Data needs for occupational epidemiologic studies†‡ Viewpoint J. Environ. Monit., 1999, 1 75N Data needs in an epidemiologic study can appear to be substantial in light of the other responsibilities of an industrial hygienist. Many of the data needed for this type of investigation, however, are already collected for other exposure assessment purposes. To increase understanding of this concept, the data needs for the major purposes for conducting an exposure assessment are identified.The purposes include determining compliance; implementing industrial hygiene programs, such as personal protective and respiratory equipment, hazard communication training, and medical surveillance; investigating health complaints and worker concerns; investigating tasks or engineering control eVectiveness; investigating toxic tort or worker compensation claims; and conducting epidemiologic studies.A comprehensive exposure assessment system is then described that incorporates the data needs for all these purposes, including epidemiologic studies. The data needs of epidemiologic studies and how the data are used are then described and illustrated with examples taken from published epidemiologic studies.Introduction In the training of industrial hygienists, little emphasis is generally placed on the data needs of epidemiologic studies, yet such data are very important because they provide the basis for accurate exposure assessments. Inaccurate assessments can aVect interpretation of the study results: overestimating exposure levels will underestimate the observed risk to the disease and underestimating exposures will overestimate the observed risk.If the assessments are accurate, however, they should result in better occupational exposure limits because they accurately describe the levels at which health risks occurred. This means that occupational limits set from these data will be low enough to protect employees and yet high enough so that financial resources are not wasted on controlling agents for which protection is not needed.If substantial representative monitoring data existed for all jobs for all situations, the data needs of an epidemiologic study would be negligible, because the monitoring data would reflect true exposures. Since, however, monitoring data are generally limited, exposures must be estimated using non measurement types of data.This report will describe the types of data that are used for epidemiologic studies that can be collected by industrial hygienists in performing their normal duties. First, however, the various types of current exposure assessment eVorts and their data collection needs will be identified to place data collection for epidemiologic studies in its proper perspective.A system that is being used to assess current exposures is then described that integrates the data needs for both current and historical assessments. The data needs of epidemiologic studies are then described with examples from epidemiologic studies to illustrate how the data can be used. Data needs of current assessments In general the reasons for assessing current exposures is to protect workers, inform them of the hazards to which they are exposed, determine compliance with regulatory, internal or consensus standards and follow policy and professional practices.Specific reasons for conducting current assessments are included below. Determining compliance with an occupational limit Compliance determinations are taken here to mean compliance with a governmental regulation or with an internal company or consensus standard.Data collected for compliance purposes are usually limited to what is necessary to ensure that the highest exposed workers and the highest exposure scenarios have been identified. Implementation of an industrial hygiene program The purpose of this type of assessment is to compare workers’ exposures to an occupational limit to determine whether implementation of some type of program, such as use of respirators, hazard communication training or medical surveillance, is necessary. Each program can have one or several levels that trigger the need for an action and the levels may diVer with the program.Data needs are similar to those needed for compliance evaluations.Task/source investigation This type of investigation is performed when analyzing exposures or sources of emissions for purposes of control evaluation. Investigation of health complaints and problems When the cause of the complaint or problem is not known, this type of investigation can require a more comprehensive eVort than other assessments and thus more types of data.In addition to epidemiologic studies, historical assessments may also be done (see overleaf ). Worker compensation or toxic tort cases A toxic tort case is a legal action to obtain compensation beyond worker compensation benefits for illness or disease arising from occupational exposures. More information is usually collected for both toxic tort or worker compensation cases than for current assessments because of the need to meet legal requirements and to place the exposure situation within the context of other lower and higher exposed workers at the worksite.Although each of these assessments has slightly diVerent data needs, there is much overlap in the types of data that †Presented at AIRMON ’99, Geilo, Norway, February 10–14, 1999.