Pooled study of occupational exposure to aromatic hydrocarbon solvents and risk of multiple myeloma

Abstract

Objectives

To investigate the association between occupational exposure to aromatic hydrocarbon solvents and risk of multiple myeloma (MM) in a large, consortium-based study.

Methods

We pooled data on 2854 cases and 10 743 controls from nine studies participating in the InterLymph consortium. Occupational exposures to benzene, toluene and xylene were assigned by a job–exposure matrix, coupled with ‘correction’ of exposure probability by self-reported or expert-assessed exposure from the individual studies. Cumulative intensity was calculated as the job-specific exposure intensity multiplied by job duration, summed across jobs. Associations were estimated using logistic regression, with inclusion of covariates for study matching factors and other potential confounders. We repeated our main analysis using random-effects meta-analysis to evaluate heterogeneity of effect.

Results

Benzene, toluene and xylene were each associated with MM. For the three solvents, the highest quartile of high-probability cumulative intensity exposure (vs unexposed) was associated with 42% to 63% increased risks of MM. Associations with toluene and xylene exposures were fairly consistent and robust to sensitivity analyses. The estimated effect for benzene was moderately heterogeneous between the studies. Each solvent’s association with MM was stronger for exposure occurring within 20 years of diagnosis than with exposure lagged by more than 20 years.

Conclusions

Our study adds important evidence for a role of aromatic hydrocarbon solvents in causation of MM. The difficulty in disentangling individual compounds in this group and a lack of data on potential carcinogenicity of toluene and xylene, in widespread current use, underscore a need for further epidemiological evaluation.

INTRODUCTION

There is a need for data characterising potential carcinogenic risks from current use chemicals. Exposure to aromatic hydrocarbon solvents is widespread both in certain occupations and in the ambient environment. These chemicals, including benzene, toluene and xylene, occur naturally in petroleum and are also found in a wide variety of petroleum-derived products such as vehicle fuels, paints, coatings, resins and adhesives. Benzene is a known carcinogen for its relationship with acute myeloid leukaemia (AML).¹ The relationship of benzene with other types of leukaemias and lymphomas is not as well established, but remains of interest due to associations with non-Hodgkin’s lymphoma and multiple myeloma (MM) in some studies. Toluene and xylene have been far less studied as potential carcinogens, and much of the epidemiological evaluation of these two solvents is based in occupational cohorts with limited exposure assessment and little individual-level information.

Several case–control studies with detailed data on job histories and workplace exposures of study participants from the general population have examined whether aromatic hydrocarbon solvents contribute to risk of MM. Although these case–control studies include large numbers of MM cases compared with industry-based cohort studies, the statistical power is limited by the low frequency of persons with high-intensity occupational solvent exposure. Pooled studies and meta-analyses of individual, regionally based case–control studies offer promise in summarising risk factors for a rare cancer like MM.

To investigate a potential role of aromatic hydrocarbons in causation of MM, we pooled data from multiple case–control studies participating in the International Lymphoma Epidemiology Consortium (InterLymph). An earlier consortium-based pooled analysis of five case–control studies found a 38% increased risk (95% CI 0.96 to 1.85) for employment in occupations with high exposure to any organic solvent,² demonstrating the potential use of our consortium-based approach. Our research on specific solvents includes nine studies with information on lifetime occupational history, comprising 2854 cases and 10 743 controls.

METHODs

Study population

The study population was pooled from case–control studies of MM participating in InterLymph. A study was eligible for inclusion in this pooled analysis of solvents if its protocol included collection of information on lifetime occupational history. A summary of the nine participating studies^3–11 is shown in table 1.

Table 1.

