Introduction

A study using data collected between 2002 and 2006, immediately before these data were collected, suggested that 1.8% of deaths in a public hospital in the Eastern Cape, South Africa were due to HIV [1] but the HIV status of those that died was not recorded. A more recent study [2] suggested that in 2006 283 thousand deaths or 42% of all deaths in South Africa were attributable to HIV and the authors compared this to other estimates for 2006 of 225 thousand using the Actuarial Society of South Africa Model [3], 250 thousand using the Thembisa model [2], 270 thousand in the Global Burden of Disease study [4,5], 350 thousand using the UNAIDS Spectrum model and 354 thousand based on changes in the age-distribution of deaths over time [6]. The considerable variation in these estimates was attributed to misclassification of AIDS deaths associated with different underlying conditions [2].

The difficulty in estimating AIDS deaths is partly due to the difficulty in deciding what proportion of deaths from a particular cause, such as TB, should be attributed to HIV [7]. The published models [2–8] all agree that the number of AIDS deaths peaked in 2006 but the model estimates of mortality vary widely because the models differ in their assumptions about survival after infection without ART. Without a direct estimate of AIDS-related deaths or of survival after infection without ART, models fitted to trend data in HIV prevalence and ART coverage cannot be used to accurately determine the overall mortality attributable to HIV.

The early response of the South African Government to the epidemic of HIV was one of confused denial [9] and although the epidemic started later in South Africa than in many neighbouring countries [10,11] the provision of ART in the public health system only began in earnest in 2006 by which time AIDS deaths had reached their peak [10]. The stigma associated with HIV and the unwillingness to acknowledge the magnitude of the epidemic meant that HIV and AIDS were often omitted from death certificates with other causes of death, such as TB, listed as the preferred alternative.

Between 2006 and 2009, when these data were collected, the Chris Hani Baragwanath Hospital (CHBH) was one of the main referral hospitals in South Africa serving a population of 1.3 million or 2.4% of the population of South Africa. At that time about 300 thousand people were dying of AIDS each year [2] so that about seven thousand would have died each year in Soweto. In the CHBH there were 3.9 thousand deaths each year, on average, suggesting that more than half of all those that died of AIDS in Soweto died in the medical wards of the CHBH. In an earlier study among children in the CHBH [12] the proportion of deaths accounted for by HIV infection increased from 7% in 1992 to 46% in 1996 with an overall odds ratio of 2.85.

Here we use a detailed and extensive set of data from the CHBH where the cause of death as well as the HIV status of those that died was established in order to determine the proportion of deaths from a range of causes that are attributable to HIV before ART became widely available in the public health sector.

Methods

Results

Of the 15,981 adults that died 15,722 were included in the analysis, 7630 men and 8092 women, while 527 cases were excluded as the cause of death occurred in less than 20 patients, and an additional 529 cases were excluded as the cause of death was not recorded.

Of the men, 48% were HIV-positive, 15% were HIV-negative, 11% were suspected to be HIV-positive and 25% were of unknown status. The corresponding figures for women were 54%, 11%, 17% and 25%, respectively. For HIV-negative men and women the modal ages at death were 64 and 74 years, respectively; for HIV-positive men and women they were 37 and 32 years, respectively (Fig 1). CD4+ cell counts were available for 41% of those that were HIV-positive and among these the median CD4+ cell count was 45/μL (90% between 3/μL and 310/μL) showing that most of the HIV-positive patients were in late stages of HIV. Despite the availability of ART in the public sector ART coverage was low in those infected with HIV increasing from 4.9% in 2006 to 8.6% in 2009.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1.

The number of A: Men and B: Women who were HIV-positive (peaks at lower ages) and who were HIV-negative (peaks at higher ages).

https://doi.org/10.1371/journal.pone.0215591.g001

The distribution of time from admission to death was not significantly different for men and women or for those that were HIV-positive or HIV-negative. Amongst this deceased population, 27% had died by the day after admission and the mortality was 11% to 13% per day after that (S1 Text, Appendix 6).

