Measuring Exposure to Environmental Tobacco Smoke in. Studies of Acute Health Effects. Marian C. Marbury,1 S. Katharine Hammond,2 and Nancy J. Haley3.
American Journal of Epidemiology Copyright © 1993 by The Johns Hopkins University School of Hygiene and Pubic Health All nghts reserved
Vol. 137, No 10 Pnnted in U.S.A
Measuring Exposure to Environmental Tobacco Smoke in Studies of Acute Health Effects
Marian C. Marbury,1 S. Katharine Hammond,2 and Nancy J. Haley3
The relations among three methods of measuring exposure to environmental tobacco smoke, questionnaires, urinary cotinine, and a passive monitor for ambient nicotine, were investigated in a study of 48 children in Minnesota in 1989. Subjects were all under 2 years of age and did not attend day care. Passive nicotine monitors were placed in the activity room and the child's bedroom for 1 week, urine samples were collected at the beginning and end of the week for cotinine analysis, and a detailed questionnaire concerning cigarette smoking was administered at the end of the week. These same measures were obtained weekly for 8 weeks for 22 of the children. Among households with smokers, concentrations of ambient nicotine and urinary cotinine were lowest when the father smoked, intermediate when the mother smoked, and highest when both parents smoked. Activity room concentrations were highly correlated with both urinary cotinine {r = 0.81) and the total number of cigarettes smoked in the house (r = 0.86). Regression equations indicated that knowing who smoked in the house was a more important predictor of ambient nicotine than knowing the amount smoked. Both urinary cotinine and ambient nicotine demonstrated variability over time, although ambient nicotine was less variable. In addition, 100% of possible ambient nicotine samples were collected in contrast to 80% of urine samples. The results of the study suggest that both urinary cotinine and ambient nicotine provide better information about the exposure of young children to environmental tobacco smoke than questionnaire data alone, and that ambient nicotine may be the more useful in this population based on its greater stability and ease of collection. Am J Epidemiol 1993;137:1089-97. cotinine; environmental exposure; environmental monitoring; epidemiologic methods; nicotine; questionnaires; smoking; tobacco smoke pollution
A growing body of epidemiologic literature has examined potential adverse health effects attributable to environmental toReceived for publication May 26,1992, and in final form February 16, 1993. 1 Section o) Chronic Disease and Environmental Epidemiology, Minnesota Department of Health, Minneapolis, MN. 2 Department of Family and Community Medicine, University of Massachusetts, Worcester, MA. 3 American Health Foundation, Valhalla, NY. Reprint requests to Dr. Marian C Marbury, Section of Chronic Disease and Environmental Epidemiology, Minnesota Department of Health, 717 S E Delaware St., Minneapolis, MN 55414. This study was supported by National Institute of Environmental Health grant R29 ES04787. The authors thank Dr. David Knebel, Dr. Jonathan Samet, and Dr. Ruth Etzel for their comments on the manuscript and Barbara Ottis for her fieldwork
bacco smoke, including lower respiratory illness in children and lung cancer in spouses of smokers (1, 2). This accumulating evidence has resulted in a recent decision by the Environmental Protection Agency to declare that environmental tobacco smoke is a human carcinogen (3). The majority of exposure estimates in epidemiologic studies have been based on questionnaire data. Environmental tobacco smoke, composed of both exhaled mainstream smoke and side stream smoke, is a complex mixture of over 3,800 chemicals in both particle and vapor phases (1) and thus cannot be measured directly. Several newer approaches that appear suitable for use in epidemiologic studies may provide more precise exposure esti-
1089
1090
Marburyetal.
