In simple scientific terms, the growth of an HIV epidemic is determined by the following factors: average probability of HIV transmission per exposure to an infectious partner, the number of exposures of susceptible persons to infectious partners per unit of time, and the duration of the infectious period1. These factors produce the reproductive rate or the number of new infections (secondary cases) generated by each case, expressed in the following equation: R0 = ßcD, where:

R0 = Basic reproductive rate
ß = Average probability of HIV transmission per exposure to an infectious partner
c = Number of exposures of susceptible persons to infectious partners per unit of time
D = Duration of infectious period

This simple formula of R0 becomes more sophisticated once the variability or heterogeneity in sexual activity (the variance of the probability distribution of new partners per unit of time) within a defined community is taken into account2. A high variance created by a core group of highly sexually active individuals may ensure that R0 exceeds 1.0 in value even when the mean rate of acquisition of new sexual partners in the wider population is small. The heterogeneity of sexual behavior, therefore, plays a very important role not only in determining the course of the HIV epidemic but also in selecting an adequate prevention and control strategy.

A schematic illustration of the relationship between incidence, prevalence, and mortality is shown in Figure 14-13. The basic reproductive number (R0) is the number of new infections caused on average by one infectious individual in an entirely susceptible population. The reproductive number at time t (Rt) changes as the epidemic progresses. At the beginning of the epidemic's growth phase, HIV incidence and HIV prevalence are likely to grow exponentially in the population at risk. As the epidemic grows, the number of people who are susceptible to HIV infection will decrease in the population at risk. At the same time, the proportion of contacts of those infectious with other members of the population who have already been infected will increase. This effect reduces the reproductive rate of the infection and will slow the growth of incidence. In this scenario, HIV incidence will eventually decline (Rt<1) while HIV prevalence continues to grow.

In the transition phase between the epidemic growth phase and the endemic steady state, HIV incidence stabilizes (Rt=1). HIV-associated mortality, however, becomes greater than the incidence of new infections and, as a consequence, HIV prevalence will decline. Once the in-flow of new HIV infections and out-flow of HIV-associated deaths reaches an equilibrium in the population, prevalence will remain stable at an endemic level.

In reality, however, the above parameters are difficult, if not impossible, to prescribe and quantify in populations. This is because the population dynamics of HIV transmission depend to a great extent on a number of factors, including the prevalence and intensity of multiple sexual partnerships, temporal patterns of sexual exchange rates (concurrent or serial), the presence of overlapping or closed sexual networks, and the prevalence of HIV and other sexually transmitted infections (STIs). Also, these factors are neither uniform within a given population nor stable over time.

Other factors influencing this "natural course" of an HIV epidemic include the response of the individual and the population as a whole to the epidemic. Changes in behavior may be induced by effective prevention strategies or by the heightened awareness of the problem when increasing numbers of members in the population develop the disease. On the other hand, reversals in initial reductions in risk behavior may occur when those initially targeted "age out," and new generations enter into their sexually active life span.

Therefore, measurements of HIV spread and trends over time cannot be interpreted without knowledge of levels and changes in risk behavior. Changes in behavior should result in changes in HIV incidence and prevalence and, given the difficulties in measuring risk behavior as discussed in previous chapters, effectiveness of programs aimed at behavior change can ultimately only be assessed when compared with trends in HIV incidence and prevalence.

A consensus is emerging among evaluation experts that prevention programs need to investigate trends in infection alongside trends in behavior that may lead to that infection4. To this end, HIV data have to be collected in conjunction with behavioral, socioeconomic, and socio-demographic data. The combined analysis of these data sets will provide the necessary context and range of information for an interpretation and explanation of the epidemiological data collected by sentinel serosurveillance surveys.

When reductions in HIV prevalence and HIV incidence are observed at the population level, however, the following key questions arise:

• Are the observed changes a reflection of the natural progression of the epidemic?
• Are the observed changes a product of behavior change?
• Are the observed changes a result of prevention interventions?

While some country surveillance systems provide convincing evidence that population-level behavioral change results in a decline or continuing low prevalence of HIV infection, a perfect correlation between the two is unlikely and, in fact, impossible to prove. As described in the general model above, other factors that are unrelated to behavioral change may be working to reduce HIV prevalence. These may include:

• saturation of the epidemic among susceptible individuals;
• increasing AIDS-associated mortality, especially in mature epidemics;
• reductions in symptomatic and asymptomatic STIs through improved treatment; and
• population differentials related to in- and out-migration patterns.

