10
Integration of Biology and Epidemiology

INTRODUCTION

Previous chapters of this report have reviewed major elements of experimental and epidemiologic studies relating to the tumorigenic and heritable effects of low-LET (linear energy transfer) ionizing radiation. The development of views on the risks to health from exposure to ionizing radiation depends increasingly upon the establishment of scientific coherence between judgments that stem from knowledge of the biological mechanisms underlying radiation-induced health effects and the direct epidemiologic quantification of such effects. The epidemiologic modeling of radiation-induced health effects for the purposes of risk estimation relies in many cases on biological concepts developed from experimental studies with cultured cells and laboratory animals. This chapter draws together the most important conclusions reached from the reviews of the data. The principal topics considered here are the intimate relationship between cellular responses to DNA damage and health effects; the possible implications of the knowledge of cancer mechanisms for projections of cancer risk over time and the transport of that risk between populations; the shape of the dose-response relationship for cancer risk at low doses; dose and dose-rate effects for cancer risk; the possible implications for cancer risk or other forms of cellular response to radiation; genetic factors in radiation cancer risk; and the heritable effects of radiation.

DNA DAMAGE RESPONSE AND CANCER RISK

Chapters 2 and 3 review the largely cellular and molecular data that strongly support the proposition that chromosomal DNA, the genetic material of the cell, represents the principal target for the deleterious effects of ionizing radiation. In brief, energy deposition from low-LET electron tracks intersecting DNA or its local environment can lead to radical-mediated disruption of covalent bonds in DNA. The cell responds to the presence of such DNA damage in a biochemically complex fashion, but the outcome of greatest importance is the repair or misrepair of critical DNA lesions. Depending on its location, misrepaired DNA damage can lead to the appearance of gene and chromosomal mutations in both somatic (Chapter 2) and germline cells (Chapter 4).

The establishment of an intimate relationship between DNA damage responses, somatic mutation, and cancer development represents one of the most important advances in cancer research during the last decade (Vogelstein and Kinzler 1993; UNSCEAR 2000b). There is good evidence that this relationship applies to a range of human tumor types arising spontaneously or induced by certain environmental carcinogens (UNSCEAR 1993, 2000b).

Epidemiologic studies (Chapters 59) show that exposure to low-LET radiation can lead to the age- and time-dependent development of a wide range of tumor types that, in general, are not distinguishable from those arising in nonirradiated populations; studies with experimental animals provide essentially the same message (Chapter 3). Therefore, an initial conclusion would be that the multistage process of cancer development after ionizing radiation is unlikely to be substantially different from that which applies generally (i.e., that the DNA-damaging capacity of radiation is a crucial element in cancer risk). DNA or chromosomal mutations and the heritable effects of radiation are summarized in a subsequent section of this chapter. This initial conclusion receives much support from reviews of data on the links between radiosensitivity and DNA damage response deficiency in humans and mice (Chapters 2 and 3) and findings of candidate radiation-associated mutations in tumors of humans and experimental rodents (Chapter 3). This broad conclusion, while not excluding other mechanistic components of radiation cancer risk, particularly at high doses, underpins many of the judgments summarized below.

PROJECTION OF RISKS OVER TIME

Although the Life Span Study (LSS) cohort (Hiroshima and Nagasaki) has been followed for more than 50 years,



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 239
10 Integration of Biology and Epidemiology INTRODUCTION portance is the repair or misrepair of critical DNA lesions. Depending on its location, misrepaired DNA damage can Previous chapters of this report have reviewed major ele- lead to the appearance of gene and chromosomal mutations ments of experimental and epidemiologic studies relating to in both somatic (Chapter 2) and germline cells (Chapter 4). the tumorigenic and heritable effects of low-LET (linear en- The establishment of an intimate relationship between ergy transfer) ionizing radiation. The development of views DNA damage responses, somatic mutation, and cancer de- on the risks to health from exposure to ionizing radiation velopment represents one of the most important advances in depends increasingly upon the establishment of scientific cancer research during the last decade (Vogelstein and coherence between judgments that stem from knowledge of Kinzler 1993; UNSCEAR 2000b). There is good evidence the biological mechanisms underlying radiation-induced that this relationship applies to a range of human tumor types health effects and the direct epidemiologic quantification of arising spontaneously or induced by certain environmental such effects. The epidemiologic modeling of radiation-in- carcinogens (UNSCEAR 1993, 2000b). duced health effects for the purposes of risk estimation relies Epidemiologic studies (Chapters 5–9) show that expo- in many cases on biological concepts developed from ex- sure to low-LET radiation can lead to the age- and time- perimental studies with cultured cells and laboratory ani- dependent development of a wide range of tumor types that, mals. This chapter draws together the most important con- in general, are not distinguishable from those arising in clusions reached from the reviews of the data. The principal nonirradiated populations; studies with experimental animals topics considered here are the intimate relationship between provide essentially the same message (Chapter 3). Therefore, cellular responses to DNA damage and health effects; the an initial conclusion would be that the multistage process of possible implications of the knowledge of cancer mecha- cancer development after ionizing radiation is unlikely to be nisms for projections of cancer risk over time and the trans- substantially different from that which applies generally (i.e., port of that risk between populations; the shape of the that the DNA-damaging capacity of radiation is a crucial dose-response relationship for cancer risk at low doses; dose element in cancer risk). DNA or chromosomal mutations and and dose-rate effects for cancer risk; the possible implica- the heritable effects of radiation are summarized in a subse- tions for cancer risk or other forms of cellular response to quent section of this chapter. This initial conclusion receives radiation; genetic factors in radiation cancer risk; and the much support from reviews of data on the links between heritable effects of radiation. radiosensitivity and DNA damage response deficiency in humans and mice (Chapters 2 and 3) and findings of candi- DNA DAMAGE RESPONSE AND CANCER RISK date radiation-associated mutations in tumors of humans and experimental rodents (Chapter 3). This broad conclusion, Chapters 2 and 3 review the largely cellular and molecu- while not excluding other mechanistic components of radia- lar data that strongly support the proposition that chromo- tion cancer risk, particularly at high doses, underpins many somal DNA, the genetic material of the cell, represents the of the judgments summarized below. principal target for the deleterious effects of ionizing radia- tion. In brief, energy deposition from low-LET electron tracks intersecting DNA or its local environment can lead to PROJECTION OF RISKS OVER TIME radical-mediated disruption of covalent bonds in DNA. The cell responds to the presence of such DNA damage in a bio- Although the Life Span Study (LSS) cohort (Hiroshima chemically complex fashion, but the outcome of greatest im- and Nagasaki) has been followed for more than 50 years, 239

OCR for page 239
240 BEIR VII most survivors who were young (under age 20) at the time a decline in the relative risk with time since exposure or of the bombings are still alive, and thus their risks at older attained age. ages, when baseline risks are greatest, have not yet been Finally, because follow-up is now reasonably complete studied. This is also true of other exposed cohorts. For leu- for all but the youngest A-bomb survivors, there is less un- kemia, risks in A-bomb survivors had dropped to negligible certainty in projecting risks forward in time than in past risk levels by the end of the follow-up period (Preston and oth- assessments. ers 1994; Pierce and others 1996). However, estimating life- time risks of solid cancer for those who are young at expo- THE TRANSPORT OF CANCER RISK BETWEEN sure requires assumptions about the time-response patterns DIFFERENT POPULATIONS of disease. Approaches that have been used in past risk as- sessments include a multiplicative projection based on the Another important issue in risk assessment is applying assumption that the excess cancer rate increases in direct risks estimated from studying a particular exposed popula- proportion to the baseline cancer rate and an additive pro- tion to another population that may have different genetic jection based on the assumption that the excess rate is con- and life-style characteristics and different baseline cancer stant and independent of the background rate. Currently risks. Specifically, the application of risk estimates devel- available data on A-bomb survivors and other cohorts make oped from Japanese atomic bomb survivors to a U.S. popu- it clear that the additive projection method is not appropri- lation is a concern. Two approaches that have been used are ate, and this method has not been used in recent years. multiplicative or relative risk transport, in which it is as- From a biological standpoint, a multiplicative projection sumed that the risks resulting from radiation exposure are of risk implies a mechanism whereby all host and environ- proportional to baseline risks, and additive or absolute risk mental factors that modify the background cancer rate have transport, in which it is assumed that radiation risks (on an an equivalent impact on radiation-induced disease. This absolute scale) do not depend on baseline risk and thus are would be the case if radiation were to act predominantly on the same for the United States and Japan. Estimates based an early stage in multistage tumorigenesis (i.e., as a tumor on relative and absolute risk transport can differ substan- initiator). By contrast, additive projection of risk would ap- tially. For example, baseline risks for stomach cancer are ply if radiation acted independently as one of many cancer- much higher in Japan than in the United States, and for this modifying factors during postinitiation cellular develop- reason, estimates of stomach cancer risks from radiation ex- ment (e.g., as a tumor promoter). Cytogenetic and molecular posure from a recent report based on absolute risk transport studies on tumorigenic mechanisms in experimental animals are nearly an order of magnitude higher than those based on (Chapter 3) suggest that acute doses of low-LET radiation relative risk transport (UNSCEAR 2000). act predominantly to initiate tumorigenesis rather than to In general, if the factors that account for the difference in promote its development. Thus, the monoclonal tumorigenic baseline risks act multiplicatively with radiation, then rela- mechanism of initiation proposed for low-LET radiation is tive risk transport would be appropriate, whereas if they act also most consistent with a multiplicative projection of can- additively, then absolute risk transport would be appropri- cer risk. In addition, epidemiologic studies of Japanese A- ate. If some factors act multiplicatively and others addi- bomb survivors and of persons exposed for medical reasons tively, the correct estimate might be intermediate to those indicate that exposure early in life results in greater risks obtained with the relative or absolute transport models. than exposure later in life, which also argues against strong Whether a factor acts multiplicatively or additively with ra- tumor-promoting activity and favors an initiation role. diation will depend on whether radiation and the factor of Although multiplicative risk projection is clearly better interest act principally as initiators of cancer or act at later supported than additive risk projection, current epidemio- stages in multistage cancer development as discussed below. logic data indicate that relative risks may decrease with in- Two approaches based on epidemiologic data can inform creasing attained age or time since exposure, especially for us regarding the most appropriate transport method. The those who were young at exposure (Thompson and others first is to compare risk estimates based on A-bomb survi- 1994; Little and others 1998; Preston and others 2002b). vors with those obtained from studies of non-Japanese Thus, it may not be appropriate to use the multiplicative pro- populations, particularly predominantly Caucasian popula- jection method without modification. Risk assessments con- tions. If estimates of the excess relative risk (ERR)1 per ducted by the United Nations Scientific Committee on the sievert are comparable, this suggests that relative risk trans- Effects of Atomic Radiation (UNSCEAR 2000), the Na- port may be appropriate, whereas if estimates of the excess tional Institutes of Health (NIH 2003), and this committee have allowed for a decline in relative risk with attained age (see Chapter 12). Because experimental animal data seldom include detailed information on age-specific baseline and ra- 1ERR is the rate of disease in an exposed population divided by the rate diation-associated cancer, these data do not inform us about of disease in an unexposed population minus 1.0.

