An innovative study presented at the 16th European AIDS Conference (EACS 2017) in Milan two weeks ago used a machine-learning algorithm (originally developed to help astronomers classify galaxies) to tease out whether there were specific groups of gay men within a large national cohort whose sexual risk behaviours followed similar trajectories over time, and if so, whether they were influenced by external factors such as new scientific data.
The analysis of gay men from the Swiss HIV Cohort Study by Luisa Salazar-Vizcaya from the University of Bonn in Germany found that an apparently steady and homogeneous increase in the proportion of men who reported ever having had condomless anal sex with non-steady partners (nscAI) between 2000 and 2016 was in fact due to quite sudden changes in behaviour in three specific groups of gay men who between them numbered less than half of the men in the study. What distinguished the three groups was that their decision to drop 100% condom use with non-steady partners happened at three different time periods.
The Swiss HIV Cohort Study comprises a large majority of all the people diagnosed with HIV in Switzerland. For the purposes of this study, the researchers selected a subgroup of 2614 gay men who were diagnosed with HIV, and so joined the study, between 2000 and 2016. The average length of time men were followed up in the study was nearly nine years and the number of men increased over time, as more were diagnosed than died or dropped out.
Over the cohort as a whole, the proportion of men who said they had ever had nscAI increased from under 20% in 2000 to approximately 60% in 2016. This appeared to be a fairly steady increase, though there was an acceleration in the rate at which men abandoned 100% condom use with non-steady partners after 2008 – the year of the Swiss Statement, the first statement ever publicly issued by researchers about the association between viral undetectability and non-infectiousness.
However, Salazar-Vizcaya emphasised, “Aggregated trends are often over-simplifications”. They therefore used a technique called Hierarchical Agglomerative Clustering to find out whether men clustered into groups that changed behaviour at specific times. This technique, which uses a computer algorithm to generate groups with similarities that are maximal within that group, and minimal between groups, was originally developed by astronomers to sort galaxies into types. The result was a “phylogenetic tree” of behaviour as regards condom use. Phylogenetic trees are more familiar as graphical way of representing genetic differences, but can be used to analyse the degree to which any attribute of individuals within the group is shared by others.
This found that condom usage among about 50% of the men did not change over time. In these individuals, about 25% in any one year reported ever having had condomless anal sex. Note that this actually represents a decrease over time in the frequency of condomless anal sex per individual, because as time goes by, follow-up time increases.
In a second group, comprising 23% of the men, the proportion reporting ever having had nscAI increased steadily from a very low base of about 10% in 2000 to over 90% in 2016. The rate of increase in reported condomless sex took a noticeable turn upwards around 2010.
The other two, smaller groups exhibited much more sudden and dramatic changes in behaviour. In one group, representing 12% of the men, reported nscAI stayed steady at about 10-15% till 2008, then increased steadily to 25% by 2013. After this it took a sudden upward jump to nearly 100% by 2016 – in other words within the space of three years, the situation changed from only one in four of these men answering "yes" to the question “have you ever had condomless anal sex with a non-steady partner?” to nearly all of them answering "yes".
In the last and smallest group, comprising 9.4% of the men, the situation was even more polarised. This group, numbering roughly 250 men, were historically the most consistent condom users: in any one year only 10% of them reported ever having had nscAI till 2015. Then suddenly, a year later, 95% of them were reporting having had nscAI; it was as if this group had, within the space of a single year, taken an en-masse decision to stop using condoms every time.
Further analysis of these groups showed no demographic differences between them in terms of age, ethnicity or education. Only one cluster, the third one, had any distinctive characteristic: its members tended to be more recently diagnosed, with an average date of diagnosis of November 2012 and a quarter of them diagnosed since the mid-2015 (though another quarter were diagnosed before 2005).
Although this is speculative, Luisa Salazar-Vizcaya hypothesised that the second group were people who from 2001 or so onwards were trying to use ‘serosorting’ as a method of avoiding transmitting HIV. An increase in the rate of their reporting nscAI occurred around 2008, the year of the Swiss Statement, as did the first increase in condomless sex reported by group 3. The sudden jump in condomless sex in group 3 happened at around the same time as the first results of the PARTNER study – findings that were widely reported in the gay media. Finally, the sudden increase in condomless sex in group 4 coincides with the publication of the results of the PROUD and IPERGAY pre-exposure prophylaxis (PrEP) studies. Were these men – who had previously been the most consistent condom users – ones who had negative partners who started taking PrEP?
These remain speculations at present, and Salazar-Vizcaya said that qualitative research was needed to discover whether changes in condom use in different men were associated with different findings being publicised.
As a general point, she added, the study showed that it was a mistake to assume homogeneity of behavioural change and therefore of motivations for change in cohorts under study: similar analyses could help to identify how different pieces of knowledge and thought processes influenced different groups, which could lead to more precision and effectiveness when it came to spreading positive health messages.
Salazar-Vizcaya L et al. Highly Dissimilar Patterns of Sexual Risk Behaviour among HIV-positive Men who Have Sex with Men: Clustering Individual Trajectories in the Swiss HIV Cohort Study from 2000 to 2017. 16th European AIDS Conference, October 25-27, Milan, abstract PS12/1, 2017.