Figure1: The elevated odds ratios (above unity) in this figure suggest that exposures to VOCs from water-based points and solvents are associated with a variety of sick building symptions (* indicates the OR is significant at p < 0.05; ** for p < 0.01).
The indoors is often regarded as safe haven from problems associated with outdoor air pollution, but a growing number of reports have suggested that exposures in indoor environments may lead to health problems. One area in which evidence has been accumulating is the relationship between working in office buildings (as opposed to industrial exposure conditions) and a variety of health effects, such as eye, nose, and throat irritation and dry, itchy skin. Such symptoms are often considered part of "sick building syndrome" (SBS), which is a complex of symptoms that may arise while inside a building (and which may improve or disappear upon leaving the building). High prevalences of these symptoms have been reported from various epidemiological investigations of SBS, but the exact nature of the chemical and physical characteristics of the indoor environments is not known, nor is the cause-and-effect relationship between the exposures and these health effects understood.
Among the suspected causes are volatile organic compounds (VOCs). Indoor air typically contains between 30 and 100 VOC species at concentrations that are readily measurable. Exposures to individual VOCs— typically at high concentrations— have been linked with sensory, pulmonary, and neurologic responses. However, in office buildings, occupants are often exposed to mixtures of individual VOCs, each at low concentration. Some researchers have hypothesized that high total concentrations of VOCs are a causative factor of SBS symptoms, and thus total VOC concentration (TVOC) can be used as an exposure metric that is related to symptoms. However, few significant associations have been found between TVOC concentrations and SBS symptoms in studies in buildings.
Indoor Environment Program researchers* have recently developed new VOC exposure metrics that account for the differences in response to the various VOCs and in the VOC sources. The response is based on the "potency" of individual VOCs to elicit symptoms, estimated from available irritancy measurements. Because the human irritancy of so few compounds has been tested, we based our potency estimates on available animal bioassays or, when no data were available, on properties of similar compounds. The most irritating VOCs were then selected and used in a principal component analysis, which yielded "clusters" of VOCs with concentrations that increase or decrease together because the compounds within a cluster have a common dominant source.
The California Healthy Buildings Study (CHBS), an investigation recently completed by LBNL scientists in 12 office buildings in or near San Francisco, provided data on indoor VOC concentrations (see article in CBS News, Spring 1994). Based on these data, we identified four clusters of VOCs associated with specific sources: motor vehicle emissions, building materials, carpet/building materials, and water-based paints and solvents, using principal component analysis. These clusters represent groups of both measured VOCs and those with the same sources that have not been measured because of limitations in technology. Both the measured and unmeasured VOCs in each cluster may cause observed symptoms. Principal component analysis iteratively develops coefficients for each VOC that reflect the association (positive or negative) of that VOC with the source. The principal components representing each cluster are composed of the sum of the original VOC concentrations multiplied by the individual VOC principal component coefficient; these sums range from roughly -3 to 3.
The new VOC exposure metrics were applied to the CHBS data set. We used SBS symptom data reported on a written questionnaire by 517 office workers who were in close proximity to 22 sites where 39 individual VOCs were measured. As a result, we developed a set of data associating the VOC clusters with SBS symptoms that could then be examined to determine which associations were significant statistically.
The method used in this examination was based on an odds ratio (OR), which represents the odds of experiencing a symptom (versus the odds of not experiencing a symptom) given increased VOC exposure. The principal components representative of VOC sources are used as independent variables in the prediction of SBS symptom outcomes in a logistic regression analysis. As with linear regression analysis, this technique describes the relationship between an outcome and a set of independent or predictor variables. However, the results of a logistic regression are binary. That is, they predict the presence or absence of a symptom. An odds ratio is calculated for each independent variable based on the regression analysis. An OR of 1.0 indicates that the odds of experiencing a symptom do not vary with VOC exposure level; conversely, an OR greater than 1.0 indicates that VOC exposure does affect symptom outcome. Those ORs with confidence intervals (error bars) that exclude 1.0 are statistically significant.
As seen in Figure 1, after adjusting for potentially confounding influences, the source identified (using principal component analysis) as water-based paints and solvents was associated with several SBS symptoms, including dermal (OR=2.2, 95% Confidence Interval 1.3-3.7) and eye (OR=1.7, 95% CI 1.1- 2.7) irritation. The ORs represent the increased odds of experiencing the symptoms, given a one-unit change in the principal component (where the range is ± 3). The more typical TVOC exposure metric used in prior analyses was not useful in symptom prediction in the adjusted model. Also not useful were metrics that took into account potency but did not rely on principal component analysis to identify clusters of VOCs.
As demonstrated by this work, sophisticated methods are needed to understand the associations between health effects and exposure to complex mixtures. These new VOC exposure metrics appear to have significant potential for helping identify causes of and potential solutions for SBS.
—JoAnn Ten Brinke
*J. Ten Brinke, J.M. Daisey, A.T. Hodgson, W.J. Fisk, Indoor Environment Program, Lawrence Berkeley National Laboratory; S. Selvin, Department of Biostatistics, School of Public Health, University of California, Berkeley; M.J. Mendell, Industrywide Studies Branch, National Institute for Occupational Safety and Health, Cincinnati, Ohio; C.P. Koshland, Department of Environmental Health.
Joan Daisey, Head
Indoor Environment Program
(510) 486-7491; (510) 486-6658 fax
This work is supported by the Office of Building Technologies and the Office of Health and Environmental Research, DOE.
EETD Newsletter Home Page
CBS Newsletter Home Page
Table of Contents for this Issue