Epidemiologic evidence for assessing the carcinogenicity of acrylamide

Epidemiologic evidence for assessing the carcinogenicity of acrylamide

Regulatory Toxicology and Pharmacology Regulatory Toxicology and Pharmacology 39 (2004) 150–157 www.elsevier.com/locate/yrtph Epidemiologic evidence ...

152KB Sizes 3 Downloads 75 Views

Regulatory Toxicology and Pharmacology Regulatory Toxicology and Pharmacology 39 (2004) 150–157 www.elsevier.com/locate/yrtph

Epidemiologic evidence for assessing the carcinogenicity of acrylamide Linda S. Erdreicha,* and Marvin A. Friedmanb b

a Exponent, 420 Lexington Av. Suite 408, New York, NY 10170, USA University of Medicine and Dentistry of New Jersey (UMDNJ), Newark, NJ 07103-2757, USA

Received 12 February 2003

Abstract Acrylamide (ACM) has recently been found in fried and baked foods, suggesting widespread public exposure. ACM is an industrial chemical that causes neurotoxicity in humans and an increase in benign tumors of the endocrine system of laboratory rats. The U.S. EPA and the International Agency for Research on Cancer (IARC) have designated ACM as a probable human carcinogen based on the bioassay data and evidence for a DNA reactive mechanism. We report here an assessment of the published epidemiological data with regard to exposure to ACM. The results of an epidemiology mortality study of heavily exposed workers published in 1999 failed to reveal any increase in total cancer in this workforce. The average total exposure in the exposed group was equivalent to over 100% of the estimated average lifetime dietary intake, assuming a U.S. diet. However, this epidemiologic information had limited power to detect modest increases in specific tumors of the type reported in the rodent studies. Although the mortality study could not have picked up the small increases in cancer or in specific cancer types predicted by EPAÕs linear extrapolation model, research on biochemical and physiological mechanisms suggests that EPAÕs assessment overstates the potency, and therefore, the risk from foods and other sources of exposure may be lower than previously anticipated. Ó 2003 Elsevier Inc. All rights reserved. Keywords: Acrylamide; Epidemiology; Risk assessment; Diet; Cancer; Thyroid cancer; Pancreatic cancer

1. Introduction Acrylamide (ACM) is a high volume industrial chemical used mainly in the manufacture of water-soluble polymers. These polymers are primarily employed by the wastewater, paper, mining, and oil industry (IARC, 1994). Currently, human exposure from these applications is very limited (EU, 2002). Acrylamide had been designated a probable human carcinogen by several scientific organizations, based on data available at the time of these assessments (IARC, 1994; U.S. EPA, 1988). Scientists in several countries have recently reported finding high concentrations of ACM in carbohydraterich foods heated to high temperatures by frying or baking (FSA, 2002; NFA, 2002a,b; Tareke et al., 2002). A WHO/FAO consultation in 2002 concluded that as a * Corresponding author. Fax: 1-212-972-9480. E-mail address: [email protected] (L.S. Erdreich).

0273-2300/$ - see front matter Ó 2003 Elsevier Inc. All rights reserved. doi:10.1016/j.yrtph.2003.12.004

result of the high levels in these foods, ACM represented a ‘‘major problem.’’ Preliminary chemical studies have shown that that ACM results from reduction of aspargine in these products (Mottram et al., 2002; Stadler et al., 2002). Because ACM exposure through food is widespread and substantial, the strength of the evidence for ACM carcinogenicity is critical. In drinking water studies, ACM increased tumor incidence in Fisher 344 rats, and in screening studies of mice (Bull et al., 1984a,b; Friedman et al., 1995; Johnson et al., 1986). In 1988, EPA calculated a unit risk using the linearized multistage model for low dose extrapolation after pooling benign and malignant tumors at all sites in female rats (U.S. EPA, 1993).1 Female rats had a greater diversity of cancer sites than male rats. This assessment can be viewed as highly conservative due to the inclusion of benign tumors, tumors that were 1 The EPAÕs IRIS summary for acrylamide was updated 1993, but the basis has not changed since the original posting in 1988.

