Tobacco Documents Online

Int Arch Occup Environ Health (1983) 52:285-300 Limitations to the Use of Employee Exposure Data on Air Contaniinants in Epide>iniologic Studies Ulf Ulfvarson Arbetarskyddsstyrelsen. National Board of Occupational Safety and Health. Ekelundsviigen 16, S-17184 Solna, Sweden Summary. The bias in the estimation of uptake of substances in the human body from exposure data gathered from ordinary workplace check-ups is dis- cussed. It is concluded that most exposure is probably overrated. This means that exposure limits based on these premises will tend to be too high. To counteract this bias in the future, filed exposure data should be accompanied with information on a number of circumstances which prevailed at the sampling. Key words: Bias - Exposure data - Air contaminants - Work sites - Permissible exposure limits Introduction The purpose ofan epidemiologic investigation is to establish a causal association between categories of events occurring in a group of persons: one category being, for example a disease in a certain proportion ofthe group and another category being an attribute or experience in a certain proportion of the group [21J. This attribute or experience in occupational health will often be an exposure to a known chemical or physical factor. The definition of noneffect levels of these factors or even the substantiation ofdose-response or dose-effect relationships is of special interest for the purpose of setting or revising exposure limits [16]. This ambition as goal unfortunately is often hampered by the inadequacy of'exposure data. The concept of exposure ofan employee to a substance in the work environ- ment may denote at least two things. It may indicate the dose of the substance absorbed in the body. It may also merely indicate the presence ofthe employee in an environment in which there is a more or less well determined concentration of' the substance, from which an uptake of the substance is deduced. The purpose ol' the following inquiry is to discuss the bias in the estimated uptake. Random error in concentration determinations, although important as such, is considered only briefly in this context.

286 U. Ulfvarson Exposure measurements may be perfonned either in order to check if the ~ qt9vtr~nnlent complies with exposure limits or, at least in theory, in order to estimate the`uptt}ke of a substance for the purpose of epidemiologic studies. It is important to separat8'thqae two goals, since the outcome of rneasurements with one or the other purpose will be quite difl'erent. Most epidemiologic studies are retrospective. Ifa prospective investigation is considered, it is probably also justified to control the exposure, which will entirely change the premises of the investigation [161. Consequently most epidemiologic studies are reduced to using data from measurements already made, and made for other purposes. I-Ience it is important to judge the relevance of the few data which may be found in the best possible way. The following factors and circumstances may affect the bias ofestimated uptake deduced from exposure data: (a) the intention of the measurement, (b) the methods used and (c) the investigation strategy. These factors afTect the representativeness of the exposure data. Additional factors affecting the uptake but not the concentration of contaminants in the air must also be considered. Such factors are (d) part-time exposure. (e) the use of respirators, (f) personnel rotation, (g) uniavourable distribution of exposure periods over time and (h) unusually hard work increas- ing the ventilation of the exposed individual. 2. ,1lethmtls 2.1 Sanrplitrg qJr Air Corrtairrinatrt.r The mcasurement ofexposure to air contaminants may be pcrti>rmed by analysing theircon- centration and variation with timc in the inhaled air of the exposed employee. This can be acconiplishetl by sampling close to the nose oithc employee. This has been a regular practice =since the introduction of portable eyuipment inclutling battery driven pumps in 1961) f 251. It is also possible to estimate exposure irom data on the concentrations in various places or "zones" where the employce is present [101 or from exposurc data with regard to various occupational titles or uniform task classes to which the employee belongs f II, 12.321. 2.2 Investigation Strategies The purpose of this section is to list and describe sampling strategies which ttre in current use or niay have been used in the past. 2.2.1:Vo Visits ".Srcrrre,qr• :When no visits at all are made to a work prace, this "strutcgy" may be a well founded choice or due to lack of' resources, or even ignorance. 2.2.? Irr.rpec•tiurr. At its best an inspection is a etualikiecl evaluation based ttpon survcyinga place ofwork and a work operation, possibly contbined with inl'erence from earlier meusureruents at the sante place of %cork or svhat are suppuseil to be similar places of work. Actual meatiure- ments may occasionally be pert'ormed at the inspection, e.g. by detection tubes or other direct reading instruments. When notations have been m,itle regarding reactions to the eniiron- ment. e.g. smeH of a substance or ucute rcactions ol'the empIoyces as irrttation, lachrvmation oreven fainlint!a conclusions as to the level ofeqposure may be clrawn. Thr fe« data gencrateel in an inspection are unlikely to he representative. The samplcs may refer to the general work room atmosphere or to the "worst casc" without any reference in the records. 2.2.3 IJenritTrarion. This indicates sampling and analysis with the purpose nf iilentifqine unknown substances in tfte air. Yhe saunplus may be taken, e.g. by a"high volume s;implrr" in a suitable position of the premises invetitigated or by other means depending on how the

