Evidence is not yet sufficient to recommend universal treatment, clinicians and community writers say

Standard way of grading evidence urged for all guidelines
This article is more than 11 years old. Click here for more recent articles on this topic

We do not yet have sufficiently strong evidence to recommend that antiretroviral treatment (ART) should be offered to all people with HIV, regardless of their CD4 count, a group of writers in AIDS Journal has concluded.

The writers criticise the current inconsistency on when to start ART in guidelines that have global influence. They suggest that if all guidelines used a rigorous standard for rating evidence, their recommendations would be more consistent, and probably more cautious about when to initiate treatment.

The writers are researchers Caroline Sabin of University College London, David Cooper of the University of New South Wales in Sydney and Mauro Schechter from the University of Rio de Janeiro, Brazil, plus Simon Collins of the UK treatment information organisation HIV i-Base.

Guideline inconsistencies

They explore inconsistencies between guidelines issued by:


virological suppression

Halting of the function or replication of a virus. In HIV, optimal viral suppression is measured as the reduction of viral load (HIV RNA) to undetectable levels and is the goal of antiretroviral therapy.

observational study

A study design in which patients receive routine clinical care and researchers record the outcome. Observational studies can provide useful information but are considered less reliable than experimental studies such as randomised controlled trials. Some examples of observational studies are cohort studies and case-control studies.


When the estimate from a study differs systematically from the true state of affairs because of a feature of the design or conduct of the study.


A serodiscordant couple is one in which one partner has HIV and the other has not. Many people dislike this word as it implies disagreement or conflict. Alternative terms include mixed status, magnetic or serodifferent.

statistical significance

Statistical tests are used to judge whether the results of a study could be due to chance and would not be confirmed if the study was repeated. If result is probably not due to chance, the results are ‘statistically significant’. 

The two US guidelines strongly recommend that treatment is started in people with CD4 counts below 500 cells/mm3 and issue a moderate recommendation that ART is offered to people with CD4 counts above that figure – i.e., that ART should be offered to everyone on diagnosis with HIV.

In contrast, the two European guidelines do not recommend that treatment is started in people with CD4 counts over 350 cells/mm3. They both make various exceptions to this, recommending treatment for pregnant women, for people diagnosed in early infection, for people with various HIV-related conditions at any CD4 count, and for people with  hepatitis B and C below a CD4 count of 500 cells/mm3 (BHIVA) or if treatment is needed/feasible (EACS).

Another exception is that both BHIVA and EACS recommend that treatment is provided for people who wish to take ART in order not to pass HIV on to others.

EACS says: “In serodiscordant couples, early initiation of ART as one aspect of the overall strategy to reduce HIV transmission to the seronegative partner should be considered and actively discussed.”

BHIVA says that it is a matter of good clinical practice to discuss the prevention possibilities of ART with all patients, not just couples, and prescribe it if requested for this reason.

WHO changed its guidelines, which are written with resource-poor countries in mind, after the AIDS article was written, raising the treatment initiation threshold from 350 cells/mm3 to 500 cells/mm3on 30 June 2013. WHO also makes exceptions to this, recommending treatment for all people with active TB and hepatitis B.

WHO also recommends that HIV-positive people in serodiscordant relationships are offered ART.

Grading the evidence

The writers of the AIDS article, as well as being concerned with the discrepancy in recommendations, are just as concerned with how the guideline writers arrive at their differing recommendations: the various guidelines disagree diametrically on whether the evidence for certain recommendations is strong, weak or insufficient.

The US guideline writers, for instance, say that the evidence for starting treatment before the CD4 count falls below 500 cells/mm3 is "strong" and is derived from “one or more published randomized clinical trials” (IAS-USA) or from “well-designed nonrandomized trials or observational cohort studies” (DHHS).

In contrast, BHIVA does not even consider that there are “sufficient data to make a recommendation” on whether people with CD4 counts over 350 cells/mm3 should start treatment. WHO says the evidence for starting people with CD4 counts over the AIDS-defining limit of 200 cells/mm3 is only “moderate”. EACS does not publish its assessment of the strength of evidence.

Why the difference? It is primarily due to the way evidence is gathered and assessed. The US guidelines essentially still rely on expert opinion, or at least the experts’ opinion of what trial results mean. Members of the guideline writing groups conduct literature reviews to identify relevant new information, synthesise the information obtained, and make recommendations which are then voted on.

The article writers comment that, even if individual bias is ruled out, this method is not transparent in terms of how evidence is weighted: to what degree is evidence from different types of studies balanced, and are some clinical outcomes (e.g., rates of virological suppression) regarded as more important than others (e.g., rates of treatment-limiting toxicity)?

