Artigo Revisado por pares

Capture-recapture for estimating the size of the female sex worker population in three cities in Côte d’Ivoire and in Kisumu, western Kenya

2010; Wiley; Volume: 15; Issue: 12 Linguagem: Inglês

10.1111/j.1365-3156.2010.02654.x

ISSN

1365-3156

Autores

Béa Vuylsteke, Hilde Vandenhoudt, Lilian Langat, Gisèle Semdé, Joris Menten, Fredrick S. Odongo, Ayubu Anapapa, Lazare Sika, Anne Buvé, Marie Laga,

Tópico(s)

HIV/AIDS Research and Interventions

Resumo

Objective To estimate the female sex worker population size in three cities in Côte d'Ivoire and in Kisumu, Kenya. Methods Capture–recapture was used, calculating size estimates by first 'tagging' a number of individuals and, through an independent recapture, calculating the proportion of overlap. The same procedures were used in all four cities. In the first phase, members of the target population were 'captured' and 'marked' by giving them a capture card. Six days later, in the same places and at the same time, a second sample was 'captured', which comprised a certain number of people who were captured in the first round. During the exercise, questions were asked to estimate the coverage of the sex worker clinics. Results Using capture–recapture, the estimated number of female sex workers was 1160 in Yamoussoukro (95% CI 1053–1287), 1202 in Bouaké (95% CI 1128–1279), 1916 in San Pedro (95% CI 1809–2030) and 1350 in Kisumu (95% CI 1261–1443). The proportion of female sex workers in Côte d'Ivoire who had visited the clinic ranged from 15% in Yamoussoukro to 30% in San Pedro and was 34% in Kisumu. Conclusions Capture–recapture was successfully applied to estimate the population size of female sex workers. These estimations were urgently needed to help mobilize an increased response to HIV, to assess programme coverage and to estimate potential impact of the targeted intervention. Objectif: Estimer la taille de la population des professionnelles du sexe dans trois villes de la Côte d'Ivoire et à Kisumu au Kenya. Méthodes: La méthode de capture-recapture a été utilisée, le calcul des estimations de taille a d'abord été effectué par le 'marquage' d'un certain nombre d'individus et par une recapture indépendante, le calcul de la proportion de chevauchement. Les mêmes procédures ont été utilisées dans les quatre villes. Dans la première phase, les membres de la population cible ont été"capturés" et "marqués" en leur octroyant une carte de capture. Six jours plus tard, dans les mêmes endroits et au même moment, un second échantillon a été"capturé", qui comprenait un certain nombre des personnes capturés lors du premier tour. Au cours de l'exercice, des questions ont été posées pour estimer l'étendue de la couverture en cliniques pour les professionnelles du sexe. Résultats: Sur base de la méthode de capture-recapture, le nombre estimé de professionnelles du sexe était de 1160 à Yamoussoukro (IC95%: 1053-1287), 1202 à Bouaké (IC95%: 1128-1279), 1916 à San Pedro (IC95%: 1809-2030) et 1357 à Kisumu (IC95%: 1261-1443). La proportion de professionnelles du sexe en Côte d'Ivoire qui avaient visité une clinique variait de 15%à Yamoussoukro à 30%à San Pedro et de 34%à Kisumu. Conclusions: La méthode de capture-recapture a été appliquée avec succès pour estimer la taille de la population des professionnelles du sexe. Ces estimations étaient urgemment nécessaires pour aider à mobiliser et à accroître la réponse au VIH, évaluer la couverture du programme et estimer l'impact potentiel des interventions ciblées. Mots-clés: professionnelles du sexe, estimation de la taille, capture-recapture, cartographie, prévention du sida, Afrique. Objetivo: Estimar el tamaño de la población de trabajadoras sexuales en tres ciudades de lCosta de Marfil y en Kisumu, Kenia. Métodos: Se utilizó el método de captura y recaptura: calculando el tamaño estimado haciendo primero un "marcaje" de un número de individuos y, posteriormente, mediante una recaptura independiente, calculando la proporción de solapamiento. Se utilizó el mismo procedimiento en las cuatro ciudades. En la primera fase, los miembros de la población diana fueron "capturados" y luego "marcados" dándoles una tarjeta de captura. Seis días más tarde, en los mismos lugares y a la misma hora, se "capturó" una segunda muestra, que incluía a un cierto número de mujeres que habían sido "capturadas" durante la primera ronda. Durante este ejercicio, se realizaron preguntas para estimar la cobertura de las clínicas para trabajadoras sexuales. Resultados: Utilizando el método de captura-recaptura, el número estimado de trabajadoras sexuales era de 1160 en Yamoussoukro (95%IC 1053-1287), 1202 en Bouaké (95%IC 1128-1279), 1916 en San Pedro (95%IC 1809-2030) y 1357 en Kisumu (95%IC 1261-1443.) La proporción de trabajadoras sexuales en Costa de Marfil que había visitado una clínica estaba entre un 15% en Yamoussoukro y el 30% en San Pedro y era del 34% en Kisumu. Conclusiones: El método de captura-recaptura fue exitoso cuando al aplicarse para estimar el tamaño poblacional de las trabajadoras sexuales. Estas estimaciones se necesitaban de forma urgente