Studies included in the InterLymph pooled study of occupational solvent exposure

study label	FHCRC-1980s	LACCMM	Population Health	Italian Multicentre	Cross-Canada	NCI-Yale	Epilymph	FHCRC-SEER	ENGELA
Study name	Fred Hutchinson Cancer Research Center-1980s (FHCRC-1980s)	Los Angeles County Case–Control Study of Multiple Myeloma (LACCMM)	Population Health	Italian Multicentre	Cross-Canada	National Cancer Institute–Yale (NCI– Yale)	Epilymph	Fred Hutchinson Cancer Research Center–Surveillance, Epidemiology, and End Results (FHCRC–SEER)	L’Etude des Facteurs Environmentauxet Genetique des Lymphomes de l’Adulte (ENGELA)
Study centre	USA: Georgia, Michigan, Washington State, Utah	USA: Los Angeles County	USA: Georgia, Michigan, New Jersey	Italy: Firenze, Forli, Imperia, Latina, Ragusa, Siena, Torino	Canada: Alberta, British Columbia, Manitoba, Ontario, Quebec, Saskatchewan	USA: Connecticut	Europe: Czech Republic, France, Germany, Ireland, Italy,Spain	USA: Michigan, Washington State	France: Bordeaux, Brest, Caen, Lille, Nantes, Toulouse
Years	1977–1981	1985–1990	1986–1989	1991–1993	1991–1994	1996–2000	1998–2004	1999–2002	2000–2004
N cases	681	274	569	258	327	182	278	177	108
N controls	1679	276	2137	1095	1428	705	2464	481	478
Source of controls	Population	Population	Population	Population	Population	Population	Population or hospital	Population	Hospital
Matching	Frequency-matched by age, sex, study centre	Pair-matched by age, sex, race	Frequency-matched by age, sex, race, study centre	Frequency-matched by age, sex, study centre	Frequency-matched by age, study centre	Frequency-matched by age	Frequency-matched by age, sex, study centre	Frequency-matched by age, sex, study centre	Pair-matched by age, sex, study centre
Age in years (mean (SD))	62.6 (10.0)	61.1 (9.0)	62.2 (10.9)	52.3 (13.5)	56.1 (15.8)	62.5 (14.0)	56.8 (15.8)	59.4 (10.3)	59.0 (9.0)
Male sex (%)	54.8	54.7	60.6	50.5	100	0	53.9	52.0	63.0
Education less than high school graduate (%)	45.0	20.4	40.8	60.2	36.1	12.3	46.3	11.6	36.5
White race (%)	76.2	68.4	56.7	Not collected	Not collected	92.7	Not collected	76.2	Not collected
Information available for adjustment of JEM-assigned exposure probability	Self-reported exposure, specific questions for ‘benzene’, ‘cleaning solvents’	Self-reported exposure, open-ended, prompted by list including ‘benzene’, ‘toluene’, ‘xylene’, ‘other degreasers or solvents’	None	Expert assessment: benzene, toluene, xylene, aromatic hydrocarbons	Self-reported exposure, specific question for ‘solvents’	Self-reported Exposure, open-ended, prompted by list including ‘benzene’, ‘toluene’, ‘xylene’, ‘aromatic hydrocarbons, unspecified’, ‘solvent, unspecified’	Expert assessment: benzene, toluene, xylene, BTX	None	Expert assessment: benzene, toluene/xylene

JEM, job–exposure matrix.

The case–control studies enrolled histologically confirmed incident primary diagnoses of MM (ICD-O-3 973, ICD-9 203) that occurred during the respective enrolment periods. Controls were identified in the general population or from participating hospitals, and were frequency-matched or pair-matched to the cases by variables including age, sex and, in some studies, study centre or race (table 1).

We obtained data on occupational history from each of the participating studies, including, for each job reported by a subject, the job title, industry, dates of employment and self-reported exposures. We also requested, where available, solvent exposures that were assessed by experts (eg, industrial hygienist) in the original, individual studies. Subjects who never held a job outside the home were excluded from the database (<1% of subjects), as were 138 subjects with missing data on education or race. The final analytical dataset included 2854 cases and 10 743 controls.

Exposure coding

All exposure coding for the pooled study was conducted blinded to case status. We linked subjects’ occupational histories to job–exposure matrices (JEMs) for aromatic hydrocarbons, specifically for benzene, toluene and xylene; these JEMs were developed for, and applied in, previous epidemiological studies.^{12 13} In order to link the JEMs, all reported occupations and industries were consistently coded using either US Standard Occupational Classification 1980 and Standard Industrial Classification 1987 codes or the International Standard Classification of Occupations 1968 and Nomenclature of Economic Activities 2000. For each type of solvent, the JEM assigned an exposure probability (the likelihood of exposure) and intensity (the average concentration of exposure), according to the job and industry. Probability levels represented low (ie, possible (1)), medium (ie, probable (2)) and high (ie, certain (3)), and intensity levels represented low (1), medium (2) and high (3). Jobs with no linkage to the JEM were considered unexposed. In some study observations where the job was given but industry was missing, we assigned the job-specific exposure intensity with low probability.

