Quantitative Assessment of Lung Cancer Risk from Diesel Exhaust Exposure in the US Trucking Industry: A Feasibility Study

ABSTRACT

The objectives of this study were to test the feasibility of identifying a population exposed to diesel exhaust in which small to moderate excesses in lung cancer could be estimated with reasonable precision and to develop a strategy to provide quantitative estimates of current and past exposures. We chose to assess the feasibility of designing an epidemiologic study based in the US trucking industry.

With cooperation of the Motor Freight Carriers Association (the trucking industry trade association) and the Inter-national Brotherhood of Teamsters (Teamsters union), 4 large unionized national trucking companies agreed to participate in the feasibility study. We obtained samples of personnel, payroll, and truck inventory records and interviewed long-term employees, record managers, ands enior management. The types of retirement records avail-able from 2 large Teamsters union pension funds were determined. A pilot questionnaire was mailed to 526employees at one terminal to obtain information on smoking behavior and job history. Short-term variations in exposure were assessed by measurement of air quality in truck cabs, loading docks, and yards in 2 large urban terminals and 4 small rural terminals. Measurements included elemental carbon (EC*) and organic carbon (OC) particles2.5 μm or smaller in diameter, and respirable particulate clusters 2.5 μm or smaller in aerodynamic diameter (PM2.5).The OC collected in high-volume area samples was further analyzed to assess the extent to which particles collected in the loading dock area came from diesel vehicles. Past studies and outside exposure databases were reviewed. 

Major determinants of exposure included an individual’s job title, terminal size, and terminal location. A gradient of exposure was identified. Smoking behavior did not differ between long-haul drivers and other workers. In 1985, the number of male union workers at the 4 companies whose job history could be characterized was 55,750, and in 1999 it was 72,666. A retrospective cohort study of workers from the cooperating trucking companies and the Teamsters union alive in 1985 with mortality assessed through 2000 would have a greater than an 80% power to detect a relative risk of lung cancer of 1.25 to 1.29 attributable to diesel exposure. Thus, epidemiologic studies can be designed to study the occurrence of lung cancer and to estimate past exposures to diesel exhaust among employees of the trucking industry

METHODS

PERSONNEL DATABASE ASSESSMENT

Company-Held Work History Records

Through visits for 1 to 2 days with senior management, personnel managers, and record managers at the 4 trucking companies, the extent of computerized records and the type and quality of information available were assessed. Samples of computerized personnel records were obtained, and the availability of paper records was determined. The availability of payroll records also was investigated. Company officials and long-term employees were interviewed both by telephone and in person to gain an understanding of the relationship between job title and job duties, and of historical changes in job-specific tasks an dhow they varied by company, terminal location, and terminal size. After each visit, study staff maintained regular communication with company personnel to obtain the information reviewed in this report. 

Teamsters Union Employment Records

Although smaller pension funds serve Teamsters in the eastern part of the United States, the Midwest and West are served by 2 large plans. The Central States Southeast and Southwest Areas Pension Fund, headquartered in Chicago,Illinois, serves 34 states, and the Western Conference ofTeamsters Pension Trust, headquartered in Edmonds,Washington, serves 13 states. We visited the headquarters of the Central States Pension Fund, met with its administrator and data management personnel, and assessed th availability and quality of work history information. Samples and descriptions of retirement records were obtained and reviewed. Discussions regarding the availability of information from the Western Conference Pension Trust were conducted by telephone, mail, and e-mail.

COMPANY DATA ON TRUCK FLEETS

The availability of computerized and paper-based truck inventory and maintenance records at each company wasa scertained from data managers and truck operations personnel, and samples of these records were reviewed. Company officials were asked to provide dates when each company first used diesel vehicles, when diesel pick-up and delivery (P&D) trucks were introduced, and when diesel forklifts were used. An overview of current and historical vehicle purchasing and retirement policies was also obtained to gauge the differences in fleet ages and characteristics between companies. The vehicle-use policy o feach company was used to assess the likelihood of older vehicles staying in service, the vehicle mix, and the location of older trucks for possible emission assessment. The following books were reviewed for information on historical fleet characteristics and vehicle-use practices, and for confirmation of dates of diesel vehicle use obtained from interviews: The Roadway Story (Cantelon and Durr 1996);Yellow in Motion: A History of Yellow Freight System,Incorporated, 2nd edition (Filgas 1971), revised edition(Filgas and Waters 1987); and Never Stand Still: The His-tory of Consolidated Freightways, Inc and CNF Transpor-tation, Inc 1929-2000 (Durr and Cantelon 1999).

COMPANY DATA ON TERMINALS 

Current addresses of terminals and information about terminal design and operations, including information on diesel fuel use and on freight loading and unloading in the dock area, were obtained from all companies. The size of each terminal was determined from the terminal codes in the personnel files, and terminals were grouped according to the number of workers assigned to them (fewer than 30,30 to 99, 100 to 199, and 200 or more workers). ArcView(Environmental Systems Research Institute 1996), a geography-based information system software, was used to overlay the terminal addresses on a US Census Bureau map with locations designated as urban or rural. This information was then used to determine the current distribution of terminals and personnel in urban and rural areas.

SMOKING AND WORK HISTORY INFORMATION AND FEASIBILITY OF COLLECTION BY QUESTIONNAIRE

The availability of information regarding cigarette smoking history in trucking company workers was investigated, as well as the feasibility of conducting a survey of current and former workers by mail. A pilot mail survey was sent to all 526 workers at a large trucking terminal inAtlanta, Georgia. Employees were initially sent a cover letter and a 4-page detailed questionnaire including questions on date of birth, height and weight, educational status, job title, smoking history, duration of work in the. trucking industry and as a Teamster, and a job matrix covering lifetime employment history. Nonrespondents received a second copy of this questionnaire with a coverletter. The third mailing to nonrespondents consisted of a cover letter and an abbreviated questionnaire (“1-page questionnaire”) with less detailed smoking questions and no job matrix. (The questionnaires are available on request as Appendix B.)

HISTORICAL DATABASES ON TRUCK FLEETS, TRAFFIC, AND AIR POLLUTION

The availability of current and historical databases containing information on factors that might contribute to diesel exhaust exposure was investigated. The feasibility and limitations of use were determined for each database to identify those that could aid in the development of exposure models.

Truck Registration and Use Data

Commercial and governmental databases containing information on vehicle registration and truck use were identified and reviewed. We looked specifically for data to characterize each company’s vehicle fleet, the distribution of trucks nationally, and factors that might influence emisssions and exposure variables.

Traffic Volume and Road Use

Databases maintained by the Federal Highway Administration and the Bureau of Transportation Statistics, both in the US Department of Transportation (DOT), were identified and reviewed to determine the nature of current and historical information on highway traffic counts and road use by different types of vehicles.

Truck Engine Specifications and Emission Factors

The availability of engine certification data was assessed by contacting engine manufacturers through the Engine Manuufacturers Association. The US Environmental ProtectionAgency (EPA) National Vehicle Fuel Emissions Laboratory also was contacted. The type and quality of vehicle and engine design specification information available from manufacturers was evaluated by review of independent publications.

Background Air Pollution

The EPA and the individual states maintain an extensive monitoring network for air pollutants across the country, and these data are available through the Aerometric Information Retrieval System (AIRS), a public-access database (http://www.epa.gov/airs/). The availability of this database and its suitability to provide historical information and variation in background particulate air pollution were reviewed. Since climatic factors such as temperature, precipitation, and wind direction can strongly influence exposure, the availability of current and historical meteorologic information from the National Oceanic and AtmosphericAdministration (NOAA) National Climate Data Center(http://www.ncdc.noaa.gov/) also was assessed.

EXPOSURE ASSESSMENT

One of the key challenges to exposure assessment is the choice of exposure markers. In this study our goal was to identify a marker that not only represents exposure to diesel exhaust, but also serves as an index of carcinogens in diesel exhaust, (ie, quantitatively related to the cancer risk). This second dimension is important because all compounds in diesel emissions are probably not carcinogenic and the carcinogenicity of the total particle may correlate poorly with the carcinogenic fraction.

A chemical mass balance method has been used to determine the relative contributions of specific emission sources to the total mass in atmospheric particulate samples (Cass et al 1984; Waalkes and Ward 1994; Schauer et al 1996; Watson et al 1998). In these tests, ambient air- borne EC appeared to be most closely related to diesel emissions (Cass et al 1984; Cass and Gray 1995; Birch andCary 1996). However, ambient air exposures are complex, with many sources of air contaminants, including EC.

Source apportionment based on details from the chemical mass balance method makes it possible to determine how much EC is coming from diesel engines. For example, diesel vehicles emit large amounts of EC, only small amounts of cholestane and related compounds, and no levoglucosan (a plant sugar), whereas gasoline vehicles emit relatively large amounts of cholestane compounds, little EC, and no levoglucosan; softwood combustion releases relatively large amounts of levoglucosan and moderate amounts of EC and cholestane. An air sample showing large amounts of EC, small amounts of cholestane and related compounds, and small amounts of levoglucosan would implicate diesel emissions as the main contributors with small amounts of gasoline vehicle emissions and softwood smoke included.

