Our analyses suggest important, biologically plausible neighborhood-level factors that may, in part, drive CA-MRSA hospitalization rates at the population-level in New York City. In sex-specific multilevel analyses, strong positive associations between higher UHF neighborhood HIV prevalence and MSM proportion with higher odds of CA-MRSA hospitalization persisted, even after controlling for individual factors. Additionally, important differences in individual and neighborhood-level risk factors for CA-MRSA hospitalization between males and females were observed.
Our analysis showed that neighborhood HIV prevalence was associated with increased odds of CA-MRSA hospitalization among females even after controlling for individual HIV-status. While the associations between HIV prevalence and odds of hospitalization with CA-MRSA were not statistically significant at lower HIV prevalence levels, a dose–response relationship is suggested. Studies, including ours, have found HIV positive persons to be at an increased risk for CA-MRSA infection and hospitalization with CA-MRSA, possible due to increased viral load, weakened immune systems, lack of antiretroviral therapy, or problems with skin integrity [4, 11, 26, 27]. This raises the possibility that neighborhood HIV prevalence may play a role in CA-MRSA transmission even among HIV-negative persons, and that the attributable fraction of CA-MRSA hospitalizations associated with HIV may be greater than previously thought.
We also found higher neighborhood MSM proportion to be associated with CA-MRSA hospitalization for both males and females. While we did observe differences by neighborhood MSM proportion at the extreme ends of the distribution, we did not observe a dose–response relationship and the association did not extend to neighborhoods in the highest quintile of neighborhood MSM proportion. A population-based study in San Francisco found that zip codes with higher percentage of partnerships being same-sex male also had higher rates of CA-MRSA infection . In New York City-based analyses, high rates of CA-MRSA observed among MSM were associated with HIV infection, crystal methamphetamine use, physical contact with someone with a skin infection, sex at private parties, and perhaps membership in social networks that include others engaged in these behaviors . Though we were unable to control for MSM status at the individual-level in our analysis of males, the finding that neighborhood proportion of MSM was associated with the risk of CA-MRSA hospitalization among females is intriguing, and suggests the possibility of a neighborhood-level effect. Epidemiologic data suggest that HIV prevalence among MSM in New York City is high (e.g., 8.8% overall, and 17.7% among MSM in Manhattan) , making it difficult to distinguish between the potentially independent role of each factor. The lack of a dose–response relationship among males or females, however, make these results difficult to interpret, and the association between neighborhood MSM proportion and CA-MRSA needs to be evaluated in future research before stronger conclusions can be drawn.
Epidemiologic studies have shown that neighborhood factors have independent associations with a number of health outcomes, including neighborhood socioeconomic status with diabetes  and heart disease , and neighborhood poverty with relapse into injecting drug use . Although neighborhood characteristics have commonly been explored with regard to chronic diseases and health outcomes, they likely also affect infectious disease risk. The extent of transmission or acquisition in a given geographic area is dependent upon a number of relevant factors that may in fact vary by neighborhood, including the number of hosts susceptible to infection and the number of hosts who are currently infected or are carriers of the microbial agent. Contacts between infectious and susceptible persons will occur in many settings, including the neighborhood in which infectious persons reside .
Our analysis has limitations that merit discussion. At the individual level, V-code 09.0 and ICD-9 CM codes indicating Staphylococcus infection have been used in published literature to identify MRSA cases in administrative data [4, 31–34] though some analyses have shown conflicting results as to the accuracy of these definitions [35, 36]. Schweizer, et al. found that use of administrative coding lacked sensitivity and have low positive predictive value for hospital-associated infections . A second study by Schaefer, et al. found low sensitivity, leading to an underestimation of the number of cases, but high positive predictive value particularly for community-associated infections . In addition, lack of laboratory data and healthcare history may have led to misclassification of susceptible versus resistant infection and of community- versus hospital-associated infection.
It was not possible to apply the Centers for Disease Control and Prevention’s definition of CA-MRSA to our dataset ; however, study inclusion and exclusion criteria were used as a proxy. Patients could not be linked across calendar years and therefore examining the full 12 months prior to hospitalization was not possible. This particularly affects those patients hospitalized early in 2006. Additionally, the definition used in this analysis likely resulted in exclusion of some CA-MRSA cases (those with certain chronic diseases also associated with HA-MRSA) but was necessary to improve the specificity of our outcome definition. Staphylococcus infections which were not recorded on the hospital discharge record or incorrect diagnosis coding could have resulted in some misclassification. Specifically, Staphylococcus infections other than CA-MRSA may have been incorrectly classified as CA-MRSA rather than excluded from analysis. Finally, the primary cause of hospitalization need not have been CA-MRSA, which may lead to an overestimation of hospitalizations primarily due to CA-MRSA.
At the UHF neighborhood level, HIV prevalence, MSM proportion, and ED usage values may be underestimates due to lack of diagnosis, reporting, or inaccurate survey responses and 2000 Census data may not accurately represent neighborhood income in 2006. Misclassification of these neighborhood-level covariates was non-differential by outcome status at the individual-level. Additionally, neighborhood measures are aggregations of individual-level information. Variation will exist among individuals living in a single neighborhood. Finally, there are many ways to designate “neighborhoods” in geographic analyses, including UHF, zip code, cities, et cetera and the specific designations used in our analysis may have affected our results. Due to small numbers of neighborhoods, we were unable to control for multiple neighborhood-level factors in the same model, 7which could result in uncontrolled confounding at the neighborhood level. We were also unable to measure individual socioeconomic status and MSM status and therefore could not control for them in modeling.