Sarsfield intensive hog farm
The town of Sarsfield consists of approximately 275 homes, and 620 residents and is in the eastern part of Ottawa, Ontario. In 2004, and intensive hog farm was established just outside this town. The building application for the farm allowed for approximately 1,000 sows to be housed on the property. However, the size of the farm may increase as plans call for as many as 3,000 sows to be housed on the 667 acre site. Concerns about the possible environmental and human health impact of this farm expressed by residents resulted in decision of the City of Ottawa to fund this self-reported health survey.
Overview of study design
The survey was designed to characterize the prevalence of several health conditions in the area of the farm shortly after farming operations had begun. With the planned expansion of the farm, it was hoped that this survey could provide valuable baseline data. In this cross-sectional study, we selected seven communities located in the eastern part of the city of Ottawa and in the general vicinity of intensive swine farm operations. To assess the HRQOL and the prevalence of health conditions in children and adults, a questionnaire was delivered to a sample of homes in these communities. The survey was presented as a general health survey, not as an assessment of the health effects of the IHF. This was done to minimize possible reporting biases by those concerned about the farm. The study was sponsored by Ottawa Public Health, and was approved by the City's ethics review board.
We compiled a complete listing of eligible private households by using the city of Ottawa's property listings file. This file was regarded, for all intents and purposes, as being complete based on comparisons of dwelling counts for the study area using these listings to that of the 2001 Canadian census. The target population consisted of all residents of the communities that are located within twenty kilometres of the IHF, namely: Bearbrook, Cumberland, French Hill, Leonard, Navan, Sarsfield, and Vars. The total estimated population for these communities was approximately 8,100. Given the proximity of the IHF to the town of Sarsfield, all residents in this town were provided an opportunity to participate. Sampling ratios for the other communities were selected so as to compare prevalence of selected health conditions between communities while adhering to time and cost constraints. For communities other than Sarsfield, households were randomly chosen thus ensuring that each household within these areas had an equal probability of being sampled.
A questionnaire was developed to collect data on the general health of residents, and also incorporated questions on health conditions for which previous research had identified associations with hog farming operations. Pilot testing of the questionnaire was done on a sample of 25 individuals to ensure that it could be easily understood, and could be completed within 30 minutes. Questionnaires were made available in both English and French. Sociodemographic characteristics for adult participants were collected including age, sex, household income, educational attainment and occupation. The age and sex of each child resident in the home (if any) were collected from one participating adult member of the household.
Adult participants were also asked whether they had any of a number of health conditions that persisted for at least 6 months, and had been diagnosed by a health professional. This list consisted of several respiratory conditions including asthma, wheeze, chronic bronchitis, wheeze, sinusitis and rhinitis. Participants were also asked about whether they had been diagnosed with cardiovascular conditions, gastrointestinal disorders, or mental health conditions such as depression, anxiety, and migraines. Questionnaire data collected enabled us to classify individuals according to their cigarette smoking habits, and their body mass index. One adult household member provided information on behalf of the children about their general health status and the prevalence of respiratory conditions.
Health related quality of life (HRQOL) was assessed with the widely-used SF-36 survey instrument[14
]. This survey instrument measures a person's ability to function while at work, at home, and in social situations. A standardized score with a range between 0 and 100 is calculated across 8 health domains, with a higher value reflecting better HRQOL.
To evaluate the extent of possible awareness bias, adult participants were asked to indicate from a pre-specified list of 18 items what had been sources of worry for them in the past year. IHF related worries included outdoor smells, air pollution, or water pollution. Finally, the survey included an open-ended question at the end whereby participants could provide any additional comments.
Administration of survey
In October 2005, questionnaire packages were assembled and delivered to the selected households by a survey team of two individuals. The survey was presented as a general health questionnaire. The survey team visited each eligible home up to three times to establish contact. The visits were coordinated to ensure that the interview teams visited the homes at different times of the day, and if contact was not established during the initial visit, a weekend visit was made. On the third visit, if contact was not made, the questionnaire package was left in the mailbox of the home.
