviral treatment: a study of linked population-based and clinical data in rural South Africa. Tropical Medicine & International Health [in press].
  • Bowling A (2005) Just one question: if one question works, why ask several? Journal of Epidemiology and Community Health 59, 342345.
  • Cleary S & McIntyre D (2010) Financing equitable access to antiretroviral treatment in South Africa. BMC Health Services Research 10, S2.
  • Frankenberg E & Jones N (2004) Self-rated health and mortality: does the relationship extend to a low income setting? Journal of Health and Social Behavior 45, 441452.
  • Grambsch P & Therneau T (1994) Proportional hazards tests and diagnostics based on weighted residuals. Biometrika 81, 515526.
  • Herbst A, Cooke G, Bärnighausen T et al. (2009) Adult mortality and antiretroviral treatment roll-out in rural KwaZulu-Natal, South Africa. Bulletin of the World Health Organization 87, 754762.
  • Herbst A, Mafojane T & Newell ML (2011) Cause-Specific Mortality Trends in Rural KwaZulu-Natal 2000–2009. Africa Centre for Health and Population Studies, KwaZulu-Natal, South Africa.
  • Houlihan C, Bland R, Mutevedzi P et al. (2010) Cohort profile: Hlabisa HIV treatment and care programme. International Journal of Epidemiology 40, 318326.
  • Idler E & Benyamini Y (1997) Self-rated health and mortality: a review of twenty-seven community studies. Journal of Health and Social Behavior 38, 2137.
  • Idler E & Benyamini Y (1999) Community studies reporting association between self-rated health and mortality. Additional studies 1995 to 1998. Research on Aging 2, 392401.
  • Idler E & Kasl S (1995) Self-ratings of health: do they also predict change in functional ability? Journal of Gerontology B Psychology Series 50, 344353.
  • Malaza A, Bärnighausen T, Tanser F et al. (2011) CD4 Count Distributions and Unmet ART Need in a General Population in Rural KwaZulu-Natal. Oral presentation at 5th SA AIDS Conference, 7–10 June 2011, Durban, South Africa.
  • Salomon J, Tandon A & Murray CJL (2004) Comparability of self-rated health: cross sectional multi-country survey using anchoring vignettes. British Medical Journal 328, 258263.
  • Soleman N, Chandramohan D & Shibuya K (2006) Verbal autopsy: current practices and challenges. Bulletin of the World Health Organization 84, 239245.
  • StataCorp (2005) Stata Statistical Software: Release 11 [computer program]. StataCorp, LP, College Station, TX.
  • INDEPTH WHO-SAGE (2010) Growing older in Africa and Asia: multicentre study on ageing, health and well-being. An INDEPTH WHO-SAGE collaboration. Global Health Action 3 (Suppl. 2), 1107. Available at: http://www.globalhealthaction.net/index.php/gha/article/view/5661/6034. Accessed on 19 July 2011.
  • Tanser F, Hosegood V, Bärnighausen T et al. (2008) Cohort profile: Africa Centre Demographic Information System (ACDIS) and population-based HIV survey. International Journal of Epidemiology 37, 956962.
  • Tanser F, Bärnighausen T, Cooke G et al. (2009) Localized spatial clustering of HIV infections in a widely disseminated rural South African epidemic. International Journal of Epidemiology 38, 10081016.
  • UNAIDS (2008) Sub-Saharan Africa AIDS Epidemic Update, Regional Summary. UNAIDS and WHO, Switzerland.
  • UNAIDS/WHO (2001) Working Group on Global HIV/AIDS and STI Surveillance. Guidelines for Using HIV Testing Technologies in Surveillance: Selection, Evaluation and Implementation. UNAIDS and WHO, Geneva.
  • Vanhems P, Lambert J, Cooper D et al. (1998) Severity and prognosis of acute human immunodeficiency virus type 1 illness: a dose-response relationship. Clinical Infectious Diseases 26, 323329.
  • Weltz T, Hosegood V, Shabbar J et al. (2007) Continued very high prevalence of HIV infection in rural KwaZulu-Natal, South Africa: a population-based longitudinal study. AIDS 21, 14671472.
  • Yu E, Kean Y, Slymen D et al. (1998) Self-perceived health and 5-year mortality risks among the elderly in Shanghai, China. American Journal of Epidemiology 147, 880890.