L.S. Erdreich, M.A. Friedman / Regulatory Toxicology and Pharmacology 39 (2004) 150–157

increased but not statistically significant, and tumors that occurred in only one species or site (U.S. EPA, 1999). IARC (1994) designated ACM a probable human carcinogen, but provided no quantitative assessment. Other researchers have calculated dose–response relationships using a multiplicative model, based on the same bioassay data (Paulsson et al., 2001). At the time of the first EPA risk assessment, epidemiology studies of ACM plant workers were limited. Sobel et al. (1986) had studied a cohort of 371 male workers. A much larger cohort of 8854 workers at four manufacturing plants included 2293 male workers whose cumulative exposure to ACM exceeded 0.001 mg/ m3 -yr (Collins et al., 1989). Neither study detected increased overall cancer in workers exposed to ACM, although Sobel reported excess cancers of the digestive tract in the subgroup with previous exposure to organic dyes and Collins detected an increase in lung cancer mortality among muriatic acid workers. These studies were considered inadequate evidence for carcinogenicity by both the IARC and EPA reviews. A 10-year update of the Collins study has been completed (Marsh et al., 1999), and has not yet been considered in any published risk assessment evaluating ACM and cancer. The purpose of this current analysis is to compare the information in the largest epidemiology study of cancer in ACM workers to possible cancer risk predicted by models based on studies in animals. We first provide a review of the strengths, limitations, and results of the latest update of the epidemiology study (Marsh et al., 1999). We then discuss the adequacy of the cohort study to detect risks of practical importance, as well as the impact of this information on the health risk assessment. We also address the suitability of using low-dose extrapolation modeling based on animal data to predict quantitative cancer risks in humans. Finally, we will consider basic aspects of biological plausibility of the cancer sites suggested by the animal and epidemiologic studies.

2. Analysis of the updated epidemiology cohort study 2.1. Description Collins et al. (1989) had assessed cancer mortality in a cohort of 8854 workers at four ACM manufacturing plants hired between 1925 and 1973. Mortality followup through 1983 showed no dose–response trends and no increases in overall cancer mortality or any cancer site that reached statistical significance. Marsh et al. (1999) updated the cancer mortality in workers at three of the four plants. He followed the cohort for mortality through 1994, 11 years longer than the previous report. The update also included a review of company records to assess the work pattern and work histories for

151

workers after the end of the previous study. This is the largest epidemiology study of workers exposed to ACM. Exposure was estimated as an average daily exposure for all jobs over time, based on air concentrations (mg/ m3 ) that had been monitored at all plants from 1977. The 1977 data was used for earlier time periods, and interviews with experienced workers obtained information regarding past jobs and processes. Exposure varied among the plants, in part because the plants used different processes to produce or use ACM. Therefore, specific jobs in each of the plants were assigned to job categories to classify exposures. However, these estimates likely underestimate actual exposure in view of the industrial hygiene monitoring that began in 1977. Standardized Mortality Ratios (SMRs) were calculated by follow-up period using national comparisons. Local (country) comparisons were used for the 1950– 1994 follow-up by plant and exposure status. Workers whose cumulative exposure was estimated to be at least 0.001 mg/m3 were identified as exposed (n ¼ 2004). An exposure–response analysis was conducted for the four cancer sites that had shown non-statistically significant excess of 20% or more in exposed workers, and deficits in unexposed workers. These cancers were reported in esophageal, rectal, kidney, and pancreatic tissues. Conditional logistic regression analysis was used to compare cancer risk within the cohort. The design, exposure assessment, and analytic methods used in this study are standard, state-of-the-art methods, and include additional approaches to adjust for biases that may be present in epidemiology studies. Internal comparisons help to reduce potential biases, such as the healthy worker effect, which may occur in comparisons with external populations. 2.2. Results of the Marsh study In the exposed workers, no tissue-specific cancer risk was elevated at a level that reached statistical significance, and cancer overall was not increased. The SMR was not elevated and statistically significant for any of the cancer sites identified a priori from animal studies; central nervous system, testes, and thyroid (Table 1). While the SMR for thyroid mortality was elevated, this was based on only three cancers and did not reach

Table 1 Cancer mortality in 2004 workers exposed to ACM (P0.001 mg/m3 -yr) Cancer site

SMR (confidence interval)

Total malignant tumors Pancreatic cancer Thyroid cancer Central nervous system Bronchus, trachea, and lung Testis and other male genital

0.98 (0.83–1.14) 1.79 (0.98–3.01) 4.27 (0.52–15.42) 0.74 (0.15–2.15) 1.09 (0.84–1.40) — (0.00–7.09)