We reviewed JEM-assigned job-level exposure probability in conjunction with additional information from the individual studies on self-reported exposure or expert assessment of exposure, and used those sources to ‘correct’ the exposure probability. In particular, we focused on increasing the specificity of solvent exposure assignments, by downgrading the exposure probability for a certain job if there was no additional evidence of exposure. We used questionnaire responses about solvents to correct exposure probabilities in four of the nine studies (table 1). For example, an item on the Fred Hutchinson Cancer Research Center-1980s (FHCRC-1980s) study questionnaire asked if the subject was ever exposed to benzene. We reviewed the questionnaire responses for all FHCRC-1980s subjects with JEM-assigned benzene exposure and downgraded the exposure probability for each exposed job (by 1) if the person did not report benzene exposure. In three of the nine case–control studies (L’Etude des Facteurs Environmentaux et Genetique des Lymphomes de l’Adulte (ENGELA), Epilymph, Italian Multi-centre), experts had assessed solvent exposures in the original studies, based on detailed job-specific questionnaires answered by the study participants. For these studies, we corrected the JEM-assigned probability by downgrading probability (by 1 or 2) if the experts had assessed a lower probability than the JEM. Only two of the nine studies did not have any additional solvent-related information for correction of exposure probability, and for these studies, the JEM-assigned probability was retained.

We evaluated solvent associations with MM according to the highest intensity of exposure across all jobs (categories of low, medium, high), the duration of exposure across jobs (categories of ≤10 years, 11–20 years, >20 years) and the cumulative intensity of exposure (split by quartiles of the distribution among all subjects). We defined cumulative intensity for each solvent as the job-specific exposure intensity multiplied by job duration, summed across all jobs. Exposure coding in the pooled study was also conducted specifically for jobs with high-probability exposure (probability=3, after correction by self-reported or expert-assessed exposure from the individual studies) and exposure during lag periods of interest (exposure occurring >20 years before MM diagnosis or reference date; exposure within 20 years of MM diagnosis/reference date). Finally, we coded exposure variables limited to jobs held before 1980, a year before the first International Agency for Research on Cancer (IARC) evaluation of benzene, in which the chemical was formally recognised as a carcinogen (a cause of AML).¹⁴ In addition to the individual aromatic hydrocarbons, we coded exposure variables for benzene, toluene, or xylene (BTX) exposure, reflecting exposure to any of the three solvents.

Statistical analysis

We estimated the risk of MM in relation to aromatic hydrocarbon solvent exposures using unconditional logistic regression to produce ORs and 95% CIs, with the unexposed subjects as the referent in each model. We considered two-tailed p values less than 0.05 to indicate statistical significance. When evaluating high-probability exposures, subjects with lower-probability exposure were retained in the model in a separate category, thereby including the same study population and reference group in each model. Tests for trend across categories of exposure were performed by modelling the intracategory median (of cases and controls combined) as a continuous variable, with values for unexposed participants set to zero. All models included a minimal set of potential confounders: age (<50, 50–59, 60–69, ≥70 years), sex (male, female), education (did not graduate from high school, high school graduate, college graduate), race (white/Caucasian, black/African American, other race, data on race not collected), study (indicator variable for each of the nine case–control studies) and study centre (for studies conducted collaboratively in multiple locations). In separate analyses, we evaluated potential confounding by farming occupation, of concern as agricultural work is arguably the most consistent of suspected occupational risk factors for MM¹⁵ and because those jobs were fairly common in the study population (14.3% of cases, 13.2% of controls). Effect estimates were produced for the entire pooled dataset, as well as by sex, age group (<50 years, ≥50 years) and continent (North America and Europe). We conducted analyses limiting to studies with and without exposure ‘correction’. We also conducted analyses after excluding one study at a time, to evaluate the influence of any individual study on the results. All pooled data analysis was conducted using SAS V.9.3 (SAS Institute, Cary, North Carolina, USA).

We repeated the main analysis using a meta-analytical approach to formally evaluate heterogeneity of effect among the studies. Effects were estimated by study with adjustment for the relevant covariates and then combined in a random-effects meta-analysis. Meta-analysis was conducted using Stata V.12 software (StataCorp, College Station, Texas, USA).

