Mapping and size estimates of female sex workers in Cameroon: Toward informed policy for design and implementation in the national HIV program

Background Due to high HIV prevalence among Female Sex Workers (FSWs) in Cameroon (36.5%), this population is especially vulnerable to HIV acquisition and transmission nationwide. Though being prioritized in the national HIV response, it would be relevant to generate statistics on the number of FSWs in order to guide HIV interventions among FSWs. Our objective was to estimate the size of FSWs within hotspots of Cameroon. Methods A cross-sectional study was conducted from September-November 2015 in selected cities in Cameroon: Bafoussam, Bamenda, Bertoua, Buea, Douala, Kribi, Limbé, and Yaoundé. A programmatic mapping was used, consisting of interviews with secondary key informants (KI) to identify hotspots of FSWs and their respective estimated numbers. Validation of size estimates was done by interviews with FSW at each hotspot. Size estimations in the councils mapped were extended to others not mapped using a Poisson regression model. Results A total of 2,194 hotspots were identified: Douala (760), Yaoundé (622), Bamenda (263), Bafoussam (194), Kribi (154), Bertoua (140), Limbé (35), and Buea (26). The estimated total number (range) of FSWs was 21,124 (16,079–26,170), distributed per city as follows: Douala 7,557 (5,550–9,364), Yaoundé 6,596 (4,712–8,480), Bafoussam 2,458 (1,994–2,923), Bamenda 1,975 (1,605–2,345), Kribi 1,121 (832–1,408), Bertoua 1,044 (891–1,198), Buea 225 (185–266), and Limbé 148 (110–148). The variability of estimates among cities was also observed within the councils of each city. The national predicted estimate of FSW population was 112,580 (103,436–121,723), covering all councils of Cameroon. An estimate of 1.91% (112,580/5,881,526; 0.47%-3.36%) adult female population in Cameroon could be sex workers. Conclusion There are considerable numbers of FSW in major cities in Cameroon. There is a need to prioritize interventions for HIV prevention toward this population in order to limit the burden of HIV sexual transmission nationwide.