OCR for page 239
INTEGRATION OF BIOLOGY AND EPIDEMIOLOGY 241 absolute risk (EAR)2 per 104 person-year (PY) per sievert taneous risks) are more comparable across animal strains are comparable, this suggests that absolute risk may be ap- than are absolute risks (Storer and others 1988). Thus, quan- propriate. However, other differences in the populations of- titative animal tumorigenesis data are most consistent with ten confound such comparisons. Most of the relevant expo- a relative risk transport model, although there are excep- sures occured for medical reasons, were generally tions. protracted, and often were at higher doses than those re- Current knowledge implies the following: (1) at low ceived by atomic bomb survivors, making it difficult to in- doses, radiation acts principally as an initiator of cancer terpret comparisons. Additional difficulties are dosimetry (Chapter 3), and (2) many of the known cancer risk factors uncertainties and statistical variation, which is quite large in such as hormonal or reproductive factors, particularly for some studies. Furthermore, although many studies report breast cancer risk, and chronic inflammation associated with estimates of the ERR per gray (ERR/Gy), few report esti- microbial infection, for stomach and liver cancers (dis- mates of the EAR per gray. Comparisons of estimates from cussed in this chapter), tend to act at later stages in multi- the LSS and medical studies are also discussed in the mate- stage tumorigenesis. In these latter cases, cancer risk modi- rial below on breast cancer and at the end of Chapter 12 af- fication is believed to be associated largely with the ter the BEIR VII risk estimates have been presented. postinitiation clonal expansion of preneoplastic or malig- A second approach based on epidemiologic data is to in- nant cells (Chapter 3). Genetic factors acting throughout vestigate interactions of various risk factors with radiation. cancer development may also modify risk (Chapter 3). However, there are few studies with available data on both Biologically based risk projection models provide a sim- radiation and other risk factors and with sufficient power to plistic, but useful, intuitive framework to evaluate the pos- distinguish multiplicative and additive interactions. Rel- sible role of radiation in populations with different distribu- evant data are reviewed below. A detailed discussion of in- tions of risk factors for specific cancer types. An example of teractions is given by UNSCEAR (2000b, Appendix H). such modeling approaches is given in Annex 10A, which In the sections that follow, the committee first discusses summarizes judgments that can be made on the transport of the type of interaction that would be expected based on con- cancer risk using the Moolgavkar and Knudson two stage sideration of whether radiation and other risk factors act pri- clonal expansion model, viewing low-LET radiation as a tu- marily as initiators or promoters. Because the correct trans- mor initiator. In simple terms, the model predicts that in the port model is not necessarily the same for all cancer sites, case of a radiogenic tumor type with a strong influence of this is followed by a discussion of cancers of each of several promoters, one would favor a relative risk transportation specific sites. The etiology of each site-specific cancer is model, whereas in the case of a tumor type with a strong discussed briefly, including the role of various risk factors. influence of initiators, one would favor an absolute risk A discussion of epidemiologic studies that address interac- transportation model. tions of radiation and other risk factors then follows. Although baseline risks for all solid cancers (as a single Etiology of Cancer at Different Sites category) do not differ greatly between the United States and Japan, this occurs because of the canceling out of As briefly illustrated in Annex 10A, knowledge of the site-specific cancers that are higher in the United States (in- mechanistic factors that underlie tumor etiology can provide cluding breast, colon, lung, and prostate) and site-specific an important input to judgments on the most appropriate cancers that are higher in Japan (including stomach and methodology for transportation of radiation cancer risk be- liver). If the correct transport models differ by site, esti- tween different populations. This section provides an over- mates of all solid cancers based on relative and absolute risk view of the etiology of a selection of radiogenic human tu- transport may not fully reflect the transport uncertainty. mors. Postirradiation Cancer Mechanisms and Choice of Stomach Cancer Transport Model Stomach cancer is a disease with a much higher back- Animal studies (Chapter 3) suggest that low-LET radia- ground incidence in Japan than in the United States (IARC tion acts principally as an initiator of tumorigenesis and is at 2002). Risk factors for gastric cancer include the presence best a weak tumor promoter. In addition, for many tumor of conditions such as chronic atrophic gastritis, gastric ul- types, relative risks (ratio of radiation-associated and spon- cer, atrophic gastritis, and autoimmune gastritis associated with pernicious anemia. These cause an excessive rate of cell proliferation in the gastric epithelium and are therefore likely to act as promoters, increasing the chance of fixation 2EAR is the rate of disease in an exposed population minus the rate of of replication errors induced by radiation and dietary car- disease in an unexposed population. cinogens (IARC 2003). Helicobacter pylori infection of the

OCR for page 239
242 BEIR VII stomach appears to be a strong risk factor for stomach can- oshima and Nagasaki during the same period were lower cer, and its effect is likely to be mainly through tumor pro- (40–44 per 100,000 in men and 11.8–12.9 per 100,000 in motion (although there is increasing evidence that it may women), particularly among women (IARC 2002). also cause tumor initiation; Parsonnet and others 1994; The major risk factors for lung cancer are tobacco con- Aromaa and others 1996; Goldstone and others 1996). En- sumption, occupational exposure to a number of carcino- vironmental risk factors include low consumption of fruit gens, and air pollution (Pope and others 2002; IARC 2003). and vegetables; consumption of salted, smoked, or poorly Geographic and temporal differences in lung cancer inci- preserved foods; and cigarette smoking (Fuchs and Mayer dence are determined overwhelmingly by tobacco consump- 1995). The majority of these agents are likely to influence tion (IARC 2003). the promotion of tumors. Tobacco smoke contains approximately 4000 specific The above considerations would therefore suggest that chemicals, including nicotine, polycyclic aromatic hydro- for stomach cancer, relative risk transport may be better sup- carbons, N-nitroso compounds, aromatic amines, benzene, ported than absolute risk transport. This is also supported by and heavy metals. Lung cancer is not thought to be attribut- a study of predominantly male peptic ulcer patients, where able to any one chemical component, but rather to the effect the estimated ERR/Gy based on patients with doses to the of a complex mixture of chemicals in tobacco smoke, which stomach of less than 10 Gy (mean 8.2 Gy) was 0.20 (95% may act at different stages of the carcinogenic process. CI 0, 0.73), very similar to that based on male A-bomb sur- Based on the mechanistic arguments above, this suggests vivors (Carr and others 2002; see Table 12-2). that neither a pure absolute nor a pure relative risk transport model is appropriate. The estimated ERRs/Gy for lung cancer in several stud- Liver Cancer ies involving medical exposures in predominately Cauca- The incidence of liver cancer (mainly hepatocellular car- sian patients are lower than those based on A-bomb survi- cinoma) is also much higher in Japan than in the United vors (Table 6-3), and this might be interpreted as indicating States (IARC 2002). The main risk factors for this disease that absolute risks are more comparable than relative risks. are chronic infection with hepatitis B or C virus, dietary ex- However, the lower ERR estimates may also have resulted posure to aflatoxins, and chronic alcohol consumption from other differences in the study populations, particularly (IARC 2003). Tobacco smoking also plays a role in the eti- the much higher doses in several of the medical studies. ology of liver cancer (IARC 2004). Pierce and colleagues (2003) evaluated the joint effect of Aflatoxins induce mutations in several genes involved in smoking and radiation exposure on lung cancer risks in A- hepatocellular carcinoma and are thus likely to be involved bomb survivors and found that they were significantly in the early or initiating stages of carcinogenesis. Hepatitis submultiplicative and consistent with an additive model. B and C infections and alcohol consumption, on the other They also demonstrate that inferences about the modifying hand, are likely to be involved in the promotion of tumors. effects of gender and age at exposure on the ERR/Gy can be They are thought to increase the risk of liver cancer through distorted if analyses do not account for smoking; this is be- inflammation that may result in liver cirrhosis. The latter is cause smoking habits in the LSS cohort depend strongly on the major clinical determinant of hepatocellular carcinoma, both factors. with 70–90% of these tumors developing in patients with By contrast, studies of lung cancer risks in underground macronodular cirrhosis (IARC 2003). miners exposed to radon (NRC 1999) or of Hodgkin’s dis- Baseline risks for liver cancer are much higher in Japan ease (HD) patients treated with high doses of radiation (Gil- than in the United States, and rates of infection with hepati- bert and others 2003) rejected additive interactions and tis B and C undoubtedly contribute to this difference. The found that multiplicative interactions were compatible with mechanistic arguments above and the limited epidemiologic the data. However, these studies may be less relevant for data tend to support the use of the multiplicative transporta- estimating the risks of low doses of low-LET radiation than tion model. those of A-bomb survivors. Underground miners were ex- posed to α-emitting (high-LET) radon progeny. In addition, the evidence for a multiplicative relation of radiation and Lung Cancer smoking comes primarily from analyses of data on miners Lung cancer is the most common cancer worldwide and in Colorado and China, where doses to the lung (in sieverts) the major cause of death from cancer, particularly among were much higher than in the LSS cohort (NRC 1999). men (IARC 2003). In the United States, based on SEER Although data on miners were compatible with a multipli- (Surveillance, Epidmiology, and End Results) registry data, cative effect and not with an additive one, the estimated the annual incidence rates, age-standardized to the world interaction was submultiplicative. HD patients were also population, were 55.7 and 33.5 per 100,000, respectively, in exposed to very high doses (mean dose to the lung 25 Gy) men and women in 1993–1997. Comparable rates in Hir- and, in addition, were subject to the immunodeficiency