Limitations to the Use of Employee Exposure Data on Air Contaminants 287 determinations are to be performed. Identification of unknown substances are meant to be just qualitative and any noted concentrations are not likely to represent normal conditions. The desire to collect enough material for a determination will probably lead to a positive bias, in such data. 2.2.4 Finding "Worst Cases". Random samples (grab samples) are taken during minutes or longer periods depending on the sensitivity of the analytical methods in the vicinity of operations and during phases of operations where high exposure is expected. The purpose is to get a measurement relevant to the "worst possible case" and also to check compliance with ceiling standards. "Whorst cases" are by definition not representative. Ifthe results are appl ied to other groups within the same industry, the bias will be positive. 2.2.5 Monitoring oJ'Time Weighted Average Concentrations. The goal is to measure the daily average exposure of selected employees during normal production and to check compliance with exposure limits [28J. The samples may be taken as a number of short-period samples at random during a day or as full period samples [19,28J. If data are applied to other groups of employees, the bias may be positive or negative depending on how the selection was made. 2.2.6 Work Area Savnpling. This may indicate static sampling in the work area in order to investigate the background concentration of the air contaminant there with no particular regard to the exact position of the employees in the localities. Such sampling will in itself not reveal exposure. The level of exposure has to be derived from other data as well. These data may be incomplete or entirely lacking. Since the occupational hygienist may be inclined to sample where he can get high results, while the employees avoid "hot spots" if they can, the bias may be positive. +, Work area sampling may also indicate static sampling made with regard to positions of employees where an average concentration is likely. In this case, much random error in estimating individual exposure is likely, but not essentially biased. 2.2.7 Emission Measurement. Measurement of the emission (weight unit oicontaminants per time unit) from different sources of contamination is applied in a few instances. Emission measurements say nothing about the exposure. The calculations involved in order to derive the exposure are based on several very uncertian factors, e.g. production, ventilation, air movements, and positions of employees in the localities. 2.'..8 Biological Sampling. When sampling and analysis of e.g. blood or urine from exposed employees is applicable, they are supposed to reflect uptake in individuals or groups in the best possible way. The relationship to exposure in the environment varies with the substance in a complicated way, however. Samples may be taken when a dangerous exposure is ex- pected, but it is also possible that all employees of a certain trade are examined on a routine basis. Strategy no.2, sometimes combined with no.3, may form the introductory phase of a planned larger exposure monitoring program. The most common of the strategies is no. I and then nos. 4,5 and 8. Strategies nos. 4 and 5 may be applied separately or combined. The impor- tance of strategy no. 6 has decreased since portable sampling equipment has been introduced, but it is still used in connection with the planning and checking up of control measures. Strategy no.7 is unusual, especially since it is difficult to arrange for entission measurements in the normal operations of a factors. A hybrid between emission and immission measure- ments has been practised in a study of welders [31J. 2.3 Compliance Testing Compliance with exposure limits may be tested in different ways. While Leidel et al. [19) advocate a very precise statistically based compliance testing, others prefer qualitative reasoning, e.g. the Swedish authorities [8]. Most suggested procedures are derived l'rom the following models. 2.3.1 Conventional Procedure. Compliance exists when the time weighted arithmetic mean is below the exposure limit. Excesses above the limit are permitted within short-term exposure

