Estimation of Source-Specific Occupational Benzene Exposure in a Population-Based Case–Control Study of Non-Hodgkin Lymphoma

Abstract

Objectives

Occupational exposures in population-based case–control studies are increasingly being assessed using decision rules that link participants’ responses to occupational questionnaires to exposure estimates. We used a hierarchical process that incorporated decision rules and job-by-job expert review to assign occupational benzene exposure estimates in a US population-based case–control study of non-Hodgkin lymphoma.

Methods

We conducted a literature review to identify scenarios in which occupational benzene exposure has occurred, which we grouped into 12 categories of benzene exposure sources. For each source category, we then developed decision rules for assessing probability (ordinal scale based on the likelihood of exposure > 0.02 ppm), frequency (proportion of work time exposed), and intensity of exposure (in ppm). The rules used the participants’ occupational history responses and, for a subset of jobs, responses to job- and industry-specific modules. For probability and frequency, we used a hierarchical assignment procedure that prioritized subject-specific module information when available. Next, we derived job-group medians from the module responses to assign estimates to jobs with only occupational history responses. Last, we used job-by-job expert review to assign estimates when job-group medians were not available or when the decision rules identified possible heterogeneous or rare exposure scenarios. For intensity, we developed separate estimates for each benzene source category that were based on published measurement data whenever possible. Frequency and intensity annual source-specific estimates were assigned only for those jobs assigned ≥75% probability of exposure. Annual source-specific concentrations (intensity × frequency) were summed to obtain a total annual benzene concentration for each job.

Results

Of the 8827 jobs reported by participants, 8% required expert review for one or more source categories. Overall, 287 (3.3%) jobs were assigned ≥75% probability of exposure from any benzene source category. The source categories most commonly assigned ≥75% probability of exposure were gasoline and degreasing. The median total annual benzene concentration among jobs assigned ≥75% probability was 0.11 ppm (interquartile range: 0.06–0.55). The highest source-specific median annual concentrations were observed for ink and printing (2.3 and 1.2 ppm, respectively).

Conclusions

The applied framework captures some subject-specific variability in work tasks, provides transparency to the exposure decision process, and facilitates future sensitivity analyses. The developed decision rules can be used as a starting point by other researchers to assess occupational benzene exposure in future population-based studies.

Introduction

Assessment of occupational exposures in population-based case–control studies is typically based on participants’ responses to questionnaires asking about their job and work tasks because the breadth of relevant jobs and time periods makes it impractical to collect measurements from the participants’ workplaces (Friesen et al., 2015). In a growing number of studies, participants’ questionnaire responses have been systematically linked to exposure estimates through the development and application of programmable decision rules (Pronk et al., 2012; Peters et al., 2014). Decision rule-based approaches for assessing exposure provide transparency to the exposure decision process by documenting experts’ assessments based on available data, literature review, and professional judgment. They also provide a mechanism for conducting sensitivity analyses and for applying the developed exposure decisions to other studies (Friesen et al., 2016). Furthermore, the decision rules can capture more subject-level differences in reported work tasks when detailed occupational questionnaires are incorporated in a study compared to the use of population-level job-exposure matrices (JEMs).

In this article, we describe a decision rule-based approach for estimating occupational exposure to benzene for participants in the National Cancer Institute-Surveillance, Epidemiology, and End-Results (NCI-SEER) case–control study of non-Hodgkin lymphoma (NHL) (Purdue et al., 2011). Benzene is a common occupational solvent that has had used in multiple industries, including rubber tire manufacturing and the printing industry. The International Agency for Research on Cancer (IARC) has classified benzene as a Group 1 human carcinogen since 1979, based primarily on sufficient evidence that it causes acute myeloid leukemia (IARC, 2012). IARC’s most recent evaluation of benzene in 2017 reaffirmed previous conclusions that there was limited evidence of a positive association with risk of NHL, including chronic lymphocytic leukemia subtype (Loomis et al., 2017). Therefore, additional research, including analysis by NHL subtype, is still needed. For future etiologic analyses, we developed decision rules for assessing the probability, frequency, and intensity of benzene exposure using the NCI-SEER participants’ occupational history responses and, when available, responses to job- and industry-specific modules. We derived rules for 12 categories of benzene exposure sources that can be used separately or summed together to estimate total annual benzene exposure. The subject-level, annual source-specific, and total exposure estimates derived here will be used to calculate exposure metrics for etiologic analyses of NHL risk.

Methods

Study population and occupational information

The NCI-SEER study of NHL has been described in detail (Purdue et al., 2011). Briefly, it is a multi-center (the state of Iowa; Los Angeles County, California; and the Seattle, WA; and Detroit, MI metropolitan areas), population-based case–control study with 1321 cases, diagnosed with incident NHL between 1998 and 2000 at 20–74 years of age, and 1057 controls matched to the cases by study center, age, gender, and race.

Prior to a home visit by a trained interviewer, participants were mailed a lifetime occupational history calendar and asked to list each job ever held for at least 12 months since the age of 16. During the computer-assisted personal interview, participants provided free-text responses to questions asking, for each job reported on the work history calendar, the job title, name of employer, dates of employment, number of hours worked, type of business or service, work tasks and activities performed (tasks), tools and equipment used (tools), and chemicals and materials handled (chemicals). Each job reported by participants in the occupational history was assigned a four-digit 1980 Standard Occupational Classification (SOC) code (US Department of Commerce, 1980) and a four-digit 1987 Standard Industrial Classification (SIC) code (US Department of Commerce, 1987). A list of the SOC and SIC codes observed in this study is available in Supplementary Table S1 (available at Annals of Work Exposures and Health online).