RESULTS

The 2854 cases were slightly more likely than the 10 743 controls to be coded with any occupational exposure to aromatic hydrocarbons in the pooled study (33.5% of cases, 31.2% of controls; online supplementary table 1). Exposure frequencies were considerably reduced when limiting to high-probability exposure, with 11.1% of controls exposed to any BTX. Exposure frequencies were higher among men than women, in older compared with younger subjects, and among those with low education or black race. Exposure assignments for toluene and xylene were extremely similar; 99.8% of subjects with exposure to xylene were also exposed to toluene, and this was also reflected in a correlation of 0.99 between cumulative intensity of exposure for the two chemicals. There was less overlap with benzene; 82.3% of subjects with benzene exposure were also classified with exposure to toluene or xylene, with correlations of 0.81 for benzene cumulative intensity with either of the other two solvents. The median exposure duration was 13 years for benzene and 10 years for both toluene and xylene. The vast majority of subjects with BTX exposure (96.8%) was exposed in jobs held prior to 1980.

Cumulative intensity of benzene exposure was associated with increased risk of MM (table 2); the association was particularly observed with the highest levels of high-probability exposure. There was 42% increased risk of MM for the highest quartile versus unexposed (OR 1.42, 95% CI 1.08 to 1.86), and the p value for trend across exposure categories was 0.01. The result was somewhat strengthened by inclusion of a model covariate for toluene or xylene exposure (OR 1.52, 95% CI 1.11 to 2.09), but was unaffected by adjustment for farming occupation (OR 1.43, 95% CI 1.09 to 1.87). In analyses of exposure defined based on jobs held before 1980, the risk estimate was slightly higher than in the main analysis (OR 1.54, 95% CI 1.16 to 2.03), although the exposure–response gradient was similar, with the association primarily limited to the highest category of exposure. There was no statistically significant association with benzene exposure intensity or duration (online supplementary table 2), although there was a suggestive trend of increasing risk by duration (p trend=0.05). The results from random-effects meta-analysis (online supplementary table 3) were similar to those from the pooled analysis, with a meta-estimate of 1.63 (95% CI 1.06 to 2.50) for the highest quartile of cumulative intensity of high-probability exposure versus unexposed (compared with OR 1.42 from the pooled analysis). The meta-analysis indicated moderate heterogeneity of the effect among the studies (I²=30.5%, p=0.17; online supplementary figure 1). In analyses by lag period (table 3), the association was stronger and the exposure–response trend more apparent for exposure occurring within 20 years of the diagnosis/reference date than for exposure lagged by at least 20 years.

Table 2.

Association of aromatic hydrocarbon solvent cumulative intensity exposure^* with risk of multiple myeloma^†‡

