• Concentrated epidemics-Epidemics with an HIV prevalence that has surpassed 5 percent in one or more sub-populations presumed to practice higher-risk behaviors but that remains below 1 percent among proxy groups for the general population, such as pregnant women.

• Generalized epidemics-Epidemics in which HIV has spread far beyond the

• sub-populations with higher-risk behaviors, which are now heavily infected, and is higher than 1 percent among proxy groups for the general population, such as pregnant women.

Table 8-1 summarizes the recommendations about which groups to survey in the various stages of the epidemic.

Groups at high risk of contracting or passing on HIV are important to track relatively often at any stage of the epidemic, and bridge groups should be monitored as soon as HIV is detectable in any significant amount among those groups with high-risk behavior. It is recommended that youth populations be surveyed every 2-3 years once a country has entered into a concentrated epidemic stage in which HIV is present in high-risk groups with whom young people might be interacting. General population household surveys are recommended on a less frequent basis because they are logistically difficult and expensive to carry out.

Some of the advantages of non-household surveys are that by going straight to the groups with the higher-risk behaviors, surveys can be done with smaller samples and less movement of survey teams, and therefore can be done more frequently. However, these advantages are counterbalanced by the fact that high-risk groups outside of households can be harder to reach, in the sense of being more difficult to locate and identify. A great deal of care must be exercised to ensure that systematic and repeatable sampling approaches are used that will help to minimize bias. People often compromise in carrying out behavioral surveys with high-risk groups that are hidden and difficult to identify by using convenience sampling techniques to reach them. The price to be paid for convenience sampling is that one can never be certain about who the data really represent, or whether apparent changes in behavior are real. Making the effort to be systematic about sampling groups that seem to be disorganized almost always pays major benefits. More details on this subject are included later in this chapter in the section on "Whom to Survey."

Table 8-2 provides examples of some mixes of target groups that have been chosen for behavioral surveillance surveys (BSS) in different countries.

Just as there are groups that are important to survey, it is important to point out that certain groups are inappropriate for behavioral surveillance. Notably, some of the groups that make sense to monitor for HIV surveillance, such as pregnant women and STI clients, make little sense for behavioral surveillance. Asking women in the later stages of pregnancy about their sexual behavior and condom use will not generally yield results in any way typical of the female population at large. Similarly, if one is trying to measure increases in condom use as a way to monitor reductions in risk behavior in a given high-risk "source" population, then sampling STI clients to represent the high-risk population produces a built-in selection bias because STI clients by definition are not likely to use condoms consistently. (More information about general population and target group surveys can be found in previously published documents3-5.)

Considering Measurement Issues

It is important to adhere to a minimum standard of rigor in the conduct of behavioral surveys, if high-quality data are to be obtained. If it is expected that behavioral indicators will shed light on understanding the dynamics and likely future course of localized HIV epidemics, then it must be recognized that it takes time and effort to do good work. Cutting corners and oversimplifying the process in the name of convenience and ease of implementation is ultimately not helpful because the information produced from such research is not meaningful for monitoring behavior change and evaluating the effects of interventions. In the discussion that follows, three important issues related to collecting high-quality information are discussed. These include:

What to Measure
The goal of tracking behaviors in the context of HIV prevention is to learn about behaviors that are critical to the spread of the epidemic. It is very useful to define these behaviors as indicators, so that they may be tracked in a consistent manner over time. Because prevention programs aim to reduce unsafe sexual and drug-taking practices, indicators are designed to track such key indicators as:

It is worth mentioning that each of these areas represent the very end-stage of behavior change in most target groups, and that there are many other important knowledge and attitudinal changes that must take place before individuals get to the point of changing these behaviors. Program planners need to be aware of all the intermediate steps that individuals need to take so that communities can get to these endpoints. Many other behavioral research techniques are available as tools to help assess the needs and progress of communities targeted by interventions. As alluded to earlier in the chapter, these include formative research using qualitative techniques and rapid quantitative surveys. However, with regard to quantitative indicators, it is important to think carefully about the limited set that are worth measuring, and the hard work required to measure them well.

