In public health, hard-to-reach populations (HRP), hidden populations [1] or most-at-risk populations [2] are mainly associated with illegal or stigmatizing behaviours such as sex workers (SW), injection drug users (IDU), men who have sex with men (MSM) or homeless people [3, 4]. These groups are usually seen as key populations to be targeted as they have an important role on the spread of communicable diseases like HIV or tuberculosis [5–7]. Thus, understanding how infectious epidemics affect them is crucial for the development of targeted and successful public health interventions. Ideally, a representative sample of the study population should be selected and their socio-demographic characteristics and risk behaviours identified. However, most HRP do not have a sampling frame because their members are “hidden”; hence, one cannot count how many they are [8, 9]. On the other hand, population-based surveys need to be very large to include enough “hidden” members to get precise estimates, which is a limiting factor mainly due to high costs. Therefore, studying HRP presents several difficulties and challenges: (a) it is extremely difficult to use probability sampling strategies to choose members to be included in the sample, and consequently, non-probabilistic sampling methods are mainly used. The great disadvantage of these methods is that since they do not select individuals randomly, the chosen elements may not be representative of the population to which they belong [10]; (b) each HRP has its own behavioural characteristics and deciding on which methods are the most adequate to use in each population is not straightforward [2]; (c) in spite of the financial support most middle and low human development (MLHD) countries have been receiving for infectious diseases control [11], they still receive inadequate funding to reduce the vulnerability of HRP [12]. Consequently, the need persists for documenting trends on the HIV epidemics for these key populations in these regions [13, 14].
Although several studies have been done in reviewing sampling methods [10, 15–18], we did not find any systematic literature review.
The aim of this systematic review was to identify all current methods used to survey the most-at-risk populations of MSM and SW. The review also aimed to assess if there was any relation between the study populations and the sampling methods used to recruit them (that is, to find if there is statistical significance between study populations and sampling methods). Lastly, we wanted to assess if the number of publications originated in MLHD countries had been increasing in the last years.
Data definitions
Populations
In public health, MSM is a term used to define men who engage in sex with other men irrespective of their sexual and gender identities. Commonly, this definition includes men who are identifies as gay, homosexual, bisexual, heterosexual and transgendered [19].
A transgender person (TG) is someone who has a gender identity different from his/her sex at birth. Transgender people may be male to female (female appearance) or female to male (male appearance) [20]. In this systematic review, the term “transgender” refers to the former definition because we did not find any study related to the latter.
Sex worker (SW) is defined as a person who receives money or goods in exchange for sexual services and encompasses male (MSW), female (FSW) or transgender (TSW) people [21].
The term MSM is widely used in the literature not only to mean men who engage in sex with other men but also men sex workers (MSW) and transgender people (sex workers or not) [22]. In order to be consistent with the current literature and for the purpose of analysis, retrieved publications were divided into subgroups consisting of female sex workers and men who have sex with men. Included in this last category are studies of male sex workers, transgendered persons, transgender sex workers and men who have sex with men.
Sampling methods
We call recruitment methods the techniques applied to select a sample of elements from a target population. In this systematic literature review, we identified 11 recruitment methods, which we grouped into three categories. More information about the retrieved methods can be seen in Additional file 5.
The first category includes non-probabilistic sampling methods where the sampled elements are chosen arbitrarily or casually. In these methods, it is not possible to estimate the probability of each element being included in the sample, and consequently, there is no way of making inferences to the population [3]. Methods that encompass the non-probabilistic category are convenience, purposive, snowball and targeted sampling.
The second category includes probabilistic sampling methods. These methods include those where every element in the population has a known probability of being included in the sample; the concept of probability sampling means that a sample has been drawn in a probabilistic way [23], and consequently, reliable estimates are produced and inferences can be made to the study population. Random digit dialling (RDD), cluster sampling, multi-stage sampling and stratified probability sampling (SPS) are included in this probabilistic category.
The third category is the semi-probabilistic; this category includes methods that we believe do not fall in either of the other two because, from a theoretical point of view, it is possible to determine the probability of each element being included in the sample; however, in practice, probabilities cannot be calculated [15, 24] and therefore these methods do not allow making reliable inferences to the (unknown) population. Internet sampling, respondent-driven sampling (RDS) and time location sampling (TLS) are included in this category.
Countries
Countries where the studies were conducted were first classified into eight UNAIDS regions [25]. Later, for the purpose of the analysis, these countries were classified in accordance with the development level, as defined by United Nations Procurement Division (UNPD) [26]. UNPD classifies countries in four levels of human development: very high human development (VHHD), high human development (HHD), medium human development (MHD) and low human development (LHD). For our analysis, we grouped the first two categories into one and named it as “very high and high human development (VHHHD)” and grouped the last two categories into another one and named it “medium and low human development (MLHD)”.