Introduction

Access to antiretroviral treatment (ART) has increased from 1 million to 17 million people living with HIV (PLHIV) globally from 2003 to 2015 [1, 2]. High HIV burden countries have accounted for the majority of this scale-up, with the number of PLHIV on ART increasing from 4 million to over 10 million in the last 5 years in Eastern and Southern Africa [3]. In both high- and low-income countries ART substantially increases life expectancy, especially for patients who start ART early [4, 5]. Given the rapid increase in coverage of life-saving treatment, a commensurate decrease in mortality in HIV-infected populations in high-burden countries would be expected. While cohort studies have demonstrated reductions in mortality among patients accessing ART, high rates of loss to follow-up (LTFU) make it difficult to fully assess the impact of ART at the population level. Complementary surveillance systems that can systematically assess population impact of ART programs on all-cause mortality on an ongoing basis are needed.

Civil registration systems in developing countries are often incomplete and causes of death poorly characterized. While a large fraction of PLHIV may be enrolled in care, high levels of LTFU, undocumented transfers, and passive reporting of deaths mean that program data underestimate mortality in the population in care if not linked to vital statistics [6] or adjusted statistically with LTFU studies [7, 8]. Hospital-based surveillance can shed light on HIV mortality among hospitalized patients, when both testing and HIV surveillance are routinely conducted [9]. Disease case surveillance systems, which could link patient records across systems, are still in their infancy in these settings. Further, the population in care is likely to have reduced mortality compared with those not in care, due to having access to medical interventions.

In urban areas where a majority of cadavers pass through mortuaries, mortuary-based studies have the potential to contribute to our understanding of excess mortality among HIV-infected persons. Although the literature is sparse on use of HIV testing in mortuary settings for surveillance purposes, De Cock et al. conducted a mortuary-based study to estimate mortality rates due to HIV and AIDS in Abidjan, Cote D’Ivoire in the late 1980s, before effective treatment for HIV existed, demonstrating that HIV was the leading cause of death in the city at the time [10]. A similar study conducted in 2001 found that nearly half of adult deaths in Pointe-Noire, Republic of Congo could be attributed to HIV, while the City of New York has been routinely testing for HIV during autopsies since the early 1990’s [11]. In Addis Ababa, Ethiopia, mortality surveillance at burial grounds demonstrated reductions in AIDS-related mortality following scale-up of ART [12].

Coverage of ART has rapidly expanded in Kenya, reaching 59% of all PLHIV by the end of 2014 [3], and is likely to be highest in the capital city of Nairobi. The high estimated coverage of death registration in Nairobi at 78% in 2014 [13], together with the fact that all registered deaths in Nairobi are notified by mortuaries, implies that mortuary-based surveillance could be representative of all deaths in the city. Relatively high ART coverage among PLHIV coupled with high levels of mortuary-based death registration and moderate HIV prevalence in adults, made Nairobi an attractive site for examining HIV-related adult mortality. We conducted a cross-sectional mortuary-based serological study to measure HIV prevalence among deaths in Nairobi. We used results from this study to estimate the population-level association between HIV and mortality for the city of Nairobi in the era of scale-up of ART towards universal access.

Materials and methods

Results

We found a pooled, unadjusted HIV positivity rate of 19.5% (95% CI 16.5–22.9%) among deaths at KNH and City Mortuaries, the positivity differing significantly by sex (14.6% among men, 29.5% among women, p<0.001). After weighting to account for missing test results and the coverage of the included mortuaries with respect to expected deaths by age and sex, the estimated HIV positivity among adult deaths in Nairobi was 20.9% (95% CI 17.7–24.6%) (Table 1).

Download:

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

Table 1. Percent HIV positivity among cadavers at City and KNH mortuaries, Nairobi 2015.

https://doi.org/10.1371/journal.pone.0181837.t001

A total of 807 deaths of persons aged 15 years and above (range 15–105 years, median 40 years) were reported during the 33 day period of data collection at the two mortuaries. These corresponded to 8,320 deaths when annualized and seasonally-adjusted. Based on a demographic projection, there were 16,173 deaths of persons aged 15 years and above for Nairobi City in 2015, indicating that after annualizing, the coverage of the sampled mortuaries was 65.0% of reported deaths, and 51.4% of expected deaths. Coverage versus expected deaths varied substantially by age and sex from 22.2% for women aged 15–24 years to 63.5% for men aged 25–44 years. Annualized deaths were then adjusted for this age- and sex-specific coverage to obtain the total adjusted deaths (Table 2).