	Any exposure (quartiles of cumulative intensity)				High-probability exposure (quartiles of cumulative intensity)§¶
		Cases	Controls			Cases	Controls
Benzene	Unexposed	1921	7513	Referent	Unexposed	1921	7513	Referent
				–	Low-probability/medium-probability exposure	637	2210	0.95 (0.85 to 1.06)
					High-probability exposure
	Quartile 1	179	769	0.91 (0.76 to 1.09)	Quartile 1	43	220	0.76 (0.54 to 1.07)
	Quartile 2	232	801	0.94 (0.80 to 1.11)	Quartile 2	76	252	1.06 (0.81 to 1.40)
	Quartile 3	231	811	0.94 (0.80 to 1.12)	Quartile 3	67	231	1.07 (0.80 to 1.44)
	Quartile 4	271	743	1.16 (0.98 to 1.37)	Quartile 4	88	209	1.42 (1.08 to 1.86)
				P trend=0.10				P trend=0.01
Toluene	Unexposed	2063	7969	Referent	Unexposed	2063	7969	Referent
				–	Low-probability/medium-probability exposure	493	1754	0.90 (0.80 to 1.02)
					High-probability exposure
	Quartile 1	164	704	0.91 (0.75 to 1.09)	Quartile 1	58	259	0.86 (0.63 to 1.16)
	Quartile 2	189	669	0.90 (0.75 to 1.07)	Quartile 2	59	224	0.96 (0.71 to 1.30)
	Quartile 3	193	671	0.94 (0.79 to 1.12)	Quartile 3	66	228	1.09 (0.81 to 1.46)
	Quartile 4	224	628	1.12 (0.94 to 1.34)	Quartile 4	92	204	1.63 (1.25 to 2.12)
				P trend=0.22				P trend=0.0007
Xylene	Unexposed	2107	8103	Referent	Unexposed	2107	8103	Referent
				–	Low-probability/medium-probability exposure	456	1640	0.88 (0.78 to 1.00)
					High-probability exposure
	Quartile 1	153	672	0.88 (0.72 to 1.06)	Quartile 1	58	260	0.85 (0.63 to 1.15)
	Quartile 2	178	650	0.87 (0.72 to 1.04)	Quartile 2	54	214	0.91 (0.66 to 1.24)
	Quartile 3	181	622	0.96 (0.80 to 1.15)	Quartile 3	67	231	1.09 (0.81 to 1.45)
	Quartile 4	215	602	1.12 (0.95 to 1.33)	Quartile 4	90	198	1.63 (1.25 to 2.13)
				P trend=0.22				P trend=0.0006
BTXf	Unexposed	1898	7391	Referent	Unexposed	1898	7391	Referent
					Low-probability/medium-probability exposure	619	2194	0.92 (0.82 to 1.03)
					High-probability exposure
	Quartile 1	219	917	0.92 (0.78 to 1.08)	Quartile 1	60	293	0.77 (0.57 to 1.03)
	Quartile 2	217	746	0.93 (0.79 to 1.11)	Quartile 2	72	241	1.08 (0.81 to 1.43)
	Quartile 3	213	805	0.87 (0.73 to 1.03)	Quartile 3	78	282	1.01 (0.77 to 1.33)
	Quartile 4	285	774	1.16 (0.98 to 1.36)	Quartile 4	103	229	1.57 (1.22 to 2.03)
				P trend=0.14				P trend=0.001

^*Cumulative intensity estimated as job-specific exposure intensity multiplied by job duration, summed across all jobs.

^†ORs and 95% CIs from logistic regression models, with adjustment for study, study centre, age, sex, education and race.

^‡P value for a continuous variable with values set as the median of each cumulative intensity quartile category.

^§Models for high-probability exposure include a separate category for low-probability/medium-probability exposure; this category is not included in trend test.

^¶Ns may be slightly lower for high-probability exposure than for any exposure, if missing duration in high-probability jobs.

^fCoded according to benzene, toluene or xylene exposure in a job.

Table 3.

Association of cumulative intensity^* of high-probability exposure to aromatic hydrocarbon solvents with risk of multiple myeloma^†‡§ within lag periods of interest

		Cases	Controls	Exposure occurring >20 years before diagnosis date	Cases	Controls	Exposure within 20 years of diagnosis date
Benzene	Unexposed	1921	7513	Referent	1921	7513	Referent
	Quartile 1	47	204	0.87 (0.62 to 1.21)	19	122	0.62 (0.37 to 1.02)
	Quartile 2	67	195	1.18 (0.88 to 1.58)	31	122	0.96 (0.64 to 1.45)
	Quartile 3	57	150	1.31 (0.95 to 1.81)	37	99	1.38 (0.92 to 2.05)
	Quartile 4	72	186	1.19 (0.89 to 1.59)	44	96	1.87 (1.28 to 2.73)
				P trend=0.16			P trend=0.002
Toluene	Unexposed	2063	7969	Referent	2063	7969	Referent
	Quartile 1	52	185	1.05 (0.76 to 1.45)	21	125	0.73 (0.45 to 1.17)
	Quartile 2	58	186	1.12 (0.82 to 1.53)	31	128	0.92 (0.62 to 1.39)
	Quartile 3	58	195	1.01 (0.74 to 1.37)	33	109	1.26 (0.83 to 1.89)
	Quartile 4	74	169	1.46 (1.09 to 1.96)	53	92	2.31 (1.61 to 3.31)
				P trend=0.03			P trend=0.001
Xylene	Unexposed	2107	8103	Referent	2107	8103	Referent
	Quartile 1	51	187	1.01 (0.73 to 1.40)	21	122	0.74 (0.45 to 1.19)
	Quartile 2	57	183	1.11 (0.81 to 1.51)	25	116	0.80 (0.52 to 1.26)
	Quartile 3	57	189	1.02 (0.74 to 1.39)	36	118	1.24 (0.84 to 1.84)
	Quartile 4	73	168	1.45 (1.08 to 1.94)	52	86	2.42 (1.68 to 3.48)
				P trend=0.03			P trend=0.001
BTX¶	Unexposed	1898	7391	Referent	1898	7391	Referent
	Quartile 1	51	201	0.94 (0.68 to 1.30)	24	145	0.68 (0.43 to 1.06)
	Quartile 2	80	245	1.14 (0.87 to 1.50)	33	141	0.89 (0.60 to 1.32)
	Quartile 3	71	203	1.16 (0.87 to 1.55)	40	128	1.21 (0.83 to 1.77)
	Quartile 4	79	197	1.30 (0.98 to 1.72)	57	109	2.07 (1.47 to 2.92)
				P trend=0.09			P trend=0.0007