Before assigning credit to behavioral change, all of these factors should be investigated with available data to determine whether they may have a leading role in HIV epidemic decline.

The difficulties in establishing evidence of a relationship between behavior change and epidemic decline is further compounded if we add the claim that the documented behavioral change is caused by intervention effects. This may also not necessarily be the case. For example, behavioral change may be caused by the increasingly visible presence of people with HIV/AIDS over time, which helps to personalize risk. This may happen independent from any programmed intervention.

Monitoring HIV Trends

Ideally, HIV spread and changes over time are tracked through measuring the number and distribution of new infections in a population. True incidence data, however, can only be obtained through large-scale cohort studies. Such studies have many drawbacks, including cost, ethical considerations, and biases due to selection and the fact that those included in a cohort will inevitably have more exposure to HIV programs or intervention efforts.

In the absence of true HIV incidence data, a system of sentinel surveillance for monitoring HIV infection among selected populations has been recommended5. HIV sentinel surveillance uses unlinked and anonymous HIV testing of residual blood specimens left over from samples collected for clinical purposes. Using this approach, it is possible to perform HIV tests without informed consent and minimize the possibility of participation bias associated with voluntary testing.

Antenatal clinic (ANC) attenders have been selected as a suitable "sentinel" group because they are thought to represent most closely the HIV prevalence of the general sexually active population. ANC data are the primary source of data for monitoring HIV prevalence trends in most developing countries, especially in sub-Saharan Africa.

Comparisons between ANC sentinel surveillance and general population serosurveys, however, have shown that data from pregnant women may differ significantly from the general population data and the relationship can go in different directions at different stages of the epidemic, and for different age groups6,7. The considerable variation in the findings suggests that extrapolations from ANC data should be made with caution. Because of the large denominator used (general population of women), even a small percentage difference in HIV prevalence between pregnant women and women in the general population could potentially result in an over- or underestimation of the total number of women infected.

ANC sentinel data also are subject to selection biases related to convenience sampling (sites are not randomly chosen), usage and coverage of ANC services, differentials in risk behaviors and contraceptive use, differences in fertility between HIV positive and HIV negative women, and other socio-demographic factors (such as age distribution of antenatal clinic attenders, level of education, and socio-economic status). Moreover, insufficient data exist on the relative importance of these factors in different settings, and even less is known about how these factors may vary over time.

Lower fertility rates have been found among women with HIV-1 infection8. It has been shown that differential fertility between HIV-infected and HIV-uninfected women can have substantial effects on ANC prevalence estimates and the total number of prevalent infections estimated from it. Especially in mature epidemics, the effect due to differential fertility is expected to be most apparent in older ANC age groups. In fact, population-based studies have shown that ANC women may underrepresent HIV levels in the general female population in such settings7.

If factors leading to selection biases remain the same over time, then serial data from ANC sentinel sites will provide a solid basis for analysis of HIV trends in that population. However, the selection biases referred to above may change over time. In that case, trends recorded over time in ANC sentinel populations may differ from those actually occurring in the general population.

Population-based studies carried out periodically in the catchment areas of ANC sentinel sites can help to evaluate these sources of bias in different country settings. Periodic surveys of this type are needed to compare data on HIV prevalence trends in the general population with those obtained from routine ANC sentinel surveillance systems. This approach allows the necessary "calibrations" of results obtained from pregnant women. In addition, these surveys also provide information on HIV levels in the male population.

From Prevalence to Incidence: Focusing ANC Sentinel Surveillance on Younger Age Groups
In mature epidemics, the majority of new HIV infections are now occurring in young people9. By concentrating resources on younger age groups in ANC surveillance efforts, it will be possible to obtain more information on relatively recent infections. Bias due to differential mortality and fertility will be of less concern in young antenatal clinic attenders. It has therefore been recommended that, while continuing to collect data in all age groups, oversampling should be attempted in the age range of 15-24 years10. Because sample sizes must be large enough to allow a more detailed age-stratification, key sentinel sites with a high volume have to be identified. This could mean that countries may have to downsize the number of sentinel sites in the interest of concentrating resources where they will be most useful.

