Increasing HIV testing frequency, and giving everyone antiretroviral therapy (ART), would not in themselves reduce HIV prevalence in US gay men, a mathematical model suggests.
These measures would produce, in the model’s baseline scenario, a 34% reduction in the cumulative number of new infections and a 19% reduction in cumulative deaths by the year 2023. This would lead to the annual number of new HIV infections in gay men almost declining to the annual number of deaths, but not quite.
The model therefore predicts that HIV prevalence would continue to grow in US gay men, albeit very slowly. This remains the situation under a number of different scenarios; even if every gay man took an HIV test every year, and everyone diagnosed with HIV started treatment within six months of infection, infections would still slightly outstrip mortality.
The model also finds that universal treatment would lead to a doubling in the prevalence of multidrug-resistant HIV, although this would not lead to an increase in deaths or progression to AIDS.
However this particular output from the model derives from data on the prevalence of primary HIV drug resistance that is more than seven years old and 'MDR' means any resistance to two of the three main classes of HIV drug that were well-established at this point, not resistance to all options currently available.
The model’s assumptions
The model was devised by researchers at the University of Southern California. It includes a number of parameters regarding the gay male population in Los Angeles, such as HIV incidence, the proportion of people in primary infection, the proportion diagnosed, the proportion diagnosed on treatment, and the proportion who progress to AIDS – although it does not directly input a figure for the proportion who have an undetectable viral load (virally suppressed).
It also inputs variable figures for the per-partner risk of HIV transmission in gay men, the frequency of HIV testing in gay men, the adherence rate in people taking ART, and the rate at which people acquire drug resistance.
These figures are all derived from observed trends in HIV infection in Los Angeles gay men between 2000 and 2010.
The researchers perfected their model by testing different combinations of inputs against the observed figures and repeatedly discarding ones that came out with results that didn’t match what actually happened over the previous decade, until they achieved the best fit.
The researchers than tested what would happen to the rates of new infections, the proportion of people with HIV who are not yet diagnosed, the proportion on ART, deaths, progression to AIDS and multidrug resistance, if they increased the frequency of testing and/or reduced the gap between infection and treatment in gay men.
Currently, in Los Angeles, gay men have HIV tests, on average, every 4.4 years. The researchers used the model to find out what would happen if this frequency was increased to every three years, every two years, and every year.
The average time between HIV infection and starting ART in gay men has been calculated as 2.5 years. The researchers modelled what would happen if this was reduced to one year or to six months.
Other figures fed into the model included the cumulative number of new HIV diagnoses, AIDS diagnoses and deaths that would occur by 2023 if nothing changed: 54,000 new infections, 49,000 AIDS diagnoses and 42,000 deaths. The current proportion of people with HIV who are undiagnosed was set at 20%.
The proportion who enter treatment with 'multidrug-resistant HIV' (MDR-HIV) was set at 3.1% which would have increased by 2023, even if nothing else changes, to 4.8%. This figure derives from data that was becoming out of date when it was published, and in addition, the model's definition of ''MDR-HIV' includes virus that today would be sensitive to a wide range of new drugs. See below for more on the model's assumptions about MDR-HIV.
The model’s predictions
At the time the model was first devised, the US Department of Health and Human Services (DHHS) HIV treatment guidelines still recommended that ART be started when a person’s CD4 count falls below 350 cells/mm3. So the first thing the modellers did was to model a scenario in which the only thing that changed was that all those diagnosed started ART according to this recommendation. This single change led to a 6% cumulative reduction in HIV infections and deaths and an 11% reduction in progression to AIDS over the next ten years. The proportion of people undiagnosed would fall to 18.5% and the proportion with multidrug resistance would increase to 6.1%.
If average testing frequency increased to one test a year, then the cumulative number of new infections by 2023 would fall to 35,800 (a 34% reduction), of new AIDS cases to 30,000 (a 39% reduction) and of deaths to 34,100 (a 19% reduction). This would reduce the number of people unaware of their infection very considerably, to only 4%. But the proportion of people with multidrug-resistant HIV would increase to 9.1%.
If, in addition, the gap between HIV infection and treatment initiation was reduced to one year, this would lead to a 42% reduction in cumulative new infections and a 28% reduction in deaths, and if reduced to only six months, to 47% fewer infections and 34% fewer deaths. But the proportion of people with multidrug-resistant HIV would increase to 11.9% and 13.7% respectively under these scenarios.
The modellers found that HIV testing and putting more people on treatment were not synergistic – in other words, that they worked independently to reduce HIV infections and deaths, but did not reinforce each other’s effects. Increasing test frequency from every 4.4 years to annually, for instance, resulted in an absolute 28% reduction in cumulative new infections regardless of the length of time between infection and starting treatment. Conversely, reducing the gap between infection and treatment from 2.5 years to six months resulted in a 13% reduction in new infections, regardless of testing frequency.
