Model suggests there are fewer people with HIV in the US than thought, and more of them on therapy

A study comparing recorded diagnoses of HIV with subsequent records of viral load and CD4 tests suggests that the number of people with HIV in the US could have been overestimated by as much as 45% – and the proportion who are on antiretroviral therapy (ART) with undetectable viral loads could have been underestimated by as much as 50%. There could be as few as 820,000 people with HIV in the US compared with the normally accepted figure of 1.2 million – and up to 55% of those could be on ART and virally suppressed, compared with the most commonly quoted figure of 30%.

Background

The 30% figure comes from a 2011 study (reported here on aidsmap.com). However, it is not just because it is five years out of date that the 30% figure has been questioned. It is also puzzling because with such a low rate of viral suppression, one would have expected a continuing increase in the number of new diagnoses – but this does not seem to be the case, with decreases in diagnoses reported for nearly every group in the last few years. In contrast, in the UK, with roughly two-thirds of people with HIV on ART and virally suppressed – more than twice the supposed US figure – the number of new infections, at least in gay men, continues to slowly increase.

The 30% viral suppression figure in the US has been blamed on a healthcare system that excludes a lot of people from care, and a recent model based on the usually accepted figures has calculated that as many as 45% of people diagnosed with HIV are not in current medical care and that 60% of new infections come from these people.

Figures from individual US cities with high HIV prevalence, however, suggest much higher rates of care and viral suppression. Studies from Seattle and New York presented three years ago at the Conference on Retroviruses and Opportunistic Infections (CROI) suggested that 57% and 44% of people with HIV in those cities, respectively, had an undetectable viral load even then; and a more recent study from San Francisco suggested a figure of 60% there – close to the UK figure and one which, with the addition of a sizeable number of HIV-negative people taking pre-exposure prophylaxis (PrEP), seems to be leading to considerable falls in new HIV infections.

The Seattle researchers found that, in their city at least, 20% of people with HIV reported not to be in care actually were in care – partly because they had moved to another area and were getting care there, but also because the way of calculating who was in care was to count the number of people who had had at least one CD4 count or viral load test during a specific four-month period.

This measure is the same as the one used to make the original countrywide estimate of the proportion of people in care. This comes from the Centers for Disease Control and Prevention’s (CDC) Medical Monitoring Project (MMP), which receives data from a selection of states and cities. The period used in the original 2011 estimate was January to April in that year. However, this may considerably underestimate the number of people actually in care if their CD4 or viral load tests are not reported or if they happen not to attend during that period.

Conversely, deriving the number of people with HIV from positive test results may overestimate the number of people with HIV if they are duplicated for some reason – for instance, under single and married surnames – or if, as is not uncommon, people who say they did not know they had HIV have, in fact, already been tested. A UK survey found that 27% of people reported (from anonymised blood samples) as having HIV but not being tested for it during a sexual health screen were, in fact, already taking ART – they just preferred not to tell their sexual health clinic. Deaths may also be under-recorded, especially if people move abroad, thus overestimating the number of people still alive.

This study

Suspecting that some of these conditions might bias the US statistics, the researchers in this study first directly compared the number of new HIV diagnoses reported in New York City both with the number of reported CD4 or viral load tests in a four-month period, and also the number of people for whom at least two HIV care appointments at least three months apart were recorded in one year.

They found that by HIV test data, 97,128 people were recorded as being diagnosed in New York City and still living there up to the end of 2012. However, by using laboratory test records, they only found 77,334 people diagnosed with HIV living in New York at the same date. This implies that test data could be overestimating the number of people with HIV in New York by 25.6% – or, alternatively, that 25.6% of people diagnosed with HIV in New York are not attending care.

Conversely, the figure for the number of people with at least two recorded medical visits (more than three months apart) in 2012 was 61,159 – implying that using recorded medical appointments as the definition of ‘being in care’ would underestimate the number actually in care by 21%.

Taken together, these two figures could underestimate the proportion of people diagnosed who are in care in New York by as much as 37%.