para ayudar a movilizar una mayor respuesta frente al VIH, para evaluar la cobertura del programa y para estimar el posible impacto de una intervención dirigida. Palabras clave: trabajadoras sexuales, estimación de tamaño, caputura-recaptura, mapeado, prevención del SIDA, África. Sex workers and their clients are a critical group in the spread of HIV infection, in every region in the world (Vuylsteke et al. 2008). Targeted interventions that aim to increase condom use and reduce transmission of STIs and HIV infection among sex workers and their clients are feasible and effective (Laga et al. 1994; Steen et al. 2000; Alary et al. 2002; Ghys et al. 2002; Wi et al. 2006). The effects of sex worker interventions on the incidence of HIV in the general population not only depend on the effectiveness of the interventions in increasing condom use and reducing STI and HIV infection among sex workers, but also on the stage of the epidemic and on the coverage of the interventions (Van Vliet et al. 1998). Coverage by interventions is a key parameter in the evaluation of any programme and requires a fairly accurate estimate of the size of the target population. However, reliable estimates of the size of high risk populations who are involved in covert, stigmatized and socially ostracized activities and who are highly mobile are difficult (Vandepitte et al. 2006). Different methods have been used to estimate the size of sex workers populations, including mapping and census, the multiplier method and capture–recapture. The latter method was originally developed to count and track wildlife populations by capturing, tagging and recapturing. Since the early 1990s the method has been used in several countries to estimate the size of hidden or difficult-to-reach human populations such as sex workers, men who have sex with men, homeless persons and intravenous drug users (McKeganey et al. 1992; Kruse et al. 2003; Khan et al. 2004; Minh et al. 2004; Geibel et al. 2007; Vadivoo et al. 2008). In 1992 the Institute of Tropical Medicine (ITM), Antwerp, Belgium, set up a dedicated clinic for female sex workers in Abidjan, Côte d'Ivoire, in close collaboration with 'Projet RETRO-CI' of the Centers for Disease Control and Prevention (CDC), Atlanta (USA) and the Ministry of Health, Côte d'Ivoire. The clinic offers HIV prevention and care services, including behaviour communication through peer education, condom promotion and STI treatment to sex workers in a non-stigmatizing and confidential environment. Since 2000 additional sites have been established and by 2008 a total of 13 prevention and care service sites were operational in eleven major towns of Côte d'Ivoire. Since 2004 this programme has been implemented by Family Health International, with technical support from ITM and financial support from the President's Emergency Plan for AIDS Relief (PEPFAR). In Kisumu, Kenya, a census and a survey were carried out in 1997–1998 among self-acknowledged female sex workers. It was estimated that 1374 sex workers were operating in Kisumu town. The HIV prevalence among a representative sample of these women was 75%. In 2006 ITM, in collaboration with Family Health Options Kenya (FHOK), developed and started implementing an intervention targeting female sex workers in Kisumu. The intervention was modelled after the intervention in Côte d'Ivoire and is funded by PEPFAR. Reliable estimates of the size of the female sex worker populations were not available in Côte d'Ivoire and outdated in Kisumu. Therefore, in 2008, we decided to conduct a series of studies using the capture–recapture method to estimate the female sex worker population in three cities in Côte d'Ivoire, Yamoussoukro, Bouaké and San Pedro; and in Kisumu, Kenya. The cities in Côte d'Ivoire were selected in collaboration with the Ministry of AIDS and the Ministry of Health. Yamoussoukro is the political capital of Côte d'Ivoire, it has a population of approximately 296 000 (census data 2006) and has been receiving an important number of internally displaced persons since the socio-political crisis of 2002. Bouaké is the second largest town of the country with an estimated population of approximately 748 000. San Pedro is an important harbour in the South of the country and has approximately 662 000 inhabitants. Kisumu is the capital of Nyanza Province, the province in Kenya most heavily affected by the HIV epidemic with a prevalence in the general population of 14.9% (National AIDS/STI Control Programme (NASCOP), Kenya, 2009). It is the third largest town in Kenya with an estimated population of 450 000 (census data 1999). Female sex workers of all ages were included in the study. Female sex workers were defined as women providing sex in exchange for money or goods. The same procedures were used in all four cities. First, a comprehensive geographical map was made of the hotspots, i.e. locations where female sex workers receive or solicit their clients, including bars and night clubs, hotels, certain streets and lodgings (for the home-based sex workers). Fieldwork for the mapping was carried out by teams of trained