Population Sampling in European Air Pollution Exposure Study, EXPOLIS: Comparisons Between the Cities and Representativeness of the Samples

www.nature.com/jea

Population sampling in European air pollution exposure study, EXPOLIS: comparisons between the cities and representativeness of the samples

TUULIA ROTKO,a LUCY OGLESBY,b NINO KUÈ NZLIb AND MATTI J. JANTUNENc aDepartment of Environmental Hygiene, National Public Health Institute, P.O. Box 95, FIN 70701 Kuopio, Finland bUniversity of Basel, Institute of Social and Preventive Medicine, Basel, Switzerland cEU Joint Research Centre, Environment Institute, Air Quality Unit, I-21020 Ispra (VA), Italy

A personal air pollution exposure study, EXPOLIS, was accomplished in six European cities among 25- to 55-year-old citizens. In order to compare the exposure results and different microenvironmental concentrations between the cities it is crucial to know the extent and effects of the population bias that has developed in sampling procedure and the sociodemographic characteristics of each measured population sample. In each participating city a random Base sample of 2000 to 3000 individuals was drawn from the census and a Short Questionnaire (SQ) was mailed to them. Two subsamples of the Respondents of the mailed questionnaire were randomly drawn: Diary sample for 48-h time±microenvironment±activity diary and extensive exposure questionnaires, and Exposure sample for the same plus personal exposure and microenvironmental monitoring. Significant differences existed between the EXPOLIS cities in the population-sampling procedure. Population-sampling bias was evaluated by comparing the Respondents with the total city populations. The share of women and individuals with more than 14 years of education is higher among the Respondents than the overall population except in Athens. Men, younger (25±34 years old) and unmarried individuals were hardest to get to participate in the study at least in Helsinki. The two subsamples differ from Respondents in having more employed and higher-educated individuals. The largest sample bias occurred at the first and easiest step of responding to the mailed Short Screening Questionnaire, and not at the last and most demanding stage of participating in the exposure measurements. Exposure data from some of EXPOLIS cities can only be compared to other cities with caution considering their large population bias or different sample selections. However the selection bias is not necessarily a problem for analyses about predictors of personal exposures or analyses within a city. Journal of Exposure Analysis and Environmental Epidemiology (2000) 10, 355±364.

Keywords: demographic characteristics, European cities, personal exposure, population samples, socioeconomic factors.

Introduction specific subpopulations (Jantunen et al., 1998). Six European cities: Athens, Greece; Basel, Switzerland; The goals of the EXPOLIS study (The Air Pollution Grenoble, France; Helsinki, Finland; Milan, Italy and Exposure Distributions within Adult Urban Populations in Prague, Czech Republic, were selected to represent different Europe) was to supply European air pollution exposure European regions, air-pollution situations and populations. data, which can be used to assess air-pollution distributions The most important benefit of a multicity study is to get data in populations, to search for the determinants of high from larger populations and wider variability (Katsouyanni exposures and to evaluate exposure distributions within et al., 1995). The target populations of the EXPOLIS study were the adult, urban populations of Europe. EXPOLIS focused on 1. Abbreviations: Base sample, random population sample to whom the active, working age, 25- to 55 (Grenoble 20±60) -year- SQ is sent; CO, carbon monoxide; Diary sample, subsample for TMAD and questionnaire application without exposure or microenvironmental old individuals, because their exposures are most affected monitoring ( indirect exposure assessment sample ) ; EXPOLIS,Air by urban traffic planning, zoning and occupational Pollution Exposure Distribution within Adult Urban Populations in conditions. The personalexposures and home indoor and Europe; Exposure sample, subsample for exposure and microenviron- outdoor and workplace concentrations of fine particles mental monitoring plus TMAD and questionnaire application (direct (PM2.5 ), carbon monoxide (CO) and 30 volatile organic exposure monitoring sample); PM, particulate matter; PM2.5, particles smaller than 2.5 m in aerodynamic diameter; Respondents, subjects who compounds (VOCs), together with extensive questionnaire returned a valid SQ; SQ, Short Questionnaire; TMAD, time±microenvir- and time±microenvironment±activity diary (TMAD) onment±activity diary; VOC, volatile organic compound. data were collected during 1 year from the autumn of 2. Address all correspondence to: Ms. Tuulia Rotko, Department of 1996 to the winter of 1997±1998. The overall EXPOLIS Sociology, P.O. Box 18, FIN-00014 University of Helsinki, Finland. Tel.: study design and methodology has been described in +358-9-1912-3884. Fax: +358-9-1912-3967. E-mail: [email protected] detail elsewhere (Jantunen et al., 1998; Koistinen et al., Received 20 December 1999; accepted 17 May 2000. 1999). Rotko et al. Population sampling in EXPOLIS study

