Occupation, exposure to polycyclic aromatic hydrocarbons and laryngeal cancer risk

Authors


  • Fax: +49-6221-56-59-48.

Abstract

Primary risk factors for laryngeal cancer are smoking and alcohol. The relevance of occupational exposures in the etiology of laryngeal cancer is not yet clarified. Some studies have suggested various occupational agents as additional causal risk factors. A population-based case–control study 1:3 frequency matched by age and gender on laryngeal cancer was carried out in southwest Germany with 257 cases (236 males and 21 females between the ages of 37–80, histologically confirmed and diagnosed between January 5, 1998 and December 31, 2000) and 769 population controls (702 males, 67 females). Occupational exposures and other risk factors were obtained with face-to-face interviews using a detailed standardized questionnaire. The complete individual work history was assessed. A detailed assessment of work conditions was obtained by job-specific questionnaires for selected jobs known to be associated with exposure to potential carcinogens. A specific substance list was used as second method for exposure assessment. Blood samples were taken from all individuals for genotype analysis. A strong effect of polycyclic aromatic hydrocarbons exposure on laryngeal cancer risk after adjustment for smoking and alcohol (odds ratio [OR] = 5.2, 95% confidence interval [CI] = 1.6–17.1) was observed for concordant exposure classified with both methods, and a clear dose-response (p < 0.01 for linear trend) for exposure duration. Our findings are supported by risks associated with occupational groups in which this exposure is a priori considered likely. A differential effect by glutathione-S-transferases-M1 genotype was found, however, small numbers do not allow firm conclusions on effect modification. Our study contributes to classifying polycyclic aromatic hydrocarbons as a risk factor for laryngeal cancer. © 2005 Wiley-Liss, Inc.

Laryngeal cancer is the second most common cancer of the respiratory tract with an age standardized (Segi World population) incidence rate for the year 2000 of about 10 in 100,000 in males and 1 in 100,000 in females in Europe. The range is from 2.1 (Sweden) to 16 (Belarus) in males, and from 0.18 (Latvia) to 1.6 (Macedonia) in females.1 According to recent data,2 the crude incidence in Germany is 7 in 100,000 in males and 1 in 100,000 in females. The mortality is considerably lower because of the relatively good prognosis. Most tumors are squamous cell carcinomas.

Major risk factors are smoking and extensive alcohol consumption. There is clear evidence that laryngeal cancer is causally associated with cigarette smoking. The risk increases substantially with duration and number of cigarettes smoked and relative risks (RR) up to 60 for smoking >40 cigarettes/day have been reported.3 A few studies also reported that RR for cancer of the larynx increased with decreasing age at start of smoking. The RR decreased with increasing time since quitting smoking.4 For passive smoking, the available studies do not allow a clear assessment.4

There is also clear evidence that extensive alcohol consumption is a causal risk factor for this cancer, in particular for supraglottic tumors.5 Use of alcohol in combination with tobacco smoking greatly increases the risk for laryngeal cancer.6

There are occupational risk factors for which there is some evidence for a link to laryngeal cancer, although the picture is not as consistent as for smoking and alcohol. Among these substances are asbestos,7, 8 mineral coal products, mineral oil,9, 10 fossil fuels,11 ionizing radiation, mustard gas, chromium-VI-compounds, wood dust,12, 13 nickel compounds, sulfuric acid, isopropyl alcohol and bis-chloromethyl ether,14 emissions in the paper-,15 textile-,16 leather-,17 and rubber industry18 and polycyclic aromatic hydrocarbons.19

Particular occupational groups showed, after adjustment for smoking and alcohol, an increased risk.20 Among these occupational groups are some with likely exposure to polycyclic aromatic hydrocarbons (PAH) through asphalt such as road and building construction workers. In a review on PAH exposure and cancer risk, Boffetta et al.19 considered the results inconclusive with respect to laryngeal cancer. In a recent analysis of a large multicenter case–control study in Italy, Switzerland and Spain on cancer of the larynx and hypopharynx combined with cases occurring in the early 1980s, Boffetta et al.21 identified a number of occupational groups with significantly increased risk, including construction workers, potters, butchers and barbers. The evidence for the carcinogenicity of PAH in experimental animal inhalation studies has been reviewed by Boström et al.22 Increased lung cancer rates were found with benzo(a)pyrene exposure in female Wistar rats.

