Mortality was high in this cohort, and most of the deaths occurred within 3 months of starting ART. Severe and moderate anemia, thrombocytopenia and severe malnutrition were found to be independent predictors of mortality. The high early mortality observed in our study is in line with other similar studies from resource-limited settings [8
]. Causes of death were not investigated in the present study; however, in a study from South Africa wasting syndrome, TB, acute bacterial infections, malignancies and immune reconstitution disease were the major causes of death [14
]. In our cohort more than half of the patients had clinical AIDS at enrollment into HIV care, and other African ART programs have also reported high rates of advanced disease [8
]. Stigma and delay in seeking health care, lack of voluntary testing and counseling services, and health system delays in referral and ART initiation are possible explanations. Thus, priority must be given to identify HIV-infected individuals and start treatment earlier in the course of their illness, before they develop severe opportunistic infections.
Anemia was a strong predictor of mortality in our study. Patients with severe anemia had nearly 15 times higher risk of dying during the first year on ART compared to those with a normal hemoglobin level. Several studies from Europe and North America have shown that anemia is an independent predictor of mortality in patients on ART, even after controlling for CD4 cell count and viral load [22
]. Recently, studies from developing countries have found the same association [9
]. Indeed, in the largest African cohort study published to date, severe anemia (hemoglobin <8 g/dL) was the strongest independent predictor of mortality in 16 198 patients receiving ART in Zambia [13
It is uncertain whether the association between anemia and mortality is causal or whether anemia is rather a marker of progressive HIV disease. It is known that the incidence of anemia increases with progression of HIV infection [23
]. Furthermore, anemia can be a feature of certain opportunistic diseases, like disseminated mycobacterial infection and parvovirus B19 [25
]. Several other etiologic factors may be involved in the development of HIV-associated anemia, including micronutrient deficiencies, immunological myelosuppression, impaired erythropoietin production and blood loss from intestinal opportunistic disease [25
]. The role of iron supplementation is controversial, as some reports have suggested adverse effects of iron excess in HIV-infected individuals in industrialized countries [26
]. On the contrary, recovery from anemia after erythropoietin treatment has been associated with improved survival [23
], but high costs limit its use in poor countries. More recently, ART has been shown to significantly reduce HIV-associated anemia in developed countries [28
]; however, this has not yet been investigated in rural Africa. Further studies are needed to explore possible interventions against HIV-associated anemia in resource-limited settings, including the role of iron supplementation.
Malnutrition was another strong, independent predictor of mortality in our study. Estimated one year mortality was nearly 50% among patients with severe malnutrition. Previously, studies from industrialized countries have shown that malnutrition in HIV infection is associated with morbidity and mortality, even after the introduction of highly active antiretroviral therapy in the late 1990s [30
]. More recently, studies from developing countries have found that malnutrition is an independent predictor of mortality in patients starting ART [8
]. However, it is not clear whether targeted therapy for malnutrition will result in improved survival [34
]. Studies of nutritional interventions in HIV patients are urgently needed in developing countries, where malnutrition is often a result of poverty and food insecurity.
We found a reduced risk of death in patients starting ART in later calendar years compared with the initial period 2003–04. A possible explanation is that many patients with severe AIDS were included in the initial period, as this was the first clinic to offer ART in the area. However, since the risk reduction persisted after controlling for clinical stage, we believe that it may also be attributed to improved skills among local staff managing HIV patients. The decline in mortality over time supports our experience that non-physician clinicians can be trained to follow-up and treat HIV-infected patients.
To our knowledge, thrombocytopenia has never previously been shown to predict mortality in African patients on ART, although a few studies from North America have described an increased risk of disease progression and death [35
]. Further research is needed to confirm our findings. WHO stage IV was not significantly associated with mortality in our study, in contrast to previous reports [1
]. However, the comparison group was almost entirely composed of WHO stage III patients, which would weaken the statistical effect of WHO stage IV on mortality. Furthermore, the accuracy of clinical staging is probably quite variable in rural Africa. It is interesting that simple and more objective indicators identified in the present study appear to have a better predictive ability than clinical stage.
A prognostic model based on hemoglobin level had a strong predictive power in our study, separating the patients into low, low intermediate, high intermediate and high risk groups (Figure ). Previously, similar survival curves for hemoglobin levels have been reported in European HIV patients, although anemia occurred less frequently [22
]. Hemoglobin is a simple and inexpensive laboratory test, which can be performed even in rural, basic clinics. We believe it can be used as a simple and practical tool for initial risk assessment in the absence of CD4 cell count and viral load. Such early prognostic information would allow a more targeted search for opportunistic infections and closer follow-up in high-risk individuals, thus reducing excess mortality. Although the exact mortality figures from the present study can not necessarily be applied to other populations, we believe the concept of using hemoglobin level to identify patients with a poor prognosis can be used elsewhere. This simple prognostic model should be tested out in other African settings to assess its generalizability.
There are some weaknesses of our study. First, mortality might be underestimated, since patients lost to follow-up probably include individuals dying at home without being reported. Although the proportion of patients lost to follow-up in the present study (9.7%) was comparable to other African studies [12
], data quality would be improved with better cohort retention. Second, the results might be affected by selection bias towards patients with more severe disease, since the study was conducted in a hospital setting. Third, some patients measured baseline hemoglobin shortly after ART initiation, which might have led to an overestimation of the prevalence of anemia in patients with a ZDV-based regimen. However, post-ART hemoglobin was only employed in a small number of patients, and it is unlikely that this has introduced any systematic bias into the study. Fourth, it is known that the generalizability of a prognostic system can be impaired if important independent predictors are left out [37
]. We lacked reliable CD4 cell counts and viral loads, which are established predictors of morbidity and mortality in patients on ART [1
]. However, our results strongly suggest that simple and available measurements can be useful alternative prognostic markers.
The main strength of our study is that it was carried out in a rural African hospital with use of national staff and inclusion of all eligible patients. Most other African ART studies have been performed in urban areas [9
], in research settings with strict inclusion and exclusion criteria [38
], or with support from an international non-governmental organization [8
]. We believe that our results better reflect the reality in a rural hospital in sub-Saharan Africa, and thus may be applicable to other similar settings.