OCR for page 239
INTEGRATION OF BIOLOGY AND EPIDEMIOLOGY 243 inherent to this lymphoma and associated with the chemo- be expected to act as tumor promoters. The above consider- therapy that was also given to many of these patients. ations would therefore suggest that the preferred transporta- In summary, the absolute risk transport model has greater tion model for breast cancer should be based on a multipli- support for lung cancer than for stomach or liver cancer. cative model. Mechanistic considerations suggest that the correct model The female breast is one of the few cancer sites for which may be intermediate between relative and absolute risk. extensive epidemiologic data on predominantly Caucasian populations are available, and this makes it possible to base risk estimates directly on Caucasian data, avoiding the need Breast Cancer to transport risks. Nevertheless, it is useful to evaluate what Breast cancer is the most common cancer and one of the these data tell as about appropriate transportation models. leading causes of death from cancer among women world- Land and colleagues (1980) conducted parallel analyses wide, with nearly 1,000,000 new cases per year. Known risk of cancer incidence data in Japanese A-bomb survivors, factors for breast cancer include reproductive factors, post- Massachusetts tuberculosis fluoroscopy patients, and New menopausal increased weight, and history of proliferative York women treated with radiation for mastitis, and found benign breast disease (IARC 2003). Differences in cancer that absolute risks were comparable for the three cohorts incidence between U.S. and Japanese populations have been whereas relative risks were much larger in the Japanese co- attributed to the tumor promotion effects of hormonal fac- hort. This was recently confirmed in a pooled analysis of tors (Moolgavkar and others 1980). breast cancer incidence in several cohorts by Preston and In addition, a strong genetic contribution to the risk of coworkers (2002a). In this study, models that were similar spontaneous breast cancer has been shown by the increased in form could be used to describe breast cancer incidence in cancer incidence among women with a family history of A-bomb survivors and in U.S. women (Massachusetts fluo- breast cancer. A number of genes involved in DNA damage roscopy patients and the Rochester infant thymus irradia- response pathways, including BRCA1, BRCA2, and less cer- tion cohort). The overall ERR/Gy was about three times as tainly ATM, have been found to confer genetic susceptibil- large in the Japanese cohort, whereas the EAR/Gy was simi- ity to breast cancer. Alterations in the activity of ATM, lar for the LSS and the U.S. cohorts. However, since fluo- BRCA1, and BRCA2 proteins may have far-reaching conse- roscopy exposure is protracted and involves lower-energy quences in the control of genetic stability and the risk of photons than A-bomb exposure, these differences in expo- tumor development. The presence of sequence variants that sures might confound the comparison. Also, in a pooled alter either the expression or the function of these genes analysis of breast cancer mortality in Canadian fluoroscopy could therefore influence gene-environment interactions patients and A-bomb survivors, neither the ERRs nor the and enhance the increased breast cancer risk in women fol- EARs were found to differ significantly between the cohorts lowing radiation exposure (see Chapter 3). (Howe and McLaughlin 1996), although the ERR for the There is no study published on BRCA1 or BRCA2 muta- combined LSS women was nearly four times that for non- tion frequency in the Japanese population. However, since Nova Scotia Canadian women. Little and Boice (1999) and the prevalence of these mutations in relatively large studies Brenner (1999) provide additional discussion of these issues of breast and breast-ovarian cancer in Japanese families is with a commentary by Ullrich (1999). similar to that in Europe and North America, it is likely that In a case-control study of breast cancer among A-bomb BRCA1 and BRCA2 mutation frequencies will be the same survivors, Land and colleagues (1994a) evaluated the inter- in Japanese and Caucasians. In Caucasians, the frequency action of several risk factors for breast cancer with radiation of BRCA1 was estimated to be 0.051% (95% CI 0.021, and found that the relationship was better described by a 0.125) and of BRCA2 0.068% (95% CI 0.033, 0.141; multiplicative model than an additive one. This, together Antoniou and others 2002). Thus, slightly more than one with the etiological and mechanistic considerations above, individual in 1000 is a carrier of the BRCA1 or BRCA2 mu- would seem to favor relative risk transport, in contradiction tation. For ATM there is no information about the frequency to the higher ERR/Gy observed in A-bomb survivors and of heterozygotes in the Japanese population. However, for noted in the preceding paragraph; these observations, how- ATM and other possible breast cancer genes, as a first ap- ever, might have come about because of other differences proximation it is assumed that there are not major differ- between the Japanese and U.S. cohorts. ences in gene frequencies among populations in Japan and In summary, mechanistic considerations and some epi- Europe or North America. demiologic data support relative risk transport. However, di- Thus, in the absence of more detailed data on mutation rect use of data on predominantly Caucasian populations re- and polymorphism frequencies in Japan and the United sults in estimates that are comparable to those based on States, the main differences in breast cancer incidence be- A-bomb survivors on an absolute risk scale, but not on a tween these two countries are judged to relate to reproduc- relative risk scale. tive history and, implicitly, to hormonal factors that would

OCR for page 239
244 BEIR VII Thyroid Cancer The etiology of leukemia is not well established. Apart from ionizing radiation, occupational exposure to agents Cancer of the thyroid is a rare disease, accounting for such as benzene can increase the risk of leukemia, as can only about 1% of cancer cases in developed countries. The exposure to some chemotherapeutic agents. Some risk fac- incidence is highest in Iceland and Hawaii (IARC 2002). tors such as Down’s syndrome and exposure to extremely Annual incidence rates in Japan and the United States are low frequency magnetic fields have been postulated as risk similar (of the order of 2–3 per 100,000 among men and 7– factors for childhood leukemia. Infection by the HTLV-1 10 per 100,000 in women, for rates age-standardized to the virus is responsible for adult T-cell leukemia, a disease ob- world population), and incidence rates have been increasing served in Japan, but rarely in the United States (IARC 2003). worldwide in the last decades. Thyroid cancer in childhood Conversely, chronic lymphocytic leukemia, a neoplasm of B is a very rare disease, with an annual incidence of less than lymphocytes, is rare in Japan but more frequent in the United one case per million per year in most developed countries. States. Thyroid cancer is about three times as frequent in women Based on the above, it is not currently possible to draw as in men, suggesting that hormonal factors may play a role conclusions about mechanisms of carcinogenesis, and there- in its etiology, although results from epidemiologic studies fore transport models, except to note that the different preva- of reproductive factors are inconsistent. Iodine deficiency is lence of infection with a number of viruses including HTLV- thought to be involved in the development of papillary thy- 1 and viruses involved in B-cell lymphomas may account for roid cancer, as may the consumption of some cruciferous a difference in the incidence of specific leukemia subtypes and goitrogenic vegetables (IARC 2003). Experimental stud- between Japan and the United States. ies have shown that excessive long-term stimulation of the thyroid gland by thyroid-stimulating hormone, such as results from iodine deficiency, can lead to tumor formation Conclusions with or without addition of a mutagenic agent (Thomas and At present, neither knowledge of biological mechanisms Williams 1991). nor data from epidemiologic studies are sufficient to allow History of goiter and benign thyroid nodules is associated definitive conclusions regarding the appropriate method for with papillary thyroid cancer risk, as is family history of transporting risks, although mechanistic considerations sug- thyroid cancer; the possible role of increased thyroid screen- gest somewhat greater support for relative risk than for abso- ing in these associations is unclear at present. Medullary thy- lute risk transport. For this reason, the committee provides roid carcinoma, a rare type of thyroid cancer, has a very estimates based on both relative risk and absolute risk trans- strong genetic component (IARC 2003). port. When a single estimate is needed, a weighted mean of Because the majority of the risk factors listed above (hor- the two estimates can be used. For cancer sites other than mones, iodine deficiency) are likely to influence the promo- breast, thyroid, and lung, the committee recommends a tion of tumors, mechanistic considerations suggest that the weight of 0.7 for the estimate obtained using relative risk preferred transportation model for thyroid cancer should be transport and a weight of 0.3 of the estimate obtained using based on relative risk transport. It is noted that the BEIR absolute risk transport with the weighting done on a loga- model for thyroid cancer risk is based on a combined analy- rithmic scale. This choice of weights, which clearly involves sis of epidemiologic studies carried out in different countries subjective judgment, was made because the mechanistic con- (U.S., Japan, Israel). The multiplicative model developed by siderations discussed above suggest somewhat greater sup- Ron and coworkers (1995b) was applied directly with the port for relative risk transport, particularly for cancer sites uncertainty that reflects international variation in thyroid (such as stomach, liver, and female breast) for which known cancer risk. risk factors act mainly on the promotion or progression of tumors. The choice reflects uncertainty regarding which Leukemia model is correct and also allows for the possibility that some Leukemia comprises about 3% of all incident cancers factors interact additively with radiation, whereas others in- worldwide; the age-standardized incidence in the United teract multiplicatively. The uncertainty involved in this States (standardized to the world population), based on choice is reflected in the subjective confidence intervals that SEER registry data, was 10.8 and 6.7 per 100,000, respec- are provided as discussed in Chapter 12. tively, in men and women during 1993–1997. Rates in Exceptions to the approach noted above are made for can- Nagasaki prefecture during the same period were similar (9.4 cers of the breast, thyroid, and lung. For breast and thyroid and 6.2 per 100,000, respectively, in men and women), while cancer, extensive data on Caucasian populations are avail- they were lower in Hiroshima (6.1 and 4.7 per 100,000, re- able and can be used directly to estimate risks. The com- spectively; IARC 2002). It should be noted that these rates mittee’s preferred models, which are described in Chap- include chronic lymphocytic leukemia, which is known to ter 12, make use of these data. For lung cancer, analyses of be rare in Japan but is more frequent in the United States. the A-bomb survivor data by Pierce and colleagues (2003)

