The proportion of people living with HIV in a population who have a detectable viral load is much more strongly associated with the rate of ongoing HIV infection in that community (HIV incidence) than the average viral load in people living with HIV (community viral load), a large study in men who have sex with men and people who inject drugs in India has shown.
This applied whether average viral load was measured among all people with HIV, among all people diagnosed and aware of their viral load, or only among people in medical care.
The findings may provide a useful measure to enable treatment programmes to monitor the effectiveness of test and treat programmes in reducing HIV incidence, but researchers say that long-term follow up is needed.
The proportion of people with HIV who were taking antiretroviral therapy (ART) was moderately associated with HIV incidence; this, the researchers say, could be used as a less-strong predictor of incidence in areas where viral load testing is not regularly available.
The findings were published in Lancet HIV online ahead of print in March 2016.
The most crucial measure of the success or failure of any HIV treatment and prevention programme is HIV incidence, the number of HIV infections acquired over a certain time, usually in the case of HIV a year. If HIV incidence and the number of people already living with HIV in a community is known then the R0 or reproduction number can be calculated, which predicts whether an epidemic is growing or shrinking.
HIV incidence is, however, hard to determine, especially in low-resource settings. HIV diagnoses cannot be used because of the variable and often long gap between someone acquiring HIV and being diagnosed with it; only if most of the people in a community test for HIV frequently and regularly will diagnoses reflect incidence.
There are incidence assays that can tell what proportion of people diagnosed caught HIV recently (generally in the last three to six months). However they depend on sensitive measures of antibody levels, are expensive, and need expert interpretation. They are therefore not practical for country- or population-level surveillance. So we need a reliable surrogate for incidence.
One measure that has been used is so-called ‘community viral load’: the average HIV viral load either in people in medical care, all people diagnosed whether currently in care or not, or in all people with HIV.
Clearly the usefulness of this measure depends on the accuracy of estimates of the proportion of people with HIV who are undiagnosed and the proportion diagnosed but not in care. Even if these can be accurately estimated, however, it has two fundamental flaws, as outlined in a 2013 paper. Firstly in a population with a high proportion on treatment, most of the people with a detectable viral load and therefore infectious will be undiagnosed, so the distribution of viral loads as measured in diagnosed people will not resemble the distribution in the undiagnosed, even if we have an accurate estimate for the proportion undiagnosed. Secondly, overall HIV prevalence makes a huge difference; in two communities where HIV-positive people have the same viral load, incidence will be a lot higher in a community where one in five people has HIV than a community where one in 500 has HIV, because the chances of encountering a person with HIV at random are 100-fold greater in the first community.
One measure that would take account of prevalence is the proportion of people with an undetectable viral load in the entire community, whether with HIV or not. But to verify this, one would have to test an entire community for HIV antibodies and viral load.
A group of researchers in India had exactly this opportunity, and the opportunity to use an incidence assay, and so decided to determine which measure of viral detectability had the best correlation with incidence.
They conducted a substudy within a large cluster-randomised trial with the lengthy name of the ‘Integrated Care Centres to Improve HIV Outcomes in Vulnerable Indian Populations’ trial, also called the NCT01686750 or ICCI trial.
ICCI is currently in progress across 22 cities in India and compares standard-of-care HIV services for two key affected populations, men who have sex with men (MSM) and people who inject drugs (PWID), with the provision of specialised, integrated HIV testing, care and treatment centres. The study uses respondent-driven sampling: key members of local MSM and PWID communities are identified and given incentives to recruit others, who in turn recruit others.
For this substudy, data collected at baseline for ICCI on HIV status and viral load were measured in 14,481 PWID at 15 sites and 12,022 MSM at 12 sites. Samples from people testing positive were also tested with an incidence assay to determine the proportion recently infected.
Participants also completed a questionnaire to measure the proportion who a) knew their HIV status b) knew their viral load c) were taking antiretroviral therapy (ART). Direct measurement of the proportion in care and ART was not possible as the data collected by ICCI are purposely anonymised.