The goals of this paper are to present, examine, evaluate participating in exposure and microenvironmentalmonitor- and discuss the following: ing. The purpose of the Diary sample is, on one hand, to evaluate the possible changes in the time use of the o Levels and causes of population-sampling biases Exposure sample during the monitoring period and, on the introduced at each step of the population-sampling other hand, to create a larger time±microenvironment± process. activity diary database for exposure-modeling purposes. o Similarities and differences in the samples and sampling The subsample subjects were drawn randomly from the biases between the different EXPOLIS cities. Base sample database. o The sociodemographic characteristics of each participat- EXPOLIS includes a large Exposure sample (N=201) ing city, and of the various population samples within in only one city, Helsinki, where the aim was to estimate each city. both population exposure distributions and exposure o The variation from the EXPOLIS experimentaldesign as differences between different subpopulations as well as applied in each city. the relative roles of the different determinants of exposure. o The impacts of the variation of the sociodemographic In the other cities, the aim was to estimate population characteristics and experimentaldesign in each city for exposure levels and distributions for comparison between future analyses and applications of the EXPOLIS the cities and for combined analysis of the pooled data. The exposure and microenvironmentaldata. Exposure samples consisted of 50 subjects in the other cities. In addition, another 50±250 subjects, depending on sampling logistics in each city, formed the less laborious Diary samples (Table 1). In Grenoble, no separate random Base sample was formed, and the study subjects were Materials and methods volunteers.

Population Sampling Short Screening Questionnaire In each city, a Base sample of 2000 to 3000 individuals was The Base samples in EXPOLIS study were contacted by a selected randomly from 25- to 55-year-old inhabitants mailed survey. The Base sample received an information (Table 1). Two subsamples of this Base sample were letter about the purpose of the EXPOLIS study and a two- drawn: subjects in subsample one (Exposure sample) to page questionnaire, which they were asked to complete and participate in exposure and microenvironmentalmonitoring send back to the local EXPOLIS center in a prepaid, and to respond to a TMAD and a generalquestionnaire, and preaddressed envelope. The purpose of the Short Screening subjects of the second subsample (Diary sample) to Questionnaire (Short Questionnaire or SQ) was to collect respond to the TMAD and the generalquestionnaire without basic background information about home and work

Table 1. Number of subjects of the different population samples in EXPOLIS study.

Base samplea Respondentsb Exposure samplec Diary sampled Helsinki random response rate 74% random from Respondents random from Respondents (N=2523) (N=1871) (N=201) (N=234) Athens random contacted untiltarget was random from Respondents random from Respondents (N=6968) reached (29%) (N=2000) (N=50) (N=50) Baselrandom response rate 49% random from Respondents random from Respondents (N=3000) (N=1458) (N=50) (N=282) Grenoble no base sample volunteers 20±60 years volunteers (N=54) volunteers (N=11) old half asthmatics, half controls, (N=65) Milan random response rate 25% selected, not from Base sample, random from Respondents (N=3009) (N=764) but from 15- to 55-year-old (N=250) office workers (N =50) Prague random from a limitede response rate 5% from Respondents, those from Respondents, those area in city center (N=141) willing to participate (N=50) willing to participate (N=2867) (N=36) aRandom population sample from census (Short Questionnaire mailed). bBase sample subjects who returned a valid Short Questionnaire. cExposure subsample, exposure and microenvironmental monitoring with questionnaire and time±activity data. dDiary subsample, no monitoring but questionnaire and time±activity data collected. ePrague district V.