Download:

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

Table 2. Annualized and coverage-adjusted surveillance deaths, Nairobi 2015.

https://doi.org/10.1371/journal.pone.0181837.t002

An estimated 2.8 million adults aged 15 years and above were residents of Nairobi in 2015, corresponding to 160,386 HIV-infected and 2,649,060 HIV-uninfected adults based on an overall HIV prevalence of 5.6% for Nairobi (Table 3). Dividing the estimated deaths from the surveillance system by the total projected population, the crude mortality rate in Nairobi among persons aged 15 years and older was 5.76 / 1,000 persons. In women, age-specific mortality estimates ranged from 1.92 / 1,000 among those aged 15–24 to 16.70 / 1,000 among women aged 45 and above. In men, mortality ranged from 3.16 / 1,000 among men aged 15–24 to 17.41 among men aged 45 and above. Overall, the mortality rate among HIV-infected persons was 21.12 / 1,000 compared with 4.83 / 1,000 among HIV-uninfected persons aged 15 years and older. Among HIV-infected women the mortality rate was 20.59 / 1,000 compared with 21.89 / 1,000 among HIV-infected men (Table 3). The standardized mortality ratio was 4.35 (95% CI 3.67–5.15), indicating that risk of death was significantly higher among HIV-infected persons compared with HIV-uninfected persons, when controlling for age and sex. The standardized risk difference was 0.016 (95% CI 0.013–0.019), corresponding to 16 additional deaths per 1,000 population among the HIV-infected after adjusting for age and sex (Table 4). The interaction between age and sex was significant, so all estimates were calculated using models stratified by sex and age group.

Download:

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

Table 3. Estimated age- and sex-specific deaths, population, and mortality rates per 1000 population, Nairobi 2015.

https://doi.org/10.1371/journal.pone.0181837.t003

Download:

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

Table 4. Risk difference (RD), standardized mortality rate (SMR) and population attributable fraction (PAF) due to HIV, Nairobi 2015.

https://doi.org/10.1371/journal.pone.0181837.t004

Combining the risk estimates with the prevalence of HIV in the population, the population attributable fraction (PAF) was 0.161 (95% CI 0.131–0.190). The estimated PAF differed between men and women, with women having a higher PAF for HIV infection than men: the PAF for women was 0.254 and for men was 0.101 (Table 4).

There were 40,382 males and 77,640 females aged 15 years and above on ART in Nairobi by December 31, 2015 according to official HIV program statistics. The Spectrum ART coverage estimate for 2015 for all HIV-infected persons aged 15 or more years in Nairobi was 73.6%, 61.1% for males and 82.3% for females, and Spectrum predicted that 13.8% of male deaths and 8.3% of female deaths (11.4% combined) would have been HIV-infected in 2015 (S1 Table).

Our sensitivity analysis looked at several key parameters that were estimated for this analysis as presented in S2 Table. The outcomes were most sensitive to HIV prevalence in the mortuaries and in the population, as well as the ratio of observed to unobserved prevalence in the mortuaries. The uncertainty interval for the SMR obtained through the uncertainty analysis was 3.04–7.19, while for the PAF the uncertainty interval was 0.118–0.221 (S3 Table).

Discussion

We conducted an analysis to combine HIV prevalence in a sample of deaths in Nairobi with population estimates in order to assess the burden of mortality associated with HIV in adults. We believe this is the first African study of its kind in the era of ART scale-up and as such it provides important insights in to the association of mortality and HIV infection in a major urban center with generalized HIV prevalence and a public ART program that reaches the majority of adults living with HIV. People living with HIV are at risk from death from multiple competing causes, including HIV. By estimating measures such as the SMR, RD and PAF this study looks beyond disease prevalence and provides estimates of the continuing impact of HIV on the adult population of Nairobi. Despite 7 in 10 HIV-infected persons aged 15 years and above receiving ART, our study showed persons living with HIV had a four-fold increased risk of mortality compared with those not infected. The population attributable fraction of 0.161 shows that about one in six deaths would be avoided in Nairobi if the increased mortality associated with HIV infection were eliminated, and more deaths would potentially be avoided in women than in men.

