To our knowledge, this is the first application of the BED-CEIA to estimate HIV-1 incidence from HIV-1 sentinel surveillance data in southeast Asia in general and in Cambodia in particular. Our results suggest that there has been significant decline in incidence from 1999 to 2002 in 2 target groups: CSWs and police. IDSWs showed the same trend, but it was highly significant in only the western region. ANCs were the only group for which the overall trend did not show any decline. These results are also consistent with the trends of HIV-1 prevalence in the sentinel surveillance program in Cambodia4
and the temporal trends of key risk behaviors in the behavioral surveillance program in Cambodia.18
The decline in incidence among CSWs, IDSWs, and police personnel may reflect the effectiveness of the intervention program to reduce heterosexual transmission from the high-risk population to the bridge population through the 100% condom campaign.19
It is not unexpected that this intervention program has had little or no impact on the transmission from those who are already infected to their usual sexual partners and wives, since testing and counseling are not widely available and condom use is low with wives and girlfriends.18
The overall HIV-1 incidence in the western region showed an alarming increase among pregnant women and no decline among CSWs. This trend could be due to the high internal migration to this region. The high HIV-1 incidence rate among CSWs in Cambodia makes this population well suited for HIV-1 prevention research, including vaccine trials.
Several biases influence the estimation of incidence due to the design of the HSS program. First, wide coverage of the surveillance program, coupled with some rotation in personnel supervisors, makes it difficult to keep the standard operating procedure comparable over subsequent years.1–4
Second, the rate of refusal varied from year to year, and the characteristics of those who refused to participate in the study are not known. If those who refused to participate in the study had higher risk behaviors than those who participated, this estimate may have been biased and thus may have affected the analyses of HIV-1 incidence trends. Third, it was assumed that the ANC group was representative of all pregnant women. This may not be true in Cambodia, because poor women are more likely to use ANCs (as opposed to private clinics), if at all. According to the Demographic Health Survey in Cambodia in 2000, only one-third of Cambodian women who had given birth over the previous 5 years received antenatal care from a medically trained person.20
A study conducted in Cambodia comparing HIV-1 prevalence data from the HIV-1 sentinel surveillance to a population-based survey of women in the catchment area showed that HIV-1 prevalence among ANCs provides an overestimate of the general population of women in rural areas.21
Therefore, any attempt to use data from pregnant women to represent the general female population should take this factor into account. Finally, since the younger women were oversampled in 2002 and younger infected women are more likely to be recently infected than older women, the results may have been biased.
Several other factors should also be considered in interpreting the results of HIV-1 testing. We found that among those samples that were brought to UCLA, some were false positives. It is then important to know what the false-negative rate would be among those negative samples; however, this is not known. The findings from current data analyses assumed that there were no false negatives among those negative samples. If this is not true, the incidence estimated here would reflect an underestimation of the incidence in each sentinel group. Another important assumption was that the rates of false positives and recent infections were similar between those specimens that were tested by the BED-CEIA and those not tested (5.4%). A sensitivity analysis was performed assuming 3 different scenarios: all missing specimens were not recent infections; all missing specimens have the same proportion of recent infections as the ones tested; and all missing specimens are recent infections. The sensitivity analysis showed that the trends were basically the same.
Lastly, some precautions need to be taken when interpreting the results of the BED-CEIA used in this study. The absolute estimate of incidence depends on the accuracy of the seroconversion interval. In trend analysis of populations, the error introduced by any inaccuracy of this interval would be systematic, and trends analyses are therefore likely to be valid. However, continued validation of BED-CEIA in cross-sectional specimens with known HIV-1 incidence will be important to assess the findings of this study. Four percent of the persons at the end stage of AIDS may be falsely classified by BED-CEIA as recent infections.14
However, persons with end-stage AIDS are less likely to participate in the surveillance program and would probably already be symptomatic.
When estimating the incidence from the prevalence of recent infections, an important assumption was made that the population was in a steady state, with no net case migration from the pool of prevalent cases, which was conditional on age at migration and age at disease onset. Since the duration of the infection detected by BED-CEIA was only 168 days, the theoretical condition of “stationary population with no net migration” as described here is likely to hold true.
In conclusion, the study found that HIV-1 incidence rates among 3 sentinel groups, CSWs, IDSWs, and police, declined over the study period between 1999 and 2002. These declines may reflect the effectiveness of the intervention programs Cambodia has implemented. However, this study also showed that the epidemic persists in pregnant women in Cambodia and in the western region. Further studies should focus on strategies to protect at-risk women in Cambodia.