Lessons learned in the conduct, validation, and interpretation of national population based HIV surveys

Transcription

1 Lessons learned in the conduct, validation, and interpretation of national population based HIV surveys Jesús M. García Calleja a, Lawrence H. Marum b,césar P. Cárcamo c, Lovemore Kaetano d, James Muttunga e and Ann Way f In the past few years several countries have conducted national population-based HIV surveys. Survey methods, levels of participation bias from absence or refusal and lessons learned conducting such surveys are compared in four national population surveys: Mali, Kenya, Peru and Zambia. In Mali, Zambia, and Kenya, HIV testing of adult women and men was included in the national-level demographic and health surveys carried out regularly in these countries, whereas in Peru the national HIV survey targeted young people in 24 cities with populations over The household response rate was above 90% in all countries, but some individuals were absent for interviews. HIV testing rates were between 70 and 79% of those eligible, with higher test rates for women. Three critical questions in this type of survey need to be answered: who did the surveys miss; how much it matters that they were missed; and what can be done to increase the participation of respondents so the coverage rates are adequate. The level of representativeness of the populations tested was adequate in each survey to provide a reliable national estimate of HIV prevalence that complements other methods of HIV surveillance. Different lessons were learned from each survey. These population-based HIV seroprevalence surveys demonstrate that reliable and useful results can be obtained, although they require careful planning and increased financial and human resource investment to maximize responses at the household and individual level, which are key elements to validate survey results. This review was initiated through an international meeting on New strategies for HIV/ AIDS Surveillance in Resource-constrained Countries held in Addis Ababa on January 2004 to share and develop recommendations to guide future surveys. ß 2005 Lippincott Williams & Wilkins AIDS 2005, 19 (suppl 2):S9 S17 Keywords: HIV/AIDS, Kenya, Peru, population-based surveys, surveillance, Zambia Introduction Since the development of an antibody test for HIV in 1984, most countries have put in place surveillance systems to monitor HIV prevalence levels and trends based in antenatal clinics (ANC). The quality of these HIV surveillance systems has varied over the years [1], making it difficult to obtain consistent and accurate estimates of HIV prevalence. Although HIV surveillance systems were set up primarily to monitor trends among population groups, the data collected by these systems have been used to estimate the burden of HIV in those countries [2]. Improved estimates have recently been made possible through the use of national population surveys to validate and calibrate these surveillance estimates, supported with the increased resources available within the new international initiatives to address HIV/AIDS [3]. In Africa, several community HIV prevalence studies have shown that HIV prevalence in the general adult population years old is similar to that of women attending ANC in the same communities [4]. Some From the a World Health Organisation, Geneva, Switzerland, the b Centers for Disease Control and Prevention (CDC), Nairobi, Kenya, the c Universidad Peruana Cayetano Heredia, Peru, the d Tropical Diseases Research Center, Ndola, Zambia, the e Kenya Medical Research Institute (KEMRI), and the f Measure DHS, ORC Macro, USA. Correspondence to Jesús M. García Calleja, World Health Organization, 20 Avenue Appia, 1211 Geneva, Switzerland. callejaj@who.int ISSN Q 2005 Lippincott Williams & Wilkins S9

2 S10 AIDS 2005, Vol 19 (suppl 2) Table 1. Completed, ongoing and planned national HIV seroprevalence surveys. Survey/year HIV specimen Method Mali DHS 2001 DBS Unlinked anonymous Zimbabwe Youth Venous Linked anonymous Peru Urban Youth 2002 Venous or urine Linked anonymous Burundi AIS 2002 DBS Linked anonymous Dominican Republic DHS 2002 Oral fluids Unlinked anonymous Niger AIS 2002 DBS Linked anonymous Sierra Leone AIS 2002 DBS Unlinked anonymous South Africa HIV/AIDS 2002 Oral fluids Linked anonymous Zambia DHS 2002 Venous Unlinked anonymous Burkina Faso DHS DBS Linked anonymous Congo Urban Adults 2003 Venous Linked anonymous Ghana DHS 2003 DBS Linked anonymous Kenya DHS 2003 DBS Linked anonymous Tanzania AIS DBS Linked anonymous Uganda AIS 2004 Venous (DBS children) Linked anonymous Cameroon DHS DBS Linked anonymous Lesotho DHS 2005 DBS Linked anonymous Malawi DHS 2005 DBS Linked anonymous Rwanda DHS 2005 DBS Linked anonymous Senegal DHS 2005 DBS Linked anonymous AIS, AIDS Indicator Survey; DBS, dried blood spot; DHS, demographic and health surveys. scientists have, however, highlighted the limitations of using sentinel surveillance in ANC for deriving national estimates. First, the populations tested do not represent the entire population at risk (e.g. ANC data excludes men and non-pregnant women). Second, pregnancy rates are affected both by HIV infection and the level of use of contraception (as well as sexual activity; age at first sex will determine teenage pregnancy rates). Third, pregnant women may not attend ANC and rates of attendance may change over time. Fourth, the sites selected for surveillance may not be representative because they were often chosen initially in urban or semi-urban locations and by convenience, whereas in Africa the majority of the population is rural, and the misclassification of women as rural or urban