Analyses were conducted across both regions and separately by region given the difference noted above. We also used mixed models to analyze the prediction of personal and indoor endotoxin exposure by the following household characteristics: dog and cat ownership, including the number of dogs or cats and whether dogs or cats were allowed in the house ; number of people living in the home, carpet , whether it was it customary to remove shoes before entering the home, observed cockroaches, observed rodents, flooding damage, surface mold or mildew, livestock, central air conditioning, and region . Personal and family characteristics were also used in the prediction models and included age group , sex, race-ethnicity, mother’s education, and family income. For predictor variables in the indoor endotoxin models, we found insufficient variability across the 12 homes for the more refined categories used in the personal models . Therefore, we dropped carpet cleaning, cockroach and rodent presence, shoe removal, livestock, and air-conditioning. We also dichotomized cat and dog ownership and family income. We began with crude prediction models adjusted for personal temperature, personal relative humidity and study region for personal endotoxin, and study region for indoor endotoxin . We then selected the best multivariate model based on stepwise backward elimination of predictors with the largest p-value over 0.05, and on model fit by AIC. Removed variables were added back singly to the final model to test the appropriateness of the final model.We found detectable endotoxin concentrations in 376 daily personal PM2.5 filters analyzed [median 0.57, range 0.002 – 25.3 EU/m3 ]. All 52 personal field blank filters showed low or non-detectable endotoxin . Within-subject coefficients of variation for personal endotoxin ranged from 69% to 224% . We also successfully extracted and found detectable endotoxin concentrations in all 317 daily Harvard Impactor filters from the stationary site active samplers. As described in Table 1, these included 97 ambient, 109 indoor and 101 outdoor home filters,greenhouse racking and 10 filters from a site in Whittier that served as both an outdoor home and ambient site during one 10-day run, and served as the central ambient site for remaining 10-day runs.
The 42 blank filters at the stationary sites showed low or non-detectable endotoxin . For the comparisons with available indoor and outdoor measurements there were 116 and 113 personal endotoxin measurements, respectively, among the 14 subjects living in those 12 homes. For the comparisons with available ambient endotoxin measurements there were 339 personal endotoxin measurements among the 45 subjects. For the analysis of personal vs. fixed site endotoxin in regression models, one subject for just one day lacked personal temperature and humidity as covariates leaving 338 ambient, 115 indoor, and 112 outdoor observations for analysis. There were all or nearly all 376 personal endotoxin measurements for the comparisons with available ambient air pollution. Ambient air pollutant measurements were nearly complete with at least 407 days for each variable available for comparison with the 423 days of ambient endotoxin measurements. Descriptive statistics regarding all of the exposures by region are shown in Table 2. Arithmetic mean and median personal endotoxin exposures were higher in Riverside than in Whittier. Consistent with this, outdoor home and ambient endotoxin were higher in Riverside than in Whittier. However, indoor endotoxin exposures were higher in Whittier than in Riverside. Although arithmetic mean personal endotoxin was higher than indoor, outdoor or ambient levels across both regions, the median personal endotoxin was only higher in Riverside. This is a reflection of the typical skewed distribution of endotoxin exposures. Indoor to outdoor endotoxin ratios of medians were clearly opposite between the two sites with a ratio < 1.0 at Riverside and a ratio > 1 at Whittier .Actual indoor concentrations reflected this difference with a much lower indoor concentration in Riverside than in Whittier. We show correlation matrixes separately for Riverside and Whittier relating personal endotoxin and stationary site endotoxin to personal and stationary site endotoxin and air pollutants . We found personal endotoxin in both Riverside and Whittier was not significantly correlated with indoor endotoxin or any of the indoor air pollutants. Personal endotoxin was not significantly correlated with outdoor home endotoxin in either Riverside or Whittier. We observed small positive correlations between personal and ambient endotoxin in Riverside but not Whittier. Outdoor home and ambient endotoxin measurements were strongly correlated. In both Riverside and Whittier, personal endotoxin showed a small inverse correlations with personal PM2.5, and small positive correlations with personal PM2.5 EC and OC, which were larger in Whittier.
Personal endotoxin positively correlated with personal temperature in Riverside but negatively correlated with personal temperature in Whittier.Personal endotoxin in both Riverside and Whittier were not significantly correlated with any of the indoor air pollutants. Indoor endotoxin in Riverside, on the other hand, was strongly positively correlated with indoor PM2.5 EC and moderately correlated with indoor PM2.5 mass and OC, whereas in Whittier these correlations were positive but much smaller. Both personal and outdoor home endotoxin in Riverside were not significantly correlated with any outdoor home air pollutant measurement. We observed a small inverse correlation between personal endotoxin and outdoor home PM2.5 in Whittier. Outdoor home endotoxin showed small positive correlations with outdoor home PM2.5, EC and OC in Whittier. In Whittier, ambient temperature and O3 were negatively correlated with personal endotoxin. In Whittier, but not Riverside, ambient endotoxin showed small positive correlations with ambient traffic-related air pollutants and temperature and small inverse correlations with relative humidity.The prediction of personal endotoxin in mixed regression models by the various stationary site measurements of endotoxin are shown in Table 4 including both sites together and separately by region. Ambient endotoxin for the 14 subjects in monitored homes, and their exposure to indoor and outdoor home endotoxin were not significant predictors of personal endotoxin. However, ambient endotoxin for all 45 subjects was a significant positive predictor of personal endotoxin. The regional models show that the overall association was attributable to measurements at both sites, although the regression coefficient for Riverside was twice as large as Whittier. However, the regression coefficient for Whittier was more significant than Riverside . Figures 1-2 show scatter plots and results of linear regression models for the relation between log transformed indoor and outdoor home endotoxin across the 10-day monitoring sessions in 4 homes in Riverside and 8 homes in Whittier. In both regions, the relation was positive, with outdoor endotoxin explaining 25-28% of the variability in indoor endotoxin. The analysis of the relation between personal endotoxin and household or subject characteristics shows a clear positive association with dog ownership in crude models adjusted for personal temperature, personal relative humidity and region .