Introduction HIV prevalence is considerably higher among sex workers compared to the general population, with an estimated 15% of HIV positive cases in the adult female population attributable to female sex work and about 106,000 attributed deaths globally [1]. Sub Saharan Africa (SSA) has the highest attributable fraction (17.8%) of HIV transmission due to female sex work, with 98,000 estimated AIDS related deaths [2]. An analysis of 16 SSA countries showed a pooled prevalence of 36.9% among sex workers; and 14-fold higher compared to other women globally [3]. In Cameroon, HIV prevalence is 3.4% [4]; with 53,057 new infections according to the National AIDS Control Committee [5]. In line with the sustainable development goals (SDGs), Goal 3, Target 3.3: "By 2030, end the epidemics of AIDS, tuberculosis, malaria and neglected tropical diseases and combat hepatitis, water-borne diseases and other communicable diseases", the national response has been strengthened in 2016 and 2017, with the support of partners (Grant Project No 993 CMR-H-MoH) [6]. With a 36.5% HIV prevalence rate among female sex workers (FSW), this population plays a very important role in the overall HIV transmission dynamics in Cameroon [7], and it is a priority population for programmatic interventions [8]. Cameroon is therefore committed to improving its HIV/AIDS response towards Key Populations (KPs), including FSWs. In the National HIV/AIDS Strategic Plan, the Government of Cameroon has henceforth prioritized programmatic interventions towards HIV prevention among FSWs [5]. It has thus become essential to quantify the size of FSWs for effective planning, monitoring and evaluation of programmatic impacts nationwide. Such information is critical for decision-making and for effective HIV prevention for FSWs both at the macro and micro levels. Although there have been attempts to estimate the size of FSWs in Cameroon, the findings were on one hand limited to selected areas in cities at the macro level [9]. Although helpful for gathering particular information, these previous approaches are unable to provide reliable size estimates. However, methods that estimate population sizes (census and enumeration, nomination techniques, multiplier methods, capture recapture) do not adequately describe the distribution and locations of populations in a geographical context [10].
On the other hand, efficient programming requires information on the estimated population size for FSWs at the aggregate level as well as granular information at the local level. Therefore, a size estimate of FSWs in Cameroon would help in ensuring that prevention services are planned and provided based on local needs.
Jing et al., compared the multiplier method and the generalized network scale-up method; and recommends the latter when sampling frames for the general population provided accurate demographic information is available [11]. Similarly, Yu et al., reviewed the methodological approaches from implementing population size estimates in 13 Asian countries [12]. After showing that all sampling methods rely on theoretical assumptions that are difficult to meet in practice, a formative assessment was recommended to help inform the appropriateness and feasibility of different size estimation methods [13]. Respondent-Driven Sampling (RDS) demonstrates sensible results when used to estimate the size of known networked populations from the National Longitudinal Study of Adolescent Health, and also when used to estimate the size of a hard-to-reach population at high risk for HIV [14]. On the other hand, Johnson et al., provided several country examples of how service multiplier methods have been used in respondent-driven sampling surveys and provides guidance on how to maximize the use of this method [15]. UNAIDS and WHO have also provided guideline for estimating the size of populations most at risk to HIV [16]. In Cameroon, a mixture of methods including 'wisdom of the masses', unique object method, and service multiplier, were used for the estimation [9]. The estimation approach used in the latter study was based on programmatic experience that populations at increased risk for HIV acquisition and transmission generally congregate and/ or meet clients in specific geographic settings [17]. A main difference between programmatic mapping and other KP mapping is that programmatic mapping is based on micro level, and identifies sites where programs can reach members of a KP (in this context sex workers). Moreover, size estimates of FSWs can be readily translated into outreach targets for local service delivery providers and planned interventions. The approach was originally designed by researchers at Center for Global Public Health (CGPH) of the University of Manitoba; and has been adapted for use in various countries in South and South-east Asia and in Africa. With minor changes made to adapt the approach to local contexts and research needs, the methodology could provide more valuable information on similar populations worldwide. The original approach for programmatic mapping was performed as previously reported [17], while applications of programmatic mapping and its implications were further described [18][19][20]. Programmatic mapping was borrowed from the Priorities for Local Aids Control Efforts (PLACE) method, first conceived in 1999 and implemented in a township in Cape Town, South Africa [21]. Its validity and reliability has been evaluated and the method promoted alongside with programmatic mapping by the Global Fund, WHO, and UNAIDS. Many studies have been carried out on programmatic mapping. Sabin et al., reviewed and showed that programmatic mapping was used in 71 countries for KP (FSW, MSM, PWID, transgender) and in 27 of the 71 countries, this approach was used on FSWs [22].
With the goal to set up an effective planning and implementation based on national estimates using reliable and feasible predictions in Cameroon, our study aimed at estimating the size of FSWs in order to guide the national HIV program design and implementation. Specifically we sought to: (a) estimate the total number of FSWs (population size) in selected sites; (b) develop national predictive estimates of FSWs; and (c) estimate the proportions of FSWs among adult female in Cameroon.

Study design
A cross-sectional study was conducted from September 5 th to November 30 th 2015 in locations of Cameroon where high-risk activities related to HIV transmission occur (hotspots) among FSWs. A FSW was defined as any female who exchanges sexual activity with a man in return for money or benefits, irrespective of site of operation (street, bars, hotel, etc.).

Study sites
Hotspots were in the following selected cities: Bafoussam, Bamenda, Bertoua, Buea, Douala, Kribi, Limbé, and Yaoundé. Even if a hotspot may have been previously mapped, it was essential to validate the hotspots by re-confirming their current status as well as identify any new hotspots that have emerged since the previous mapping.
Several criteria were used to prioritize the cities/towns for the mapping, taking into account the Government priorities for intervention towards FSWs. The criteria included: (i) the size of the city (both large and small cities); (ii) the prevalence of HIV in the city; and (iii) the presence of existing programs for FSWs that could be leveraged.

Study procedure
The target area for mapping was defined by selecting the cities to be mapped. Once up-to-date geographical maps were acquired, the entire area under study was divided into smaller data collection units, referred to as "zones", which were demarcated on the maps and developed using a combination of existing geographical or administrative divisions (e.g. councils/subdivision) and physical landmarks. The methodology uses two sequential steps to identify hotspots, known as locations where FSWs seek sexual partners and/or engage in sexual activity: Level 1 (L1) included systematic information gathering from secondary key informants (KIs), who were persons in direct contact with FSWs. Based on the answers provided by KIs, a list of potential hotspots was developed. Level 2 (L2) involved the validation and profiling of identified hotspots through interviews and focus group discussions with FSWs.