One of the challenges faced in measuring indicators is in knowing how to define such concepts as "high-risk" partners and "condom use." There are many different ways to define each of these. For example, high-risk partners could include paid sex partners or casual partners that one does not know very well. But a supposed "non high-risk partner," such as a spouse, could also turn out to be high-risk, if that person is engaging in unprotected sex with an HIV-infected person. Similarly, there are many different ways to define condom use, including condom use at last sex and "consistent" condom use, which can be thought of as using condoms each and every time there is sexual contact. Condom use is also likely to vary depending on who the sexual partner is, so this must also be factored into the indicator.

Because there are so many different ways to define indicators, it is useful to have a standardized set that can be shared among different groups involved in HIV monitoring and evaluation. Working together to define indicators, it is possible to use collective experience to learn what works well and what does not. Evaluation professionals around the world are pursuing ongoing efforts to refine and improve the recommended set of indicators for monitoring HIV prevention programs. More information on recent recommendations is available in Chapter 10, "Indicators and Questionnaires for Behavioral Surveys," as well as in other published documents 5,6.

Chapter 1 discussed the issue of setting targets for indicators. Targets refer both to the magnitude of effect that it is possible to measure and to the level of attainment to be measured by the indicator. The smaller the change it is desired to measure, the larger the sample size that will be needed to do so. This can sometimes pose a problem for evaluators, because it is usually not practical or affordable to work with large samples. This means that if small changes are occurring, such as a 5 percent increase in the percent of men who used a condom the last time they had sexual intercourse with a high-risk partner, or a 5 percent decrease in the proportion of men visiting commercial sex workers in the past 12 months, these changes might not be detectable with the usual sample sizes. Or, if they were detected, they might not be considered significant in a statistical analysis. It is easier to measure larger changes, but prevention programs may not be capable of achieving those large changes, especially during short periods of time. In these instances, what may look like program failure may really be a failure of the methodology to be sensitive enough to measure the change. In addition, despite our best efforts, we do not always know how much behavior change is reasonable to expect in different settings and among different populations because of the complex and unpredictable nature of human sexual behavior.

We must therefore exercise caution in not over-interpreting behavioral indicators, and also in setting reasonable measurement goals in terms of feasibility and cost. It is generally recommended that repeat surveys attempt to measure change on the order of 10-15 percentage points. However, it is most useful to look at the trends over several rounds of data collection rather than to focus on the difference between any two specific data points. This is because of the strong possibility that any individual data point might well be the result of random fluctuation or chance. It is better to rely on several data points to tell the story than just one or two. This is where the real power of repeated measures comes into play.

It must also be kept in mind that at a given point in time, indicators may reach their maximum levels of change, at which point the focus must switch to the equally important event of sustaining behaviors at their present level. When indicators are close to their optimum value, it becomes futile and impractical to keep trying to measure smaller and smaller increments of change. If, after several rounds of behavioral surveillance, behavior changes seem to be leveling off, it may be time to reduce frequency from once a year to once every 2 years. If, on the other hand, risky behaviors persist at high levels, then renewed intervention efforts and continued behavioral surveillance are called for.

Figures 8-1 and 8-2 illustrate repeated measures of behavioral indicators in which positive results were achieved. One is among sub-populations in Tamil Nadu, India, and the other among female sex workers in Cambodia7,8.

Whom to Survey
The issue of how to sample "hard-to-reach" target groups, such as injecting drug users, sex workers, and in particular, mobile and migrating populations, is one of the most difficult challenges that exists for those involved in behavioral surveillance. Although it is frequently said that it is not possible to do random sampling with these "moving targets," it is definitely possible to use sound, systematic sampling approaches. Such approaches, if used consistently from one round to the next, increase the likelihood of obtaining reliable estimates of indicators with a minimum of bias. Given the urgent need for scientifically defensible data on behavioral trends in the groups most affected by the epidemic, there is a need to move from non probability-based to probability-based sampling to the extent feasible.