On the basis of source apportionment findings and exposure data from other investigators, we chose EC as our primary marker of diesel emissions, and OC and PM2.5 as markers to track exposures to other general air contaminants. The quality of EC as a diesel marker in occupational settings varies because proximity to the source is an important determinant of exposure intensity. For workers whose jobs bring them close to operating diesels, diesel exhaust is a major component of the air contaminants soEC would be a good marker of exposure. For workers with no nearby diesel sources, EC may be a poor marker because more than half of the EC in the ambient air may come from other combustion sources such as home heating or wood fires (Cass et al 1984; Waalkes and Ward 1994; Watson et al 1998). Combining source apportionment with use of EC as a marker compensates for the lack of a unique marker of diesel emissions.

Independent of its application as a marker for the quantity of diesel emissions in the air, the suitability of EC as a marker for human carcinogens in diesel exhaust is difficult to assess. It is not clear whether EC itself is a human carcinogen, or the organic compounds in diesel exhaust are the primary carcinogens, or EC and OC together are the agents of cancer risk. No carcinogenic pollutant is unique to diesel emissions; rather, all of the common combustion sources overlap with diesel emissions. The relative concentrations vary, but virtually all are present.

Some of the goals of the feasibility study were to determine the applicability of potential sampling methods, to verify that personal exposure monitoring is feasible, and to obtain limited data on current exposures. We used real- time and integrative sampling to obtain personal, area, and source data for PM2.5. Particulate samples were analyzed to measure concentrations of EC and OC.

Exposure Assessment Methods

Exposure Assessment Methods
Personal exposure to PM2.5 was measured by filter sam pling using the Personal Environmental Monitor (PEM) (SKC, Eighty Four PA), which has a small impactor and 37- mm Teflon filter with a pore diameter of 0.2 µm. Mass col- lected on filters was determined by gravimetric analysis using an analytic balance (Micro-Gravimetric M5, Mettler Instruments Corp, Hightstown NJ) and Pallflex Teflon 37-mm fiber filters (Pall Corp, East Hills NY), and Tissue-Quartz 25-mm fiber filters (URG, Chapel Hill NC). The filters were weighed after humidity equilibrium was attained (after at least 24 hours) in a chamber. At the end of sampling, the filter was taken back to the laboratory. After humidity equilibrium was attained, the filter was reweighed to determine weight gain. A second set of PM2.5 samples was collected using a similar sampler with a Tis- sueQuartz filter. These samples were analyzed for EC and OC using the National Institute for Occupational Safety and Health (NIOSH) 5040 thermo-optical method (Birch and Cary 1996; Cassinelli and O’Connor 1998) at the laboratory of Dr James Schauer, University of Wisconsin, Madison.

Real-time measurements of PM2.5 were made with the DustTrak (TSI, Shoreview MN) using laser light scatter into detect airborne particles after passage through an impactor that removes particles larger than 2.5 µm in diameter. Concurrently with the DustTrak monitor, a Q-Trak (TSI) was used to obtain real-time data on carbon monoxide (CO), carbon dioxide (CO2), temperature, and relative humidity. The Q-Trak uses a set of physicochem- ical sensors. The sampler inlet was placed on the back of the seat, at shoulder level, next to the driver, so the sample represented the air quality in the center of the cab.

Large-volume integrative samples for detailed chemical analysis were collected using a 47-mm TissueQuartz filter sampler with a large cyclone separator (less than 2.5-µm- diameter cutoff) operated at 16.7 L/min for 8 to 12 hours. Gas chromatography/mass spectroscopy (GC/MS) was used by Dr Schauer’s laboratory to analyze these samples for a wide range of specific organic compounds (Schauer et al 1996). This analysis involved extensive sample preparation, including spiking the filters with 7 deuterated internal recovery standards, and 5 extraction steps.

Extracts were combined and reduced in volume, and finally half of the combined extract was derivatized with diazomethane to esterify organic acids. Both the derivatized and underivatized extracts were analyzed on aGC/MS (GC model 5890 and MS model 5972, Hewlett-Packard), using a capillary column 30 m × 0.25 mm in diameter (HP-1701). More than 100 compounds are quantified with a relative error of 20%. The following chemical compound groups are evaluated: n-alkanes, polycyclic aromatic hydrocarbons, substituted phenols, guaiacol and related compounds, syringol and related compounds, n-alkanoic acids, n-alkenoic acids, alkane dicarboxylic acids, aromatic carboxylic acids, resin acids, levoglucosan and other sugars, and other organic compounds. The cost and complexity of the analysis limit its application to com- posited samples from work locations where the types of sources are likely to be approximately constant.

Exposure Measurement Strategy

Testing was conducted at 2 large terminals in the Atlanta area during October 1999. In total, 45 PM2.5 personal samples (with EC and OC measurements) and 14 DustTrak personal samples were collected for forklift operators, mechanics, hostlers, long-haul drivers, local (P&D) drivers, and dockworkers directly involved in daily terminal activity. For area samples, 90 PM2.5 (with EC andOC), 7 source apportionment (later composited for anaysis), and 58 DustTrak samples were collected from the loading dock, shop, fueling lane, and clerk’s office. An upwind background sample, called the “yard sample,” was also collected at each site. Sampling methods were compared by hanging several different samplers side-by- side in different areas. Concurrent with the gravimetric monitoring, real-time monitoring of airborne particles was done with a DustTrak to obtain 1-minute averages of light- scattering respirable PM2.5. One-minute averages of CO,CO2, temperature, and relative humidity were also measured with the Q-Trak. Exposures also were assessed at 4 small rural terminals in New England by sampling in the dock area and upwind in the yard in April to May 2000 with the collection of 18 PM2.5, 14 EC and OC, and 16 DustTrak area samples.

The vast majority of personal samples and area samples were collected over 8 to 12 hours. A limited number of side-by-side personal samples also were collected by equipping a worker with 2 personal samplers to assess the relationship between URG personal measurements of EC and OC and PEM gravimetric measurements for personal sampling. The area samples were taken at similar locations in the terminal area, 5 feet from the deck level, with 4 different collecting devices (size-selective samplers): Dust-Trak, Q-Trak, personal sampler, and PEM.

Previous NIOSH Studies

In 1988 and 1989 NIOSH sponsored Health Hazard Evaluation studies by Zaebst and coworkers to characterize exposures to diesel emissions in the trucking industry (summarized in Zaebst et al 1990). A total of 327 personal and area samples were collected in 7 large truck freight terminals. These data represent a critical historical database on conditions in the late 1980s. The individual data points given in the report appendices are an important resource.
We also used the NIOSH studies to help assess other sources of exposure variability so that we could identify current and historical factors that will define differences in exposure across the industry.

STATISTICAL ANALYSIS

Smoking and Work History Questionnaire

Chi-square tests were performed to compare questionnaire respondents and nonrespondents and various parameters describing smoking behavior by job title.
Means were compared using t tests. Logistic regression models were used to identify predictors of current- smoking or ever-smoking status. Linear regression methods were used to assess predictors of average cigarette consumption.

Exposure Assessment

Results of the environmental sampling surveys were expressed as geometric mean (GM) and geometric standard deviation (GSD) because the distribution of sampling results is approximately log-normal. The results of real- time sampling from the DustTrak (plots of PM2.5 level versus time) are also presented for various locations. Analysis of variance was used to compare between-terminal variations. Although some of the limited data collected in this study was evaluated statistically, the focus was on the feasibility of monitoring exposure. The scope of the project was limited by the sponsor. As a result, the analysis of measurement error is based primarily on published data and NIOSH Health Hazard Evaluation Reports.

RESULTS

PERSONNEL

Company Databases

Job Activities

The companies invited to participate in he feasibility study are the 4 largest national, unionized, less-than-truckload freight carriers in the United States. These companies have provided shipping services for large packages and freight to industrial, commercial, and residential customers since the 1920s and 1930s. They send a P&D driver to pick up the freight, which is often on pallets. The freight is returned to the local terminal, where it is consolidated onto a trailer with other loads bound for similar parts of the country. Next, one or more long-haul drivers drive the trailer to the regional hub terminal closest to the delivery point. The trailer load is broken down into smaller loads that are either delivered to their destinations or sent to smaller local terminals from which P&D drivers deliver them to their destinations.

The major duties in the trucking industry that involve possible exposure to diesel exhaust are grouped into job categories (Table 1). Because these companies are all unionized, job duties were the same across the 4 companies with only minor differences for each job title. Our discussions indicated that major job-specific duties had not changed significantly over time for most jobs. Union rules permit workers to bid on terminal-based jobs, which are awarded on the basis of years of seniority working at a terminal. Therefore, movement between terminals is limited once workers are hired, and job categories tend to remain stable throughout each worker’s employment tenure.

Mechanics experienced the greatest change in job duties over time. Whereas in the past they performed most major engine and body repair on company trucks in house, currently they are responsible primarily for preventive maintenance, such as oil and tire changes. The division between dockworkers and P&D drivers is not always sharp. Some workers perform both duties on a day-to-day basis, as needed, and have the job title P&D/dockworker or combination worker. These workers are usually stationed at smaller terminals, and their actual job duties depend on the requirements at the terminal on a given day.