Approximately 1 year into the interviewing phase of data collection, the study incorporated job- and industry-specific questionnaire modules into the interview. The modules were designed to supplement the occupational history questionnaire by collecting detailed information about workplace tasks and exposures, and their frequency, related to the use of solvents, including benzene, over the entire duration of employment in a given job or industry. Modules were incorporated for 21 occupations (e.g. painter, welder) and 11 industries (e.g. printing industry, textile industry) with potential for solvent exposure (Table 1). Jobs reported by participants who were enrolled after the modules were incorporated triggered the modules based on a search for key words in the free-text occupational history responses. If the key word search suggested more than one potential module, the interviewer selected the most relevant module for that job. If the interviewer was unsure of which module to assign, a generic solvent exposure module was selected. To reduce participant burden, a maximum of five modules were administered per participant. Modules were only assigned for jobs held for at least 12 months, prioritizing participants’ longest-held jobs.

Table 1.

Job- and industry-specific modules and availability of questions directly or indirectly associated with 12 categories of benzene exposure sources.

			Benzene exposure source categoriesa
Module	Number of jobs	%	Combustion	Degreasing	Gasoline	Glue	Ink	Laboratory work	Paint	Petroleum	Printing	Stain/ varnish	Stripping paint	General solvent use
Job-specific modules (n = 21)
Teacher (TE)	173	2.0						X						X
Assembly worker (AW)	138	1.6		X		X			X			X
Janitor (JA)	90	1.0												X
Engineer (EN)	75	0.9												X
Mechanic (ME)	75	0.9		X		X			X				X
Machinist (MA)	69	0.8		X
Laborer (LA)	55	0.6		X					X			X
Carpenter (CR)	40	0.5				X			X			X	X
Repairer of industrial machinery (IM)	38	0.4		X		X			X				X
Handyman (HY)	34	0.4		X		X			X
Gas station attendant (GS)	32	0.4		X	X				X				X
Electrician (EL)	30	0.3		X		X							X
Welder (WE)	24	0.3		X					X				X
Production inspector (PI)	21	0.2				X			X
Painter (PA)	19	0.2							X			X	X
Plumber (PL)	19	0.2		X		X			X
Cabinet maker (CA)	14	0.2				X			X			X	X
Aircraft mechanic (AM)	12	0.1		X		X			X				X
Chemist (CH)	7	0.1						X
Packing machine operator (PM)	5	0.1				X
Tile setter (TS)	3	<0.1				X								X
Industry-specific modules (n = 11)
Printing industry (PRI)	32	0.4		X			X				X			X
Textile industry (TXI)	20	0.2		X			X				X			X
Dry cleaning/ laundry industry (DLI)	17	0.2		X										X
Chemical industry (CHI)	9	0.1		X										X
Oil refining industry (ORI)	9	0.1			X					X
Rubber industry (RUI)	6	0.1			X									X
Electroplating industry (EPI)	4	0.1		X									X
Furniture industry (FUI)	3	<0.1		X					X			X	X
Pulp and paper industry (PPI)	3	<0.1					X				X			X
Leather industry (LEI)	1	<0.1		X		X								X
Semiconductor industry (SCI)	1	<0.1		X										X
Generic solvent module (SOL)	40	0.5		X		X			X			X	X
No module assigned	7709	87.0
Total	8827	100

^aDescriptions of the benzene source categories are available in Table 2.

Exposure assessment

The exposure assessment approach was conducted in four stages, shown in Figure 1 and described in detail below. Stage 1 was the development of the decision rules. Stage 2 was the application of the decision rules to assign source-specific probability of exposure to the participants’ jobs. Stage 3 was the assignment of exposure frequency and annual exposure intensity to benzene-exposed jobs for each exposed person-year. Stage 4 was the calculation of annual source-specific and total benzene exposure for all exposed person-years. We developed a source-specific exposure assignment approach, using a framework similar to Callahan et al.’s (2018) assessment of source-specific lead exposure because of the potential for multiple benzene source categories within a job and to simplify the assignment of decision rules

Figure 1.

Source-specific exposure assessment approach.

Stage 1: development of decision rules to assign benzene exposure

Step 1: To develop decision rules, we first identified scenarios in which benzene exposure may occur by gathering a team of industrial hygienists with expertise in benzene exposure assessment. They conducted a literature review and identified scenarios for which there was a potential for occupational benzene exposure in US workplaces during the time period relevant to our study (1940–2000). Our primary sources of exposure information were a previous literature review conducted by van Wijngaarden and Stewart (2003) that identified the uses of, and exposure to, benzene in North American industries for which quantitative exposure data were published and Hein et al.’s (2007) paper describing a statistical model for benzene exposure that used the published data identified in the aforementioned literature review. We then supplemented the exposure information provided in these papers with additional published and gray literature. We categorized the identified scenarios where benzene exposure may occur into 12 source categories based on work tasks performed and chemicals used: combustion, degreasing, gasoline, glue, ink, laboratory work, paint, petroleum, printing, stain and varnish (‘stain/varnish’), stripping paint, and general solvent use (Table 2).