Odds and odds ratios

The final estimates for the ORs, DAFs and PAFs for HIV are shown in Table 1 and Fig 2 and Fig 3. For cryptococcosis, Kaposi’s sarcoma, pneumocystis and tuberculosis the DAFs are range from 80% to almost 100%. For gastroenteritis, anaemia, meningitis and lymphoma the DAF range from 50% to 75%. For other respiratory conditions, sepsis, genitourinary and disorders of the skin and bone the DAF ranges from 30% to 50%. For digestive conditions the DAF is 16%.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. The odds ratios for being HIV-positive for particular causes of death compared to controls.

https://doi.org/10.1371/journal.pone.0215591.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. HIV disease-attributable fraction (DAF) and population-attributable fraction (PAF) for different causes of death.

https://doi.org/10.1371/journal.pone.0215591.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. The total number, prevalence of HIV, odds ratios for HIV in those with conditions associated with HIV versus those with control conditions not associated with HIV, the disease (DAF) and population (PAF) attributable fractions for HIV.

Point estimates with 95% confidence limits. (Details in S1 Text, Appendix 8).

https://doi.org/10.1371/journal.pone.0215591.t001

HIV-attributable deaths from TB and other respiratory conditions together account for 20.7% ± 0.9% of all deaths. HIV-attributable deaths from cryptococcus, gastroenteritis, meningitis, sepsis and genitourinary conditions attributable to HIV each account for between 2% and 3% of all deaths and together they account for 11.7% ± 0.6% of all deaths. HIV-attributable deaths from Kaposi’s sarcoma, pneumocystis, anaemia, lymphoma, diseases of the skin and bone and of the digestive system each account for less than 1.5% of all deaths and together account for 3.3% ± 0.4% of all deaths.

Of the deaths included in the analysis, excluding those that were not coded or were dropped, the prevalence of HIV was 61% and the prevalence of AIDS related conditions was 69%. The HIV-attributable fraction in the whole sample was 36% and in those that were HIV-positive the HIV-attributable fraction was 60%. In 2006 the estimated number of deaths in South Africa was 691k in a study by Bradshaw et al. [2] and 738k in a study by the Joint United Nations Programme on HIV and AIDS (UNAIDS) [8]. If we take the average, 715k deaths, this study would suggest that the HIV-attributable deaths were about 257k, close to the 2006 estimates of 250k in the Thembisa model [2] and 270k in the Global Burden of Disease study [4,5].

Discussion

We estimate the excess mortality from conditions known, or suspected from the literature, to be HIV/AIDS related. Our imputation of HIV status goes some way to avoiding classifying deaths by their exposure status bearing in mind that in some cases HIV/AIDS is given as the cause of death. The data in Table 1 and Fig 3 show that almost all deaths from Kaposi sarcoma, cryptococcosis or Pneumocystis jirovecii infections could be attributed to HIV. Kaposi sarcoma was rare before the HIV epidemic struck [14] and odds ratios for being HIV-positive compared to controls was 56 (23–176) with a DAF of 94% (92%–95%) which concurs with previous work in Soweto, Johannesburg in which the corresponding OR for being HIV-positive was 47 (32–70) [15] and 89% of patients with Kaposi sarcoma were HIV positive. Likewise cryptococcosis or Pneumocystis jirovecii infections were extremely rare pre-HIV, almost always occurring in persons who were immune compromised [16]. ORs for being HIV-positive for lymphomas in this analysis of 2.9 (1.7–5.0) resemble those found in a previous case control study where the OR for being HIV-positive was 5.9 (4.3–8.1) for non-Hodgkin and 1.6 (1.0–2.7) for Hodgkin lymphoma [15].

The prevalence of HIV in the control group is high and comparing this to the prevalence of HIV in ante-natal clinic surveys between 2006 and 2009 in the Johannesburg Municipality suggests an OR for being HIV-positive in our control group as compared to those in the ante-natal clinic surveys of 3.2. However, the prevalence of HIV among pregnant women in the Johannesburg Municipality may have been lower than in Soweto and it could also be that HIV causes such a force of mortality that even for deaths that are unrelated to HIV, such as car accidents or lung cancer, being HIV positive would cause additional medical complications leading to a greater case fatality. The high prevalence of HIV in older people which reaches 7% in those over 70 years of age (S1 Text, Appendix 7), is particularly striking since their life expectancy without ART is only about 2 years [17]. The problem of HIV infection in older people has been noted but only touched on briefly in the literature [14]. In the control arm of a study of newly diagnosed cancers at CHBH and Johannesburg Hospitals between 1995 and June 2004 the prevalence of HIV in those aged 65 years or more was 2.6% [18]. Various explanations have been put forward [12] but the reason for this high prevalence in older people remains unclear. As shown in S1 Text, Appendix 7, increasing the odds-ratios by a factor of 3.2 increases both the DAFs and the PAFs but the overall pattern and the general conclusions remain unchanged.