In order to try and eliminate bias and add transparency, WHO and BHIVA have switched to a system called GRADE, which stands for the Grading of Recommendations Assessment, Development and Evaluation.

This requires guideline writers to define in advance the outcomes they consider most significant before they review the literature, so they cannot change their views on the relative importance of outcomes later. This means, for instance, that if they want to make a decision on whether or not to initiate treatment at CD4 counts above 500 cells/mm3, then studies looking at treatment outcomes for other CD4 count thresholds would only be regarded as indirect evidence. The evidence provided by specific studies can be upgraded or downgraded after they are analysed, but only according to very specific criteria.

What the cohort studies say

Sabin and her co-authors point out that there has never yet been a true randomised controlled trial addressing the question of whether to start treatment at a CD4 count above 350 cells/mm3 as opposed to waiting until the CD4 count reaches this figure.

There have been four observational cohort studies, but only two of these, the US NA-ACCORD study and the European CASCADE study, found there were more deaths in people starting ART at CD4 counts below 350 cells/mm3 than in people starting it at higher CD4 counts, and only NA-ACCORD found a benefit in starting therapy at CD4 counts over 500 cells/mm3, compared with starting below that figure.

The difference in results is substantial: NA-ACCORD found that participants were 94% less likely to die if they started at CD4 counts over 500 cells/mm3 rather than under that figure, while CASCADE found no benefit at all; the other two studies, the When to Start Cohort and the HIV-CAUSAL study, found no statistically significant benefit even to starting at CD4 counts over 350 cells/mm3 versus under it, although they did find trends that did not reach the boundary of statistical significance.

There must be some confounder that is causing such varying results, and the most likely candidate may be that participants in cohort studies are not randomised; they don’t start or not start ART purely by chance, and the reasons they start may influence the outcome.

So people who start ART above a certain CD4 threshold may have done so because they were already ill, which could in itself be associated with higher mortality. Alternatively, they may have done so because they tested sooner; this would mean they were less likely to be already ill and had more time in which to consider starting, which could be associated with lower mortality.

The question of whether to start at higher CD4 counts or even drop CD4 criteria altogether therefore remains controversial, not least because current studies may be designed to look at outcomes such as mortality and viral suppression, but may be less able to pick up on factors that would weigh against early treatment, such as long-term drug-related side-effects.

Other studies have found an association, not between mortality and taking treatment as such, but between mortality and maintaining an undetectable viral load; but in this case, the writers say, the cause of failure to suppress HIV may be lifestyle factors such as poor adherence, drink and drugs, or homelessness, which could cause more deaths in themselves rather than such deaths being directly due to unsuppressed HIV at high CD4 counts.

The writers acknowledge that the evidence that HIV therapy makes people much less likely to transmit HIV is compelling, but argue that the one randomised controlled trial that proved this beyond doubt, the HPTN 052 trial, compared the prevention benefit of starting at a CD4 count over or under 250 cells/mmand therefore cannot directly address the risk/benefit ratio of starting at higher CD4 counts. This means it remains essentially up to the patient, advised by their doctor, to decide whether the possibility of transmission of HIV to others or the possibility of drug-related side-effects is more important to them.

Trials underway

There are two randomised controlled trials underway that will answer at least part of this question. The START trial and the TEMPRANO trial will both randomise people to start ART at CD4 counts above 350 cells/mm3 or wait till their CD4 count reaches that figure. Results are expected in 2016 and 2014 respectively (though TEMPRANO is solely being conducted in west Africa so may be seen as not applying to other settings).

Neither study, however, will look at starting at higher CD4 thresholds and will not tell us if it is clinically better to offer ART to all people diagnosed with HIV. While another randomised controlled study to decide this could be designed, the risk/benefit ratio at high CD4 counts could be so finely balanced that a trial would have to be unfeasibly large.

A decision to offer universal ART may end up being taken for reasons other than clinical benefit, such as simplicity, public health benefit, or because people taking ART are usually less likely to disappear from care.

In the meantime, however, the writers of the AIDS article argue, it is important for guideline writers to rely strictly on best-quality evidence and not on their own expert opinion. They therefore recommend the adoption of the GRADE system for all HIV treatment guidelines.


Sabin CA et al. Rating evidence in treatment guidelines: a case example of when to initiate combination antiretroviral therapy (cART) in HIV-positive asymptomatic persons. AIDS 27(12): 1839–1846, 2013.