288 U. Ulfvarson limits [6,3J. The short-term exposure limits will function as a limit to the acceptable range in readings when the mean is close to the exposure limit and thus limit the chance of false statements of compliance. 2.3.2 Testing o!'Sicuisrical /ftporhesi,S. Comparison H'ith the C'onlTdence Linrits. The normal or log-normal distribution model is used to calculate the upper and lower confidence limits of the mean at some reasonable confidence level. It is suggested that the lowercontidence limit be used primarily by the compliance officer to test possible noncompliance with exposure limits and that the upper confidence limit he used primarily by the eniplot'cr to test possible compliance with exposure limits-[19J. A shortcoming of'this procedure is that the range in readings is not controlled. Therefore the power of the test should also be considered 1281. 2.3.3 (h•ere.vposrrre Risk. The risk of overexposure on any occasion when the apparant exposure on that occasion is in compliance with the exposure limit has been calculated con- sidering various premises [2U[. It is suggested that an "action level" of half the exposure limit be used. 2.4 Decision dlodels It is a tact that the eagerness with which exposure limits have been enforced has varied con- siderably with time and from one nation to another, as it depends among other things on the legal status ofthe exposure limits. The consequences of noncompliance with exposure limits therefore have been very different in dil7'erent situations in the past. The decisions critical to the bias ofold exposure data regard the possibility ofrepeating measurements and the control measures taken. II' effective control measures result from a decision of noncompliance o ith exposure limits, the old exposure data wifl rapidly be rendered obsolete. If aphlied lo Iuture situations they will be potiitively biased. It' mcasuremcnts are repeated only when hiRh concentruions are registered and f'urther measurements cancelled when a single luw reading is observed, negative bins is incvitable. A detailed decisions scheme has been developed for the National Institute of Occu- pational Safety and lfealth [ I9J. When the exposure limit has been e.ececdecl. the person is inforrnrd about it and control measures taken. 'I hen new measurements are performed at tegular intervals until repeated results show concentrations below the "action ie%;;l". When the "action level", but not the exposure limit, is exceeded measurements are performed at regular ititervals until repeated results show concentrations below the "action lovel". If this program is enfi>rced the influence of' chance in measurements and decisions will be minimi7ed. The data generated will probably be representati~e for most exposed employees on the premises. Important to the usefulness of exposure data are also the decisions taken when random samples are fitr below the exposure limit. In most cases all further mcasur,- ments will be cancelled when concentrations are well below the exposure limit and lhc number ot'data in such situations theref'ore will be very limited. 3. General Idepresentatir•eness of I yposure Measurements 3.1 Distribution of Concentrntion.s o/'AirCattnnuitatNs Concentrations of samples of contaminated air usually have po,sitiv(~ly skewecl distributions. It is now rathcr>;enerally,tcknowletiged that these distributionsare approximately log-normal [i9,28J, i.e. the logttrithms of' the concentrations are normally distributed [2J. As tun cxample, the general agreement with the log- normal distribution of a large aunount ofconccntration data in connection with welding has been confirmed [31). The relative positions of the mean, median (= geometric mcan) and rnode (= most ('reyuent value) at c~' ' t2" 2, cl' and c~'-"' (,u and a being the mcan and

Limitations to the Use of Employee Exposure Data on Air Contaminants 289 probability density Fig. 1. Relative positions of arithmetic mean and median in the log-normal distribution measure units Table 1. The area on the left of the arithmetic mean under frequency curves of log-normal distributions GSD Area with varying geometric standard deviation 1.5 0.58 2 0.64 2.5 0.68 3 0.71 3.5 0.73 4 0.76 standard deviation of the logarithms of the variate) emphasize the positive skew- ness of the distribution, cf. Fig.1. Some authors also refer to the geometric standard deviation, GSD=e° [19,20]. A simple relation obtains between the quantiles of the log-normal distribution and the corresponding quantiles ofa nor- mal distribution with mean = 0 and standard deviation = 1. If vq is the quantile of order q of the normal distribution then the q:th quantile of the log-normal distribution will be e"+`'a' ° The skewness of distributions of air contaminant concentrations found in industrial operations has been investigated by Leidel et al. [20). Also, quoting other authors, they conclude that the median category of GSD's of particulate sampling data from a great number of measurements in various branches was 1.60 to 1.69 and the median category GSD's of gas and vapour sampling data 1.50 to 1.59 [20]. The difference in intra- and interindividual variations may be exemplified with the dust exposure of 63 welders performing shielded electric arc welding on stainless steel [30]. Dust was sampled inside their face guards in the morning and in the afternoon. The GSD within days (same welder) was 1.2 and between welders 1.6 on the average. An illustration of the consequences of the skewness is given in Table 1 giving the relative frequencies with which log-normal distributed observations will fall below the arithmetic mean when the GSD varies. If the distribution is normal

