Some key considerations for surveys of key populations

Editor’s notes: A key challenge for epidemiological research involving key populations is to find a representative sample.  Whereas national surveys such as demographic health surveys (DHS) and PHIA can use the total population to create a sampling frame from which to draw individuals at random, researchers interested in key populations have to use a range of methods, all of which have limitations as well as strengths.  Internet and app-based surveys may accrue large numbers, but may have significant biases in terms of who chooses to answer such questionnaires.  Venue-based sampling allows data to be collected from people who happen to be at the venue at the same time as the researchers.  Respondent driven sampling has become increasingly popular as a method to reach individuals that might otherwise be hard to include in studies.  Increasingly sophisticated statistical methods have been developed to adjust estimates, and in particular their precision, according to characteristics of respondents found in the sample.

This month we have three respondent driven studies that highlight different methodological aspects as well as shedding light on key populations in Africa and Asia.  Hladik et al. conducted a major survey of female sex workers in Kampala, Uganda.  Unfortunately, it has taken some time for this study to be published, as the original questionnaires were completed in 2008/9 and it is plausible that many aspects of sex work have been changing over the past decade.  Nonetheless, the authors succeeded in enrolling almost 1000 female sex workers from the capital city using a respondent driven sampling approach.  The authors paid close attention to methods that could maximize the validity of the data they collected as well as ensuring that participants were protected.  Formative research laid out acceptable incentives to participate, as well as approaches to discuss sensitive or taboo areas and to ensure that all the women understood what was being asked in particular questions.  Finger scanners were used to generate unique identification numbers, so that women could be tracked during the study, and these files were subsequently deleted.  This approach was widely accepted, as it has been in many programmes offering services that benefit from a linked identifier.  However, any approach that creates identifiers for populations that are often discriminated or legislated against needs to be examined critically to ensure that any risks to participants are well understood, particularly for research that is not going to bring any direct benefits to the individual participants.  Although the study’s findings are not particularly surprising, they remind us that sex workers in Kampala need to remain a vital part of the HIV response.  Not only are they affected by a high prevalence of 33%, rising to 44% among those over 25 years old, but they are also subject to horribly high rates of violence including both rape and beating in up to one third of the women in the one month prior to the interview.  The study highlights particular factors that might help identify women in most need of HIV and other services.  Women with less education, who rely entirely on sex work for their income and who have never tested for HIV are all more likely to be HIV positive.

In nearby Malawi, Wirtz et al. point out that many respondent driven samples of key populations, such as that from Hladik et al., are only able to collect data from one particular city or region, and that this can lead to misinterpretation if the results are generalized to whole countries.  The authors conducted a large study of gay men and men who have sex with men in seven different communities across Malawi.  They found considerable heterogeneity leading to an overall estimate that the risk of HIV was approximately twice as high in gay men and men who have sex with men as in the general population of men of the same age.  The study managed to enrol a total of almost 2500 men through respondent driven sampling in the different districts.  However, this was at the expense of having to collect data over a considerable time period, with the study team moving from district to district.  As the authors acknowledge, the risk is that data collected in the most recent time period may not be equivalent to data collected four years previously.  The authors did find that the highest rates of HIV among gay men and men who have sex with men were not always where they have been presumed to be.  In particular tourist areas and some rural areas had higher rates than some of the cities that are usually the focus of key populations programmes.  Once again, the finding that so few gay men and men who have sex with men knew their status and were linked to treatment may not be surprising but is still shocking.  Only 1% of men found to be positive reported that they were aware of their status.  The authors point out the tension between public health and policy in a country where homosexuality is criminalized.  If HIV is to be prevented, this tension will need to be resolved.

The third respondent driven sampling study also highlights heterogeneity.  Verdery et al. used additional statistical methods to study the network characteristics of people who use drugs in two cities in the Philippines (Cebu and Mandaue).  The “small world” phenomenon explains how in more closed settings everyone knows everyone else, and among people who use drugs, many people form part of overlapping networks of needle sharing that allow for rapid propagation of infection.  Developing such methods could allow respondent driven samples to yield greater insights in to the epidemiology of HIV in key populations.  However, issues of representation both of the sample interviewed and of the broader geographic population of interest will remain important.  Quantitative research is certainly essential to understand the population sizes of key populations, and their prevalence, incidence and risk factors of HIV infection.  However, research into policy formation; social science research to understand the larger context of HIV and implementation science to determine how better to offer services that engage individuals in HIV testing and care remain a high priority. 

Burden and characteristics of HIV infection among female sex workers in Kampala, Uganda - a respondent-driven sampling survey

Hladik W, Baughman AL, Serwadda D, Tappero JW, Kwezi R, Nakato ND, Barker J. BMC Public Health. 2017 Jun 10;17(1):565. doi: 10.1186/s12889-017-4428-z.

Background: Sex workers in Uganda are at significant risk for HIV infection. We characterized the HIV epidemic among Kampala female sex workers (FSW).

Methods: We used respondent-driven sampling to sample FSW aged 15+ years who reported having sold sex to men in the preceding 30 days; collected data through audio-computer assisted self-interviews, and tested blood, vaginal and rectal swabs for HIV, syphilis, neisseria gonorrhea, chlamydia trachomatis, and trichomonas vaginalis.