Linking Behavioral Data Collection And HIV Serosurveillance

To minimize biases, not jeopardize the validity of the serological data, and gather less biased behavioral data on the population as a whole, it is recommended that blood samples and risk behavior interviews be obtained from different individuals4. To establish a better association between behavior and HIV prevalence in the community, however, the data on HIV serostatus among ANC women and behavior in the general population should be drawn from the same source population. This means carefully defining the population from which a key sentinel surveillance site (such as a large urban antenatal clinic) draws its attendees and collecting behavioral data from a random selection of households in the same catchment area. If this is done as part of national or regional behavioral surveys, it may mean deliberately oversampling in the catchment populations of key sites (that is, the population that is served by the particular site in question).

To link the behavioral data with the HIV prevalence data, it is therefore recommended that a minimum set of socio-demographic questions be asked of all antenatal clinic attenders at sentinel sites. These data would include age, parity, last birth interval, level of schooling, occupation, and length of time living in the area (as an indicator of migration). These parameters can then be compared with those collected in the population-based behavioral surveys, allowing any systematic differences between the two groups to be identified and adjusted for in the analysis.

Country Examples of Behavioral Change and HIV/STI Decline
The following section describes and discusses the experience of Uganda and Thailand–two countries where enough reliable behavioral and biologic data have been generated through national surveillance systems to strongly indicate the relationship between large-scale population behavioral risk reduction and declines in HIV and STIs.

Uganda
A rapidly disseminated epidemic occurred in Uganda during the 1980s. By 1992, HIV prevalence among female attenders of selected ANC clinics in Kampala had reached about 30 percent. Other urban sites in the country reached similar levels.

In 1993, HIV prevalence began a slow decline among the ANC clinic population in Kampala, and by 1996 it had reached 15 percent, half of the prevalence only 4 years before. Furthermore, HIV prevalence among women attending a antenatal clinic in Mulago showed significant declines in all age groups except persons aged 38 years and above. The overall prevalence rate decreased from 28.1 percent in 1989-90 to 16.2 percent in 1993. As shown in Figure 14-2 for the example of pregnant women in Nsambya, the declines were most pronounced in the youngest age groups11.

What caused this decline in prevalence? Although several factors may have worked together to produce this decline, the strongest lies in the documented evidence of extensive behavioral change at the population level in Uganda. A representative adult survey in the country in 1989 indicated substantial risk among both men and women: 38 percent of men and 19 percent of women reported at least one non-regular sex partner in the past year. Similarly, a majority of both male and female youth were sexually active (69 percent and 74 percent, respectively).

During the years following this national survey, Uganda initiated a massive HIV prevention program consisting of a variety of strategies. Simultaneously, people with AIDS began visibly appearing in society, buttressing interventions with human evidence, which helped to increase perceived risk and move individuals toward behavioral change.

A follow-up national survey in 1995 of the same universe of adults conducted by the Uganda Ministry of Health indicated that behavioral risks had significantly declined12,13. Both adult men and women had cut their risk by about one-half: only 15 percent of men and 6 percent of women reporting having a non-regular sex partner in the past year. Youth populations had similarly reduced their risk, with a greater percentage of them delaying onset of intercourse: In 1995, fewer than 25 percent of 15-year-old boys and girls reported ever having had sex, down from about 50 percent in 1989 (Figure 14-3).

Condom use increased during the same period: In urban areas, 61 percent of men and 48 percent of women reported condom use in non-regular partnerships.

Thailand
The first case of HIV in Thailand was discovered in 1984. By 1989, prevalence among sex workers in the northern city of Chiang Mai was 44 percent and among injecting drug users almost 40 percent. Over the next few years, an extensive national surveillance system documented prevalence increases among both high-risk groups as well as among the general population. HIV infection among brothel-based sex workers climbed steadily until 1995 when over one-third were HIV-positive. In that same year, 12 percent of "indirect" sex workers found in bars, restaurants, and lounges tested HIV-positive. During the same time period, infection levels in the general population also climbed. By 1993, HIV levels in 21-year-old male military recruits throughout the country had reached levels of almost 4 percent, with levels of 10 percent and more in some parts of the country. A year later, over 2 percent of women in ANC clinics tested HIV positive.

