Contents, preface, executive summary, chapters 1 and 2

Following its meeting in Oxford, UK in 1997, Committee 1 of the International Commission on Radiological Protection proposed a Task Group to prepare a report on low-dose extrapolation of radiation-related cancer risk estimates based largely on higher-dose epidemiological data, and the possible implications for radiological protection. The Commission accepted this recommendation and established a Task Group, which began its work in April 1998.

The membership of the Task Group was as follows:

Corresponding members were:

R. Cox, J.H. Hendry, C.R. Muirhead, and R. J. Preston of Committee 1 contributed additional text to the report.

The membership of Committee 1 during the period of preparation of this report was:

(a) The present report considers the evidence relating to cancer risk associated with exposure to low doses of low linear energy transfer (LET) radiation, and particularly doses below current recommended limits for protection of radiation workers and the general public. The focus is on evidence regarding linearity of the dose–response relationship for all cancers considered as a group, but not necessarily individually, at low doses [the so-called linear, non-threshold (LNT) theory], and the possibility of a universal threshold dose below which there is no risk of radiation-related cancer. According to the LNT theory, the same number of radiation-related cancers would be predicted in a population of a given size exposed to a certain small average radiation dose and in an otherwise similar population many times times larger and exposed to a proportionally smaller average dose. According to the threshold theory, the radiation-related risk in the larger population would be zero if its average dose was sufficiently small.

(b) The present document has been preceded by other recent reports, notably those of the United Nations Scientific Committee on the Effects of Atomic Radiation (UNSCEAR, 1993, UNSCEAR, 2000) and the US National Council of Radiation Protection and Measurements (NCRP, 2001). These reports recommended that radiation protection should continue to be guided by the LNT theory. The Task Group concurs with those recommendations.

(c) This report is organised by scientific discipline, beginning with epidemiological studies of exposed human populations (Chapter 2). Epidemiological studies offer the most directly relevant information for risk-based radiation protection. The major scientific issues, as illustrated by the example of cancer incidence from all solid tumours combined in the Life Span Study population of atomic bomb survivors, are: (1) establishment of the existence of a dose-related risk in this population; (2) modelling radiation-related risk as a statistically uncertain parametric function of dose, modified by other factors such as sex, exposure age, attained age, and time following exposure; (3) extrapolation of estimated risk to other potentially exposed populations, with possible different baseline cancer rates; (4) projection of the risk in the population to the end of its natural life; and (5) extrapolation of risk estimates from moderate-to-high dose levels of acute exposure, characteristic of the most informative atomic bomb survivor data, to the far more common low-dose and/or protracted exposures that occur in occupational and general settings. Consideration of each of these issues leads to more refined risk estimates; however, because information about each is uncertain, the overall uncertainty of the improved estimates is increased. There is limited evidence of increased cancer risk associated with acute exposures of the order of a few tens of mGy, and this will be discussed in the report. However, firm epidemiological evidence of radiation cancer risk comes from studies that involve exposures of >100 mGy. Other evidence may be used to place an upper limit on the value of any universal threshold that may exist. Also, the risk of mortality and morbidity from all solid cancers combined is proportional to radiation doses down to approximately 100–150 mGy, below which statistical variation in baseline risk, and small and uncontrollable biases, tend to obscure evidence concerning radiation-related risk. Extrapolation of risk estimates based on observations at moderate-to-high doses continues to be the primary basis for estimation of radiation-related risk at low doses and dose rates.

(d) The fundamental role of radiation-induced DNA damage in the induction of mutations and chromosome aberrations, and the apparent critical involvement of aberrations and mutations in the pathogenesis of cancer provides a framework for the analysis of risks at low-dose and low-dose-rate exposures (Chapter 3). A characteristic type of damage produced by ionising radiation (IR) involves multiple lesions within close spatial proximity. Such clustered damage can be induced even by a single radiation track through a cell. Although cells have a vast array of damage response mechanisms that facilitate the repair of DNA damage and the removal of damaged cells, these mechanisms are not foolproof, and emerging evidence suggests that closely spaced lesions can compromise the repair machinery. Also, while many of the cells containing such radiation-induced damage may be eliminated by damage response pathways involving cell-cycle checkpoint control and apoptotic pathways, it is clear from analysis of cytogenetics and mutagenesis that damaged or altered cells are capable of escaping these pathways and propagating.