OCR for page 239
INTEGRATION OF BIOLOGY AND EPIDEMIOLOGY 245 support an additive interaction of smoking and radiation. the present report has placed much emphasis on the mecha- Since differences in smoking habits undoubtedly contribute nistic data that can underpin such judgments. to the differences in baseline risks in Japan and the United The following data and conclusions are given in Chap- States, this finding supports the use of absolute risk trans- ters 1, 2, and 3 and are most pertinent to radiation risks in port. Furthermore, lung cancer analyses of A-bomb survivor the dose range 0–100 mSv where epidemiologic and animal data based on EAR models may provide a more reliable data are inadequate. evaluation of the dependence of radiation risk on factors such First, there is evidence that most cancers are monoclonal as gender and age at exposure than do ERR models, as dis- in origin (i.e., they develop from progeny of a single abnor- cussed above. As indicated in Chapter 12, relative risk trans- mal cell; UNSCEAR 1993). Whatever molecular mecha- port estimates are based on ERR models, whereas absolute nism is envisaged for radiation, at very low doses (e.g., 0– risk transport estimates are based on EAR models. Thus, for 5 mGy low LET), increases in dose simply increase the lung cancer, the weighting scheme used for most other solid probability that a given single cell in the tissue will be inter- cancers is reversed, and a weight of 0.7 is used for the esti- sected by an electron track which will have a nonzero prob- mate obtained with absolute risk transport and a weight of ability of inducing a biological effect. Therefore, at these 0.3 for the estimate obtained with relative risk transport. very low doses, a linearity of response is almost certain For sites other than breast, thyroid, and lung, it is likely (Chapter 3). that the correct transport model varies by site. However, the Second, given the intimate relationship established be- committee judged that current knowledge was insufficient tween DNA damage response, gene or chromosomal muta- to provide separate approaches for other specific sites. tions, and cancer development, the form of the dose-re- sponse for mutation induction in single cells should be broadly informative for cancer initiation. Data from a large- FORM OF THE DOSE-RESPONSE FOR RADIATION scale study noted in Chapter 2 suggest a linear relationship TUMORIGENESIS between low-LET dose and chromosomal mutation down to Follow-up of cancer incidence and mortality in Japanese around 20 mGy. A-bomb survivors (the LSS study) continues to provide the A central question addressed in this report is the nature most informative epidemiologic data on the shape of the of critical DNA lesions after low-LET radiation and the ex- dose-response for solid tumors and leukemia (Chapter 6), tent to which they may be repaired by the cell without er- although studies of large-scale populations with low-dose rors. This is a crucial judgment in radiation tumorigenesis chronic exposures are increasingly informative about the ef- since, at the level of cancer-associated gene or chromo- fects of low doses. somal mutation, the presence of a true dose threshold de- Atomic bomb survivor data for solid tumors combined mands totally error-free DNA damage response and repair. provide statistical evidence of a radiation-associated excess The detailed information available on the importance of at doses down to around 100 mSv; these combined data are a chemically complex DNA double-strand break (DSB) in- well described by a linear no-threshold dose-response, al- duced by a single ionization cluster for postirradiation bio- though some low-dose nonlinearity is not excluded (Pierce logical effects (Chapter 1), together with the predominance and Preston 2000; Preston and others 2003). Indeed, dose- of error-prone nonhomologous end joining (NHEJ) repair response relationships for individual tumor types in the LSS in postirradiation cellular response, argues strongly against can differ, and for nonmelanoma skin cancer the dose re- a DNA repair-mediated low-dose threshold for cancer ini- sponse is highly curvilinear with an excess seen only at doses tiation (Chapters 1–3). The same data provide a strong higher than around 500 mSv. The LSS dose-response for leu- counter to pro-threshold arguments based on the relative kemia is also clearly curvilinear, with a statistically signifi- abundance of spontaneously arising and radiation-induced cant excess being evident at doses around 200 mSv. DNA damage. Those arguments fail to take account of the The above human data well illustrate the problems of lim- quality of the repair achievable for simple and complex ited statistical power that surround epidemiologically based forms of DNA damage. conclusions on the shape of the low dose-response for radia- In principle, complex DNA DSBs may be repaired with tion cancer risk and how it might vary between tumor types. full fidelity by homologous recombination (HR) pathways. Similar difficulties surround judgments based on data ob- Since HR operates almost exclusively between sister chro- tained using experimental animals; many studies are broadly matids in cells that have newly replicated their DNA (Chap- consistent with a linear no-threshold dose response, but there ters 1 and 2), the cell has a limited cell cycle window for are a number of examples of highly curvilinear, threshold- such error-free repair. At any one time, only a small frac- like relationships (Chapter 3). tion of stem-like target cells in tissues are expected to re- It is abundantly clear that direct epidemiologic and ani- side within this postreplication window—many will be in a mal approaches to low-dose cancer risk are intrinsically lim- nonreplicative, quiescent state (e.g., Potten and Hendry ited in their capacity to define possible curvilinearity or dose 1997; Kountouras and others 2001; Young 2004). On this thresholds for risk in the range 0–100 mSv. For this reason basis HR-mediated error-free repair is unlikely to be the

OCR for page 239
246 BEIR VII dominant feature of in vivo cellular response and tumor with the activity of cellular DNA repair than with that of induction. postirradiation whole-tissue remodeling, thus drawing Finally, evidence is emerging that the DNA deletions that together dose-rate effects at the cellular and whole-animal are characteristic molecular footprints of NHEJ-mediated levels (Chapter 3). misrepair and gene loss in cultured cells are also seen as Animal tumorigenesis data and related information from early events in radiation-induced tumors in rodents; there is life-shortening studies (Chapter 3) may be used to provide also preliminary evidence pointing toward the involvement judgments on DDREF that vary up to a value of 10 or more of NHEJ misrepair in the genesis of early arising RET gene (UNSCEAR 1994). However, when those tumor types that, rearrangements in post-Chernobyl childhood thyroid cancer atypically, depend strongly on cell killing are excluded and (Chapter 3). analysis is restricted to doses up to a few grays, the DDREF When considered together, these in vitro and in vivo data values obtained are in the range of 2–3 (Chapter 3). These are seen to provide a scientifically coherent linkage between values are similar to those of gene mutation and, thereby, error-prone postirradiation repair of chemically complex broadly consistent with the recurring theme of a close asso- DNA DSBs in target cells in vivo and tumor induction. ciation between DNA damage response, mutation induction, Mechanistic uncertainties remain, but the weight of avail- and cancer. able evidence would argue against the presence of a low The biological picture overall is that cellular and animal dose threshold for tumor induction based on error-free repair data relating to protracted radiation exposures provide a con- of initial DNA damage. In summary, the committee judges vincing argument for the inclusion of DDREF in judgments that the balance of scientific evidence at low doses tends to about cancer risk at low doses and low dose rates. The ani- weigh in favor of a simple proportionate relationship be- mal data showing reduction in carcinogenic effectiveness, tween radiation dose and cancer risk. including life shortening, following protracted exposure con- stitute the strongest element in this argument; the coherence of the mechanistic data adds additional weight. DOSE AND DOSE-RATE EFFECTS ON TUMOR An alternative approach is to estimate DDREF on the INDUCTION basis of the degree of curvature of the dose-response for ex- Since much of the informative epidemiologic data on low- cess cancer after acute irradiation. Conventional radiobio- LET radiation cancer risk derives from the study of acute logical theory holds that the initial linear (α) term of a lin- exposures, it is necessary to make somewhat indirect judg- ear-quadratic (αD + βD2) dose-response (where D is the ments about the magnitude of the expected reduction in risk dose) will represent the low-dose and low-dose-rate re- associated with low doses and dose protraction. This reduc- sponse. Accordingly, the α and β terms of the acute dose- tion in risk is conventionally described by the dose and dose- response may be used to provide an estimate of DDREF. rate effectiveness factor (DDREF). As illustrated and dis- Note that the BEIR V committee did not apply a DREF (sic) cussed in Chapter 2 (see Figure 2-1), the reduction in risk for in its analysis of solid tumor data and used a linear-quadratic low doses (DEF) and the reduction in risk for dose protrac- model for leukemia (NRC 1990). Also, the UNSCEAR tion (i.e., low dose rates; DREF) are assumed to be equal; (2000) committee commented that the LSS data suggested a therefore, the term DDREF is used for estimating effects for “value of about 1.5” for the DDREF. In its report, the Inter- either low doses or low dose rates. national Commission on Radiological Protection (ICRP Information from cellular and molecular studies strongly 1990) stated that “the Commission has decided to recom- suggests that dose and dose-rate effects of low-LET radia- mend that for radiation protection purposes the value of 2 be tion are determined largely by the activity of DNA damage used for the DDREF, recognizing that the choice is some- response process in cells. For the induction of gene and chro- what arbitrary and may be conservative.” mosomal mutations in cultured somatic cells, values for The committee has taken a computational approach to DDREF generally fall in the range of 2–4 (Lloyd and others the estimation of DDREF that is based on a Bayesian analy- 1992; Thacker 1992; UNSCEAR 1993, 2000b; Cornforth sis of combined dose-response data. The data sets consid- and others 2002), although higher and lower values have ered were (1) solid cancer incidence in the LSS cohort of been recorded in some mutation systems. Together, these Japanese atomic bomb survivors; (2) cancer and life short- data are consistent with the view that the temporal abun- ening in animals; and (3) chromosome aberrations in hu- dance of radiation-induced DNA damage is a major factor in man somatic cells. the efficiency or fidelity of DNA repair and hence the fre- quency of induced mutation (Chapter 2). Derivation of the Dose and Dose Rate Reduction Factor by In vivo effects of dose protraction or fractionation are Bayesian Analysis likely to be more complex, but available data on animal tum- origenesis show that the reduction of tumor yield with dose The BEIR VII cancer risk estimates are based on risk fractionation is determined by processes that operate on a models derived primarily from analyses of data on the LSS time scale of up to 24 h. This time scale is more consistent cohort of Japanese atomic bomb survivors. Historically, and