residents is frequent [5 8]. Some of these limitations in deriving national HIV estimates from ANC surveillance data can be overcome by adjusting some factors [9], but recent papers comparing ANC surveillance results and populationbased surveys have questioned the limitations of extrapolating HIV national estimates from HIV sentinel surveillance [10 13]. National AIDS programmes have thus been under increased pressure by decision-makers to implement nationally representative HIV surveys as they are considered to provide better HIV estimates for the general population. Rwanda and Uganda conducted national HIV population-based surveys in the mid-1980s. However, the researchers involved in these pioneering studies themselves questioned some of the aspects of the study design, particularly the testing of only one individual per household, and instead recommended that future studies test all eligible members of sampled households [14,15]. With the availability of more resources and new laboratory methods for HIV testing (using dried blood spots or saliva), 15 countries have conducted, and at least 10 are either planning or are in the midst of implementing, national HIV population-based surveys (Table 1). In some cases there have been special surveys to estimate HIV prevalence, whereas in other cases HIV testing has been added to demographic and health surveys (DHS). The results of these surveys allow a better understanding of HIV epidemiology by providing greater information on the behavioral risk factors and the demographic and geographical patterns of HIV infection within a country, and in this way they serve to calibrate ANC surveillance results [16]. They are, however, more costly than sentinel surveillance and are therefore impractical for tracking intermediate term trends as they are conducted approximately every 5 years. This paper reviews the methods and conduct of HIV population-based surveys carried out in some countries in Africa and Latin America in order to describe issues that influence participation and the representativeness of this type of survey. It was initiated through an international meeting on New strategies for HIV/AIDS Surveillance in Resource-constrained Countries held in Addis Ababa on January 2004 [17]. The detailed results of these surveys are reported elsewhere [18 20]. It is intended that the lessons learned in undertaking these surveys may help other countries planning national HIV population surveys in order to improve the quality of the resulting information. Sampling methods and participation bias The interpretation of results and the accuracy of prevalence estimates depend in large measure on the participation biases within the surveys. Three questions are important in assessing these biases: who did the survey miss?; how much

3 Validity of national population-based HIV surveys García Calleja et al. S11 does it matter that they were missed?; what can we do to address coverage issues? In considering the first question, who was missed, it is important to recognize that the sampling approach in national population-based seroprevalence surveys generally involves at least two stages: (i) selecting a representative sample of geographical units (usually census enumerated areas) from a frame of all such units in the country; and (ii) selecting a representative sample of households from among all the households living in the geographical units selected during the first stage. All individuals living within the households selected at the second stage who meet a specific criterion (e.g. who are within a specific age range) are eligible for inclusion in the survey interview and testing. Because such surveys are typically household based, specific groups are automatically excluded from the sample frame, for example, individuals in prison, in the military, away at school, in refugee camps or the homeless. Other reasons why the household-based population may not be fully represented in the survey include: an out-of-date sampling frame; errors with the household listing or selection; and problems in selection or in identifying eligible respondents within the household. Geographical areas may also be deliberately excluded from the sample, particularly if they are remote or otherwise difficult to access during the survey. In addition to these issues that affect the selection of the survey sample, there are many reasons why, once the sample is chosen, an eligible individual may not be interviewed or tested. From the perspective of the household, the interviewer may not be able to find the dwelling where a household was listed, the dwelling may have been vacated or destroyed between the time of the listing and the survey, there may be no competent individual at home or the entire household may be absent when the interviewer visits. In turn, household members who are eligible may be absent or otherwise unable to take part. Individuals who are asked to participate may also refuse, either to be interviewed or to be tested or both. These issues of absence and refusal to be tested will be addressed below in the country-specific experience. Moreover, in households of single individuals it may be more difficult to find the occupants because they are absent for work or other reasons. Often these people are at greater risk of HIV. Analysis of the participation bias is necessary to answer the second question: how much does it matter that they were missed? For those who are absent for an individual interview, some limited information may be available about the household and their sociodemographic characteristics. For those who consent to an interview but refuse to provide a specimen for testing, surveys