We found that the decreases in annual measures of mean and total CVL in San Francisco were significantly associated with temporal decreases in the number of new HIV diagnoses and accompanied by a decline in estimated HIV incidence by over one-third from 2006–2008. Although the trend in reduction of HIV incidence was not statistically significant, due to several sources of variation in the estimation, the trend across three consecutive years is consistent with the parallel and highly statistically significant decline in newly reported and diagnosed HIV cases and with trends in several indicators of potential transmission. A particularly rapid decrease in CVL occurred from 2005 to 2008, a period characterized by an increase in ART uptake among people living with AIDS from 74%
to 90% 
, availability of more effective, potent, and tolerable ART regimens, and a significant increase in the population rate of virologic suppression from 46.8% of all HIV/AIDS patients in 2005 to 78.1% by 2008 (). The trends are also concurrent with changes in laws facilitating HIV testing 
,policy initiatives promoting expanded HIV testing 
, increasing acute HIV detection 
, partner services 
, and a reduction in the rate of unknown HIV infection. Taken together, the findings support our primary hypothesis that reductions in CVL resulting from increased ART uptake, expanded ART options, and greater virologic suppression can in turn reduce HIV incidence at the population level.
Our analysis also supports the use of CVL as a surveillance biomarker for treatment effectiveness and HIV transmission risk in a given jurisdiction or among a group of individuals. Differences in mean CVL among different groups are consistent with our understanding of the known disparities in HIV/AIDS morbidity and mortality in San Francisco, particularly with respect to low-income neighborhoods, African-Americans, and transgendered individuals 
. The geographic disparities in CVL are consistent with known disparities in socioeconomic status and may also reflect neighborhood-level environmental factors 
that structure HIV risk and access to prevention and care, such as availability of alcohol and drugs, transportation, and proximity to quality health care programs. These factors may be most evident in the low-income and relatively isolated Bayview with its very high mean CVL. As expected, the Castro had the overall highest total CVL given its high HIV/AIDS case density. Yet, the high total CVL in the Tenderloin is worthy of further exploration as it likely reflects both a high number of persons living with HIV coupled with disparities in any of several factors, including the uptake of ART, adherence to ART, or engagement in care. Our findings suggest that targeting HIV prevention to areas with the greatest total CVL, such as the Castro, should produce the greatest reductions in overall HIV incidence, while addressing inequities in both access to care and prevention services should decrease the relative disparities in HIV incidence among the subpopulations and in the geographic areas in San Francisco with highest mean VL.
There are several limitations to our study, including the well-understood problems with surveillance data. In particular, our dataset only included HIV-positive individuals who were diagnosed and reported in the HIV/AIDS case registry and had at least one VL measurement between 2005 and 2008. We were not able to include individuals who are not yet diagnosed. However, the estimated 15–20% of HIV-positive persons who are unaware of their infection is lower in San Francisco than the 25% unaware nationally in 2004 
. Our analysis also does not include those who are diagnosed but not yet reported to the HIV/AIDS case registry, but this is estimated to be less than 5% of all cases 
. We also could not include all persons who are acutely infected with HIV, who may contribute significantly to amplified HIV transmission 
, as many of these individuals are not diagnosed in the acute stage of infection. Omission of acute infections would affect the longitudinal trend in mean or total CVL if (1) acute infections compose a large proportion of the total number of cases or (2) the rate of new infections rapidly changed. However, we currently estimate that acute infections comprise a small minority of all prevalent HIV infections 
in San Francisco. If we were able to improve the detection of acute HIV infections and they are as important a part of transmission as hypothesized by some investigators 
, we might even see a stronger correlation between CVL and incidence rates. The increase in the correlation would be very slight, however, because acutely infected persons are a very small minority of the total population included in CVL. In addition, viral load results were missing for 26% of persons in the registry. We used multiple imputation to replace those missing values, but this requires the strong assumption that the data are missing at random, given the available covariates. This assumption may not hold, so that the MI estimate of mean CVL in any given year remains biased. However, the associations between trends in CVL and new HIV infections could be approximately unbiased, provided the bias is comparable across years. Another limitation is that HIV incidence estimates calculated by the CDC method involve several imprecisely estimated inputs. For example, not all recent HIV diagnoses have remnant samples available for BED assay testing, resulting in an imprecise assessment of recent vs. chronic infection; similarly, not all cases have HIV testing history reported, which is needed to adjust incidence estimates. Thus, while point estimates of HIV incidence declined by over one-third, neither the trend nor the association with CVL reached statistical significance. In addition, we did not have the power to examine whether differences in mean CVL by sub-group were related to disparities in HIV incidence by sub-group. We also acknowledge the important possibility of ecologic fallacy. This is an observational study and there is limited ability to directly assess causality. For example, we cannot determine if transmissions occur from treated or untreated individuals. Similarly, the association between community viral load and new HIV infections could be confounded by different levels of risk behavior among distinct sub-populations. Although we lack routinely collected population-level risk behavior data and could not adjust for that association in our analysis, we do have data on sexually transmitted diseases, such as rectal gonorrhea, as a proxy for sexual risk behavior. Moreover, a decrease in new cases may reflect testing patterns or risk behavior levels. However, newly diagnosed HIV cases decreased concurrently with a significant increase in HIV testing uptake and a reduction in the number of people unaware of their HIV diagnosis, despite increased cases of rectal gonorrhea. That new HIV infections declined during a period when sexual risk behaviors increased, as suggested by increased cases of rectal gonorrhea, lends further credence to our hypothesis that the reduction in HIV infections was driven primarily by reductions in community viral load rather than reductions in sexual risk behavior. Despite these limitations, we believe our data from multiple independent sources provide persuasive evidence that the recent increases in HIV testing, treatment coverage and efficacy, and viral load suppression are having an effect on the HIV epidemic, and that CVL is a useful biomarker of these effects. In summary, reductions in San Francisco mean and total CVL following comprehensive public health measures to increase HIV testing and ART uptake were consistent with apparent declines in HIV incidence and strongly associated with declines in newly diagnosed HIV infections.
There is current, heated debate on whether “test and treat” strategies can eradicate the HIV epidemic or reduce HIV incidence at a population level. To be clear, we do not suggest that our data prove that treatment alone can accomplish these aims. Rather, we believe achieving a high level of ART coverage is an integral part of a comprehensive “highly active HIV prevention” 
approach incorporating behavioral, biomedical, public health 
and structural interventions to the HIV epidemic that has manifest individual health benefits and, we believe, potential prevention effects. While the relative amounts of incidence reduction achievable by behavior change interventions versus VL suppression may be an area of contention, there should be consensus that both are needed. Given the continued HIV transmission that occurs in all areas of the world regardless of whether uptake of ART is high or not, programs intervening on risky behavior (including both sexual and syringe-sharing) also need to be scaled up and improved 
. And until all of the estimated 9.7 million persons 
who need HIV treatment are able to access ART, the urgency to scale up treatment remains as a matter of social justice and human rights, regardless of its prevention effects 
In the meantime, we recommend that public health departments and other organizations consider using CVL as an indicator of the overall success of ART uptake and for HIV prevention efforts mediated through increased HIV testing, early linkage into care, improved engagement and retention in care, and increased virologic suppression. Because CVL is temporally upstream of new HIV infections, it may be particularly useful as an outcome measure for multi-level HIV prevention trials or community-level interventions, including behavioral, “test and treat,” “ART as prevention,” or earlier ART initiation 
strategies at the population level. Additionally, mapping the spatial distribution of CVL may delineate disparities, new “hotspots” or areas with particularly high HIV incidence, allowing a rapid response to target resources and interventions to populations at greatest risk.