Document n934eNJBOMrqxeMrRgzOGg7V6

The German Environmental Survey 1990/92 (GerES 11): Sources of personal exposure to volatile organic compounds KURT HOFFMA", CHRISTIAN KRAUSE, BERND SEJFERT and DETLEF ULLRICH Fedend Environnrental Agmq, Instiwe& Watler; Soil and Air Hygiene,RO.Box 330022,I4 191 Berlin, Germany Introduction The number of studies on the occurrenceof variousvolatile organic compounds (VOCs) in indoor air has increased in recent years (Miksch et al., 1982; Mslhave, 1982; Ginnan et al., 1986; Lebret et al., 1986; Seifert et al., 1989; muse et al., 1991; Brown et al., 1993; C m p and Madany, 1993; Fellin and Otson, 1993; Schreiber et al., 1993; Heavner et al., 1995, 1996). Although people tend to spend most of their time indoors, indoor VOC concentrations do not necessarily reflect personal exposure adequately. Even modelling personal exposure using indoor monitoring and time-activity data is not without problems. At the present stage, the most reliable information on personal exposure will be obtained by personal monitoring in which the pollutants of interest are measured close to the breathing zone of a person. In previous studies dealing with personal -monitoring (Wallace et al., 1988, 1991; Proctor et al., 1991) only 24 h exposures were analysed. Wallace et al. (1994) stated that such short-term measurements are not I . Abbmiatiam: GcrES. German EnvirnUnental Sumy; VOC. volatile organic compouads; TVOC, total volatile organic compornb; ETS. envhmmental tobrrap smoke; CI,canfideace intaval, GM,pcomccric man; LOQ, limit of ~ t i f i c a t i o n . 2. Address a11 collgponda~t~o:~Dr. Band Seifat, F e d d Eavirop mental Agency, Institute for Water, Soil and Air Hygiarc, P.O. Box 330022, 14 191 Berlin, Germcay. Tel.: +49-30-89O3- 1320. Fax: +49- 30-8903-1830. E-mail: baad.sCifnt@Ubsde Received I5 April 1999, rtccptsd 8 Febnrary 2000. suitable to estimate long-term distributions of exposure to mostof the analysed VOCs. The samplingperiodshould be at least 1 week includimg the weekend, because exposure may partly be due to activities that people do only occasionaliy. in fact, longer-term exposure measurements. are impomnt to generate as accurate as possible annualid (or even lifetime) exposure distributions. Long-term sampling has not only been used in GerES I (Krause et al., 1987) but also in the NHEXAS study (PelliPari et al., 1995). In the 6amework of the second German Environmental Survey (GerES IIa) conducted in the Western part of Germany (old states before reunification) a study was conducted to chantcterise population exposure to about 70 VOCs (Hoffmann et al., 1996). Here we report and discuss the results obtained for benzene and a number of its homologues. Metbods Selection of the Srudy Participants In GerES IIa, 2524 adults aged 25 to 69 years were randomly selected using a two-stage procedure stratified according to community size, gender and age (Seifertet al., 2000). From this population sample a subsample of 113 subjectswas drawn at random to study personal exposureto VOCs. The participating subjects were distributed over 36 sample points to cover different areas of the Westem part of . (D Hofianndal. Table 1. Comparison between study population and total population. catceory sur fhnale male Study population (%) Total population (%) (N=113) (N=35.041.000)* 48.7 51.0 51.3 49.0 Age (yean) 25-29 30-39 40-49 50-59 60-69 15.9 24.0 22.1 19.5 17.7 13.6 23.2 20.0 23.5 18.9 West German resident population, aged 25-69 years, on the basis of the micro-census 1991. Germany. The distribution by gender and age in the subsample was similar to that in the population (Table 1). The number of persons engaging in exposure related activities like smoking and refuelling can be found in Tables 9, 11 and 13. Personal Sampling i O W - 3 5 0 0 diffusive samplers (3M, St. Paul, MN,USA) were used as personal samplers to collect VOCs. A member of the field staff instructed the subjects how to use the personal sampler. The subjects then wore the monitor close to the breathing zone over seven days placing it at the bedside when sleeping. At the end of the sampling period, the sampler's collecting surface was covered with the lid provided by the manufacturer for this purpose. The sampler was then wrapped tightly in an aluminium foil and sent to the laboratory for analysis. Subjects were asked to complete a special questionnaire and to note the time they stayed in the living room, bedroom, kitchen, bathroom, and in other rooms out of home as well as in traffic and outdoors. The three rooms they stayed in most of the time had to be described more precisely with regard to floor space, height, ventilation habits, type of heating system, presence of tobacco smoke, type of windows as well as type and age of wallpapers, carpets and furniture. Moreover, activities like renovation and painting during the 7-day period had to be reported. Besides these information, each subject completed the basic questionnaire of GerES 1Ia which included detailed questions about the household, characteristics of the residential quarter, occupation, smoking habits and other lifestyle characteristics. Analysis and Quality Control The diffusive samplers were analysed by a gas chromatographic procedure according to German VDI guideline 3482/4 (VDI, 1984). This procedure includes sampling on activated carbon, desorption with CS2 and separation using a nonpolar capillary gas chromatographic column. To confirm the identification of compounds which was done via the retention indices and retention index differences, another aliquot of the sample was analysed using a second gas chromatograph equipped with a column of slightly different polarity. The total procedure used has been described in more detail by Seifert et al. (1989). Seventy-four VOCs, including n- and iso-alkanes, aromatic hydrocarbons, halocarbons, terpenes, esters, ketones and alcohols were analysed. Each VOC was calibrated individually using standard compounds. The peak area obtained from the