‡US Government Copyright.76N J. Environ. Monit., 1999, 1 Viewpoint Table 1 Data needs of various types of exposure assessments Type of data Type of exposure assessment Compliance Industrial Task/source Health Toxic Epidemiologic health complaints tort studies program Job +a + + + + + Dept. title + + — (+) (+) + Date + + (+) + + + Agent + + + + + + Measurement result + + (+) + + + Type of measurement + + + + (+) (+) Tasks (+) (+) + + + + Frequency and duration (+) + + + + + Location (+) (+) + + + (+) Respirator, PPE use (+) (+) (+) + + + Equipment (+) (+) + (+) (+) (+) Level of controls (+) (+) + (+) (+) + Process description (+) (+) (+) (+) + + Source — — + (+) (+) (+) Work practices — — + (+) (+) (+) a+=Usually collected; (+)=often collected; —=usually not collected.are collected (Table 1) and many of these data are collected informally but are simply not documented. Thus, if the other five types of assessments are being made, the basic data for an epidemiologic study are already being collected. The diVerence, therefore, between the other types of assessments and assessments for epidemiologic studies is not the type of data, but rather the extent of the coverage (i.e., the number of jobs or workers).The coverage of jobs or workers is generally limited when only one agent is of concern, but as the number of agents under scrutiny increases the number of jobs or workers likely to be covered in the data collection process increases. Epidemiologic studies, in contrast, usually cover all (or all exposed) workers to one or more agents.The challenge then is to develop a data base that retains in an organized fashion all the data collected for all the assessments routinely made for all agents being assessed. One could envision this data base, for example, as a spread sheet that has the types of data on the horizontal axis and the jobs at a work site on the vertical axis. As more information is collected, the job-data cells are filled.Over time, more of the data base would be completed and less eVort would be needed to collect new data. An alternative to developing this data base in an ad hoc fashion is to take a systematic approach to evaluating all exposure scenarios. Such an approach has been taken by a major US chemical company.1 The approach used follows the same decision-making process that is followed in epidemiologic studies assessing historical exposures.Process for assessing current and historical exposures There are four basic steps in assessing exposures: identifying the agents, identifying the exposure metric, developing exposure groups and assessing exposures. Identifying the agents Although compliance needs often drive what exposures are monitored for current assessments, all agents in the workplace should be identified to ensure that no hazardous situations are missed.Agents should include raw materials, support materials (e.g., solvents, catalysts, etc.), intermediates, by-products (both intentional and accidental) and final products. Typically, when performing agent identification, the process, location and job titles coming into contact with each agent are implicitly identified.It is suggested that this identification be made explicit. Identifying the exposure metric Eight-hour time-weighted averages (TWA8) are usually estimated, but other metrics such as peak or dermal exposures must also be considered in some environments. Retention of some statistic(s) that describes the distribution of the exposure is an important component of the assessment approach.Means and standard deviations of the exposures are generally calculated, but other metrics, such as the 90th percentile can be used.1 Developing exposure groups It is useful to group workers for whom exposures can be predicted outside of the frequency and duration of the exposure. This approach is diVerent from that of the homogenous exposure group,2 which relies on statistical evaluations of monitoring data and has often been taken to mean a collection of people who have the same job title and the same TWA8.It has been shown, however, that this definition is not as straightforward as many investigators originally had thought, because workers with the same job title often have very diVerent exposures.3 One can think of several reasons for this. One is that the workers in the same exposure group may be performing diVerent tasks, may be performing the same tasks in diVerent locations with diVerent process or control equipment or have diVerent frequencies or durations of tasks. Another reason for the large variability may be that they may make similar products but work in diVerent locations with diVerent processes.Another source of variability is the duration and frequencies of the tasks. Our definition of exposure group in this report is more like the recently used term of ‘similar exposure group’, i.e., ‘having the same general exposure profile for the agent(s) being studiedJ. Environ. Monit., 1999, 1 77N Viewpoint Table 2 Example of estimating exposure to chlorine in a chloralkaline plant for a laboratory technician using a model Step 1.Data collection from interviews— Task/exposure group Level of control Frequency Duration Analyze samples Local exhaust; PPE 1 per week 2 h Step 2. Identify hazard of chlorine from toxicity data on scale of 1–6— Chlorine estimated to be a hazard of 6.Step 3. Estimate exposure level Step 3a. Estimate weight for the controls (entire table not presented)— Weight Description of control 0 Closed system: no potential for release to work area 1 Closed system: potential for release at identified points; eVective engineering controls in place at open points 2... Open system: eVective engineering controls in place... Step 3b. Estimate the weight for frequency and duration (entire table not presented)— Weight Description of frequency and duration for 8 h shift 1 <1 per month or <5 min per d 2 At least 1 per month for 5 min–1 h per d 3... 