Work History 

Each company has a computerized personnel record system that contains, for each worker, name, social security number, race, sex, date of birth, date of hire, union membership status, current and previous job titles, terminal code (which can be linked to an address), and a record of when the person was actively working or on extended leave due to layoff or illness. The companies differ regarding the dates that records were computerized and the extent of the information available. Between 1979 and 1993, three of the companies computerized the complete job histories for all active employees. The company that computerized in 1993 also included the work histories for all workers employed on or after 1980 even if they were not employed when the database was established. The fourth company started its computerized database in 1972 and included all jobs held beginning in 1971. If an employee worked at another company acquired by one of these 4 companies, then the date of hire reflects that person’s start date at the original company.

Additional information is available from each company in the form of paper or microfiche records. All 4 companies maintain noncomputerized personnel records for current employees and for other employees within 3 years of termination. The original job applications, which include previous job history outside the company, are also available for these employees. Detailed paper records and job applications are available only after 1995 for one of the companies because its warehouse was destroyed in a storm. However, employee identification cards with basic identity information, job title, terminal location, union seniority date, date hired, last day worked, and reason for termination were saved and can be used to fill in some missing information.

The availability of payroll records listing hours worked for each job assignment was also investigated. These data may be useful for determining distribution of time spent between dock work and P&D for combination jobs. Archived computerized payroll records are not available for 2 of the companies. One company computerized its payroll records starting in 1972, and the other has noncomputerized payroll records from 1979 to 1995 and computerized payroll information since 1995.

The existence of information regarding cigarette smoking was also explored. Although the DOT requires drivers to pass a physical examination every 2 years, information on cigarette smoking is not collected. Furthermore, the companies do not maintain medical or other health- related records that might contain smoking data.

Although the yearly listings of job and terminal location were available in the company personnel databases, the companies did not have the in-house computer support to extract these data. Instead, 3 of the 4 companies provided computerized cross-sectional snapshots of the data for either December 31 or November 30 of 1985, 1990, 1994 or 1995 and for 1998 or 1999. Two of these companies also provided data for 1980, and data for additional dates were acquired from some companies. The fourth company pro vided a copy of its computerized personnel database of all employees on the payrolls in 1980 and every year subsequently. Only current (1999) or last job title (before retirement) and terminal location were included. Therefore, for this company the cross-sectional data were extracted and it was assumed that job title was stable throughout a worker’s tenure. All identifiers were removed before we obtained these data.

Computer Records 

Results of the computerized personnel record review were summarized for the years 1985, 1990, 1995, and 1999 (Table 2). Assuming conversion from gasoline-powered trucks to diesel-powered vehicles by 1985 (see Table 5), long-haul drivers with long-term employment in these companies would have had at least 20 years’ experience of driving diesel trucks by the time of this study. Most workers in the industry are male and white although the proportion of nonwhite employees has been increasing gradually over time. In 1985, most of the union women (89.2%) had clerical jobs, thereby limiting our ability to include them in a meaningful fashion in an epidemiologic study of the health effects of diesel exhaust. The missing indicators for union status in Table 2 are due to omissions in the data provided by one of the companies.

The nonunion workers include management and casual labor, and most of the missing date-of-birth and date-of-hire information in the databases reflects the inclusion of casual labor in the data extraction. One of the companies was missing job titles for 70% of its employees in the 1985 data extraction, whereas in 1990 this information was missing for only 6% of its workforce. Because job titles are generally stable in the industry, the 1990 data were used to fill in the missing information for 1985. In summary, extensive computerized records are available from each company, including for workers with longtime employment in the trucking industry during the time diesel trucks were primarily used.


The numbers of male unionized workers were determined for the 4 companies and for specific job titles in 1985, 1990, 1995, and 1999 (Table 3). For the company missing union status indicators, all employees with these job titles were considered to be union members and thus were included in the totals. In 1985 the number of male workers categorized as likely belonging to the Teamsters union was 55,750, and this workforce increased in later years as the industry consolidated through acquisitions.

The percentage of long-haul drivers remained relatively constant (between 24% and 27% of the workforce) between 1985 and 1999. The proportion of workers classified solely as P&D drivers fell slightly as the proportion of combination workers (P&D/dockworkers) increased. The proportion of mechanics also decreased, which was consistent with the practice of contracting the heavy repair work to outside vendors. Most of the workers listed under the job titles of management and trainee were from the company for which the union status indicator was missing and therefore may not have been union members. In other companies, some of the managers of smaller terminals and persons whose job title was listed as trainee were union members; for completeness, therefore, all workers with these job titles from this company were included in the totals. In any study using these data, the union membership of these workers could be verified by a more detailed search of each company’s personnel records.

The age distribution, mean years worked, and the third quartile of the distribution of years worked were determined for all union male workers for whom date of birth was available in the 4 companies in 1985 and 1999 (Table 4). More than half of the entire unionized workforce and of the long-haul drivers were between the ages of 35 and 54 in both years. The age distribution indicates, however, that the midpoint was slightly older in 1999 than in 1985 and that the long-haul drivers tended to be older than the rest of the population. In both 1985 and 1999, nearly 50% of the workers (those 65 years of age of older), including the long-haul drivers, had more than 20 years of employment in their current company. The distributions of age and years worked in 1990 and 1995 were similar to the distributions in 1999 (data not shown).

Teamsters Union Employment Records

Information on the numbers and vital status of relevant members of the Central States Pension Fund and theWestern Conference Pension Trust was obtained. In January 1999, 183,661 retired Teamsters were receiving bene- fits from the Central States Pension Fund, and inNovember 1999 approximately 188,000 active members were contributing to the pension funds. The 4 companies had 36,010 employees contributing to this fund at this time (19.2% of fund membership), and the 21,889 former workers from these companies made up 11.9% of the fund retirees receiving benefits. In 1999, there were 7,113 deaths among all Central States Pension Fund members receiving benefits.

In 1999, employers were contributing to the Western Conference Pension Trust for 270,744 employees, with 14,228 members (5.3%) coming from the 4 large unionized carriers. Overall, 169,667 Teamster retirees were receiving benefits from the Western Conference Pension Trust, but the proportion of retirees from the 4 large unionized carriers was not obtained for this report. Therefore, of the 72,666 active Teamsters union members in the 4 companies in 1999, 49.6% belonged to Central States Pension Fund and 19.6% belonged to Western Conference PensionTrust, for a total of 69.2% of the workers currently contributing to one of the two large Teamsters union retirement plans. Approximately 25 small funds accounted for the other 30.9% of the workers’ retirement benefits.

The 2 pension funds we studied maintain computerized records of the contributions made by each unionized trucking company on behalf of a worker, including dates and company name. In the Central States Pension Fund, the only specific job histories available are the self-reports provided by each worker when he or she completes an application at the time of retirement. The applicant lists job title and dates of service with each employer; this information is not verified against company records. In the Central StatesP ension Fund, starting in 1992, these records are available from an image retrieval system, and prior to 1992, on microfiche. No job history information is collected by theWestern Conference Pension Trust. No other specific details about job title and location are maintained by either fund, and medical information and information on personal habits, such as cigarette smoking, are not collected.
Thus, although the pension funds cannot provide many details about a participant’s job history or risk factors, they can provide information on total years of creditable service as a Teamster, beyond what is available from the current or last employer, and they provide an independent means of verifying data on work in the unionized trucking industry.

COMPANY RECORDS ON TRUCK FLEETS

Vehicle and Equipment Records

Each company has 2 computerized systems containing vehicle and equipment information: asset registers and maintenance records. The extent and availability of these records vary by company. Asset registers record the following information about every truck and forklift purchased or sold by the company: vehicle serial number, make, model, year, and engine type as well as location assignment for forklifts and P&D vehicles. The asset registers are computerized and available for all vehicles ever owned by one of the companies, from the 1980s for 2 companies, and for the last 5 years from the other. Additional information may be retrievable from paper records starting as early as 1959 in one company. Furthermore, long-time personnel in the asset and maintenance departments often are able to reconstruct much of this information from memory, because companies typically purchased only one or two types of trucks per year. The maintenance departments keep information on vehicle component changes, such as engine overhauls and air-conditioning removals. These data supplement those provided by the asset registers, and although some are computerized for some companies, they are available mainly in paper form.
An historical listing of the number and design of diesel P&D vehicles assigned to each terminal and the composition of the long-haul fleet can be constructed for each company. Long-haul vehicles are not assigned to a terminal and are used throughout a company’s system. However, the long-haul drivers are assigned to a specific home terminal, which indicates the region where they drive.