152

L.S. Erdreich, M.A. Friedman / Regulatory Toxicology and Pharmacology 39 (2004) 150–157

statistical significance. The study population was all male, and thus provided no information on mammary cancer. The SMRs for four cancer sites (esophagus, rectal, kidney, and pancreas) were examined in a more detailed exposure–response analysis, based on non-statistically significant increases in the exposed group. Using three exposure levels, the highest at P0.3 mg/m3 -yr to reflect current and proposed regulated exposure levels at the beginning of the study, and local county comparisons, the only statistically significant increased SMRs were for pancreatic cancer (Table 2). SMRs for the four levels of cumulative exposure did not show a dose–response pattern, but reached statistical significance at the highest cumulative exposure level, P0.3 mg/m3 -yr, based on nine cases. SMRs for the four levels of mean exposure also showed an increase for the highest level (5 cases), which had a broad confidence interval that did not include 1.0. No statistically significant increased SMR was found for any category of time since first employment, duration of exposure, or duration of employment. There was a positive trend for time since first exposure and duration of exposure in the internal analysis of pancreatic cancer. 2.3. Strengths and limitations As observational studies, epidemiologic studies have both strengths and limitations. The strength of epidemiology is that it focuses on people, the species of interest, and real world exposures, and thus avoids the pitfalls of species and dose extrapolations. The main factors limiting the existing epidemiologic information on ACM are exposure uncertainties, insufficient control for potential confounding factors, and lack of sufficient statistical power to detect increases in less common cancer sites, which includes some sites reported to occur in laboratory animals. Industrial hygiene monitoring was introduced into this ACM workplace in 1977, and these data were considered representative of exposures back to start up time. However, because exposure in the 1970s may already be low, exposure prior to this time may be underestimated. In this study the paucity of data on smoking limits the interpretation of pancreatic cancer results. Smoking is a known risk factor for several cancer types, including the

pancreas, and if not controlled can confound the analysis (Anderson et al., 1996; Rothman and Greenland, 1998). Smoking history was known for less than a third of the individuals in the study. In studies of workers, there is always concern about the presumed better health status of employed populations in comparison to the general population that is used as the comparison group. This consideration may be relevant to the SMR comparisons, but this potential bias is avoided in the internal cohort comparisons used for pancreatic cancer dose–response modeling. In addition, the healthy worker effect is of less concern for cancer than for non-cancer causes of death and the effect decreases with length of follow-up, and thus would be expected to be minimal in this study (Fox and Collier, 1976). As noted by the authors, the results of the two comparisons are generally similar. 2.4. Statistical power Table 3 displays the minimum true relative risk likely to be detected with 80% power for all cancer and selected cancer sites in this cohort. Marsh et al. (1999) stated that the follow-up study of the ACM cohort had ample power to detect even a small increase in all cancer, and in lung cancer, in the total population. The 95% confidence intervals (CI) from the study indicate that it is reasonable to rule out increases in overall cancer of 5% or greater for the total cohort and 14% or greater for the exposed cohort. The epidemiology study of ACM workers was not sufficiently powerful to detect the low risks predicted by low-dose extrapolation modeling of animal data. The

Table 3 Estimating power of the Marsh et al. (1999) study to detect increased relative risk; total cohort, population N ¼ 8508, a ¼ 0:05 Minimal detectable relative riska 80% power

Cancer types (expected number of cases)

1.10 1.40 1.15 1.61 4.07

All cancer (870) Pancreas (44) Bronchus, trachea, lung (310) Brain and central nervous system (22) Thyroid (1.4)

a Calculations based on Breslow and Day (1987, Eq. 7.1) for an external cohort.

Table 2 Risk estimates for pancreatic cancer using external control and internal controla Cumulative exposure (mg/m3 -yr)

Obs deaths

SMR (local population comparison)

Risk ratio (relative risk regression) (internal cohort analysis)

<0.001 0.001–<0.003 0.03–<0.30 >0.30

30 3 2 9

0.80 2.77 0.73 2.26

1.00 3.14 (0.92–10.71) 0.77 (0.18–3.26) 2.63 (1.23–5.60)

a

From Marsh et al. (1999).

(0.54–1.14) (0.57–8.09) (0.09–2.64) (1.03–4.29)

L.S. Erdreich, M.A. Friedman / Regulatory Toxicology and Pharmacology 39 (2004) 150–157

expected relative risk of cancer in the exposed cohort was calculated to be 1.015, based on the EPAÕs unit risk estimate and the lifetime average daily dose from the mean worker exposure concentration, reported by Marsh et al. (1999). (0.00034 mg/kg: see Appendix A). This study had low statistical power to detect excess risks for many of the cancers of interest such as thyroid gland, or testis and male genital organs, and this was reflected in the broad 95% confidence intervals for these cancer types. These estimates indicate that if there is a true cancer risk, the magnitude may be below the level that a cohort study such as this one could have detected. However, extrapolation modeling is not the sole criterion for assessing the value of epidemiologic and toxicological studies, as discussed below.