356 Journal of Exposure Analysis and Environmental Epidemiology (2000) 10(4) Population sampling in EXPOLIS study Rotko et al. environments, socioeconomic status, commuting and some response rate in Prague was so low (5%) that the personalcharacteristics and to evaluateeach subject's Respondents cannot be considered a representative sample eligibility (living and working in the study area), and of the citizens. All Respondents were contacted by phone availability for the study. In Helsinki some of the and those who agreed to participate formed the Exposure information was obtained directly from the census, namely and Diary samples. gender, birth year, home type, home area, number of adults In Grenoble an ongoing study on the PM2.5 exposures in the house and number of children in the house. In other and daily symptoms of nonsmoking asthmatics and controls cities this information was collected with the SQ. was adapted to yield PM2.5 exposure results which can be related to the PM2.5 personaldata from other EXPOLIS Differences in Population Sampling in the EXPOLIS Cities cities. Contrary to the other EXPOLIS cities, in Grenoble Population sampling varied between the cities and the only one main Exposure sample was studied. The Exposure differences from the general EXPOLIS sampling design are sample was studied in two phases: a summer phase (phase presented here (Table 1). Helsinki followed the general 1: May±July 1996) and a winter phase (phase 2: January± sampling design and the Base sample was formed from March 1997). The phase 1 Exposure sample consisted of 40 Finnish-speaking citizens, aged 25±55 years and living in volunteers. The phase 2 Exposure sample consisted of 41 the Helsinki metropolitan area (Helsinki, Espoo, Vantaa, volunteers plus 11 Diary volunteers. In the phase 2 Kauniainen). After a reminder mailing, computer-assisted Exposure sample, 27 volunteers have already participated telephone interviews were performed by the CATI- in phase 1. Thus, the total Grenoble Exposure sample laboratory of the Regional Institute of Occupational Health consisted of 54 subjects, 27 with repeated measurements in Kuopio to contact those that had not responded to the and 27 subjects measured once. Subjects were recruited mailed questionnaires. A 74% response rate was achieved. with the help of the Grenoble Hospital pneumology service In addition, 11 volunteers, recruited independently of the and through a public appeal in a local newspaper: half of the EXPOLIS random sample among the participants of the volunteers were asthmatics, half controls (20±60 years ULTRA study (Penttinen et al., 2000), were included in old). Grenoble results are not presented in Table 4 due to an the Exposure sample. Basel also followed this sampling incomparable sampling design. design and after a second recall 49% response rate was In order to evaluate how well each population sample achieved. represents the overall population the distributions of gender, In Athens, a private (opinion polling) company was age, maritalstatus and education of Respondents of employed to find 2000 individuals, and visit their homes to EXPOLIS Base samples (25±55 years old) of each city complete the SQ. For the Exposure sample only nonsmo- are compared to the same age (25±54 years old) population kers were selected and only one of every 8 to 10 contacted of the city, respectively. In Helsinki some demographic agreed to participate. For the Diary sample potential characteristics of the Base sample are available from census. subjects were contacted by mailing, approximately 200 at Therefore, it is possible to evaluate the nonresponse bias in a time. The response rate was in the order of three out of Helsinki samples. The Respondents to the SQ were 200, so when all 1950 subjects had been contacted only 30 compared for specified characteristics between the EX- had responded. Consequently, the remaining 20 subjects POLIS cities and, also, the proportions of women and men were chosen randomly from the Base sample and contacted among Respondents are compared in the EXPOLIS cities. by home visits, which agrees well with the original The Diary and the Exposure subsamples are compared to the sampling design. Respondents to evaluate what selection biases may have In Milan the response rate in the Base sample of 3009 been caused by the rather involving requirements for the subjects was only 25%. Since in Milan over 75% of the Diary sample subjects and the quite invasive procedures and working population work in offices or similar microenvir- requirements of the Exposure sample. Pearson's chi-square onments, it was decided to evaluate the exposure for this tests were carried out to test if the proportions can be occupational category only. The Exposure sample was considered to represent the same population. Data was selected from office workers of a set of public and private analyzed using STATA Statistical Software, StataCorp buildings located in Milan. These buildings had been 1997, Release 5.0 and 6.0, Stata Corporation, College previously evaluated in former studies (Carrer et al., 1997) Station, TX, on Windows 95 and Windows NT. and can be considered representative of the different building types. Fifty office workers (not from the Base sample) were selected among the workers of these Results buildings and the adopted criteria were age (25±55 years old), place of residence (Milan) and job (only office Respondents Versus City Populations workers). In Prague the Base sample was selected from the Table 2 shows that women are somewhat overrepresented in inhabitants in area Prague V in the center of the city. The Grenoble, Prague and Helsinki. Respondents' age distribu-

Journal of Exposure Analysis and Environmental Epidemiology (2000) 10(4) 357 358 ok tal. et Rotko

Table 2. Respondents of base samples versus 25- to 54-year-old city populations in EXPOLIS cities.