instrument's computer system was checked manually for correct baseline setting. The baseline was corrected in case of need. Quantification was carried out by means of an internal standard method (ISM) using cyclooctane and 1,2,3-trichloropropane as internal standads. The ISM included correction for potential dilution errors, varying solvent desorption efficiency and errors in gas chromatographic injection volumes. A field blank was taken with every 10th sample in order to test for correct sample handling, transportation and storageconditions. Furthermore, laboratory blanks were run continuously. In one case interfering compounds were found in the field blank, and their concentrations were markedly higher than those of the laboratory blanks. Since this blank sample exhibited a specific VOC pattern which was also found in the accompanying real sample, the latter was excluded from further treatment. Consequently, all results are based on 113 data sets. In the evaluation of the results of a European interlaboratory experiment on the performance of diffusive samplers (De Bortoli et al., 1986) in which eight laboratories participated, the mean relative interlaboratory standad deviation has been calculated for a variety of compoundsrepresentingdifferent chemical classes, namely - -1-butanol, 1,1,2 trichloroethane, 1 octene, butylacetate, - -3 heptanone, o-xylene, alpha pinene, and n-decane. The mean relative interlaboratory standard deviation was found to be about 13% for concentration levels between 20 and 1000pg/m3and a samplingperiod of four days. In addition, it was found that for nonpolar compounds (0-xylene and ndecane) the passive samplingresults were generally in close agreement with those of active sampling whereas for the six polar compounds named above the results of active sampling using thermal desorption after collectionon Tenax exceeded those of passive sampling by 13% to 56%. In the interlaboratory comparison mean desorption efficiencies of 78% to 105%were obtained for loadings of 10 to 100 pg/ sampler. In our own laboratory,desorption efficiencies were between 93% and 105%. Only in the case of 1-butanol was the desorption efficiency much lower and more variable, namely between about 30% and 65%. This variability led us 116 Journal of Ana&& and Environmenfd Efidemiolqv (2000) 1q2) to abstain fiom correcting the alcohol concentration levels. It was felt, however, that even the uncorrected data, although underestimating the real concentrations, should be included in Table 5 because they represent useful information. To check the validity of the passive samplers results, a test series was conducted in a test chamber using test gases -at low concentration levels. The model compounds, 1,1,1- trichloroethane, m-xylene, n heptane, benzene, toluene, and n-hexane were studied at concentrations below 25 pg/ m'(U1lrich and Nagel, 1996). In these tests passive sampling was carried out in parallel to active sampling which was used as a reference. Passive samplers were exposed to these concentrations between 3 days and 3 weeks. If a mean uptake rate of 27.9 cm3/minwas used to calculate the results for the passive samplers, the overall mean deviation between active and passive sampling was -8.5% with a range between +2% and - 19%. For high- boiling compounds (n-decane to n-dodecane) it has to be anticipated that the deviation is larger and may attain about 40%. In another study which has been completed since (Ullrich et al., 1999) relative standard deviations of concentrationsobtained by the passive sampling procedure including the sampling step were determined h m 18 duplicate samples. Relative standard deviations varied between 5% (benzene) and 31% (n-dodecane). In addition, quality assurance was achieved by participation in extemal intercomparisonexperiments (Hoffmann et al., 1996). From the information obtained in all these quality control activities it can be concluded that for the reported results a measurement w r between about 1Wh and 50% has to be assumed, depending on the compound and the concentration level. Statktical Data Treatment The statistical data analysis was performed using SPSS software (SPSS, 1993) and included regression analysis as well as descriptive methods. As usual, regmsion models were derived on the basis of unweighted data. To be compatible with the regression results, tables with data across categories of model predictors were also based on unweighted records. The Shapiro-Wilk test was applied to test the distribution type of the data (for more details see Results and Discussion section). Results and discussion Statistical parameters that describe the concentration distribution of most of the determined VOC are given in Tables 2-5. Not included are 4-methylheptane, 1,2,5- -trimethylhexane, 2-ethyltoluene, 1,2,4,5-tetiamethylben- =ne, tetraline, 1,4 dichlorobenzene, trichloromethane, tetmchloromethane, bromodichloromethane, 1,1,2 -trichloroethane, chlorodibromomethane, tribromomethane, a-ter- -pinene, isobutylacetate, 4-methyl-2-pentanone, hexanal, 2 methoxyethylacetate and 2-ethoxyethylacetate as mom than two thirds of the respective concentrationswere below the LOQ.Rather than determining the LOQ for each VOC individually, it was felt that it was more appropriate to choose one LOQ value for all VOCs. The LOQ was set to 1 pg/m3. This is an estimate and higher than statistically derived LOQ. The estimate takes into account some variability of the blank value and uncertaintiesof the "real" -baseline of small chromatographic peaks. Corrections h r blanks had only to be applied in the case of benzene and n hexane which were contained even in pure CS2. Table 6 gives the results h r different groups of VOCs. The concentrationof a group was calculated by summation of the concentrations of all analysed compounds belonging to the respective group. In a similar way, the total volatile organic compounds (TVOC)concentration was obtained as the sum of the concentrations of all 74 individually calibrated VOCs. In