^*Cumulative intensity estimated as job-specific exposure intensity multiplied by job duration, summed across all jobs. Note that a subject can be included as exposed during both lag periods, with cumulative intensity calculated separately according to the number of years worked in exposed jobs during each period.

^†ORs and 95% CIs from logistic regression models, with adjustment for study, study centre, age, sex, education and race.

^‡P value for a continuous variable with values set as the median of each cumulative intensity quartile category.

^§Models include separate categories for low-probability/medium-probability exposure and for high-probability exposure outside of lag period (results not shown); these categories are not included in trend test.

^¶Coded according to benzene, toluene or xylene exposure in a job.

The results for toluene and xylene were very similar to each other due to the overlapping exposure categorisations. For brevity, we highlight the results for toluene in the text. There was a pattern of increasing MM risk with greater cumulative intensity of toluene or xylene exposure (table 2), primarily when considering high-probability exposure. For toluene, there was 63% increased risk of MM for the highest quartile versus unexposed, and a statistically significant trend across exposure categories (p trend=0.0007), although small elevations in the other quartile categories were not statistically significant. The observed association with the highest exposure quartile was only slightly attenuated with adjustment for benzene exposure (OR 1.59, 95% CI 1.18 to 2.15), and there was no confounding by farming occupation. Toluene or xylene exposure based on jobs held before 1980 produced nearly identical associations with MM as the main exposure definition. Highest-intensity exposure and duration of exposure were also associated with increased risk of MM, for the highest categories of high-probability exposure (online supplementary table 2). Random-effects meta-analysis of cumulative intensity exposure (online supplementary table 3) showed similar results to the pooled analysis, with a meta-estimate for toluene of 1.79 (95% CI 1.35 to 2.37) for the highest quartile of high-probability exposure versus unexposed, and no indication of heterogeneity between studies (I²=0%, p=0.73; forest plot shown as online supplementary figure 2). As for benzene, analyses by lag period revealed stronger associations for exposures occurring within 20 years of the diagnosis/reference date than for exposure lagged by at least 20 years; nevertheless, there was a significant association with MM in both lag periods (table 3). The results for BTX exposure reflect a combination of the associations with the three individual solvents (tables 2 and 3).

We focused sensitivity analyses on the results for cumulative intensity of high-probability exposure (figures 1 and 2; online supplementary figure 3). Associations were stronger in men than women; however, results for women were based on sparse data. Risk increases were observed in both younger (<50 years) and older age groups. Risk increases were seen in both the six North American and three European studies, although estimates based on the European data were far less precise. Results were not greatly sensitive to exclusion of any one particular study. Positive associations were most reduced with exclusion of the Cross-Canada study, which was composed of only men. The strongest risk increases were observed with exclusion of the two studies lacking information for ‘correction’ of JEM-assigned exposure (benzene: OR 1.96, 95% CI 1.32 to 2.90; toluene: OR 2.27, 95% CI 1.50 to 3.45), particularly for exposure within 20 years of diagnosis (benzene: OR 2.73, 95% CI 1.61 to 4.64; toluene: OR 3.24, 95% CI 1.93 to 5.45).

Figure 1.

Association of benzene exposure with risk of multiple myeloma (cumulative intensity of high-probability exposure, highest quartile vs unexposed). ENGELA, L’Etude des Facteurs Environmentaux et Genetique des Lymphomes de l’Adulte; FHCRC-SEER, Fred Hutchinson Cancer Research Center–Surveillance, Epidemiology, and End Results; FHCRC-1980s, Fred Hutchinson Cancer Research Center-1980s; LACCMM, Los Angeles County Case–Control Study of Multiple Myeloma.