Results: A total of 942 FSW were enrolled from June 2008 through April 2009. The overall estimated HIV prevalence was 33% (95% confidence intervals [CI] 30%-37%) and among FSW 25 years or older was 44%. HIV infection is associated with low levels of schooling, having no other work, never having tested for HIV, self-reported genital ulcers or sores, and testing positive for neisseria gonorrhea or any sexually transmitted infections (STI). Two thirds (65%) of commercial sex acts reportedly were protected by condoms; one in five (19%) FSW reported having had anal sex. Gender-based violence was frequent; 34% reported having been raped and 24% reported having been beaten by clients in the preceding 30 days.

Conclusions: One in three FSW in Kampala is HIV-infected, suggesting a severe HIV epidemic in this population. Intensified interventions are warranted to increase condom use, HIV testing, STI screening, as well as antiretroviral treatment and pre-exposure prophylaxis along with measures to overcome gender-based violence.

Abstract [1]  Full-text [free] access [2]


Geographical disparities in HIV prevalence and care among men who have sex with men in Malawi: results from a multisite cross-sectional survey.

Wirtz AL, Trapence G, Kamba D, Gama V, Chalera R, Jumbe V, Kumwenda R, Mangochi M, Helleringer S, Beyrer C, Baral S. Lancet HIV. 2017 Jun;4(6):e260-e269. doi: 10.1016/S2352-3018(17)30042-5. Epub 2017 Feb 28.

Background: Epidemiological assessment of geographical heterogeneity of HIV among men who have sex with men (MSM) is necessary to inform HIV prevention and care strategies in the more generalised HIV epidemics across sub-Saharan Africa, including Malawi. We aimed to measure the HIV prevalence, risks, and access to HIV care among MSM across multiple localities to better inform HIV programming for MSM in Malawi.

Methods: Between Aug 1, 2011, and Sept 13, 2014, we recruited MSM into cross-sectional research via respondent-driven sampling (RDS) in seven districts of Malawi. RDS and site weights were used to estimate national HIV prevalence and engagement in care and in multilevel regression models to identify correlates of prevalent HIV infection. The comparative prevalence ratio of HIV among MSM relative to adult men was calculated by use of direct age-stratification.

Findings: 2453 MSM were enrolled with a population HIV prevalence of 18·2% (95% CI 15·5-21·2), as low as 4·1% (2·2-7·6) in Mzuzu and as high as 24·5% (19·5-30·3) in Mulanje. The comparative HIV prevalence ratio was 2·52 when comparing MSM with the adult male population. Age-stratified HIV prevalence showed early onset of infection with 11·8% (95% CI 7·3-18·4) of MSM aged 18-19 years HIV infected. Factors positively associated with HIV infection included being aged 21-30 years and reporting female or transgender identity. Among HIV infected MSM, less than 1% reported ever being diagnosed with HIV infection (0·9%, 95% CI 0·4-2·5) and initiated antiretroviral treatment (0·2%, 0·2-0·3).

Interpretation: HIV disproportionately affects MSM in Malawi with disparities sustained across the HIV care continuum. These issues are geographically heterogeneous and begin among young MSM, supporting geographically focused and age-specific approaches to confidential HIV testing with linkage to HIV services. 

Abstract access [3] 


Social network clustering and the spread of HIV/AIDS among persons who inject drugs in two cities in the Philippines

Verdery AM, Siripong N, Pence BW. J Acquir Immune Defic Syndr. 2017 Sep 1;76(1):26-32. doi: 10.1097/QAI.0000000000001485. Epub 2017 Jun 22.

Introduction: The Philippines has seen rapid increases in HIV prevalence among people who inject drugs. We study two neighboring cities where a linked HIV epidemic differed in timing of onset and levels of prevalence. In Cebu, prevalence rose rapidly from under 1% to 54% between 2009 and 2011 and remained high through 2013. In nearby Mandaue, HIV remained below 4% through 2011 then rose rapidly to 38% by 2013.

Objectives: We hypothesize that infection prevalence differences in these cities may owe to aspects of social network structure, specifically levels of network clustering. Building on prior research, we hypothesize that higher levels of network clustering are associated with greater epidemic potential.

Methods: Data were collected with respondent-driven sampling among males who inject drugs in Cebu and Mandaue in 2013. We first examine sample composition using estimators for population means. We then apply new estimators of network clustering in respondent-driven sampling data to examine associations with HIV prevalence.

Results: Samples in both cities were comparable in terms of composition by age, education, and injection locations. Dyadic needle sharing levels were also similar between the two cities, but network clustering in the needle sharing network differed dramatically. We found higher clustering in Cebu than Mandaue, consistent with expectations that higher clustering is associated with faster epidemic spread.

Conclusion: This paper is the first to apply estimators of network clustering to empirical respondent-driven samples, and it offers suggestive evidence that researchers should pay greater attention to network structure's role in HIV transmission dynamics.

Abstract access [4]

HIV transmission [6], Key populations [7]
Africa [8], Asia [9]
Malawi [10], Philippines [11], Uganda [12]
  • [13]