(e) Cellular consequences of radiation-induced damage (Chapter 4) include chromosome aberrations and somatic cell mutations. The processing and misrepair of radiation-induced double-strand breaks, particularly complex forms, are responsible for chromosome/gene alterations that manifest as chromosome aberrations and mutations. Current understanding of mechanisms and quantitative data on dose and time–dose relationships support a linear dose–response relationship at low doses (i.e. LNT). Considered as a whole, the emerging results with regard to radiation-related adaptive responses, genomic instability, and bystander effects suggest that the risk of low-level exposure to IR is uncertain, and a simple extrapolation from high-dose effects may not be wholly justified in all instances. However, a better understanding of the mechanisms for these phenomena, the extent to which they are active in vivo, and how they are inter-related is needed before they can be evaluated as factors to be included in the estimation of potential risk to the human population of exposure to low levels of IR. In addition, although there are intrinsic uncertainties at low doses and low dose rates, direct epidemiological measures of radiation cancer risk necessarily reflect all mechanistic contributions, including those from induced genomic instability, bystander effects, and, in some cases, adaptive responses, and therefore may provide insights about these contributions.

(f) Experimental approaches using animal models (Chapter 5) are well suited to precise control of radiation dose and dose rate, as well as genetic background and other possible modifiers of the dose–response relationship, and can facilitate precise determination of biological outcomes. Recent studies using newly developed animal models; cellular, cytogenetic and molecular data for acute myelogenous leukaemia (AML), intestinal tumours, and mammary tumours; and cytogenetic and molecular studies on the induction of AML and mammary cancer support the view that the essential radiation-associated events in the tumourigenic process are predominantly early events involving DNA losses targeting specific genomic regions harbouring critical genes. As such, the response for early initiating events is likely to correspond to that for the induction of cytogenetic damage. On this basis, mechanistic arguments support a linear response in the low-dose region, i.e. the process should be independent of dose rate because interactions between different electron tracks should be rare. Quantitative analyses of dose–response relationships for tumourigenesis and for life shortening in laboratory animals also support this prediction. These studies also support a dose and dose-rate effectiveness factor (DDREF) for reduction of estimated risk per unit dose based on acute, high-dose data in the range of about 2 when data are extrapolated to low doses from effects induced by doses in the range of 2–3 Gy. Extrapolation of results from less than 1 Gy would result in lower DDREF values.

(g) Chapter 6 presents a formal exercise in quantitative uncertainty analysis, in which the different uncertain components (as identified in Chapter 2) of estimated cancer risk associated with low-dose, low-LET radiation exposure to a non-Japanese population, in this case that represented by the US National Cancer Institute’s SEER (Surveillance Epidemology and End Results) registry, are combined. Attention is paid to the resulting uncertainty distribution for excess relative risk (ERR) per Gy, with and without allowing for the uncertain possibility of a universal low-dose threshold below which there would be no radiation-related risk. In the example that involves risk from all cancers combined including leukaemia, except for non-melanoma skin cancer, the major sources of uncertainty are statistical variation in the estimated ERR at 1 Gy for the atomic bomb survivors, subjective uncertainty (informed by experimental and epidemiological data) about the DDREF to be applied at low doses and dose rates, and the postulated uncertainty concerning the existence of a universal threshold at some dose above that for which the calculation was being made. Unless the existence of a threshold was assumed to be virtually certain, the effect of introducing the uncertain possibility of a threshold was equivalent to that of an uncertain increase in the value of DDREF, i.e. merely a variation on the result obtained by ignoring the possibility of a threshold.

(h) The conclusions of this report are given in Chapter 7. While existence of a low-dose threshold does not seem unlikely for radiation-related cancers of certain tissues, and cannot be ruled out for all cancers as a group, the evidence as a whole does not favour the existence of a universal threshold, and there seems to be no particular reason to factor the possibility of a threshold into risk calculations for purposes of radiation protection. The LNT theory, combined with an uncertain DDREF for extrapolation of risk from high doses, remains a prudent basis for radiation protection at low doses and low dose rates.