the summation, concentrations of compoundsbelow LOQ were set at 0.7L.OQ. The factor 0.7 was chosensince in the case of asymmetricdistributions it is abetter choicethan the normally used factor0.5 (Hallez and Derouane, 1982). As could be expected h m theoretical considerations (Ott, 1990) and the results of previous studies (Lebret et al., 1986; Krause et al., 1991; Wallace et al., 1991) the concentration of most of the VOCs had asymmetric distributions with a small portion of extremely high values which can be well approximated by logarithmic normal distributions. Statistical tests applied to the present data confirmed the lognormality of the distributions. Tests h r lognormality were performed as tests for normality using the logarithmicallytransformed data. Since the sample size was moderate, the Shapiro-Wilk test was p r e f d to the Kolmogorov and the x 2 test. Using the Shapiro-Wilk test, the null hypothesis of normality was not rejected for the large majority of the logarithmically transfomred VOC concentrations. Tables 2-6 give the summary statistics for the different VOCs. Besides geometric mean (GM), arithmetic mean ( A M ) , loth, 5oth, and 90th percentiles, the 95th percentile was presented to focus on persons with high exposure. The geometric mean seems to be the preferred statistical location measure for the given data since it is robust against single elevated values. To describe the estimation error, a 95% confidence interval (CI-GM) was added. The standard deviation of the GM can be easily calculated from the confidence interval and the given sample size if wanted. Toluene was the most frequently observed compound in the present study. This is in accordance with the results of other studies of personal and indoor air (e.g., Krause et al., 1991;Proctoretal., 1991) andcanbeexplainedbyboth its J o m d of Eqmmre Ao.ljnis and Endmamnml E- (2000) 1q2) 117 Tabk 2. Summary slatistics: Alkanes, including cycloalkanes (pglm') [N= 1131. compound n < L O Q PI0 P50 P90 P95 A M GM CI-GM n -Hexane -n Heptane n-Octane -n Nonane n-Decane -n Undecane -n - D o d u x n e n Tridecane n -Tetradcane -n Pentadme 2-Methylp~ntane 3-Med?ylp~~~tan~ -2.3 Dimethylpentane -2 M e t h y l b e -3 Mcdrylhexane -2 Methylheptane 3 -Mcthylheptane Iso-nonane I Iro-nonanc [I M4Ylcydopen-e cycldrexane Methvlcvcldmtane 0 6 9 20 40 13.4 10.3 (9.2- I 1.5) 0 3 5 22 49 14.7 6.I (5.1-7.3) 26 < I 28 20 4.3 2.I (1.7-2.5) 9 I 3 14 26 6.0 3.3 (2.7-4.0) 3 2 5 23 41 9.7 5.2 (4.3-6.3 ) I 2 4 21 29 9.9 5.2 (4.4-6.2) 2 3 5 12 21 8.0 5.6 (4.9-6.5) I2 < I 26 IO 3.2 2.2 (1.9-2.5) 0 2 3 5 7 3.6 3.3 (3.1 -3.5) 0 2 3 5 5 3.5 3.4 (3.2-3.5) 0 7 13 39 1SO 29.2 16.3 (13.9-19.1) 5 2 6 30 53 11.6 6.3 (5.2-7.6) 2 2 4 11 41 8.I 4.1 (3.4-4.8) 34 < I 26 19 4.3 I .7 (1.4-2.1) I 2 4 14 60 < I < I 4 45 10.3 4.8 (4.0-5.7) 9 2.2 I .3 ( 1. I - 1.5) 71 < I < I 3 7 I .7 1.1 (1.0-1.2) 8 1 2 4 6 2.4 2.0 (1.7-2.2) 62 < I < I 3 5 1.5 1.1 ( 1.0-1.3) 0 2 3 8 23 5.6 3.9 (3.5-4.5) 0 2 3 I1 20 20.4 3.8 (3.1-4.6) 2 2 3 20 54 11.5 4.4 (3.6-5.4) N=samptesize;n<LOQ~numbrrofvalucsbelowthelimitofquantificatio(nLOQ; fordctlilsxetcxt); PIO.P90,P95=percentiles;GM=gcomcbicmean; CI-GM=95% confidence interval for GM.Values below LOQ ate set to 0.7. LOQ for alculation purposes. presence in outdoor air and its frequent use indoors as a solvent in paints, lacquers, printing inks, adhesives and other household products (Kumai et al., 1983; Sack et al., 1992). Somewhat unexpected was that 2-propanol and limonenewere next in rank with mean concentrationshigher than 30 pg/m3. All together, the geometric mean of the TVW concentration was 584 pg/m3. If VOCs are considered by chemical class, aromatics were the most Tabk 3.,Summary statistics: Aromatic compounds (pglm3) [N=113]. Compound n<LOQ PI0 P50 P90 P95 AM GM CI-GM &nmne Toluene Ethylbenzme m- lp-Xylene -o Xykne Iso-/n-pmpylbenzene - -3 14 Ethyltoluene -1.2.3 Trimethylbenzene -13.4 T r i m e t h y l b e 1~3,5-Tri~y~e StYmK Naphlhslme 4-PhenYkYctohexme 0 0 0 0 1 12 I 5 0 2 24 4 0 5 II 32 69 37 7 16 25 I3 38 24 37 23 <I 2 12 35 21 32 208 382 21 106 55 283 17 67 8 I2 21 35 8 I2 22 44 7 13 78 34 67 13.5 130.2 24.0 50.5 13.6 4.6 12.5 4.9 11.8 4.8 5. I 2.3 4.8 10.5 13.9 8.5 19.9 6.5 3.5 8.3 3.8 7.3 3.7 2.1 2.1 4.1 (9.3- 11.9) (62.7-87.0) (7.1-10.2) (16.4-24.1 ) (5.4-7.7) (3.1-3.9) (7.2-9.7) (3.4-4.3) (6.2-8.5) (3.2-4.1 ) (1.8-2.5) (2.0-2.3) (4.4-4.9) N=samplc Sip; n<LOq=number of values below LOQ (for details see text); PIO. P50, P90,P95-perCentilcs; AM=arithmetic mean; GM=geomctric mean; CI-GM=95% confidence interval for GM.Values below LOQ are set to 0.7. LOQ for calculation purposes. 118 Journd of ExposureAnalysic and EnvironmentalEpidemidqp (2000) lO(2) . The Gerinan Environmeatal Survey 1990/92 (GaES 11) Hofiann etal. (D Tabk 4. Summary statistics: Aliphatic halocarboap and terpenes (pg/m3) [NE1131. Compound n<LOQ PI0 PSO P90 P95 AM GM CI-GM 1.1,l -Trichloroahane Trichloroahylene Tehachloroethylene -Q P h I C p -Pinene 3-canae Limonene Tapinme-ertefsct --y Teminene 39 75 33 1 3 12 0 22 25 I I5 20 6.6 2.I (1.7-2.7) < I 6 8 2.7 I .2 ( 1.0-1.4) 2 6 22 3.8 2.0 (1.7-2.4) 5 34 74 18.3 6.7 (5.4-8.3) 4 I2 20 8.I 4.6 (4.0-5.4) 3 17 30 9.8 3.4 (2.8-4.2) 32 124 155 53.5 34.1 (28.7-4.6) 3 9 15 5.0 3.0 (2.5-3.6) 2 5 6 2.5 1.9 (1.7-2.2) N = w n p l e s k , n<LOQ=number of values below LOQ (for details see text); PIO. P50, P90. P95=jtercu1tilqAM=erithmaic mesa; GM=gcometrk mean; CI-GM=95%confidaKe intavpl for GM. V W bdow LOQ M set to 0.7. LOQ fi# calculation puposes. prevalent compounds, followed by oxygen-containing compounds and alkanes (Table 6). Correlation between VOCs To detect correlations between VOCs of one and the same chemical class the Pearson correlation coefficient was calculated for different pairs of VOCs. Since the Pearson correlation coefficient is highly influenced by one or two high values in a data set, logarithmic concentrations were used asthe data basis. In this way, the influenceof outliersis reduced without the'loss of information that would have