In 1994, hopeful signs started to emerge of decreasing new infections in the country. That year, the overall prevalence among army conscripts decreased to about 3 percent, and the reductions were even greater in the North, where the prevalence had been highest (Figure 14-4). Furthermore, new cases of five other STIs seen in government hospitals and STI clinics decreased by over 80 percent between 1989 and 1996 (Figure 14-5).

What evidence do we have that these major reductions in STIs, including HIV, were caused by behavioral change? The available behavioral data show that in Thailand as in Uganda, these reductions were preceded by significant behavioral change. In Thailand, this took the form of reduced patronage of sex workers by males as well as increased condom use in commercial sex relationships (Figure 14-6). This temporal linkage of behavioral change followed by biologic reductions is an important factor in strengthening the causal relationship.

For example, only 24 percent of national army recruits reported visiting a sex worker in 1995, down from 57 percent in 199116. Condom use questions attached to the national HIV surveillance survey indicated that between 1989 and 1993, reported condom use increased from 14 percent to 94 percent of commercial sex acts. These behavioral changes most definitely slowed down HIV transmission, but it was not until 1994 that they were reflected in decreased HIV prevalence rates.

Strengthening the Link Between Behavior Change and Epidemic Decline

To detect and build evidence for behavioral change leading to HIV epidemic decline, methodologies must be implemented to provide the necessary data. In their most recent guidelines for "Second Generation Surveillance Systems," UNAIDS and WHO have stressed a combination of regular and systematic HIV surveillance combined with STI and behavioral surveillance10. Establishing these monitoring systems will allow analyses similar to those highlighted in Uganda and Thailand, which illustrate significant population-level behavioral change preceding HIV prevalence decline.

The state of Tamil Nadu, India, has integrated HIV and behavioral surveillance so that both behavioral and HIV trends are monitored in key groups. The HIV surveillance system, supported by the Government's Tamil Nadu State AIDS Control Society, is monitoring HIV rates in both high-risk and general population groups to assess epidemic spread. These groups include symptomatic or asymptomatic STI patients, truck drivers (through to 1997 only), tuberculosis (TB) patients, and ANC clinic attenders. Conversely, the Tamil Nadu behavioral surveillance system, sponsored by the USAID-funded AIDS Prevention and Control Project (APAC) of Voluntary Health Services (VHS), has been monitoring risk behaviors since 1996 among selected population sub-groups whose risk reduction is key to reducing HIV levels in both high-risk and general population groups. These groups include female sex workers, truckers, male and female factory workers, and female and male students (through 1997 only).

HIV surveillance has shown steadily rising HIV prevalence rates in all these groups. By 1998, HIV had climbed to 14.7 percent among STI clinic attenders, 7.7 percent among TB patients, 9.4 percent among truckers (in 1997), and 0.95 percent among ANC clinic attenders.

The results from the sub-groups selected for behavioral surveys have indicated substantial behavioral risk reduction. Both truckers and male factory workers reported fewer non-regular sex partners, including commercial partners, between 1996 and 1998, and both of these groups, together with the female sex workers, reported higher levels of condom use in commercial sex during the same period.

While these changes have definitely slowed the spread of the epidemic, they have not yet resulted in a decline of the overall HIV prevalence level. It is the task of the future waves of systematic data collection, both HIV and behavioral, to detect when this critical downward trend in the epidemic will occur.

Conclusion

The theoretical considerations and country examples above have shown the importance of combining serosurveillance data with data from behavioral studies to understand patterns and trends of HIV in populations. Furthermore, as also discussed in the guidelines for second generation surveillance of HIV infection, it is essential to analyze data–both on HIV status and behavior–by age. A focus on young people allows the collection of data that more closely, and with less bias, correspond to recent developments in the epidemic. Thus, the impact of program efforts on HIV transmission can be detected much earlier in these groups than in interventions that include the whole population.

Models and thought have also focused on the question of how much behavior change is necessary to bring the epidemic to a halt17. Clearly, if risk behavior is extremely rare, the chances of acquiring or transmitting HIV infection will be very low and an epidemic will not develop. However, this concept holds only true on a population level, and by no means can guarantee that an individual will be safe when exposing him or herself by engaging in risk behavior. HIV is present in virtually all countries and regions of the world, and prevention thus needs to aim at all risks that may occur.