Back to Article Outline

1. Introduction 

(1) The purpose of the present report is to summarise scientific evidence relevant to the quantification of cancer risk associated with radiation exposure at (effective) doses of interest for radiation protection, particularly doses below current recommended limits for protection of radiation workers (e.g. 20 mSv/year) and the general public (e.g. 1 mSv/year). As a rough rule of thumb, effective doses of the order of 1 Sv, 100 mSv, 10 mSv, 1 mSv, and 0.1 mSv may be called ‘moderately high’, ‘moderate’, ‘low’, ‘very low’, and ‘extremely low’, respectively. However, in common usage, and in this report in particular, ‘low’ and ‘high’ are usually relative terms, i.e. shorthand for ‘relatively low’ and ‘relatively high’, which may refer to ranges of different numerical values depending on the context.

(2) Ionising radiation (IR) exposure is an established cancer risk factor. Compared with other common environmental carcinogens, it is relatively easy to determine organ-specific radiation dose and, as a result, radiation dose–response relationships tend to be highly quantified. Nevertheless, there can be considerable uncertainty about questions of radiation-related cancer risk as they apply to risk protection and public policy, and the interpretations of interested parties can differ radically. A major reason for disagreement is that public and regulatory concern is often focused on exposures at radiation doses far lower than those at which useful information about cancer risk can be obtained directly, i.e. than can be obtained by studying populations with such exposures. Thus, risk estimates promulgated by expert committees, for example, are usually based upon epidemiological dose–response data obtained at doses ranging up to 0.2 Gy, 0.5 Gy, 1 Gy, or higher, and the resulting estimates are then extrapolated, with appropriate caveats, to lower doses. The extrapolation rules are based, in part, upon epidemiological observations, such as the degree of curvature of fitted linear-quadratic dose–response models for leukaemia and solid cancer morbidity among atomic bomb survivors, and on models derived from experimental systems.

(3) The discussion in the present report is concerned ultimately with the biological effects of IRs of low linear energy transfer (low LET), such as photons (gamma rays and x rays) and electrons (beta particles) of various energies, as contrasted with high-LET radiations such as neutrons and alpha particles. However, some biological effects that have been observed mainly in connection with high-LET exposure are clearly relevant to questions of cancer risk at low levels of low-LET radiation.

(4) Currently, the ICRP radiation protection philosophy is based on the so-called linear, non-threshold (LNT) theory. According to this theory, total radiation-related cancer risk is proportional to dose at low and moderately low doses (of the order of 200 mGy or less) and dose rates (less than 6 mGy/h averaged over the first few hours) (EPA, 1999, UNSCEAR, 1993). The theory is not universally accepted as biological truth. However, because it is not actually known what level of risk is associated with very-low-dose exposure, this theory is considered by many to be a prudent rule of thumb for public policy aimed at avoiding risk from unnecessary exposure.

(5) A logical conclusion from the LNT theory is that at a sufficiently low dose D and sufficiently large population size N, exposure of N people to average dose D would result in the same number of radiation-related cancers as exposure of k×N people to average dose D/k, for arbitrary k>1. This logical consequence has been used to justify the concept of ‘collective dose’, that the product of average dose and the number of people exposed is proportional to the number of radiation-related cancers. The concept of collective dose is sometimes used to support a moral argument against widespread use of technologies or practices that would, according to the LNT theory, involve individual exposures at doses so low that any associated risk, from the standpoint of the individual, would be far smaller than other risks that are casually taken in everyday life. A so-called threshold theory, according to which there is no radiation-related risk associated with exposures at doses below some universal threshold dose, would obviate concern about exposures at doses below the threshold and, specifically, arguments based on the concept of collective dose. Aside from collective dose, however, it is worth emphasising that the practical importance of the LNT vs threshold question is associated with doses at which the associated risks, if they exist, are high enough to be of ‘legitimate’ concern, as determined by the usual social and political processes.