Even though the deaths examined did not cover all wards and all hospital deaths, such as deaths from cancer of the cervix, an AIDS-defining condition, and a mixture of deaths attending the intensive care unit, it is clear that the proportion of deaths in this series that are attributed to HIV are high, ranging from 20% of deaths from digestive causes to almost 100% for Kaposi sarcoma, Pneumocystis jirovecii and cryptococcal infection.

Tuberculosis, for which the PAF due to HIV is in excess of 90% is the most common opportunistic infection among people infected with HIV [18–20] and TB co-morbidity is very likely in Soweto where tuberculosis is endemic. Specific diagnoses of the organisms underlying meningitis, gastroenteritis and respiratory conditions were not always available and a correct diagnosis requires a culture which is costly and time consuming [21]. In sub-Saharan Africa, there is limited literature on the causative agents of meningitis, gastroenteritis and other infection related deaths but the aetiology of these is different in those with and without HIV [21].

Anti-retroviral therapy causes a decline in the incidence of several AIDS defining conditions, most clearly in the case of TB [22], but full immune recovery is only achieved if treatment is started immediately after infection when CD4+ cell counts are still high [23]. Of all the deaths between 2006 and 2009 in the medical ward of CHBH up to 36% could have been averted if people living with HIV had been started on anti-retroviral therapy early in the course of their infection. In countries with well-established treatment programs, the life expectancy for people with HIV approaches that seen in HIV-negative people [24] but in 2006–2009 the policy was not to start people on ART until their CD4+ cell-count had fallen to less than 200/μL [15,18] by which time their immune system was severely compromised. In this study 6.3% of those that were HIV-positive were on ART which matches the 6% coverage of ART among South African adults at that time [10]. The mean age at death for those infected with HIV was 38 years for men and 33 years for women, suggesting that millions of life-years were lost due to the low coverage and late provision of treatment. Providing ART much earlier in the epidemic would have saved the lives of very many young adults while minimizing the economic impact of HIV, and reducing the burden on the health system [25]. A recent study of productivity losses due to premature mortality from cancer showed that the cost per cancer death in South Africa was US$101k. The 82 deaths from Kaposi’s sarcoma alone will have cost US$8.3M in lost productivity [26]. If we estimate the cost of keeping one person in CHBH for one day at about US$100 and noting that the average survival after admission was 6.7 days, then the hospital costs of the 2.5 thousand HIV-attributable deaths each year will have amounted to US$1.7 million per annum.

Now that ART has been provided quite widely in South Africa it will be of great importance to repeat this study in the same or a similar hospital as this will provide a direct estimate of the impact of the roll-out of ART on mortality from AIDS related conditions in South Africa.

An important strength of this study is that a confidential data sheet was used with an option for the clinician to include comments, in addition to listing the cause of death, to allow clinicians to record causes of death and express clinical opinions that may otherwise have been omitted from routine death certificates. Only consultants who were involved with the medical care of patients at CHBAH reviewed and assigned the causes of death. Laboratory results that may not have been available at the time of death such as HIV status and microbiological cultures were all reviewed at a later date and included in the data. The consultants recorded the causes of death in a manner that allowed for ICD coding without being effected by coder interpretation.

A major limitation of this study is the reliance on clinical records to establish the cause of death which is known to be prone to inaccuracies and which may be further compounded when death certificate data are used. While certain cases were discussed amongst the hospital consultants prior to arriving at a cause of death this ‘peer review and consensus’ was ad hoc and mainly confined to ‘problem’ cases.

Supporting information

Acknowledgments

The original study and all of the data collection was done by AB. The mathematical analysis was led and coordinated by BW. All authors contributed to the design of the analysis, the interpretation of the data, and writing the manuscript. All authors state that they have no conflict of interest and all authors have approved the final manuscript. This project arose from the 2016 Clinic on the Meaningful Modelling of Epidemiological Data (MMED), which is part of the International Clinics on Infectious Disease Dynamics and Data (ICI3D) Program.