that link the behavioral information with the biological results are able to assess the individual behavior associated with the risk of HIV and compare it with those who have accepted the test. We reviewed the methods, the levels of participation and the lessons learned in the implementation and analysis of surveys in Mali, Zambia, Kenya and Peru. In the three African countries, the HIV testing was carried out in DHS. In Zambia, the HIV results were only linked to basic demographic characteristics, whereas in Kenya, the HIV results, although anonymous, were collected and linked to the full range of respondent data. In Peru, the survey was conducted only in mid-size cities as part of an intervention to prevent sexually transmitted diseases (STD) in young people. Biomarkers for HIV and other STD were linked to personal characteristics and to all data available. National and international ethical committees reviewed and approved the HIV testing protocols for the surveys carried out in the four countries. Mali demographic and health survey 2001: informed consent and participation The Mali DHS 2001 was the first to include HIV testing as part of a DHS. As knowledge of HIV was low in Mali, at the start of the survey there was concern about whether informed consent was possible and whether it was feasible to include a biomarker without interfering with the survey process [19]. In this survey, haemoglobin was checked on the spot and a dried blood spot was collected on filter paper for later HIV testing. The level of knowledge of HIV was low among the population, and there was concern that participants would have difficulty providing informed consent for testing of two diseases (anaemia and HIV) that were poorly known or understood. In the survey, however, 85.2% of eligible women and 75.6% of eligible men were found, gave consent, and were tested for HIV using dried blood spots. A voucher was given to participants to obtain free voluntary counseling and testing (VCT) at a nearby centre. The concurrent Mali Informed Consent Study evaluated the process of obtaining consent and the impact on participation through observation of the informed consent procedures. A total of 192 participants were interviewed in 10 selected clusters [20]. Over 90% understood that they were being asked for a blood test for anaemia and AIDS, and there was direct evidence that over two-thirds knew participation was voluntary. Whereas 30% said they knew nothing about AIDS, two-thirds had heard about ways of transmission. Men were more likely to refuse the test and to ask questions after the informed consent statement. There were weaknesses in the process of presenting the informed consent statement, partly because of the contrast between the language of the formal statement and local linguistic norms. Most of the field editor/health workers simplified and adapted the language in presenting the informed consent statement: 69% explained without reading; 13% read with explanations; 6% only read the statement; and another 13% did not present the statement.

4 S12 AIDS 2005, Vol 19 (suppl 2) Lessons on the consent process Although there were inconsistencies in the way information was presented in order to obtain informed consent before collecting the specimens, the first lesson from this survey is that people are willing to give consent for testing, even in countries with a low prevalence of HIVand where there is not much knowledge about HIV infection. Second, the informed consent statement needs to be clear, concise and precise. Moreover, it needs to use appropriate language to the local culture. Understanding the informed consent would probably increase participation rates. Zambia: a survey in a high prevalence setting The Zambia DHS was conducted in The survey involved a nationally representative sample of 8050 households from 320 clusters (100 urban and 220 rural). In a subsample of one-third of the households, all female respondents aged years and male respondents aged years were asked to consent to participate in the syphilis and HIV testing components of the survey [21]. Consent for testing was obtained separately for syphilis testing and for HIV testing. Venipuncture was performed to obtain blood for syphilis testing, and for those who also consented to HIV testing dried blood spot samples were prepared. The survey response rates were high, with values over 95% for the household and female interviews and nearly 90% for the male interviews. Coverage rates for blood testing were significantly lower, especially for men: 79% of 2689 eligible women consented to the HIV testing compared with 73% among the 2418 eligible men (Table 2). Coverage rates for syphilis were nearly identical to those of HIV. Women were slightly more likely to refuse HIV testing (16 versus 15% for men), but were significantly less likely to be absent at the time of consent for blood testing (3 versus 8% for men). Regional differences were significant for refusal, ranging from 9% in Northwestern Province to 28% in Eastern Province. Absence rates had both a regional and sex participation bias, ranging from 0.5% for both women and men in Luapula Province, to significant differences in Copperbelt (4% for women and 15% for men), Western (3% for women and 13% for men) and Lusaka (3% for women and 11% for men). Overall, HIV prevalence in Zambia was 15.6%, with 17.8% of women and 12.9% of men infected for a female to male ratio of 1.5. The urban rural prevalence ratio was almost 2 to 1 for both men and women, and there were significant regional variations [21]. Lessons from the Zambia demographic and health survey An analysis of participation bias is critical for the