Several researchers have investigated the relation between genetic predisposition and laryngeal cancer. Studies on genetic polymorphisms in alcohol dehydrogenases (ADH) as risk modifiers for head and neck cancer were inconclusive,23 but neither ADH1B or ADH1C modified the alcohol-associated risk of laryngeal cancer in a Caucasian population.23 Glutathione-S-transferases (GST) M1 and T1 are involved in the detoxification of tobacco carcinogens. Previous studies concerning the relevance of the highly prevalent homozygous deletion of the genes (GSTM1 null or GSTT1 null, respectively) to risk of head and neck cancer have yielded inconsistent results.23, 24 Gene–environment interactions have been investigated with smoking as the primary environmental variable and do not show a clear picture.23, 24 In a case–control study on lung cancer, Risch et al.25 reports a protective effect for the GSTT1 null genotype. An interaction with smoking exposure, however, was not found.

A population-based case–control study involving 257 incident cases of laryngeal cancer and 769 population controls was conducted. The main goal of our study was to identify occupational risk factors, in particular PAH exposures as hypothesized in earlier smaller studies26 and to investigate a possible role of glutathione-S-transferases as risk modifier for this exposure.

MATERIAL AND METHODS

Study design and study population

This population-based case–control study was carried out in the Rhein-Neckar-Odenwald region of southwest Germany comprising the cities of Heidelberg, Mannheim, Ludwigshafen, Darmstadt and Heilbronn, with a population of about 2.7 million. Cases and controls were restricted to Germans up to 80 years of age who were registered as citizens in the study region. Ascertainment of all incident cases of histologically confirmed cancer of the larynx began on May 1, 1998. Treatment of laryngeal cancer is only done in the clinics of the cities listed above from which cases were obtained. Local practitioners were additionally contacted to check for possible cases sent to other more distant clinics and to verify complete case ascertainment. Incidence rates in districts within the study were calculated and compared internally and with German incidence rates. Given the population size and study period, the expected number of cases was 244 males and 38 females.

Case ascertainment continued until December 31, 2000. The final sample size included 257 cases (236 males, 21 females). Eleven cases refused participation, and 20 patients could not be interviewed because of speaking impairment or poor health status (response rate = 89.2%). Given the population size, and assuming a similar incidence as from a cancer registry in a neighboring federal state, this is very close to the expected number of cases and supports the study being population-based. Clearance was received by the Ethical Committee of the University and informed consent was obtained from the participants through collaborating physicians. Population controls were selected randomly from the population registries of the study areas. Frequency match for age and gender was 1:3. Controls were contacted through a letter in which the study purpose was explained and dates for a possible interview were suggested; about 1 week later they were contacted by telephone for an appointment for interview. Information on smoking, alcohol consumption, occupational exposure, family history of cancer and nutrition was collected with a comprehensive, standardized questionnaire that has been used in almost identical form in previous large studies.27, 28 Eligible controls (1,233) were contacted and 769 complete interviews were carried out (response rate = 62.4%). All interviews were conducted by 5 well-trained interviewers under standardized conditions. Cases and controls were interviewed at similar proportions.