(6) Historically, the LNT vs threshold controversy has been associated with public policy issues related to exposures that are widespread but (typically) low for individuals, such as local and worldwide exposure to radioactive fallout from aboveground nuclear test explosions carried out by different governments, mainly during the 1950s (Caron, 2004, Lewis, 1957, Lewis, 1963). The threshold theory, as applied to IR and to fallout exposure in particular, drew some of its legitimacy from the field of chemical toxicology, where thresholds are the rule (Brues, 1958, Brues, 1960), whereas the LNT theory is more consistent with findings from experimental radiation mutagenesis. As described by Caron (2004), the intellectual positions taken by proponents of the opposing sides during the fallout controversy of the 1950s (no compelling evidence of increased cancer risk at low radiation doses vs no compelling evidence against a radiation-related increase in cancer risk) are very similar to the situation at the present time. Some differences discussed in this report include the present general acceptance of a mutational basis for carcinogenesis, and evidence that radiation-related mutations tend to be more complex than more common mutations associated with endogenous and other causes.

(7) The present report has been preceded by other surveys of the biological and epidemiological information that underlies our understanding of low-dose risk and its estimation by extrapolation from data obtained at higher doses, notably and recently the comprehensive reports of the United Nations Scientific Committee on the Effects of Atomic Radiation (UNSCEAR, 2000, UNSCEAR, 2000) and the US National Council of Radiation Protection and Measurements (NCRP, 2001). The existence of these reports has allowed the present ICRP Task Group to be somewhat less comprehensive in its coverage of the field than may otherwise have been necessary, and to concentrate on updated coverage of developments in areas of epidemiology, fundamental biology, experimental radiation mutagenesis and carcinogenesis, and uncertainty analysis.

(8) Studies of cancer risk following exposure of human populations are the most obvious sources of information applicable to radiation protection policy. However, as discussed in Chapter 2, generalisation of risk information obtained from one exposed population to other populations with different characteristics and potentially exposed to radiation from different sources, at different doses and dose rates, requires the use of dose–response models to describe the behaviour of risk as a function of radiation dose, as well as possible modification of the dose–response relationship by individual and environmental factors. It also requires making assumptions that are often based on uncertain information.

(9) Chapter 3 deals with events believed to be fundamental to radiation carcinogenesis: radiation-induced DNA damage and its repair. In particular, Chapter 3 discusses the nature of radiation-induced damage and damage response pathways including repair of DNA double-strand breaks (DSBs), cell-cycle checkpoint control, early sensors of DNA damage, and signal transduction after irradiation. Questions of particular relevance for the current investigation are comparability of molecular damage from radiation exposure and endogenous causes, and comparability between radiation-related damage from IR at high vs low doses and dose rates with respect to mechanisms, pathways, and fidelity of repair.

(10) Cellular consequences of radiation-induced damage are discussed in Chapter 4. Rates of radiation-induced chromosome aberrations and somatic cell mutations were among the earliest quantitative measures of the cellular effects of IR, and studies of these outcomes have been highly informative about the dose–response relationship over a wide range of doses, and about effects of dose rate and fractionation. Induction of bystander effects in cells not directly irradiated, genomic instability in the progeny of irradiated cells, and adaptive responses are radiation-related phenomena that evoke questions about the generality of inferences based on cellular studies.

(11) Considerations of statistical power, and possible bias due to unobservable and uncontrollable confounders, govern the extent to which useful epidemiological information can be obtained at exposure levels of regulatory interest, and some degree of extrapolation is unavoidable. Experimental approaches using animal models, discussed in Chapter 5, offer considerably more control of radiation exposure and dose, genetic background, and modifying factors including other exposures, and can facilitate very precise determination of biological outcomes. On the other hand, analogies between radiation-related risks in human beings and inbred strains of experimental animals are necessarily limited. Low statistical power for low-dose studies is problematic for experimental and epidemiological studies alike, but indirect approaches, based on protraction and fractionation of exposure resulting in moderate to high cumulative doses, offer insights into low-dose effects. Experimental studies can, of course, be replicated to provide a firmer basis for insights into mechanisms, tissue-modifying factors, and quantitative dose–response relationships.