1–2 h per day Step 3c. Calculate score— Score=6 for chlorine hazard ×2 for control ×2 for frequency and duration=24 Step 3d. Compare score to Occupational Exposure L imit (entire table not presented)— Score Exposure level (range) compared to occupational exposure limit <20 <0.10 20–40 0.10=<0.25 41–80 0.25-<0.50...Conclusion— The exposure level to chlorine for this exposure group is 0.10–0.25 the occupational exposure limit because of the similarity and frequency of the tasks they perform, the materials and processes with which they work and the similar way they perform the tasks’.4 Thus, with this definition in mind it may be necessary to break a job into several smaller groups: for example, welders of mild steel vs.chromium or machinists operating drills with and without enclosures. One can think of several other variables or determinants of exposure that help distinguish between groups of workers. Determinants used by one employer in the US to group workers include: area or geographic location, job classification (i.e., title), job assignment, department, task, craft, product, process, batch or lot, project, container, production unit and ambient air (i.e., cross contamination).1 Not all of these determinants are used for every job being assessed in that company and there may be others that are more appropriate to other workers.Other determinants that have been found in the published literature include characteristics of the substance (fluid type), facility (ventilation, size or type of equipment, production rates), and individual or job (distance to operations, frequency of task and task duration) and meteorological conditions (humidity, wind speed).1 Assessing exposures This step has traditionally been accomplished by monitoring employees,2,4,5 but this is a very resource-intensive approach.An alternative to monitoring is modeling. In the area of current exposure assessment, modeling has generally been described in terms of mass balance,2,4 but has been used in few other applications. In contrast, when assessing historical exposures, monitoring is obviously not an option, so some type modeling (whether it is explicit or not) is generally performed.Generally these models have been based on a deterministic approach, rather than mass balance approach.1 In such models important determinants of exposure are identified a priori and weights identified from measurement data or from the published literature or other information are assigned to the determinants.This approach can also be used when assessing current exposures and can substantially reduce the amount of monitoring necessary. For either current or historical assessments, models can be used to develop scores that can then be compared to an occupational exposure limit to indicate a range of exposures (e.g., <0.25–0.75 times the limit) or they can be used to develop point exposure estimates in units of industrial hygiene measurements (e.g., ppm).The same determinants listed above to identify exposure groups can be used in estimation models. In developing exposure groups and estimating exposures there are two diVerent approaches that can be followed. One is to base the exposure group on workers, in which case the exposure of the group is the same as the exposure for the worker.Such an approach may be appropriate when the exposures, tasks, duration, etc., are homogeneous for the full workshift for all workers in the group. When this is not the case, however, it may be easier to view the group not as comprising individual workers but rather78N J. Environ. Monit., 1999, 1 Viewpoint comprising combinations of determinants (such as welding without ventilation on stainless steel tanks for 3 h and oiling equipment for 5 h).The limitation with this approach, however, is that there can be many diVerent variables aVecting an individual’s exposure. As a result, after all the determinants have been identified for all workers, there may be only one worker comprising an exposure group.The way to resolve this rather substantial limitation is, then, to consider the exposure group as a set of one or more tasks with unique exposure determinants outside of frequency and duration (in the above example, a group welding without ventilation on stainless steel tanks and a group oiling equipment). Once the exposure for the exposure group is determined, the linkage to workers can be made.The exposure level of the individual is calculated by weighing the exposure level by the time (frequency and duration) the individual spends in all groups with that agent. The various exposure groups and the frequency and duration values for all agents can be described as a worker’s exposure profile. An example of a model used to estimate current exposures has been described.1 The company using the model has over 35 work sites in the US with 10 000 workers and 50 000 chemical agents.To estimate exposure levels, the determinants of frequency, duration of exposure and level of control are identified. Each determinant is weighted based on predefined criteria (Table 2). Also included in the algorithm is a relative score for the hazard that accounts for how likely the agent will exceed its occupational exposure limit compared to other agents. The determinant and hazard scores are multiplied to derive the exposure score for that exposure group.To derive the worker’s exposure to an agent, the exposure scores of each group with an exposure to that agent are summed across all exposure groups with that agent.This score is then compared to a ranking using the 90th percentile of an occupational exposure limit (<10%, 10–24%, 25–49%,50–100%, >100% of the occupational exposure limit). Using an algorithm to estimate workers’ exposures does not mean that monitoring is not necessary. It is necessary on a regular basis to ensure that the estimation procedure is valid. A random sample of exposure groups stratified by the exposure level was monitored by the company developing this system.When the estimates were compared to monitoring data, 65% of the estimates were categorized correctly and 95% were within one exposure category, that being the more protective category. Other reasons to monitor include to investigate the eVectiveness of controls, to respond to worker concerns and to comply with particular occupational standards that mandate monitoring.This process of assessing exposures has significant advantages compared to assessing exposures through monitoring. For example, in the system described above, to monitor all possible exposure scenarios (assuming five samples per scenario) to all possible chemical agents in the company was estimated to cost 37.5 million dollars and take 1820 person years to collect the 1 250 000 measurements necessary to assess all the exposure groups.1 In contrast, using the system described here, 750 exposure groups were developed and assessed for all 50 000 agents in just two personyears.Another advantage is that once exposure estimates are developed for all agents for all workers, with a good management system organizing and retaining this information, it becomes quite simple to identify which workers are out of compliance, which require medical surveillance, which require hazard communication training, etc.This type of program also eliminates much duplication of eVort because most of the data identified in Table 1 usually does not change often and the system is easily updated by adding any monitoring results and copying the unchanged information.In addition, the eVect of changes in the work place on exposure groups can be easily assessed and added to the data base with minimal documentation. Finally, this type of approach ensures collection of the data needed for epidemiologic studies. Data needs of epidemiologic studies Most epidemiologic studies have been described as following one of two procedures to estimate exposures.6 In one, means of monitoring data are calculated.Jobs not monitored are assumed to have the same concentrations as those monitored. In the second procedure a description is provided indicating that measurements and other information on changes in the worksite were used, but exactly how this was done is usually missing.Reports in the literature have described data needs for epidemiologic studies, but have not generally described how the data should be used.7,8 The data needs for an epidemiologic study and how the data can be used is provided below, generally in the context of a recent epidemiologic study of acrylonitrile workers. Standardized job and department titles and dates First and foremost of the data necessary for epidemiologic studies is the jobs, departments and dates of the study subjects.Department title is important because many employers use the same job title (e.g., operator) for workers across diVerent departments, and department names can distinguish between tasks, locations, agents and other important exposure determinants.Dates are important because of changes in processes and exposures over time. The usefulness of job titles, departments and dates without monitoring data, however, is generally limited. Historically many investigators only analyzed health risks by job or department title. Although successful in many cases in finding excess risks, it has been suggested that the risks were found only because the health risks were large.9 Today, we are less likely to find risks of such magnitude and finding smaller risks using such limited data is also less likely.The information from the work histories should be standardized, so that the same terms be used for all members of an exposure group both in the work histories and in the industrial hygiene records.Without such a link the usefulness of the monitoring data can be limited. For example, in a mortality study of aircraft maintenance workers, the monitoring data were associated with job titles and shop names (i.e., the term used to approximate department).10 The personnel records did not, however, identify shop name, but rather department codes (an organizational level diVerent from shops).Some department codes could be identified with a shop, but many others could not. The monitoring data indicated diVerences in exposure levels across the shops. Because of the inability to link the shops with departments and because the patterns of exposure (i.e., frequent, infrequent, continuous or intermittent) were fairly consistent for a job title across departments, health risks were evaluated by pattern of exposure rather than by exposure level.No relationship was found with trichloroethylene, the exposure ofJ. Environ. Monit., 1999, 1 79N interest in this study, and any cause of death.11 Whether this lack of an association was because trichloroethylene truly does not increase the risk of dying or because there was so much misclassification that the association was missed could not be determined.In an acrylonitrile study the job and department titles identified in the work history records were not standardized.12 Several person-years were spent collapsing the 127 000 job/ department/plant titles (that included words like operator A or oper A) into 3600 unique job/department/plant titles. There were other problems with the work history records.Several of the plants identified in the personnel records cost accounting codes, which indicated which department paid for the time of the