OCR for page 239
INTEGRATION OF BIOLOGY AND EPIDEMIOLOGY 247 with the exception of leukemia, there has been little statisti- to convert a risk estimate from the high-dose linear approxi- cal evidence of a need for curvature in the LSS dose-response mation to the more appropriate low-dose linear approxima- models and substantial reliance on models in which risk is tion, as shown in the figure. The association between the simply a linear function of radiation dose (Pierce and Preston form of the dose-response at acute doses and the effects of 2000; Preston and others 2003). There is stronger evidence dose-rate is discussed in Chapter 2 and in Annex 10B. of curvature from radiobiological considerations and experi- This DDREF clearly must depend on what is meant by mental results. The DDREF has been used in the past as a high dose and should not be mistakenly thought of as a uni- device for allowing risk estimates to conform to this expected versal low-dose correction factor. Furthermore, of particular curvature but without abandoning the LSS linear models (ICRP interest here is what might more appropriately be called an 1991; NCRP 1993; EPA 1999; UNSCEAR 2000b; NIH 2003). LSS DDREF, where a curvature adjustment to risk estimates A rationale for DDREF is illustrated in Figure 10-1 for a from LSS-estimated linear models is based on a wide range setting that mimics a simple animal experiment on cancer of doses. The line analogous to the “high-dose linear ap- induction by acute-dose low-LET radiation in which risks proximation” of Figure 10-1 is the one that results from lin- are observed only at two doses: zero and some particular ear model estimation with the LSS data. If a certain degree “high dose.” If the true dose-response relationship is con- of curvature is presumed, then it is possible to define an LSS cave up to that dose, as the radiobiological data tend to sug- DDREF that correctly adjusts LSS linear risk in order to gest, then a line connecting the excess risk at high dose to the estimate cancer risk at low doses. Such a definition is pro- origin would tend to have a larger slope than a line that ap- vided after the discussion of a numerical characterization of proximates the dose-response curve at doses near zero. The dose-response curvature upon which it is based. DDREF in this case is the ratio of these two slopes (i.e., the If, over some dose range of interest, the dose-response risk per unit of dose at high dose divided by the risk per unit curve can be approximated by a linear-quadratic (LQ) func- of dose at low dose). If this ratio is known then it can be used tion, αDose + βDose2, then the slope of the high-dose linear 1.0 The true dose-response curve for ERR (or EAR) as Low Dose Range 1 a function of dose is thought to have some inward curvature, with possible leveling off at higher doses. 0.8 Excess Relative Risk of Cancer, ERR 2 A linear approximation at some high dose has slope s H. 4 The DDREF is the ratio 0.6 of the two slopes, sH / sL . If known, it can be used to convert a risk estimate (at dose D) from the high dose linear approximation, ERRH = 0.4 s H D, to the low dose one, ERRL = sL D .ß Since the low dose and low dose rate slopes are equivalent it can also be 0.2 3 used to convert a risk A better line for low-dose estimate from high dose to risk estimation is the tangent low dose rates. That is, ERRL of the dose response curve at =ERR LDR =ERR H /DDREF. 0, with slope sL. 0.0 0 1 2 3 4 Radiation Dose (Gy), D FIGURE 10-1 A hypothetical dose-response curve with a linear approximation for low doses (i.e., the tangent of the curve at dose zero) and a linear approximation based on risk at one particular high dose (i.e., the line that passes through the origin and the true dose-response curve at the high dose), when the high dose is taken to be 2 Gy. The DDREF at this high dose is the larger slope divided by the smaller slope. An explanation of why this low-dose effect also applies to low-dose-rate effects is provided in Chapter 3.

OCR for page 239
248 BEIR VII approximation (sH in Figure 10-1) at a particular high dose The two definitions of DDREF as a function of LQ curva- DH is α + βDH, the slope of the low-dose linear approxima- ture must be clearly distinguished: the fixed high-dose tion (sL in Figure 10-1) is α, and the DDREF corresponding DDREF (or UNSCEAR definition), DDREF = 1 + θ × high to DH is their ratio, 1 + (β / α)DH (UNSCEAR 1993). A natu- dose, and the LSS DDREF defined by the estimation process ral numerical quantity for curvature characterization, there- in the preceding paragraph. The first is a function of θ and fore, is β / α, which is not tied to any particular high dose. some particular high dose. The second is a function of θ and This ratio is referred to here as the LQ “curvature” and is the LSS data. Their relationship, as illustrated in Table 10-1, represented by the symbol θ (i.e., the reciprocal of the so- indicates that the LSS DDREF based on A-bomb survivors called crossover dose). with doses of 1.5 Sv or less is roughly equivalent to the fixed If the correct curvature, θ, is known, then an LSS DDREF high-dose DDREF at an effective high dose of about 1 Sv. In may be defined through the following steps: an LQ model other words, in terms of the familiar UNSCEAR single high- for ERR or EAR is estimated from the LSS data in such a dose definition, one can act as if the nonzero LSS doses were way that the curvature is constrained to be θ, that is, by fit- concentrated at a dose of 1 Sv. ting the relative risk model αLQ(Dose + θDose2) for fixed θ Table 10-1 may be used as an aid in interpreting radiobio- and with unknown linear component αLQ. A separate linear logical evidence for curvature. If, for example, radiobiology model is estimated from the same data: αLDose, with linear data indicate that a DDREF of 2 is appropriate for adjusting component αL. The LSS DDREF is the estimate of the ratio risks based on a linear model derived at the single high dose of the two linear components, α L / αLQ. The resulting of 2 Sv, then the implicit curvature is 0.5 Sv–1 and the corre- DDREF can be used to convert a risk estimate based on the sponding LSS DDREF is 1.5. linear model projection to one based on the linear compo- The committee estimates LSS DDREF in this report by nent of an estimated LQ model with curvature determined combining radiobiological and LSS evidence concerning by a given choice of the value of θ. Figure 10-2 illustrates curvature via a Bayesian statistical analysis and applying the the definition for two possible choices of this value. definition of LSS DDREF to the result. As detailed in Annex 1.5 Low Dose Range 1.0 Excess Relative Risk 0.5 Linear fit, 0-1.5 Sv LQ fit when curvature = .3, 0-1.5 Sv LQ fit when curvature = .7, 0-1.5 Sv 0.0 0.0 0.5 1.0 1.5 2.0 Adjusted Dose (Sv) FIGURE 10-2 Illustration of LSS DDREF. Plotted points are the estimated ERRs for solid cancer incidence (averaged over sex, for individu- als exposed at age 30 at attained age 60) from LSS subjects in each of 11 dose categories. The vertical lines extend two standard errors above and below the estimates. The solid line is a linear fit to the data for dose range 0–1.5 Sv, with slope αL = 0.56. The other two curves are estimated LQ models for the same dose range, when the curvature, θ, is constrained to be 0.3 Sv–1 (resulting in estimated linear coefficient αLQ = 0.43) and 0.7 Sv–1 (resulting in estimated linear coefficient αLQ = 0.32). The LSS DDREFs that result from these are 0.56 / 0.43 = 1.3 and 0.56 / 0.32 = 1.8, respectively.

OCR for page 239
INTEGRATION OF BIOLOGY AND EPIDEMIOLOGY 249 TABLE 10-1 UNSCEAR Definition of DDREF and LSS Table 10-2 summarizes the graphical results of Figure DDREFa Corresponding to Three Values of Curvature 10-3. A single estimate of curvature is the median of the posterior distribution: 0.5 Sv–1, corresponding to an LSS UNSCEAR DDREFb DDREF of 1.5. On the basis of these analyses, there is little disagreement between the radiobiological and LSS estimates Curvature High Dose High Dose High Dose LSS of LSS DDREF. While a quadratic term in an LSS LQ model (θ, Sv–1) = 1 Sv = 2 Sv = 3 Sv DDREFc is not significantly different from zero (twosided p-value = 0.5 1.5 2.0 2.5 1.5 .2, for the 0–1.5 Sv dose range), the single best estimate of 1.0 2.0 3.0 4.0 2.1 LSS DDREF from the LSS subset is 1.3. If the radiobiologi- 2.0 3.0 5.0 7.0 3.1 cal estimate of 1.5 seems low, the committee believes that it is due not to a new interpretation of radiobiological curva- aFor incidence of solid cancers and based on doses between 0 and 1.5 Sv, ture but rather to the use of an LSS DDREF that is specific to as in Figure 10-2. bDDREF = 1 + θ × high dose. the needs of LSS linear model adjustment to account for the cFrom estimating LQ models forced to have curvature θ. curvature. As evident in Table 10-1, a DDREF suitable for LSS adjustment is less than that expected for low-dose ex- trapolation of estimates based on high doses of 2 to 3 Sv. The Bayesian approach formalizes the connection among 10B, the radiobiological information comes from mouse ex- the DDREF, the LQ curvature in radiobiology, and the LSS periments, via models estimated from direct cancer risk data data. However, there are two reasons for the continuing un- and models estimated from cancer-associated life-shorten- certainty in the estimation of DDREF: (1) there is substantial ing data. The resulting posterior distribution for possible inconsistency and imprecision in the data from animal ex- values of LSS DDREF is displayed in Figure 10-3. periments; and (2) the curvature estimates from radiobio- 1.5 Radiobiological prior LSS Likelihood Combined posterior 1.0 Probability Density 0.5 0.0 Theta: 0 1 2 3 4 LSS DDREF: 1 2 3 4 FIGURE 10-3 Results of a Bayesian statistical analysis of dose-response curvature and associated LSS DDREF values. The probability density labeled “radiobiological prior” expresses the belief about curvature deduced from animal data, as detailed in Annex 11B. Regions of high density correspond to more believable values of curvature. The LSS likelihood is the likelihood function of curvature θ from the data displayed in Figure 10-2. The “combined” density is the Bayesian posterior obtained by updating the radiobiological density to account for information from the LSS data. The scale below the plot shows the implied values of LSS DDREFs corresponding to the θ scale. NOTE: The committee judges it preferable to choose a cutoff dose that lies within the lower rather the higher portion of the possible range.

