The analysis of the Cross-Border Project data reveals consistent reductions in drug-related HIV risk behaviors, HIV prevalence, and HIV incidence among IDUs in the project sites in Northern Vietnam and Southern China. Since this is neither a randomized trial nor a quasi-experimental study but rather an ecological study based on cross-sectional data, we cannot draw firm conclusions that our interventions caused the positive trends we have observed. However, this problem is reduced in that the Cross-Border Project is implementing structural interventions, which are appropriately evaluated using community cross-sectional data. In addition, the consistency in trends across the primary outcomes strongly supports a positive intervention effect.
The Cross-Border findings suggest that peer-based interventions for IDUs that provide large numbers of needles/syringes by multiple means (direct distribution, pharmacy vouchers, and secondary exchange) without limits on numbers per contact can control HIV epidemics among IDUs. There have been similar findings from other places, for example, New York City where reductions in HIV incidence among IDUs followed large-scale implementation of peer outreach and needle/syringe programs 
The successful use of current IDUs as peer educators is also a noteworthy feature, as such PEs were demonstrated to have efficient access to and trust among the target population. Coverage of the Cross-Border interventions has been quite high (60%–70% of IDUs) in all sites but some reductions in coverage were observed in Ning Ming, China in the most recent cross-sectional surveys. This, and concomitant reductions in the average monthly numbers of needles/syringes provided, have probably resulted from reduced levels of funding for the interventions in Ning Ming. However, these changes have not yet occasioned any rebound in HIV prevalence or incidence among IDUs. This suggests that the interventions are truly structural in that they have changed behavioral norms in the IDU community and thus no longer need directly reach very high percentages of IDUs in order to maintain low levels of HIV risk behavior and associated HIV transmission. The success of such structural interventions depends greatly on building and maintaining community support and collaboration with police and other officials. The importance of such support has been noted in assessments of other needle/syringe programs in Vietnam 
and has been successfully maintained over the long term by the Cross-Border Project 
There are several other issues and potential limitations regarding the design of the study and the methods of collecting and analyzing the primary outcome indicators that require discussion. It is possible that our survey samples are not fully representative of IDU populations in the sites. The use of snowball sampling methods may introduce sampling bias and dependence that may influence analytic results. However, we are unable to explore or adjust for such influences because, in part to protect participants’ confidentiality, we did not record which participants came from which snowball strings. Nevertheless, in studies of IDU populations, it is common practice to use targeted or snowball samples because these are often the best methods that are feasible. This point is supported in several extensive literature reviews 
Self-reported behavioral data may be subject to biases. For example, the very low reported rates of needle/syringe sharing in the Vietnam sites from the very first survey waves may reflect responses based on social desirability. In fact, this seems likely given the high HIV incidence among new injectors in the early survey waves.
High risk sexual behaviors remain prevalent among IDUs, indicating a need to intensify interventions in this domain. Sexual partners of IDUs may be at particular risk. Indeed, cross-sectional surveys in this group, conducted to evaluate companion interventions for women at risk, revealed HIV prevalence of 2%–5% in Ning Ming and Ha Giang and a sharply higher 22% in Lang Son (Abt Associates, unpublished data). Cross-sectional surveys of sexual partners of IDUs in Hanoi reveal HIV prevalence between 9% and 14% 
In general, biological measures are more reliable than self-reported behavioral data for assessing the effectiveness of interventions. A strength of our evaluation is that we have collected biological measures. The Cross-Border data reveal major reductions in HIV prevalence among IDUs in Lang Son and Ha Giang and more modest declines in Ning Ming. Declining prevalence may reflect factors unrelated to intervention effects, such as deaths or mobility in the target or surveyed population. However, the declines in prevalence in our project sites are not likely due to underlying trends, since other provinces of northern Vietnam and in Yunnan Province, China, without such interventions did not experience similar declines in prevalence. These comparisons support the conclusion that the Cross-Border interventions have had a positive effect on declining HIV prevalence among IDUs.
HIV incidence is the most important indicator of effectiveness but few evaluations of HIV prevention interventions for IDUs have examined incidence trends. The Cross-Border project analysis reveals reduced HIV incidence in all sites although there are some differences between the new injector and BED-based incidence estimates. Incidence among new injectors declined sharply in all sites in the survey waves following full implementation of the interventions 
but there appear to be some rebounds in later waves, especially in Ning Ming. We believe that these later results may be unreliable, particularly given the much lower incidence for these same waves based on BED testing. The new injector estimation method relies on an assumption that a new injector was HIV-negative at the time he initiated injection. While this may have been a safe assumption for the early survey waves when the HIV epidemic was growing sharply among IDUs, it may not be reliable for the later waves when the epidemic had become more mature and more sexual transmission had begun to occur. In these later waves, it is possible that an individual could already have been infected with HIV through sexual contact by the time he initiated injection, which would distort the incidence estimate upward. As a result, we are inclined to give more credence to the BED-based incidence estimates, especially for the later survey waves.
It is also important to recognize the possible problems with the BED-based estimates. The BED assay can be non-specific, particularly in African settings. False-recent BED test results can occur in persons with long-term infection and those on ARV treatment 
. Such errors would tend to produce artificially inflated incidence estimates. However, in a recent study of BED performance in Thailand, where the epidemic is dominated by similar viral subtypes as found in northern Vietnam and southern China, very few false-recent cases were observed 
. Application of adjustments for false-recency in that study reduced the incidence estimates to the conventional cohort-based observed incidence, suggesting that false-recency may be less problematic in Southeast Asia, or in the viral subtypes that predominate in this region. While debate continues on the most appropriate adjustment factor to apply for specific settings 
, we selected the Welte adjustment 
, as recommended by CDC and the Office of the Global AIDS Coordinator 
and as used in the Thailand study 
. Moreover, in our surveys, participant-level data were available allowing removal from analysis of cases that could generate false-recent results on the BED assay.
In any event, overestimation is not a problem for BED-based estimates in the later survey waves, which are already extremely low, except for the Ha Giang 96-month survey in which the three BED-based “recent” infections, if correctly identified, would have resulted in a sharp incidence increase because of the small numbers involved. As reported above, we found that these three cases were all false-recents on the BED test, resulting in a 0% annual incidence estimate for the Ha Giang 96-month survey.
Artificially high estimates in the earlier surveys could undermine the validity of the observed downward incidence trend. However, the BED and other assays to detect recent infections are ideally suited to longitudinal surveys such as this where the objective is to determine trends over time because assay-based biases that might be present at one time would likely persist at others, making it very likely that the downward trend observed in the BED-based incidence is real even if some overestimation existed. Moreover, the estimates from the new injector analysis, the survey responses regarding individuals’ duration of infection, and trends observed after the application of the Welte adjustment 
all corroborate the higher incidence rates in the earlier survey waves and thus support the downward trend.
Combining the incidence trends from the new injector analysis and the BED testing, we find sharp declines in HIV incidence among IDUs in all sites. The only other study of IDUs in Vietnam that has measured HIV incidence was in Thai Nguyen Province, where an intervention without needle/syringe provision was being implemented. HIV incidence in a cohort study conducted there between 2005 and 2007 was 5% per year 
, higher than the rates found in the later Cross-Border survey waves.
Despite the limitations of the study, we conclude that the Cross-Border interventions have played an important role over an eight-year period in controlling HIV transmission among IDUs and, as a result, offer a model of HIV prevention for IDUs that should be considered for large-scale replication.