Helsinki Athens Basel Grenoble Milan Prague Helsinki Respondents Athensb Respondents Baselc Respondents RhoÃne-Alpesd Respondents Milane Respondents Praguef Respondents metro- (25±55 years) (N=1, (25±55 years) (N=88, (25±55 years) (N=2 384, (20±60 years) (N=600, (25±55 years) district V (25±55 years) politan areaa (N=1871) 093,699) (N=2000) 518) (N=1458) 700) (N=65) 840) (N=764) (N=72, (N=141) (N=449, 515) (%) (%) (%) (%) (%) (%) (%) (%) (%) 167) (%) (%) (%) Gender Men484448495048504550484840 Women525652515052505550525260

Age 25±34 years 36 32 37 38 38 35 43j 45j 35 35 33 43 35±44 years 31 32 34 34 33 33 25 23 30 31 33 27 45±55 years 33 36 29 28 29 33 32j 32j 35 34 34 30

ora fEpsr nlssadEvrnetlEpidemiology Environmental and Analysis Exposure of Journal Civil status Married 58h 68 75 73 na na na na na 67 62g 58 Single, divorced 42h 32 25 27 na na na na na 33 38g 42 or widow

Education Mandatory school37 g,35i 25i 26g 27 na 17 na 0 63g 18 22g 1 Apprenticeship 14g,19i 18i 42g 43 na 34 na 15 28g 36 26g 17 A-levels/university 49g,46i 57i 32g 30 na 49 na 86 9g 46 51g 81 na=data not available. aCity of Helsinki, Urban Facts, Helsinki Region in Statistics 1996. b

NationalStatisticalService of Greece. in sampling Population cAnnualstatistics of Basel-Stadt 1997. dData Shop Eurostat (d2age80). eNationalInstitute of Statistics, Italy, ISTAT. fCzech StatisticalOffice. gPopulation >15 years old. hOnly Helsinki city area (N=259,974). iTo get a crude estimate to compare Helsinki occupation data to education the following transformation was made (in the Helsinki Short Questionnaire version the subject was asked his/her occupational

category, but in the other cities years of education to define the socioeconomic status of the subject; the years of education were asked from the Exposure and Diary samples also in Helsinki). It was EXPOLIS

20)10(4) (2000) assumed that mandatory schoolcoincides with occupation group ``other,'' apprenticeship coincides with worker and A-level/university education coincides with white-collar employees. j20±34 years and 45±60 years old. study Population sampling in EXPOLIS study Rotko et al. tion represents well the overall population, except in Prague, Helsinki Nonresponse Bias where the youngest (25±34) age category replied more The SQ was mailed to the Base sample (N=2523) in actively than the oldest (45±55) one. In Helsinki married Helsinki of which less than 2% (N=47) had moved and did subjects responded more actively compared to Helsinki city not get the questionnaire, and 24% (N=605) did not population. Education statistics from the city population answer (Table 3). The proportion of Respondents (C) was concerns inhabitants older than 15 years, therefore these significantly different (Pearson's chi-square test) in all data cannot be compared directly to Respondents' working detected demographic characteristics from those not reached age (25±55 years) sample. Both 16- to 24-year-old and by mail(A) and those who did not answer to the SQ (B). older than 55 years populations are less educated than the Specifically, the proportion of men, the youngest (25±34 working age population. However, it can be assumed that years old), unmarried, with no children and the proportion the Respondents are somewhat more educated than the of those living in small apartments ( 60 m2), decreased overall population at least in Milan and Prague. Grenoble from those not reached by mail(A) to nonrespondents (B) and Prague have the highest education level among and further to the Respondents (C). Of those who answered Respondents. the SQ and were contacted by phone to participate 12%

Table 3. Helsinki nonresponse bias (Base sample N=2523).

Aa, N=47 Ba,b, N=605 Ca, N=1871 c1a, N=73 c2a,c, N=108 c3a,d, N=424 c4a, N=1266 (%) (%) (%) (%) (%) (%) (%) Gender Men 60 56** 44 38 37 42 46 Women 40 44 56 62 63 58 54

Age 25±34 47 40** 32 41 24 34 32 35±44 30 30 32 37 29 31 32 45±55 23 30 36 22 47 35 36

Marital status Married 17 36** 68 49 72* 75** 66 Not married 60 46 21 34 16 14 22 Divorced 23 17 10 15 10 10 10

No of adults in household 1 15 34** 26 40 22* 21* 27 2 or more 85 66 74 60 78 79 73

No of children in household No children 74 67** 55 63 61** 49** 56 Children 26 33 45 37 39 51 44

Home area 60 m2 52 49** 37 49 31 35 38 >60 m2 48 51 62 51 69 65 61

Home type Single-family house 23 23** 33 25 38 38* 32 High rise 77 77 65 75 61 61 68 aA=Not reached by mail, B=No answer, C=Responding to the Short Questionnaire, Respondents from which c1=Dropped (not reached by phone), c2=Refused to participate, c3=Participated, c4=not contacted for participation. bPearson's chi-square test for sum of A+B compared to C, *p0.05, **p0.001. cPearson's chi-square test for sum of c1+c2 compared to c3, *p0.05, **p0.001. dPearson's chi-square test for sum of c1+c2+c4 compared to c3, *p0.05, **p0.001.