interpretation of results. The Zambia DHS utilized special sample weights for analysis of the HIV and syphilis test results, taking into account differences in coverage of testing by province and sex. Rates of testing were also analysed by background characteristics, with minimal differences by age group, marital status, number of sexual partners and education. There was, however, a significant difference based on HIV test status: 74% of those ever tested for HIV were tested in the survey, whereas the survey test rate among those not previously tested was 88% of those who wanted to know their status and 60% of those who did not. This suggests that some refusals are caused by either not needing a test as a result of previous testing, or stigma and not wanting a test. As only sex, age and residence were linked to the HIV test results, other adjustments or an analysis of behavioral and other demographic characteristics were not possible. A second lesson from the Zambia DHS was in the comparison of general adult HIV prevalence with that of sentinel surveillance through two approaches. First, 29 clusters located within catchment areas of 16 antenatal sentinel surveillance had an HIV prevalence of 19% compared with 20% in the ANC. Second, urban rural prevalence differences were useful in adjusting for the urban geographical bias of the sentinel surveillance system. Peru national urban youth survey: high participation and linking sexual contacts As part of a baseline assessment for a community randomized intervention trial, data on STD prevalence and risk behaviors and the structure of sexual networks Table 2. Survey participation rates in Mali, Peru, Kenya, and Zambia. Mali Peru Kenya Zambia Households sampled for HIV testing Household interviewed (%) 97.6% 90.5% 86.8% 88.5% Household response rate (%) 97.6% NA 96.3% 98.2% Response rate among eligible women (%) 94.9% 92.0% 94.0% 96.4% Response rate among eligible men (%) 83.8% 90.5% 85.5% 88.7% NA, not available.

5 Validity of national population-based HIV surveys García Calleja et al. S13 were collected through a general population survey. Twenty-four (80%) of the 30 Peruvian cities with populations greater than inhabitants were selected for the survey. The target in each city was a random twostage cluster household sample of 250 men and 250 women and a consecutive sample of 50 male sexual partners and 50 female sexual partners. A census in the selected clusters was carried out to identify households with eligible members, selecting a random sample of households with at least one eligible member, and within the selected households, selecting the eligible member with the most recent birthday. Inclusion criteria for participants in the random sample included men and women from the ages of 18 to 29 years old, resident of the selected city for at least 6 months before the survey. A consecutive sample of adult (aged 18 years or older) sexual partners living in the same household as the participants in the random sample was included in the survey. Consenting participants completed a face-to-face demographic questionnaire and a self-administered sexual behavior questionnaire [17]. Participants were asked to provide venous blood or, if unwilling, oral fluid. Men were also asked to provide urine, and women were asked to provide self-administered vaginal swabs or, if unwilling, urine. Of individuals selected for participation, (90.5%) agreed to participate in the testing. In addition, 2347 of 2663 sex partners selected (88.1%) agreed to participate (Table 2). Almost 77% of participants were tested, self-administered questionnaires were returned by all but 176 of the participants (1.2%) (Table 3). Eightyfour per cent provided blood or oral fluid, 87% of men provided urine, and 84% of women provided a vaginal swab or urine [17]. The baseline survey documented a number of risky sexual behaviors among sexually active Peruvians, including a high frequency of sex with female sex workers (FSW), a low frequency of condom use with casual partners and FSW, and a high percentage of men having unprotected sex with men. Overall, HIV is more frequent in men (0.4%) than in women (0.1%). The most important risk Table 3. Participation rates (completed questionnaire and HIV testing) among the population eligible for testing by sex, Peru, Kenya, and Zambia. Peru Kenya Zambia Mali Women tested (%) Refused (%) Absent (%) Missing results NA Number eligible Men tested (%) Refused (%) Absent (%) Missing results NA Number eligible NA, not available. factor for HIV infection for men was sex with other men, and for women, sex in exchange for money [17]. Lessons learned from the Peru urban youth survey Several factors contributed to the high participation rates in this national urban youth survey. First, the selfadministered sexual questionnaires contributed to confidence in the confidentiality of results. Second, there was good collaboration and coordination of many institutions, including the participation of local resident health professionals as interviewers and sample collectors. Third, an oral fluid method of HIV sample collection was offered should there be refusal of a blood test for HIV, and urine tests were used as an alternative for STD testing. Kenya demographic and health survey 2003: learning in the field and anonymous linked results The 2003 DHS in Kenya was the first to include a comprehensive HIV/AIDS module and HIV testing [18]. The survey sampled 9865 households in 400 clusters (129 urban and 271 rural). In half of the households, all men aged years and all women aged years were interviewed and asked voluntarily to provide capillary blood from a finger stick, collected