While the classic concept of core groups for transmission of HIV and other sexually transmitted infections has a lesser meaning in most of the high level and generalized epidemics in sub-Saharan Africa, in other regions of the world this concept still has important implications for understanding and preventing the infection. There, risk behavior, and consequently HIV infection, is usually concentrated in defined sub-populations, such as sex workers and their clients, men having sex with men or injecting drug users. While risk levels in the population overall may be low (on average), which prevents the epidemic from taking off at high levels, HIV is often concentrated within the populations at highest risk. Thus the simplistic model described in the beginning of this chapter has to be revisited when applied to populations at large. In such situations it is more appropriate to speak about several epidemics that may well occur in parallel in different populations in a country. In fact, genetic analysis of HIV specimens isolated from different populations in Thailand have shown that epidemics in drug users seem rather distinct from those in homosexual men or female sex workers and their clients. While the spread of HIV in low-risk populations is primarily triggered through sporadic infections through contacts between members of low- and high-risk groups, the epidemics within these high-risk populations are characterized by a rapid spread of HIV in a relatively short period of time and higher HIV prevalence levels.

The determinants of HIV transmission described at the beginning of this chapter establish the theoretical basis for current HIV/AIDS prevention efforts. The challenge for program designers lies in trying to identify the most effective ways to decrease HIV transmission by influencing these determinants and to translate this theoretical concept into feasible interventions in the field.

References

1. May RM, Anderson RM. Transmission dynamics of HIV infection. Nature 1987;326(6109):137-142.
2. Anderson RM. Some aspects of sexual behaviour and the potential demographic impact of AIDS in developing countries. Soc Sci Med 1992;34(3):271-280.
3. Joint United Nations Programme on HIV/AIDS (UNAIDS). Trends in HIV incidence and prevalence: natural course of the epidemic or results of behavioural change? Geneva: UNAIDS; 1999. Best practice collection 99.12E.
4. Family Health International/IMPACT and Joint United Nations Programme on HIV/AIDS (UNAIDS). Meeting the behavioural data collection needs of national HIV/AIDS/STD programmes. Proceedings from a joint IMPACT/FHI/UNAIDS workshop. Arlington (VA) and Geneva: Family Health International/IMPACT and UNAIDS; 1998.
5. WHO/GPA/DIR/88.8. Sentinel surveillance for HIV infection: a method to monitor HIV infection trends in population groups. Geneva: World Health Organization; 1988.
6. Fontanet AL, Messele T, Dejene A, et al. Age- and sex-specific HIV-1 prevalence in the urban community setting of Addis Ababa, Ethiopia. AIDS 1998;12(3):315-322.
7. Fylkesnes K, Ndhlovu Z, Kasumba K, et al. Studying dynamics of the HIV epidemic: population-based data compared with sentinel surveillance in Zambia. AIDS 1998;12(10):1227-1234.
8. Gray R, Wawer M, Serwadda D, et al. Population-based study of fertility in women with HIV-1 infection in Uganda. Lancet 1998;351(9096):98-103.
9. Taha TE, Dallabetta GA, Hoover DR, et al. Trends of HIV-1 and sexually transmitted diseases among pregnant and postpartum women in urban Malawi. AIDS 1998;12(2):197-203.
10. 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.
11. STD/AIDS Programme, Ministry of Health. HIV/AIDS surveillance report. Kampala (Uganda): Ministry of Health; June 2000.
12. Uganda Demographic and Health Survey. Calverton (MD): Macro International Inc.; 1996.
13. Asiimwe-Okiror G, Opio AA, Musinguzi J, et al. Change in sexual behaviour and decline in HIV infection among young pregnant women in urban Uganda. AIDS 1997;11(14):1757-1763.
14. Armed Forces Research Institute of Medical Sciences, Thailand; 2000.
15. Venereal Disease Division, Ministry of Public Health, Bangkok, Thailand; 2000.
16. Nelson K, Celentano D, Eiumtrakol S, et al. Changes in sexual behaviour and a decline in HIV infection among young men in Thailand. N Engl J Med 1996;335(5):297-303.
17. Morris M, Dean L. Effect of sexual behavior change on long-term human immunodeficiency virus prevalence among homosexual men. Am J Epi 1994;140(3):217-232.