Study design and population
We conducted a case–control study of children living in and around the rural community of Humboldt, Saskatchewan, Canada during the fall and winter seasons from 2005 to 2007 with some catch-up in the spring. All rural schools in the same school district as schools within the community boundary of Humboldt were approached. The total population of the town of Humboldt and its surrounding rural municipality was approximately 6,100 people. Of the towns surrounding Humboldt that had schools included in the study, the population ranged between approximately 200 to 800 people. The major source of industry for the region was classified as agriculture and resources. Subjects were recruited from a previously conducted population-based cross-sectional survey of respiratory health of 6 to 18 year old children which had been completed in 2004. Potential cases comprised all subjects reporting doctor-diagnosed asthma or wheeze in the past 12 months. Two potential controls for each case were randomly selected from among children who were not considered cases. Once selected, an invitation letter was mailed and potential participants were contacted by telephone a maximum of seven times. Following contact, cases and controls status was confirmed by a screening questionnaire that enquired about asthma diagnoses, recent asthma or wheeze events including symptoms, episodes and breathing medication use.
Data collection for this study included an interviewer-administered questionnaire, dust collection from the home for endotoxin, and saliva collection to assess recent tobacco smoke exposure. The Health Research Ethics Board – Panel A (University of Alberta) and the Biomedical Research Ethics Board (University of Saskatchewan) approved the study as did the local school boards. Prior to taking part, parents and children completed consent and assent forms, respectively.
The questionnaire, based primarily on previously validated and standardized questionnaires [14–16], and those used in previous respiratory health studies in Saskatchewan subjects [10, 17, 18] was administered by a trained interviewer to a parent of the subject. Information was collected on respiratory health, socio-demographic factors, general health, family history, birth characteristics, lifestyle, housing characteristics and environmental exposures. Subjects were classified into age groups (≤12 years vs. >12 years) to be comparable with other childhood asthma studies where age ranges were typically between 6–12 years [18, 19], and to be able to look at effect modification by age. Season of testing was defined by the date of the home visit and was recorded as spring (March, April, May), fall (September, October, November) and winter (December, January, February) to account for the potential differences in endotoxin and allergy levels by season. Personal history of an allergic condition was determined by presence questionnaire report of hayfever, eczema and respiratory allergies. Respiratory allergies were based on the question “Has this child ever had an allergy (hives, swelling, and/or wheezing) to any of the following: house dust/grain dust/pollen/trees/grasses/mould or mildew/dog/cat/birds or feathers/farm animals?”
Collection and analysis of dust samples to quantify endotoxin exposure
Samples of household settled dust were collected from the floor of the room where the child spent most of his/her free time (play area); and from the child’s mattress (mattress). Dust was collected following the International Study of Asthma and Allergies in Childhood (ISAAC) protocol . A sock filter made of Connaught satin with a pore size of approximately 5 to 10 μm  was used to collect dust. Floors that were mostly carpet had 2 m2 vacuumed for 4 minutes. Completely smooth floor (eg. hardwood, laminate, or linoleum) had 4 m2 vacuumed for 4 minutes. Dust collection from the bed was completed after all duvets, blankets and sheets that the child slept under were removed as per ISAAC protocol . The length and width of the bed was measured and the whole area of the bed was vacuumed for 2 minutes. Dust was collected by using one of two vacuum cleaners (both were Solaris vacuums made by Miele – S514). Both vacuum cleaners were equipped with a HEPA filter, and temperature and humidity were measured at the time of vacuuming.
According to the ISAAC protocol, vacuums must be capable of at least 800 W of power . For this study, the two vacuums each had a power rating of 900 W. The flow rate vs. static pressure relationships for the vacuums was experimentally tested at the College of Engineering at the University of Saskatchewan. A typical second order polynomial relationship between flow rate and static pressure (R2 for all tests exceeded 0.99) was observed in all cases. The performance of the vacuums evaluated post study was found to be comparable to pre-testing evaluation.
Assays to determine endotoxin levels were completed following the protocol by Gereda et al.  Laboratory technicians were blinded to the case–control status of the subject and to all other data collected. Quantity of endotoxin was measured using the kinetic chromogenic Limulus assay (Cambrex Bio Science, Kinetic QCL, Walkersville, MD). For 49 samples, endotoxin analyses were conducted with the following dilutions: neat, 10, 100 and 10 with a spiked sample. These were all completed in duplicate. The spike recovery was 113%. For the remainder of the samples, analyses were completed in duplicate only using the 10 and 100 dilutions. The coefficient of variation between samples was 9.32%. Endotoxin measures were log-transformed prior to statistical analysis.
Endotoxin levels were expressed as concentration [endotoxin units/mg of dust collected (EU/mg)] and load [endotoxin units/m2 of area vacuumed (EU/m2)]. Both were reported as there is inconsistency in the literature about which expression is most appropriate to report . While concentration is most often presented, it has been suggested that load may more accurately describe the burden of exposure .
Collection and analysis of saliva samples to quantify cotinine levels
Tobacco smoke exposure was determined by salivary cotinine levels. Subjects were asked to spit into a specimen container without the use of gum, Teflon, or other materials that would stimulate the flow of saliva. Up to 5 ml of saliva were collected. Analysis for cotinine was conducted using saliva cotinine microplate enzyme immunoassay kits (Cozart plc, United Kingdom).
Analysis was completed using STATA version 9.0 (College Station, TX: StataCorp LP). Initially we compared personal and environmental characteristics between cases and controls based on frequencies and proportions. Following this, we assessed the correlation between play area and mattress endotoxin levels using the correlation coefficient with statistical significance adjusted for within family correlation. We also identified the predictors of play area and mattress endotoxin levels using linear regression after log-transforming the endotoxin measures, using generalized estimating equations (GEE) to account for within family correlation, and adjusting for potential confounders. Finally, we assessed the association between the measures of endotoxin and case-status.
In this final set of analyses, the outcome for the multiple logistic regression model was case–control status. A multiple logistic regression model was fitted for each measure of endotoxin that was independent of the other models. Model 1 assessed play area endotoxin concentration; Model 2 assessed play area endotoxin load; Model 3 assessed mattress endotoxin concentration; and Model 4 assessed mattress endotoxin load. Additional variables were included based on statistical significance, clinical importance, and the effect the removal of that variable had on the beta coefficients of other variables in the model. The additional variables included in the Models were: age group, sex, daycare attendance, presence of home air filter, smoking during pregnancy, smooth floors in the bedroom in the first year of life, season of testing, and tobacco smoke exposure. Levels of cotinine (ng/ml) were eventually categorized as high and low post analysis based on the median cotinine level (1.24 ng/ml) due to a highly skewed distribution of results that could not be normalized after log transformation. Other potential confounders, such as parents’ education status, consumption of unpasteurized milk, type of farm exposure (none, grain, livestock), and presence of mold or dampness in the home were tested but not included in the final models as they did not influence the associations of the other variables in the model, and they did not have independent associations with case–control status. Interaction terms of clinical importance were also considered. These included potential interactions between endotoxin and sex, age group, and tobacco smoke exposure. All assessment of interactions was determined a priori. Throughout the analyses, GEE with an exchangeable working correlation were used to account for clustering within families. Occasionally, the model would not converge and an independent working correlation was used. Because of the importance of a personal history of allergy, interaction between endotoxin exposure and the personal history of allergy was assessed and the analysis was repeated after stratification by allergic history.