on filter paper and dried in the field. VCT was organized through mobile teams serving all clusters outside of Nairobi (there are 29 VCT sites). Survey respondents and other members of the community who wanted to be tested were referred either to existing VCT centres or to mobile centres. The Kenya DHS household response rate was 96%. Although the survey interview response rate was also comparatively high, above 90% for women and 85% for men (Table 2), the proportion of eligible respondents tested was 76% of women and 70% of men (Table 3). There was also a large difference in the testing coverage between respondents in urban and rural areas, with lower testing rates among urban dwellers, especially men, less than 60% of urban men tested (Table 4). Men were more likely to be absent at the time of the survey; 15% of men compared with 9% of women were neither interviewed nor tested, whereas 5% of eligible men and women were present for the interview but absent for the testing. Refusal rates were 14% for women and 13% for men, with greater numbers of refusals in urban areas [18]. The overall HIV prevalence was 6.7%, with significantly greater prevalence among women (8.8%) compared with men (4.5%). HIV prevalence was also significantly greater in urban areas [18].

6 S14 AIDS 2005, Vol 19 (suppl 2) Table 4. Participation in HIV testing by sex and residence among those eligible. Residence In the first month of the survey, the rate of testing was low (64%) [22]. Response rates improved in the second month, after initiating national media advertising, additional community mobilization, and closer monitoring of test acceptance by teams in the field. Refusal rates declined from over 20% in the first month of the survey to approximately 10% by the third month (Fig. 1). In addition, absence rates for both men and women also dropped by half in the same time period. Strengthening VCT outreach during the survey implementation may also have a positive effect. The VCT results indicated that substantial numbers of the population were eager to know their HIV status, as a total of individuals, including both participants and nonparticipating community members, came and requested HIV testing from mobile VCT sites [23]. Out of those individuals, 6617 were men, and 3472 were women. The prevalence in the mobile VCT sites was similar among men, 5% compared with 4.9% in male survey participants, but the women coming to VCT had a greater prevalence (13% compared with 8.8% in survey participants), suggesting a participation bias in seeking VCT for women but not for men. Percent Women refused Mali Zambia Kenya % Tested % Tested % Tested Women Urban (%) Rural (%) Men Urban (%) Rural (%) Men refused Women missing Men missing Fig. 1. Rates of refusal of testing and absence during the course of the Kenya demographic and health survey. Refusal rates declined from less than 20% to approximately 10% after beginning community mobilization and better publicity. Month 1; month 2; month 3. In the Kenya DHS 2003, HIV test results were linked to the behavioral and demographic information without any identifiers for household, cluster, or district to protect the anonymity of the participants. This allowed the development of regression models to predict HIV status for respondents who were sexually active, utilizing 35 questionnaire variables to compare the 3272 women and 2917 men who were interviewed and tested with the 769 women and 661 men whowere interviewed but not tested. The results indicated that those who were interviewed but not tested (refused or not found) had a lower HIV risk and may have had a lower HIV prevalence than those who consented and were tested. Adjustment based on this regression analysis reduced the prevalence estimate for women from 8.8 to 8.5% and for men from 4.9 to 4.8%. Even if those who were absent and were neither interviewed nor tested had a greater prevalence than those interviewed or tested, the predicted difference from the results of those tested was negligible [24]. Lessons from the Kenya demographic and health survey Participation significantly improved in the Kenya DHS after additional efforts were made to inform the communities about the nature of the survey, including through the media and local community leaders. In some clusters, cultural fears that blood might be used for witchcraft led to high refusal rates until there was community discussion and assent by community leaders. Second, large numbers of men and higher-risk women availed themselves of mobile VCT services that were provided in all 350 clusters outside of Nairobi. Even in communities where there was a VCT centre, some participants expressed a preference for the anonymity of outsiders providing the service. Third, it is possible to link HIV results and behavioral information without compromising identity, and this allows multivariate analysis of participation bias in the survey. Discussion All of the surveys described were able to provide significantly improved estimates of the burden of HIV in their respective countries. The prevalence of HIV varied in the general population in the three DHS: 16% in Zambia, 7% in Kenya; and less than 1% in Mali. Prevalence was also below 1% in urban youth in Peru. For Kenya and Zambia, there were significant differences between women and men and urban compared with rural areas. Participation bias remains the greatest challenge to the validity of these surveys, with significant differences in