research assistants and peer health educators. At each hotspot, the exact location was geo-mapped using a Global Positioning System tool (Garmin GPS 60 and GPSMap 60 Cx in Côte d'Ivoire; GPS Navio in Kisumu). A site questionnaire with questions on number of female sex workers usually present at peak hours, and the busiest time of the day and the week was completed with assistance of a key informant (bar owner, peer leader, etc.). While in Cote d'Ivoire these data were entered on hard copy questionnaires, in Kisumu, Personal Digital/Data Assistants (PDAs) were used for data collection. Geo-referenced data and information on the sites were entered into an Excel file and recorded on a map. In Côte d'Ivoire ArcView GIS 3.2 was used, in Kenya Google Earth (basic Google-maps website: maps.google.com; http://www.gpsvisualizer.com/map input). These maps were then used to plan the capture–recapture exercise. Towns were subdivided into smaller areas with a limited number of hotspots. Study teams consisting of peer educators and research assistants were each assigned a list of hotspots in a particular area. There were 10 teams of three persons operating in each city in Côte d'Ivoire and 25 teams of two persons in Kisumu. All received 3 days' training on study procedures and ethics and spent 1 day prospecting their hotspots and confirming peak times. In all four cities, capture was done on a Saturday. On that day, teams visited their assigned hotspots at peak hours. The female sex worker peer educators identified and approached each female sex worker individually, and obtained verbal informed consent for study participation. Consenting sex workers were asked two or three simple questions by a research assistant and were given a capture card (see Figure 1 for study procedures). If the sex worker had already received a study card from a study team that same day, at the same or another place, she was not given another card, to avoid double counting. If the sex worker admitted she had already received a card from a friend or other source, she was considered a new capture. The study card consisted of two parts. One part was used to record limited information on each participant, the removable part, a picture, was given to the participant by way of 'tag'. The numbers of women who refused, who were unable to consent, and who were seen more than once during the same round, were recorded on a separate form. At the end of the interview, female sex workers were encouraged to visit the dedicated health services. All women approached, regardless of whether they consented to participate or not, were given condoms. Study procedures. (a) First round (capture) *Cote d'Ivoire: Do you know the sex worker clinic in town? Have you ever visited it? Kisumu: What is your age? Do you know the sex worker clinic in town? Have you ever visited it? If yes, have you visited it during the last 12 months? (b) Second round (re-capture). *Cote d'Ivoire: Do you know the sex worker clinic in town? Have you ever visited it? Did you receive a capture-card last week? Kisumu: What is your age? Do you know the sex worker clinic in town? Have you ever visited it? If yes, have you visited it during the last 12 months? Did you receive a capture-card last week ? **Côte d'Ivoire: in case she did not received a capture card, an identical card as the capture card was given. The re-capture took place 6 days later, on a Friday. The same teams visited each hotspot of their assigned area at approximately the same time as the previous week and followed the same procedures (Figure 1). In addition, all participating women were asked if they had received a card the previous week (and were thus captured during the first round). Or, N = C1*C2/R. Some sex workers were not willing to participate in the study or were unable to provide consent. This proportion of women Prefuse was estimated as the total number of refusals (including those unable to provide consent because of language barriers or being too drunk) divided by the total number of instances a sex worker was approached (including those who were approached more than once on the same day etc.) over the two visits. The total number of female sex workers was then estimated as Ntotal = N/(1−Prefuse). Confidence intervals for the estimates of N and Ntotal were obtained using the non-parametric bootstrap method. This is a computer-intensive, resampling-based approach that allows the estimation of confidence intervals without the need for distributional assumptions (Buckland & Garthwaite 1991). In the non-parametric bootstrap, the observed data are resampled to create a series of 'pseudo-experiments'. For each of these 'pseudo-experiments', N was estimated and the resulting distribution of estimates was used to construct the 95% confidence interval for N and Ntotal. Population size estimates were also obtained during the mapping exercise through the site questionnaire. The number of sex workers usually present at peak hours was estimated by key informants and recorded in that questionnaire. The total number for all sites was calculated and used as alternative