Implementing team
A Technical Advisory Group (TAG) was set up to oversee the execution of this study and to support the study implementation (Decision N˚1106/D/MINSANTE/CAB/SG/STBP/NACC/ SP/SPE of June 15, 2015). The TAG, headed by the NACC, consisted of members from World Bank, UNAIDS, USAID, University of Manitoba's CGPH, local technical experts, local stakeholders (donors and non-governmental partners), representative of NGO working with FSWs. The TAG was under the supervision of the Minister of Public Health.

Key informants
Members of the TAG agreed to the identification of relevant key informants (KIs). KIs are persons who are likely to have information on the profiles of the locations and estimates of the number of participants engaging in high-risk activity. Based on their involvement in the relevant high-risk activity mapped, KI's were classified into two types: Primary KIs were FSWs and were used for Level 2 data collection. Secondary Key informants were persons who are in direct contact with FSWs, or those likely to provide information on their location and are used for Level 1 data collection. Some of the following people acted as secondary KIs: MSM, Drug users (injectable and non-injectable), Taxi drivers, Local food sellers, Brothel owners/madams, Watchmen/security staff, Nanga boko (Street children), Hotel/lodge workers, Bar workers/owners/patrons, Porters, Petty shop owners, Cyber man (a controller of an internet network service shop), Drug peddlers/pushers, Pharmacists, Lottery sellers, Streets/toilets sanitary workers, Networks of MARPs; NGO staff, Health care service providers, Gov./law enforcement officials, Street families, Beggars, Public/private transport staff, Callbox men (mobile phone airtime vendor) Construction workers/laborers (Fig 1) at various places such as parks, transit stops, shopping malls, and night clubs.

Level 1 activity
Level 1 was the first phase of field activity, in which information was collected on the various high-risk activities and individuals within each zone, by interviewing secondary KIs. This enabled the development of a comprehensive list of hotspots where FSWs congregate within each zone. The major aim was to generate a list of hotspots where such activities take place and a range of estimates (minimum and maximum) of participants in high-risk activity. For each mentioned location, KIs were asked a set of questions about the characteristics of the hotspot (public place, brothel, night club, etc.) and an estimate of the number of FSWs who can be found there (minimum and maximum). The level 1 interview was conducted at various major markets, parks, streets, etc., within a specified zone to administer questionnaires on high-risk activities taking place within that zone.

Level 2 activity
Level 2 consists of validating information collected in Level 1, by visiting each "hotspot", and interviewing FSWs (primary KIs) who were present at those hotspots. Hotspots were validated by determining whether they still exist or are currently inactive, and by identifying FSW typologies and estimating the number of each FSW typology at that hotspot on average and during peak times. Level 2 fieldwork also involved collecting geo-coordinates (using GPS devices) for each active hotspot validated. Using hotspot lists prepared in Level 1, teams validated hotspots by interviewing FSWs at those hotspots at the time that those hotspots were active. Following planning from the field supervisor, mobilizers identified FSW and introduced them to the team. Validation of hotspots in Level 2 was to confirm two key areas of inquiry: firstly the existence of an active hotspot and secondly a more accurate estimate per hotspot. All hotspots found in L1 were validated in L2; with interviews being conducted in all hotspots. In very few cases (8.2%), primary KIs were not available, and secondary KIs were interviewed instead. Each hotspot was visited twice, at least, before a decision to validate information from secondary KIs was made. Because KPs are often stigmatized and marginalized, the team worked with community members throughout the study by hiring these individuals as peers workers (social mobilisers), in order to ensure more access to hidden segments of sex worker and community participation.

Data quality assurance
A monitoring and quality assurance system was designed together with a timeline to complete the data collection activities. Supervisors monitored the fieldwork by visiting each field team on a daily bases. Twenty percent (20%) of hotspots were randomly selected daily for validation by the supervisor. For both levels (L1 and L2), monitoring forms were used to assess the quality of work by site coordinator, supervisor and data entry operators. The forms were filled by the respective team members to keep track and record of the daily activities, and then were submitted to the respective line manager. About 30% of the forms were rechecked, edited for wrong entry, out of ranges values, skip patterns, irrelevant data, and spelling of text answers, then were validated by supervisors. Feedback was provided to interviewers on the basis of the errors identified in the forms. A double entry of data was performed for consistency.