7 Validity of national population-based HIV surveys García Calleja et al. S15 absence and refusal between women and men, between urban and rural areas, and between regions within the respective countries. An analysis of these participation biases is necessary to validate and interpret the results of the surveys. As shown in Table 2, the household response rate ranged from 96 to 98% in the DHS, whereas the proportion of eligible participants interviewed ranged from 86 to 98%. Among those eligible for HIV testing, however, the rate of testing ranged from 73% in Kenya to almost 77% in Peru. This participation bias was approximately equal because of absence or refusal at the time of blood testing in Kenya and Zambia. In Peru, almost 10% did not complete the questionnaire and the rest refused to provide specimens. Especially noteworthy were the higher participation rates among women compared with men, much of which is because of the absence of men from the household. Participation rates were higher among men in Peru, perhaps because they were given a choice of sampling (urine, saliva or blood). Clearly, in all four surveys, substantial proportions of the eligible populations did not participate in the testing component of the survey. The second question, do the missing persons matter? is relevant in assessing the representativeness of results of the testing from these surveys. At both the household level and the individual level in all four surveys, refusal and absence were the main reasons for the non-response. Unfortunately, there is very limited information on the relationship between absence and refusal during a seroprevalence survey and an individual s serostatus. Missing persons matter if their HIV level is significantly higher or lower than that found among the group that was tested. A study conducted in Zambia found an effect of mobility among younger men but not among women [25]. Other studies have shown that highly mobile individuals typically have higher HIV levels than the less mobile. In a review of absences in a four-city study in Cameroon, it was found that women who had traveled more than a month have lower HIV prevalence that those who were present (8.6 versus 9.8%) [26]. On the other hand, men who have been absent for more than a month have significantly greater HIV prevalence than those who had not been away for a long period (7.6 versus 1.4%) [27]. In another analysis conducted in Kisesa (Tanzania), a cohort study found that mobility was associated with greater HIV prevalence rates, with a relative risk of 1.5 for men and 1.2 for women (B. Zaba, personal communication). However, it is difficult to conclude and to extrapolate these conclusions to all population surveys. First, the two studies document that mobility is directly related to HIV status but they do not document: (i) that the population absent at the time of the survey is significantly more mobile than the group that was tested; or (ii) that the mobile absent population differs significantly from the mobile population that was found and tested in the survey. Unlike the two studies that are geographically limited in scope, in a cross-sectional national survey, individuals away from one locality at the time of a crosssectional national survey have the opportunity to be represented in the locality to which they travelled. This acts to reduce considerably the likelihood that the missing population is significantly biased with respect to mobility. But at the same time, the population missing for the testing may still include a somewhat greater proportion of highly mobile individuals than the population tested, thus a greater HIV prevalence. Another important point is that size matters in determining whether biases among the missing group will have a significant impact on a national prevalence rate. There can be subgroups within the missing population that have higher than average HIV rates, but if those groups together do not constitute a significant proportion of the overall population, this may have little impact on the national rate. In addition, there are certainly some groups that have lower than average rates, and it is the balance between these two groups that determines the potential impact of the missing group on the overall rate. This balance will vary depending on the culture, factors related to non-response that may have no relationship to HIV status. That brings us to the third question, how can participation of survey respondents be increased? One factor that would seem to be intuitively important is the testing modality. An analysis of non-response patterns in these three surveys, however, does not provide clear-cut evidence that the testing modality (capillary blood, whole blood, or oral fluids) influences response levels, and the relationship is not straightforward. In all three surveys, the percentage of eligible respondents who agreed to HIV testing is above 70%, and whereas Peru offered alternatives for HIV testing (blood or oral fluid), Zambia reported higher response rates. The results of other surveys would suggest, however, that less intrusive testing modalities are linked to lower non-response. For example, in a previous population-based survey in Zambia, using saliva, refusal rates were below 10% [28]. Refusal rates were also lower for a DHS survey in the Dominican Republic that used an oral device for HIV testing, with an overall response rate of 90% [29]. However, in another recent national population-based survey conducted in South Africa, the response rate for HIV testing was below 70% although they used oral fluid [30]. Therefore, for general population surveys, factors in the design and implementation of the survey to increase the response rate other than the type of sampling are important. That may differ from surveys among hard to reach populations where specimen collection may be more difficult [31].