From the preliminary information provided, we constructed a timeline for the conversion to diesel long-haul vehicles, P&D trucks, and forklifts (Table 5). Each company started using long-haul diesel trucks in the 1950s, and all 4 companies had converted from gasoline long-haul vehicles by 1965. Diesel vehicles were introduced into the P&D fleets beginning in 1972. Depending on the company, the year that the P&D fleet was 100% diesel varied considerably (1980 to 1992), but the majority of combined fleets were diesel by the late 1980s. Diesel forklifts were in general use by the mid 1980s.They were then phased out between 1990 and 1996 and were replaced with propane units as part of contract negotiation with the Teamsters union in the early 1990s. According to information from one company, diesel forklifts were most likely to be used in the large distribution centers but not in the smaller local terminals. Company records can be used to identify where diesel forklifts were used.

Historical Changes in Vehicle Use

Vehicle replacement and utilization practices differ among the 4 companies. Three of the 4 companies have historically converted long-haul vehicles to P&D vehicles after 10 to 15 years of service and then used them for an additional 10 to 15 years. Each company keeps its vehicles in service for different lengths of time, and within each company this policy has changed historically. The trend is toward keeping the vehicles on the road longer. Because of these policies, the age of the fleets varies substantially between companies. For example, data from one company showed that several hundred trucks purchased in 1975 to 1979 are still in service, whereas another company keeps its vehicles for only 5 years.

Maintenance policies for all 4 companies generally call for restoring vehicles to original operating conditions rather than replacing the engines (although different engines may be installed in some of the older vehicles). Any replacement of engines is recorded and archived. The long-haul vehicles have been air conditioned for many years (for example, new vehicles purchased by one company have been air conditioned since approximately 1980). Since P&D vehicles are not air conditioned, the air conditioning is removed when a long-haul vehicle is converted to local use.

TERMINAL CHARACTERISTICS

Through analysis of annual reports, personnel records, terminal operation records, and other company sources (such as maps and company histories), the total number of terminals and the size of each terminal at each of the 4 companies starting in 1985 were determined. Current terminal addresses were obtained from all companies and historical terminal addresses from one company. Samples of terminal blueprints illustrating design features were obtained from some of the companies. We have determined that it is feasible to obtain this information on all relevant terminals.

Terminal codes and the number of employees assigned to each terminal are available from the personnel databases. In 1985 there were 2,200 work locations in the personnel data- bases; the number increased to 2,427 by 1990. Since 1990, the trend has been toward a reduction in the number of smaller terminals, and the total number of work locations decreased to 1,770 in 1995 and 1,337 in 1999. However, because some of the work locations specified in the personnel files are at the same physical address or terminal (in other words, office areas and repair shops may have unique codes), the number of total terminals is overestimated. For example, there were only 1,267 unique terminals in the address files in 1999. Therefore, historical address information (available from all companies) should be used to determine unique terminal addresses in a larger study. Of the 1,267 unique addresses in 1999, most were terminals with fewer than 30 employees (Table 6). Because background air pollution (and respirable particle levels) vary with urban or rural location, the addresses of the 1999 terminals were overlaid on a map of the United States (Figure 2) and classified as urban or rural as defined by the US Census Bureau. Of the 1999 employees, 58.5% worked in rural locations, mainly at large terminals with 100 or more employees (Table 7). Examination of air pollution data, local truck traffic volume, and number of trucks assigned may result in finer gradations of exposure potential, which will permit better resolution of any exposure effects.

An alternative source of terminal address data is the RJ Polk Company Truck Group (Southfield MI), which provides data on the number of trucks assigned to each terminal address and detailed information on each truck. These data indicated that terminals with few trucks also had small workforces in the company personnel databases. Of 1,232 registration addresses identified in September 1999 by RJ Polk Company, 626 terminals (50.8%) had 1 to 4 trucks assigned and 247 terminals (20.1%) had 5 to 9 trucks assigned.

Current records of diesel fuel use and of freight volume by terminal are available, but historical records are limited. Companywide summaries of fuel used are available for the last 3 to 10 years, depending on company policy, and are mostly in paper form. Terminal-specific information on the amount of freight moved across a dock is available for the last 3 to 12 years, and companywide summary data are available for the last 10 to 15 years, also in paper form.

SMOKING AND WORK HISTORY QUESTIONNAIRE
Response Rates

The smoking and work history questionnaire and the mailing protocol are described in the Methods section. The original sample comprised 526 subjects who worked at the same terminal. However, 2 workers had been terminated, and 14 had either missing or incorrect address information. Of the remaining 510 subjects (489 male, 21 female), 247 (48.4 %) responded after 3 mailings. In total, 175 of the 4-page questionnaires and 72 of the 1-page questionnaires were returned. The response rate among the 294 male white unionized workers was 49.7% (146 respondents). The 37.9% response rate among 145 unionized workers of other races was significantly less (P = 0.02). Based on the job title provided by the company, the response rate among 213 long-haul drivers, regardless of race, was 49.8% (Appendix A, Table A1). This rate was slightly higher than the rate
 among the other male unionized workers combined (42.0%) but not significantly so (P = 0.10). This slight difference in response rates for long-haul drivers persisted even when the rates were stratified by race. Subsequent analyses were based on the total population of 201 male white and nonwhite union workers who responded to the questionnaire. Data are presented in detail in the tables of Appendix A.

Respondent Population Description
Self-Reported Job Title Versus Company Job Title

To assess the extent that the job title listed in company personnel records can be used to determine a worker’s actual job title and duties, company job titles for these workers were compared with the job titles obtained by self-report on the mail survey (Table A2). Agreement between self-report job title and company job title for the long-haul drivers, mechanics, and hostlers was nearly perfect. The distinctions between the P&D drivers, the dock-workers, and the P&D/dockworkers were less clear. This company classifies nearly all of its P&D drivers a sP&D/dockworkers. However, a P&D/dockworker might drive a P&D truck most of the time and thus call himself a P&D driver, or he might work predominantly on the dock and call himself a dockworker. If the exact job titles available for all workers are compared, then 79.6% of the self-reports agreed with the job title in the company records. If P&D driver, dockworker, and P&D/dockworkers are considered the same job, then there was 94% agreement.

Personal Characteristics 

The mean age of the male union respondents was 51.8 ± 9.6 years. Characteristics of the long-haul drivers were compared with those of other workers combined (Table A3). The long-haul drivers were significantly older than the other workers (P < 0.0001), which is consistent with the pattern observed in the entire workforce. Fewer long-haul drivers had educational training beyond high school (P = 0.07). Body mass index, a possible indicator of dietary and exercise habits, was similar across job titles in this population.

Smoking History

Smoking behavior was examined by self-reported job title (Table A3). Of all respondents, 33 (16.4%) were current smokers, 93 (46.3%) were former smokers, and 75 (37.3%) were never smokers. The prevalence of smoking (current or ever) and the average number of cigarettes smoked per day over time were generally similar across union job categories although long-haul drivers tended to smoke more (TableA3). The numbers of workers in the job categories other than long-haul drivers were too small to allow further analyses.
When the job categories were grouped into long-haul drivers versus other jobs, these variables were not significantly different (Table A4). The long-haul drivers had also smoked longer than had other workers (P = 0.0004). However, they were also the oldest and thus had had more opportunity to smoke. This difference was no longer statistically significant after duration of smoking was adjusted for age (adjusted mean was 24.2 years of smoking for long-haul drivers and 19.8 years for all other workers, P = 0.13). In multivariate logistic regression models that included age, educational status, and body mass index, the self-reported response to “long-haul driver (yes/no)” was not a significant predictor of ever or current smoking (P = 0.59 and 0.86, respectively).

Age, educational status, and body mass index for all workers were also not significant predictors of smoking behavior in the multivariate model (P > 0.20) and were similar between smokers and nonsmokers when examined independently (Table A6). Overall, these pilot data suggest that smoking behavior is similar between the long-haul drivers and other workers, and among the other workers, and smoking behavior is independent of job title.

Work History

Information regarding job history was obtained in 2 different ways on the questionnaire: simple direct questions and a detailed job matrix. Workers were asked what year they started working in the trucking industry, what year they joined the Teamsters union, and their current job title. Calculation of years working for the current company was based on the date of hire provided by the company. Respondents also completed a job matrix listing their past job titles, whether the job was full-time or part-time, dates of service, company name, and major job duties. In analyzing the job matrix data, we assumed that a part-time job occupied 50% of a worker’s time on the job during the indicated time period.

Comparisons between number of years worked in the trucking industry, years as a Teamster, and years worked for the current company were based on the 125 male union workers for whom complete information was available. On average, workers started in the trucking industry in their mid to late 20s (Table A5), so at the time of the survey these workers had worked in the industry for more than 20 years. Approximately 4 to 5 fewer years were spent as a member of the Teamsters union, and even fewer years were spent working in their current company (Table A5). This difference in number of years worked as a Teamster and for the current company is probably a function of the terminal and company that was selected for this survey. The company had greatly expanded in recent years, and before 1995 the terminal belonged to an acquired unionized company. Almost half of the workers at this terminal had worked for the previous company, and we were able to account for this in our analyses. In this terminal, 41.8% of the workers had been employed for less than 10 years and only 15.9% had been employed for more than 20 years. This is a younger age distribution than that of the rest of the company. This company had only recently started to hire large numbers of new workers as it expanded.