mates. These approaches include air monitoring of environmental tobacco smoke constituents and biologic indicators of absorption (4). The utility of questionnaires for assessing exposure is inherently limited. The validity of questionnaires depends on the subject's ability to accurately recall and report smoke exposure or the smoking habits of others. Even if valid, questionnaires provide only an estimate of source strength (the amount of pollutant being emitted) and duration of exposure. However, the room concentration of environmental tobacco smoke additionally depends on room size, room air exchange rates, and adsorption of environmental tobacco smoke constituents onto room surfaces. In addition to room concentration, personal exposure depends on the subject's time-activity pattern in relation to smoking, including the subject's proximity to the smoker during the time that the smoking is occurring, the amount of time spent out of the house, and exposure to environmental tobacco smoke outside the home. Air monitoring of environmental tobacco smoke constituents provides a more direct estimate of room concentration or, if the monitor is worn by the subject, of personal exposure. Vapor phase nicotine, which is a specific marker of environmental tobacco smoke exposure (5), can be measured with either active or passive samplers. Active samplers use a pump to pull airborne contaminants through a collection device, while passive samplers rely on molecular diffusion to deliver the contaminant to the collection medium. Passive samplers are less intrusive and easier to use and thus, given equivalent accuracy, are usually preferable for epidemiologic studies. A recently developed passive sampler for vapor phase nicotine has been shown to produce results comparable to those of an active sampler (6). Experience with using the passive sampler for measuring atmospheric nicotine in epidemiologic studies is quite limited. It can be used as either an area or a personal sampler. As an area sampler, it provides a measure of the actual room concentration of nicotine. Like all area samplers, however, it continues to sample
whether or not the study subject is in the room and thus may not provide an accurate estimate of personal exposure. Biologic indicators provide a measure of dose, the amount of the contaminant that is absorbed into the body. While various biologic indicators of environmental tobacco smoke exposure can be measured, cotinine, a nicotine metabolite, has been shown to be the most sensitive and specific marker for separating individuals who are and are not exposed to passive smoke (4). Cotinine also may provide an integrated measure of exposure from all sources, including sources outside the home. However, previous studies have suggested that cotinine exhibits a high degree of both /«ter-individual variability (7), probably due to differences in uptake, distribution, and metabolism, and intraindividual variability (8), when cotinines are repeatedly obtained from the same subject. For these and other reasons, the use of cotinine as a quantitative measure has been questioned (9, 10). In preparation for an investigation of environmental tobacco smoke and lower respiratory illness in children under 2 years of age, we conducted a study to compare methods of characterizing exposure to environmental tobacco smoke, including questionnaires, urinary cotinine measurement, and ambient nicotine measurement. Our purposes were to examine the relations among the three measures and their variability over time and to assess feasibility and logistic issues. Although ventilation rates were not measured, the study was conducted during a period when rates were expected to be stable. MATERIALS AND METHODS
Study subjects were children under 2 years of age who attended Park Nicollet Medical Center for either a respiratory illness or wellchild care and who were not enrolled in day care. Potential subjects were identified from billing forms. Telephone calls were made to their parents to explain the purpose of the study and enroll those who agreed to partic-
Assessing Exposure to Environmental Tobacco Smoke
ipate. To increase the proportion of households with smokers in the study, we also invited the participation of parents of young children who had been enrolled in a previous study. A research assistant visited the home of each child to administer a household characteristics questionnaire to one of the parents (usually the mother), place passive nicotine samplers, and obtain a urine specimen from the child. The homes were revisited a week later to retrieve the samplers, obtain a second urine specimen from the child, and administer a short questionnaire on smoking patterns in the house during the previous week. The parents of a sample of these children were invited to participate in a longitudinal study. During the longitudinal study, passive nicotine monitors were replaced, urine specimens were obtained, and the smoking assessment questionnaire was administered each week for 8 consecutive weeks. Urine specimens were collected with a pediatric urine collection bag. If the specimen could not be obtained during the visit, the mother was instructed to remove the bag and place the specimen in a plastic container, which was then kept in the freezer until the next visit. Cotinine concentrations in urine samples were determined by competitive inhibition radioimmunoassay with a modification of the method originally described by Langone et al. (II). This method has interassay and intraassay variations of 5 percent, with a sensitivity of I ng/ml. To adjust for urine dilution, urine cotinine concentrations were standardized to creatinine concentration and expressed as cotinine: creatinine ratios. Because of data quality problems with the laboratory we initially used (not the American Health Foundation), we do not have data from 20 percent of the samples we collected overall and 17 percent of the samples in the longitudinal study. Passive nicotine samplers were placed according to a standardized protocol in the room designated by the respondent as the major activity room. Samplers were also placed in the child's bedroom in homes with cigarette smokers. Since the nicotine sam-