(12) Chapters 2–5 highlight statistical variations inherent in estimates obtained by fitting parametric models to epidemiological and experimental data, but also more fundamental uncertainties about important factors that cannot be ignored, but about which there may only be limited information. The implications of these uncertainties for conventional estimates of radiation-related cancer risk, especially at low doses and/or low dose rates characteristic of exposures most commonly encountered by radiation workers and the general public, are investigated in Chapter 6. The approach taken is an exercise in quantitative uncertainty analysis similar to approaches used in a number of recent exercises by expert committees concerned with such risks. Central to the approach is recognition of the fact that radiation protection is a political process, responsive to the interests and perceptions of stakeholders with differing points of view, and relying upon a knowledge base that is extensive but also uncertain. Acceptance of this fact implies that it is important, for the benefit and information of participants and stakeholders in the radiation protection process, to identify sources of uncertainty and to quantify the implications of such uncertainty for estimated risk. Among the questions addressed is the impact on radiation protection policy of treating the existence of a universal low-dose threshold for radiation-related cancer risk as an uncertain possibility.

Back to Article Outline

2. Epidemiological considerations 

2.1. Introduction 

(13) As for other areas of epidemiological research, the study of radiation-related cancer risk began with clinical observations, the earliest of which may have been the 16^th century identification by the physician Georg Bauer (more often known by his Latinised name, ‘Agricola’) of a specific condition, which he called ‘Joachimsthal Mountain Disease’, among miners in the Joachimsthal region of the present-day Czech Republic. The disease, the description of which now appears consistent with radon-related lung cancer but which could also include other lung diseases such as silicosis (NAS/NRC, 1999), was thought by Agricola to be caused by ‘metallic vapours’ in mine atmospheres. Roentgen’s discovery of x rays in 1895, Becquerel’s discovery of natural radioactivity the following year, and the subsequent use of both in science, medicine, and industry led to the recognition, documented by case reports early in its history, that radiation exposure may be harmful (Doll, 1995). The Court Brown and Doll study (1958) of mortality among British radiologists (Berrington et al., 2001; Smith and Doll, 1981), which demonstrated a significantly increased risk of cancer mortality among radiologists who had registered with a radiological society before 1921 and who were therefore likely to have received higher doses than radiologists who began their practice later, is an example of an influential study in which the fact of exposure was related to risk but individual dose estimates were not available. However, experimental studies of radiation effects such as cell inactivation, mutation, and carcinogenesis have taken advantage of the experimenters’ ability to regulate, with precision, radiation dose to target cells or tissues. Similarly, epidemiological investigations of exposed populations have benefited enormously from information enabling scientists to reconstruct individual, and even organ-specific, radiation doses. Benefits include the estimation of dose–response relationships and of the modification of such relationships by individual properties such as sex, age, lifestyle, and genetic inheritance. Thus, dose reconstruction is a fundamental component of the epidemiology of radiation carcinogenesis, and tends to be well worth the often considerable effort and expense required.

(14) ‘Risk’ is a concept in common use that is often applied to the past and future experiences of individuals, but a numerical risk value can be estimated and verified only on the basis of population rates, e.g. by comparing cancer rates in a population exposed to a given radiation dose with rates in an otherwise comparable population that is either not exposed or exposed to a much lower radiation dose. Thus, when we speak of an individual’s risk, we are really referring to a property of a population similar to that to which the individual is assumed to belong.

(15) The implications of risk for public policy, and for radiation protection in particular, are controversial, largely because risk estimates are uncertain and because there are legitimate interests in avoiding radiation-related risks and in maintaining radiation-related benefits and/or avoiding costs associated with unnecessary exposure reduction. A person who may be at risk of radiation-related cancer will naturally insist on proof that the risk either does not exist or is small enough to be tolerated in view of the presumed benefit. A person whose interest is in maintaining the benefit, or avoiding costs associated with reduction of exposure, will demand proof that there is a risk that is high enough to be of concern. The problem is inherently political, and its fair resolution requires information about risk, including its uncertainty, framed so as to address the concerns of both viewpoints.