Only 43 (34.4%) of these 125 workers completed the job matrix in sufficient detail to allow for meaningful analysis. Among the 20 long-haul drivers in this subgroup,
 25.4 years (± 8.9 SD) of their entire job history in trucking was spent as a long-haul driver; among 16 P&D/dock-workers, 22.1 years (± 8.0 SD) of their entire job history was in similar jobs in the trucking industry; and among 6 mechanics, 27.7 years (± 6.9 SD) of their entire work his- tory was spent as a mechanic. Therefore, when considered together, these workers had spent many years employed in the trucking industry, mostly as a Teamster in the same or a similar job.

HISTORICAL DATABASES ON TRUCK FLEETS,TRAFFIC, AND AIR POLLUTION

Many government departments and agencies maintain current and historical databases on factors that might contribute to the development of exposure models. Information that could be utilized for this purpose is also available from commercial sources. The scope and quality of information in these databases vary widely, however, and many of them are inappropriate for use in epidemiologic studies. We have included information on selected large databases as well as their level of appropriateness for use in epidemiologic studies.

Truck Registration and Use Data

The most comprehensive external truck registration database was the RJ Polk Company Truck Group, which offers the list for purchase. This database provides a detailed national listing of new truck registrations by company and address starting in the early 1990s. For several states, registration address cannot be provided because of state regulations. Truck model, cab specifications (eg, cab over engine or engine in front of cab), engine type and model, and detailed engine specifications are available. The RJPolk Company also maintains a database updated quarterly starting in the early 1990s that provides the same information on all trucks in service based on registration address and company name. Several other companies and organizations, including the American Trucking Association, Transportation Technical Services (FredericksburgVA), and Trinc Transportation Consultants (Washington DC), publish yearly aggregated data for all trucking companies, including total number of trucks.

The Truck Inventory and Use Survey database was considered to be a potential general source of information to profile trucks in service in the United States. Others have used this database to profile the operational characteristics of the US truck fleet. First developed in 1963 by theDepartment of Commerce, the database has been updated every 5 years. A survey mailed to a random sample of private and commercial truck owners is used to collect data on a single truck. Owners are asked to estimate the number of miles traveled by the vehicle, the numbers of trailers usually hauled, and the type and weight of a load. This database contains company-based or terminal-specific truck data, but it does not have information about the types of trucks used by each company because information is collected only on one truck in a fleet. Therefore, the data reported do not reflect the distribution of trucks operated by each company, and how accurately these data represent long-haul and P&D trucks driven by unionized trucking company workers is uncertain.

Traffic Volume and Road Use

The DOT maintains information on traffic counts on all of the federal and statewide highways starting in 1970.
Summary information is also available for some urban and rural main roads. This database includes traffic volume (in vehicle miles traveled) by vehicle type and class (car, motorcycle, truck, or bus), classification of interstate (urban or rural), vehicle weight, and average daily traffic count. These variables are available on a monthly or yearly basis. Information on traffic volume on smaller highways and local roads is available only at the state or county level, and availability and amount of information varies by county and state.

Information on Air Quality

The EPA and the individual states maintain an extensive monitoring network for various air pollutants across the country, and these data are available through AIRS, a public-access database (http://www.epa.gov/airs). The pollutants monitored include total suspended particles, particulate matter less than or equal to 10 µm diameter (PM10), and PM2.5, depending on the site. The national database was started in 1983, but measurements of PM2.5 have only recently been added. Some states and cities had monitoring data as early as the late 1960s, and this information was included in AIRS. These data can be used to assess exposure to ambient air pollution and to understand the changes in magnitude of pollution over time. The location of each monitoring station is given by latitude and longitude and can be linked to other geography-based information such as terminal address. The limited quality of information for earlier periods must be considered in any study.

Climatic factors such as temperature can strongly influence exposure. Information on climate can be obtained from the NOAA National Climate Data Center. This data- base includes hourly, daily, monthly, and yearly information on temperature and barometric pressure. Available in computerized form from 1961 forward, this resource can be linked to the trucking terminals using the latitude and longitude of the monitoring stations.

Truck Engine Certification Data and Emission Factors

The EPA certification program for heavy-duty engines tests diesel engines by using an engine dynamometer. Engine testing started in 1971, but particulate matter emissions were not regularly assessed until 1988, the year that a particulate matter standard went into effect for heavy-duty diesel engines. Manufacturers perform these tests on their engines and submit the data to EPA, which may independently conduct further testing. Diesel engine manufacturers were contacted via the Engine Manufacturer’s Association and asked to provide engine certification data for selected vehicles used by several of the companies in the past. Beginning in 1988, the emissions clearly met the required standard for each year. The data available before 1988 were minimal. The extent of these data was not specifically ascertained because the expo- sure models developed for this proposal were not based on engine certification data (see later section on exposure model development). The relevance of engine certification data for predicting in-use emissions has not been established because the test process may not represent actual operating conditions.

Chassis dynamometer testing may also be used to obtain data on emissions directly from heavy-duty diesel vehicles using simulated driving conditions. Limited testing has been done in the past because truck emission standards have not been based on these test results. Chassis dynamometer particulate matter test results for heavy-duty vehicles (trucks and buses) have been summarized by Yanowitz and coworkers (2000) using data available starting in 1975. This review showed a general reduction in particulate matter emissions since 1975. For emissions in a given year, variability among engines was consider- able. Because many preregulation vehicles are in operation in the participating companies, a statistical model can be developed to test whether truck model year can be used to estimate personal exposure to diesel exhaust.

The Diesel Truck Index (Truck Index, Santa Ana CA) database was used to assess the availability of data on design specifications for heavy-duty truck and diesel engines. This paper-based yearly manual also provides detailed information on truck model, cab design, engine model, and various engineering specifications for trucks from many manufacturers starting in 1980. Current data are also available from the engine and truck manufacturers. Engine manufacturers are also able to provide model and engineering specifications for diesel engines manufactured before 1980.

Historical specification data on vehicles and engines are available from the library of the American Truck Historical Society (2000) in Birmingham, Alabama. This library contains over 1,000 books, 25,000 pieces of vintage truck sales literature, over 30,000 vintage truck magazines, and 5,000 owner, repair, and sales manuals and is available to members of the society. This society, with over 21,000 members, is dedicated to preserving the history of companies and individuals involved in the trucking industry. It sponsors yearly meetings and truck shows at which members display trucks that are more than 25 years old. These gatherings of several hundred old vehicles could also provide an opportunity to obtain chassis dynamometer emissions data on older vehicles to compare with current vehicles.

EXPOSURE ASSESSMENT


Current Exposure Assessment

Field tests were conducted to evaluate various sampling methods, to verify that personal exposure monitoring was feasible, and to obtain some limited data on current exposures to diesel exhaust. Real-time and integrative sampling were used for personal and area monitoring of PM2.5. Particulate samples were analyzed to measure concentrations of EC and OC. Trucking industry and terminal operations were reviewed to identify the possible sources of diesel emissions and sources of EC, OC, and PM2.5, the markers selected as indices of exposure. See the Methods section for further discussion of these markers.

Field tests, including both personal and fixed location sampling, were conducted for 1 week in Atlanta in 2 large urban terminals operated by 2 of the participating companies. Fixed location sampling was also conducted for several days in 4 small rural terminals in New England. Cooperation by local management, union representatives,
 and workers was excellent. The GM and GSD exposures toPM2.5, OC, and EC by job and work site were determined (summarized in Table 8). The GM ratio of EC to OC was also determined for each job and work location. Similar measurements were made at fixed locations in the 4 small, rural New England terminals to compare the background and within-terminal EC and OC levels relative to the large urban terminals in Atlanta.

Exposure Findings 


The background levels were represented by yard samples that were taken upwind of the terminals at the property line (Table 9). These samples represent the quality of the ambient air entering the properties. This was especially important because a number of other truck freight terminals are in the area, the terminals are located on one of the runway approaches to the international airport, and a major 8-lane interstate highway is within a half mile. In general, levels of the individual markers were lower in yard samples than in corresponding indoor samples from the same terminal.

Exposures by Job Title 

The jobs differed in exposure levels (Table 8 and Figure 3): long-haul drivers had the lowest and dockworkers had the highest. The differences in exposure to EC, the diesel exhaust marker, were relatively small, but the occupational PM2.5 exposures were substantially higher than outdoor levels. The 3 exposure measures were related but not well correlated, except for EC and OC, which had an R2 = 0.81 for log–log correlation. The level of PM2.5 was weakly negatively correlated with levels of both EC (−0.49, R2 = 0.24) and OC (−0.367, R2 = 0.135) for log–log correlation. Assuming that EC represents diesel exposure in these settings, these results show thatPM2.5 would be a poor marker of diesel exposure and that OC would add little information (although the sampling numbers were small). 

The driver samples represent different traffic situations.
The long-haul drivers were driving in the evening on suburban and rural highways, whereas the P&D drivers were driving into Atlanta during the day and were frequently in heavy traffic. These differences were reflected in the personal sampling results for all 3 contaminants measured. The long-haul drivers had higher EC exposure levels than
 the yard levels. The P&D drivers had much more exposure to EC, OC, and PM2.5. The EC values suggest that drivers had exposures to diesel exhaust from traffic and possibly from small engine leaks into the cab, although the extent of exposure from possible leaks was not assessed. The EC comparison between the long-haul drivers and P&D drivers gives an indication of the expected rural-to-urban gradient in driver exposures to diesel emissions, which was approximately 2-fold.