Each case and control was asked to provide a blood sample for molecular genetic analysis. Overall, 86.4% provided a sample (253 of 257 cases and 664 of 769 controls). Forty milliliters were collected in 3 aliquots. EDTA-blood samples were centrifuged and buffy coat fractions were stored at −80°C until DNA extraction (QIAamp DNA Blood Midi Kit, Qiagen, Hilden, Germany). DNA samples were stored at 4°C until genotyping. Genotyping was carried out using genomic DNA isolated from peripheral lymphocytes and employing PCR/RFLP-based methods.23, 29 The multiplex GST genotyping method included a control β-globin amplicon (larger than the GSTT1 and GSTM1 bands) in each amplification, and all analyses included positive and negative controls for each batch. Polymorphisms of the genes GSTM1, GSTT1, ADH1B (called ADH2 previously) and ADH1C (called ADH3 previously) were determined. These polymorphisms were investigated in all cases for which a blood sample was provided (245 cases) and for a similar number of controls (251 controls). These were selected from all 658 controls who provided a blood sample.

Smoking data were assessed by lifetime smoking periods for which daily, weekly and monthly tobacco consumption of cigarettes was asked (rare uses of cigars, cigarillos and pipes were added according to their average weight relative to that of cigarettes). Daily alcohol consumption was calculated from the alcohol data obtained by interview (daily, weekly and monthly alcohol consumption 10 years before interview for all common alcoholic beverages), assuming the following ethanol content: beer, 5 g/l; wine, fruit wine or sparkling wine, 10 g/l; aperitif and liquors, 20 g/l; and spirits, 40 g/l.

Occupational exposure assessment

Two methods for the assessment of occupational exposure were employed. The first was based on 3 different sources of the questionnaire: a detailed occupational history of all jobs held for at least 6 months; an exposure checklist for known and suspected carcinogens of the respiratory tract; and 34 job-specific supplementary questionnaires (JSQ) addressing specific exposures in job- (or branch of industry) oriented questions. Estimates for lifetime exposure hours by substance were calculated based on JSQ. Published occupational hygiene data were used to infer semi-quantitative scores of exposure intensity for specific job tasks. The quantification procedure was a modification of the method described elsewhere.30 The performance of this method with respect to asbestos has been evaluated.31 Industries and job titles were coded according to the standard classifications provided by the German Statistical Bureau (“Statistisches Bundesamt”)32 and the International Labor Organization (ILO).33 The analysis of job history was based on these codes, which were grouped into categories as given in Table II, respectively, on an ever–never-basis and by duration as described elsewhere.27

Assessing exposure to PAH was based on a detailed assessment of work conditions for selected jobs known to be associated with PAH exposure in the branches: roofer and installer of house siding, insulation installer, workers in road construction and civil engineering and building construction (concrete worker, mason, plasterer). In these industries, exposure to PAH is likely through use of tar. The agriculture/forest industry is another relevant PAH-exposure occupation because PAH-containing carbolineum in pesticides/herbicides has been used frequently in the past. The questionnaires contained specific questions on use of such materials. Published occupational hygiene data were used to infer semi-quantitative scores of exposure intensity for specific job tasks. These data refer to the calendar period after which tar was replaced by bitumen that contains considerably less PAH.34

The second method was based on direct questions on use of specific substances. For PAH exposure, the following substances were used as indicator for exposure: coal tar and pitch, carbolineum, coal or graphite electrodes and carbolineum in pesticides/herbicides. This method is simpler, but it is considered to be less specific, because incorrect reporting of exposures are more likely to happen. Both approaches were used and compared in our analysis.

Statistical analysis

All odds ratios (OR) given are based on a conditional logistic regression model conditioned on a gender × age classification (5- year age groups).35 Adjustment for smoking or alcohol consumption was carried out as indicated in Tables I–IV. To assess the magnitude of confounding by smoking and alcohol, OR estimates with and without adjustment are presented, however, the interpretation of the occupational risks is based on the adjusted values.