1091
pler can also be worn as a personal monitor, we asked the mothers of children in the longitudinal study to keep the sampler either on or next to the child for 1 week to assess the feasibility of personal monitoring in this age range. The sampler consists of a sodium bisulfate-treated filter held within a 37-mm polystyrene cassette with a windscreen. Nicotine is eluted from the filters into an aqueous solution of ethanol, the pH being adjusted with NaOH. The nicotine is then concentrated into ammoniated heptane and quantitated by gas chromatography using nitrogen-selective detection, with methods described elsewhere (12). This method has a limit of detection of 0.1 ng/m3. The smoking assessment questionnaire included questions on the number of cigarettes smoked in the house by each household member or visitor, the rooms where smoking took place, and the amount of time the child was exposed to smoking either in a car or otherwise outside the home during the past week. We also asked the parents for an hour-by-hour description of the child's timeactivity pattern on a typical day during the past week, including how many hours were spent in the child's bedroom, the activity room, and the kitchen, and how many hours were spent outside the house. Data analysis
The three exposure variables of primary interest were activity room nicotine concentrations, cotinine levels in the urine obtained at the end of the week, and the total number of cigarettes reported smoked in the house over the previous week. As the data were not normally distributed, we calculated Spearman's correlation coefficients to describe the relations among these three measures. Information on time-activity was used to construct a time-weighted average (TWA) of the form: TWA = ((no. of hours spent in bedroom x bedroom concentration) + (no. of hours spent in activity room or kitchen x activity room concentration) + (no. of hours spent outside the house x zero))/(24 hours). We used linear regression to identify the elements of the questionnaire that were most
1092
Marburyetal.
predictive of nicotine concentrations and urinary cotinine levels. We first took the logarithmic transform of all three measures after adding the value of one to correct for zero values. The dependent variables and the total number of cigarettes smoked in the house were all analyzed as continuous variables. We constructed three indicator variables representing who smoked in the house. The baseline exposure category included children with nonsmoking parents combined with those whose parents limited their smoking to the basement or outdoors. This pooling was based on the results of the bivariate analyses (see below). For predicting urinary cotinine levels, we also defined indicator variables to indicate whether the child had been exposed to smoke outside the home and whether the child had been exposed to smoke while riding in the car. We used information on where cigarettes were smoked in the house only to verify that mothers and fathers smoked in the same rooms in the house and to identify those homes where parents restricted their smoking to the basement or outdoors. These homes are referred to as the "low exposure" group. We used the approach followed by Brunekreef et al. (13) to determine the number of measurements necessary to adequately characterize exposure. This method relies on computation of the variance ratio X, the ratio of the within-subject or error variance to the between-subject or true variance. A measure that has a smaller value of X, which indicates that the between-subject variability is greater than the within-subject variability, is better able to distinguish the exposure of subjects. In addition, X can be used to calculate the number of measurements that are required to reduce the underestimation in the regression coefficients to some level p. The number of measurements is calculated as (/VO - P))^- We used p = 0.9 in our estimates. RESULTS Cross-sectional study
Thirteen girls and 35 boys under 2 years of age, all non-Hispanic white, were enrolled
in the study. In 23 homes neither parent smoked; five homes were in the low exposure group. The homes of the 48 children enrolled in the cross-sectional study were initially visited between January 27 and April 18, 1989. In the activity room of houses with smokers, the 1-week average nicotine concentration was 5.8 Mg/m3, the median was 3.0 ng/ m3, and the range was from 0.1 to 28.6 ng/ m3. The average nicotine concentration in the bedroom was 2.7 Mg/m3, the median was 2.1 Mg/m3, and the range was nondetectable to 7.2 Mg/m3- Cotinine concentrations in the urine specimens collected at the end of the week ranged from nondetectable to 5,095 ng/mg of creatinine, with a mean of 709 ng/ mg of creatinine and a median of 200 ng/ mg of creatinine. All three measures varied significantly, depending on which parent(s) smoked. Concentrations of ambient nicotine and urinary cotinine were higher when the mother smoked than when the father smoked and highest when both parents smoked (figure 1). This same gradient was not observed for the total number of cigarettes smoked in the house; in households where the father smoked, more cigarette smoking was reported than in homes where the mother smoked. Concentrations of all three measures were more than twice as high when both parents smoked because five of seven of these households had visitors who also smoked, in contrast to two of six houses where the mother smoked and none of seven houses where the father smoked. Concentrations of ambient nicotine and urinary cotinine in the homes of the low exposure group were higher than in the homes without smokers, but substantially lower than in homes where smoking was not restricted. Correlations among all four measures of exposure were significant in both the entire population and the smoking population (table 1). The same pattern of correlations was seen in both the total and the smoking populations, although some correlations were marginally lower in the latter group. The two urinary cotinines taken a week apart were also highly correlated with each other (r = 0.89).