OCR for page 239
250 BEIR VII TABLE 10-2 Maximum Likelihood Estimates of More generally, since a linear model fits the LSS data Curvature and Corresponding Estimates of LSS DDREFa over the entire range (for cancers other than leukemia), it is and the Posterior Median from the Bayesian Analysis that important to question why the expected curvature fails to Combines the Two materialize and whether the absence of curvature necessarily implies that the LSS low-dose slope is too large. It could be (95 % LSS (95% that a linear relationship is the result of some cancelation of Estimate of θ interval) DDREF interval) inward curvature and high-dose leveling-off. It is not obvi- ous that the linear relationship resulting from such cancela- Radiobiology animal experiments 0.5 Sv (0.1, 3.2) 1.5 (1.0, 4.4) LSS data (0–1.5 Sv dose range) 0.3 Sv (–0.1, 1.5) 1.3 (0.8, 2.6) tion overestimates low-dose risk. Combined (posterior) 0.5 Sv ( 0.1, 1.2) 1.5 (1.1, 2.3) Given these unresolved issues, it is comforting that the estimate of LSS DDREF is consistent with the best-fitting NOTE: The 95% intervals are confidence intervals (likelihood ratio) in the LQ model based on LSS data alone; that is, low-dose risk first two rows and Bayesian posterior probability intervals in the last row. estimates based on LSS linear models with DDREF adjust- aFrom radiobiological animal experiment results and LSS data. ment will be essentially the same as risk estimates based on the best-fitting LQ model from LSS data over the range 0 to 1.5 Sv. In Figure 10-2, for example, it is clear that the linear component of an LQ curve with curvature 0.5 Sv–1 over the low-dose range is consistent with the data. The difference logical data and from LSS data are sensitive to the range of between that estimate and one based on the unadjusted linear doses used for estimation. model will be small relative to the size of the associated As shown in Annex 10B and evident in Figure 10B-1, confidence interval. there is a statistically significant difference in curvatures for The DDREF analysis has used LSS data on solid cancer the different mouse strains, sex, and cancer outcome combi- incidence. A recent similar analysis on cancer mortality nations investigated (p-value < .001). Some results indicate (Preston and others 2004) has provided the somewhat larger large curvature, some no curvature, and some curvature in estimated curvature 0.94 Sv–1 (90% CI 0.16, 8.4) for the best- the opposite direction. The combined effect is weak evidence fitting LQ model over the range 0 to 2 Sv. Since there is for small curvature. Because of the wide variability, the considerable imprecision in the calculations, this result is analysis is sensitive to the particular studies chosen and the not inconsistent with the committee’s conclusions. approach for estimating a curvature that is presumed to be In summary, the approach used by the committee to make constant across studies. an analytical judgment about the value of DDREF has em- The numerical results also are not robust because of the ployed a combined Bayesian analysis of dose-response cur- somewhat arbitrary choice in dose range subset for estimat- vature for cancer risk using animal radiobiological data and ing linear-quadratic models, both from animal experiments human evidence from the LSS. The committee found a be- and from LSS data. If the LQ model is fitted to a dose range lievable range of DDREF values for adjusting linear risk es- that includes doses at which leveling off of the dose-response timates from the LSS cohort to be 1.1–2.3. Based on this curve has occurred (as illustrated in Figure 10-1), the result analysis, the committee elected to use the value of 1.5 for will be biased for the intended purpose. If the dose range is solid tumors; also, a linear-quadratic model was used for too low, meaning it excludes doses for which the LQ ap- leukemia. The committee recognizes the limitations of the proximation is still good, the estimates will be less precise data and the uncertainties in estimating the DDREF. than what is possible but will not lead to any bias. Given these consequences, it is judged preferable to choose a cut- OTHER FORMS OF CELLULAR AND ANIMAL off dose that is too low rather than one that is too high. The RESPONSE TO RADIATION cutoffs of 1.5–2 Gy for animal experiments and 1.5 Sv for LSS data were chosen subjectively, based on the belief that This report has given much attention to biological re- these were sufficiently low that leveling off is not of great sponses to radiation that, although not well understood, may concern. Nevertheless, the fact remains that the relationships influence the development of views on tumorigenic mecha- shown in Figure 10-3 would be quite different if different nisms and the modeling of epidemiologic data. dose ranges were used. The cutoff of 1.5 Sv for the LSS data is important for Adaptive Responses assessing curvature. The resulting LSS DDREF is appropri- ate for adjusting risks from linear models based on the same Adaptive responses to radiation are represented in a range dose range. Since the LSS estimated linear model is insensi- of studies that purport to demonstrate that a low priming tive to the choice of subset however, the particular choice of dose of radiation can influence the subsequent response of dose range upon which to estimate the linear model is not cells or experimental animals to subsequent challenge by a critical. second higher dose. It is claimed by some that these adaptive

OCR for page 239
INTEGRATION OF BIOLOGY AND EPIDEMIOLOGY 251 responses will reduce low-dose cancer risk substantially, about the contribution of induced and persistent genomic perhaps to zero, or even be beneficial to health (see Calabrese instability to postirradiation tumor development, there is at and Baldwin 2003 and references therein). present no meaningful way in which the phenomenon can be Cellular data and mechanistic considerations on adaptive included in the general interpretation of epidemiologic data responses are reviewed in Chapter 2. From this review it is and, thereby, the derivation of new estimates of low-LET concluded that adaptive responses are not expressed robustly cancer risk. in cells and that a mechanistic basis for the phenomena, par- ticularly in the form of well-characterized DNA damage re- Bystander Cellular Effects sponse, has yet to be established. This situation may be con- trasted with the detailed knowledge that has accrued on many Chapter 2 details the almost exclusively cellular data for other aspects of DNA damage recognition or repair and cel- high-LET radiation, indicating that cellular damage response lular response (see Chapters 1 and 2). Accordingly, cellular signals may be passed from an irradiated cell to a non- and mechanistic data on adaptive responses are as yet insuf- irradiated neighbor. There are few consistent data sets for ficient to develop specific judgments on the fundamental low-dose, low-LET radiation. The stress-related mechanisms aspects of low-dose cancer risk. that have been suggested to underlie postirradiation signal Recent animal studies on adaptive responses to radiation transfer via cellular gap junctions or cell culture medium are and cancer risk are considered in Chapter 3. These studies not well understood. In addition, the in vivo expression of provide some evidence that under certain conditions, a low bystander effects and their impact on tumor development priming dose of radiation can modestly influence the rate of have yet to be adequately addressed. For these reasons, the development of certain tumors. However this response is not committee judges that current knowledge of these phenom- accompanied by a reduction of overall lifetime cancer risk. ena is insufficient for the purpose of interpreting epidemio- Uncertainties remain about the specific conditions of irradia- logic data and developing judgments on cancer risks at low tion under which this form of adaptive response is expressed, doses of low-LET radiation. and its mechanistic basis is a matter of speculation. Accord- ingly, these animal data, although of considerable scientific GENETIC SUSCEPTIBILITY TO CANCER interest, are not sufficiently well developed to influence the modeling and interpretation of epidemiologic data. The data reviewed in Chapters 1 and 3 provide coherent evidence from cellular, animal, and clinical or epidemiologic studies that inheritance of certain germline gene mutations Induced Genomic Instability can predispose to radiation-induced cancer. The qualitative Induced genomic instability is a term used to describe a linkage between such epidemiologic or clinical and experi- set of cellular phenomena whereby radiation exposure alters mental data are particularly strong for rare, strongly express- the state of a cell in a way that generally leads to a persistent ing human mutations. However, with current knowledge, elevation of mutation rate over many cell generations. The experimental data cannot quantitatively inform about the cellular data reviewed in Chapter 2 highlight the inconsis- magnitude of the increased radiation risk in such genetic dis- tent mode of expression of this phenomenon and the current orders. Accordingly, only broad judgments are possible— lack of information on the cellular mechanisms that might be principally that strongly expressing human mutations of rel- involved. It is notable that many of these data sets relate to evance to radiation cancer risk are too rare to an appreciably cells established in culture for many years. Despite these distort population-based estimates of low-dose risk as de- problems of interpretation, there has been speculation that rived from epidemiologic data (Chapter 3). radiation-induced genomic instability might make a signifi- The implication for population risk of weakly expressing cant contribution to cancer induction in vivo and thereby but potentially common variant genes is a most difficult is- confound the interpretation of epidemiologic data. Chapter 3 sue. Genetic studies with mice (Chapter 3) provide evidence considers the in vivo expression of radiation-induced ge- of the potential complexity of germline gene-gene interac- nomic instability, possible mechanistic links with cancer in- tions in radiation tumorigenesis. However, human molecular duction in animal models, and the expression of such insta- epidemiologic studies in this area are at a very early stage of bility in radiation-associated human tumors. Although some development, and no specific judgments are possible on the uncertainty remains, these in vivo data strongly question the extent to which common genetic variation influences epide- proposition that radiation-induced, genome-wide instability miologic measures of radiation risk. The general judgment plays a major role in radiation tumorigenesis. One possible made in Chapter 3 is that the potential impact of such variant exception to this is the instability of altered telomeric se- genes on radiation cancer risk in the population will depend quences at chromosome termini that may trigger ongoing on a complex interplay between their frequency in the popu- cycles of chromosomal associations and rearrangement lation, their tissue specificity, and the strengths of the gene- (Chapters 2 and 3). However, given the great uncertainty gene and gene-radiation interactions that may apply.