Table I. Distribution of Age, Smoking, Alcohol Consumption and Educational Status by Gender and Cases-Control Status in a Population-Based Case-Control Study (Germany, 1998–2000)
FactorCasesControls
MaleFemaleMaleFemale
n(%)n(%)n(%)n(%)
Age (years)
 36–4020.829.530.434.5
 41–4562.5  192.757.5
 46–50156.4  486.8  
 51–552510.6419.07410.51725.4
 56–605422.929.515421.969.0
 61–653916.5628.612317.51319.4
 66–704117.4419.010414.81217.9
 71–754016.914.812117.246.0
 76–80145.929.5568.0710.4
Smoking (pack-years)
 052.1419.016723.83653.7
 >0–10104.2314.316924.11420.9
 >10–20198.129.511015.7811.9
 >20–408134.3628.613919.8710.4
 40+12151.3628.611716.723.0
Time since quitting smoking (years)
 0–115866.91885.733748.04668.7
 2+7833.1314.336552.02131.3
Alcohol consumption (g ethanol/day)
 ≤255724.21257.130343.25176.1
 >25–505121.6419.016924.1913.4
 >50–753916.5314.311316.169.0
 75+8937.729.511716.711.5
Educational level (years of school education)
 9 or less20687.31885.743662.14364.2
 10166.829.510515.01522.4
 More than 10145.914.816122.9913.4
Total2361002110070210067100
Table II. Distribution of Job Groups (Ever Held Job and Longest Held Job) and or in a Population-Based Case-Control Study (Germany, 1998–2000)
 CasesControlsOR1Adj OR295% CI
  • 1

    OR stratified by age and gender.

  • 2

    OR stratified by age and gender, adjusted for smoking and alcohol consumption.

Job group (ever held job)
 Road construction worker2297.66.42.4–17.3
 Painter, varnisher23164.93.31.6–6.8
 Unskilled worker21174.32.71.3–5.8
 Paper mill worker, printer18232.52.41.2–5.1
 Chemical worker26372.21.81.02–3.3
 Carpenter, brick mason57882.21.91.2–2.8
 Cook, food, beverage and tobacco processer18331.71.70.84–3.3
 Metal production and processing worker40671.91.60.97–2.6
 Farmer, orchad worker, wine grower41791.71.60.97–2.5
 Joiner, wood worker16431.21.60.75–2.9
 Miner12182.11.40.60–3.1
 Textile or leather worker11320.991.20.54–2.8
 Service workers (e.g. hairdresser,  housekeeper, waiter)24431.81.20.62–2.2
 Storage and transportation worker591371.41.10.77–1.7
 Electrician24780.901.10.63–1.9
 Plumber, sheet metal worker, machine-tool  operator, welder611761.040.980.67–1.4
 Protective service worker14341.20.970.49–1.9
 Technician, engineer311630.510.750.48–1.2
 Clerical worker482140.620.670.45–0.99
 Stationary engine operator12400.880.670.32–1.4
 Merchant (goods or insurance)431590.770.650.43–0.98
 CasesControlsOR1Adj OR295% CI
Job group (longest held job)
 Road construction workers955.15.51.3–22.4
 Paper mill worker, printer9102.72.80.90–8.7
 Chemical worker12113.42.61.06–6.5
 Metal production and processing worker14133.42.30.95–5.7
 Electrician10251.21.60.69–3.9
 Painter, varnisher8112.41.50.56–4.7
 Service workers (e.g. hairdresser,  housekeeper, waiter)10142.21.40.47–4.3
 Farmer, orchad worker, wine grower9290.891.40.56–3.4
 Carpenter, brick mason17242.11.30.64–2.8
 Storage and transportation worker29601.51.30.78–2.2
 Plumber, sheet metal worker, machine-tool  operator, welder31741.30.990.60–1.6
 Technician, engineer241090.630.980.58–1.6
 Clerical worker201390.400.540.32–0.93
 Merchant (goods or insurance)12710.470.320.16–0.64
 214595   
Table III. PAH Exposure and Laryngeal Cancer Risk in a Population-Based Case-Control Study (Germany, 1998–2000)
ExposureCasesControlsOR1Adj OR295% CI
n%n%
  • 1

    Odds ratio, stratified by age and gender.