Both English and French language questionnaires were made available. Each adult member of the household was asked to complete a questionnaire, while one adult was asked to supply information for all children that resided in the home. Participants were asked to complete the questionnaire at their leisure, and return it with a postage paid envelope that was provided. We accepted all questionnaires returned by December 31, 2005.
In May 2006, a second SF-36 questionnaire was mailed to households that indicated they would be willing to participate in subsequent surveys. This allowed us to evaluate changes in health related quality of life that might exist due to seasonal variations. No analyses of the second survey are presented in this paper, largely because no appreciable differences in SF-36 scores were observed using a matched design approach.
Responses from the questionnaires were entered into a database using a double data entry method. Where discrepancies were found, manual resolution was used to ensure the proper value would be used in the analysis. The number of questionnaires for which manual resolution was undertaken was relatively small (approximately 30 questionnaires).
Questionnaires were delivered to a total of 829 households, and of these, 37 refused to accept the materials. At least one questionnaire was completed and returned by mail by 419 (50.5%) households. Data were collected for 741 adults and 285 children. From these questionnaires, we excluded those (n = 7) where the residential address could not be ascertained and a further 11 for whom their age could not be ascertained. When compared to Canadian census data for the study region, survey participants were more likely to be female, and over the age of 45, and have French as their mother tongue. Rates of employment were similar between census and survey data for the area.
Descriptive statistics for each questionnaire item were generated to identify outliers or data anomalies. For each participant with available address, geographical information system (GIS) methods were used to calculate the distance between their homes and the IHF. We created a categorical variable to classify participants according to their residential distance to the hog farm. For adults, the distance categories were < 3, 3 - < 9, and ≥ 9 km. We choose these cutpoints for three important reasons. First, the selection of the cutpoints of <3, 3-<9 and 9> km grouped the number of participants into nearly equal groups (231, 236, and 256) thereby optimizing statistical power to compare disease prevalence rates across three distance groupings. Second, the study was initiated, in part, from concerns about the potential health concerns among residents of Sarsfield. The selection of a 3 km buffer essentially placed all residents of Sarsfield in the smallest distance category, while grouping participants from the other villages in the upper distance groupings. In particular, the mean distance between the residence and the IHF among Sarsfield participants was 2.1 km (Range 0.5 -3.1). As there were far fewer children than adults in the study, and given the prevalence of the health conditions under study, and the need to adjust for other covariates we opted to use two rather than three distance categories. The third reason for selecting a range of within 3 km was that recent research has demonstrated a high prevalence of reporting malodour among those who lived within 1.5 miles (2.4 km) from industrial swine operations [8
]. In our study, we had data on a smaller number of children than adults, and therefore, lacked the statistical power to model these same three distance categories. Instead, we used the two groupings of <3, and ≥ 3 km.
Multivariable logistic regression was used to evaluate associations between residential distance to the hog farm, and the self-reported prevalence of various health conditions. Odds ratios generated from these regression models were adjusted for the possible confounding influence of other risk factors including: age, sex, cigarette smoking, and body mass index (BMI). BMI was treated as a potential confounding factor as several studies have found associations between it and respiratory symptoms, as well as mood disorders including depression. However, it made little difference to our measures of association and ultimately was dropped in the final models. We also examined the influence of socioeconomic status using household income on the risk estimates as previous work has shown that the validity of self-reported measures of health may vary by SES, and SES is associated with other determinants of health including diet, hygiene, and smoking. The standard errors of the odds ratios were adjusted for correlations arising from multiple questionnaires returned from the same household through the use of the method of generalized estimating equations. Statistical significance was assessed through inspection of the 95% confidence intervals of the odds ratios. We applied similar methods to compare the prevalence of childhood respiratory conditions.
Analysis of variance methods were used to compare SF-36 scores according to proximity to the IHF. The method of least squares was used to produce estimates for each of the SF-36 domains within each distance category adjusted for age, sex and income. To evaluate the possible role of awareness bias, stratified analysis was performed by grouping subjects according to whether they experienced any IHF related worries, or not.