Table 2.

Description of the 12 categories of occupational benzene exposure sources identified based on a literature review.

Category	Description of potential for benzene exposure	Key references
Combustion	A component of products of incomplete combustion of coal, oil, gasoline, waste, biomass, and carbonaceous man-made materials.	IARC (2013)
Degreasing	Added to or an impurity in petroleum distillates used in parts washing, manual degreasing, and other metal and plastic cleaning tasks; an impurity in commercial grades of toluene, n-hexane, and cyclohexane used for cleaning and degreasing.	Elkins and Pagnotto (1956); NIOSH (1977); Fedoruk et al. (2003; Williams et al. (2008)
Gasoline	A natural component of gasoline; an additive, along with toluene, ethylbenzene, and xylene (known as BTEX), used to increase gasoline octane rating.	Runion (1975; Verma and des Tombe (2002)
Glue	Present in solvent-based cements and adhesives, in particular rubber-based contact cements and other adhesives requiring fast dry times.	EPA (1978)
Ink	Added to or an impurity in thinners for solvent-based inks, in particular those requiring fast-drying times	Novick et al. (2013)
Laboratory work	Solvent used in laboratory settings for chemical and clinical processes, most notably chemical extractions and tissue processing.	EPA (1978)
Paint	Present as a solvent or impurity in oil-based coatings and associated cleaning agents; present as a solvent or impurity in solvent-based paint thinners, particularly for spray application; an impurity in industrial grade toluene, a widely used solvent in the paint industry.	NIOSH (1974); EPA (1978)
Petroleum	A natural component of crude oil and petroleum-derived products.	Runion (1988); Verma et al. (2001)
Printing	Added to or an impurity in impurity in solvent-based cleaning agents for printing presses, plates, and blankets; evaporation from printing inks while presses are running and from ink fountains.	ATSDR (2007); Novick et al. (2013)
Stain/varnish	Present as a solvent or impurity in oil-based stains, varnishes, lacquers and associated clean agents; present in solvent-based thinners, particularly for spray application of lacquers; solvent for ethyl cellulose resins used in lacquers.	Browning (1953); EPA (1978)
Stripping paint	Added to or an impurity in solvent-based paint removers for oil-based paints, stains, varnishes, and lacquer.	EPA (1978)
General solvent use	Industry-specific use other than cleaning/degreasing parts; includes chemical, leather, textile, rubber, pharmaceutic, and pesticide manufacturing; use as a solvent for fats, waxes, resins, oils, plastics, rubber; a component of other solvents, either as a known ingredient or impurity, including industrial-grade toluene, n-hexane, n-heptane, cyclohexane, and naphtha.	Browning (1953); Elkins et al. (1963); AIHA (1970); EPA (1978); IARC (2012)

Step 2: Based on our literature review, we identified 47 occupation groups (e.g. mechanic, painter) and 23 industry groups (e.g. petroleum refining, printing) with potential for benzene exposure. We linked these groups to the relevant SOC and SIC codes observed in our study population (Supplementary Tables S2 and S3, available at Annals of Work Exposures and Health online).

Step 3: We developed a list of words that identified tasks, tools, and chemicals that may be associated with potential benzene exposure. We then extracted those words from participants’ free-text responses to the occupational history questions and converted them into standardized variables using an approach previously developed by Friesen et al. (2014a). A total of 70 occupational history exposure variables were developed that captured 30 tasks (e.g. vehicle repair), 16 tools (e.g. vehicle mechanic’s tools), and 24 chemicals (e.g. gasoline) reported by participants (Supplementary Table S4, available at Annals of Work Exposures and Health online).

Step 4: Using the benzene-exposed scenarios identified in the literature review as our starting point, the exposure assessment team identified combinations of the SOC and SIC codes, the 70 standardized variables for tasks, tools, and chemicals, and the module responses that represented unique benzene exposure scenarios. For each unique combination of information, and using expert judgment and the literature review from Stage 1: Step 1, we developed a set of decision rules to assign time period-specific probability for each of the 12 benzene source categories separately. Probability was defined as the estimated proportion of workers in a given scenario exposed to benzene at a time-weighted average exposure level > 0.02 ppm. This threshold level was defined to exclude environmental exposures and represented the lowest arithmetic mean concentration value reported in the database developed by van Wijngaarden and Stewart (2003). Probability of exposure to each benzene source category was assigned on a five-category, ordinal scale of none (0 to <5% of workers exposed), low (5–49%), medium (50–74%), medium-high (75–89%), and high (≥90%). Decision rules for assigning a job high or medium-high probability of exposure to a specific benzene source category are listed in Table 3, and the complete decision rules are listed in Supplementary Table S5 (available at Annals of Work Exposures and Health online).

Table 3.

Programmable decision rules identifying criteria for assignment of medium-high (MH) or high (H)^a probability of benzene exposure, by source category.