person, rather than the actual department titles. Historical records linked most of the codes to the departments but because the codes were on paper and not organized by code number, the linkage task was tedious and some records were missing.In another plant the personnel records identified jobs with the title of ‘operator’. The monitoring data, however, indicated specific job assignments (e.g., control room, reactor, and purification operators) with substantially diVerent exposure levels. Two long-term workers were therefore interviewed about the type of assignment each study subject with this job title had.This allowed the investigators to develop more accurate exposure estimates than if the mean of the monitoring results of all operators had been used to represent the exposure. The industrial hygienist has little control over the personnel records. It is therefore important to identify in the industrial hygiene records the terms used in the personnel records or to develop a dictionary to link the terms in both records.In addition, it is all too easy to abbreviate job titles when collecting data (whether in the work history files or for industrial hygiene purposes). Care must be taken to ensure the abbreviations are standardized. An alternative to using job titles would be the use of codes. Monitoring data The second most important type of information is probably monitoring data and the documentation that typically accompanies the data. Documentation should include date, job title, person, department, sampling/analytical those occurring under typical operating conditions.When the comments accompanying the monitoring data were reviewed, it was found that very often the unusual condition was identified as having been due to a spill, a process upset, shut-down of equipment, etc.Long-term workers were interviewed to identify the frequency with which these conditions occurred and these frequency estimates were used to weight the diVerent types of measurements in the calculation of the mean. For example, a measurement taken during spills that occurred once a month was given a weight of 0.03 (12/365 days) in the calculation of the mean.If no other unusual conditions were designated, the measurements identified as having been taken under typical conditions were given the weight of 0.97 (353/365 days). In cases where workers could not estimate the frequency of the occurrences these measurements in the mean calculations without weighting. This was likely to have biased the means but it is not clear to what extent or which direction.It obviously would have been much better to have had the frequency estimates of these types of occurrences identified at the time of the occurrence, rather than 10 years later. In seven of the eight plants in that study personal monitoring data were available back to about 1978, but before that time no monitoring data were available.In one plant, however, area measurements were available back to 1960. Job descriptions that contained estimates of time spent at diVerent locations were used with any corresponding area measurements to estimate the full-shift exposures. Because area measurements may not represent personal exposures the estimates were indicated as being of low confidence and these were dropped in one of the epidemiologic analyses to explore whether they changed the observed disease risks.Documentation of the sampling and analytic method is important because diVerent methods have diVerent validity and reliability performances. Several investigators have reported adjusting measurements of silica and asbestos when combining measurement data from diVerent methods (e.g., Rice et al.and Dement et al.15,16). The plants in the acrylonitrile study also used several diVerent sampling and analytic techniques. The workers from all the plants were to be combined into a single epidemiologic analysis so that it was technique, duration of the sample, respirator use, typical or atypical conditions, type of sample (area or personal, full-shift, peak or short-term exposure level ), the purpose of the monitoring, the work shift and any comments that increase understanding of the data.Means of monitoring data are generally used as the exposure estimates in epidemiologic studies. There is always concern, however, when calculating a mean in the absence of good documentation, that the mean may not reflect the true long-term average exposure.This could occur for a number of reasons. The monitoring data could have been done for diVerent periods of time, reflecting diVerent components of an exposure (e.g., diVerent tasks). The monitoring could have been done under diVerent seasons reflecting diVerent meteorological or air flow conditions. It could have been done under diVerent operating conditions, i.e., normal operating conditions, scheduled upsets, such as reactor cleanout between diVerent products, or unscheduled upsets, such as spills.If the measurements reflecting these diVering conditions were taken at the same frequency with which the conditions occurred, unweighted means are appropriate. If, however, measurement of these conditions was taken more or less frequently than the conditions actually occurred, calculation of an unweighted mean could result in a biased exposure estimate.The bias could be either upward or downward depending on the conditions being monitored. Excluding measurements of unusual conditions from a mean, however, is also not appropriate because all of the measurements do reflect actual exposures received. In the study of acrylonitrile workers the statistical diVerences were evaluated among means of longer duration (>360 min) and those of shorter duration (<60, 60–119, 120–239 and 240–359 min), between means taken in the winter and in the summer, and between means representing typical and unusual operating conditions.13 There was little diVerence in the means based on the duration of the measurement or the season, although for the latter, the measurements were much more variable in the summer than in the winter.