3. Discussion 3.1. Comparison with risk calculated from extrapolation models Neither experimental nor epidemiologic data can rule out small risks that are derived from extrapolations to very low-dose levels. Extrapolation modeling based on animal bioassays has been used as a conservative regulatory tool in the absence of better information, not because these estimations are wholly valid as predictors of human cancer, or because increases in risk of all cancers are actually expected. EPAÕs current draft cancer risk assessment guidelines discourage extrapolating below the range of the level of the observed data (U.S. EPA, 1999). Granath et al. (2001) question whether ‘‘unacceptable risks’’ could have been detected in the epidemiology study. They defined these as risks from all types of cancers, as calculated from the low-dose extrapolation model based on animal data, specifically female rats. They calculated relative risk estimates for all cancer of 1.01–1.03 in the cohort, and 1.05–1.15 in the exposed subgroup. We calculated a risk ratio of 1.01 in the cohort, based on the EPA model (Appendix A). They note that because uptake through the skin occurs in addition to inhalation, it is possible that the true risk increments for overall cancer in this cohort are higher, perhaps 10– 20%, which may well have been detectable in this cohort (see Table 3). Based on size of the overall cohort, this study is estimated to be able to detect an SMR of 1.10 overall and 4.10 for the thyroid, the most prevalent animal tumor. However, this exposure estimate has been averaged over a population that included unexposed workers, which introduces uncertainty. Quantitative risk assessments based on mathematical models and animal experiments have a significant role in predicting potential incidence of disease in human populations at low doses, although they are based on

153

statistical, more than biological, considerations. Because it is the role of quantitative risk assessment to safeguard public health, the parameters of these extrapolation models are based on highly conservative assumptions, and are specifically designed to be Ôworst caseÕ scenarios. The relationship between animal studies at elevated doses and human responses at small doses is not always clear. The generic risk assessment models may not reflect physiological and pharmacological factors or mode of action specific to the chemical. The role of protective mechanisms such as DNA repair or metabolic and chemical detoxification is seldom quantified. These biological factors may play a substantial role in the nature of the site affected and in the dose–response relationship, and therefore, the organs affected and the type of adverse effect resulting from chemical exposure may differ between laboratory animals and humans (U.S. EPA, 1999). Implicit in the use of these mathematical models is that ACM exerts its effects through a DNA reactive mechanism. The mechanistic data do not support a DNA reactive mechanism at this time (JIFSAN/ NCFST, 2002). Although the result of extrapolation modeling is typically described as cancer risk, omitting mention of tissue type or anatomical site, there are no reliable examples of agents that cause an increase in cancers at many sites. Cigarette smoking and ionizing radiations are often given as examples to support the concept of a universal carcinogen, but this is incorrect. Cigarette smoking is not a relevant example because it includes multiple exposures; the smoke includes metals, particulates, gases, and PAHs. The argument regarding ionizing radiation is also of dubious relevance because ionizing radiation is both genotoxic and cytotoxic, and the multiple cancer sites were identified in people who had high exposures. At lower environmental exposures, only associations with leukemia are well documented. 3.2. Relevance of the Marsh cohort study for dietary risk assessment The epidemiology study of ACM workers was not sufficiently powerful to detect the low risks from this workplace exposure predicted by extrapolation modeling, however, other considerations indicate that the results provide relevant information regarding the question of ACM in the diet and cancer. The average cumulative workplace exposure to ACM was 0.25 mg/m3 -yr, which is equal to 912.5 mg assuming an intake of 10 m3 /day and 100% absorption. Daily dietary exposure is estimated to be 0.033 mg/day, which equals 843 mg over a 70-year lifespan (WHO, 2002). Therefore, the average cumulative workplace exposure (presumably by inhalation) in this cohort is equivalent to over 100% of the estimated average lifetime dietary intake, assuming a U.S. diet. It is the largest existing

154

L.S. Erdreich, M.A. Friedman / Regulatory Toxicology and Pharmacology 39 (2004) 150–157