Dockworkers and mechanics had higher personal exposures to PM2.5 than did either type of driver. Dockworkers were the most highly exposed of all workers. Probably the heavy vehicle traffic in the dock area produced consider- able diesel exhaust, propane forklift truck exhaust, and surface dust particles from tires. The source apportionment analysis of the composited high-volume filters from the dock area showed a chemical composition associated with idling diesel trucks (JJ Schauer, personal communication, 2000). This demonstrated that source apportionment analysis was feasible, but the data were too limited to draw detailed conclusions. Personal exposures of the dock-workers were generally higher than the dock area measurements presented in Table 8. This is typical of jobs in which the source of exposure is localized.

Mechanics’ exposure is associated with the sporadic operation of truck engines when the tractors are brought in and out of the shop area plus any exhaust in the area from refueling, which was done in a part of the shop building. While the PM2.5 and OC exposures for mechanics were higher than for long-haul drivers, the EC exposure for mechanics was similar to that for the long-haul drivers. This suggests a difference in sources for the particulate matter and organic compounds between these jobs.

Area samples were also collected at 4 small terminals in rural New England. The size and location (urban versus rural) of a freight terminal should affect the level of expo- sure because fewer diesel emission sources operate around smaller terminals in rural areas. This is clearly shown inTable 9, which compares the EC and OC levels measured in the large terminals in Atlanta with those in the 4 small New England terminals. A difference of 6 to 7 fold was seen between the GM exposures to EC for the Atlanta andNew England terminals in both the dock and yard samples. The background EC level measured upwind was substantially higher in the urban location. The large terminals showed a GM 2.0-µg/m3 increase from yard traffic, whereas the small terminals showed a 0.3-µg/m3 increase. The 2-fold difference in EC levels between the yard and dock locations for both Atlanta and New England terminals was not statistically significant (P > 0.05). These data show that even in rural locations the existence of diesel sources near the terminal can increase exposure levels.

The differences in OC were much smaller than the differences in EC between the urban and rural measurements. In the urban areas, the background levels of OC were 58%
 higher and the dock levels were 26% higher. The relative differences between the terminal and yard were larger for the rural terminals, indicating a larger impact from the terminal emission sources. The increase above background for both EC and OC is likely to be proportional to the number of vehicles assigned to the terminal and to the volume of freight handled.

Interpretation of the measurements made during the feasibility study is tentative because both the total number of observations and the number of sites visited were small.
The limited data did not permit a formal statistical analysis. These results strongly suggest, however, that this cohort experienced a wide range of exposures to diesel exhaust and that the source-receptor approach is useful for analyzing the exposure. The observed variability and mean EC levels showed little overlap between the large urban and small rural terminals. This indicates that the exposure contrast will be sharp, avoiding the need to rely on statistical differences. The differences were also large relative to measurement error of the monitoring techniques. If a large number of samples are collected under a wide variety of conditions, there are likely to be other subtler differences associated with characteristics of the expo- sure situations.

Mean estimates of current ambient exposure to EC from diesel exhaust ranged from 1.9 to 5.6 µg/m3 in southern California (Cass and Gray 1995); historical peaks of 15 to
 30 µg/m3 have been reported in the same area (Cass et al 1984). Concentrations of EC on the sidewalk, collected in 1996 in Harlem NY, ranged from 1.5 to 6.0 µg/m3, and variations were directly related to diesel vehicular traffic (Kinney et al 2000). Therefore, our results in trucking company workers occupationally exposed to diesel exhaust are relevant to selected general population exposures.

An important feature of the exposures observed was that they cover and extend above the range of normal urban population exposures. As a result, the study findings will be able to address the critical question of risk derived from low-level population exposures. In addition, inclusion of source apportionment methods in the study protocol makes it feasible to characterize the source of the EC and to partition the mix of pollutants associated with the range of common combustion sources: cars, trucks, home heating, power generation, and secondary pollutant formation.

Temporal Variations in Exposure 

Direct reading instruments, the DustTrak for PM2.5 and the Q Trak for CO, CO2, temperature, and relative humidity, were used to measure changes in particulate concentration over several hours and up to a full shift in the terminal areas of the dock and shop and in the cabs of operating vehicles. The objectives were to determine variability of conditions over time and to detect specific sources of potentially confounding materials such as cigarette smoke. CO2, temperature, and relative humidity are indicators of room occupancy and ventilation. Human breath is the main source of CO2 in an occupied room. When doors or windows are opened, temperature and humidity will change toward outdoor conditions, which rarely equal those indoors. As a result, the time profiles of temperature reflect changes in ventilation with outdoor air. This was seen in the analyses of data within the operating truck cab, in which it was clearly evident when the windows were opened or closed.

Figure 4 shows a typical time profile of variation in PM2.5 levels in the dock area of one of the Atlanta terminals compared with those in the yard upwind. The spikes of particulate matter in the dock area relative to the yard represent the diesel emissions entering the work area from trucks outdoors. A similar pattern was seen at the dock area of a rural New England terminal (Figure 5). Note that the dock area levels were much lower than those in urbanAtlanta, confirming the findings shown in Table 9 for the integrated samples.

Real-time monitoring in the truck cabs was used to observe how traffic and driving location affected the PM2.5 level. Figure 6 shows the PM2.5 levels during a single delivery run. None of the exposure differences during the trip was dramatic, including the difference between stop- and-go traffic and urban highway driving. Apparently, the turbulent mixing is sufficient to smooth out short-term variations. Real-time monitoring was also used to observe the effects of cigarette smoking (Figure 7). Each relatively brief spike represents one cigarette smoked with the windows open based on recorded time of cigarette smoking (a study team member accompanied the driver and recorded conditions during the trip). There was little tendency for the smoke to accumulate during these warm-weather tests, which would not be the case if the windows were closed.

Comparisons were made between the PM2.5 levels measured with the DustTrak and the standard PEM filter sampler. The weak correlation (R2 = 0.07) is consistent with the sensitivity of the DustTrak’s response to particle size distribution, which is likely to vary across terminal work sites and from day to day in the truck cab. The plan was to obtainPEM measurements of average PM2.5 concurrently with the DustTrak measurements and to use the PEM value as a 1-point calibration to set the scale for the responses. The calibration factor is the ratio of PEM measured value to mean DustTrak value for the same time period. This calibration was performed for each time profile. We determined that the DustTrak is more useful as an indicator of changes over time than as an indicator of the absolute PM2.5 levels in this setting.

Comparison to Zaebst NIOSH Data 

A rich source of historical data was found in the NIOSH Health HazardEvaluation reports by Zaebst (1989a,b,c,d,e,f ) and oworkers (1990). These data provided not only a time point for comparison with the current sampling data set, but an opportunity to identify variability across terminals because 6 larger terminals were evaluated.

An overview comparison was made for EC between the Atlanta data and the Zaebst data, as shown in Table 10. Generally, our observations were comparable to those obtained by Zaebst for warm weather, but there were some differences. Exposures of long-haul drivers were much lower in our Atlanta data than in the Zaebst data, which may reflect the effect of new diesel technology because the majority of the Atlanta vehicles were less than 5 years old. Why the Atlanta measurements for dockworkers using propane forklift trucks were so much higher was not clear, although the number of samples obtained in our data set was small.

The Zaebst data permitted analysis of the variability across terminals compared with workers within terminals using a 1-way analysis of variance. In general, the majority of variation was observed between terminals. For example, for dockworkers at terminals where diesel forklift trucks were used, the variation in EC exposure (GSD) between terminals was 1.21, and the variation in GM exposure across workers was 1.17. The variability across terminals was generally small (ie, between-terminal GSD < 1.4), as shown in Table 10. Values of GSD are multipliers and are not additive like the SD. For example, 1 GSD above the GM is given by GM × GSD. The larger GSD values indicate increasingly skewed distributions, whereas values below 1.4 are nearly symmetric and are indistinguishable from normal distributions. Two groups had large GSDs, the mechanics in cold weather and the dockworkers using propane or gasoline forklifts. In both cases, the numbers of observations were small and the groups were heterogeneous. In the Zaebst data, 2 of the terminals did not use diesel-powered forklifts: one used propane and the other used gasoline. For the mechanics, one of the terminals had very high exposures in cold weather, reportedly because the shop was “very closed up,” which would trap any exhaust emissions. Weather had little effect on the ter-minal docks sampled by Zaebst because there were no doors or they were never closed. As a result of the low variability from terminal to terminal, it may be feasible to sample a small number of representative terminal types and obtain reasonable exposure estimates for the workers at those terminal types.