Figure 2.

Association of toluene exposure with risk of multiple myeloma (cumulative intensity of high-probability exposure, highest quartile vs unexposed). ENGELA, L’Etude des Facteurs Environmentaux et Genetique des Lymphomes de l’Adulte; FHCRC-SEER, Fred Hutchinson Cancer Research Center–Surveillance, Epidemiology, and End Results; FHCRC-1980s, Fred Hutchinson Cancer Research Center-1980s; LACCMM, Los Angeles County Case–Control Study of Multiple Myeloma.

DISCUSSION

In our consortium-based pooled analysis of nine case–control studies, we found evidence for increased risk of MM from occupational exposure to aromatic hydrocarbons. These associations were strongest with high probability of exposure and with enhancement of exposure specificity using subject-specific information from their occupational histories. There was negligible heterogeneity between studies of the effects associated with toluene and xylene exposures, and these associations were observed within multiple subgroups of the study population and across all analyses, further indicating consistency. A moderate heterogeneity across studies was observed for the association with benzene. The overlap between specific solvent exposures was substantial, limiting interpretation.

The objective of our pooled analysis approach was to enable a statistically well-powered examination of infrequent occupational solvent exposures in relation to a rare cancer. Three of the case–control studies included in our pooled analysis had previously examined occupational benzene exposure. The Italian Multicentre study (263 cases, 1100 controls) found increased MM risk for medium-intensity/high-intensity benzene exposure (OR 1.9, 95% CI 0.9 to 3.9), and the risk estimates were more elevated with at least 10 years’ exposure duration (OR 4.1) or latency (OR 2.3).⁵ Cocco et al observed non-significant 40% increased risk of MM in association with benzene exposure in the Epilymph study (277 cases and 2460 controls).⁶ There was no association in the ENGELA study conducted in France (56 cases, 313 controls).⁷ Toluene and xylene were associated (without statistical significance) with increased MM risk in the Italian Multicentre study with longer duration exposure,⁵ but there was no association in Epilymph.⁶ Clearly, our analysis of nine studies had adequate statistical power to detect modest associations and to characterise heterogeneity of the effect between case–control studies conducted in various regions of North America and Europe, improving on individual, regional analyses. The previously published results differ in some instances from our estimated study-specific effects (online supplementary figures 1 and 2), primarily due to disagreement between the JEM used in our study and expert-assessed exposure in the individual studies. We capitalised on such disagreement to improve exposure coding in our pooled study, as described below.

A concern with the use of JEMs is that persons in the same job and industry receive the same exposure assignment, whereas in reality, exposures may differ between people in the same job based on different duties or workplace practices. We strategised to decrease misclassification in our pooled study by enhancing the specificity of exposure coding. In general population-based case–control studies with low exposure prevalence (as for occupational solvents), even small numbers of individuals incorrectly classified as exposed can result in bias that attenuates effect estimates towards the null value.^{16 17} To that end, we used questionnaire data and expert assessment of solvents from the individual studies to ‘correct’ the JEM-assigned exposure probability, for a more stringent (specific) definition of exposure in the pooled study. The increased risks with aromatic hydrocarbons were essentially only observed with the ‘corrected’ high-probability exposures. Furthermore, the associations were strongest when limiting to the seven studies with information available for exposure correction. The pattern of increasing strength of association with enhanced specificity of exposure coding strengthens the plausibility of our findings; however, it is impossible to know whether the increased ORs are due to a more specific (causal) effect or introduced bias from using self-reported exposures. Self-reporting of exposure can suffer from misclassification due to poor or biased recall. We are assured by the fact that such issues are unlikely to have affected subjects’ reporting of job title,¹⁷ used as the primary basis for exposure coding in our study by linkage to the JEM, despite possibly affecting the self-reported exposures that we used for enhancing the specificity of exposure assignments. We are further assured that the associations we observed are not entirely due to misclassification or bias from self-report, as we did not observe a positive association in every study with exposure correction based on self-report (such as for benzene exposure in the FHCRC-1980s study), and, likewise, there were positive associations observed in studies without self-reported information available for exposure correction (ie, with probability based on the JEM alone, as for benzene and toluene exposure in the Population Health study).