population size estimates. Both studies were approved by the Ethical Committee of the University of Antwerp, Belgium. Ethical clearance for the study in Côte d'Ivoire was also obtained from the National Ethical Committee of Life Sciences and Health, Abidjan and the Protection of Human Subjects Committee of Family Health International. The protocol was scientifically reviewed by the Associate Director of Science Office of the National Center for HIV, Hepatitis, STI and tuberculosis prevention at the US Centers for Disease Control and Prevention, Atlanta (CDC). The protocol for Kisumu was approved by the Institutional Review Boards of the Kenya Medical Research Institute and CDC. The three capture–recapture surveys in Côte d'Ivoire took place between February and November 2008. In Kisumu the capture–recapture exercise was done in September 2008. In all sites, mapping was finalized 2–4 weeks prior to the capture. During the capture–recapture, most sex workers were met in bars, clubs and dance venues. Of all sex worker encounters in Yamoussoukro, 66% were in bars, clubs and dance venues; for Bouaké and San Pedro, these figures were 62% and 51%, respectively. In Kisumu the proportion was the highest (87%). Table 1 presents the estimates of the size of the female sex worker populations in the four cities based on the capture–recapture exercise. A substantial number of sex workers were encountered twice during the same round, but on a different site. The total numbers of sex workers captured during round 1 and during round 2 was similar in the four cities. The proportion of sex workers recaptured (R/C2) varied from 38% in Yamoussoukro to 57% in Bouaké. The number of refusals, both in round 1 and round 2, was significantly higher in Kisumu than in the three cities in Côte d'Ivoire. A comparison between the population size estimates resulting from the short questionnaire applied to key-informants during mapping and estimates resulting from the capture–recapture is presented in Table 2. Both methods yielded different results in each city, and no pattern was seen. Table 3 shows the coverage of the interventions. The proportion of female sex workers in Côte d'Ivoire who knew about the dedicated services ranged from 36% in Yamoussoukro to 53% in San Pedro and 15%–30% had actually visited the clinic in town. These figures tended to be higher in cities where services had been established for a longer time. In Kisumu, 58% of sex workers had already heard of the services and 34% had actually visited the clinic. In the past, most studies used mapping and census to estimate the number of sex workers (Vandepitte et al. 2006). Census is a straightforward method and less time-consuming than capture–recapture, because it can be done during the mapping. However, census methods are not well suited to hidden and very mobile populations. Double counting because of high mobility may result in an overestimation, and a large invisible population will result in an underestimation of the population size. In our study, size estimates reported during mapping were lower in some cities, and higher in others. In the absence of a 'gold standard', it is difficult to draw a conclusion on the most reliable method. Capture–recapture was previously described as a relative simple method to assess the size of high-risk populations (Kruse et al. 2003). However, in our experience, capture–recapture was harder to conduct well than its initially simple-looking method would have suggested. Capture–recapture has both operational and methodological challenges. Although the assistance of sex worker peer educators was crucial for the successful conduct of the capture–recapture, a team of experienced researchers was needed to train research assistants, coordinate, standardize and supervise field work. There are four conditions that need to be fulfilled to give reliable estimates with the capture–recapture method (Family Health International, 2003). We designed the methods and procedures of the study to adhere to these conditions as much as possible. First, recapture was done 6 days after the capture visit. Given this short time-frame, the number of sex workers moving in or out of town between the two rounds is likely to be very small, meeting the condition of a closed population. In addition, capture and recapture dates were chosen away from events which may involve a large influx or outflux of sex workers during the exercise. The second condition is that all members of the female sex workers population have the same probability of being captured. Comprehensive mapping was done and capture was realized during peak hours on week-ends, in all mapped sites. However, some sex workers have inherent lower probabilities to be caught than others. Call girls, for instance, who work exclusively from home, were not approached in this study and their number could not be estimated. The third condition is the independency of samples (Chao et al. 2001). To limit dependency, we chose different days of the week to conduct the two rounds, as sex workers may have their habits, e.g. on Saturdays a sex worker may always