8 S16 AIDS 2005, Vol 19 (suppl 2) From the lessons learned in the surveys presented we can conclude that a carefully designed publicity campaign should be incorporated into the survey planning. This should include both mass media coverage and local government officials. Providing participants with information about the survey objectives and the testing procedures and with confidentiality protection will increase participation. The selection and training of the field staff are key variables. In all four surveys cited in this paper, health personnel were recruited to collect the HIV samples. Other surveys, however, have shown that lay personnel may be acceptable depending on the testing modality. Close supervision of the performance of survey teams is also important because field staff performance may be strongly related to non-response. The workload of the staff involved in HIV testing is also of importance. For example, the Kenyan field staff had a 50% greater workload than the Zambian field staff. The results of the four surveys presented in this review show that it is feasible to undertake national representative HIV population-based surveys with response rates adequate to provide reliable and useful results. Appropriate consent processes with an adequate number of well-trained staff, publicity to inform the community, and careful multivariate analysis with linked datasets to describe better those who refuse and those who are absent will assist future surveys to minimize participation bias and inform programmes of prevalence and trends of HIV infection and risk behavior. References 1. Garcia-Calleja JM, Zaniewski E, Ghys PD, Stanecki K, Walker N. A global analysis of a decade s trends in the quality of HIV sero-surveillance. Sex Transm Infect 2004; 80 (suppl. 1): Walker N, Stanecki KA, Brown T, Stover J, Lazzari S, Garcia- Calleja J, et al. Methods and procedures for estimates HIV/ AIDS and its impact: the UNAIDS/WHO estimates for the end of AIDS 2003; 17: UNAIDS Global Report Report on the Global AIDS Epidemic. 4 th Global Report. UNAIDS The UNAIDS Reference Group on Estimates Modeling and Projections. Improved methods and assumptions for estimation of the HIV/AIDS epidemic and its impact: recommendations of the UNAIDS Reference Group on Estimates, Modeling and Projections. AIDS 2002; 16:W1 W Fylkesnes K, Ndhlovu Z, Kasumba K, Musonda RM, Sichone M. Studying dynamics of the HIV epidemic: population-based data compared with sentinel surveillance in Zambia. AIDS 1998; 12: Zaba B, Boerma JT, White R. Monitoring the AIDS epidemic using HIV prevalence data among young women attending antenatal clinics: prospects and problems. AIDS 2000; 14: Glynn JR, Buve A, Carael M, Musonda RM, Kahindo M, Macauley I, et al. Factors influencing the difference in HIV prevalence between antenatal clinic and general population in sub-saharan Africa. AIDS 2001; 15: Gregson S, Terceira N, Kakowa M, Mason RM, Anderson RM, Chandiwana SK, Carael M. Study of bias in antenatal clinic HIV-1 surveillance data in a high contraceptive prevalence population in sub-saharan Africa. AIDS 2002; 16: Ghys PD, Brown T, Grassly NC, Garnett G, Stanecki KA, Stover J, Walker N. The UNAIDS estimations and projection package: a software package to estimate and project national HIV epidemics. Sex Transm Infect 2004; 80 (suppl. 1):i5 i Assefa T, Davey G, Dukers N, Wolday D, Worku A, Messele T, et al. Overall HIV-1 prevalence in pregnant women overestimates HIV-1 in the predominantly rural population of Afar Region. Ethiop Med J 2003; 41 (suppl. 