Data management and analysis
After the data set was properly coded, the entire data was entered into CSPro 6.2. Once the data entry was completed, it was exported to Microsoft Excel 2013 and IBM-SPSS Statistics (Version 21. Armonk, NY: IBM Corp.) for statistical analysis.
Calculation of estimates of FSWs. The crude estimates were obtained as an average of minimum and maximum number of FSWs on normal/peak days, and were adjusted to account for mobility (duplicates) of FSWs. For example, if FSWs are solicited at two or more hotspots, the number of FSWs estimated for the geographic unit that contains those hotspots will increase [23]. For a zone i, let C i be the crude (unadjusted) estimate of FSWs, obtained by summing the estimate of FSWs from each hotspot of the zone; P i the probability of a FSW operating in several hotspots, M i the mean number of places where a FSW operated. The adjusted estimate N i (accounting for the mobility of FSWs) was derived using the following simple mathematical function: Estimates from all zones in a city were subsequently summed up to generate estimates for the whole city. This size estimation formula was initially developed by the University of Manitoba, and adopted in programmatic guidelines by Global Fund in 2010; and this formula was previously used [24].
Predicting national estimates of FSWs. Council-wise FSW population size estimates were obtained using a Poisson regression modeling from the mapped councils. The dependent variable was estimates of FSWs (E i ); while the square root of adult population (A 1=2 i ) and logarithm of adult population (Log(A i )) were used as independent variables (predictors) from the 26 councils of the eight cities (the index i was for each council). The national estimate was obtained as the sum of predicted estimates of FSWs of each of the 360 councils in Cameroon (including the 26 councils used for estimation). The predicted interval for the national estimates (NE) was obtained as NE � 2s ffi ffi ffi ffi m p [24] where σ was the standard deviation of the predicted estimates of each of the m councils (m = 360); s ffi ffi ffi ffi m p was the standard deviation of a sum, and s= ffi ffi ffi ffi m p the standard deviation of a mean. Residuals from the model diagnostics showed that predictions from cities mapped were close to the estimates obtained from the mapping.

Developing hotspot maps
The entire data, which has been collected, validated and triangulated, was displayed on zonal maps that were later merged into city maps. A combined city map was further developed using the geo-coordinates collected at Level 2 to show the spatial distribution of FSW population and its estimates. For purposes of confidentiality, hotspot maps for every city developed as part of this study were only used for programmatic perspectives. The council/sub-division maps were also developed (please contact National AIDS Control Committee (NACC) to access data, more information within the Data Availability Statement in order to identify hotspots (Fig 1). Fig 1 shows the full representation of the programmatic methodology, from literature review to report development and dissemination.

Ethical considerations and safety of FSWs
Ethical clearance for the study was obtained from the "Comité National d'Ethique de la Recherche pour la Santé Humaine" (N˚2015/07/617/CE/CNERSH/SP of July 17 2015). Administrative authorizations were obtained from each local representative of the Minister of Public Health for the cities involved. Programmatic mapping, by its nature, gathers information about places and locations where risk activities take place, so potential hazards to the field teams and respondents were addressed. Divisional Officers, Police Heads, Mayors were notified. Identity cards (temporary) signed by health or police authorities were issued to field workers to facilitate fieldwork. Considerable efforts were made to preserve the safety of respondents because sensitive questions in a public place could cause discomfort or even put respondents at risk. The teams were trained to ensure that interviews were conducted with respect to privacy. A non-identifying coding system was used to track study data while assuring non-disclosure of participants' identity.
Recruitment of participants was conducted only after describing the study procedures and obtaining written informed consent. During the process of obtaining informed consent, participants were clearly informed that participation is voluntary and that non-participation would have no negative consequences in terms of access to programs or services. In addition, names of respondents were not asked for purposes of confidentiality and anonymity.
All survey-related materials (e.g. completed questionnaires and maps) were kept in a secure and locked cabinet at the survey office. Table 1 shows the number of zones where FSWs can be found in various cities in Cameroon as well as the number of interviews conducted, average interview per zone and the number of hotspots identified, as compared to the Cameroon Central Bureau of the Census and Population Bureau (BUCREP) [25].