Journal of Exposure Analysis and Environmental Epidemiology (2000) 10(4) 359 360 Table 4. Comparison of the exposure and diary subsamples to the respondents in EXPOLIS cities with chi-square tests. ok tal. et Rotko Helsinki Athens Basel Milan Prague Respondents, Exposurea, Diaryb, Respondents, Exposurea,d , Diaryb, Respondents, Exposurea, Diaryb, Respondents, Exposurea,e , Diaryb, Respondents, Exposurea, Diaryb, N=1882 N=201 N=234 N=2000 N=50 N=50 N=1458 N=50 N=281 N=814 N=50 N=250 N=141 N=50 N=34 (%) (%) (%) (%) (%) (%) (%) (%) (%) (%) (%) (%) (%) (%) (%) AGE 25±34323135353832352829354439505035 35±44323628332834333840313030273435 45±55363337313434333431342632231629

Self smoking 28 26 25 52 0d ** 12 ** 34 16 ** 27 ** 34 38 30 26 22 18

No. of adults in household 1 2622219 1262818291222*9222615 2 596462555456596659515650525059 3ormore151417363438131612362241262426

Workplace One building/room 71 82 * 78 56 64 54 57 54 66 * 73 100e ** 65 66 76 73 ora fEpsr nlssadEvrnetlEpidemiology Environmental and Analysis Exposure of Journal Outdoors 4531261072350191083 (including traffic ) Other (multiple 9 6 7 1 0 0 17 14 15 7 0 13 6 6 3 locations daily) Home or not working 16 8 10 32 30 36 19 30 16 15 0 3 18 10 21

Education Mandatory school22 f 18f 17f** 27 4 ** 14 17 12 7 ** 14 8 8* 1 0 0 Apprenticeship 19f 20f 15f 43 38 46 34 31 36 37 33 38 17 8 15 A-levels/university 58f 63f 69f 30 58 40 49 57 57 48 59 55 81 92 85 ouainsmln in sampling Population Commuting in winter c Car 40 44 43 na 46 42 20 9 17 43 34 45 12 12 6 Bus or tram 36 35 38 na 22 14 39 35 41 26 52 ** 26 70 78 65 Train or metro 14 8 * 16 na 6 8 4 9 4 21 24 24 5 0* 6

Home location

City center high traffic na na na na na na 12 16 10 22 10 22 84 88 82 EXPOLIS

20)10(4) (2000) City center low traffic na na na na na na 22 18 22 24 24 24 6 4 9 Suburban high traffic na na na na na na 34 34 36 33 30 33 9 8 6

Suburban low traffic na na na na na na 31 32 32 21 36 21 1 0 3 study Population sampling in EXPOLIS study Rotko et al.

(N=73) could not be reached (c1) and 18% (N=108) Exposure sample in Milan consisted of selected office (24% of the Exposure sample and 11% of the Diary workers. Significant differences in the proportion of sample) refused (c2). The Participants (c3) differed smokers were in Baseland Athens. In Athens this was significantly from the other subjects, who could not be because the number of smokers was restricted in Exposure reached by phone (c1) or who refused to participate (c2), sample in Athens. The Exposure sample, compared to all by having more married subjects and, more often, children. Respondents, included a greater proportion of those living Participants (c3) also differ from other Respondents (c1, alone in Milan, a greater proportion who work in one c2 and c4) by having a larger proportion that live in single- building or room in Helsinki and Milan (by study design) family houses compared to high-rise buildings. Women and and a greater proportion who stay at home in Basel. The the oldest (45±55 years old) refused participation in the Exposure sample, compared to all Respondents, also study more frequently and the youngest (25±34 years old), included less train or metro commuters in Helsinki and unmarried, living alone in small apartments with no children Prague and more who commuted by bus or tram in Milan. were those who most often could not be reached by phone. The Diary sample, compared to all Respondents, included Married people, people with children and people living in fewer smokers in Athens and Basel. Furthermore, the Diary single-family houses agreed to participate more frequently sample, compared to all Respondents, included a greater in the Exposure or Diary sample than Respondents in proportion of those who worked in one building or room in general. Basel. In Basel and Milan the Diary sample included less subjects with only mandatory school education than the Comparison between EXPOLIS Cities Respondents sample. The Respondents to the SQ can be compared between the cities based on Table 4. The proportion of smokers among the Respondents is highest in Athens and lowest in Helsinki Discussion and Prague. The proportion of single-adult households is highest in Basel and Helsinki, lowest in Milan and Athens. Response rates in American mailsurveys publishedin Working at home or not working at all among the medical journals (219 articles) have been around 60%, and Respondents was most common in Athens (32%), in average 13% increase have been noticed in response rate followed by Basel and Prague. In Helsinki 84% of the when mailed or phoned reminders have been used (Asch et Respondents worked outside of home. Working outdoors or al., 1997). In a lifestyle survey (Hill et al., 1997) the in traffic was most common in Athens and Prague (over response rate of a postalquestionnaire was 58% and when 10%). For commuting to work in winter, over 40% used those who did not answer were telephoned, the total their own cars for at least a part of the way in Helsinki and response rate was increased to 81%. In EXPOLIS, mail Milan, only 20% in Basel and 12% in Prague. Commuting survey was used because it is easy and cheap and the by bus or tram was most common in Prague (70%) subjects were asked no sensitive questions. Mailed followed by Basel and Helsinki. Metro and train were used reminders (all cities), telephone interviews (Helsinki) most frequently in Milan and Helsinki. In Prague the Base and door-to-door interviews (Athens) were used to sample was selected from densely built area with high traffic increase the response rates. Complementary telephone volume. In Basel almost half and in Milan more than half of interview and the fact that Finnish citizens are known to the Respondents live close to a street with high traffic respond well to population surveys of public interest, volume. explain the high overall response rate (74%) in Finland. In other cities response rates varied from 5% (Prague) to 49% Comparability of the Subsamples (Basel) (Table 1). No major differences can be seen between Exposure and The random sample in Helsinki represents the Finnish- Diary samples compared to all Respondents (Table 4), speaking population ( >90% of the total population), though both Exposure and Diary samples were more between 25 and 55 years old in the Helsinki Metropolitan educated than all Respondents in each EXPOLIS city. The area (Table 2), despite some minor differences, which can