  • 2

    Odds ratio, stratified by age and gender, adjusted for smoking [log(packyears+1)] and alcohol [average daily consumption]. 95% CI for Adj OR.

Based on JSQs
 No23892.675498.01.01.0 
 Yes197.4152.04.12.31.05–5.2
Based on substance list
 No23290.374096.21.01.0 
 Yes259.7293.82.81.60.85–3.1
Exposure duration, JSQs (hours of exposure)
 023892.675498.01.01.0 
 >0–1,30041.681.01.61.060.28–4.0
 1,300155.870.97.03.81.3–11.1
Table IV. PAH Exposure and Laryngeal Cancer Risk; Stratified By GSTM1 Genotype (Males) in a Population-Cased Case-Control Study (Germany, 1998–2000)
GenotypePAH exposureCasesControlsOR1Adj OR295% CI
n%n%
  • 1

    Odds ratio, stratified by age and gender.

  • 2

    Odds ratio, stratified by age and gender, adjusted for smoking [log(packyears+1)] and alcohol [average daily consumption]. 95% CI for Adj OR.

GSTMI null
 Based on JSQsNo10793.012697.71.01.0 
Yes87.032.33.61.70.42–6.8
 Duration (hours of exposure)010793.012697.71.01.0 
>0–130021.821.61.30.930.12–7.0
>130065.210.78.53.050.34–27.3
 Based on substance listNo10793.012193.81.01.0 
Yes87.086.21.20.80.26–2.45
Total 115100.0129100.0   
GSTMI non-null
 Based on JSQsNo10090.110198.11.01.0 
Yes119.921.97.065.81.04–32.4
 Duration (hours of exposure)010090.110198.11.01.0 
>0–130021.810.952.62.20.15–32.1
>130098.110.9511.79.90.98–99.1
 Based on substance listNo9585.610097.11.01.0 
Yes1614.432.96.44.00.93–17.1
Total 111100.0103100.0   

Smoking was considered as the cumulative number of cigarettes smoked (pack-year (py); 1 pack-year ≅ 20 cigarettes/day for 1 year ≅ 7,300 cigarettes). It was included as a log-transformed continuous variable (log(py + 1)). Residual confounding through smoking was minimized by comparing the results using other transformations of the smoking dose, including a categorization into 5 categories. The transformation used gave the best fit and also in most cases the maximally reduced estimates for the occupational variables of interest. Time since smoking cessation was included as binary variable “having stopped smoking at least 2 years before diagnosis/before interview”. Average daily alcohol consumption was included as continuous variable. Occupational groups were analyzed as binary variables and PAH exposure additionally in categorical form. School education was considered as an additional variable in 3 levels according to the German educational system (≤9 years, “Hauptschule”; 10 years, “mittlere Reife”' and >10 years, “(Fach-)Hochschulreife”).

Because the presence of PAH in tobacco smoke has been shown previously4 and glutathione-S-transferases have been shown to be relevant to their detoxification, gene deletion polymorphisms in GSTT1 and GSTM1 were analyzed with respect to an effect modification of PAH exposure. These analyses were restricted to the subgroup with genotype information.

The statistical software package SAS (version 8.2) was used for all analyses. PROC PHREG was used for conditional logistic regression modeling.

RESULTS

Table I summarizes the distribution of age, smoking, alcohol consumption and education by gender for cases and controls that differed considerably between groups. The mean age was 62.5 years for cases and 62.7 for controls. Five male (2.1%) and 4 female (19.0%) cases were never-smokers, compared to 167 (23.8%) and 36 (53.7%) of the controls. Among the ever-smokers, fewer cases had stopped smoking. For alcohol consumption, a different distribution for cases and controls in males and females was found. Our study confirmed previous knowledge on these factors for laryngeal cancer risk with OR of 32.7 (95% CI = 15.1–71.0) for smoking more than 40 pack-years compared to never-smokers, with a significantly reduced risk for ex-smokers (OR = 0.37, 95% CI = 0.25–0.54) for ex-smoker compared to current smoker and for an increased risk for high alcohol consumption (OR = 2.3, 95% CI = 1.5–3.6, >75 g ethanol/day) as obtained from a joint model with alcohol and smoking, stratified by age and gender. Educational level was also associated with laryngeal cancer risk. More detailed results on the joint effect of smoking and alcohol consumption are given in Ramroth et al.36