(16) As epidemiological investigations of radiation-related cancer risk have evolved over time, emphasis has shifted from the discovery that radiation is indeed a cancer risk factor, to demonstration of a radiation dose–response relationship, to identification of factors that modify the dose–response relationship, to examination of assumptions inherent in the risk estimation process. IR exposure is a known, and well-quantified, human cancer risk factor. Nevertheless, estimation of cancer risk following radiation exposure is a very uncertain process for most cases of regulatory and/or popular concern. One reason is that risk estimates are usually applied to exposed populations different from those on which the estimates are based. Another is that public and regulatory interest is usually with exposures at radiation doses far lower than those at which useful information about risk can be obtained by studying populations with such exposures.

(17) The current report is mainly concerned with the possibility that cancer risk may be increased following exposure to IR. There is a great deal of information about this question. A second possibility, which is also a matter of concern, is that exposure may be associated with increased transgenerational cancer risk. Various epidemiological and laboratory studies have examined whether risks of cancer are raised in offspring following parental radiation exposure. These studies have been reviewed in detail elsewhere (Boice et al., 2003; COMARE, 2002, COMARE, 2004). Cellular and animal studies indicate that the induction of cancer in the offspring of irradiated parents is possible in principle. However, the findings in mice have not been consistent. No effect has been seen in some strains (Cattanach et al., 1995), whereas a raised risk has been observed that is greater than that predicted by the conventional induction rate for gene mutations in other strains (Nomura, 1982).

(18) Epidemiological studies conducted in several countries do not provide convincing evidence to suggest that occupational radiation exposure alone results in an increased incidence of childhood cancer in the offspring of male workers; data for the offspring of female radiation workers are too sparse to draw conclusions (COMARE, 2002). In the case of a cluster of childhood leukaemia cases among children in the village of Sellafield, UK, possibly associated with paternal employment at the nearby Sellafield nuclear reprocessing plant (Gardner et al., 1990), such an association received little or no support from other data (Wakeford, 2002). A better case can perhaps be made in the context of the well-documented phenomenon of increased levels of childhood leukaemia in so-called new towns, in which there has been an influx of residents from different areas; the postulated mechanism is an unknown viral aetiology affecting previously unexposed residents (Doll, 1999, Doll et al., 1994). In addition, follow-up of about 40,000 offspring of the Japanese atomic bomb survivors has not shown any association between the incidence of cancer in children and young adults and parental dose (Izumi et al., 2003). Thus, the subject of transgenerational risk, while a legitimate subject of scientific investigation, is insufficiently developed to provide much information on risks associated with low-dose radiation. It is briefly discussed in Chapter 5 in connection with radiation-induced genomic instability, but is not pursued further in this report.

2.2. Dependence of cancer risk on radiation dose 

(19) There is reasonably good epidemiological information on cancer risk following acute exposures in the range 0.2–5 Gy and (for partial-body exposures) above. There are numerous epidemiological studies of populations containing ‘high-dose’ subsets with radiation doses in this range. These populations include patients treated with radiation for benign and malignant disease; patients who received extensive diagnostic radiography over a lengthy illness, such as tuberculosis patients treated with lung collapse therapy monitored by frequent fluoroscopy examinations; people who received substantial exposures because of their occupations, such as uranium miners exposed to radon decay products in mine atmospheres, and instrument dial painters who ingested radium contained in luminescent paint; and survivors of the atomic bombings of Hiroshima and Nagasaki, Japan. These studies, and, in particular, inferences based on the moderate-to-high-dose component of the populations under study, form the primary epidemiological basis for estimation of radiation-related risk. Comprehensive reviews of epidemiological information on radiation-related cancer risk have been published recently (NCRP, 2001, UNSCEAR, 2000).