It is not always necessary to have a comprehensive list of all sub-population members to do probability sampling. In fact, because of the sensitive nature of the data being collected, which in many instances involves identifying individuals engaging in illicit behaviors and interviewing them about those behaviors, it is, in fact, preferable not to have a list. This is because of the imperative need to respect the privacy and ensure the confidentiality of all respondents who give their consent to participate in behavioral surveys. All that is needed to do probability sampling is information (or maps) of sites where individuals from the sub-population in question can be accessed. Probability methods can be used for all groups for which a sampling frame of sites or locations where group members congregate can be constructed. For groups for which sampling frames of sites/locations cannot feasibly be created, network or snowball sampling approaches can still be used in such a way as to improve the reliability of estimates. Detailed information on how to conduct sampling for surveillance can be found in Chapter 9, "Sampling Strategies for Monitoring HIV Risk Behaviors," and also in published guidelines for behavioral surveillance 5.

How to Obtain Valid Results
Equally as challenging as sampling hard-to-reach groups is the quest for valid results. If it cannot be assured that the data collected are meaningful, then it does not matter how perfect the sampling may be. The results will still not be useful. Apart from sampling error and selection bias, two main sources of error interfere with the ability to gather valid data. One comes from the people collecting the data and the other from the survey respondents. Both contribute to systematic error and both are avoidable if care is taken to exercise quality control during fieldwork.

It is commonly said that people do not tell the truth about their sexual behaviors, and that they exaggerate, withhold information, or refuse to admit to behaviors that are culturally unacceptable. Women, especially, are thought to be reluctant to talk about sex. However, interviewing techniques exist that increase the likelihood of honest sharing of information. Well-trained interviewers who are well-equipped to handle situations that arise in the field are generally able to encourage honest responses and collect data that are credible and consistent with evidence from other sources.

To increase the likelihood of honest responses, interviewers must be thoroughly trained in open and non-judgmental questioning techniques and in accurate recording of responses. The amount of training required will vary depending on who is carrying out the survey. Where peers of those in the respondent group are selected as interviewers, they may be less likely than professional researchers to appear judgmental. Without adequate training they may, on the other hand, also be more prone to recording or coding responses in a way that reflects their own opinions or behavior. Sometimes, using same-sex interviewers or members of the sub-population, such as MSM or IDUs, to interview other members of those sub-populations can make the total difference between being able to obtain valid information or not. There is some evidence that use of computer-generated questionnaires also improves the likelihood of valid responses9. During the stage of adapting the questionnaire and organizing the fieldwork, these aspects of the survey, such as the profile of interviewers for obtaining the most valid results, the proper environment for conducting interviews, and the time it takes to complete the interview should be pretested. At this stage, it is sometimes necessary to conduct some focus group discussions or in-depth interviews with members of the target group to understand ways to improve validity of responses.

Building rapport with sub-populations with whom behavioral research is being conducted is an essential element of success. When researchers have not worked closely with marginalized communities or taken the time to build support for the research process, the result is very often a lack of cooperation and ultimately an inability to obtain high-quality information. There are great benefits to building the support of the communities being studied by involving them in the research process. In addition to getting more meaningful information for monitoring behavior change and evaluating the effects of prevention efforts, learning to access and build trust in the communities to be surveyed will open the doors for later prevention efforts. Failure to do so can backfire severely and contribute to an atmosphere of hostility and distrust that can last for several years.

Among the numerous issues that can help to obtain valid survey results, many relate to the survey instrument itself. Some things that can be done to improve questionnaires are:

There is some evidence to suggest that people do respond more truthfully when self-administered questionnaires are used, as opposed to face-to-face interview9. However, it is not always necessary that the whole questionnaire be self-administered. Sometimes it is enough for only the most sensitive questions to be self-administered. The important thing is that that the same method of questioning, whether it be self-administered or face-to-face, be used for each respondent in a given sample population.