Notes to Table 4: na=data not available (question not included in the Short Questionnaire). aPearson's chi-square test Exposure/total, *p<0.05, **p<0.01. bPearson's chi-square test Diary/total, *p<0.05, **p<0.01. cMultiple choices allowed. dSmokers were restricted in the Exposure sample according to study design. eExposure sample was limited to office employees by study design in Milan. fIt was assumed that mandatory schoolcoincides with occupation group ``other,'' apprenticeship coincides with worker and A-level/university educ ation coincides with white-collar employees.

Journal of Exposure Analysis and Environmental Epidemiology (2000) 10(4) 361 Rotko et al. Population sampling in EXPOLIS study be corrected by weighting. Therefore the exposure results be usefulfor study cost reasons. Peoplehave other priorities can be generalized for the population and used for than just minimizing their efforts, namely curiosity, interest, modeling. The 49% acceptable response rate of the Basel being a part of something deemed important and different, Base sample should be considered to be a borderline case having some interesting conversation piece at the workplace for accurate representation of the generalpopulation, and coffee break or lunch table, i.e., a social competitive edge. needs to be corrected by weighting for age, gender, Within these conflicting priorities, Diary sample, although residence and educational differences. Clearly, all three less demanding, is also less rewarding than Exposure samples in Athens and even more in Prague need to be sample, and the SQ, although quickly and easily answered, considered as unrepresentative of the generalpopulation. carries almost no rewards at all Ð just a duty from an We do not have a representative population sample from anonymous body that can be easily ignored. It was in Grenoble due to the different study design of nonsmoking generaleasier to get women and educated individualsto asthma patients and volunteer controls. Inferences made participate in this demanding population study in each about exposures in Athens, Prague and Grenoble to the participating centre. general population as well as comparison with other cities Considerable effort was made to ensure comparable data must be done with caution and after considering the possible from the very different EXPOLIS cities including devel- exposure bias caused by gender, age, residence and opment of identicalequipment, protocolsand question- educationaldifferences between the samplesand the general naires, common training, etc. Similar harmonization was population. The Milan exposure results can be compared to not achieved in forming the Base, Exposure and Diary those of office workers in Milan and in the other cities, but samples due to the administrative and cultural differences not to the more general population samples. across the cities, especially the differences in the attitudes In Payerne, the Swiss SAPALDIA study centre, where towards less (Diary sample) or more demanding (Ex- nonrespondents had been contacted and interviewed by posure sample) participation. However EXPOLIS residen- telephone, the nonrespondents had in average lower social tialoutdoor measurements represent more accuratelythe class and lower education level than the respondents. No spatial variability of the outdoor concentrations (at least significant difference could be seen regarding age, gender, for PM2.5) of each city than fixed-site monitor data nationality, civil status or smoking status (Martin et al., considering the large number of the sampling sites 1997). In a lifestyle survey (Hill et al., 1997), a sample of representing the distribution of the residences of the nonresponders was also contacted by phone for evaluating population. However, in Prague outdoor measurements selection bias. Nonresponders had a higher share of men, were located only in the center of the city. Workplace smokers and the young. These results are similar to Helsinki measurements are comparable between the cities keeping where the proportions of men, young (25±55 years old), in mind that the workplaces of educated individuals and unmarried, those not having children and living in small therefore cleaner work environments are somewhat over- apartments were all significantly higher for nonresponders represented and in Milan the monitored workplaces than responders (Table 3). The selection of nonresponders represent only offices. Comparison of indoor and personal towards single, young individuals living in small apart- measurements between EXPOLIS cities requires taking the ments, mostly located in city centre, is likely to increase major exposure determinants of each air pollutant into their exposures to air pollutants relative to responders account and adjusting for socioeconomic factors to reduce (Oglesby et al., 2000). sample selection and population bias especially in Milan The largest sample bias seems to be introduced at the first and Prague. The latter concern applies to the questionnaire and easiest step of responding to the SQ and agreeing to and time±activity data. Because the number of smokers participate, and not at the last and most demanding stage of was restricted in Athens, uncorrected exposure results exposure sample at least in Helsinki. Interestingly (Table would underestimate the exposure of the city population to 3), the ``demographic'' biases, gender and age, in smoking-related pollutants; however, population data to responding to the SQ Ð the first self-selection stage Ð weight for smoking are readily available. Subsamples and are slightly reduced in the Diary and Exposure samples Ð exposure data should not be compared crudely across the the second self-selection stage. In contrast, the ``socio- EXPOLIS cities or be used directly for distribution economic'' biases, maritalstatus, children,size and type of simulations without adjusting or weighting the data. residence, are further increased from the first to the second However the selection bias is not a concern in analyses self-selection stage. It also seems that the more demanding about predictors of personalexposures, especiallywith Exposure sample represents the population better or at least multivariate models, or analyses within a city. as well as the less tedious Diary sample. The Duan and It is difficult Ð especially in Central and Southern Mage (1997) idea of splitting the more demanding Europe Ð to obtain and motivate representative population exposure and less demanding diary sample does not seem samples for demanding air pollution exposure surveys. The to improve sample representativeness, although it may still tasks that a survey inflicts upon the study subjects may