Benzene source categoryb	Relevant modulesc	Time period (probability assigned)	Module criteria		Occupational history criteriad
			Module questions	Module responses	Relevant task variables flagged? (Category)	Relevant tool variables flagged? (Category)	Relevant chemical variables flagged? (Category)
Degreasing	IM, MA, ME, PL, WE, SOL, AM, AW, EL, GS, HY	1940–1979 (MH)	Did you ever clean or degrease metal or plastic parts with chemicals?	Yes; Yes	None required	None required	None required
			When you cleaned metal or plastic parts, did you ever use avgas, gasoline, gunk, jet fuel, naphtha, paint thinner, toluene, or VM&P?
	LA	Any (H)	Did you work in a production area of the plant?	Yes; Yes to at least one	None required	None required	Yes: Benzene
			Did you ever clean out tanks, vats, or chemical sewers?
			Did you ever clean other types of equipment or machinery?
Gasoline	GS	Any (H)	None		Yes: Handle fuel or pump gasoline	None required	Yes: Fuel, gasoline, or petroleum
Laboratory work	CH	1940–1959 (H)	Did you ever work in a lab?	Yes; Yes	None required	None required	None required
		1960–1979 (MH)	Did you use or handle aromatic compounds for at least 1 h a week?
Paint	IM, ME, PL, WE, PA, SOL, AM, AW, CA, CR, GS, HY	1940–1959 (H)	Did you ever use oil-based paint?	Yes	None required	None required	None required
		1960–1979 (MH)
Stain/ varnish	PA, SOL, AW, CA, CR	1940–1959 (H)	Did you usually use oil- or solvent-based varnishes?	Yes to at least one	None required	None required	None required
		1960–1979 (MH)	Did you usually use oil- or solvent-based stains?

^aRules for the remaining probability ratings are available in Supplementary Table S6 (available at Annals of Work Exposures and Health online).

^bProbability of exposure for the following source categories were assigned via expert review and do not have explicit decision rules: combustion, ink, petroleum, and printing. Decision rules for assigning probability of exposure greater than medium were not developed for the source categories glue, strip paint, and general solvent use.

^cDefinitions of the module abbreviations are provided in Table 1.

^dDescriptions of the task, tool, and chemical categories associated with benzene exposure that were extracted from participants’ free-text responses to the occupational history are provided in Supplementary Table S5 (available at Annals of Work Exposures and Health online).

In addition, decision rules for some combinations of variables were not assigned a probability and were instead identified by the team as requiring expert job-by-job review. This included jobs that could not be assigned a single probability rating because of potential exposure heterogeneity based on the module responses (e.g. jobs assigned the laborer module in which the participant reported cleaning equipment and tanks). Jobs identified as requiring expert review also included jobs assigned one of the 11 industry-specific modules (representing rare exposure scenarios) and jobs for which the module and occupational history responses provided conflicting information (e.g. jobs held prior to 1980 for which the participant responded ‘no’ to the module question ‘Have you ever used glue?’, but for which occupational history variables related to glue exposure were flagged).

Stage 2: assigning probability of exposure to participants’ jobs

Step 1: For jobs with module responses, we applied the time period and source-specific programmable decision rules using a hierarchical procedure that assigned the rules in order from highest to lowest probability estimate. When the responses for a job triggered multiple rules for a source, the highest probability was assigned.

Step 2: Using the module responses for job groups defined based on combinations of SOC, SIC, and variables extracted from the free-text occupational history questions, we calculated (i) the proportion of workers within a job group conducting an exposed task and (ii) the median task duration. These job-group estimates were then used in decision rules to assign probability of exposure to jobs without module responses. Jobs without modules for which there was insufficient data to derive a median estimate from similar jobs with module responses were identified for job-by-job review. The job-group frequency estimates were used in Stage 3.

Step 3: All jobs identified in using decision rules as requiring expert review were assessed one at a time by an industrial hygienist based on a review of the literature, the job-group medians from jobs reasonably expected to have performed similar tasks, and professional judgment. Job-by-job review was also used as a quality control step to verify the decision rule assignments for (i) all jobs assigned a high probability of exposure and (ii) all jobs held prior to 1980 that were assigned medium or low probability that had possible benzene exposure scenarios identified in the occupational history variables. This was done to maximize both the specificity and sensitivity of identifying jobs with high probability of benzene exposure. All job-by-job review assignments were then reviewed by a second industrial hygienist and any discrepancies were resolved by consensus.

Stage 3: assigning frequency and intensity of exposure

We assigned annual source-specific frequency and intensity estimates only to jobs assigned a source-specific probability of medium-high or high (≥75% of workers exposed). Subjects assigned low or medium probability of exposure based on activities will be retained as a separate group from the unexposed group for future etiologic analyses but will not contribute to cumulative exposure estimates to maximize both sensitivity and specificity, as was previously done in Bassig et al. (2015) for evaluations of benzene and NHL and in Purdue et al. (2011) for evaluations of trichloroethylene and NHL.

Step 1, Frequency: We assigned frequency of benzene exposure to each job with medium-high or high exposure probability. Frequency of benzene exposure was defined as the estimated proportion of time spent working on a task in which a participant was exposed to benzene at an estimated concentration level of >0.02 ppm (0–100% of work time, continuous scale). We used proportion of time, rather than hours per week, to facilitate the application of derived job-group medians from those jobs with module responses to jobs without module responses. We accounted for the number of hours worked per year in our subsequent calculations of concentration (see Stage 4: calculation of annual concentration). Proportion of time was calculated by dividing the number of hours spent working on a task (e.g. degreasing with a benzene-containing solvent), as reported in the module responses specific to each benzene source, by the total number of hours per week worked, as reported in the occupational history. If the number of hours spent working on a task was <1.0 h per week (i.e. <2.5% of a 40-h work week), we lowered the source-specific probability to ‘medium’ (50–74% of workers exposed) and did not assign an exposure frequency or intensity because the benzene concentration could not be distinguishable from the threshold background exposure. Jobs without source-specific frequency information from the module responses were assigned a proportion of time exposed using job-group medians from similar jobs with module responses (see Stage 2, Step 2) or estimates based on expert judgment when job-group medians were not available.

