A steep HIV prevalence decline was observed in young men and women aged 15–24 years in the studied populations. Prevalence declined also in urban men and women aged 25–29 years, whereas it was mostly stable in older age groups. In age group 15–49 years, a marked prevalence decline appeared in urban men and women, but the rural decline was modest and not significant. HIV-related mortality is likely to have contributed substantially to the overall decline. A 3-years cohort study based on the 1995 survey found HIV-related mortality to be generally low among young people, except in urban young women where HIV was found to have already affected mortality substantially (unpublished). In the absence of incidence data which was not possible to collect in this study due to ethical limitations, the observed prevalence declines in young people can be a proxy of change in incidence, and selective mortality seemed only to represent a small proportion of the steep decline in young urban women[21
]. Similar patterns of declines have been reported in the region recently[1
]. Furthermore, the declines were strongest in groups with secondary school education or more and were associated with sexual behavioural change.
It is possible that differential infection patterns among non-participants over time could have biased our estimates, but the magnitude and direction of this effect can only to some extent be assessed[2
]. The non-participation due to refusal to provide saliva remained low in all survey rounds, and significantly below refusal levels as experienced in the Demography and Health Survey in Zambia 2001/2002 when using blood as the basis for HIV testing [16
]. Absence was the most significant cause of non-participation, particularly in men. We think that the group of participants reporting to be highly mobile could be used as sentinel of HIV infection for men who were absent. Mobility and migration have been suggested to be factors fueling HIV transmission, but there are relatively few studies documenting it[23
]. Mobile groups in this study had significantly higher likelihood of infection in men overall, but the risk difference was marginal in young men. However, we observed that even in these mobile groups there were significant prevalence declines in urban men and less prominent in the rural men. In younger people, we are persuaded to believe that if this bias was present, its effect was very minimal because the difference in odds of infection by mobility was non-significant and the major reason for absence was being away at school. Rather, it is likely that the prevalence among young people could even have been over-estimated due to the fact that those not found because of school attendance are less likely to be infected[25
]. Notwithstanding the presence of selection biases due to non-response, they are unlikely to be an important factor explaining the sharp HIV prevalence declines among young people. We also note that where significant trends have been observed, all of the trends have been declining, irrespective of the factor being considered. This implies that there is no evidence of any interaction among the factors considered in this study.
External validity is a critical challenge when data from selected communities are extrapolated to the whole population. We found evidence supporting the interpretation that the HIV declines observed in these selected communities approximate well with those of the general population. Firstly, the prevalence levels of selected urban and rural communities matched with respective national estimates, and this was one of the criteria for selecting them[17
]. Secondly, national ANC-based estimates show declines among young women[5
]. Thirdly, we have previously reported that ANC-based trends under-estimate declines in the general population[10
]. The main explanation for this reduced representativeness of ANC-based data was substantial delayed age at first birth among women in the general population.
HIV prevalence is a reflection of accumulations of infections over time, but during this same period, the population might change considerably[2
]. Trend therefore reflects a time averaged dynamic balance between incidence, migration and mortality. The observed decline in prevalence could have been influenced by any or a combination of these factors. However, we realise that the changes were marked in young people where mortality is low and where there was no difference in odds of infection observed between in and out-migrants as observed elsewhere in the region[23
]. Furthermore, the decline observed in higher educated young people who have adopted safer sexual practices is plausible because they have grown up during a time when prevention messages on HIV transmission were readily available, and this influenced their sexual behaviour[11
]. In addition, we observe that if this trend continues, it has great potential to dictate further declines among educated people. HIV-related mortality's contribution to overall prevalence declines might reflect the impact of differential HIV-related mortality by age groups[2
]. Long survival was unlikely to have any impact because access to anti-retroviral therapy was limited in this population before 2004.
Men acquired infections later than females and reached peak levels after age 30 years, and even among men aged 50–59 years prevalence levels remained very high. In this region, a gender-generational-power imbalance exists between casual and regular sexual partners. Men are usually older than their female partners and this age difference has been found to be "the major behavioural determinant of the more rapid rise in HIV prevalence in young women than in men"[27
]. Most of the infections in young women are from older men in whom there has been a power change through age rise related affluence[27
]. In order to control the HIV epidemic, breaking this cycle should be the cardinal aim of prevention programmes.
There are still limited data from sub-Saharan Africa linking HIV prevalence to parallel changes in risk behaviours[29
]. We collected behavioural data and HIV status concomitantly and work on detailed parallel sexual behaviour patterns and HIV trends are the focus of a separate paper (awaiting publication). We found reductions in high risk behaviours among young people and that these were particularly in higher educated and urban groups. These findings confirm the assumption that effects of educational attainment on risk of HIV infection is likely to be exerted through mediator factors such as more consistent condom use, lower likelihood of sexually transmitted infections and less number of sexual partners[11