go to a certain bar. If she is captured on a Saturday in that bar, she will have a higher probability to be re-captured on the next Saturday in the same bar. However, we could not eliminate all dependency; women captured during the first round for instance may be more likely to be captured again if they recognize the field assistants. A positive dependency among samples, the most likely situation, will lead to an underestimation of the number of sex workers. The fourth condition is the accuracy of the re-capture history. The picture cards used in our study were neutral, not valuable enough to solicit misreporting, but attractive enough not to be forgotten. The correct identification of sex workers at the hotspots is a key factor to the success of the capture–recapture. The involvement of sex worker peer educators as field workers was of utmost value, as was also reported in Madagascar (Kruse et al. 2003). Peer educators not only are in the best position to identify other sex workers, but they also inspire confidence to potential participants. However, it cannot be excluded that a certain number of sex workers were missed because of non-identification. The number of refusals was much higher in Kisumu than in Côte d'Ivoire. There may be different explanations for this, including misclassification, bad timing and lack of trust in the research. In Côte d'Ivoire, peer educators confirmed that participants were active sex workers before asking consent. This was not done in Kisumu, so we cannot exclude that some women approached by the peer educators in Kisumu were not involved in sex work and refused participation for that reason. In sites where the team operated late at night, sex workers were already busy negotiating with their clients and were more likely to refuse participation. Finally, peer educators in Kisumu thought the refusal rate had to do with mistrust, not knowing what would happen to the data collected, rather than misclassification. To deal with the refusals, we assumed consistent refusal among sex workers. The correction used in the analysis (for Ntotal) assumes that women who refuse to participate do so consistently at all encounters at both rounds, thus constituting a subpopulation of women who can never be captured. The size of this population is estimated by the number of refusals divided by the number of contacts (totalled over both visits). However, if women were to refuse participation at random, e.g. when they are in conversation with a potential client, the unadjusted (N) estimate would have been more correct. A third possibility is that the population who refuse participation changes between capture and recapture visits. This would occur when some women who refuse participation at the capture visit consent to participate at the recapture visit (e.g. because they are encouraged by peers). In this situation, the total population estimate may be biased. Other challenges during the implementation of this study were police raids which happened during re-capture in Bouaké and San Pedro, despite the prior visits of the investigator's team to the police in all cities to inform them about the study. In Kisumu, heavy rainfall hampered field work during the second round. One of the major advantages of the capture–recapture is the direct contact with sex workers, allowing a few simple questions. In our study, we were able to assess the coverage of the intervention, which was rather low in all cities. This may partly be explained by the inclusion of indirect, occasional, sex workers in the capture–recapture survey. Interventions tend to target predominantly direct, self-acknowledged sex workers, because they are easier to reach than indirect sex workers. Efforts should be made to increase coverage by scaling up prevention outreach to all identified hotspots, by intensifying one-to-one and group session contacts with peer educators and by improving the quality and attractiveness of services. Geographical mapping of hotspots was essential to plan the capture–recapture exercise but also resulted in an updated practical tool for improving prevention outreach by local implementing NGOs. In this perspective, it is important that researchers and programme managers jointly undertake a capture–recapture exercise not only for estimation purposes, but also for planning and redirecting field activities. In conclusion, capture–recapture was successfully applied to estimate the population size of female sex workers both in Côte d'Ivoire and in Kisumu, Kenya. These estimations were urgently needed to help mobilize an increased response to HIV, to assess programme coverage and to estimate potential impact of the targeted intervention. However, in the absence of a 'gold standard', estimates can not be validated and may be under- or overestimated. More work needs to be done on comparing different methods to estimate the size of sex worker populations. The study was funded by the Directorate-General for Development Cooperation (DGDC), Belgium and the President's Emergency Plan for AIDS for AIDS Relief (PEPFAR), United States of America.

Referência(s)