Other quality control issues relate to interactions with the groups being surveyed. Often these surveys are conducted among communities that are on the margins of society or the law, and therefore reluctant to open up to strangers. Working through NGOs who have relationships with the populations in question is an essential component of this work. In some cases, such as with MSM or IDUs, it will be necessary to use members of the community itself (or those working with them) to do the interviewing.

In terms of the actual data collection, care must be taken to ensure that interviews are conducted in strict privacy and out of the earshot of friends and family members. Intensive supervision of interviewers by experienced staff is also a necessity.

Despite the best efforts to control the fieldwork, the nature of self-reported data is such that it is not possible to verify the results objectively. Therefore, it is necessary to use more than one method to assess behavior change-the concept of multiple method triangulation. In the case of behavioral surveillance, it involves the use of qualitative methods after the survey to help interpret the findings. It also involves the use of biological data (HIV and STI prevalence) and other data, such as recorded condom sales or increased use of STI treatment facilities, to validate the findings of behavioral surveys.

Meeting Special Challenges

Although household surveys with general populations have been conducted for a number of years in many countries in the context of evaluating HIV prevention programs, the experience of repeating surveys among hard-to-reach populations with higher levels of HIV risk is new to most countries. Such surveys require skill, sensitivity, and the backing of the communities involved, all of which take time to develop. Some of the special challenges involved in doing these surveys are discussed below.

Ethical Issues for Hard-to-Reach Groups
Ethical issues are always a concern, but may be even more of an issue when dealing with some of the high-risk groups that are most at risk of HIV. Confidentiality is important for all survey participants, but when the participant of the survey is a group involved in illegal activities, such as sex work, injecting drug use, or illegal migration into another country for work, the importance of protecting privacy is magnified. Sometimes, the factors that make these groups so hard to reach are the very ones that put them at elevated risk of HIV infection in the first place. Yet if they were to be sidestepped for collecting information simply because they are difficult to access and complicated to deal with, then this would be tantamount to sidestepping the epidemic itself.

Researchers must be highly sensitive to the reality that for some of the groups being studied, the survey itself could pose a danger because it might expose members of the group to authorities who would fine or imprison them. Social discrimination as a side-effect of the survey also cannot be easily controlled. Even if results cannot be linked to individuals, if they are linked to a community of people, they can still be quite damaging, and provide the impetus for increased stigmatization.

Special efforts must be made to ensure that survey participants understand their rights and the risks involved, and that every effort is made to ensure that the community will benefit from the data collection effort. One of the simplest, but often overlooked ways to do this, is to involve the community in planning the survey and in disseminating its findings. This can help reduce the perception, frequently held by target groups, that they are merely being used as sources of information that will ultimately benefit other groups.

Who Should Conduct the Research?
Developing the technical capacity of a local institutional base to conduct behavioral surveillance is of paramount importance to ensure continuity over time. The involvement of independent private research firms without a vested interest in the continued use of the data can be problematic and can threaten the sustainability of the system. While this does not necessarily have to be the case, it is clear that without sufficient capacity building and commitments from the national government and international donors, a high-quality surveillance system cannot be maintained. In addition, although many countries are increasing their focus on decentralization and participation at the regional, district, or provincial level, including supporting surveillance, a need still exists for a solid institutional base at the central level to maintain national standards. In some countries, this is being handled by making provincial AIDS committees central to the data collection process and by training a team of interviewers at the local level who can be involved in surveillance on a continual basis from one survey round to the next.

Coordination of Behavioral and Biological Data Collection
It is a good idea to collect behavioral data in the same catchment areas where HIV surveillance is occurring, so that both behavioral and biological trends from the same locations can be observed over time. For the purposes of regular monitoring and evaluation it is not, however, advisable to attempt to collect biological specimens and behavioral data from the same individuals.