(Other investigators, however, have found diVerences due to temperature.14) The occurrence of unusual conditions, however, resulted in higher means than Viewpoint80N J. Environ. Monit., 1999, 1 Viewpoint crucial that the measurement results be on the same measurement scale.For example, combining measurements from one company that used a sampling or analytic technique that increased the true exposure by 25% with measurements from an unbiased method could cause an exposure–response relationship to be missed. One option would be to not use the biased data; usually, however, data are usually too few to exclude some of them. Monitoring was therefore conducted by the study investigators in all the plants using the same sampling and analytic method and protocol to compare with the company measurement data.17 Some diVerences were found by method and by plant but in general, the diVerences were small.One plant, however, used a method that diVered from the other plants and did not compare well with the investigators’ data.The company retained the validation study it had performed many years before. Because there was nothing that indicated incorrect quality control procedures, the data from this plant were used at face value. Tasks or job descriptions Because tasks provide a major source of information on exposure, it is useful to understand what tasks were performed to place the results of the monitoring in their proper context.This procedure is particularly important when the number of data are limited, because a mean based on small numbers (e.g.,<5) may be biased away from the true mean. Job descriptions are a source of task information. Understanding the tasks is even more important when monitoring data are not available. In this case, knowing what the tasks were, where they were performed, under what ventilation configuration and at what frequency and duration they were performed helps the investigator identify what monitoring data (such as those on another job) best represents the true exposures of the unmonitored job.Alternatively, a measurement mean of one job may need to be adjusted to estimate the exposure of a second unmonitored job.Knowing the exposure profiles of both jobs (i.e., the exposure determinants and weights) increases the likelihood of identifying the diVerences between the jobs, thereby increasing the accuracy of the modified estimate.1 As indicated above, job descriptions with estimates of time at diVerent locations were used with area measurements when no personal monitoring data were available in one plant in the study of acrylonitrile workers.Peak exposures are often of interest because they can provide information on acute or chronic hazards.18 In the acrylonitrile study there were too few data measuring peak exposures to use them for all but a few jobs.1 Job descriptions were therefore reviewed to identify which jobs were performing tasks likely to have peaks.The frequency and the highest reasonable peak based on the TWA8 estimate were estimated. These estimates were used in the epidemiologic analysis to determine if peaks were associated with mortality. A significant source of exposure for some agents can be received through the skin or can be ingested, yet for most substances there is no information on these routes of exposure. As a minimum, in an epidemiologic study an indication should be made as to whether exposure through one of the routes may occur, and if so, how frequently.In the acrylonitrile study, because no data were available on dermal exposures, job descriptions were used to develop an estimate for the frequency of occurrence.12 Using this information with the concentration of acrylonitrile in the liquid being handled, a dermal score was used in the epidemiologic analysis as a surrogate for dermal exposure.Changes in the workplace In many epidemiologic studies monitoring was not conducted before the mid-1970s; yet exposure estimates may be required for many years before this time. In such cases, a mean is oftentimes calculated for the earliest year for which monitoring data are available, and then used as a baseline estimate for every year going back through time until a change is found that is thought to have aVected exposure levels for that job. The eVect of that change is estimated and used to develop the estimate which is assigned until the next change occurred.This process continues to the beginning of the study. Changes that aVect exposure levels can include changes in engineering controls, such as enclosures or local exhaust, work practices (e.g., going from manual charging to automatic charging), process changes (e.g., equipment, operating conditions and products), production rates and protective equipment use.Identifying the impact of that change is, however, diYcult. In a study of machinists, measurement means were used to determine diVerences in exposures after installation of enclosures on grinders and other metal machines.19 Using measurements was also the approach taken in the acrylonitrile study where it was possible, but for most changes monitoring data were not available.1 The eVect of the change was estimated based on personal experience, historical documents and interviews, but having measurement data describing the changes would have