study that provides information relevant to humans about exposures to ACM at levels that exceed actual dietary exposures Both the workers and the members of the reference population were probably exposed to ACM in their diet. Data are not available to support inferences about dietary differences in ACM between these factory workers and the general population. Epidemiology studies of dietary intake of acrylamiderich foods were published after the presence of ACM in food was reported (Bosetti et al., 2002; Mucci et al., 2003; Pelucchi et al., 2003). Bosetti et al. (2002) reported an association and positive trend for laryngeal cancer with dietary intake of fried protein-rich foods. This hospital based case-control study was conducted in Italian and Swiss towns, using cases that were histologically confirmed. However, it included no information on ACM intake. This study was one of a series of case-control studies of diet and incident cases of cancer, which were reevaluated for intake of fried/baked potatoes to provide information regarding ACM (Pelluchi et al., 2003). The reanalysis found no association or positive trend for frequency of potato consumption and several cancer types: oral cavity, esophagus, larynx, colon, rectum, breast or ovary. Odds ratios were between 0.8 and 1.1. Mucci et al. (2003) completed a population-based case-control assessment of cancers of the large bowel, kidney, and bladder in Sweden. No association was found between any of these cancer sites and overall ACM intake in the diet, or more frequent consumption of foods with high or moderate levels of AMC. Dietary intake of ACM was estimated to be 0.027 mg/day in controls. No studies of North American populations have been reported. An efficient approach to increase power is to continue the follow-up of the existing ACM cohort. In the absence of readily accessible data on the age distribution of the cohort, we assessed the minimal detectable relative risk after 10 years of additional follow-up (1995–2004) by assuming an increase in cases proportional to the previous 10 years for the cancer sites listed in Table 3. Gains in minimal detectable relative risk were negligible; from 1.09 to 1.08 for cancer overall, 1.41 to 1.37 for pancreatic cancer, and from 1.60 to 1.55 for brain and central nervous system. However this approach is likely to underestimate expected cases because it does not account for aging. 3.3. Assessment of pancreatic cancer results Overall, positive statistical associations were reported only for pancreatic cancer for workers who had experienced the highest mean exposure or cumulative exposure. This relationship with pancreatic cancer shows no clear evidence of dose–response trend or increased risk with cumulative exposure. The internal analysis indi-

cated a positive trend for increasing duration of exposure, although there were only 4 or 5 cases in each group. In this, as in many epidemiologic studies of cancer, multiple comparisons are made, which could lead to a chance finding. Cigarette smoking is accepted to be a causal risk factor for pancreatic cancer (Anderson et al., 1996; Potter, 2002). Therefore the possible role of cigarette smoking must be considered in assessing the effects of occupational exposure. Data on smoking were available for only 35% of the ACM cohort; of these, 76% were smokers (Collins et al., 1989). All of the 14 exposed cases of pancreatic cancer in the study were reported to have been smokers for a period of three months or more. This information suggests that smoking may have contributed to the risk of pancreatic cancer and detracts from concluding that there is a causal relationship with ACM. On the other hand, the observation that there is no apparent difference in the SMRs for lung cancer, the cancer most strongly related to smoking, between the exposed and unexposed workers, suggests that exposed workers did not smoke more than unexposed workers. Other suspected risk factors for pancreatic cancer, including dietary factors such as high fat intake or low folate intake, and diabetes, were also uncontrolled in this study (Anderson et al., 1996; Potter, 2002; Stolzenberg-Solomon et al., 2002). There is no clear causal interpretation of this positive statistical association, given the absence of complete smoking histories. To date, this association has been seen only in one study, and the study included multiple comparisons that detract from the statistical significance. The studies in laboratory animals have not found increases in cancer of the pancreas (Friedman et al., 1995; Johnson et al., 1986). Several of the tumors reported in at least one study were hormonally related sites including mammary gland, uterus, and tunica vaginalis. Although the pancreas is an endocrine organ, human pancreatic cancer is almost universally of the exocrine pancreas (Potter, 2002.) 3.4. Assessment of biological plausibility Whenever epidemiologic data are unclear, it is important to assess the weight of evidence supporting the observations in the animal bioassays and their relevance to predicting cancer in humans. The toxicology of ACM in the laboratory rat is well understood. The following is a brief summary of issues currently under discussion regarding biological plausibility (JIFSAN/NCFST Workshop on Acrylamide in Food, October 2002). Orally administered ACM increased the incidence of several tumor types in male rats: mesothelioma of the tunica vaginalis (TVM), adrenal pheochromocytomas (one study), and thyroid adenomas. In female rats, incidence was increased for gliomas and astrocytomas