occurred in the ranking process needed to calculate Spearman's rank correlation coefficient. The higher the correlation coefficient between logarithmic concentrations of two compoundsthe less the ratiosof the compound levels in all samples vary. It can be assumed that high correlations of VOCs are due to emissions of the same sources. As a result of an extensive correlation analysis subclasses of highly correlated compounds could be derived (Table 7). Compounds belonging to the same subclass all showed correlations that were significantlydiff-t fmm zero at the 0.001 level of significance. The Pearson correlation coefficients within the classes varied between 0.67 and 0.99. Very high correlation coefficients were obtained for CS-aromatics. VOCs not belonging to any of the mom frequently encountered classes were, in general,not highly correlated with other VOCs as, very likely, they have some specific emission s o m s or their own source pattern. The result of the correlation &lysis was checked by fktor- analysis, anothermultivariate statisticalmethod. The classes of compounds obtained by factor analysis coincide largely with the classes given in Table 7., Tabk 5. Summary statistics: Oxygen-containiing compounds (pg/m3) [N=1131. Compound Ethylscaate -n Butylacetate -Methyl ethyl ketone 3 Heptanone Methyl bemonte -n Butanol' Iso-butanol' Iso-amyl alcohol' - -2-Ethyl 1 hexanol' 2 -Propanol' -r Butyl mahylcthcr n<LOQ PI0 PSO P90 P95 AM GM CI-GM ~~ 1 5 12 46 180 69.8 15.4 (12.4- 19.1) 29 < I 4 31 135 35.3 4.4 (3.2-5.9) 16 < I 4 44 92 24.1 5. I (3.9-6.7) 60 < I < I 4 6 1.7 I .2 ( 1.1 - 1.4) 35 < I 2 5 7 3.1 I .8 (1.6-2.1) 60 < I < I I5 31 6.9 2.0 (1.5-2.5) 67 < I < I 7 23 22.0 1.6 (1.3-2.1) 60 < I < I 2 3 I .4 1.1 ( 1.0-1.2) 25 < I 4 8 14 4.4 3.0 (2.5-3.5) 1 7 40 188 326 80.0 39.I (30.8-49.6) 57 C l < I 7 16 4.1 I .8 (1.4-2.2) N=samplc s k , n < L O Q = n u m k of values below LOQ, (for details rec text); PIO,P50, P90, P95=petcentiles; AM=ui(hmetk man; GM=g- CI-GM=95%c ~ a f i inbwd for GM. Vlhvr below LOQ M set to 0.7. LOQ fi# cphlptioa m. T h e conmbations given undagtimate the real concentrations (see text Eor details). 119 6)HqdlhronnetuL The Geman Envinmmental Survey 1990/92 (GerES 11) Tabk 6.Summary statistics: Sums of VOC concentdons by chemical class (pg/m') [N = I I31. ~ Chemical class ~~ ~~ ~ PI0 PSO P90 P95 AM GM CI-GM Alkanes. including cycloalkanes 56 I 07 336 641 I87 122 (105-140) Aromatic hydrocarbons 78 150 608 944 286 I80 ( 155-208) Aliphatic halocarbons 7 13 38 68 21 16 (14-18) Terpenes 23 57 I88 303 98 64 (55-75) Oxygen-containing compounds 44 106 438 739 308 I27 (105-152) Total volatile organic compounds (TVOC) 275 51 1 1480 2810 901 584 (509-671 ) -N=sample size; PIO.P50,P90,P95=percentiles; AM =arithmetkmean; GM=geometric mean; CI GM=95%confidence interval for GM, TVOC=sum of the concmhationsof all 74 VOC. Sources of Exposure to Benzene To determine and quanti@ the main factors that influence the benzene level, regression analysis was applied. About I00 variables comprising relevant questionnaire data were considered for entry into the regression model. Using a stepwise selection procedure a model with five predictors was derived (Table 8). The model function is a product of factors since it was obtained by exponential transformation from a linear model derived for the logarithmic benzene concentration. This model transformation was necessary because the benzene concentration did not meet the requirement of regression analysis to be normally distributed, whereas the logarithmic benzene concentration hlfilled this condition. The two most important predictors in the model relate to the presence of environmental tobacco smoke (ETS) indoors. They account for 20% of the observed variance of the benzene concentration. Two automobile-related activities, namely rehelling and the time spent in automobile traffic, were hrther predictors. Together they explain 12% of the variance of benzene. The fifth predictor is a variable expressingpopulation density in the area which accounted for 7% of the observed variance. The presence of this predictor in the model is likely to be due to elevated emission rates of benzene from automobile exhaust gases in areas with high population density. Altogether, 39% of the observed variance of benzene are explained by the model. The remaining 61% of the variance includes errors in measuring the benzene concentration and incompleteness in allowing for all possible sources of exposure. As a consequence of the latter point some specific but unknown sources of exposure are not considered in the model. On the other hand, the two known main determinants of benzene concentration, ETS and automobile-related activities, ~ f e subject to a number of variations details of which are unknown, however. For example, the effect of 1 h spent daily in automobile traffic varies depending on, e.g., traffic density, type and age of vehicle, ventilation conditions, and tempel-ature. In contrast to usual linear regression models, the interpretation of a multiplicative regression model is based on a percentage rather than an absolute change. An increase or decrease of the value of any predictor given in Table 8 always implies a percentage change of the benzene level. For instance, an increase by 1 h of the daily time spent in automobiletraffic, which means that predictorD (see Table 8) is increased by 1, results in a relative increase of the expected benzene concentration by 21% since the corres- Tabk 7. Subclasses of highly correlated VOC. Subclass Compounds C6 -alkanes C7 -alkanes C8 -alkanes C9-Cll -alkanes - -C12-$I3 -alkanes CI4 .C I5 alkanes CI-aromatiCs 0-arOmaticJ selected terpenes n- hexane, iso-hexane, methylcyclopentane -n heptane, iso-heptane, methylcyclohexane n-octane. isooctane -n nonane, iso-nonane. n-decane, n - d e c a n e n-dodecane, n-hidecane -n-teeadecane, n-pentsdccane ethylbenzene. m /p-xylene, o-xylene iso-In-propylbcozem, 3- /4-ethybhrene, 1,2,3-himethylbenzene, - -12.4 himahylbenzem, 1.3.5 trimethylbenzene a-pinene, &pinene. 