The optimal time to start treatment for HIV/AIDS has been a contentious issue since the introduction of Highly Active Antiretroviral Treatment (HAART). Initially a "hit hard and early" strategy was promoted 1. Because of concerns about long term toxicity and fear of developing drug resistant viruses, delayed treatment starts were later recommended in clinical guidelines 2. The delayed treatment policy implied that, in the absence of particular disease manifestations, treatment should not be started before CD4 counts dropped below 200 cells/μl. However, recent evidence indicates that this policy reduces survival compared to earlier treatment start. The World Health Organisation (WHO) revised the ART guidelines for resource constrained settings accordingly and re-introduced a "hit hard and early" strategy. In the revised 2009 guidelines, it is recommended that HAART is initiated on all HIV patients with CD4 counts below 350 cells/μl, regardless of symptoms 3. Despite this change of recommendations, few low income countries have revised the national ART guidelines and many still recommend that initiation of HAART in asymptomatic HIV-infected persons are delayed until the CD4 count drops below 200 cells/μl 4. Recent evidence from high income countries support even earlier initiation of treatment - before CD4 count drops below 350 cells/μl 56. A clinical trial in Haiti recently demonstrated that deferring treatment until CD4+ T cell counts drops below 200 cells/μl, rather than providing HAART at CD4 counts between 200 and 350 cells/μl, increases death risk nearly four times 7. However, there is little information to guide this important clinical decision in low income settings.

The debate regarding optimal timing of treatment start has tremendous implications for HAART demand, and subsequently, on the estimated treatment coverage in different settings. Towards the end of 2008, only 3 million people out of 33 million with HIV were given HAART 8. In low income countries, treatment is still mainly provided to the sickest patients. Median baseline CD4 counts at initiation of HAART have been found to be between 100-150 cells/μl in several low income countries 9101112131415. In contrast, a population based study from 2007 indicates that 42% of all HIV patients in Malawi had a CD4 cell count under 350 cells/μl, while 22% had under 200 cells/μl 16. Shifts to an early treatment start strategy will increase the need for HAART, but few people actually receive HAART. Because of the huge gap between treatment coverage and needs, health outcomes from different treatment indications need to be assessed systematically.

Life years gained by different CD4 starting points is necessary information for making informed choices about early or late start of treatment. Studies in low income countries have found that patients starting HAART early (CD4 <350 cells/μl) have life expectancies from 9.4 to 17.2 life years and that life expectancies are 6.8 - 14.9 with late treatment strategies (CD4 < 200 cells/μl) 171819202122. Only one study adjust for lead time bias and report the size of the health benefit from the treatment; Cleary et.al. found that HAART yielded a net health benefit of 10 life years when it was initiated at the point when CD4 was below 200 cells/μl. The increase in the length of survival for patients starting treatment in earlier stages of HIV may reflect either earlier treatment initiation or delay in time of death. Therefore, to avoid overestimates of the actual impact of early treatment start, it is necessary to eliminate this potential confounding effect, which is often referred to as lead time bias. A recent eART-linc collaboration study found lead time to be as long as 4.6 life years 23. Studies need to adjust for lead time bias and report net life years gained from different starting strategies in order to provide adequate information to make a rational clinical decision on this issue in a low income setting.

Our overall aim is to inform the ethical dilemma of choosing between providing HIV treatment to patients with potential best outcomes and giving fair chances to the more severely ill patients with less potential benefit. The objective of this study is therefore to estimate the life years gained with early and late treatment strategies in low income countries. In the absence of randomized clinical trials, we model life years gained by using best available evidence from observational studies.

We did a structured literature review and combined findings in a pooled-analysis to find the best evidence underlying the different treatment strategies. Based on these findings, a Markov life cycle model was constructed to estimate the expected remaining life years with and without HAART stratified in three baseline CD4 strata at the time patients first present for care: CD4 <50 cells/μl or 50-199 cells/μl or 200-350 cells/μl. The model can be applied on any population, and we test it on a Tanzanian HIV population since information on HIV prevalence and life tables were easily available 2425. We eliminate lead time bias by calculating survival with and without treatment for three CD4 strata, representing three different stages of the disease, and subtracting a stratified net health benefit for each CD4 strata.

The set of mutually exclusive health states in the model and the various events that can occur in the history of HIV are illustrated in Figure 1. In our model, all patients have CD4 counts below 350 cells/μl and are assigned to the initial health states called "CD4 200-350 and alive", "CD4 50-199 and alive" or "CD4 <50 and alive". The Markov-cycle tree model is evaluated by expected value simulation, and a cycle length of one year is applied. At the end of each cycle, patients may continue in the initial health state or move from one state to either of the two events called "HIV related death" or "Age related death" or to a lower CD4 stratum, according to transition probabilities (see below). We considered 100 Markov cycles (years) to ensure that the whole cohort has moved to the death state by the end of the analysis. We incorporate dynamic changes in CD4 cell count occurring after patients enter the model for those not initiating HAART. We did not allow HIV patients initiating HAART to jump down to lower CD4 strata since studies indicate a large CD4 recovery the first four years after HAART initiation 26. The effect from this CD4 recovery will be incorporated in the pooled HIV-related mortality-rates.

