Revisiting the Methodology of Measuring HIV Community Viral Load

Abstract

To the Editors: Clinical trials have demonstrated that HIV transmission can be reduced in discordant heterosexual couples by control of viremia in the infected partner,1,2 and mathematical models predict that incidence can ultimately be eliminated by a strategy in which all persons receive antiretroviral therapy (ART) at diagnosis regardless of CD4 count.3,4 “HIV treatment as prevention” was chosen by Science as its 2011 Breakthrough of the Year.5 Similar to the use of viral load to indicate an individual’s risk of transmitting HIV, a new concept, community viral load (CVL), has been defined as an aggregate measure of the viral load of a population or subpopulation in a geographic area and proposed as a biological marker to monitor population-level transmission risk.6,7 Recent studies have observed an association between reduction of CVL and decrease in HIV incidence or new diagnoses.6,8,9 The United States National HIV/AIDS Strategy recommends the collection of data needed to calculate CVL and measuring the viral load in specific communities.10 We argue that, as currently conceived with the population included and the statistics used in its calculation, CVL is limited, difficult to interpret, and may not be appropriate for comparison across communities. CALCULATING CVL: WHICH POPULATIONS SHOULD BE INCLUDED? Some studies have calculated CVL based on the population of diagnosed and reported patients who had a viral load test result and excluded those who did not.7,9 Restriction of analyses to patients with reported viral loads confines the CVL to the population receiving viral load monitoring, ie, the population in care. The results may not reflect the true transmission risk in the community, because the proportion of patients without a reported viral load not only is large in many populations but also differs across communities.11 In a community with a large or increasing proportion of persons out of care or undiagnosed, CVL calculated from persons in care may provide false reassurance in that there may be a dramatic reduction in observed CVL but a small decrease or even an increase in the community transmission risk. One study included all patients who were diagnosed and reported to the local public health authority.6 Missing viral load values were imputed based on the assumption of missing at random, meaning that the probability of missing viral load can depend on other observed variables, eg, age, sex, race/ethnicity, transmission risk, CD4 count, and AIDS diagnosis, but not on viral load itself. The assumption may be difficult to defend because a proportion of in-care patients will be on ART and virologically suppressed, whereas few out-of-care patients would fall into either category. CALCULATING CVL: WHICH STATISTICS SHOULD BE USED? A variety of measures of central tendency and dispersion have been used to calculate CVL. Some studies selected the median.8,9 However, the median becomes less helpful as ART expands, and an increasing proportion of patients achieve virologic suppression. Time trends in median CVL will be difficult to follow-up when more than half of the population has an undetectable viral load, because the additional patients newly achieving virologic suppression will not be reflected in a median of 0 copies/mL. To counter this problem, other studies have used the mean.6,7 However, the mean’s sensitivity to extreme values (common in viral load distributions) limits its ability to represent the best central location of the data. More importantly, the mean CVL actually does not provide information about community transmission risk when the calculation does not consider the ratio of HIV-positive, or more accurately, persons with uncontrolled HIV infection, to HIV-negative persons in the community. For example, men who have sex with men (MSM) in New York city have a higher community transmission risk and higher HIV incidence than heterosexuals, but the surveillance data show that the MSM community has a lower mean CVL (12,722 copies/mL vs. 16,201 copies/mL in 2010) because MSM are more likely than others to be in care and to achieve and maintain viral suppression.12 Another measure that has been used by researchers is the total CVL, which is the sum of the reported and imputed viral loads.6,7 Measuring an individual viral load is the same as measuring the concentration of a solution, and the concentration here is the number of copies of HIV RNA in 1 mL of plasma of an individual. It is not mathematically proper to add up concentrations, and it is difficult to understand what per milliliter refers to when the unit copies per milliliter is used in total CVL, eg, 200 × 106 copies/mL. Some studies have suggested that targeting HIV prevention to communities with the greatest total CVL should produce the greatest reduction in overall HIV incidence.6,7 These analyses did not account for variation in population and subpopulation size, eg, heterosexuals may have a higher total CVL than injection drug users because they represent a larger population even though their overall prevalence is lower. Thus, a higher total CVL should be used with caution when targeting prevention services as it might misleadingly divert funds from a small, high-prevalence, high-risk group to a larger one with a more attenuated prevalence. ASSOCIATION BETWEEN CVL AND HIV INCIDENCE Several ecologic analyses have observed that decreases