Assessing Exposure to Environmental Tobacco Smoke
1093
Reported number of cigarettes smoked In house 227.6 (67.1)
Light Father SMOKER
None
Activity room nicotine concentrations 12.1 (9.0)
Light
Father
Mother
Urinary cotlnlne concentrations
None
Light
Father
Both
16S0 (1825)
Mother
Both
FIGURE 1. Means of three measures of environmental tobacco smoke exposure by household smoker Minnesota, 1989. Numbers in parentheses, standard deviation. Cr, creatinine.
TABLE 1. Spearman's correlation coefficients among three measures of exposure in the cross-sectional study: Minnesota, 19S9
Activity room nicotine (ng/m3) Bedroom nicotine t^g/m3) Urinary cotinine (ng/mg of creatinine)
Bedroom nicotine
Urinary cotinine
Total* cigarettes
0.91 (0.90)*
0.81 (0.86) 0.83 (0.80)
0.86 (0.77) 0.80 (0.77) 0.80 (0.74)
' Numbers r\ parentheses, coefficients among smokers only.
To assess which elements of the questionnaire data were most predictive of activity room nicotine concentrations and urinary cotinine levels, we constructed linear regression models for each. Three models were constructed for activity room nicotine: one with total cigarettes as the only independent variable, one with the three indicator vari-
ables representing who smoked, and one with all four variables. We constructed similar models to predict urinary cotinine levels. In addition, a fourth model included indicator variables for exposure outside the home and in the car. For both activity room nicotine and urinary cotinine, the indicator variables for
1094
Marburyetal. Longitudinal study
who smoked were better predictors than the total number of cigarettes smoked in the house. The model with all four variables described the data only marginally better than the one with the indicator variables alone, indicating that knowing who smokes is more important than knowing the number of cigarettes smoked. Exposure to cigarette smoking outside the home, but not to smoke in a car, was also an important predictor of urinary cotinine levels. Since previous studies have demonstrated that children under 2 years of age spend the majority of their time in their bedrooms (14), we measured bedroom nicotine concentrations. Bedroom concentrations were highly correlated with activity room concentrations throughout the study. To determine whether these data improved our exposure estimates, we combined data on nicotine concentrations with time-activity data to calculate a time-weighted average exposure. This time-weighted average was highly correlated with both activity room and bedroom nicotine concentrations; the magnitude of its correlation with urinary cotinine (r = 0.80) was about the same as that of activity room nicotine and urinary cotinine.
Twenty-seven of the 48 households enrolled in the cross-sectional study were invited to participate in the longitudinal study. Of the 22 households that remained in the longitudinal study for at least 7 weeks, three had no cigarette smokers and two were from the low exposure group. We were unsuccessful in collecting urine samples 20 percent of the time; 100 percent of nicotine samples and questionnaire data were obtained. With the 17 percent loss sustained because of the initial laboratory problem, we had cotinine values for 69 percent of possible samples. At least four cotinine values were available from 15 homes with cigarette smokers. Sixteen of 22 mothers agreed to try using the nicotine sampler as a personal sampler on the child for a week. At the end of the week, 10 reported that they had kept the sampler with the child 25 percent of the time or less. Only three had successfully used it as a personal sampler more than 75 percent of the time. Both activity room nicotine and urinary cotinine concentrations demonstrated considerable variability within households (figures 2 and 3), with urinary cotinine being
• a M O
£
1
20
o o
o u
a
B
10-
•
I•
c a a