OCR for page 239
252 BEIR VII HERITABLE EFFECTS OF RADIATION adverse effects cannot be compared readily to what are for- mally called genetic diseases. As in the BEIR V report (NRC 1990), estimates of the The total numbers of children included in the analyses to risks of adverse heritable effects of radiation exposure are ascertain radiation effects were about 41,000 in the “unex- made indirectly through extrapolation from mouse data on posed” and 31,000 in the “exposed” groups, although the rates of radiation-induced germ cell mutations using popula- numbers were different for different indicators (e.g., ~8000 tion genetic theory and a set of plausible assumptions (see children each in the exposed and control groups for balanced Chapter 4). These estimates are expressed as increases in the structural chromosomal rearrangements and sex chromo- frequencies of genetic diseases relative to their baseline fre- somal aneuploidy; ~41,000 in the exposed and ~31,000 in quencies in the population. The method that is used for this the exposed groups for malignancies in F1). purpose is referred to as the “doubling dose method.” Equa- Although no statistically significant effects of parental tion (10-1) below summarizes the method: radiation exposures were found, Neel and colleagues (1990) estimated doubling doses on the basis of data for five of the Risk per unit dose = indicators (i.e., UPO, F1 mortality, F1 cancers, sex chromo- P × [1/DD] × MC × PRCF, (10-1) somal aneuploids, mutations) that would be consistent with the findings. In order to do this, several assumptions had to where P is the baseline frequency of the disease class under be made (discussed in Annex 4G). The oft-quoted DD esti- consideration, DD is the doubling dose (i.e., the dose of ra- mated from these data, corrected for low-dose or chronic, diation required to double the rate of spontaneous mutation low-LET radiation conditions is 3.4 to 4.5 Sv. in a generation, estimated as a ratio of rates of spontaneous The perception remains that the above estimate of the DD and induced mutations in defined genes), MC is the muta- is indicative of far lower heritable risk than that implied by tion component (a measure of the responsiveness of the dis- the DD of 1 Gy used by the present BEIR committee and ease class to an increase in mutation rate), and PRCF is the UNSCEAR (2001; since 1/DD, the relative mutation risk per potential recoverability correction factor (the fraction of in- unit dose, is a smaller fraction with the Japanese DD). It duced mutations that are compatible with live births and should be stressed that comparison of the DDs alone does cause disease). not present the correct picture of risks for the following rea- This report incorporates several important advances that sons: (1) the Japanese DDs are estimated retrospectively have been made since the publication of BEIR V (NRC from empirical observations using measures of genetic ill 1990), among which are those that relate to the four quanti- health that are totally different from those used by this ties mentioned above. It suffices to note that the estimates committee; besides, these measures have not shown any sig- for P, MC, and PRCF are different for Mendelian and multi- nificant differences between the control and radiated groups; factorial diseases; however, the DD estimate of 1 Gy (for and (2) the DD of 1 Gy used by the present committee (and low-dose or chronic low-LET exposure) is common to both by UNSCEAR 2001) is based on data on mutations in de- classes of disease. fined genes and is used prospectively as one of the four fac- The risk estimates provided in Chapter 4 are about 3000 tors in predicting the risk of genetic diseases. Nonetheless, to 4700 cases of excess genetic disease per million first-gen- the principal message that emerges from the Japanese epide- eration progeny per gray of radiation to the parental genera- miologic studies and the present risk estimates projected tion. Compared to the natural (i.e., baseline) risk of genetic from mouse data on radiation-induced mutations is the diseases of 738,000 per million live births in the population, same—namely, that at low or chronic doses of low-LET ir- the radiation risk (per gray) is very low (about 0.4 to 0.6% of radiation, the genetic risks are very small compared to the the baseline). baseline risk in the population. As mentioned earlier, the results of the extensive genetic epidemiologic studies of A-bomb survivors in Japan have shown no adverse effects in the progeny that could be attrib- SUMMARY uted to the radiation exposures (of the order of 0.4 Sv) sus- The principal objective of this chapter is to highlight the tained by most survivors. The indicators of adverse effects ways in which cellular, molecular, and animal data consid- used in these studies were untoward pregnancy outcomes ered in this report may be integrated with epidemiologic find- (UPOs), mortality of live born children through a period of ings in order to develop coherent judgments on the health about 26 years (exclusive of those resulting from malignant effects of low-LET radiation. Emphasis is placed on data tumors), malignancies in the F1 children, frequency of bal- integration for the purposes of modeling these health risks. anced structural rearrangements of chromosomes, frequency The principal conclusions from this work can be summa- of sex chromosomal aneuploids, frequency of mutations af- rized as follows: fecting protein charge or function, sex ratio shift among chil- dren of exposed mothers, and growth and development of F1 • Current knowledge of the cellular or molecular mecha- children. The important point here is that these indicators of nisms of radiation tumorigenesis tends to support the appli-

OCR for page 239
INTEGRATION OF BIOLOGY AND EPIDEMIOLOGY 253 t t cation of models that incorporate the ERR projection over h(t ) = µ(t )∫ {v( s) X ( s) exp ∫ [α(u) − β(u)]du}ds , 0 s time. • The choice of models for the transport of cancer risk where µ(t) and ν(t) denote, respectively, the first and second from Japanese A-bomb survivors to the U.S. population is mutation rates at time t; α(t) is the rate of division of inter- influenced by mechanistic knowledge and information on mediate (or initiated) cells; and β(t) is the rate of death or the etiology of different cancer types. Where specific epide- differentiation of intermediate cells at time t. miologic evidence is lacking, the committee recommends If ionizing radiation and the other main risk factors for that the weights attaching to relative and absolute risk trans- the tumor of interest are predominantly cancer initiators, port should be 0.7 and 0.3, respectively. their effect would be modeled additively on the first muta- • A combined Bayesian analysis of A-bomb epidemio- tion rate µ, as follows: logic information and experimental data has been employed to provide an estimate of the DDREF for cancer risk. The µ(t, dose | risk factor) = µ(t) + γ dose + ε risk factor. committee found a believable range of DDREF values to be 1.1 to 2.3 and uses a median value of 1.5 to estimate solid The resulting relative risk would then be of the form: cancer risks. • Knowledge of adaptive responses that may act to re- RR(t, dose | risk factor) duce radiation cancer risk was judged to be insufficient to be = h(t, dose | risk factor) / h(t | risk factor) incorporated in a meaningful way into the modeling of epi- = [µ(t) + γ dose + ε risk factor] / [µ(t) + ε risk factor], demiologic data. The same judgment is made in respect of the possible contribution to cancer risk of postirradiation while the absolute risk (AR) would be of the form: genomic instability and bystander signaling effects. • Genetic variation in the population is a potentially im- AR (dose | risk factor) = h(t, dose | risk factor) – h(t) portant factor in the estimation of radiation cancer risk. t Strongly expressing cancer-predisposing mutations are = [µ(t ) + γ dose + ε risk factor] ∫ {v(s) X (s) exp 0 judged from modeling studies to be too rare to distort popu- t lation-based estimates of risk appreciably but are a signifi- ∫ [α(u) − β(u)]du}ds − [µ(t ) + ε risk factor] s cant issue in some medical irradiation settings. The position t t regarding potentially more common variant genes that ex- ∫ {v(s) X (s) exp ∫ [α(u) − β(u)]du}ds 0 s press only weakly remains uncertain. t t • The estimation of the heritable effects of radiation by ∫0 ∫ = γ dose {v( s) X ( s) exp [α(u) − β(u)]du}ds 0 the committee takes advantage of new information on hu- = γ dose h(t ) / µ(t ) man genetic disease and on mechanisms of radiation-induced = δ(t ) dose h(t ). germline mutations. The application of a new approach to genetic risk estimation leads the committee to conclude that According to this formulation, the effect of radiation would low-dose induced genetic risks are very small compared to tend to be independent of the other risk factors on the AR baseline risks in the population. scale. The AR per sievert could therefore be transported from • The committee judges that the balance of evidence from one population to another. epidemiologic, animal, and mechanistic studies tends to fa- vor a simple proportionate relationship at low doses between radiation dose and cancer risk. Uncertainties in this judg- ment are recognized and noted. ANNEX 10A: APPLICATION OF THE MOOLGAVKAR AND KNUDSON TWO-STAGE CLONAL EXPANSION MODEL TO THE TRANSPORT OF RADIATION CANCER RISK In the case of tumors whose background incidence is strongly influenced by initiating factors, one would expect the EAR to be directly transportable from one population to another. If one considers, for example, the Moolgavkar and Knudson two-stage clonal expansion model (Moolgavkar and Knudson 1981; Moolgavkar and Luebeck 1990) shown in Figure 10A-1, the hazard function h(t) at time t is given FIGURE 10A-1 The two-stage clonal expansion model. SOURCE: approximately by the following formula: Luebeck and others (1999).

OCR for page 239
254 BEIR VII If, on the contrary, the background incidence of a given should be adjusted to correspond to the linear component of tumor type is heavily influenced by host or environmental the estimated LQ model, which is exactly what the DDREF promoting factors (e.g., breast cancer, stomach cancer), the presented in this chapter is designed to do. effects of these factors can be thought to affect the expan- Figure 10B-1 shows data from mouse experiments that sion (increasing α to a value αr, decreasing β to βr, or both) fitted to the model above (data from Table 6 of Edwards of the clone of initiated or transformed cells and, thus, the 1992). These data show that the slope in the linear dose- expression of tumors. The resulting relative risk would then response for chronic exposure approximates the linear com- be of the form: ponent of the LQ model for acute exposure. RR(t, dose | risk factor) Details of DDREF Estimation = h(t, dose | risk factor / h(t | risk factor) t t An LQ model for ERR or EAR, with curvature con- (µ(t ) + γ dose)∫ {v( s) X ( s) exp ∫ [α(u) − β(u)]du}ds strained to be θ, may be written as αLQ[Dose + θDose2]. 0 s = t t A Bayesian statistical analysis is used to update information µ(t )∫ {v( s) X ( s) exp ∫ [α(u) − β(u)]du}ds 0 s about dose-response curvature from animal carcinogenesis = (1 + γ / µ(t ) dose). studies with the information concerning curvature from the LSS cohort of Japanese A-bomb survivors (over the dose This formulation is independent of the magnitude of the ef- range 0–1.5 Sv). A posterior distribution for LSS DDREF fect of promoting factors on the cell division and mortality follows directly from this, via its definition as a function of rates αr and βr. Hence the ERR can be exported directly from θ. The LSS DDREF is essentially 1 + θ for the 0–1.5 Sv dose one population to another. range and for values of θ of interest here. Pierce and Vaeth Expressed in simple terms, low-LET radiation (viewed (1991) provide a more detailed discussion of this relation- here as a tumor initiator) will tend to act additively with ship over different dose ranges. other tumor initiators and multiplicatively with tumor pro- Two forms of animal experiment data were used to esti- moters. Thus, in the case of a radiogenic tumor type with a mate curvature: estimated cancer risks and mean survival strong influence of promoters (e.g., stomach cancer), one times (referred to as life-shortening data). These are two dif- would favor an RR transportation model, while in the case of ferent summarizing results from the same experiments, so a tumor with strong influence of initiators, one would favor they are not independent but address the curvature in differ- an AR transportation model. ent ways. LQ models for risk as a function of dose can be The preceding formulations are consistent with more gen- estimated for each separate cancer and combined to form a eral analyses of the nature of risk relationships involving single estimate of curvature, θ. On the other hand, the life- exposure to two carcinogens (Kodell and others 1991; shortening studies ignore cause of death and therefore repre- Zielinski and others 2001). sent a cumulative effect of all radiation-induced deaths, the majority of which are cancer related. By using the relation- ship between survival rate and risk, the curvature of interest ANNEX 10B: EVIDENCE FOR THE CONNECTION can be estimated from these, as detailed below. BETWEEN DOSE EFFECTS AND DOSE-RATE EFFECTS The estimated risks of relevant cancers, plotted versus IN ANIMAL EXPERIMENTS radiation dose in Figure 10B-2, were extracted from the sum- First consider fractionated acute exposures. If the relative mary tables of Edwards (1992), but exclude (1) the results in risk due to the sum of K acute exposures of equal dose, D / K, Tables 1 and 2 because those risk estimates were not ad- administered at separate times, is the sum of the individual justed for competing causes of death; (2) results for doses relative risks, and if an LQ dose-response model describes greater than 2 Gy; and (3) results on lymphomas, ovarian the effects at each fraction, then the total relative risk due to cancer, reticulum cell carcinoma, and nonmyeloid leuke- all K exposures is mias, because these are thought to arise via atypical biologi- cal mechanisms, as discussed in Chapter 3, or to reflect an RRTotal = K {α(D / K) + β(D / K)2] = αD + βD2 / K. ill-defined combination of cancer types. The risks presented here are based on acute exposures only. Thus, for a given total dose D, the importance of the qua- There is substantial evidence that the curvature, θ, is not dratic term diminishes with increasing number of fractions the same in all 11 situations (p-value < .0001, from a likeli- of exposure. The RR due to a protracted exposure may be hood ratio test). Despite this evidence, the model with com- thought of, at least approximately, as the limit as K ap- mon curvature explains 97% of the variability and the model proaches infinity. In this way, the total RR due to a protracted with different curvatures explains 98% of the variability in exposure is simply αD, where α is the linear coefficient in estimated risks, so the practical significance of the different the LQ model. Therefore, if a risk estimate corresponding to curvatures may not be too important. Note in Figure 10B-2 a protracted exposure D is based on an LSS linear model, it that although the LQ curves seem to be highly divergent, the