in CVL have been accompanied by a reduction in new HIV infections or new diagnoses.6,9 It is difficult to resist associating the 2 trends but equally difficult to assess the association, despite its biological and epidemiological plausibility. When ART is not available in a community, both median and mean CVL are expected to approximate the average viral load set point. As soon as some patients, even a small number, start ART and lower their individual viral loads, the median and mean CVL will drop. With the expansion of ART in industrialized countries and the developing world,13 we would expect to see a decrease in the mean CVL in almost every community, but a decrease in HIV incidence has not followed everywhere, eg, a decreasing trend in detectable viral load in patients receiving ART accompanied by increases in HIV diagnoses and estimated HIV incidence has been observed in Australia.14–16 This is because HIV incidence is driven by the size of the infectious reservoir and risk behaviors in the community. The infectious reservoir is composed of persons with uncontrolled viral load, a group that includes persons with HIV who are undiagnosed, diagnosed but out-of-care, and diagnosed and in care but not virologically suppressed. In a community with a small proportion of HIV-infected persons receiving HIV care, a decrease in observed mean or total CVL among in-care patients would not be able to significantly reduce the infectious reservoir; in a community where ART is widely available, a decrease in the size of the infectious reservoir may be negated by an increase in risk behaviors.16 AN ALTERNATIVE: PREVALENCE OF UNCONTROLLED HIV INFECTION To measure the population-level transmission risk, we need to measure the relative size of the infectious reservoir by including in our calculations both persons who are able to transmit the virus and persons who are at risk of infection (ie, HIV-negative individuals). Patients with undetectable HIV viral load are at low risk of transmitting their virus1,2 and may be considered not to be part of the infectious reservoir. We propose to introduce a new indicator, prevalence of uncontrolled HIV infection, which was also proposed by Kelley et al17 as transmission potential prevalence. We define it as the number of persons with detectable viral load divided by the number of individuals in the population or subpopulation. In jurisdictions where surveillance of HIV diagnosis is virtually complete and both detectable and undetectable viral load test results are reportable, surveillance data can be used to calculate this indicator. The method faces a number of challenges: (1) estimating the size of the undiagnosed and diagnosed but out-of-care populations; (2) determining the distribution of viral load among those without a reported value including persons who are undiagnosed, diagnosed but out of care, and diagnosed and in care but with no viral load test results; and (3) obtaining denominator data for subpopulations such as heterosexuals and MSM. In areas where surveillance data are not complete, viral load results are not reportable, or the size of a subpopulation of interest cannot be accurately estimated, cross-sectional household or high-risk subpopulation serosurveys could be conducted and viral load measured in antibody-positive specimens.18 This method also faces many logistical and methodological challenges that include specimen handling and sampling and nonresponse bias. Regardless of the method chosen, to present the results, a figure could be constructed to show the overall HIV prevalence and prevalence of controlled and uncontrolled HIV infection. The 3 HIV prevalence rates provide information on HIV burden, treatment outcome, and population-level transmission risk, respectively. The proportion of infected persons who are virologically suppressed, another measure for treatment outcome that has been widely used, can be expressed as the prevalence of controlled infection over the overall HIV prevalence. Figure 1 shows that, in a theoretical community, overall HIV prevalence increased from 4.0% in year 1 to 4.6% in year 5, the prevalence of controlled HIV increased from 0.5% to 3.5%, and the prevalence of uncontrolled HIV decreased from 3.5% to 1.1%. Because of the expansion of ART and decrease in mortality among people living with HIV/AIDS, overall prevalence is expected to continue to rise. We also expect a decrease in the prevalence of uncontrolled infection and an increase in the prevalence of controlled infection with the expansion of HIV treatment.FIGURE 1: Overall HIV prevalence and prevalence of uncontrolled and controlled HIV infection in a theoretical community.CONCLUSIONS As currently conceived with the population included and the statistics used in its calculation, CVL is limited and difficult to interpret. It may be used to track the success of treatment efforts in a population over time when undiagnosed and diagnosed but out-of-care persons are included in the calculation, but the measure will not be appropriate for the comparison of population-level transmission risk across communities if HIV-negative individuals are not included. To measure the population-level transmission risk, we should measure the relative size of the infectious reservoir by including both persons who are able to transmit the virus (ie, HIV-positive individuals with uncontrolled HIV infection) and persons who are at risk of infection (ie, HIV-negative individuals).