The distribution of the polymorphism for glutathione-S-transferases shows small differences between cases and controls, with a slight overall increase for the GSTT1-null risk allele (OR = 1.3, 95% CI = 0.7–2.4). An insignificantly increased risk of the GSTM1-null risk allele is seen in females (OR = 1.4, 95% CI = 0.25–8.1) but not in males (OR = 0.91, 95% CI = 0.59–1.4) when analyzing both genders separately. Results are given in Risch et al.23

The pattern of employment history is given in Table II. It gives the job titles that cases and controls had for at least 6 months during their life (“ever held job”). The sum exceeds the total number because an individual may have worked in several areas. Categories with 10 cases or more are presented. The lower part of Table II shows the longest held job. Odds ratios are given for each occupational group in comparison with the rest. OR1 gives the OR stratified by age and gender, OR2, to which the CI refer, is adjusted for smoking and alcohol. There are a number of occupational groups associated with an increased risk. Working in the road construction was associated with the highest risk. Several other occupational groups were also associated with a significantly increased risk, such as painters, unskilled workers, paper mill workers, workers in the chemical industry, carpenters and brick layers. Results were similar for the longest held job and ever held job, however, the latter yielded smaller CI due to the higher prevalence. In general, a positive confounding with tobacco and alcohol consumption was demonstrated because the estimates were, with few exceptions, reduced toward one after adjustment. White collar workers had significantly decreased risks. These results point to specific substances that are associated with particular groups.

The result of the assessment of exposure to polycyclic aromatic hydrocarbons is displayed in Table III and Figure 1. The κ coefficient for the dichotomous variable “exposed yes/no” is 0.4. There is a wide range of exposure duration from anecdotal exposures up to 30 years permanent exposure. Based on the JSQ the median exposure time among the exposed is 1,120 hr in controls and 4,620 hr in cases. Thirteen cases (5.5% of male cases) and 6 (0.9% of male controls) were classified as exposed based on Methods One and Two, respectively. One female was classified as low exposed (painting of fences with carbolineum in the 1950s). Both methods of exposure yield an estimated increased risk for the exposed group with a significantly increased OR of 2.3 (95% CI = 1.1–5.2) based on the JSQ. There is a confounding effect of smoking and alcohol that is stronger for the PAH specific OR than for the job-specific results. There is no significant difference in age of first exposure in cases and controls.

Figure 1.

Estimated duration of exposure to PAH in males by assessment method (in hr) in a population-cased case–control study (Germany, 1998–2000).

Figure 1 graphically shows the concordance between both methods and the OR associated with different groups. Individuals that were classified as exposed based on both methods showed a very high OR of 5.2 compared to those with no exposure on either method (p = 0.007, 95% CI = 1.6–17.1). For those who were classified by either of both methods only, the risks were not increased. Fifty individuals (18 cases, 32 controls) were classified as exposed by one method only.

It is interesting to analyze whether risks observed for occupational groups are those for which a PAH exposure was more likely, therefore the distribution of occupations and branches among those classified as PAH exposed by calculating the duration of respective employment periods was also checked. The highest%age was found for the occupational branch “building and ground construction” with 61.1% followed by “machinery and vehicle production” (10.0%) and “farming and forestry” (9.6%). The corresponding job titles are carpenter and brick mason (28.3%), road and ground construction worker (14.5%) and farmer (9.5%). These are among the groups that also showed an overall increased risk.