3-canne 'Pearson correlation coefficients for the IogarithmicaIIy transformed concentrations. Correlation coefficients* 0.75-0.93 0.82-0.94 0.89 0.67-0.93 0.67 0.76 0.97-0.99 0.71-0.96 0.63-0.80 120 Journal of Elpoxwe Ana&& and Envhnmenttrrl EpjdemMogv (2000) lO(2) The GcfmmEnvironmental Survey 1990/92 (GerES 11) tD~ o f i m e i r . Tabk 8. Predictorsand regression model for the benzene concentrationin inhaled air. Symbol Redictor ~~ ~ ~ 0 Subject to occasional or fraluent smoking indoors (110-0y,es=l)' F Subject to frasueat smoking indoors (no=O, yes=l)' D Daily time in automobile traffic (in hours) P Retuelii (no-0, yes=l) L Living in or next to, an area with apmtmcnt houses (no=O, yes=l) Regression model for the bmzene concentration ( Y): x I.66Fx 1.21Dx1.28Px I.& (pg/m3) Percent VarianCC explained 8% 12% 6% 6Yo 7Yo R2=390/. Y=5.11 x 1.38' R2=Coefiknt of determination. 'Persons that BIC subject to h q w n t smoking indoom arc charecteriscd by 0=1and F = l , whereas persons that arc subject to o c c a p i d smoking indoors BIC *sed by o=I and F=O. ponding factor in the model equation is 1.21. Refuelling once in a week, i.e., an increase of predictor P by 1 causes an increase of the predicted benzene concentration by 28%. A person staying mainly in rooms where he/she or others are frequently smoking has to expect a benzene level in air that is 129%higher than a person who does not stay in such rooms. This relafive change follows from multiplication by both factors concerning the presence of ETS indoors. The constant 5.11 pg/m3 of the model equation represents the basic level of exposure to benzene which is not due to smoking indoors or any other of the benzene-related predictors. The influence of the different predictors on the benzene level could also be demonstrated by bivariate statistical analysis. In Table 9 the mean benzene concentrationswere given for subgroups defined by the predictors of the regression model. However, it should be noted that bivariate description is less appropriate to predict exposure than is multivariate statistical analysis, such as regression analysis. For instance, Table 9 would suggest that refuelling activities cause an increase of the benzene level by about 50% while the regression model indicates only an increase of 28%. In this case the result of bivariate analysis is erroneous because it does not take into account that there exists a correlation between the two important automobile-related variables. Obviously, people who refuel their car are drivers who spend more time in automobile traffic than others. The finding of the present study that the presence of tobacco smoke indoors and automobile-related activities are the main sources of personal exposure to benzene supports earlier results on different levels- of benzene concentrations in the indoor air of smokers' and nonsmo- kers' homes (Krause et al., 1987). It is also in agreement with the results of the TEAM study (Wallace et al., 1988; Wallace, 1990). In the present study roughly half of the subjects refuelled their car during the sampling week. Although this activity needs only a short time, its contributionto the cumulativeexposure during the sampling period is considerablebecause high benzene concentrations can be observed at gas stations. As an example, 500 pg/m3 were found by Rijmmelt et al. (1989) in Munich. While close to 40% of the variance of the benzene concentrationcould be explained, this percentage was much lower for toluene (18%). This may be due to the large number of possible somesof toluene indoorsand outdoors. The only significant predictors of toluene obtained in the regression analysiswere the occurrenceof paints or lacquers at the workplace and the use of adhesives. A larger sample size than 113 may be required to achieve a higher explanation of variance and a larger number of significant predictors. Sowres of E x p o s u ~to C8-Ammatia - - -Ethylbenzene, m / p xylene and o xylene were highly correlated (see Table 7), which means that the three compounds occurred in characteristic proportions. The concentration ratio of ethylbenzene to o-xylene was approximately 1.3 with a 95% confidence interval from 0.8 to 2.1, while the estimated ratio of m- /p-xylene to o- xylene was 3 with a 95% confidence interval h m 2.2 to 4.4. These data reflect the proportions of the C8-aromatics concentrationsin crude oil. For comparison purposesit may - - Subject IO smoking indoors never occasional mat Doily time in outomobile M 520 min 21-60 min >60 min c RefLeling no Y= Living in or next IO, on area wih aptamen1 houses no Yes 41 7.8 54 10.8 18 19.6 31 8.3 45 10.4 37 13.0 56 8.6 57 12.8 62 9.0 51 12.8 Jownd of AM@& .nlElrvlronnarrl Epilnnlolog,(2000) IO(2) 121 HoBanneral. TIE German Environmental Suntey 1990192 (GcrES 11) be mentioned that Scheff et al. ( 1989) found a ratio of 1.5 form-lp-xylene to o-xylene in auto exhaustwhile the ratio between ethylbenzene and o-xylene was 0.5. From published data (Wallace et al., 1987; Hajimiragha et al., 1989) it may be concluded that the ratio of ethylbenzeneto the other C8-aromatics is somewhat higher in cigarette smoke than in auto exhaust. Again, stepwise regression analysis was applied to identi@ the main sources of exposure. The predictors selected from more than 100variableswere identical for the three individual C8-aromatics and their sum. In Table 10 the five predictors and the model for the sum of C8aromatics are given. The models for the individual compounds differed only slightly from the equation given and, therefore, are omitted here. The regression model accounted for 60% of the observed variance of the total C8-aromatics concentration. Spending time in workshops and warehouses and the Occurrence of paints or lacquers at the work place were the major contributors to the model. The three corresponding predictors explained 44% of the observed variance. Reading newspapers or magazines and living or working near print- works or printing