For each dog owned, personal endotoxin exposure approximately doubles. Interestingly, compared with having no dogs, the strongest and only significant association with personal endotoxin in crude models was for dogs that were only occasionally indoors. This contrasts the finding for cats since the only significant association was for having cats that were often indoors compared with having no cats. Other variables were significantly positively associated with personal endotoxin in the crude models,equipment for growing weed and they included reports of flooding damage and sex . Nominal associations included increasing personal endotoxin by the number of household residents and lower personal endotoxin among Hispanics. The final selected multivariate model included only cat and dog numbers adjusted for personal temperature, personal relative humidity and region. We found a relative increase in endotoxin for each dog of 1.76 , and for each cat of 1.39 . Residence of dogs and cats were not included due to expected dependent relations with the number of animals Chi-Square p-value < 0.0001. Adding back single excluded variables to this final model did not improve the fit of the model and showed that each of those variables were non-significant including flooding damage and male sex . Flood and cat were positively associated with each other . As a result, cat number confounded the association with flood and the association with cats also decreased by 37.5% as well . The analysis of the relation between indoor endotoxin and household or subject characteristics shows that unlike the personal exposure models, dog and cat ownership was not associated with indoor endotoxin . Only three variables were significant in the crude models,reports of flood damage, which was unexpectedly associated with lower endotoxin, Hispanic subjects associated with higher endotoxin , and high school or lower education level in mothers that was associated with lower endotoxin. The final selected multivariate model included only flooding damage and lower education levels in mothers.Our results suggest that fixed site measurements of endotoxin in the home environment may not adequately represent daily personal exposures. The finding of a positive association between ambient and personal endotoxin exposure is not particularly relevant to research used to investigate relations of respiratory health to endotoxin , but it does have some relevance regarding potential impacts of regional sources on personal exposure. It is possible that the limited sample size was insufficient to detect an association of personal with home endotoxin. Evidence in support of that view is that when we limited the analysis of prediction of personal endotoxin by ambient endotoxin to the monitored homes , associations were non-significant but point estimates were similar to those for the 45 subjects . Nevertheless, although we had a limited sample size in the 14 subjects, the findings for the relation of personal endotoxin exposure with indoor home endotoxin exposure , suggest that other micro-environments and personal activities are important to assess.
Given that our analysis was based on daily exposures using measurements all conducted with active 24-hour samplers, our conclusion that any one fixed site measurement may not adequately represent personal exposure applies to short-term exposures that may be involved in the acute exacerbation of asthma. We assessed the potential importance of other locations and physical activity by using previously reported data on quarter-hourly time-activity reports from an electronic diary that each subject filled out throughout follow-up. We found that on average, around 73% of time was spent at home indoor, 1.7% at home outdoor, 12.6% at school indoor, 1.8% at school outdoor, 4% in-transit, 4.3% indoor elsewhere, and 2.6% outdoor elsewhere. Out of an estimated average of 40 min per day of diary reported moderate to strenuous activity , 82% occurred while away from home. Such higher levels of activity may be important in promoting personal endotoxin exposure as a result of the so-called “personal dust cloud.” This is a phenomenon where localized personal activities lead to increased PM exposure by re-suspension of settled PM, which brings the breathing zone of subjects into closer contact with PM from various sources. The highly skewed distribution of personal endotoxin we observed may be partly due to the generation of personal clouds that results from subject activity, including activity around sources of resuspended dust. Our findings of a general lack of correlation between personal and home micro-environmental endotoxin are consistent with the findings of Rabinovitch et al.. In a panel of school children with asthma, they found geometric mean personal endotoxin was higher than indoor or outdoor school endotoxin levels, and personal endotoxin was not correlated with these stationary site measurements.The present results show a positive association between personal endotoxin and the number of dogs and cats owned, as expected, and this substantiates the utility of the personal exposure measurements. This finding is consistent with a sub-study of 10 children by Rabinovitch et al. who found personal endotoxin exposures were significantly higher in 3 households with dogs and one with cats compared with 6 households with no furry pets. We found the association of personal endotoxin was strongest among subjects with dogs that were only occasionally indoors. This could be attributed to entrainment of debris from the outdoor environment into the indoor environment, including fecal matter. However, we found no association between indoor endotoxin and dog or cat ownership. This may be due to either the smaller sample size or that personal exposure is more dynamic as would be expected from the generation of personal dust clouds.