New ECDC modelling tool to estimate HIV incidence
The ECDC has recently launched a new model for estimating the number of people living with HIV. Could you explain your role in the development of this model?
SHM had a leading role in the development of this method together with colleagues in the UK and in Switzerland. Initially we did a literature search on what methods were available at that time and could be of potential interest. Then we took two methods which we further developed and tested on data from a couple of pilot countries. These methods were then implemented in a user-friendly tool (with the help of an external programmer) that could be used by epidemiologists and other professionals involved in HIV surveillance. The first version of the tool is now available via ECDC’s website.
Why was there a need for a different method?
The most widely used method currently in use is UNAIDS Estimation and Projection Package (EPP). This method works quite well for countries with generalised HIV epidemics where HIV is found across the whole population in a country. Essentially, in such epidemics, it is sufficient to measure HIV prevalence (proportion HIV-positive) in a representative survey, for instance in households, and then extrapolate to the entire population in a particular country to estimate the number living with HIV in that country. Of course, there are many challenges associated with such an approach, but because the proportion with HIV is large – for example, up to a quarter of the population in southern Africa – you don’t need to survey a lot of people.
This is very different in Europe where HIV is concentrated in certain risk groups like men who have sex with men, drug users, and migrants. EPP could still work in this case provided there are sufficient data about each of the risk groups. As such data are unavailable in many countries, it is very hard to estimate the number living with HIV in European countries. Doing a survey to measure HIV prevalence when the HIV prevalence is low would mean that you need to survey an awful lot of people, in each risk group, to get reliable estimates. Also, information is needed on the number of people in a risk group, which is very hard to determine. For instance, when are you considered to belong to the group of men who have sex with men? How many times do you need to have sex with a man? And what is a drug user? Someone who takes drugs now or in the past, intravenous or not?
On the other hand, many countries in Europe have a well-developed surveillance system that records new HIV diagnoses on an annual basis with information on e.g. age, sex, most likely route of transmission, and CD4 count at the time of diagnosis. These data give an indication of the size of the HIV population, but only those who are diagnosed. A substantial proportion of HIV-positive people are still undiagnosed, either because they were only recently infected or because they never test for HIV.
How does this new method differ to other methods currently in use?
The new method does not use data based on surveys or estimates of the size of risk groups. It only uses routinely available surveillance data on HIV diagnoses, CD4 counts, and whether there was a concurrent AIDS diagnosis. Hence, the major advantage is that the data are already there. There is no need for costly surveys that need to be repeated every few years. Based on these data, the method estimates the number of newly acquired HIV infections per year, the time between infection and diagnosis, and the number who are still undiagnosed.
Of course, as with all methods, you need to make some assumptions. An important ingredient is the rate at which CD4 counts decrease in untreated HIV-positive people, which is fairly well known from data from observational cohorts. Also, you need to make some assumptions on how HIV testing rates may have changed over calendar time. The method is therefore not a black box.
As I mentioned before, two methods have been incorporated in the new ECDC HIV modelling tool. The second method is much simpler and needs fewer data; just a single calendar year with good quality data is sufficient. This method estimates the number of people with very low CD4 counts who are in immediate need of treatment.
By the way, the method is not really new, as similar approaches have been used before, for instance in France and in the UK. What is new is that the method comes with a user-friendly interface and only needs routine data. Maybe even more importantly, the method uses the kind of data that countries in Europe should be collecting and reporting to ECDC. Seeing that collecting such data serves a purpose, i.e., yielding estimates of the total number living with HIV, shows that the data are not just collected and stored, but are also really useful. This may encourage countries to improve their HIV surveillance.
What does the new model mean for figures in the Netherlands, in particular?
For the Netherlands, estimates of the total number of people living with HIV, including those undiagnosed, are now smaller than the numbers that were given before. Earlier numbers were based on estimates by UNAIDS, but it was unclear how they actually derived those estimates. This is something we hope to gain clarity on next year together with UNAIDS.
New figures for the Netherlands are also lower than an estimate published this summer in a study led by RIVM. They used yet another method to arrive at estimates for the Netherlands ,which was also based on prevalence surveys, and on other information such as the number of people in care. This method, multi parameter evidence synthesis (MEPS), is a powerful method as well, because it uses a wide range of available data to estimate the number with HIV. However, a drawback is that it still relies on surveys and population estimates in different parts of the country. In the Netherlands, we have many data for cities such as Rotterdam and Amsterdam and, for example, for Amsterdam, the ECDC method agrees very well with the MPES method. However, outside these cities, data are far sparser and therefore estimates have a larger uncertainty.
All in all, the lower estimate of people living with HIV means that the number with undiagnosed HIV is much smaller than previously reported. That is, of course, good news!
What are the future plans for the ECDC model?
The tool is now available via ECDC for use by anyone who wants to. I really look forward to hearing whether the tool has been useful for other countries in Europe. A group in Spain and even the Ministry of Health in Singapore have already used the tool and their feedback has been extremely useful. Meanwhile, we are planning a few extensions to the tool. One of these is to make it possible to use the tool when data are available over multiple years, but not for the whole epidemic. At the moment, the tool needs a lot of historical data, but if these data are not available now, they never will. The plan is that with ten or fifteen years of data it should also be possible to make estimates of the undiagnosed population, especially when testing rates are high and assuming that most HIV infections are found within ten years of infection.
More information about the new ECDC HIV modelling tool can be found at: http://ecdc.europa.eu/en/healthtopics/aids/Pages/hiv-modelling-tool.aspx