increased confidence in the estimates.Frequency and duration of exposure Full-shift monitoring results may represent the exposure level over a day, but they, by themselves, provide no information on the number of days a person holding the job is exposed. Yet, epidemiologic studies of chronic disease such as cancer generally focus on longterm exposures.In the acrylonitrile study, monitoring data were available for engineers, but the engineers in these plants often spent only one or two days in the operating unit and spent the rest of their time in an unexposed oYce.12 Means from these data were therefore adjusted by weighting the mean by the number of days in the acrylonitrile unit over a year, which was obtained from interviews.For other estimates, frequency and duration of exposure were used with area measurements to estimate exposures when personal measurements did not exist. Frequency of dermal contact was used to develop a dermal score (see below). Process descriptions Information on process description can be used to place measurement data in context and to characterize jobs with the same title diVerently.In the acrylonitrile study, such information was used to estimate the dermal exposure by identifying the concentration of acrylonitrile throughout the process to which workers may have had dermal contact.12 Use of respiratory and personal protective equipment Because respirators and personal protective equipment can aVect exposures, their frequency of use and their eVectiveness should be consideredJ.Environ. Monit., 1999, 1 81N Viewpoint when estimating exposure levels. In the study of acrylonitrile workers, full-time use of respirators was mandatory in only a few jobs.12 For those jobs two exposure estimates were calculated: one without regard to respirator use and one where the estimate was lowered to account for respirator use.The eVectiveness of gloves provided to the workers was not documented in the company records, so that gloves were not used to modify the dermal estimates. Odor and health eVects Although odor and health eVects suggest exposure levels, actually using them to estimate an exposure level is problematic. There is a large amount of variability in odor perception because it can be aVected by age, smoking, the presence of other chemicals and other factors. Moreover, perception of odor only suggests a minimum level, not an actual level, because the odor may not change at higher concentrations.The use of health eVects presents similar problems. In the acrylonitrile study, this type of information was, for the most part, not used.12 For a few jobs with no monitoring data and for which there was no other way to estimate an exposure level, the odor threshold of acrylonitrile was used as the estimate.Quality control and laboratory reports Quality control reports may indicate the presence of a contaminant (e.g., asbestos in talc) and can indicate what exposures occurred (and how frequently) to quality control staV. Worker compensation reports These reports often provide exposure information on the process, tasks or exposure levels.In a study of aircraft maintenance workers, these reports were found to contain much useful information on the exposure environment including job titles, ventilation controls, frequency and duration of exposure and measurement results.10 What made these particularly useful is that the information was collected closer to the time of exposure than epidemiologic data collection eVorts, so that confidence in the validity of this information was greater than information collected at the time of the study.Medical records Medical records may provide information similar to what was described earlier for assessing exposures for worker complaints or concerns.They may contain biological monitoring results that can be used to estimate airborne exposure levels or used directly as the exposure metric. For example, blood lead levels were used as the exposure metric in a mortality study of workers in a variety of lead-using industries.20 Medical records may also provide information on nonoccupational factors, such as smoking. Unfortunately for exposure assessment purposes, the industrial hygienist may not be able to get access to the records.It may be possible, however, to have the information abstracted by medical personnel with personal identifying information deleted. Other data Other production-related data that may be useful include production records, inventory records and shipping and receiving reports.These can indicate changes in the workplace that may aVect exposure levels. Seniority lists may useful if personnel records are unsatisfactory. Plant layouts can identify where measurements were taken and engineering records can be used to identify controls.16 Health and safety committee reports can also provide information on the workplace. Discussion In this report, examples are described of how historical information was used in an epidemiologic study.Every study is diVerent, having diVerent goals and diVerent information available. The more information available the better the exposure estimates are likely to be and therefore the better reflection of truth the study results are likely to be. If any of the data that were described above are not available when assessing exposures for an epidemiologic study, interviews of long-term workers can be used to supplement the information.For example, in the acrylonitrile study, the title of operator was used in the personnel records to represent jobs of substantially diVerent exposures. Two long-term workers were therefore asked to independently identify which job assignment 90 individuals