362 Journal of Exposure Analysis and Environmental Epidemiology (2000) 10(4) Population sampling in EXPOLIS study Rotko et al. affect their behavior during the monitoring (Boudet et al., The selection bias is not necessarily a problem for 1999). It is essential, therefore, to collect information about analyses about predictors of personal exposures, especially the exposure relevant outdoor and indoor environmental, with multivariate models, or analyses within a city. behavioraland time±microenvironment±activity patterns Ð the determinants of exposure Ð of larger and more representative samples of the studied population. This Acknowledgments information can be obtained from census data and from larger and less demanding population surveys, the repre- EXPOLIS was supported by EU contracts ENV4-CT96- sentativeness of which also needs to be ensured. The 0202 and ERB IC20-CT96-0061 (Prague); Academy of population exposure estimates should be based on the Finland contract No: 36586, intramural KTL projects Nos: exposure survey data corrected to represent the more 2169, 2127 (Helsinki); Bundesamt fuÈr Bildung und general population. Uncorrected exposure levels and Wissenschaft BBW No: 95.0894; Schweizerischer Natio- distributions should only be applied, when the representa- nalfonds 32-048922.96 (Basel); EidgenoÈssische Tech- tiveness of the exposure survey sample has been demon- nische Hochschule ZuÈrich (ETHZ), (Departement strated, or when the differences between the exposure Umweltnaturwissenschaften Position No: 47112); French survey sample and the general population have been Department of Environment (contract 96096), French demonstrated to be not relevant to exposure. NationalEnvironment Agency (ADEME contract In conclusion: 9693035), Union RoutieÁre de France and Grenoble The largest sampling bias is not necessarily introduced CommunauteÂ de Communes (Grenoble); other national by the most demanding tasks for the study subjects, and research funds, and intramuralfunding from the participat- other factors than personalinconvenience may affect the ing institutes. subject's decision to participate. Participating EXPOLIS cities and institutes in Europe are Women and more-educated individuals are overrepre- NationalPublicHealthInstitute, Department of Environ- sented among the Respondents to a mailed SQ compared to mental Hygiene, Kuopio, Finland; University of Basel, the generalpopulations of the cities. Institute of Socialand Preventive Medicine, Basel,Switzer- Significant ``demographic'' (age and gender) and land; University Joseph Fourier, Medical School, Grenoble, ``socioeconomic'' (maritalstatus, children,size and type France; University of Athens, MedicalSchool,Athens, of residence) biases are introduced through nonresponding Greece; University of Milan, Institute of Occupational to a mailed SQ in Helsinki. However, these ``demographic'' Health, Milan, Italy; RIVM Department of Chronic biases are slightly reduced, but ``socioeconomic'' biases Diseases and EnvironmentalEpidemiology, Bithoven, the further increased in participation to the more demanding Netherlands; Regional Institute of Hygiene of Central Diary and Exposure tasks of the survey. Bohemia, Laboratory of Genetic Ecotoxicology, Prague, People with higher education and employed outside of the Czech Republic; VTT, Chemical Technology, Espoo, home are overrepresented among the rather involving Diary Finland. sample (intensive questionnaires and 48-h time±activity diary) and the quite invasive Exposure sample (the previous tasks plus exposure and microenvironmental References monitoring) compared to the less demanding Respondents sample. Consequently, unweighted population exposure Asch D., Jedrzewski K., and Christakis N. Response rates to mailsurveys estimates would be biased towards the exposure patterns published in medical journals. J. Clin. Epidemiol. 