Development of Models to Estimate Current Exposures

One goal of this feasibility study was to determine whether it is possible to extrapolate past exposures. The proposed strategy for this extrapolation was development of a statistical model of current exposures using defining factors that determined the exposure conditions. These factors could then be used to determine past exposure conditions by modeling the effects of changes in the factors. The original plan included additional sampling to characerize the statistical variability, but that plan was dropped when the budget was reduced by the sponsors. As a result, this analysis is largely descriptive rather than quantitative. A full study would be necessary for sufficient characterization of this statistical variability.

Rationale 

Previous studies of diesel exposure have not included statistically representative samples of exposure for subjects in the epidemiologic cohort, which has limited the accuracy of exposure assignments. Data from a statisical survey of exposures across a stratified random sample of work environments in the trucking industry will make it possible to develop quantitative statistical models. These will describe variations in exposure intensity, such as EC, as a function of each work site’s characteristics (exposure factors). The predictive capabilities of this model and its error structure can be assessed with calibration and validation data collected at the same time. Such studies were beyond the reduced scope of the feasibility study but would be included in a larger exposure assessment.

A source-receptor model can be used to describe exposure to diesel exhaust in a particular occupational environment or work site.  Such a model defines the relationship between the concentration of contaminants in an individual’s breathing zone, the emission of contaminants by sources, the transport and losses of the emissions, and the atmospheric processes that modify and dilute the emissions. The source-receptor model can be a regression model with terms for each of the model pro- cesses and factors that modify exposure intensity. This is a common approach for modeling air-pollution exposure (Kauppinen et al 1994), and Smith and colleagues (1993) have used it to describe occupational exposure of gasoline transportation workers.

Separate source-receptor models would be developed for 3 major trucking industry work sites: vehicle cab, terminal dock area, and shop (see Figure 1). In these models, the diesel engine is considered the source and the exposed worker is the receptor. The model has a term for each of the 3 sources: background air pollution, local outdoor emissions that infiltrate the work area, and indoor sources within the work area. The local outdoor and indoor source terms are defined by types and intensities of emission sources and modifiers. For example, the local outdoor source term can be defined by the number and types of trucks in the yard area of a terminal. These elements are multiplied by factors that quantify infiltration (such as the distance from the source or degree of enclosure of the indoor workplace). These factors are empirically determined by regression analysis of measured exposures as a function of workplace covariates. For this approach to be applicable to the estimation of historical exposures, the covariates must be factors that are available from historical records. Although current indoor exposures could be characterized on the basis of tracer gases and truck emissions, the tracer studies could not be done for all of the terminals where subjects have worked, and some terminals no longer exist. Alternatively, the more practical empirical factors in the model might be outdoor source, defined by number of trucks assigned to a terminal, and infiltration, defined by number of doors in the building; both factors are available from historical records held by the trucking companies.

Examples of data-derived exposure factors (Table 11) were developed from area levels (Table 9), using a simple concept: observed levels in the dock areas inside the terminals are the sum of the background level plus the emissions infiltrating from the truck activity in the yard, as shown below:

Dock Area Exposure = (Background) + (Local Emissions by Trucks in Yard)

Because the background levels were measured upwind in the yard, the exposures due to truck activity in the yard could be estimated by subtracting the background from the observed dock area levels as shown. Large terminals have much more truck activity in their yards than do small terminals: hence more EC emissions. The volume of truck traffic can be estimated from the number of trucks assigned to a particular terminal at each point in time, which is available historically from truck company records and outside sources. A modified relationship would be developed as shown below for location i at time t, where β is the observed regression coefficient:

Dock Area Exposure[i,t] = [Background for Location i,t]  + [Number of Trucks Assigned to Terminal i,t]

If the EC data collected during the feasibility study represent the current time period, then the data can be recombined to estimate the dock area exposures for various combinations of terminal size and location. Further, the average exposures at large and small terminals in urban and rural locations can also be calculated. This simple example (Table 11) illustrates the principles involved and shows that there is nearly an order of magnitude range in exposures expected across these combinations.

Data Sources 

In addition to data collected during a detailed survey of current exposure by monitoring, data can be obtained directly from a company’s records and from external databases. These can be used to form expo-ure prediction equations and to extrapolate historical exposures. The companies’ computerized and paper records on fleet characteristics, along with the RJ PolkCompany Truck Group database, can provide both the age of the fleet and how many trucks of different types were assigned to a given terminal at a given time. The DOT information on traffic counts on all federal and statewide highways could be used to create an index of pollution exposure from traffic sources linked to terminal addresses.

General outdoor background air pollution, including but not limited to PM2.5, can be quantified on the basis of data from the AIRS database. The AIRS data have limitations because of changes in methods used to obtain them over time and because less precise methods were used to collect the older data. The utility of these data will have to be assessed. Results from scientific studies contemporaneous with the older AIRS data may be useful in deriving factors to calibrate between older air pollution measurements and current measurements of EC. It was beyond the scope of this study to determine the feasibility of such work. The latitude and longitude of each monitor are provided and can be linked to the terminal addresses; interpolation techniques can be used to extrapolate information for terminals without a nearby monitor to obtain measurements directly. Information from NOAA National Climate Data Center can be used to account for climatic factors. These data will be useful for air pollution modeling to estimate background exposures in areas distant from monitoring stations. Some interpolation will be needed to estimate background levels at all of the terminals.

Source-Receptor Exposure Factor Models 
Development of these models was beyond the scope of the feasibility study but would be part of a larger study. Random effects models can be developed for exposure factors for each type of work site and for EC, OC, and PM2.5 using the general form shown below:

Exposure Intensity = (Background) + sum (Indoor Emissions) + sum (Outdoor Emissions)

Each of the terms in parentheses will be associated with specific combinations of factors derived from historical data on the terminals, trucks, and characteristics of the surrounding areas. This general model can be applied to each of the specific job and work-site combinations and can be used to estimate exposures in jobs, work sites, and terminals that were not measured. A term for weather or climate can also be considered for each model: The Zaebst data showed large differences in exposure measurements for warm and cold weather. A seasonally weighted average exposure can be estimated for each terminal by the weather parameters of its area. In addition, the models can be constructed so as to use historically available data to estimate past exposures and to validate the exposure extrapolation against the older data collected by Zaebst.
Historical changes in factors may produce corresponding changes in both the composition and the intensity of exposure. Historical exposures can be extrapolated from the current exposure models by substituting values with values taken from historical written and electronic documents. The same exposure factors used to predict EC could be evaluated to determine whether they can also be used to predict OC and PM2.5 as alternative markers of exposure to diesel or confounding materials.

Model for Vehicle Cabs 

The general model can be modified to reflect the specific sources of emissions experienced by drivers. Model factors are somewhat different for long-haul drivers than for P&D drivers.
Exposure Intensity (EC µg/m3) = 0 + 1 ( Background) + 2 (Truck Factor) + 3 (Outdoor Emissions Entering Cab)


Background levels can be determined by assessment of terminal location (urban versus rural and region of the country); current background EC, OC, and PM2.5; and EPA and state environmental sampling data in locations that are not measured. It is important to demonstrate that values in EPA and state data are comparable to those measured. Long-haul drivers drive predominantly on highways outside cities and towns, and the observed levels in the feasibility study and the Zaebst data were consistent with this. Therefore, their background exposure can be assumed to be low and equivalent to rural levels measured by region of the country. The background for the P&D driver is the local air pollution in the location where deliveries are made (such as in the areas surrounding the terminals). Outside cities, deliveries in rural areas and small towns usually have low background air pollution as indicated by rural sampling done in these tests, by the EPA, and by some of the states. The value for background pollution can be determined by reevaluating the measured current background levels in urban and rural terminals or by consulting historical local air pollution data, when available, in the AIRS database.

The truck factor is an addition, which is a function of the trucking company, terminal location, and truck model, year, and cab design. The company-specific diesel timeline (Table 2) can be used to determine the probability that a given long-haul driver or P&D driver was driving a diesel truck during a given period of time. For both types of drivers, engine emissions may also enter the cab through leaks at the floor level. The amount of emissions entering the cab depends on the quantity of emissions entering the engine compartment, any leaks in the exhaust system, and the size of holes or gaps into the cab. Such gaps may be breaks in the gear-shift rubber boot or the seal between the cab and engine compartment, both of which depend on the type of cab, company repair and maintenance practices, and age of the vehicle. These will affect the distribution of in-cab exposures, with higher exposure more likely to be associated with older vehicles. As it will not be possible to specifically identify vehicles with exhaust leaks into the cabs, one goal of a detailed exposure assessment could be to determine if cab design is a significant predictor of exposure, and how exposure is broadly affected by model year and maintenance practices. These variables, specific to the company and to the terminal location, can be determined from company databases.

The outdoor emissions entering the cab are a function of the number and type of vehicles in traffic. Dilution of on the-road emissions entering the cab depends on the distance between vehicles and their average speeds, both of which affect the amount of turbulent mixing. Open or closed windows and the use of ”fresh air” vents that draw in contaminated air also contribute to exposure, but practices vary among drivers and cannot be assessed for individuals in the cohort. At highway speeds, emissions are rapidly diluted by turbulent mixing, whereas in stop-and- go traffic (more likely to be experienced by P&D drivers in urban locations), emissions may accumulate around the slow-moving vehicles. Our real-time monitoring of PM2.5 found no evidence of this, however. A potential index of outdoor emissions can be based on heavy truck counts per mile collected by individual states or DOT.