In their 2017 assessment of benzene, IARC found sufficient evidence that benzene causes AML/acute non-lymphocytic leukaemia, and limited evidence for a relationship between benzene and other lymphohematopoietic cancers, including MM.¹ IARC noted that most cohort studies found no association with MM¹⁸; however, one meta-analysis found that the association between benzene and MM was strongest in cohort studies with presumably higher-quality case classification, exposure assessment and greater exposure contrasts.¹⁹ While our pooled study results provide some support for a causal association between benzene and risk of MM, inconsistencies limit the strength of our conclusions. The association was found only at the highest levels of exposure and was moderately heterogeneous between the studies. In addition, there were null or merely suggestive effects observed in analyses of exposure intensity and duration metrics, and with a longer lag between exposure and diagnosis. Nevertheless, there was an apparent exposure–response gradient with benzene exposure occurring less than 20 years before diagnosis.

The stronger association with more recent exposure is some-what consistent with observations in cohort data in which relative risks of benzene in relation to MM declined with longer follow-up (eg, >20 years) in some studies, particularly when exposure had terminated decades earlier.²⁰ This finding may imply a late-stage biologic mechanism for benzene-induced myeloma-genesis, such as through altered immunity. Such a mechanism is supported by studies that observed immune effects in workers following exposure, such as decreases in circulating leucocytes and antibodies,²¹ and lowered CD4⁺:CD8⁺ ratio.²² Furthermore, well-documented increases in MM incidence among patients with AIDS and certain autoimmune diseases¹⁵ identify immune alteration as an important biologic mechanism in MM aetiology. Benzene alteration of immune competence could theoretically promote tumour development, acting at a later stage than initiating genotoxic changes (which may also be induced by benzene).¹⁸ An alternate explanation for associations limited to more recent exposure is competing risks, if subjects exposed to aromatic hydrocarbons were more likely to die earlier, from causes other than MM. Finally, enhanced or biased reporting of more recent exposures may have resulted in stronger associations with the shorter lag.

Toluene and xylene were individually evaluated in the IARC monograph programme in the late 1990s, by which each solvent was categorised as group 3, ‘not classifiable as to its carcinogenicity in humans,’ in part due to inadequate evidence in humans.^{23 24} Epidemiological studies have been particularly lacking for toluene or xylene in relation to MM. The evidence for a potential role of toluene and/or xylene in causation of MM was moderately strong and consistent in our pooled study. There was little heterogeneity of the effect among the studies, and the associations were robust to exclusion of any one of the studies. Positive associations were observed with cumulative intensity of exposure, as well as with the metrics of exposure intensity and duration. Furthermore, increased risks were observed with both distant exposure (>20 years) and exposure more proximal to diagnosis (≤20 years), perhaps suggesting biologic mechanisms acting in both early-stage and later-stage myelomagenesis.

In conclusion, our study adds important evidence for a role of aromatic hydrocarbon solvents in causation of MM. The difficulty in disentangling individual compounds in this group, in addition to a lack of data on potential carcinogenicity of toluene and xylene, in widespread current use, underscore a need for further epidemiological evaluation.

Key messages.

What is already known about this subject?

• Benzene is a carcinogen known for its relationship with acute myeloid leukaemia; however, its association with other lymphohaematopoietic cancers is less clear. Several previous studies and meta-analyses have observed an association between benzene exposure and risk of multiple myeloma, but the evidence has been inconsistent. Other aromatic hydrocarbon solvents, such as toluene and xylene, have been little studied as potential human carcinogens.

What are the new findings?

• Exposure to each of the aromatic hydrocarbons was associated with increased risk of multiple myeloma. The results for toluene and xylene were rather consistent between studies included in our pooled analysis and in various subgroups of the study population, including men and women, and older and younger persons. A similar association with benzene was moderately heterogeneous between the studies. For all three solvents, associations were strongest for exposure occurring within 20 years of diagnosis, compared with exposure lagged by more than 20 years.

How might this impact on policy or clinical practice in the foreseeable future?

• Our study adds important evidence for a role of aromatic hydrocarbon solvents in causation of multiple myeloma. The difficulty in disentangling individual compounds in this group hinders conclusions for effects specific to any one of the compounds. Toluene and xylene are in current widespread use, and a lack of data on these compounds, coupled with our suggestive findings, begs for further epidemiological evaluation to support assessment of policy.