shops accounted for 10% and 6% of the variance, respectively. The model allows to quantify percent changes of C8aromatics levels as a function of the predictor values. The cumulative exposure to C8-aromatics increases by 22% if the time spent daiIy in workshops and warehouses is prolonged by 1 h. Workers who occasionally are subject to the influence of paints or lacquers at work have to expect C8-aromatics concentrations in the air they breathe that are higher by 69% than persons not subject to such conditions. The lowest C8-aromatics levelobtained fiom the regression Table 10. predictors and regression model for C8-aromaticsin inhaled air. Symbol Redictor ~~ ~ Percent variance explained D Daily 'time in workshops 20?! and warehouses (in hours) 0 Occasional or frcqucnt occurrence of paints Q?? or lacquers at workplace (no=O, yes= I)' F Frequent occurrence of paints 15% or lacquers at workplace (no=O, yes=l)' T Daily time spent on reading newspapers IOO? -and magazines (in hours) P Living or working near print works 6% or printing shops (no=O, yes= I ) R2=600/. Regression model for the sum of C8-aromatics concenbations (Y): Y= 1 9 . 3 I~.2ZDx I .69Ox6.4fx I S T x 1.94' (pglm' ) R*=Coefficientof determination. Trequcnt ormmnce of paints or lacquers is characteised by O= 1 end F= I. whereas occasional occmence of paints or lacquers is c h d s e d by O= 1 and F=O. Table 11. C8-aromatics concenhations in inhaled air (pg/m3) across categories of the model predictors. Subgroup -~ N Geometric means Ethyl- m - l p - benzene Xylene o- Xylene -C8 Am- matics 7Tme in workshops and Warehouses 53 hlday 100 6.8 > 3 hlday 13 48.0 15.9 114 5.3 30.4 28. I 195 Paints or lacquers at wor&place no or not 98 6.9 employed occasional 8 14.8 frequent 7 79.5 16.2 5.4 38.4 171 12.7 41.8 28.6 66.2 298 Reading newspapers and magazines 545 minlday 86 8.0 18.7 6.1 >45min/day 27 10.5 24.2 8.0 32.9 42.9 Living or working near IO print-works or printing shops no 104 7.7 18.1 5.9 VCS 8 30.9 67.3 21.0 28.4 88.6 model is given by the model constant, 19.3 pglm3. This minimum will result if all predictor values are zero. In Table 11 the mean concentrations of ethylbenzene, m-lp-xylene, o-xylene and their sum are given for subgroups defined by the predictors of the regression model. There is a striking uniformity of the three compounds in the percentage changes of the geometric means between subgroups. It is clear that in some cases, e.g., for those exposed to paints or lacquers at work, the size of the subsampleis not large enough to permit the use of the geometric mean as a reliable estimate of the mean exposure of the corresponding subpopulation. The main result of the multivariate statistical analysis was that exposure to C8-aromatics is substantially influenced by emissions from paints, lacquers and printing inks. This is consistent with the results of Kumai et al. ( 1983) and Sack et al. (1992) on the content of organic solvent components in various products. The authors found that ethylbenzene, m - / p-xylene and o-xylene were contained in more than half of the paints and lacquers under study, partially in considerable concentrations. Wadden et al. (1986) suggested that 90% of the solvents in paints and lacquers are C8-aromatics and toluene. It was somewhat surprising that other known emission sourceslike smoking and automobileexhaust did not appear in the final regression model. A possible explanationwhich, however, is a purely statistical one is that the sample size in the present study does not allow a model with more significant predictors. In fact, if the three dominant predictors covering occupational exposure were removed 122 Journol of Eqvosure Ano&sis and Environmental Epidemiolo~(2000) lO(2) I 8 . n e Ckrhan Environmental Survey 1990192 (GerES 11) 6)~ o f i m e t a l . Symbol - F`redictor pavnt Variance explained D Daily time in woiirshops and 28% warehouses (in hours) R Renovation or paintingin the penon's 13Y* sunomdings (no=O, yes= I ) N Number of cigarettes smoked per day 5% L Living in or next to, an area with apartment houses (no=O. yes=l) 3% R2=49% Regression model for the sum of C9-mmatics mncentratim ( 0 : Y= 1 8 . 7 ~1.24Dx I . U R x1.OZNx 1.2SL (pg/m3) R*=Coefficientof determination. fiom the model, some variables concerning smoking and automobile-related activities would enter the regression model. However, such a model would only explain about 30% of the observed variance of the C8-aromatics. This degree of explained variance is similar to that of the repssion models derived in the TEAM study (Wallace et al., 1988). It should be noted that in the case of competing models which explain different portions of the variance of the VOC concentration by considering different pathways of exposure, the percentage of the explained variance could be increased by combining the predictors of the competing would k v e to be studied. S o ~ r c oef~EX~OSUtoWC 9 - A n ~ t ~ t i c . ~ For the individual compounds, iso-In-propylbenzene, 3-J 4-ethyltoluene, 1,2,3-trimethylbenzene, 1,2,4-trimethylbenzene, and 1,3,5-trimethylben, and for the sum of C9-aromatics including 2-ethyltoluene, stepwise regression analysis was carried out to determine the most important factors that influence human exposure. The final regression models for the individual compounds contained three or four predictors and did not differ very much one from the other. Hence, only one model equation is given in Table 12 for the sum of C9-aromatics, together with the corresponding predictors. Almost half of the variance of the total C9-aromatics concentrations can be explained by the regression model. Spendingtime in woTL8hops and warehouseswas the major contriiutor to the model with a contribution of 28Y0. Renovation activities in the person's surroundings (at home, at work place) accounted for 13% of the observed variance. Intensity of smoking and population density explained 5% and 3% of the variance, respectively. The model can be used to predict levels of C9-aromatics in a