L.S. Erdreich, M.A. Friedman / Regulatory Toxicology and Pharmacology 39 (2004) 150–157

of the central nervous system (one study), fibroadenomas and adenocarcinomas (one study) of the mammary gland and of the thyroid, adenocarcinomas of the uterus (one study), and papillomas and carcinomas of the oral cavity (one study) (Friedman et al., 1995; Johnson et al., 1986). Tests in animals done at the maximum tolerated dose (MTD) are assumed to be relevant to humans in the absence of information to the contrary, such as data on metabolic or mechanistic differences between experimental models and humans. Physiological and toxicological evidence provides support for the idea that the mechanism of induction of benign fibroadenomas and their characteristic occurrence as tumors in aging F344 rats is not predictive of a carcinogenic mechanism in humans. More specifically, these tumors are formed as a result of perturbation of metabolic pathways that are probably unique to the aging Fisher 344 rats used in these studies, and therefore may not be relevant to humans (CTFA, 2003). For example, TVMs are likely to be a rodent-specific response due to a biochemical cascade that does not occur in humans. ACM induces perturbations in the levels of circulating hormones (Ali et al., 1983; Khan et al., 1999; Uphouse et al., 1982). Rats lack circulating proteins that bind hormones and serve as a reservoir for these hormones. Rats are substantially more sensitive to perturbations in hormone physiology than humans (U.S. EPA, 1998). A primary requirement for linear low-dose risk assessment is that the chemical be DNA reactive. The DNA reactivity can be measured by mutagenicity or directly (i.e., chemically). In vitro studies that show evidence of effects on DNA have been considered in rating ACM as a probable carcinogen. The positive observations in these genotoxicity studies were noted in cell systems or in animals exposed to doses significantly greater, in some cases several orders of magnitude greater than the doses evaluated in the two-year bioassays. However, ACM was not a directly acting mutagen in bacterial or mammalian assays at non-cytotoxic concentrations. ACM is clastogenic at doses at or near cytotoxic levels, but the genetic toxicity studies in rats and rat cell lines are negative. The likely cause of the clastogenic effect is protein binding through either protamine binding or krp binding rather than DNA reactivity (Sega et al., 1989 ; Sickles et al., 1995). The research on mode of action suggests that the animal data overstate the risk to humans.

4. Conclusion The absence of detectable increases in cancer in a population of workers whose occupational exposure exceeded typical dietary exposure is a significant factor is assessing carcinogenicity to humans. The epidemio-

155

logic data do not support the hypothesis of ACM as a multisite carcinogen, or the idea of ACM as a carcinogen specific for the target organs suggested by the animal studies. The value of the existing epidemiologic information is limited by insufficient power to detect modest increases in several of the tumors that appeared in the rodent studies. Currently, there is unlikely to be a worker cohort that is large enough, or has sufficiently elevated exposures, or sufficient numbers of women to study adequately the more uncommon cancers such as thyroid, or hormone responsive cancers in women. Existing methods could be used to increase power in future epidemiologic studies, particularly those designed to address the question of the carcinogenicity of ACM in food. Dietary information about food purchases and food intake is available, which offers the ability to compare cancer rates among ethnic or socioeconomic population groups whose diets may differ, and who thus may have different levels of intake of ACM. These correlations would offer a preliminary or screening approach, which could be followed by case-control epidemiology studies that include dietary histories for any cancer sites that appear to be increased in populations that have higher exposures. Such studies are subject to the usual difficulties of dietary studies, but are feasible in terms of the required number of subjects; it would take 200–400 cases to detect a 50–100% (odds ratios of 1.5 or 2.0) increase in risk. Although it is not possible for any single study to prove the absence of risk, studies of reasonable power can provide important support for a lack of an association. Well-designed epidemiology studies can provide evidence regarding the upper limits of risk, based on the studyÕs power of detection.

Acknowledgments This study was funded by SNF SA and Exponent. The authors acknowledge the timely input and helpful comments provided by Dr. G. Marsh.

Appendix A Calculating excess cancer risk for ACM from the occupational exposure in Marsh et al. (1999), using the animal-based EPA oral cancer potency factor. This inhalation risk estimate as reported on EPAÕs Integrated Risk Information System (U.S. EPA, 1993) is based on the oral cancer potency (slope factor) of 4.5 per mg/kg/day. No inhalation bioassays using ACM are available and EPA recommends not using the inhalation unit risk estimate (based on the oral slope factor) when exposure is greater than 8 lg/m3 . The excess risk cancer for ACM is calculated for workersÕ estimated lifetime