Frequency and intensity information for the benzene source categories paint and stain/varnish were combined to develop one overall frequency and one overall intensity estimate because of the limited availability of measurement data related to stain/varnish activities and the similarities in module questions and reported work tasks relevant to paint and stain/varnish activities. We refer to the combined source as ‘paint/stain/varnish’.

Step 2, Intensity: Prior to assigning exposure intensity, we divided each job into unique calendar years (employed person-years). We then assigned annual intensity estimates separately for each benzene source category for each job/calendar year with a medium-high or high probability of exposure (hereafter, exposed person-years). Exposure intensity was defined as the average concentration (in ppm) that workers were exposed to while conducting benzene-exposed work tasks. Whenever possible, intensity estimates were derived using published personal air measurement data and assigned on a continuous scale. When measurement data for a given benzene source scenario were sparse, we assigned intensity based on extrapolation of measurements from similar exposure scenarios. If no relevant measurement data were available, we used expert judgment to assign intensity as the midpoint value of following categories: <0.02 (unexposed); 0.02 to <0.1; 0.1 to <1.0; 1.0 to <5; 5 to <10; 10 to 25 ppm. Intensity levels >25 ppm were assigned a value of 25 ppm. For the source categories paint/stain/varnish and gasoline, we had sufficient published monitoring data to develop mixed-effects meta-analysis models to systematically combine multiple sources of summary measurement data. This approach is described in Supplementary Appendix A (available at Annals of Work Exposures and Health online). For the source category degreasing, we used the approach described in Supplementary Appendix B (available at Annals of Work Exposures and Health online), in which we first identified the likely benzene content in common degreasing solvents for the specific solvents and degreasing activities reported by participants in the module responses and then took into account differences in the intensity of exposure based on degreasing method and solvent temperature using published measurements.

Stage 4: calculation of annual concentration

Source-specific annual benzene concentrations were calculated for each job, j, and benzene source category, s, using Equation (1):

$${\displaystyle \begin{array}{l}{\displaystyle {[\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}{\mathrm{n}}_{j,s}]}_{\mathrm{A}\mathrm{n}\mathrm{n}\mathrm{u}\mathrm{a}\mathrm{l}}=\mathrm{F}\mathrm{r}\mathrm{e}\mathrm{q}\mathrm{u}\mathrm{e}\mathrm{n}\mathrm{c}{\mathrm{y}}_{j,s}\times \mathrm{I}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{n}\mathrm{s}\mathrm{i}\mathrm{t}{\mathrm{y}}_{j,s}\times \mathrm{F}\mathrm{T}{\mathrm{E}}_j}\end{array}}$$ (1)

Here, frequency (proportion of work time) and intensity are the source-specific metrics described in the above sections. FTE, or full-time employee equivalent, was calculated as the number of hours worked in a given job per year, as reported in the occupational history, divided by 2080 h, which represents a standard work schedule of 40 h per week for 52 weeks per year. We assumed an FTE equal to 1.0 (job median) for jobs that were missing data on the number of hours worked. We capped the FTE at a maximum value of 2.0 for those reporting more than 80 h worked per week. When the intensity and frequency resulted in a source-specific annual concentration that was <0.02 ppm (our exposure threshold), we lowered the source-specific probability to ‘medium’ and assigned a source-specific concentration of 0 ppm. A total annual benzene concentration for each job was calculated by summing all source-specific annual concentration estimates.

Results

Of the 2170 participants who completed an occupational history, 1322 (60.9%) were enrolled in the study after incorporation of the job- and industry-specific modules. A total of 8827 jobs, covering the time period 1946–2001, were reported by all participants; of those, 1118 (13%) were assigned one of the 33 modules (Table 1). The teacher module was the module applied to the most jobs (173 jobs, 2.0%).

Overall, decision rules identified 711 (8%) of the 8827 jobs as requiring expert review for one or more sources (not shown). We show the number of jobs to which the applied decision rules assigned a non-zero probability estimate or identified as requiring expert review for each source and module in Table 4. Expert review was identified as most frequently needed for the source categories glue (208 jobs, 19%), strip paint (88, 8%), and general solvent use (80, 8%). Of the 1118 jobs with module responses, the decision rules most often assigned a non-zero probability of exposure for the source categories degreasing (310 jobs, 28%), painting (152, 14%), and stain/varnish (85, 8%).

Table 4.

Number of jobs to which the decision rules applied a non-zero probability of exposure or identified as requiring expert review, stratified by job module and benzene source category.