1997: 50 (10): and levels of these groups. 1129±1136. Boudet C., Zmirou D., KuÈnzli N., and Oglesby L. Subjects adapt time± The residential outdoor air measurements of EXPOLIS activity patterns during participation in a personalexposure assessment are expected to be comparable between the cities albeit the study. Epidemiology 1999 (submitted). variability in population sampling, except for Prague. Carrer P., Alcini D., and Cavallo D. Daily personal exposure to air The workplace air measurements are expected to be pollutants of office workers in Milano. In: Woods et al. (Eds.), comparable between the cities except for Milan, where all Proceedings of ``Healthy Buildings '97,'' Washington, 1997, Vol. 2, pp. 249±254. monitored subjects work in offices, and should only be Duan N., and Mage DT. Combination of direct and indirect approaches for compared to office employees in the other cities. exposure assessment. J. Exposure Anal. Environ. Epidemiol. 1997: 7 The home indoor air and personal exposure measure- (4): 439±70. ments need to be corrected statistically to better represent Hill A., Roberts J., Ewints P. et al. Non-response bias in a lifestyle survey. the generalpopulations of the cities or defined subgroups. J. Public Health Med. 1997: 19 (2): 203±220. Before such corrections are applied, the effect of each Jantunen M.J., HaÈnninen O., Katsouyanni K., KnoÈppelH., KuÈnzli N., Lebret E., Maroni M., Saarela K., SraÂm R., and Zmirou D. Air population bias on the specific home indoor air or personal pollution exposure in European cities: the ``EXPOLIS'' study. J. exposure needs to be assessed. Exposure Anal. Environ. Epidemiol. 1998: 8 (4): 495±518.

Journal of Exposure Analysis and Environmental Epidemiology (2000) 10(4) 363 Rotko et al. Population sampling in EXPOLIS study

Katsouyanni K., Zmirou D., Spix C., Sunyer J., Schouten J.P., PoÈnkaÈA., method and participation in the cross-sectionalpart of the Swiss Study Anderson H.R., Le Moullec Y., Wojtyniak B., Vigotti M.A., and on Air Pollution and Lung Diseases in Adults. Soz.-PraÈventivmed Bacharova L. Short -term effects of air pollution on health: A 1997: 42: 67±84. European approach using epidemiological time -series data. Eur. Oglesby L., Rotko T., KruÈtli P., Boudet C., Kruize H., Jantunen M.J., and Respir. J. 1995: 8: 1030±1038. KuÈnzli N. Personal exposure assessment studies may suffer from Koistinen K., Kousa A., Tenhola V., HaÈnninen O., Jantunen M., Oglesby L., exposure relevant selection bias. J. Expos. Anal. Environ. Epidemiol. KuÈnzli N., and Georgoulis L. Fine particle (PM2.5) measurement 2000 (in press). methodology, quality assurance procedures and pilot results of the Penttinen P., Timonen K., Mirme A., Ruuskanen J., and Pekkanen J. ``EXPOLIS'' study. J. Air Waste Manage. Assoc. 1999: 49: 1212±1220. Ultrafine particles in urban air and respiratory health among adult Martin B., Ackermann-Liebrich U., Leuenberger P. et al. SAPALDIA: asthmatics. Eur. Respir. J. 2000 (in press).

364 Journal of Exposure Analysis and Environmental Epidemiology (2000) 10(4)