person's breathing air and to quantify the efects of changing predictor values. Prolonging the time spent daily in workshops and warehouses by 1 h is connected with an increase of the expected level of C9-aromatics by 24% Tabk 13. C9-mmatics coocentrations in inhaled air (pg/m3) scross categories of the model predictors. N 7Tme in workshops and wamhousa 5 3 hlday 100 >3 hlday 13 Geometric means Iso- i n - 3-+4- 155- 1.2.4- 1.33 - c9- propylbenzene Ethyltoluene TrimethylbenzaK Trimahylbe-nzene Trimethylbenzene Aromatics 3.1 7.2 3.4 9.1 26.4 9.5 6.2 3.3 24.4 9.2 25. I 84.7 Renovationlpainting in the surroundings no 91 Y S 22 3.0 6.3 7.0 3.4 17.4 6.3 6.0 3.2 15.6 6.2 24.6 56.8 Number of cigmttes smoked 0 (nonsmoker) 1-20 cigareacslday >20 cigprerteslday ai 19 7 3.3 3.9 5.8 1.5 3.6 10.0 4.0 18.8 6.5 6.3 3.3 9.4 4.1 20.5 6.5 26.2 34.1 61.8 Living in or nert to, an area with apartment houses no 62 3.2 , 7.1 Ye 51 3.8 10.2 3.3 4.6 6.3 3.2 8.6 4.4 25.3 33.9 ~ J o d of Exp#rrr Analpsfr and E d m d E p u a J . l r g (2000) lO(2) 123 since the corresponding model factor is 1.24. Renovation activities during the sampling period increases the cumulative exposure to C9-aromatics by 64%. The effect of smoking is less pronounced. The sum of the C9-aromatics concentrations increases by 2% if the number of cigarettes smoked per day is increased by one. According to the model given in Table 12this means that a person who smokes,e.g., 10 cigarettes daily has to expect a C9-aromatics level in breathing air that is about 22% higher than that of a nonsmoker. The basic level of the sum of C9-aromatics is 18.7 pg/m3. It should be noted that the Occurrenceof paints or lacquers at work is an additional predictor for the two compounds 1,2,3-trimethylbenzene and 1,3,5-trimethylbenzene. To illustrate and confirm the results of regression analysis, Table 13describes mean C9-aromatics concentrations for subgroups defined by the predictors of the model. Persons who stay more than 3 h in workshops and warehouses daily have elevated C9-aromatics concentra- tions in the air they breathe. Also, the effect of renovation activities is remarkable and almost equal to the increase prognosticated by the model which indicates its independence of the other predictors. Elevated C9-aromatic levels in a person's breathing air connected with a long stay in workshops and'warehouses can be explained by emissions from solvent-containing industrial products like paints, lacquers and glues. The influence of renovation activities on human exposure to C9-aromatics is probably due to emissions from solvents. The predictor of smoking intensity reflects the exposure to C9-aromatics contained in tobacco smoke. Thus, two known emission sources of C9-aromatics, namely emissions from solvents and tobacco smoke can be shown to be relevant for human exposure by regression analysis. A third known emission source, C9-aromatic compounds in fuels and exhaust gases (Sigsby et al., 1987; Scheff et al., 1989) had only a negligible influence on a person's exposure and was therefore not considered in the regmsion model. Also, bivariate statistical considerations did not show associations between C9-aromatics levels and automobile-related activities like travelling in an automobile and refbelling. Conclusions Passive sampling over 1 week is a suitable tool to measure personal exposure. The results of such personal monitoring, in combination with information on timeactivity patterns can be used to establish population exposure distributions for VOCs if a large enough number of individuals has been studied. Multivariate regression analysis was applied successfblly to determine major sources of personal exposure to 8 number of aromatic VOCs. For benzene, 40?! of the total variance could be explained with exposure to environmental tobacco smoke and automobile-related activities having the highest influence on exposure. In the case of C8aromatics, even 60% of the variance could be explained. Here, work place conditions proved to be the essential influencing factor for exposure followed by the contribu- tion of handling printed matter in one way or the other. Besides the time spent daily in workshops and ware- houses, renovation and painting activities, and - to a lesser extent - smoking were associated with a significant increase of the exposure to C9-aromatics. For these compounds almost half of the variance could be explained by the regression model. The results of this study underline that personal exposure to aromatic VOCs is to a large extent influenced by both individual activities, especially those that relate to the automobile and to tobacco smoke, and work place conditions. Acknowledgments Financial support of this study by the Federal Ministry for the Environment, Nature Conservation and Nuclear Safety is gratehlly acknowledged. References Brown V.M., Cmmp D.R., Gardiner D., and Yu C.W.F. Long term diffusive sampling of volatile organic compounds in indoor air. Envimn. Tmhnol. 1993: 1 4 771-777. Crump D.R.and Madany 1.M. Daily variations of volatile organic compound concentrationsin residential indoor air. In: Saarela K. et al. (Eds.), PIoceedings 6th Int. Conf. Indoor Air Quality and Climate (INDOORAIR '93)Helsinki. July 4-8.1993, Vol.2, pp. 15-20 1993. De Bortoli M.. Melhave L.. Thorsen MA., and Ullrich D. European interlaboratory comparison of passive samplers for organic vapour monitoring in indoor air. Commission of the European Communities: Environment and Quality of Life. Report EUR 10487 EN, Luxembourg, 1986. Fellin P., and Otson R. Seasonal trends of volatile organic compounds (VOC)in Canadian homes. In: Saarela K. et al. (Eds.), Proceedings 6th int. Conf. Indoor Air Quality and Climate (INDOOR AIR '93) Helsinki, July 4-8, 1993, Vol. 2, pp. 117-122: 1993. Girman J.R. Hod- A.T., and Newton A.S. Emissions of volatile organic compounds from adhesives with indoor applications.Envimn. Int. 1986: 12: 317-321. Hajimiragha H., Ewers U., Bmkhaus A.. and Boettger A. Levels of benzene and other volatile aromatic compounds in the blood of nonsmokm and smokem. int. Arch. Occup. Enrimn, Heolth 1989 61: 5 13-5 18. Halla S.. and Derouane A. Novelle &ode de baitement de dries de donntes tronquceS dans I'hde de la pollutions ahnosphbique. Ski. Total Envimn. 