We conducted a large, multi-year, multi-city, exploratory study to investigate exposures to primary OC compounds and the association between their chemical classes and emergency hospital admissions among an elderly population. Using a hierarchical two-stage modeling approach, we found consistent associations between grouped primary OC compounds and cause-specific hospital admissions. Our most consistent associations were found between cyclohexane and iso-/anteiso-alkane exposures and CVD- and respiratory-related admissions, respectively. Since the primary sources of cyclohexanes are motor vehicles, our results provide support for the growing literature showing motor vehicle-associated adverse CVD effects
[14, 40], as main sources of cyclohexanes are motor vehicles. At the same time, however, we saw either null or negative associations for hopanes, also markers of motor vehicle pollution, suggesting that the OC chemical structure might be important to toxicity. Moreover, our findings of positive associations between total respiratory and pneumonia admissions and iso-alkanes, a group that has not otherwise been examined in epidemiologic studies, are consistent with previous studies that found community-acquired pneumonia to be associated with inhalation of tobacco smoke
. Given the lack of studies examining the relation of these OC classes and health, our findings should be viewed as preliminary and should be validated in additional studies.
Despite this, our findings of adverse cyclohexane impacts are supported by toxicologic studies, which have linked traffic-related particles with numerous adverse outcomes. There is evidence that the redox chemistry of organic particles plays an important role in the biological effects of ambient PM2.5. Nel et al.
, for instance, found organic compounds from diesel exhaust to induce reactive oxygen species (ROS) in macrophages and bronchial epithelial cells, as well as apoptosis and necrosis in the epithelial cells in the bronchial region. Furthermore, Li et al.
 showed that expression of heme oxygenase-1, a marker for oxidative stress, is directly correlated with high organic carbon content of UFPs. Although there have been no studies examining the toxicological effects of cyclohexanes, it is possible that cyclohexanes may lead to adverse health effects through similar biological processes, since they, like PAHs, can also be transformed by photochemical reactions into more polar products
[44, 45]. If so, it is possible that cyclohexanes are acting as surrogates of atmospheric processed, oxidized compounds.
Few studies have examined health impacts from specific organic carbon chemical classes, with these studies focusing on PAHs and hopanes. In a Los Angeles cohort of 60 elderly subjects with coronary artery disease, Delfino et al.
 found significant associations between intermediate markers of inflammation and weekly exposures to PAHs and hopanes. Kraus et al.
 reported significant associations between exposures to hopanes and PAH and shortness of breath and between PAH and increased C-reactive protein in a cohort of MI survivors in Germany. These findings differ from those in our study, as we found null associations between PAHs and hopanes and increased hospital admissions. The contrasting findings likely result from several key differences in our studies. The earlier studies used different study designs and focused on different health outcomes and examined potentially more susceptible populations, for which primary OC-mediated health risks may differ. In addition, the earlier studies measured a larger number of traffic-related PAHs, e.g. 24
, which may provide more power to detect associations.
Support for our findings is provided by the consistent trends found across exposure windows, with larger associations for both cyclohexanes and iso-/anteiso-alkanes found at exposure windows of 6-days and longer. These longer exposure windows may reflect biological processes, such as inflammation, which have been shown to occur over week-long time periods
. Alternately or in addition, significance at longer exposure windows may reflect increased classical error associated with shorter averaging times
. The wider CIs at longer moving averages could also reflect loss of power; as the averaging period increases the variability in pollutant concentrations decreases.
The robustness of our findings is further supported by consistent findings from our sensitivity analyses. When we restricted our inclusion criteria to species with ≥75% observations above the LOD, results for cyclohexanes and hopanes were similar, while results for PAHs and n-alkanes became negative and positive, respectively, and associations between iso-alkanes and total respiratory admissions became insignificant. The exclusion of species with lower concentrations could result in either decreased noise, as in the case of the PAHs and n-alkanes, or decreased power, as in the case of the iso-alkanes, depending on how conservative the LOD is and the fraction of observations below the LOD. It is also likely, nonetheless, that the exclusion of several species will make the results non-interpretable as a group. For example, in the sensitivity analysis, we only included one softwood and biomass combustion marker (retene) and a general combustion marker (fluoranthene) in the PAH group. It is likely that these two components are not representative of the PAH group, since more condensed PAHs with a wider range of molecular weights and sources were excluded.
Our sensitivity analysis on the choice of the τ
2 provides further insight in our results. For total CVD admissions, our results were stable and did not vary with τ
2. In contrast, our results for total respiratory hospitalizations were less stable, as evidenced by the second stage effect estimates and their confidence interval widths that differed widely across different values of τ
Our study has several limitations. First, ambient measurements were made at a single monitoring site in each city. All pollutants measured are emitted from local sources and thus expected to be spatially heterogeneous. The resulting exposure error, nonetheless, is likely to bias our results toward the null
. There might still remain some residual confounding by time-varying variables; this is unlikely, however, given our use of a case-crossover design
[33, 34]. Furthermore, many species did not meet our inclusion criteria and thus were not included in the analyses. Additionally, we only included particulate OC species. As a result, chemical groups were incompletely represented, as the chemical groups also include species in the gaseous phase. It is likely, however, that gases and particles exert different toxicities. Also, to increase power, we selected our controls leaving 3 days between control days, resulting in overlapping exposure periods between the cases and controls for the longer moving averages examined, and thus potential bias toward the null
. Our results are limited by the number of compounds measured and included in the analysis and should be, therefore, interpreted in light of these limitations.
To our knowledge, this has been the first study to examine association between primary organic compounds and hospital admissions. Overall, our results suggest that primary OC compounds, such as cyclohexanes, alkanes and iso-/anteiso-alkanes, may play an important role in PM2.5 toxicity and lend some support for previous findings of associations between hospitalizations and mobile sources. Furthermore, our findings add to the current research, provide insight for pollutant groups that have not previously been studied and guidance for the design of future epidemiologic and toxicologic studies.