To estimate HIV prevalence in a specific population, it is important to conduct a random selection. Ideally, one would have a database of the entire population and randomly select individuals from it to form a random sample. The prevalence estimated in this sample will then be representative of the prevalence in the target population.

In our case, there is no database of MSM frequenting gay social venues (bars, saunas, backrooms). We must therefore use specific probabilistic sampling methods and account for them in the analyses.

The time-location sampling (TLS) method is well-suited to the population targeted by our study. It requires first identifying all gay social venues in each of the cities studied. This inventory of potentially participating venues in each city was compiled by the regional prevention officers of the National Intervention Team for Prevention and Health in the Workplace (Enipse).

Sampling is then conducted in two stages: first, venues are randomly selected, along with time slots (day and time) for conducting interviews. The more frequented a venue is, the higher the number of time slots to be randomly selected for that venue; second, individuals frequenting these venues are randomly selected.

In the analyses, we accounted for the sampling by specifying survey weights⁽¹⁾. The frequency with which randomly selected subjects visit the establishments is also a factor to consider: the more often an individual visits the surveyed establishments, the more likely they are to be interviewed. Failing to account for this could bias the estimates, especially since, in our study, this frequency of visitation was linked to HIV status.

The use of the time-location sampling method has now become widespread for this type of survey. However, while survey weights are sometimes used in analyses, it is very rare to see studies that account for the frequency of visits to locations as we did in this study.

This study demonstrated that, at the individual level, frequenting these establishments had a significant effect on HIV status in Paris, Lille, and Nice. The more frequently these establishments were visited, the higher the probability of being HIV-positive. Hence the importance of accounting for venue attendance among randomly selected individuals when estimating HIV prevalence. In Montpellier and Lyon, no significant effect of venue attendance on HIV status was observed.

We compared HIV prevalence results with and without the use of survey weights and venue attendance. In the three cities where venue attendance was positively associated with serostatus, the HIV prevalence estimates obtained without accounting for sampling and attendance differed from those obtained by accounting for these two factors. For example, in Nice, the HIV prevalence estimated in this way was 25% and 17%, respectively. This difference is less pronounced for the other two cities.

Furthermore, the variance in estimates of overall HIV prevalence across the five cities was lower when sampling and attendance were taken into account (approximately 50% lower). Failing to account for these parameters can therefore lead to the erroneous conclusion that there are significant differences in HIV prevalence between cities.

One of the objectives of this paper was to provide detailed information on conducting a survey targeting hard-to-reach populations who frequent identifiable establishments, in order to equip researchers conducting such studies with the necessary tools. Thus, we described the preparatory work (identifying locations and gathering the information needed for random selection), the recruitment of participants, and the questions to ask in order to estimate the number of visitors to these locations. In addition to describing the methodology, we estimated the sampling design effect. This information is very useful because it allows those who use this method to calculate sample sizes when implementing similar sampling designs.

Finally, this methodology is applicable to any survey targeting people who frequent identifiable locations. It has already been applied by Santé publique France to the population of drug users.

It could be considered for application to other populations, such as migrants frequenting reception centers, recipients of food assistance, or homeless individuals frequenting emergency shelters.