OCR for page 239
INTEGRATION OF BIOLOGY AND EPIDEMIOLOGY 255 Risks of Lung Adenocarinoma; BALB-c Female Mice 1 1 acute exposure 1 2 acute exposures 35 4 acute exposures 20 acute exposures 2 chronic exposure Risk of Lung Adenocarcinoma (%) 30 2 4 25 1 C 20 20 1 1 C 15 1 1 C 1 1 1 0.0 0.5 1.0 1.5 2.0 Cumulative Dose (Gy) FIGURE 10B-1 Risks of lung cancer versus dose from experiments in which doses were administered fractionally or chronically. Each plotted point corresponds to an estimated risk from one experiment. The plotting symbol shows the number of fractions (number of separate acute exposures) or “C” if the administration was chronic. The curves show an estimated LQ model for risk from dose D administered in K fractions, αD + βD2 / K, for five different values of K (where K is taken to be infinity for chronic exposure). dotted lines tend to intersect most of the error bars—the evi- at various doses (Storer and others 1979). Indications of a dence for different curvatures is not as extreme as it might dose-rate effect from these data stem from the observation appear from simple visual inspection. that the mean survival times are longer for chronically By acting as if there is a single value of θ, the evidence exposed mice than for acutely exposed mice given the same for it is summarized by the likelihood function labeled “Ani- total dose. However, to extract specific information about mal Experiments” in Figure 10B-4. This is a scaled profile curvature, it is necessary to understand the connection be- likelihood function for θ from a model in which the risk tween age-specific failure rate and survival time. The human estimates of Figure 10B-2 are normally distributed with vari- risk models estimated with the LSS data are for age-specific ances that are proportional to the reciprocal of the squared failure rates, also known as hazard functions. Interest here standard errors. The means are modeled to depend on the therefore concerns LQ models for hazard functions. Data on particular condition—corresponding to each of the 11 graphs mouse survival times may, in principle, be used directly to in Figure 10B-2—with linear-in-dose coefficients that estimate the hazard function, by employing standard statisti- depend on the particular condition and with a quadratic cal tools of survival analysis, but the unavailability of the coefficient that is θ divided by the linear coefficient. Thus, raw data precluded this approach by the committee. If the the different conditions have different linear and quadratic survival times are assumed to follow an exponential prob- terms, but the ratio of the quadratic to linear term is held ability distribution, then the hazard function is the reciprocal constant. of the mean survival time. By using this exponential assump- The “life-shortening” data used here are the mean sur- tion (which is probably incorrect but useful nonetheless for vival times of mice exposed acutely and chronically to γ-rays extracting information about curvature, at least roughly), the

OCR for page 239
256 BEIR VII Myeloid Leukemia; Female RFM Mice Harderian Gland; Female RFM Mice Pituitary; Female RFM Mice 9 10 15 3.0 Risk (%) Risk (%) Risk (%) 10 8 2.0 7 5 6 1.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 Dose (Gy) Dose (Gy) Dose (Gy) Lung; Female RFM Mice Uterine; Female RFM Mice Mammary; Female RFM Mice 3.5 3.5 Risk (%) Risk (%) 28 Risk (%) 2.5 2.5 24 1.5 1.5 20 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 Dose (Gy) Dose (Gy) Dose (Gy) Myeloid Leukemia; Male RFM Mice Lung; Male RFM Mice Harderian Gland; Male RFM Mice 4 30 10 Risk (%) Risk (%) Risk (%) 8 3 6 26 2 4 2 22 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 Dose (Gy) Dose (Gy) Dose (Gy) Mammary; Female Balb-c Mice Lung; Female Balb-c Mice 20 35 Risk (%) Risk (%) 16 25 12 15 8 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 Dose (Gy) Dose (Gy) FIGURE 10B-2 Estimated risk of cancer versus radiation dose from various mouse experiments. SOURCE: Data from A.A. Edwards (1992) for cancer site, mouse strain, and sex combinations. Vertical bars extend two standard errors above and below each estimate. Solid curves are estimated LQ models based on each condition individually. Dotted curves are the best-fitting LQ models when curvature is constrained to be the same for all 11 conditions. curvature of interest can be ascertained by fitting an LQ dependent data sets, these data sets are not independent. In- model to the reciprocal of the mean survival. stead, an average of the two is obtained, shown as the solid Figure 10B-3 shows the reciprocal mean survival times curve in Figure 10B-4, to represent an average effect based plotted versus dose, with different plotting characters for on the two ways of dealing with the data. The maximum means based on acutely and chronically exposed mice. Also likelihood estimate of θ from the average likelihood is 0.5, shown on the plot are the fits to the model that has the age- corresponding to an LSS DDREF of 1.5. specific death rate equal to a constant plus αDose for chroni- Evidence of curvature at the cellular level comes prima- cally exposed mice and the same constant plus α(Dose + rily from studies of chromosomal aberrations in human cells. θDose2) for acutely exposed mice, following the reasoning Table 10B-1 shows estimated LQ models for the regression in the first section of this Annex. (The estimates are maxi- of chromosome aberration induction on dose. These results mum likelihood estimates based on normality of the recipro- may be included weakly, by specifying a probability distri- cal means, which are estimates from a large number of mice.) bution with mean and variance equal to the sample mean and The estimates depend highly on the dose range considered. sample variance of the three curvatures in the table. The re- This presents a difficulty since leveling off of the dose-re- sult of including such a distribution in the averaging of Fig- sponse is expected (as shown in Figure 10-1), but the dose at ure 10B-4 is to increase the width of the resulting average which leveling off occurs is difficult to determine, both theo- likelihood, with little effect on the center of the distribution. retically and empirically. The decision was made by the com- Since they do not alter the results and because of the extra mittee to use the 0–1.5 Gy dose range, but this is subjective theoretical demand in incorporating cellular data into mod- and open to debate. els for human cancer rates, chromosome aberration data were The (profile) likelihood function for θ is shown in Figure not included in the analysis. 10B-4. It is evident that the life-shortening data indicate The final step in the LSS DDREF estimation involves slightly more curvature than the direct cancer risk results. combining animal radiobiological information with LSS in- While it is appropriate to multiply two likelihoods from in- formation about curvature. The average likelihood in Figure

OCR for page 239
INTEGRATION OF BIOLOGY AND EPIDEMIOLOGY 257 A Acute exposure: A Chronic exposure: C 0.0024 A Reciprocal of Mean Life Length in days 0.0022 A A 0.0020 C A A 0.0018 A C C A C 0.0016 AA 0 1 2 3 4 Dose (Gy) FIGURE 10B-3 Life-shortening data from Tables 1, 2, and 3 of Storer and others (1979). Plotted are the reciprocals of the mean life lengths of RFM female mice versus the dose of exposure, with different plotting symbols for acute (A) and chronically (C) exposed groups. The curves are the result of estimation of an LQ model for age-specific mortality rate fit to the 0–1.5 Gy dose range. 10B-4 is used as a “Bayesian prior distribution” in a Baye- graph indicates the likely values of curvature and corre- sian analysis. The resulting “posterior distribution” that re- sponding values of the LSS DDREF. sults when this prior is multiplied by the LSS likelihood func- It should be evident that further study beyond that accom- tion is shown in Figure 10-3. The posterior density in the plished here could possibly lead to a better summarization of radiobiological information about curvature than provided in Figure 10B-4. Even a more thorough study, however, TABLE 10B-1 Estimates of Linear and Quadratic would similarly be obstructed by the subjectivity involved in Coefficients from Chromosome Aberration Induction the choice of dose range upon which LQ models are fit, by Studies and the Implicit Curvatures the inconsistency of animal experiment results, and by the difficulty in translating mouse results to human cancer rates. Human Cell Type Radiation α β θ The posterior density can be used directly to describe the uncertainty in LSS DDREF for the uncertainty analysis in Lymphocytesa Cobalt-60 0.015 0.06 4.0 Chapter 12. This distribution probably understates the un- Lymphocytesa 250 kV X-rays 0.04 0.06 1.5 certainty in knowledge of LSS DDREF, however, because Primary human fibroblastsb Cesium-137 (acute) 0.059 0.029 0.5 of the various subjective choices involved. In an attempt to Primary human fibroblastsb Cesium-137 (chronic) 0.019 be more realistic about the state of knowledge of LSS aFrom Lloyd and others (1992). DDREF, an inflated variance (of the distribution of the loga- bFrom Cornforth and others (2002). rithm of LSS DDREF) is used in the uncertainty analysis.

OCR for page 239
258 BEIR VII FIGURE 10B-4 A summary of radiobiological evidence for curvature: the (profile) likelihood for curvature from the risk estimate data in Figure 10B-2, the (profile) likelihood from the life shortening data in Figure 10B-3, and their average.