This study describes data of three years of follow up of 9040 HIV-infected adults initiated on anti-retroviral treatment at the Themba Lethu Clinic, Johannesburg, South Africa. This high number of patients demonstrates the ability of rolling out a successful HAART programme despite being in a resource limited environment, and this, at a rapidity and scale that compares with reports from other African countries [6, 11, 12].
Despite the fact that stavudine is associated with significant complications, a large proportion of low and middle-income countries still use stavudine based HAART as first line therapy [8, 9], mainly because of the cost implications of alternative drugs. In the present study nearly 30% of patients had to switch to non-stavudine based regimens, due to major side effects. These findings concur with those from another large study in South Africa where 21% of patients switched regimens over a similar time period .
This study provides estimates of the pattern of toxicities, the predominant ones being peripheral neuropathy, symptomatic hyperlactataemia, and lipoatrophy. Our incidence rates of peripheral neuropathy (12.1/100 person-years) were much higher compared to other African studies: 5.2/100 person-years in Rwanda  and 2.8/100 person-years in another site in South Africa . The variability in these rates could be because there are no grading protocols for the severity of peripheral neuropathies. Our incidence rates of lactic acidosis were similar to a study from another province in South Africa, 1.6 versus 1.9/100 person-years . The proportion of lactic acidosis was slightly higher compared to another study from Botswana , 2.5% versus 1% in our cohort. When compared to this same study , the proportion developing symptomatic hyperlactataemia were much higher in our sample (5.7% versus 2%). The median time to the development of lactic acidosis was a little later in our study when compared to another study from South Africa (10.8 months versus 7.5 months) .
In terms of the chronic toxicities, incidence rates of lipoatrophy were 4.6/100 person-years on stavudine based therapy and 3.0/100 person-years on the non-stavudine based therapy. One study from Rwanda , where patients were also on stavudine based regimens showed a similar incidence rate of lipoatrophy to that reported in the present study at 4.7/100 person-years, while one other study from South Africa  showed a lower incidence rate of 1.4/100 person-years. Another study from Rwanda showed a much higher proportion (34%) of patients developing lipoatrophy . The reason for this wide variation in lipoatrophy rates specifically, is possibly due to different diagnostic criteria used for identifying cases. Thus, in the present study and in the studies showing low but similar rates [6, 13], only cases that were severe enough to warrant a regimen change were noted. In the study showing a much higher proportion, milder cases of lipodystrophy that did not require a regimen change, were also recorded . An objective case definition of lipodystrophy has been developed ; however, it requires access to DEXA and CT imaging technology, which is often not available in resource-limited settings. Therefore, an alternative consensus definition for the diagnosis of lipodystrophy is required for such environments.
A very small number of patients presented with diabetes (0.3%) and dyslipidaemias (1.3%). However, this could be because lipid or glucose levels were only tested when clinically suspected due to cost implications, therefore probably underestimating the true prevalence of dyslipidaemia and diabetes.
The major strength of this study is the large sample size. This cohort is similar to many other resource-limited settings where there is rapid scaling-up of comprehensive HIV care and HAART. It allowed for a relatively long duration of follow up of up to three years and a fairly high retention rate of 70%, with all clinicians working on common protocols for defining the various HAART-associated toxicities. However, despite this, these findings must be considered in the light of potential limitations. It is possible that only severe toxicities that warranted a change in regimen may have been reported and this may have led to an underestimation of the rates of some of the HAART-related toxicities, particularly lipoatrophy. Also, plasma glucose and serum lipid levels were not routinely measured and thus the true rates for glucose intolerance, diabetes and dyslipidemia were not attained. Lower reported rates of the chronic toxicities could also be related to the rates of death (5%) and loss to follow up (24%), the majority of which occurred within the first six months on treatment as described in a previous report .