Antiretroviral treatment has been shown to dramatically reduce premature mortality. A prior study in Central Africa conducted at the height of the HIV epidemic when HIV prevalence was similar to Nairobi, but before ART was widely available, found nearly half of adult deaths were HIV-infected, while an East African cohort study in the late 1990’s found a 20-fold increased mortality risk among the HIV-infected before widespread access to ART. A national study in South Africa found consistent declines in population all-cause mortality associated with early expansion of ART access from 2005–2009. In our study, conducted after over a decade of ART scale-up, we found about 1 in 5 deaths was HIV-infected, yet people living with HIV continue to be at significantly increased risk of death with a greater than four-fold increased risk of mortality. This is especially true for women in Nairobi who have reached 82.3% ART coverage in 2015 –surpassing the UNAIDS fast-track target of 81% of all PLHIV on treatment.

This study had several limitations. The mortality data were collected in early 2015, but the distribution of deaths available for abstraction at the time of the study was for 2014. The coverage of the two participating mortuaries was moderate, but may not have been representative of all deaths in the city. The calculation of mortality rates assumes that measured deaths were from the resident population, however deaths in Nairobi are not necessarily drawn from the resident population of the city. Some Nairobi residents die outside the city limits, and some non-Nairobi residents die within the city. The latter group is likely to be over-represented at KNH since it is a national referral hospital and routinely receives patients transferred from other counties.

Although these two facilities do not include all deaths in the city, they do include a significant fraction of deaths that occur in Nairobi. Although there may be some seasonal factors that affect mortality, Nairobi does not generally suffer from seasonal malaria epidemics or other infectious disease outbreaks that are likely to be over- or under-represented during the study period. After seasonal and annual adjustments for total deaths, our study covered 51.4% of expected deaths based on demographic estimates, but covered 65.0% of reported deaths, but coverage varied substantially by age and sex. Our sensitivity analysis found that outcome measures were sensitive to assumptions about HIV prevalence and representativeness of sampled mortuaries, however our uncertainty analysis showed that overall conclusions were robust in the face of this uncertainty, lending confidence in the robustness of the findings that the risk of dying is greater in the HIV-infected population than in the HIV-uninfected population, and that a considerable fraction of deaths at the population level can be attributed to HIV in Nairobi.

Though a lower proportion of HIV-infected men access treatment, their lower HIV prevalence and competing causes of mortality result in HIV being a less frequent cause of death. A complete diagnosis, treatment and adherence history for all deaths could help elucidate to what extent undiagnosed infection, inadequate treatment access or late presentation, poor adherence, or treatment failure are responsible for the mortality differentials seen among PLHIV. Mathematical modeling has shown that the majority of deaths among PLHIV are likely to be among those not on treatment, even in high ART coverage settings, underscoring the need to reduce barriers to treatment initiation and improve retention on ART. Thus it was surprising to find a higher HIV-associated SMR among women than men, in spite of their greater access to ART. This may be due to over-representation of deaths due to external causes in our sample, which differentially affect men. If this is the case, our estimates of HIV mortality among men are conservative. Inclusion of additional mortuaries in future studies would help control this potential bias. Despite study limitations, it appears that mortality among PLHIV in Nairobi continues to be high in spite of high estimated ART treatment coverage. This high mortality among PLHIV may help explain why HIV prevalence has not increased in recent years in spite of continued new infections and a rapid expansion of public HIV care and treatment services [20].

Conclusion

Although sub-Saharan Africa has dramatically decreased mortality due to HIV through expansion of ART, in order to further reduce HIV-associated mortality, high-burden countries may need to reach very high levels of diagnosis, treatment coverage, retention in care, and viral suppression. Though hospitals are no longer overwhelmed with AIDS patients, Kenya has not yet achieved the goal of an AIDS-free generation.

Supporting information

Acknowledgments

We would like to acknowledge the study investigators and consultants for their participation in the development of the protocol and oversight of the study or development of specimen collection techniques, the study staff, laboratorians, data abstractors, mortuary staff that collected specimens from City Mortuary and KNH mortuary, those who assisted with KNH medical file reviews, and administrative support.

The Nairobi mortuary HIV surveillance study was conducted by National AIDS and STI Control Programme (NASCOP), University of Nairobi Institute of Tropical and Infectious Diseases (UNITID), University of California, San Francisco (UCSF), and U.S. Centers for Disease Control and Prevention (CDC/Kenya, CDC/Atlanta).