with this job title worked.12 They were able to identify the jobs of 90% of these workers.Of these, they both knew 60 workers and for these 95% of the job assignments were in agreement. In spite of these encouraging results, it would have been much better to have had the actual information recorded when the assignments were made. It took time to conduct this eVort. Furthermore, when information is derived solely from workers’ memory it is subject to error.There are several reasons for memory error.Workers can forget particular events or dates when the events took place. Error can also arise fromthe desire people often have to report what they think the investigator wants to hear. For example, it is thought that workers typically emphasize ‘bad’ working conditions and minimize normal conditions, although such error can be reduced by careful probing.Another error arises from workers’ perceptions, which may not be based in reality and therefore can distort the evaluation of the workplace. For example, some workers lose their sense of smell with some chemicals andmayminimize the frequency of high exposures in the workplace, whereas others who are more sensitive to odors may emphasize the ‘high’ exposures.Another type of error is based on a selectivememory: the importance of some events may be overestimated and others may be underestimated. For example, incidences of spills may be remembered with a greater frequency than what really occurred. For these reasons, it is best to collect information on exposure determinants at the time they exist.Although collection of all these data appears overwhelming, as suggested in Table 1, much of it is probably being collected already but is just not being documented. Using a systematic evaluation of the work place with an assessment systemsuch as was described here can make the process more manageable. A laptop computer that can be used to enter data as they are being collected can facilitate the data collection process.Finally, retention of the data in a computer software data base (e.g., see Stewart et al.21 1992) can allow easy retrieval and make the assessment process more eVective. Good data, without good assessment techniques, of course, have limited utility. It is crucial therefore that the industrial hygienist use as rigorous, careful and documented assessment techniques as possible.One system was described here. A review of the literature on exposure assessment and on82N J. Environ. Monit., 1999, 1 Viewpoint epidemiologic studies6 may suggest others. Regardless of the approach used, the techniques should be validated wherever possible. References 1 P. A. Stewart and M. Stenzel.Exposure assessment in the occupational setting, submitted. 2 N. C. Hawkins, S. K. Norwood and J. C. Rock, in A Strategy for Occupational Exposure Assessment, American Industrial Hygiene Association, Akron, OH, 1991. 3 S. M. Rappaport, Selection of the measures of exposure for epidemiology studies. Appl. Occup. Environ. Hyg., 1991, 6, 448. 4 J. Damiano and J. R. Mulhausen, in A Strategy for Assessing and Managing Occupational Exposures, American Industrial Hygiene Association, Akron, OH, 2nd edn., 1998. 5 Anon., Workplace atmospheres— Guidance of the assessment of exposure by inhalation to chemical agents for comparison with limit values and measurement strategy, European Committee for Standardization, European Standard EN689, Brussels, Belgium, 1995. 6 P. A. Stewart, P. S. J. Lees and M. Francis, Scand. J. Work Environ. Health, 1996, 22, 405. 7 P. A. Stewart, A. Blair, M. Dosemeci and M. Gomez, Appl. Occup. Environ. Hyg., 1991, 6, 280. et al, Evaluation of side-by-side pairs of acrylonitrile personal air samples collected using diVerent sampling techniques, submitted. 18 S. R. Woskie, P. Shen, E. A. Eisen, M. H. Finkel, T. J. Smith, J. R. Smith and D. H. Wegman, Am. Ind. Hyg. Assoc. J., 1994, 55, 207. 19 S. R.Woskie, T. J. Smith, S. K. Hammond and M. H. Hallock, Appl. Occup. Environ. Hyg., 1994, 9, 612. 20 A. Anttila, P. Heikkila, E. Nykyri, T. Kauppinen, E. Pukkala, S. Hernberg and K. Hemminki, J. Occup. Environ. Med., 1996, 38, 131. 21 P. A. Stewart, D. Lemanski, D. White, J. Zey, R. F. Herrick, M. Masters, J. Rayner, M. Dosemeci, M. Gomez and L. Pottern, Appl. Occup. Environ. Hyg., 1992, 7, 820. Patricia Stewart National Cancer Institute EPS 810 Bethesda MD 20892 USA Mark Stenzel OxyChem Dallas TX 75244 USA 8 H. Checkoway, J. M. Dement, D. P. Fowler, R. L. Harris, S. H. Lamm and T. J. Smith, Am. Ind. Hyg. Assoc. J., 1987, 48, 515. 9 D. Wegman, in Exposure assessment for epidemiology and hazard control, ed. S. M. Rappaport and T. J. Smith, Lewis Publishers, Chelsea, MI, 1991, pp. 159–74. 10 P. A. Stewart, J. S. Lee, D. E. Marano, R. Spirtas, C. D. Forbes and A. Blair, Br. J. Ind. Med., 1991, 48, 531. 11 R. Spirtas, P. A. Stewart, J. S. Lee, D. E. Marano, C. D. Forbes, D. J. Grauman, H. M. Pettigrew, A. Blair, R. N. Hoover and J. L. Cohen. Br. J. Ind. Med., 1991, 48, 515. 12 P. A. Stewart, D. Zaebst, J. N. Zey, R. F. Herrick, M. Dosemeci, R. Hornung, T. Bloom, L. Pottern, B. Miller and A. Blair, Scand. J. Work Environ. Health, 1998, 24, suppl. 2, 42. 13 P. A. Stewart, P. S. J. Lees, A. Correa Villasenor and P. N. Breysse, Appl. Occup. Environ. Health, 1998, 13, 546. 14 S. R. Woskie, T. J. Smith, S. K. Hammond, M. B. B. Schenker, E. Garshick and F. E. Speizer, Am. J. Ind. Med., 1988, 13, 395. 15 C. Rice, R. L. Harris, J. C. Lumsden and M. J. Symons, Am. Ind. Hyg. Assoc. J., 1984, 45, 689. 16 J. M. Dement, R. L. Harris, M. J. Symons and C. M. Shy, Am. J. Ind. Med., 1983, 4, 399. 17 J. Zey, P. A. Stewart, R. Hornung, R. Herrick, C. A. Mueller, C. McCammon