Hotspots identification and distribution of secondary key informants
Cities with larger populations had the larger number of zones. Of the 100 zones, Douala and Yaoundé each had 36 zones while Bafoussam and Bamenda each had 6 zones and finally, Limbé, Bertoua and Kribi had 4 zones each.
Douala was the city with the largest number of hotspots identified (37.4%) and Yaoundé, having comparable number of interviews as Douala, had less hotspots (29.4%). Buea had the lowest number of hotspots (2.2%). The average number of interviews among the cities included in this study varied between 57.5 in Bamenda and 77.2 in Bafoussam.

Prediction of FSWs estimation in the studied sites
The predicted FSWs per 1000 adult male varied from 13.91 (Littoral) to 30.88 (East). The regions of South and East had the highest number of FSWs per adult female with 27.47 and 28.28 respectively, then highest percentages of FSWs (see Table 4).

Locations of FSWs
The hotspots appeared to be unevenly distributed across and within cities and often regrouped into clusters. Most hotspots were located in the urban area of each city. The maps generated from this study found that most cities had hotspots concentrated in the commercial centers and along the major roads and highways leading in and out of the cities. In larger cities such as Douala, this pattern was persistent but did not preclude activities from being dispersed to other areas of the city. In Douala and Yaoundé, some hotspots were also located in rural settings.

Discussion
Estimating the size of FSWs in a country with generalized HIV epidemics is of great importance to set-up programmatic interventions toward a better prevention of HIV transmission within such a high-risk population within a resource-limited setting context. To our knowledge, this study is the first in Cameroon to estimate the size of FSWs at micro (hotspot) level for programming purposes. The comparison of the size of FSWs between different cities was better appreciated by calculating the ratio of FSWs per 1000 adult male/female. These estimations showed the highest ratio of FSWs in Kribi (55.39) and Bertoua (41.85). Kribi is a touristic city visited and inhabited by many wealthy foreigners because of popular beaches. Many large scale developmental projects are currently being implemented in Kribi, such as its port. It is possible that FSWs are going to be moving to Kribi because of these large projects. Many hotspots are found in the area neighboring an oil producing factory showing a concentration of FSWs. Sex work does not seem to be stigmatized in Kribi, as culturally, the population may consider this work as an actual profession. On weekends, many FSWs residing in Douala and Yaoundé may go to Kribi to solicit clients. Estimates of female sex workers Many large scale developmental projects (such as the Lom Pangar Dam) are being implemented in Bertoua. There are also many operating forestry enterprises exploiting wood in that city. Many truckers (also high risk populations) who transport goods from Cameroon to other Central African countries reside in Bertoua.
The high rate of FSWs per 1000 adult male (30.06) in Bafoussam may be explained considering that, in the west region, funerals and other traditional ceremonies are usually organized on weekends. Thus, many FSWs may use this opportunity to congregate in Bafoussam in search of clients.
Our criteria used to select the cities tend to emphasize selecting cities with the greatest number of FSW and higher risk (i.e. where normally FSW programs would be implemented). This can potentially lead to bias in selecting the cities with larger populations of FSW for mapping, which can overestimate the national FSW population when generalizing from the 8 cities to the rest of the country, even taking into account population proportions. This trade-off between practical criteria to city selection and generalizability has been taken into account by using the square root of adult population (A 1=2 i ) and logarithm of adult population (Log(A i )) as covariates in a Poisson regression; in order to avoid an exponential growth. Odek et al., used a linear regression in Kenya, with the estimates of FSW per 1000 adult population as dependent variable; adult population and its square as two independent variables [24]. The density of FSWs (per capita) being very different across cities in Cameroon is clearly an indication that in addition to population size, many other variables should enter into play as other predictors when estimating the size of FSWs. Some of these variables include: (a) prevalence of HIV, (b) whether the area is urban or rural, (c) being a touristic site, (d) its economic opportunities, and (e) location along a major highway. Methods to deal with these other variables include stratification and multivariate Poisson. In either case, the sample size (n = 26) is insufficient to carry out such analyses.
The  (23,477) in urban areas (nearly forty percent of FSW are in urban areas) [9]. The cities involved were Bamenda, Bafoussam, Bertoua, Douala, Kribi, Ngaoundere and Yaoundé. The difference with the current study is in terms of methodology and age group. It should be noted that Papworth et al used a mixture of methods including 'wisdom of the masses', unique object method, and service multiplier. Unlike programmatic mapping whose methodology is based on hotspots (micro level), Papworth et al. focused at macro level with 15-45 years as age group. Furthermore, Papworth et al., estimated that 1.88% (1.15-2.61%) of the female population in Cameroon are FSWs comparable to 1.91% (0.5-3.43%) obtained from the current study for adult female [9]. Odek et al., found a national urban FSW population estimate of 138,420 (107,552-169,288) covering all towns with a population of over 5,000 in Kenya [24]. They also found that approximately 5% of the urban female population of reproductive age in Kenya could be sex workers [24]. For this study, out of 26 councils, 5 were rural: Bafoussam 3, Bamenda 1, Bamenda 3, Bertoua 2, and Yaoundé 7; thus the sample of cities here is mixture of urban and rural. Since the predictions are based on estimating the model for only 26 councils for eights selected cities (not on hotspots level), fits of this model could be improved if more councils were mapped.
The South and East regions have higher predicted proportions of FSWs with 2.75% and 2.83% respectively. These results may partially explain the high HIV prevalence in these regions; 7.2% in the South region and 6.3% in the East region (National Institute of Statistics and ICF International). These results can serve in predicting the number of condoms to be distributed per region, as well as the minimum package of interventions for each region.