Number of jobs
Module (N jobs)a	Degreasing		Gasoline		Glue		Lab		Paint		Solvent		Stain/varnish		Strip paint		Combustionb	Inkb	Petroleumb	Printingb
	DR > 0c	ERd	DR > 0c	ERd	DR > 0c	ERd	DR > 0c	ERd	DR > 0c	ERd	DR > 0c	ERd	DR >0c	ERd	DR > 0c	ERd	ERd	ERd	ERd	ERd
AM (12)	11	1			0	5			6	0			0	1	1	3
AW (138)	53	0			25	16			13	0			7	0	0	1
CA (14)	0	2			0	12			5	0			8	0	1	3
CH (7)	1	1			0	0	5	2	0	0			0	0	0	0
CR (40)	0	4			0	32			28	0			24	0	6	12
EL (30)	14	1			0	8			0	1			0	1	2	7
EN (75)	0	14			0	14			0	15	0	15	0	14	0	14
GS (32)	26	0	27	5	0	1			8	0			0	0	4	4
HY (34)	16	3			1	8			11	9			0	4	0	0
IM (38)	29	1			14	3			14	0			0	0	4	7
JA (90)	0	6			0	3			0	6			0	5	0	3
LA (55)	6	1			0	0			5	2			5	1	0	0
MA (69)	26	0			0	3			0	0			0	0	0	2
ME (75)	67	0			3	31			24	1			0	1	10	26
PA (19)	0	2			0	0			19	0			19	0	0	1
PI (21)	6	1			3	0			2	2			3	0	0	0
PL (19)	8	0			0	10			4	0			0	0	0	0
PM (5)	0	0			0	2			0	0			0	0	0	0
SOL (40)	22	2			8	0			13	0			10	0	5	3
TE (173)	14	2			0	6	11	0	0	11	0	9	0	1	0	0
TS (3)	0	0			0	2			0	0	2	0	0	0	0	0
WE (24)	11	0			0	0			0	0			9	0	0	2
CHI (9)											0	9
DLI (17)											0	17
EPI (4)											0	4
FUI (3)
LEI (1)
ORI (9)
PPI (3)
PRI (32)	0	32			0	32
RUI (6)											0	6
SCI (1)
TXI (20)					0	20					0	20
All (1118)	310	73	27	5	54	208	16	2	152	47	2	80	85	28	33	88	0	32	9	32

^aDefinitions of the module abbreviations are provided in Table 1.

^bProbability was assigned only through expert review.

^cDR > 0 = probability of exposure assignment of low, medium, medium-high, or high based on the decision rules.

^dER, identified for expert review based on the decision rules.

Medium-high or high probability of overall benzene exposure was assigned to 287 of the 8827 jobs (3.3%) (Table 5) and was attributable to one benzene source category for 245 jobs, two source categories for 41 jobs, and three source categories for 1 job (not shown). High probability was assigned to 133 of the 8827 jobs (1.5%) and was attributable to one source category for 124 jobs and two source categories for 9 jobs. These jobs with medium-high or high probability from any benzene source category correspond to 2.4% of employed person-years (Table 6). The source categories most commonly assigned medium-high or high probability of exposure were degreasing (0.6% of employed person-years), gasoline (0.6%), paint (0.4%), and stain/varnish (0.4%). Medium-high to high probability of benzene exposure was most prevalent prior to 1980.

Table 5.

Distribution of probability of exposure assignments by benzene source category for 8827 jobs in the NCI-SEER NHL Study.

Source category	Number of jobs by probability rating
	None	Low		Medium		Medium-high		High
	N	N	%	N	%	N	%	N	%
Degreasing	8484	206	2.3	80	0.9	57	0.7	0	0
Glue	8650	138	1.6	31	0.4	7	0.1	1	<0.1
Paint	8504	124	1.4	152	1.7	33	0.4	14	0.2
Stain/varnish	8723	50	0.6	17	0.2	22	0.3	15	0.2
Stripping paint	8719	93	1.1	14	0.2	1	<0.1	0	0
Laboratory work	8799	22	0.3	4	0.1	2	<0.1	0	0
Gasoline	7844	635	7.2	254	2.3	1	<0.1	93	1.1
Printing	8765	8	0.1	21	0.2	29	0.3	4	0.1
Ink	8792	2	<0.1	23	0.3	6	0.1	4	0.1
Solvent	8219	541	6.1	37	0.4	25	0.3	5	0.1
Petroleum	8809	0	0	11	0.1	1	<0.1	6	0.1
Combustion	8822	0	0	1	<0.1	4	0.1	0	0
Highesta	6676	1369	15.5	495	5.6	154	1.7	133	1.5

^aHighest overall probability rating from any benzene source category.

Table 6.

Descriptive statistics of source-specific probability and intensity estimates for the 71 659 employed person-years of NCI-SEER NHL participants.