Some of the reasons for this are that:

When all is said and done, the logistical and ethical difficulties involved in combining these surveys should be considered in light of the value added, which may be negligible in comparison to the benefits of doing them separately.

Conclusion

It is fair to say that behavioral data provide some of the most useful information available in the fight against AIDS, especially because at present, behavior change is the only weapon available for breaking the transmission cycle. However, good behavioral data are not easy to acquire. Without carefully planned data collection strategies and a strong commitment to establishing high-quality data collection systems, programs will continue to collect volumes of useless data.

Although many different types of behavioral research are needed by HIV/AIDS prevention programs, collecting high-quality behavioral indicators fulfills many objectives, including monitoring the dynamics of the epidemic, identifying the groups most at risk and helping describe the patterns of their behavior, and guiding program planning and evaluating program effects. One of the most useful and powerful ways to collect behavioral data is through repeated cross-sectional surveys in groups that are important to the spread of the epidemic.

Repeated behavioral surveys should be conducted for the general population every 4-5 years and more frequently among high-risk groups that are the focus of HIV prevention efforts. Selecting which target groups to monitor should be carefully thought through, taking into account the stage of the epidemic, the presence of various high risk and vulnerable groups, and the data needs of the country.

Because many of the populations that are in the center of HIV spread are difficult to identify and access, attention should be given to working in partnership with the communities at risk, and using systematic, repeatable approaches to sampling and surveying that will produce reliable, valid and unbiased data.

Ethical considerations are particularly important because many of the target groups that are most useful to monitor are also highly marginalized groups that are sometimes engaged in illegal activities. Without their trust and cooperation, successful survey work cannot be achieved.

Behavioral surveillance is now recognized as an essential component of second generation surveillance that supports the improved interpretation of epidemic trends. Therefore, every effort should be made by countries to establish an institutional base for conducting behavioral surveillance and steps should be taken to strengthen the technical capacity to sustain high-quality data collection systems.

References

1. Brown T, Sittitrai W, Carl G, et al. Geographic and social mapping of commercial sex: a manual of procedures. Honolulu and Bangkok: Program on Population, East-West Center, and Thai Red Cross AIDS Research Center; 2000.
2. WHO/CDS/CSR/EDC/2000.5 and UNAIDS/00.03E. Guidelines for second generation HIV surveillance. Geneva: World Health Organization and Joint United Nations Programme on HIV/AIDS; 2000.
3. Family Health International/IMPACT and Joint United Nations Programme on HIV/AIDS (UNAIDS). Meeting the behavioral data collection needs of national HIV/AIDS and STD programmes. Proceedings from a joint IMPACT/FHI/UNAIDS workshop. Arlington (VA) and Geneva: Family Health International/IMPACT and UNAIDS; 1998.
4. Mills S, Saidel T, Bennett A, et al. HIV risk behavioral surveillance: a methodology for monitoring behavioral trends. AIDS 1998;12(Suppl 2):S37-S46.
5. The BSS Development and Implementation Group. Behavioral surveillance surveys: guidelines for repeated behavioral surveys in populations at risk of HIV. Arlington (VA): Family Health International; July, 2000.
6. UNAIDS/00.17E. National AIDS programmes: a guide to monitoring and evaluation. Geneva: Joint United Nations Programme on HIV/AIDS (UNAIDS); 2000.
7. AIDS Prevention and Control Project (APAC), Voluntary Health Services, Chennai, India. HIV risk behavior surveillance survey in Tamil Nadu: report on third wave, 1998; Supported by USAID.
8. National Center for HIV/AIDS, Dermatology and STDs (NCHADS). Cambodia Ministry of Health. Cambodia's behavioral surveillance survey 1997-1999 (BSS I-III).
9. Des Jarlais DC, Paone D, Milliken J, et al. Audio-computer interviewing to measure HIV-risk behavior among injecting drug users: a quasi-randomised trial. Lancet 1999;353(9165):1657-1661.