Model for Terminal Dock and Shop Operations 

The general model can be modified as shown below to reflect the specific sources of emissions experienced by dock and shop workers:

Exposure Intensity (EC µg/m3) = 0 + 1(Background)  + 2 (Indoor Emissions in Terminal Dock or Shop) + 3 (Yard Emissions)

The background for a terminal is the local air pollution in its location: city versus small town or rural area. The AIRS data can be used to estimate the background levels. Outside cities, the terminals in rural areas and small towns are expected to have background air pollution levels consistent with rural measurements. Comparison between measured background levels and local air pollution data can determine how well the available pollution data correlate with measurements of terminal-area background exposures.

Indoor emissions in the dock area are defined by the number of forklifts within the dock area. The forklift emissions are released into the volume of air in the terminal, which is proportional to the floor space or the number of doors. This factor can be examined as a function of company data on each terminal’s loading dock area, number of loading dock doors, and number of forklifts assigned. Historically, diesel forklifts were used in larger terminals in 3 of the 4 companies (see Table 2). Therefore, for the reconstruction of historical exposure, a scaling factor based on data from Zaebst et al (1991) can be developed to account for this exposure for the years diesel forklifts were used.

Indoor emissions in shop are defined by number and type of tractors or trucks being repaired. The emissions are released into the air volume of the shop so the size affects the exposure intensity. The size of the shop can be obtained from company records, and the number of vehicles can be estimated from the number of mechanics assigned to that location. These factors can be examined in the exposure assessment as potential predictors of exposures.

The quantity of outdoor yard emissions entering the building is affected by the number and type of vehicles operating in the yard and by the amount of dilution or mixing in the yard before the emissions enter the terminal or the shop. The average number of diesel vehicles and terminal size can be examined as predictors of exposure at each location.

For P&D/dockworkers who may work on a loading dock or drive a P&D truck depending on terminal size and location, exposure assessment would require weighting time spent on the dock versus time driving. This proportion would be applied to the exposure model developed for each separate work location. Company payroll records can be used to identify hours spent working on a loading dock or driving a truck and to determine an exposure profile (mean and variation) based on terminal size and location for a typical P&D/dockworker.

Estimation of Historical Exposures

Two definitions of long-term exposure to diesel exhaust can be assigned to each individual. First, exposure can be defined categorically on the basis of yearly job title, terminal size, and terminal location (that is, region of the country and urban versus rural area). Second, exposure can be defined quantitatively as cumulative dose of EC using the statistical model derived from the exposure assessment discussed in the previous section. The company personnel files provide chronological job title and terminal information. Teamsters union pension fund records of contributions made by each unionized trucking company on behalf of a worker supplement this information for retired workers.

Previous epidemiologic studies have used job title alone to categorize exposure to diesel exhaust (Steenland et al 1990; Cohen and Higgins 1995; Bhatia et al 1998). For this cohort, the results showed that job alone might give highly misclassified exposure assignments. This definition should be refined by including terminal characteristics and formulating job-terminal exposure categories for historical periods. The different duties encompassed by the job title determine the potential contact that an individual has with diesel vehicles. However, size and location of the trucking terminal also appear to strongly influence exposure and the actual definition of job duties covered by the job title. Large terminals have greater activity and number of vehicles, which leads to higher exposures. At smaller terminals there is also less distinction between dockworkers and P&D drivers because a given worker often per- forms both sets of duties. Therefore, a dockworker at a small terminal does not have the same exposures as one at a larger terminal. In addition, the urban or rural location of the terminal influences the exposures experienced by all employees, including the drivers. Terminals in urban areas have higher levels of background air pollution owing to the other sources around them. Drivers based at these terminals are more likely to drive congested city routes than drivers from rural terminals. Long-haul drivers are assigned to a home terminal, indicating the region of the country in which they are most likely to drive. Intercity highways and those in the heavily populated East Coast and West Coast areas are more congested than those in less populated parts of the country. All of these factors influence exposure to diesel exhaust and therefore should be taken into account when grouping trucking employees into exposure categories. Dockworkers and clerks at small rural terminals appeared to have the lowest exposure, with exposures at or near background levels.

Several semiquantitative exposure indices can be developed from the duration an individual has worked in the above exposure categories. These permit evaluation of important exposure contrasts. Because there is little cross- over among the categories, risk associated with years of work at small terminals can be compared with risk associated with years of work in large terminals to determine whether the cumulative risk differs by terminal size. SinceP&D trucks are assigned to specific terminals and terminals vary in the year when they first used diesel trucks, years driving a diesel P&D truck can be used to index driver exposure. Similarly, diesel exposure for mechanics and dockworkers can be calculated on the basis of an individual’s assignment to a specific terminal. Years of use of diesel-powered forklifts can also be determined. Years of driving a P&D truck for city delivery can be compared with years of rural delivery to determine whether driving in traffic increases the risk beyond driving by itself. These categorical and semiquantitative exposure categories have a major advantage in their historical accuracy of potential exposure because they are based on company records and do not depend on the accuracy or precision of retrospective exposure estimates.

Quantitative historical exposures can be backward extrapolated from a statistical evaluation of current exposures. The same statistical model can be used based on historical changes in the exposure factors. In developing this approach, important elements are (1) a set of factors for which historical data are available over the whole time period of interest, and (2) data that document the qualitative and quantitative effects of changes that can be used for model validation. Our feasibility study demonstrated that this is generally possible, but all historical factors that might be used to extrapolate exposure in the models developed may not be available. For example, air pollution data for the 1970s are limited, and specific highway and regional traffic count data are not generally available before the 1970s. The effect of these incomplete data on the extrapolation of historical exposure will depend on the importance of the individual factors in the exposure model. Possibly surrogates for these incomplete data can be determined. However, these limitations would affect calculation of only a small fraction of cumulative exposure if diesel exposure were to be assessed through 2000 and beyond in the design of a cohort study.

Other exposure-modifying factors cannot be modeled explicitly but may be important in evaluating the measurement data and extrapolating past exposures. For example, local variations in the mix of vehicles and traffic patterns will affect the exposure relationship. Cities and urban areas can have substantial variation in local industries and geography. The effects of these limitations can be mini- mized with a sufficiently large statistical sample that cuts across the whole industry. Such a sample would permit reasonably precise historical estimates of average exposure intensity within categories.

In the ideal case, we would want extensive measurement data to validate our extrapolation model. However, occupational and environmental measurements of EC are limited, especially data collected with the best methods, and are not available prior to the mid 1980s. Several studies have shown that reflectance from deposits on filter paper and the old method for measuring air pollu- tion by light transmission through deposits on filter paper (coefficient of haze) are highly correlated with EC measurements (Cass et al 1984; Kinney et al 2000). As a result, it may be possible to obtain estimates of EC for comparison with our model in the large cities where coefficient of haze was routinely measured.

Changes in emission sources are one key area of concern. Truck, car, and industrial emissions have all changed over time because of the ongoing efforts to reduce air pollution. These changes have been documented to different degrees, but generally the data include only relatively coarse assessments of total particulate matter and toxic gases. The source-receptor model implies that changes in sources will be reflected in changes in exposure at the receptor. Reductions in total emissions must be assumed to reduce component emissions, which may only be approximately true but difficult to verify.

Estimation of Measurement Error

Exposure measurement error will likely affect our assessment of the true relative risk of lung cancer attributable to diesel exhaust exposure. Errors depend on the quality and detail of the data on the sampling sites and historical operations. This error is most likely to be nondifferential (ie, unrelated to disease status). Thus, our observed relative risks for a dichotomous exposure will be underestimates of the true relative risk (Rothman 1981). However, the effect of error on the assessment of dose response may go in either direction (Dosemeci et al 1990; Birkett 1992).

The magnitude of these effects on the observed point estimate of the relative risks can be estimated. A correction factor, γ, can be derived from the variability of EC levels between jobs, between workers in the same job, and within an individual working at the same job over time (Rosner 2000). The true logistic regression coefficient, β*, would then be estimated as the observed β divided by γ, and the confidence intervals of the corrected relative risk would be wider than those of the observed risk.

The data from this feasibility study were too limited to allow direct estimation of measurement error between jobs, urban and rural locations, and large and small terminals. However, we used the GSD between all jobs mea- sured during this feasibility study and estimates of theGSD between and within workers obtained from published data and NIOSH Health Hazard Evaluation Reports to illustrate a few possible examples if lung cancer risk were compared between 2 jobs (Table 12). In our extreme example, if we assume a GSD of 1.4 between jobs, and a combined GSD of 2 between and within workers, the β would be underestimated by 5-fold. The preliminary expo- sure data showed a range of 6 to 7 fold in exposures between large urban truck terminals and small rural terminals. That broad range will allow some imprecision in the numerical estimates of exposure and maintain the ability to detect differences in risk associated with exposure if the GSDs used in these categories are similar to those obtained for job- exposure categories in a larger study. A full study would include more extensive assessment of measurement error.