Source category	Employed person-years by probability of exposure									Estimated annual benzene attributed to occupation (ppm)
	None	Low		Medium		Medium-high		High		Medium-high or High					High
										Percentile					Percentile
	N	N	%	N	%	N	%	N	%	25th	50th	75th	95th	Max	25th	50th	75th	95th	Max
Degreasing	68 469	2102	2.9	672	0.9	415	0.6	0	0.0	0.12	0.23	0.40	1.3	1.6	—	—	—	—	—
Glue	70 063	1322	1.8	249	0.3	23	0.0	2	0.0	0.16	0.50	0.97	3.7	3.7	0.25	0.25	0.25	0.25	0.25
Paint	68 202	1555	2.2	1582	2.2	242	0.3	77	0.1	0.03a	0.06	0.09	0.11	0.19	0.03	0.08	0.09	0.11	0.19
Stain/varnish	70 408	660	0.9	327	0.5	190	0.3	73	0.1
Stripping paint	70 399	1133	1.6	123	0.2	3	0.0	0	0.0	0.13	0.13	0.13	0.13	0.13	—	—	—	—	—
Laboratory work	71 391	201	0.3	59	0.1	7	0.0	0	0.0	0.05	0.05	0.10	0.10	0.10	—	—	—	—	—
Gasoline	61 504	6488	9.1	3235	4.5	1	0.0	431	0.6	0.04	0.08	0.10	0.21	1.3	0.04	0.08	0.10	0.21	1.3
Printing	71 081	150	0.2	227	0.3	177	0.2	24	0.0	0.24	1.2	2.4	23	23	0.03	23	23	23	23
Ink	71 308	74	0.1	180	0.3	72	0.1	24	0.0	1.4	2.3	2.6	14	14	2.4	11	14	14	14
Solvent	67 170	4010	5.6	305	0.4	132	0.2	41	0.1	0.14	0.75	1.0	5.8	6.4	0.94	1.6	3.2	6.3	6.4
Petroleum	71 527	0	0.0	84	0.1	2	0.0	45	0.1	0.09	0.20	0.40	0.70	0.70	0.09	0.20	0.40	0.70	0.70
Combustion	71 579	0	0.0	12	0.0	67	0.1	0	0.0	0.09	0.09	0.11	0.11	0.11	—	—	—	—	—
Highestb	51 253	12 958	18.1	5720	8.0	1049	1.5	679	0.9	0.03	0.10	0.40	3.2	23	0.04	0.08	0.12	3.2	23
Total concentrationc										0.06	0.11	0.55	3.2	34	0.05	0.09	0.20	3.2	34

^aFrequency and intensity information for the source categories paint and stain/varnish were combined to develop one overall estimate for paint and stain/varnish.

^bHighest overall probability rating from any benzene source category.

^cTotal annual concentration is the sum of all source-specific annual concentration estimates.

The median total annual concentration from all benzene source categories among jobs with a medium-high or high probability of exposure was 0.11 ppm (interquartile range: 0.06, 0.55); the median total annual concentration among jobs with a high probability of exposure was 0.09 ppm (interquartile range: 0.05, 0.20) (Table 6). The proportion of total annual concentrations of >1.0 and >2.0 ppm among jobs with a medium-high or high probability of exposure was 18 and 10%, respectively. The source categories with the highest median annual concentration among jobs with a medium-high or high probability of exposure were ink and printing (2.3 and 1.2 ppm, respectively).

Discussion

We conducted the first benzene exposure assessment in a case–control study that assigned benzene estimates using source-specific decision rules, which will facilitate future etiologic analyses and aid the transparency of the exposure assessment process. Similar to previous decision rule-based assessments conducted in NCI case–control studies (Pronk et al., 2012; Friesen et al., 2014b; Callahan et al., 2018), we developed rules that prioritized subject-specific information when available, increasing our ability to capture within-job exposure heterogeneity. In contrast, due to the mid-study incorporation of modules, we had fewer jobs with completed modules than these previous assessments. As a result, this assessment had a greater reliance on the occupational history variables in developing decision rules and on the use of job-group medians to limit the use of job-by-job review to only those exposure scenarios that required closer review.

The limited availability of module data in our study population (available for only 13% of jobs) resulted in an increased importance of the occupational history variables in developing both the probability decision rules and the job-group medians applied to jobs without module responses. The incorporation of the occupational history responses, after their systematic extraction, in the decision rules was essential in both reducing the total number of jobs that required expert review and identifying those jobs that truly needed expert review, such as those with inconsistencies between occupational history and module responses or between job group and expected tasks. The information in the responses also increased the ability to capture within-job exposure variability compared with traditional JEM-based approaches that do not consider free-text responses. Furthermore, the occupational history responses were used in defining job groups. Although job groups for which we derived medians of response patterns (e.g. task frequency) and exposure metrics were primarily based on occupation and industry groups, some groups were further stratified based on information available from the occupational history variables related to tasks, tools, and chemicals to partially reduce within-job heterogeneity. This data-based approach required that we assume that the distribution of job tasks in those jobs with modules was similar to those that did not receive modules but reduced our need for job-by-job expert judgment and extrapolation from other populations.

Expert job-by-job review was important in the exposure assessment process but minimized to focus on the jobs most in need of review. A subset of the programmable decision rules was developed by the exposure assessment team to systematically identify jobs for which exposure information was insufficient to assign estimates based on the decision rules and for which they believed job-by-job review was needed. In addition, we also used job-by-job review to confirm high exposure scenarios identified using decision rules to ensure specificity and sensitivity in the exposure assignments. Overall, job-by-job review was limited to 711 (8%) of the 8827 total jobs in the study, reducing the overall assignment burden compared with reviewing all jobs and making it feasible to have each identified exposed job reviewed by two or more experts.