On stratification by GST genotypes an increased risk with PAH exposure and an increasing risk with exposure duration is clearly visible. For GSTT1, however, a detailed analysis was not possible because of the low frequency of the GSTT1 null genotype carriers. For GSTM1 we also observe an increase in risk associated with PAH exposure. It is considerably more pronounced for the GSTM1 non-null genotype carriers (OR = 5.8; 95% CI = 1.04–32.4) compared to 1.7 (95% CI = 0.4–6.9) for the GSTM1 null genotype carriers. This difference seems large. Due to low numbers, CI are wide and a test for gene–environment interaction is not significant. A causal interpretation of this finding does not seem to be warranted.

DISCUSSION

Case–control studies have contributed significantly in the past to the identification and quantification of occupational exposures as cancer risk factors.37 In particular, for less common cancers this study type is often the only possible design because occupational cohort studies are usually too small in most cases to identify risks for rare cancers.

There are well-known limitations of case–control studies that may lead to various sources of bias or limited power. A careful planning and conduct of the study can only reduce these biases. One of the major difficulties is an appropriate exposure assessment. Because biomarkers for historical exposures are usually not available, dioxin being one of the exceptions due to the long half life in the body, exposure assessment remains commonly based on interview. Other external sources of exposure information are difficult to obtain in case–control studies because the study subjects usually have a wide range of occupational histories. In our study the same occupational exposure assessment methods were used previously in large German lung cancer studies. The more detailed method, based on JSQ, has shown to be more specific because the interviewee gives a detailed job description and the exposure assessment is based on a priori knowledge of exposures associated with common work conditions at the calendar period of the respective job. The second, more rapid method, has a further disadvantage for PAH because a direct question “Have you been exposed to polycyclic aromatic hydrocarbons?” is not appropriate for obvious reasons. The substances used here as surrogate for PAH, “coal tar and pitch products, carbolineum, coal and graphite electrodes”, may also be too technical for some interviewees, and they are neither fully sensitive nor specific for PAH. Nevertheless, the result based on the simple method also yields increased risks, and because there are no obvious reasons for differential misclassification between cases and controls, the observed OR is rather biased toward the null.

Among other sources of bias in case–control studies are low response rates in controls that may lead to a non-representative sample of controls, and confounding/residual confounding due to missing or insufficient control for other risk factors. Our response rate was 62.4%, which in our view is very satisfactory in comparison to other studies. It does not, however, rule out a possible bias. The response rates are lower for the youngest and for the oldest age groups, and are independent of gender.

When combining both methods of exposure assessment, there is a group of 13 cases and 6 controls that have been classified as exposed by both methods. The estimated exposure duration in these individuals were also higher and the OR is very high with 5.2 (p = 0.006, 95% CI = 1.6–17.1). This group can be considered exposed with high certainty. As discussed above, with either exposure assessment method, a classification error cannot completely ruled out. In general, this leads to an underestimation of the OR if this does not depend on the case/control status. Brenner and Blettner38 have argued that the OR could be overestimated with a “dual measurement strategy.” They conclude that restriction of the analysis to individuals with concordant exposure measurements produce higher OR than all other analytic strategies with dual measurements they have investigated, and they are also closer to the true OR than any other strategy. They consider 2 measurements with the same instrument and the case when both are above a certain cutpoint. This is different to the situation here. It is difficult to assess whether an overestimation has occurred in our case. We have carried out a number of plausibility checks and obtained similar effects on PAH-associated risk that supports our result. The result also speaks in favor of combining different methods of exposure assessment as a means to increase specificity. It is known that for low exposure prevalence the specificity is more relevant than the sensitivity as to the bias of the OR estimate.

Our findings are supported by risks associated with particular occupational groups in which the substance exposures in question are a priori considered likely. In particular, the high OR in road construction workers support the findings on PAH.