Evidence-based programmatic implications from the study
The current study estimated the number of FSWs that could be reached by service delivery providers working with FSWs in the eight cities with the highest priority for HIV prevention in the country. Our findings are therefore of relevance for implementing interventions following the Continuum of Prevention, Care and Treatment of HIV/AIDS with Key Populations (CHAMP) project in Cameroon.
Within the UNAIDS' 90-90-90 targets, Cameroon will considerately increase the number of people screened for HIV, and the ART program is expected to double the number of people receiving antiretroviral treatment. The results are expected to identify priority areas of intervention and guide the choice of community-based organizations (CBO) responsible for the implementation. Recommendations will serve for other HIV-related studies such as integrated biological and behavioral surveillance (IBBS) studies in the country. Because FSWs constitute a mobile population, planned interventions should start as early as possible, using current mapping as a routine exercise in order to help NGOs to achieve effective interventions. Mapping of other councils should continue in order to have reliable predictive estimates of the size of FSWs at the entire national level. Implications of our findings therefore include providing: (a) denominators that can be used for setting coverage targets and program reach; (b) critical information that can be used as indicators for implementing agencies; and (c) sampling frame for operational studies. The Ministry of Public Health in Cameroon has endorsed these estimates of FSWs for planning, budgeting, monitoring and evaluation of interventions in Cameroon in subsequent years [27][28][29].

Study limitations
Some limitations could be reported from our study. For instance, our estimation was based on information from KIs, which is an approach less accurate compared to a census design, which is generally unfeasible in Cameroon since FSWs are stigmatized and they fear being recognized by family members and friends. Also because sex work is illegal, they also have fear of being arrested by the police, thus making it difficult to reach them. In addition, this study could not capture hidden FSWs who do not operate in public places (e.g. home-based FSWs), virtual places, and internet-based facilities. The adult population alone does not suffice in sizing FSWs in a city. Thus, predictions of this study can only be considered proxy of the true size of FSWs in Cameroon. The entire area under study was divided into smaller data collection units, referred to as "zones", in order to generate findings closer to the census approach and to maximize the probability of not missing any hotspot. Finally, the sample size was small to carry out multiple Poisson regression with many other covariates.
The formula used herein for estimation accounts for mobility of FSW between hotspots in a zone. However, it does not make any adjustment for mobility between cities, which infers likelihood for double counting. In contrast, this bias is highly minimal because very few FSWs move from one city to another to find clients. Further investigations should be undertaken to account for such evidence and their programmatic implications.

Conclusions
There are considerable numbers of FSWs in major cities in Cameroon. There is thus a need to prioritize interventions for HIV prevention towards these populations in order to limit the burden of HIV sexual transmission nationwide. Moreover, our findings would serve as baseline for informing the national HIV program on planning, implementation, monitoring and evaluation targeting FSWs in Cameroon.