290 U. Ulfvarson TCA in urine mg/litre 40-1 20 ~ \ ~`\ J_~i~^ \___' year 45 47 49 51 53 55 57 59 65 67 68 69 71 72 74 76 Fig.2. Yearly geometric means ofTCA in urine samples taken in health controls on employees exposed to trichlorethylene in Sweden 1945-1976 this frequency by definition is 50%. When the skewness increase the relative frequency of observations below the arithmetic mean increases, e.g. there is more than a 70% chance in random sampling of observing a single variate less than the arithmetic mean when the geometric standard deviation is 3, a high but not unusually high spread. Only ifa number of observations are made, will the mean converge with the true overall mean. The importance of the arithmetic mean is obvious considering that, in full period sampling, the arithmetic mean of the pefiod is automatically registered and that in principle the accumulated uptake of the human body is proportional to the arithmetic mean of the period under observation. This will be further discussed below. 3.2 Variations in Exposure over Lon; Periods In Sweden and other industrialized countries there has been a general trend towards lower employee exposure to air contaminants in the work environment at least during the 1970's,as indicated by the continuous revisions of the exposure limits. The trend is shown in those cases where long-term series of incidences of occupational diseases or exposure measurements are available its reported for silicosis and quartz exposure [1]. As a measure ofexposure to trichloroethylene, trichloroacetic acid (TCA) has been determined in ttrine for a long time (3j. Each year since the 1940's, hundreds of urine samples have been sent to the National Institute of Public I Iealth, later the Institute of Occupational Health, in Sweden for analysis of TCA. The yearly geometric means of these analyses are plotted in the Fig. 2. The curve indicates a general long-term increase in the exposures on to the end ol'the 1960's and then a signilicant decrease. The yearly means ofthe con- centrations of lead in blood samples taken in health controls in Finland during a number nt' vears has been reported [17]. From 1969 on to 1976 there has been a .teady and ,ub.stantial decrease.

Limitations to the Use of Employee Exposure Data on Air Contaminants 291 Table 2. Investigations in a number of representative paints factories in 1976 and 1979. Range of the sum of concentrations of solvents, each standardized with the corresponding permis- sible exposure limit value in common operations before and after an extensive cleaning-up program [4,26,29] Operation Before control (1976) After control (1979) Charging of raw materials 0.02-16.0 0.15- 0.54 Pigment dispersion 0.2 - 4.4 0.43- 0.78 Tinting 0.1 - 2.0 0.3 - 2.4 Filling of cans 0.02- 6.6 0.12- 1.1 Cleaning of equipment with solvents 0.5 -30.0 0.33-47.0 It is obvious that a lot of changes habe-been imposed on the individual factory, e.g. as a result of measures taken to limit exposure or as a consequence of changes in the production. It is necessary to have accurate knowledge of the investigated factory or branch if exposure data is to be correctly interpreted. As an example, two surveys in the paint industry in 1976 and 1979 [4,26,291 are presented in the Table 2. Between these two years large-scale cleaning up operations were in progress in the whole branch. As a result the exposure to solvents decreased con- siderably in several types of operations. The group of persons occupied with the solvent cleaning of equipment, who were the worst exposed employees, showed no change in their conditions, however. The conclusions of these examples is that if exposure data are generalized over a long period of time, the estimated uptake of substances will probably be considerably in error. The sign of this bias will vary depending on the long-term devel9pment of the exposure to the substance in particular and how the generali- zation is made. 3.3 Variations in Exposttre with the Time of the Year When there is a season effect on exposure this must obviously affect the in- ference from measurements during one season to other seasons. An example of effects due to the time of the year is given in the Fig. 3, showing the exposure of welders to welding fumes. Welding the same material and with the same method is performed in a rather similar way irrespective of where it is done [31]. In the figure geometric means of exposure of different welders in different enterprises working with three independent methods have been plotted against the month of the year in which the measurements were made. The differences are significant according to analyses of variance performed with the logarithms of the con- centrations. A second example is given in the Fig.4. Monthly geometric means of TCA concentrations in urine samples mentioned above show a significant seasonal effect on exposure. An explanation of the season effect on the two entirely different exposure situations in the examples may be the changes in the general ventilation during a year. When the outdoor climate is mild in Sweden in May through August, the air exchange through windows and doors may be important, while there is a ,