The overall prevalence of high probability of benzene exposure (≥90% of workers exposed) in our study population was 5.4% (117 of 2171 participants). This was consistent with the prevalence of benzene exposure reported in the Epilymph case–control study of lymphomas, in which 5.6% (138 of 2462) of control participants had ≥90% likelihood of benzene exposure (Cocco et al., 2010) but was lower than the 14.8% of subjects considered exposed in the prospective Shanghai Women’s Health Study cohort (Friesen et al., 2012). These may be due to geographic differences in the distribution of jobs in China versus the United States.

A novel aspect of our assessment was the development of estimates separately for 12 source categories of benzene exposure. This source-specific framework increased the transparency of the exposure decision process and simplified aggregation of exposure from multiple sources. The decision rules were most effective (i.e. proportion requiring expert review was low) for assigning probability of exposure to the source categories degreasing, painting, and stain/varnish. The module questions related to these source categories collected detailed exposure information that allowed us to develop well-defined decision rules. For example, the degreasing questions asked about the use of several specific solvents (gasoline, naphtha, Stoddard solvent, etc.). The source categories for which the decision rules most often identified the need for expert review (glue, strip paint, general solvent use) were based on module responses that did not often include detailed questions on relevant tasks. This demonstrates the value of developing detailed, task-oriented module questions that can capture exposure variability within a job title.

Medium-high or high probability of overall benzene exposure was attributable to two or more source categories in 42 (15%) of the 287 jobs assigned medium-high or high probability; the source categories most frequently observed together were ink and printing, and paint and stain/varnish. Our approach also allowed us to compare the prevalence and intensity of benzene exposure across source categories. The most common source categories assigned medium-high or high probability of benzene exposure above the 0.02-ppm threshold were degreasing and gasoline. The occupation groups primarily associated with these source categories included gas station attendants, vehicle mechanics, and industrial mechanics. This was consistent with the findings of a population-based case–control study of cancer in Montreal, Canada, in which exposure to benzene was attributed primarily to the occupations of mechanics and repairmen and service station attendants (Gérin et al., 1998).

The intensity estimates developed for this study were based on published data and not measurements collected in the participants’ workplaces. Estimates of exposure intensity were derived using published measurement data when available. Our primary source of exposure measurement data was van Wijngaarden and Stewart’s (2003) review, which provided benzene exposure levels for a wide variety of jobs and industries within the same geographic region as our study (North America). We considered using the benzene determinants of exposure models described by Hein et al. (2007); however, because the variables used in the models could not easily be extracted from our study participants’ occupational histories, we developed a series of meta-regression mixed-effects models to evaluate exposure differences by specific tasks or jobs that were more easily extracted from the occupational questionnaires.

In our population, exposure intensity levels among participants with jobs assigned medium-high or high probability were generally low. Overall, 26, 18, and 10% of these exposed participants had one or more years exceeding 0.5, 1.0, and 2.0 ppm, respectively. These intensity levels were lower than those developed for a multi-center case–cohort study of benzene-exposed workers in China (Portengen et al., 2016), in which maximum exposure levels exceeded 3 mg/m³ (0.94 ppm) for 76% of the study population, and in the Shanghai Women’s Health Study cohort (Friesen et al., 2012). The lower intensity levels in our study compared with the two studies in China were likely due to differences in country (United States versus China, respectively) and sample type (full-shift or task-based personal samples versus short-term area samples, respectively) of the measurements used to derive the estimates, and sampling strategy (inspection versus research). In addition, in this study, we were able to capture some subject-specific differences in the frequency of conducting benzene-exposed tasks, which was not done in the two studies in China.

The strengths of our approach were its transparency and its use of information beyond the standard occupation group, including use of free-text description of work tasks and structured questions on work tasks, their frequency, and the use of specific chemicals, to derive subject-specific exposure estimates. The extent to which this assessment captured exposure information that would not have been captured with a JEM will be evaluated in the future. Our approach also had several limitations that should be noted. Similar to previous decision rule-based approaches, the rules were based on determinants and responses to open-ended questions that could easily be recalled by the participants, and the intensity estimates were derived from published exposure data that were not specific to the participants’ workplaces and that may not be fully representative of typical workplace exposures. As has been observed in studies for other exposures, benzene exposure data prior to 1970 were sparse, limiting our ability to estimate time-trends during the earliest time period of interest. Measurement data were not available for all benzene exposure scenarios and thus intensity estimates were sometimes extrapolated from scenarios that the exposure assessment team deemed similar. In addition, because of sparse data, intensity estimates for the source category degreasing, which incorporated differences in intensity by degreasing method and solvent temperature, were at times extrapolated from measurement data collected for toluene or other related solvents and then adjusted for the percent benzene content.

In summary, we described a programmable decision rule-based approach for estimating occupational benzene exposure that prioritized subject-level data when available and used data-derived job-group medians for jobs without modules. We supplemented the decision rules with job-by-job review performed by two or more exposure assessors to maximize the specificity of capturing jobs with medium-high and high probability of exposure. The applied framework captured subject-specific variability in work tasks wherever possible, provided transparency to the exposure decision process, and will facilitate sensitivity analyses in future etiologic evaluations. This approach and resulting decision rules can be used to assist researchers in developing benzene estimates for other studies but would require adaptation for geographic differences and inclusion of exposure scenarios not found in this population.

Disclaimer

The lead author, Pamela J. Dopart, worked for the Division of Cancer Epidemiology and Genetics at the time the project was funded. The authors declare no other conflict of interest relating to the material presented in this article.