1): Fabiani M, Fylkesnes K, Nattabi B, Ayella EO, Declich S. Evaluating two adjustment methods to extrapolate HIV prevalence from pregnant women to the general female population in sub-saharan Africa. AIDS 2003; 17: Neequaye AR, Neequaye JE, Biggar RJ, Mingle JA, Drummond J, Waters D. HIV-1 and HIV-2 in Ghana, West Africa: community surveys compared to surveys of pregnant women. West Afr J Med 1997; 16: Kwesigabo G, Killewo JZ, Sandstrom A. Sentinel surveillance and cross sectional survey on HIV infection prevalence: a comparative study. East Afr Med J 1996; 73: Dunn D. Nationwide community-based serological survey of HIV-1 and other human retrovirus infections in a Central African country. WHO AIDS Tech Bull 1989; 2: Kengeya-Kayondo JFA, Amana A, Naamara W. Anti-HIV seroprevalence in adult rural populations of Uganda and its implications for preventive strategies. In: Fifth International Conference on AIDS. Montreal, Canada, 4 9 June Boermat JT, Ghys P, Walker N. HIV estimates from national population-based surveys: A new gold standard for surveillance systems? Lancet 2003; 362: CDC, UNAIDS, WHO. Proceedings of the Conference New Strategies for HIV/AIDS Surveillance in Resource-constrained Countries. Addis Ababa, Ethiopia, January 2004; in press. 18. Central Bureau of Statistics, Ministry of Health, Kenya Medical Research Institute, Centers for Disease Control, and ORC Macro. Kenya demographic and health survey 2003: preliminary report. Calverton, Maryland, USA: Central Bureau of Statistics, Ministry of Health, Kenya Medical Research Institute, Centers for Disease Control, and ORC Macro; Cellule de Planification et de Statistique (Ministère de la Santè) et Direction Nationale de la Statistique et de l Informatique. Mali Demographic and Health Survey (MDHS-III) Yoder PS, Konate MK. Obtaining informed consent for HIV testing: the DHS experience in Mali. Calverton, Maryland: ORC Macro; Central Statistical Office, Central Board of Health, and ORC Macro. Zambia demographic and health survey Calverton, Maryland, USA: Central Statistical Office, Central Board of Health, and ORC Macro; Muttunga JN, Otieno F, Marum L, Odoyo J, Mutura C, Chebet K. Uptake of dry blood spots sampling in the Kenya Demographic and Health Survey 2003: challenges of culture, stigma and attitude. In: 13th International Conference on AIDS and STDs in Africa. Nairobi, Kenya, 26 September 2003 [Oral abstract]. 23. Marum L, Ngare C, Odoyo J, Gathendu B, Muttunga J, Chebet K. Learning HIV status in population based prevalence surveys: lessons from mobile VCT in the Kenya Demographic and Health Survey In: 13th International Conference on AIDS and STDs in Africa. Nairobi, Kenya, 26 September 2003 [Oral abstract]. 24. Way A, Cross A. Evaluating the impact of no response on the KDHS HIV prevalence estimates. Presented to Ministry of Health Surveillance Stakeholders Meeting; Nairobi, Kenya November Fylkesnes K, Musonda RM, Sichone M, Ndhlovu Z, Tembo F, Monze M. Declining HIV prevalence and risk behaviors in Zambia: evidence from surveillance and population-based surveys. AIDS 2001; 15: Lydié N, Robinson NJ, Ferry B, Akam E, De Loenien M, Abega S. Mobility, sexual behavior, and HIV infection in an urban population in Cameroon. J Acquir Immune Defic Syndr 2004; 35: Buve A, Carael M, Hayes RJ, Auvert B, Ferry B, Robinson NJ, et al. Multicentre study on factors determining differences in rate of spread of HIV in sub-saharan Africa: methods and prevalence of HIV infection. AIDS 2001; 15 (suppl. 4): S5 S14.

9 Validity of national population-based HIV surveys García Calleja et al. S Fylkesnes K, Kasumba K. The first Zambian population bases HIV survey: saliva based testing is accurate and acceptable. AIDS 1998; 12: Encuesta Demográfica y de Salud Centro de Estudios Sociales y Demográficos (CESDEM) [Demographic Health Survey], Secretaría de Estado de Salud Pública y Asistencia Social (SESPAS) Comisión Ejecutiva para la Reforma del Sector Salud (CERSS); Shisana O, Stoker D, Simbayi LC, Orkin M, Bezuidenhout F, Jooste SE, et al. South African national household survey of HIV/AIDS prevalence, behavioral risks and mass media impact detailed methodology and response rate results. S Afr Med J 2004; 94: Respess RA, Rayfield MA, Dondero TA. Laboratory testing and rapid HIV assays: applications for HIV surveillance in hard to reach populations. AIDS 2001; 15 (suppl. 3):S49 S59.