Taken together, these data meant it was possible to compare HIV incidence in the two populations with:
The average age of the MSM recruited to the study was 26 and of the PWID 30: just 1.6% of the PWID were women (232 individuals). There was very little overlap between the two populations: just 1% of the MSM had ever injected drugs and 3.7% of the male PWID had had sex with another man.
In terms of HIV risk, while 56% of the MSM had had condomless sex with a man over the last six months, 41% of them had also had condomless sex with a woman – in fact a higher proportion than in PWID, where 39% had had condomless sex with a woman in the last six months. Thirty-five per cent of the PWID had shared needles in the last six months.
HIV prevalence in MSM was 8.6% (one in 11.6) and in PWID was 19.5% (one in five). HIV annual incidence in MSM was 0.87% (one infection per 115 MSM a year) and 1.43% in PWID (one infection per 70 PWID a year).
Among the 1034 MSM who tested positive (anonymously) for HIV, the average viral load was 6300 copies/ml (3.8 logs) and in the 2853 PWID it was 10,000 copies/ml (4.0 logs). In those aware that they had HIV it was 800 and 2500 copies respectively (2.9 and 3.4 logs); and in those aware and in care it was 500 copies/ml (2.7 logs) in both MSM and PWID.
The proportion of all participants (including HIV negative ones) found to have HIV viral loads over 150 copies/ml was 5% in the MSM and 14.2% of the PWID. Note that a measurement of the proportion of people with detectable viral load in the whole population necessarily incorporates prevalence.
Fifty-eight per cent of MSM with HIV and 72% of PWID with HIV had viral loads over 150 copies/ml.
Forty per cent of the HIV-positive MSM and 28% of the HIV-positive PWID said that they were taking ART: this was by self-report, but these proportions are almost exactly the same as the proportion with viral loads below 150 copies/ml.
How well did these measurements correlate with HIV incidence as measured by assay? By far the strongest correlation, with a correlation coefficient of 0.81, was the proportion of the entire community with viral loads over 150 copies/ml. The researchers calculated that for annual incidence to fall by 1% (which in the case of MSM would bring it down to zero, or 0.43% in PWID) the percentage of people with viral loads over 150 copies/ml in the whole community would have to fall by 4.34% (in other words down to 0.66% in MSM and 9.86% in PWID).
Average viral load in people with HIV, and in people who by self-report knew that had HIV, were both moderately well correlated (coefficients of 0.51 and 0.54% respectively). However, as has been pointed out, these measurements are no easier to estimate than the proportion of people in the community with viremia without widespread viral load testing. Even people who know they have HIV will not necessarily know their viral load if a high proportion are not in care. The average viral load in people in care is easier to estimate, as long as viral load testing is routine, but – largely because most people in care had low viral loads – it was not correlated with incidence (correlation coefficient 0.29, p = 0.14).
The proportion of people with HIV who were on ART, however, was moderately well-correlated with incidence (coefficient -0.54) and is an easy quantity to measure even in situations where viral load testing is not available.
The researchers therefore confirmed something that, in most models, has been a supposition rather than a fact – that the more HIV-positive people there are on ART and virally supressed, the lower HIV incidence will fall. The researchers calculated that to achieve a 1% drop in incidence, the proportion of people living with HIV on ART would have to increase by 19.5%, in other words to 59.1% in MSM (which would bring incidence down to zero) and 47.7% in PWID.
There is only one study worldwide where a direct, statistically watertight relation has been shown between the proportion of people in ART in a population and falls in HIV incidence: in the ANRS 12249 trial in KwaZulu Natal in South Africa.
A model by Andrew Hill and colleagues presented at the 2014 International AIDS Conference in Melbourne did find a correlation between the proportion of people with HIV on ART in a country and its HIV diagnosis and mortality rates – though this was very indirect evidence.
This Indian study bridges the gap between evidence from a closely-monitored trial and evidence from global surveillance, by finding a clear correlation between treatment, viral suppression and HIV incidence in large populations. It thus makes it easier for HIV treatment advocates to make a direct link between putting as many people as possible on treatment and subsequent falls in the rate of new HIV infections.
Solomon SS et al. Community viral load, antiretroviral therapy coverage, and HIV incidence in India; a cross-sectional, comparative study. Lancet HIV 3: e183-190. See abstract here. 2016.
