In this collaborative study of five treatment programmes in four countries in sub-Saharan Africa, the mortality of HIV-infected patients starting ART could be compared with that estimated for the corresponding non-HIV–infected general populations. In these countries, a large proportion of deaths among young and middle-aged adults are HIV-related. We found that mortality during the first 2 y of ART was more than 18 times higher than in the general population not infected by HIV. However, there was large variability between prognostic groups and over time: in patients with very low CD4 counts and advanced clinical disease, mortality was increased over 300 times in the first 3 mo of treatment, whereas in the second year of ART, patients who started with high CD4 counts and less advanced disease had mortality rates that were comparable to those estimated for non-HIV–infected individuals.
We used excess mortality rates as well as SMRs and thus took the background mortality in the general population into account. The calculation of expected numbers of deaths was restricted to people not infected with HIV, which is crucial when the prevalence of the exposure (HIV infection) in the general population is high and SMRs are large 
. The mortality of over 13,000 patients was analyzed, including women and men, teenagers and middle-aged people, and patients with severe and less pronounced immunodeficiency. Our results should therefore be applicable to many other patients receiving ART in sub-Saharan Africa. We used estimates of non-HIV–related mortality from the WHO Global Burden of Disease project 
. Beginning with the year 1999, WHO has been producing annual life tables for all member states. A key use of these tables is the calculation of healthy life expectancy, the basic indicator of population health published each year in the World Health Report 
One limitation of our study is that the reference rates for HIV-unrelated mortality are unlikely to be completely accurate for the source populations from which the HIV-infected patients originate, and that errors in the calculation of expected number of deaths are not reflected in the confidence limits of SMRs and excess mortality rates 
. The five ART programmes included in this study are public sector scale-up programmes, which serve disadvantaged urban populations. Data from the 1970s and early 1980s suggest that adult mortality is lower in urban Africa than in rural Africa 
. The generally lower mortality rates observed in urban settings may, however, conceal pockets of poverty and high mortality among urban dwellers 
. Nevertheless, the use of national rates may have lead to estimates of the expected number of HIV-unrelated deaths that are too high, and SMRs and excess mortality rates that are too low. Given that reliable local mortality data are not available, we believe that the data from the Global Burden of Disease project are the best reference data available. Of note, the estimates used in this study for South Africa are in line with those from other analyses. For example, a recent modelling study of the demographic impact of HIV/AIDS in South Africa by the University of Cape Town and the South African Medical Research Council estimated that in 2006, 71% of deaths in the 15–49 y age group were due to HIV infection 
. Similarly, a study of AIDS-related mortality in rural KwaZulu-Natal estimated that 127 of 186 deaths (68%) were attributable to AIDS in 2004 
. A demographic surveillance study using verbal autopsy in the Agincourt subdistrict, rural South Africa, also found that HIV and tuberculosis were the leading causes of death in people aged 15–49 y 
Our study has other limitations. Complete ascertainment of risk factors and deaths and complete follow-up of patients is difficult to achieve in treatment programmes in low-income countries 
. Loss to follow-up was particularly high in one programme in Malawi, however, this is probably due to a higher rate of transfer out of patients in this programme. At present we cannot distinguish between loss to follow-up and transfer to another programme; this will be remedied in the next update of the database. We used multiple imputation to deal with missing baseline CD4 cell counts and loss to follow-up. This method assumes that missing values can accurately be predicted using the available data. In other words, the probability of missing no longer depends on the missing value after taking the available data into account (“missing at random” in Rubin's terminology 
). The plausibility of this assumption is unverifiable, but it is clear that mortality is increased in patients lost to follow-up 
, and unlikely that this can fully be captured by the clinical stage and CD4 cell count at baseline. Of note, sensitivity analyses excluding the sites with high rates of loss to follow-up from Malawi and Côte d'Ivoire gave similar results.
Follow-up was limited to 2 y in the present analyses, reflecting the relatively recent scale up of ART in sub-Saharan Africa, and it is possible that mortality will increase again in HIV-infected patients with longer duration of treatment. The short follow-up also meant that life expectancy of patients starting ART could not be examined. The ART Cohort Collaboration of HIV cohorts in Europe and North America recently estimated that life expectancy at age 35 y among patients on ART not infected through injecting drug use was 33 y 
. These questions will be addressed in future analyses of the IeDEA databases. Finally, our analysis did not consider differences between the HIV-infected and non-HIV–infected populations other than gender and age. In industrialised countries, there are important differences in the prevalence of risk factors, for example smoking, between infected and noninfected populations. In sub-Saharan Africa, where the epidemic is generalised and transmission by heterosexual contacts, differences in lifestyle factors are unlikely to be a major source of bias.
How do these SMRs compare with other population groups at increased risk of death due to unhealthy lifestyles, occupational exposures, or chronic conditions other than HIV infection? Few data are available for sub-Saharan Africa. White South African gold miners, compared to the white male population, had an SMR of 1.3, because of excess mortality due to lung cancer, chronic obstructive lung disease, and liver cirrhosis 
. Among male British doctors born in the 1920s, the probability of dying from any cause in middle age was three times higher in smokers than lifelong nonsmokers 
. Similarly, an analysis of the National Alcohol Survey in the US showed that regular, heavy drinkers had mortality rates from all causes that were 2.2 times higher than those observed in lifetime abstainers 
. The mortality of people with a body mass index (BMI) over 35 kg/m2
is increased by factor 1.5 to 2.5, compared to those with a BMI between 20 and 25 kg/m2
, and a similar increase in all-cause mortality is found in physically inactive people compared to physically active individuals 
. In a population-based study in Turin, Northern Italy, women with type 1 diabetes had an SMR for all causes of 3.4 and men an SMR of 2.0 
. The SMRs found in these patients and populations exposed to risk factors are thus quite comparable to those found in some of the patient groups included in our analysis.
Excess mortality was greater among men than among women. A recent analysis from the ART in Lower Income Countries (ART-LINC) collaboration found that although women are more vulnerable than men to becoming infected with HIV, they were equally or more likely than men to start ART 
. Women were younger and started treatment at a less advanced clinical stage, which could partly explain their lower excess mortality. Gender inequities in health may affect men as well as women: traditional masculine roles cast men as taking risks, being unconcerned about their health, and not needing help or healthcare 
. Conventional views of gender inequality might have made it easier for women than men in some settings to become engaged with HIV diagnosis and treatment services 
. Clearly, continued efforts are needed to empower women and secure their rights to treatment and care for HIV infection. However, more attention needs to be paid to HIV-infected men.
Although some HIV-infected patients starting ART in sub-Saharan Africa experienced mortality rates that were comparable with those experienced by other patients with a chronic condition, early mortality in adults starting ART continues to be high in sub-Saharan Africa 
. Many patients start treatment late, with a history of AIDS defining illnesses and low CD4 cell counts. Leading causes of death include tuberculosis, acute sepsis, cryptococcal meningitis, malignancies, and wasting syndrome 
. Of note, the Starting Antiretrovirals at three Points in Tuberculosis (SAPIT) trial recently showed that mortality among patients co-infected with tuberculosis and HIV can be reduced by 55% if ART is provided with TB treatment 
. Although our study cannot determine the CD4 cell count when ART should be started in order to minimise mortality, much of the excess mortality observed in our study would probably be preventable with timely initiation of ART. Further expansion of public health strategies to increase access to ART in sub-Saharan Africa is therefore urgently needed. In collaboration with the Global Burden of Disease project, the IeDEA network will continue to monitor mortality of HIV-infected patients starting ART and compare their mortality to that of the general population not infected by HIV.