292 i of averages within methOds a zo0a 1 iCA in urine v @9/litre 40 30 zu to~ • Mrtal src weld- , in stainless steel (Nov 1974-July 1975) p Tig in fll (Oec 1974-Oct 1975) o Mig indJtl o Stationary samnlin9 (Nov 1974-June 6 A 9 10 11 11 6 7 6 9 l0 ll 12 1976) month nonth U. Ulfvarson Fig.3. The effect of the season. Geometric means ot' con- centrations of welding fumes in the inspiration air of the welders inside the face guard and at stationary sampling places in the work rooms. The samples were taken in 40 enterprises 1974-1976. No local exhaust ventilation (spot ventilation) was used. The monthly geometric means of each method is given as percent ot'the grand geometric mean ol'the method, for metal urc welding in stainless steel 3.8 mg/ml, for Tig in Al 1.1 mg/m1, for Mig in At 10.4 mg/m} and for the background in the work rooms 3.8 mg/m' I FiR.4. The effect of the season. Geometric means of trichtor- ncetic acid in urine taken in health controls on employees exposed to trichlorethylene in Sweden the years 1945, 1947, 1949, 1951, 1965, 1966, 1967, 1968 and 1969 tendency among employees to turn off the forced ventilation to avoid draught when outdoor temperatures are low in January through March. The variations between the days of the week is sometimes appreciable us observed e.g. in dry cleaning enterprises and metal industries where the exposure to trichloro- ethylene was at its peak in the middle o!'the week (3]. The situation with respect to work load in the industry may also influence the exposure. This situation follows a typical pattern overthe yearand should b-, con- sidered for the particular branch under study. In the above example ofexposure to dust in welding, the are time factor was observed. There was no significant ttittcrence in dil'I'erent months in the arc time factor indicating that the work intenslty did not cause the variation.

Limitations to the Use of Employee Exposure Data on Air Contaminants 293 3.4 Errors in the Determination of Samples of Air Contarninants The magnitude of the errors in the determinations of samples made one or two decades ago are rarely stated. This is so partly because a statistical view on analy- tical results is a rather recent accomplishment, even today not universally ac- cepted by all analytical chemists. Nowadays it is common to provide descriptions of analytical methods with data regarding precision [5]. Coefficients of variation in repeated analyses with a number of frequently used methods operated by the same operator are reported in the range 5-10% [5, 10]. This is an overestimation of the precision in the deter- mination, however, when different laboratories and different methods and instruments are used. In a series of interlaboratory calibrations, samples containing predetermined quantities of quartz, asbestos fibers or organic solvents where determined [18]. The limit of acceptance of a laboratory was for quartz and asbestos about ±30% from the predetermined value and for organic solvents ± 15%. The standard deviations and ranges in determinations of organic solvents are presented in the Table 3. This table shows that some laboratories make large errors, although fortunately this is rather rare. The variation due to analysis is usually much less than the variation due to the sampling error at different times of e.g. the day. If a sample is taken e.g. in the inhaled air of an employee during one day and n determinations are made of the sample then the variance sZ of the mean of the n determinations is composed of the variance si in the true concentration of each sample and the variance s'Z in the determination of each sample: sZ=si +sZ/n [14]. In most cases only I deter- mination is made ofa sample and the equation is reduced to s-2=sf+s2. Usually the estimation of sZ has to be based on repeated measurements ofa few samples. It can be safely assumed that most methods of determination have co- efficients of variation far below e.g. 30%. To exemplify the importance of a co- efficient of variation of 30% in the analytical determination it is assumed that the GSD of the true concentration values in one case is 1.60. Then the total GSD due to variation in the concentrations and in the determination will be about 1.70. The- conclusion is that low precision in the determinations (e.g. about 30%) is usually not a serious drawback. The possibility ofa considerable systematic error is a more important threat to the relevance of exposure data. The chances are good that large systematic errors have not been introduced in present day deter- minations since the methods have improved greatly during the past decade. 3.5 Random Errors It is a well known fact that if the values of the independent variable in e.g. a regression analysis are encumbered by random errors, the regression coefficient found will be smaller than the regression coellicient determined from values of' the independent variable without errors [141. Random errors in measuring an exposure variable therefore tends to bias the slope of an exposure response line towards zero [9]. In practice this effect is usually unimportant in comparison with other biases discussed here, even in those rare cases where enough data are avail- able to make a regression analysis of exposure vs response.

s Table 3. Result of interlaboratory calibration (181. Samples on charcoal tubes corresponding to S I of air with a concentration in the range 5 times the permissible lin:it to one-fifth of the lintit, s=the standard deviation, r=the range (lowest and highest valucs) Sub- Set I Set 2 Set 4 Set 5 st:mce Number of partic- ipatins labs s'!o r'% Number of partic- ipating labs s% ro,o Number of partic- ipating labs s% 1°u Number s"/o of partic- ipating labs rlu/o StVrene 14 11 37-137 2-1 12 77-310 29 12 55-299 - - - Tri 9 11 82-181 - - - - - - 26 6 77-178 Xylene 13 ll 31-134' 24 12 67-127 30 8 8-264 25 9 69-130 •' One laboratory reportcd concentrations about 10 times too high 9S09VSSzoz