We calculated expected remaining life years without HAART and net life years gained from HAART by different CD4 starting points (figure 2). From figure 2 it can be seen that the sickest patients have 0.7 remaining life years without treatment and net life years gained from HAART is 7.2 life years. Patients with CD4 counts 50-199 cells/μl are expected to live 2.0 life years without treatment and have a net health benefit of 7.6 life years from HAART. Patients with CD4 counts 200-350 cells/μl have 4.8 remaining life years without treatment and gain an extra 14.5 life years with early treatment start.

Results from the one-way sensitivity analyses are shown in table 3, and show that outcomes are highly sensitive to assumptions about death risk for patients receiving HAART. High and low values of death risks with and without HAART were taken from the upper and lower bounds of the weighted 95% confidence intervals of the pooled-analysis. Applying the most optimistic reductions in death risks from HAART yields estimated survival of 24.3, 9.2 and 14.5 life years per person if starting HAART at CD4 200-350, CD4 50-199 and <50 cells/μl, respectively. The corresponding figures using the most pessimistic death risks were 8.5, 6.4 and 4.4 life years. The effects of death risks without treatment, and assumptions about mean age of patients were also tested. Outcomes turned out to be much less sensitive to these assumptions (table 3).

Our study suggests that providing HAART early is likely to yield most life years if everyone receives treatment. This is in concordance with the recommendations in the revised 2009 WHO guidelines. However, few governments in low income countries are capable of providing optimal treatment to all if everyone with CD4 < 350 cells/μl are considered to be eligible. Given the severe budget constraints low income countries face, a policy where fewer patients are considered to be eligible is the more common practice.

In terms of stratified net life years gained from HAART, we believe our study to be unique. From international published databases, we were unable to identify any studies which describe stratified net life years gained from HAART with various starting points. Five studies reported life expectancy for HIV patients when starting treatment at CD4 < 200 cells/μl and results ranged from 6.8 life years to 14.9 life years, which is comparable to our total survival (without treatment + HAART) of 9.6 life years in figure 21719202122. Four studies reported life expectancy when starting HAART at CD4 < 350 cells/μl and results ranged from 9.4 to 17.2 life years, which is slightly lower than our total survival of 19.3 life years (figure 2) 17182022. However, our estimated health benefit from HAART was reduced to 14.5 life years (figure 2) when we adjusted for lead time bias. One study reported life years gained by increasing HAART initiation threshold from CD4 counts below 250 to 350 cells/μl, and found undiscounted survival gain to be 1.04 life years 18. This is lower than our estimated 6.9 life years and could be explained by the fact that we apply a threshold of 200 cells/μl rather than 250 cells/μl.

The Tanzanian ART guideline from 2005, which is still being used in 2009, recommends treating asymptomatic patients when the CD4 count drops below 200 cells/μl 42. Asymptomatic patients with higher CD4 counts and greater potential to benefit may be excluded. The revised 2009 WHO guideline considers all patients with CD4 counts below 350 cells/μl eligible for treatment, no patient groups are excluded. This goal is ambitious for resource constrained settings, and implies that up to 573 000 adult HIV patients would be eligible for HAART in Tanzania. Much more resources would be needed to scale up national ART programs to this level 43.

This study demonstrates that HIV patients live longer with early start of antiretroviral treatment in low income countries, and highlights the ethical dilemma of choosing between providing HIV treatment to patients with potential best outcomes and giving fair chances to all the more severely ill patients with less potential benefit. Since low income countries have many constraints to full coverage of HAART and more people will be eligible with an early start strategy, the results of this study is a good starting point for a more transparent and reasoned debate when drawing explicit eligibility criteria during further scale up of HAART.

The authors declare that they have no competing interests.

All authors fulfill the criteria of authorship and contributed collaboratively to conception and design, analysis and interpretation of data, drafting the article, revising it critically for important intellectual content and final approval of the version to be published. We assure that there is no one else who fulfils the criteria of authorship that has not been included as an author.

KAJ was principal investigator, designed the study, performed the systematic literature search, developed the Markov model, led the analysis, and was lead author for the paper. BR was a co-investigator and participated in the planning, analysis and writing of the paper. OFN sought funding for the study, contributed to design, planning, analysis and writing of the paper. All authors read and approved the final manuscript.