Cohort studies are not particularly useful for diseases with low incidence. Although laryngeal cancer cannot be considered as rare, the number of cases in published cohort mortality studies are usually too low to allow firm conclusions because survival is relatively good. A cohort study among shipyard workers in Italy39 showed a significantly increased risk for laryngeal cancer, however, these workers have been exposed to a variety of substances in addition to PAH such as asbestos fibers, welding fumes and gases, silica dust and solvents. In our study, asbestos exposure was also associated with an increased risk, however not significantly, and the PAH OR estimates did not change with additional adjustment for asbestos exposure.

PAH exposures have been more prevalent before 1980. Changing workplace conditions and improving industrial hygiene or regulatory limitations of tar use have yielded a reduction in exposures. The observed prevalence in our study cannot be used to estimate an attributable risk or to estimate future numbers of cases caused by PAH exposure. Because the majority of cases occur at ages over 60 years, however, most of them started their working career before 1960 and thus had the risk of exposure if they had worked in the respective industries.

Because smoking and excessive alcohol consumption both are strong risk factors for laryngeal cancer we did a careful investigation on residual confounding. Odds ratio estimates with and without adjustment for smoking and alcohol are presented. The confounding effect, in particular of smoking, was clearly visible. Most OR estimates reduced toward one after adjustment. We also checked for different methods of adjustment, using smoking as categorical variable, and by using both duration of smoking, average number of cigarettes smoked, and age at start of smoking. The adjustment for smoking and alcohol was done by identifying the dose-response curve that gave the best fit using the method of fractional polynomials.40 The adjustment method used in the results gave the best fit to the data, and the corresponding estimates for job groups and PAH exposures were in general smallest when using this adjustment. Differences, however, were negligible for different methods of adjustment. This has also been shown previously by simulation and by real data examples.41 We conclude that the observed association cannot be explained by confounding effect of alcohol and tobacco.

We have also included other potential confounding factors in the regression models, such as educational status, nutrition and family history of cancer (although nutrition and family history of cancer did not yield a further reduction of the PAH OR estimates). Educational status was assessed as school education in 3 levels. After adjustment for smoking and alcohol, high educational level was associated with a significantly reduced risk. The risks associated with PAH exposure changed little after additional adjustment for education. The OR for exposure using the JSQ assessment reduced slightly from 2.3 to 2.1. All individuals with high exposure to PAH belonged to the low educational group.

Our findings regarding the gene–environment interaction do not justify a causal interpretation. The finding is not supported by previous analyses within the same study on smoking and GSTM1 where one would expect a similar finding. No differential effect of smoking on lung cancer risk by GSTM1 genotype was found in these previous analyses.23 In a lung cancer case–control study in China, Lan et al.42 found an increased risk of the GSTM1 null genotype in a region where indoor smoky coal emissions contain high levels of PAH. In our study, however, the higher risk of laryngeal carcinoma was associated with the GSTM1 non-null genotype. Previous similar findings have lead to the hypothesis that GST-genotypes may be important in modulating not only the detoxification of PAH-carcinogens but also the elimination of potentially chemopreventive substances contained in the diet.25 Due to the small number of individuals this hypothesis can not be explored further in our study.

In summary, our study provides evidence that PAH exposure is a relevant risk factor for laryngeal cancer. It is not possible to relate the risk to particular genetically predisposed groups. Several occupational groups showed considerably increased risks.

Acknowledgements

We acknowledge the contribution of Dr. M.-L. Groß who organized the interviews of patients and controls and of the team of interviewers and data base managers. The study would not have been possible without the cooperation of clinicians from the following clinics: Universität HNO-Klinik Heidelberg (Prof. Weidauer), Universität HNO-Klinik Mannheim (Prof. Hörmann), HNO-Klinik der Städt. Kliniken Darmstadt, (Prof. Reck), HNO-Klinik der Städt. Kliniken Ludwigshafen (Prof. Münker) and HNO-Klinik der Städt. Krankenanstalten Heilbronn (Prof. Naumann).

Ancillary