Cautions, caveats and conclusions
The modellers do make one other crucial assumption: they assumed that when gay men become aware of their HIV infection, they very considerably reduce their sexual risk behaviour. In the model the researchers reduced the likelihood of someone transmitting HIV by an average of two-thirds post-diagnosis, a figure based on US studies. These reductions have not necessarily been matched by figures from other parts of the world, and the researchers found that sexual risk behaviour was the assumption fed into the model that had the biggest influence on cumulative new infections.
The researchers do point out that their model lacks certain subtleties. Firstly, it doesn’t attempt to stratify HIV risk in different age groups, ethnic groups, or by risk behaviour – it assumes all sexually active gay men are approximately at the same risk of HIV. Secondly, it does not add in any allowance for the possible future use of pre-exposure prophylaxis (PrEP). And thirdly, they point out that "mathematical models are only as good as the available data used for the parameters and calibration".
It is also interesting that they use estimates of the average time between diagnosis and treatment as their parameter for the influence of treatment on prevention, rather than using the more direct figure of the estimated proportion of people with HIV with an undetectable viral load. They explain that this is because we do not have high-level evidence for the efficacy of viral load suppression as a prevention measure in anal sex. However, they do feed in an assumption that anyone starting treatment at a CD4 count over 350 cells/mm3 who mainains adherence becomes 96% less infectious.
The researchers suggest that, given that starting people on treatment earlier leads to a prediction of higher rates of multidrug-resistant HIV, and given that increased testing and more treatment do not seem to be synergistic, it might be better to concentrate on getting people to test more frequently rather than treating everyone diagnosed.
The finding that the number of patients with MDR-HIV will increase, however, is based on very old data. The figure of 3.1% the model uses for the proportion of people who start therapy with MDR-based HIV is derived by taking the median figure for primary MDR resistance from a single review (Van de Vijver) which was published in 2007, and includes no data collected after 2005. Even in this review it was noted that MDR resistance was lower in other parts of the world than the US. In addition, it was just after this point that studies started to find that drug resistance in people with HIV was starting to decline, and epidemiologists soon confirmed that it had in fact been doing so for several years (see Health Protection Agency). This trend has been sustained in more recent studies.
This brings the finding of the model that each 10% increase in average testing frequency, or each 10% decrease in average time between infection and starting ART, leads fairly consistently to a 0.45% absolute increase in the proportion of people starting therapy who have MDR-HIV, into some question.
In addition, however, they also use an outdated definition of multdrug resistance, namely resistance to two of thre three drug classes in use at the time of the 2007 review, nucleoside and non-nucleoside reverse transriptase inhibitors (NRTIs and NNRTIs) and protease inhibitors (PIs). Even this review noted that it was not taking into account resistance to the then-recently developed fusion inhibitor enfuvurtide (T-20, Fuzeon). Since then ARVs of two other classes (integrase inhibitors and CCR5 inhibitors) have been developed, as have a number of drugs of established classes that work against HIV with resistance mutations to those classes.
However, even with their own assumptions about resistance, the researchers found that the development of multidrug resistance actually had relatively little clinical effect and that even projecting the model into the far future, which would lead to a 23% rate of multidrug-resistant HIV, would not lead to more HIV cases or deaths than we have currently.
The model includes, buried within its parameters, a number of other interesting assumptions, which are not entirely explained. It assumes, for instance, that if people start therapy at a CD4 count over 350 cells/mm3 their adherence rate will be just under 90% but that if their CD4 count is under 350 cells/mm3 their adherence rate will be nearly 99%. It is not clear what data these inputs are based on.
However, even though some of its parameters are based on somewhat out of date findings, this model, by basing its assumptions carefully on what has actually been observed to happen in gay men, may avoid exaggerated predictions of the success of ‘test-and-treat’ for which some other models have been criticised.
Sood N et al. Treat and treat in Los Angeles: a mathematical model of the effects of test-and-treat for the MSM population in LA County. Clinical Infectious Diseases, early online publication, doi: 10.1093/cid/cit158. See abstract here and supplementary data here (requires payment). 2013.
Van de Vijver DAMC, Wensing, AMJ and Boucher CAB. 'The Epidemiology of transmission of drug resistant HIV', in Hahn B et al (editors), HIV Sequence Compendium 2006/7, pages 17-36. Thoretical Biology and Biophysics Group, Los Alamos National Laboratory. LA-UR 07-4826. 2007.
Health Protection Agency. HIV drug resistance in the United Kingdom: data to end of 2005. Health Protection Report 1(31): 2007.