156

L.S. Erdreich, M.A. Friedman / Regulatory Toxicology and Pharmacology 39 (2004) 150–157

exposure based on the cancer potency factor, and then the excess cases are used to calculate the relative risk (RR) predicted for that population. 1. Excess risk Risk ¼ Lifetime average daily dose ðLADDÞ  Cancer potency; LADD ¼ ðConcentration ½mg=m3   R  D  AÞ=Y  W ; where C, air concentration of ACM (mean daily exposure); R, respiratory rate (m3 /day); D, exposure duration in days; A, bioavailability, which is assumed to be 100% in absence of definitive data to suggest otherwise; A, days of exposure (average); Y, days in lifetime; W, body wt in kg. LADD ¼ fð0:098 mg=m3  10 m3 =day  620 days fi:e:; 1:7 yearsgÞ=ð25; 550 day  70 kgÞg ¼ 0:00034 mg=kg=day; Risk ¼ 0:00034  4:5 ¼ 0:00153: For a cohort of 8508, this risk predicts 13.01 extra cancer cases. 2. Expressing the excess risk as RR RR (estimated by the SMR) is the ratio of the rate in the exposed to that in the unexposed or reference population, the background rate. This is equivalent to the number of cancers observed in the exposed cohort to the number expected based on rates in the referent population. SMR ¼ Observed in exposed/Expected in exposed, based on reference population (background rate). Marsh et al. (1999) reported an SMR ¼ 0.98, based on 853 cases observed in the cohort population. The expected number of deaths for all cancer then ¼ 870.4 cases. The animal-based RR for the ACM cohort (i.e., the expected RR, if the animal-based potency were applied to predict the excess cases in the exposed) would be: Relative risk ¼ ð870:4 þ 13:01Þ=870:4 ¼ 1:015: References Ali, S.F., Hong, J.S., Wilson, W.E., Uphouse, L.L., Bondy, S.C., 1983. Effect of acrylamide on neurotransmitter metabolism and neuropeptide levels in several brain regions and upon circulating hormone levels. Arch. Toxicol. 52, 35–43. Anderson, K.E., Potter, J.D., Mack, T.M., 1996. Pancreatic cancer. In: Schottenfield, D., Fraumeni, J.F. (Eds.), Cancer Epidemiology, and Prevention, second ed. Oxford Press, New York. Bosetti, C., Talamini, R., Levi, F., Negri, E., Franceschi, S., Airoldi, L., La Vecchia, C., 2002. Fried foods: a risk factor for laryngeal cancer? Br. J. Cancer 87 (11), 1230–1233. Breslow, N.E., Day, N.E. (Eds.), 1987. International Agency for Research on Cancer (IARC). Statistical methods in cancer

research. Volume II The design and analysis of cohort studies. IARC Scientific Publication No. 82. IARC, Lyon, France. Bull, R.J., Robinson, M., Stober, J.A., 1984a. Carcinogenic activity of acrylamide in the skin and lung of Swiss-ICR mice. Cancer Lett. 24, 202–212. Bull, R.J., Robinson, M., Laurie, R.D., Stoner, G.D., Greisiger, E., Meier, J.R., Stober, J.A., 1984b. Carcinogenic effects of acrylamide in sencar and a/j mice. Cancer Res. 44, 107–111. Collins, J.J., Swaen, G.M.H., Marsh, G.M., Utidjian, H.M.D., Caporossi, J.C., Lucas, L.J., 1989. Mortality patterns among workers exposed to acrylamide. J. Occup. Med. 31, 614–617. Cosmetics, Toiletries and Fragrance Association (CTFA) Cosmetic Ingredient Review panel (CIR), 2003. Safety of Polyacrylamide, amended final report. Wasington DC. (available at http://www.cirsafety.org/publications.shtml). European Union (EU), 2002. Opinion of the scientific committee on foodon new findings regarding the presence of acrylamide in food. Scientific Committee on Food, Health and Consumer Protection Directorate-General, European Commission. Available from . Food Standards Agency (FSA), 2002. Study of acrylamide in food: background information & research findings. Press Briefing 17.05.02. Available from . Fox, A.J., Collier, P.F., 1976. Low mortality rates in industrial cohort studies due to selection for work and survival in the industry. Br. J. Prev. Soc. Med. 30, 225–230. Friedman, M., Dulak, L., Stedham, M., 1995. A lifetime oncogenicity study in rats with acrylamide. Fundam. Appl. Toxicol. 27, 95–105. Granath, F., Ehrenberg, L., Paulsson, B., Tornqvist, M., 2001. Cancer risk exposure to occupational acrylamide. Occup. Environ. Med. 58, 608. International Agency for Research on Cancer (IARC), 1994. Acrylamide. In: IARC Monographs on the Evaluation of Carcinogenic Risks to Humans; Some Industrial Chemicals, vol. 60. IARC, Lyon, France. Johnson, K.A., Gorzinski, S.J., Bodner, K.M., Campbell, R.A., Wolf, C.H., Friedman, M.A., Mast, R.W., 1986. Chronic toxicity and oncogenicity study on acrylamide incorporated in the drinking water of Fischer 344 rats. Toxicol. Appl. Pharmacol. 85 (2), 154–168. Joint Institute for Food Safety and Nutrition/National Council for Food Safety and Technology (JIFSAN/NCFST) Workshop on Acrylamide in Food, October 2002. Available from . Khan, M.A., Davis, C.A., Foley, G.L., Friedman, M.A., Hansen, L.G., 1999. Changes in thyroid gland morphology after acute acrylamide exposure. Toxicol. Sci. 47, 151–157. Marsh, G.M., Lucas, L.J., Youk, A.O., Schall, L.C., 1999. Mortality patterns among workers exposed to acrylamide: 1994 follow up. Occup. Environ. Med. 56, 181–190. Mottram, D.S., Wedzicha, B.L., Dodson, A.T., 2002. Acrylamide is formed in the Maillard reaction. Nature 419, 448–449. Mucci, L.A., Dickman, P.W., Steineck, G., Adami, H.-O., Augustsson, K., 2003. Dietary acrylamide and cancer of the large bowel, kidney, and bladder: absence of an association in a populationbased study in Sweden. Br. J. Cancer 88, 84–89. National Food Administration (NFA), 2002a. Summary—acrylamide in heat-processed foods. Available from . National Food Administration (NFA), 2002b. Acrylamide in foodstuffs, consumption and intake. Available from . Paulsson, B., Granath, F., Grawe, J., Ehrenberg, L., Tornqvist, M., 2001. The multiplicative model for cancer risk assessment: applicability to acrylamide. Carcinogenesis 22, 817–819. Pelucchi, C., Francheschi, S., Levi, F., Trichopoulis, D., Bosetti, C., Negri, E., La Vecchia, C., 2003. Fried potatoes and human cancer. Int. J. Cancer 105, 558–560.