1982 22: 1IS- 123. Heavner D.L., Morgan,W.T., and @den M.W. Determinationof volatile organic compoundsand ETS apportionment in 49 homes. Envimn.Inf. 1995: 21: 3-21. 124 Jovrrrol of Eqwsum Analp& and EnvironmentalEpidemiollogt (2000) lO(2) Tbe &Eminnumntal Survey 1990/92 (GaES11) mHoJinannetal. Heavner D.L.. M o q W~.T., and Ogdea M.W.Dearminntim of volatile organic compounds and rcapirabk suspeadtd partidate miter m New Jersey and Pennsylvania homes and worlrplaces. Envhn. Int. 1996: 2 2 159-183. Hoffinann It. SchwabeR, KrauseC., Schulz C.. Seifat B., and Ullrich D. Ummlt-Survey 1990/91, Band Iv: personengebundcne Exposition gcgenriber Whtigen organischen Verbindungen in den .Itea Bunde slaendan (Pasoad exposure to volatile organic oompwods in west Gamwy). WaBOLu-Hdte 4/1996, ISSN 0175-4211. Instihdc for Water, Soil and Ai Hygiene of the Federal Envimnmcntal Agency, Berlin. 1996. KrauJe C.. Cbutsch M.,Henke M.,Huba M.,Kliem C., Leislre M.et al. Umwelt-Survey Band Ilk: W o h n - I ~ n u a ~ mR:aumluft (Living -spaces: indoor air). WaBoLu Heftc 4/ 1991. Institute for Water, Soil and Air Hygiene of the Federal Health Oftice. W i n , 1991. Krause C., Mailahn W., Nagel R.Schulz C., Seifert B., and Ullrich D. Occurrenceof volatile organic compoundsin the air of 500 homes in the Federal Republic of Germany. In: Seifert B. et d. (Eds.). F'rocdhgs 4th Int. Conf. Indoor Air Quality and Climate (INDOOR AIR '87) Berlin, A w t 17-21. 1987. b l . I, Pp. 102-106: 1987. K d M., K o i A. Saito IC,sakuni H., Inoue T.,and TaLeuchi, Y.et al. A nationwide survey on organic solvent components m vluious aolvent products: Part 2. Heterogeaous products such as pints, inks andadhesivcs.Ina! Health 1983:21 (1):85-197. Lebrrt E., Van de Wiel HJ., h P., Noij D.. and Boleij J.S.M. Volatile organiccompouDds in Dutch homes. E n v i m . Int. 1986 1 2 323-332. Miksch R.R., Hollowell C.D., and Schmidt H.E. Trace organic chemical contaminants in office spaces.Envimn. Int. 1982: 8: 129-137. Melhave L. Indoor air pollution due to orgamic gases and vepours of solvents in building materials. Environ. Int. 1982: 8: 117-127. Ott W.R. A physical explanation of the lognormality of pollutant c o a c e a J~. A i r Wmte Manage. Assoc. 1990: 40: 1378- 1383. PellizLpri E., Lioy P., Quackenboss J., whitmom R.Clayton A.. Freeman N., Waldman J., 'Ihome~K., Rode C.. Md W~lcoslryT.P~pllptioabased exposure mcannmentr in EPA region 5: aphase I fieldsiudy in support of the National Human Exposure Assessmeat Survey. J. Expos. Anal. Environ Epidemiol. 1995: 5 (3): 327-358. Proctor C.J., W m N.D.,Bevan M.A.J.. and Baka-Rogm J. A comparison of mahods of assessing exposure to environmental to- smoke in non-smoking British women. Environ. lnt. 1991: 17: 287-291. R6mmelt H., Kessel R. Pfaller A.. Sigl H., and Fruhmann G. Arbcitsplak und Urnweltbelaslung durch Treibstoff-Emissionen an GroMenbtellen. Manch.Med. Wmhenschr. 1989: 13I (22):437-440. Seck T.M.. Steele D.H., Hammerstrorn K., and Remmers J. A survey of household products for volatile organic compounds.Amos. Embun. 1992: 26A. 1063-1070. Scheff P.A., Wadden R.A., Bates B.A.. and Aronian P.F. Source hguprints for mccptor modeling of volatile organics. J. Air Pollut. conbdASS=. 1989: 3 9 469-478. S c h i J.S., House S.,Robonic E., S d 0..Hudsoa C., !3yk M.. and Laubcr J. An investigation of indoor air CoatuninetiOa in r e s i d a ~ ~ ~ above Qy cleawr~R. kk Anal. 1993: 13: 335-344. Seifat B.. Becker IC., Hofiinann K.. KfB16c C.. and Schulz C. This issue, 2000. Seifert B., Mailnhn W. Schulz C., and Ulbich D. seasoaal variation of cancentdons of volatile organic compounds m scleded Gaman homes. Envimn. Int. 1989: IS: 397-408. Sipby J.E., Tqada S.. and Ray W. %latile organic compouad emission bom 64 in-usc passenger cars. Environ. Sci. Tmhnot! 1987: 21: 466-475. SPSS Basc System Syntax Ref- Guide, Release 6.0. SPSS, 1993. Ullrich D..and Nagel R. Comparim of d i h i v e sampling with different sampling periods and sbort term active sampling at low VOC coacmtrations. In: Pmceedings 7th M Conf. I& Air Quality and Climate (INDOOR AIR '96).Nagoya,Japan. August, 1996. UUrich D..Weiland S.K. and %fat B. VOC in homar of childm with asthma - a case-control siudy of indoorair quality. In: Roceedinc 8th Int Conf Indoor Air Quality and Climate (INDOOR AIR '99) Edinburgh, 8-13 August, 1999. VDI (Association of Germsn Engineem) Mcawement of &r9ccnur emissions, gas chnwatographic detaminetion of organic wmpounda using capillary columns, samplingby enrichment on activated c h o n - with SOIVCIIV~DI ouideline3482/Prut 4, &uthVal- &rli 1984. Wadden A.R., Uno I., and Walwratsu S.Source discriminationof short- tam h y d n x d o n samples mcasurrd aloft. Envinm. Sci. T d n o l . 1986: 20: 473-483. Wallace LA. Major sources of exposure to bavreae and other volatile Organic Rkk Anal. 1990: I O 59-64. Wallsce L.A., Duan N.. and Ziegenfirs R Can long-term exposure distriiutians be predicted fmm short-term messurrments?Risk Anal. 1994: 14: 75-85. W a l b L.A.. Nelson W. Ziegenfus R, Pellipari E., Michel L,and - Whitmae R et al.The Lm Angeles TEAM Study:personalcxposmes, ~door-o~tdoorair COC~CCO&&~IS, and breath c~1cm- of 25 volatile organic compounds. J. &asurn Anal. Envitvn. Epidemiol. 1991: I: 157-192. Wallace L.A., Pellizzari E., Hartwell T.Lf, Perria R,and Ziegenhs R Exposures to benzene and other volatile compounds h m active and passive smoking. Arch. E n v i m . Health 1987: 42: 212-279. Wallace L.A., Pellizzari E.D., Hartwell T.D.. Whimore R., Zelon H.. Perria R., and Sheldon L. The California TEAM Study: B~eath co~xntrntionsand personal exposure to 26 volatile compounds in air and drinking water of 188 residents of Los Angeles. Antioch, and Pittsburg, PA. Amos. Envimn. 1988: 22: 2141-2163.