The researchers were able to do a similar exercise for 19 other jurisdictions in the US ranging from populous and high-prevalence areas (such as the state of Illinois) to rural and low-prevalence areas (such as North Dakota). However, while they had diagnosis and laboratory test reports for these 19 areas (which were chosen because they had the most complete data), they did not have appointment attendance figures.

They therefore provided two estimates: in one, the under-reporting of ‘being in care’ was assumed to be the same as in New York; in the other, they assumed that both lab test results and appointment records were under-reported by 15% each. This would increase the number actually in care, and would mean the difference between this figure and the number diagnosed with HIV from HIV test records would be lower.

They found that if CD4 and viral load results and appointments were recorded as reliably in New York, then the number of people diagnosed with HIV derived from HIV test results would in all cases overestimate the true figure – and in some cases, by a huge margin. For instance, it would mean that the number of people actually diagnosed with HIV living in West Virginia, Minnesota, Hawaii and Washington DC was actually less than half of that derived from HIV test results – and in Illinois, less than one third.

Or – and this is the alternative reading – that in these five areas, an unusually high proportion of people with HIV is not currently in care.

If CD4 and viral load results and appointments were regarded as being recorded 15% less reliably than in New York City, then in some states – Delaware, Iowa, Nebraska, North Dakota and South Carolina – the number of people diagnosed with HIV derived from HIV tests and from CD4 and viral load tests would match. In short, nearly everyone diagnosed with HIV in those states would actually be in care.

In Washington DC and Illinois, however, the true number of people diagnosed with HIV would still be overestimated by 100% if CD4 and viral load test results were a better guide to true HIV diagnosis figures than HIV test results, and by 50% in West Virginia, Minnesota and Hawaii.

Impact on the HIV care cascade

If viral load and CD4 count results are a better guide to the number of diagnosed people living with HIV, then this would have a huge impact on the so-called HIV care continuum or cascade.  

The CDC estimates that there are 1.2 million people living with HIV in the US of whom just over a million (86%) are diagnosed. However, it estimates that only 46.5% of the diagnosed, or 40% of the total, are retained in care, and this means only 30% of the total have a suppressed viral load.

If the researchers’ assumptions are correct, then the true total of people living with HIV in the US is only 819,000 – a third lower than the CDC estimate. But the proportion retained in care would be 84% of the diagnosed and 72% of the total. This in turn would imply that 55% of all people with HIV in the US are on ART and have a suppressed viral load – not that much lower than the UK figure.

In making this new cascade, the researchers used a method of estimating the proportion of people out of care by noting that patients return to care at a fairly constant rate and that this can be used to estimate the number out of care. They used the case of Seattle – a city with excellent records – to estimate the proportion of diagnosed people with HIV who are retained in care. This is 84%, which may seem high, but actually includes people who have not had a CD4 or viral load test result for over four years. When these are removed, the proportion rises to 88% and in New York using the same method, to 91%.

The 55% viral suppression rate may be higher than this in some cities if the same methods are used to weed out ‘false’ records of people living with HIV. It would be 67% in Seattle (about the same as the UK), 70-72% in New York, and as much as 78% in San Francisco, which could explain most of its fall in HIV diagnoses.

Conclusions

The truth is likely to lie somewhere in between, as some areas may have a high proportion of diagnosed people out of care while in others there may be a lot of people recorded as living with HIV who are not actually in that area, alive, or even actual people. In particular, the true number of people living with HIV in the US is likely to lie somewhere between the CDC’s 1.2 million and this study’s 819,000.

The researchers say that their study, “is not an attempt to precisely quantify the number of HIV-infected persons in the US but rather to examine the extent to which the current estimates change when derived from evidence available at the state and local level.” They note that as the CDC received more accurate figures, they revised their own estimate down from 1.178 million in 2008 to 1.145 million in 2010. However, they suggest that this figure is still an overestimate due to duplicate records and undercounting of deaths, and that CD4 and viral load test and appointment records should be used to make more accurate local estimates.