L.S. Erdreich, M.A. Friedman / Regulatory Toxicology and Pharmacology 39 (2004) 150–157 Potter, J.D., 2002. Invited commentary: pancreas cancer—we know about smoking, but do we know anything else? Am. J. Epidemiol. 155, 793–795. Rothman, K.J., Greenland, S., 1998. Modern Epidemiology, second ed. Lippincott–Raven, Philadelphia, PA. Sega, G.A., Alcota, R.P., Tancongco, C.P., Brimer, P.A., 1989. Acrylamide binding to the DNA and protamine of spermiogenic stages inb the mouse and its relationship to genetic damage. Mutation Res. 216, 221–230. Sickles, D.W., Welter, D.A., Friedman, M.A., 1995. Acrylamide arrests mitosis and prevents chromosome migration in the absense of changes in spindle microtubules. J. Toxicol. Environ. Health 44, 73–86. Sobel, W., Bond, G.G., Parsons, T.W., Brenner, F.E., 1986. Acrylamide cohort mortality study. Br. J. Ind. Med. 43, 785–788. Stadler, R.H., Blank, I., Varga, N., Robert, F., Hau, J., Guy, P.A., Robert, M.C., Riediker, S., 2002. Acrylamide from Maillard reaction products. Nature 419, 449–450. Stolzenberg-Solomon, R.Z., Pietinen, P., Taylor, P.R., Virtamo, J., Albanes, D., 2002. Prospective study of diet and pancreatic cancer in male smokers. Am. J. Epidemiol. 155, 783–792.

157

Tareke, E., Ryber, P., Karlsson, P., Ericsson, S., Tornqvest, M., 2002. Analysis of acrylamide, a carcinogen formed in heated foodstuffs. J. Agric. Food Chem. 50, 4998–5006. Uphouse, L.L., Nemeroff, C.B., Mason, G., Prange, A.J., Bondy, S.C., 1982. Interactions between handling and acrylamide on endocrine response in rats. Neurotoxicology 3, 121–125. U.S. Environmental Protection Agency (USEPA), 1988. Draft Preliminary Assessment of Health Risks from Exposure to Acrylamide. Office of Toxic Substances. Washington, DC. March. U.S. Environmental Protection Agency (USEPA), 1993. Integrated risk information system (IRIS). Acrylamide (CASRN 79-06-1). Available from . U.S. Environmental Protection Agency (USEPA), 1998. Assessment of thyroid follicular cell tumors. EPA Document No. EPA/630/R-97/ 002, March 1998. Available from . U.S. Environmental Protection Agency (USEPA), 1999. Guidelines for carcinogen risk assessment. Risk Assessment Forum, Review Draft. Washington, DC. NCEA-F-0644. WHO, 2002. FAO/WHO consultation on the Health Implications of Acrylamide in Food, Geneva, 25–27, June 2002.