Community viral load is flawed indicator of infectiousness, say experts

Proportion with detectable viral loads may be a better guide

Gus Cairns
Published: 31 May 2013

Four US researchers from the University of Chapel Hill in North Carolina, including Myron Cohen, principal investigator of the HPTN 052 study, have criticised the use of 'Community Viral Load' (CVL) as an indicator of the success of HIV treatment as prevention, especially as a measure of the average infectiousness of people with HIV.

CVL will tend to underestimate the proportion of people with high viral loads in the population, they say; in addition, how it translates into the actual rate of new infections (incidence) is highly dependent on the overall number of people with HIV in the population (prevalence).

They recommend instead that estimates of the proportion of people in the whole population who have detectable viral loads (VLs) by conventional testing, or the proportion of people with viral loads above a certain threshold, such as 1000 copies/ml, be used instead.

CVL has been used in a number of studies as a measurement of the degree to which HIV testing and treatment is affecting the average viral load in the HIV-positive population. It has been expressed as the mean viral load in a population, the geometric mean (the mean logarithm of the viral load, where, say, 3 logs is 1000), or as the sum of all the viral loads in a specific cohort of people. It is usually taken from viral load test results from clinics.

In several studies a statistical association has been found between falls in the CVL and falls in new HIV diagnoses, and this has been cited as evidence for the success of using antiretroviral therapy to bring down HIV incidence within the community studied ('treatment as prevention', or TasP).

The writers of this paper, however, critique the use of CVL and question its ability to predict or track falls in incidence.

The first problem is that although some of the studies that have used CVL have tried to include an estimate of viral load in undiagnosed people, this is, as the writers say, "dependent on viral load in diagnosed people being related in some degree to viral load in the community."

The undiagnosed may be a very different group of people to the diagnosed and in addition, in places like the US, some studies have also found that a high proportion of the detectable HIV in the population is carried not by the undiagnosed but by those who have been diagnosed but then dropped out of care. Both the undiagnosed and those lost to care may be disproportionately young, female, heterosexual and/or black, and may have different viral load and CD4 count characteristics anyway.

Taking the undiagnosed and people who have disappeared from care into account, the authors calculated that 'true' CVL in a typical US setting may be twice as high as CVL estimated from hospital test results – 30,000 rather than 15,000, in their modelled scenario. Even in a best-case scenario like San Francisco, where there are high rates of diagnosis and retention in care, they calculate that CVL could be underestimated by about 15%.

In addition, this does not take account of the fact that the highest viral loads in any group of people with HIV are in those who have only just acquired HIV – who are also the least likely to be diagnosed. The authors cite a well-documented African study that estimated that 38% of infections came from people who had just acquired HIV themselves. There are ways of estimating the proportion of undiagnosed people with high viral loads, but these tend to have wide margins of uncertainty and it would be prohibitively expensive to detect most acute HIV infections.

CVL in itself also does not mean a lot unless HIV prevalence – the proportion of people in the community with HIV – is taken into account. Take two populations where, in one case, 5% of people have HIV and, in the other, 0.1%. Even if the CVL of people with HIV is the same in both populations, if people meet each other at random, they have 50 times more chance of encountering a person who is infectious in one community than in the other. This is why it is very difficult to bring down the rate of new infections (HIV incidence) in groups of people with very high prevalence.

Even in a population with relatively low rates of people in acute infection, how viral loads are distributed may make a big difference to ongoing transmission. The authors give an example of two groups of ten people. Each group has an average viral load (CVL) of 10,000 copies/ml but in one group everyone has a viral load near that figure while in the other nine have an undetectable viral load and one, perhaps undiagnosed, member has a viral load of 100,000 copies/ml. In the former group while the distribution of risk behaviour – who is risky and who is safe – may vary, it may not make much difference to the overall 'infectiousness' of the group. In the latter group, though the CVL does not vary according to behaviour, the group's 'infectiousness' varies widely according to whether the one member with a high viral load takes risks or not.

Finally, as the authors remark, it is impossible with certainty to attribute a change in a group (HIV incidence) to changes in individuals in that group (reductions in their viral load).

To give an example: in one of the British Columbia studies there was certainly a strong correlation between falling diagnoses in people who inject drugs and the proportion of them on treatment (with, consequently, a lower CVL in this group). However, as the authors point out, improvements in access to antiretroviral therapy could be accompanied by improved access to injecting equipment and methadone. Or the link could be more indirect: improved access to ART might also put them in touch with support services that enable them to manage their risk behaviour. Viral load and HIV infections may not be directly linked at all but could both be caused by a third factor; so an increase in the proportion of people on treatment (factor A) and a reduction in risk behaviour (factor B) could both be due to an ageing population who are both more likely to be in care and to have less casual sex.

The authors end by recommending that, rather than using the average viral load in a population, a better tool for predicting HIV incidence in modelling studies would be the proportion of people in the whole population who have HIV viral loads above the limit of detection, or above a cut-off point such as 1000 copies/ml. This takes HIV prevalence into account and also, because few people not in care will have low viral loads, is almost a proxy for the proportion of people in care and their retention in care. It does, however, still depend on accurate estimates of the proportion of people who remain undiagnosed.


Miller WC, Powers KA, Smith MK and Cohen MS. Community viral load as a measure for assessment of HIV treatment as prevention. The Lancet Infectious Diseases 13: 459-64, 2013.

Community Consensus Statement on Access to HIV Treatment and its Use for Prevention

Together, we can make it happen

We can end HIV soon if people have equal access to HIV drugs as treatment and as PrEP, and have free choice over whether to take them.

Launched today, the Community Consensus Statement is a basic set of principles aimed at making sure that happens.

The Community Consensus Statement is a joint initiative of AVAC, EATG, MSMGF, GNP+, HIV i-Base, the International HIV/AIDS Alliance, ITPC and NAM/aidsmap