This is consistent with findings from another study of this cohort, in which a substantial proportion of individuals delayed starting HAART even when national guidelines recommended initiation of treatment [17], and a recent UK analysis showing that a high proportion of patients who experience a CD4 decline to <200 cells/μL do so while under regular follow-up [18]. Among those initiating HAART at a low CD4 cell count, the median buy U0126 follow-up

after diagnosis was 5 years, suggesting that rapid decline in CD4 cell count is not the main explanation for this. Late presenters are known to have a high risk of clinical progression in the first 3 months after HIV diagnosis, regardless of HAART initiation. As we wished to capture

the inherent efficacy of HAART rather than any consequence of poor adherence or loss to follow-up, our main analyses were restricted to individuals who remained under follow-up and on treatment at each time-point – our question, therefore, was whether patients who managed to remain alive and under care throughout this high-risk period this website could ultimately achieve as good an outcome on treatment as other patients. We did, however, perform sensitivity analyses to assess the robustness of our findings to patients who were lost to follow-up after the first 3 months. On the whole, our conclusions remained unchanged in these analyses. However, as loss to follow-up rates were (somewhat Phosphoribosylglycinamide formyltransferase unexpectedly) highest in ideal starters, clinical progression rates were significantly higher in ideal starters in these sensitivity analyses. However,

we do not believe that this higher loss to follow-up rate really reflects a higher clinical progression rate in this group – it is more likely that patients with a higher CD4 cell count felt more able to discontinue treatment, attend less frequently (with reduced viral load monitoring) or transfer their care to other centres. There are some important limitations of this study to note. Firstly, we have considered two arbitrary time-points (48 and 96 weeks) in order to be consistent with those used in many randomized trials. Alternative analyses that we could have used may have considered the time to an initial virological response, time to virological rebound or time to clinical progression after starting HAART. However, these approaches can be heavily affected by the frequency of monitoring; if monitoring is less (or more) frequent in late presenters, then the degree of bias that is introduced may be greater. Secondly, as already noted, our analyses excluded patients who presented late but who did not start HAART, either because they died very soon after diagnosis or because they chose to remain untreated. Thus, our outcomes cannot be applied to all patients who present late, but only to the group who survive long enough to initiate HAART.