(20) Some benchmarks of radiation exposure levels are given in Table 2.1. Yearly natural background effective doses in normal background areas are 0.4 mSv from cosmic radiation, depending upon altitude (the dose from a typical round trip between New York and Paris by commercial airline would be 0.03 mSv); 0.5–4 mSv from radioactivity in rocks and soil, depending on local geology; 0.25 mSv from naturally occurring radionuclides in the human body; and of the order of 1.2 mSv effective dose (∼10 mSv equivalent dose) to the lung from inhaled radionuclides (radon, thoron, and their decay products) (UNSCEAR, 2000). Common diagnostic examinations produce effective doses ranging from 0.01 mSv for x rays of a foot or hand, to 4 mSv for a barium enema (Mettler and Upton, 1995), to 25 mSv for a paediatric computer tomography scan of the abdomen if adult settings are used (Brenner et al., 2003). An astronaut may get ∼2–3 mSv tissue-weighted effective dose on a typical 3-day space shuttle mission, and about 50 mSv on a 60-day tour in the international space station (NCRP, 2000). Estimated acute, neutron-weighted doses to the colon (weighted dose=gamma dose plus 10 times neutron dose) from the atomic bombings of Hiroshima and Nagasaki ranged from less than 1 mGy to nearly 6 Gy for survivors who were exposed within 3 km of the explosions and who were still alive in October 1950. Among survivors with estimated doses between 5 mGy and 4 Gy, the average was 200 mGy (RERF, 2003). An acute, uniform whole-body dose of 5 Gy is very likely to be fatal without prompt medical attention, but partial-body radiation therapy for cancer often requires organ doses an order of magnitude higher. Fractionation or protraction of exposure can allow higher doses to be tolerated in terms of acute effects. Cumulative occupational exposures among monitored radiation workers were about 20 mSv in several major studies (Gilbert, 2002), and the recommended upper limit for radiation workers is 20 mSv/year averaged over 5 years, and no greater than 50 mSv in any one year (ICRP, 1991). However, yearly effective doses at the Mayak plutonium facility approached 1 Sv for some workers during the earlier years of production (Akleyev and Lyubchansky, 1994, Khokhrykov et al., 2000).

Table 2.1. Some sources and amounts of ionising radiation exposure [unless noted, from Mettler and Upton (1995)]

CT, computer tomography; LSS, Life Span Study.

Computed using data set downloaded from Radiation Effects Research Foundation website (RERF, 2003).

(21) Dose–response data (e.g. pertaining to cancer morbidity) can be described in a number of ways, such as by arranging observations in order of dose, grouping them into consecutive dose intervals, and plotting cancer rates by dose interval (Fig. 2.1). Sophisticated modelling is not strictly necessary to establish the existence of a dose- response relationship; that can be done by a test of increasing trend, usually obtained by fitting the data to a simple model, e.g.:

(2.1)

(2.2)

• 
  • View Large Image
  • Download to PowerPoint

• Fig. 2.1. 

  Dose-specific excess relative risk of solid cancer among atomic bomb survivors, 1958–1987, by interval of neutron-weighted, estimated radiation dose to the colon. Fitted dose–response functions correspond to statistical tests of increasing trend according to the linear (relative risk=1+αD) and log-linear [relative risk=exp(βD)] dose–response models. The baseline risk is adjusted for city of exposure (Hiroshima or Nagasaki), sex, and 5-year intervals of exposure age and age at observation for risk, using a saturated model.

(22) Here, ERR(D) is excess relative risk at radiation dose D, and α and β are unknown parameters. In testing for an increasing trend using model (2.1), the dose–response is ‘statistically significant’ when the evidence is statistically inconsistent with parameter values α less than or equal to zero or, for an analysis according to (2.2), with parameter value β less than or equal to zero. These simple models can be used in tests of overall tendency, or trend, and do not suffice to establish the shape of the dose–response curve. In Fig. 2.1, in fact, neither of the fitted functions agrees particularly well with the plotted, dose-specific data points, especially at high doses, but both simple models serve to establish the existence of a dose–response relationship.

(23) If statistical significance is not achieved by a trend test, it can be inferred that the evidence in favour of the existence of a dose–response relationship is not strong, or that any relationship is too complex to be represented by such a simple parametric function. It cannot be inferred that there is no positive dose–response relationship, unless the trend is statistically significant in the negative direction; inadequate statistical power, because of an inadequate sample size for the range of doses covered, can result in failure to achieve statistical significance in the presence of a positive dose–response relationship (see Section 2.4.2).

(24) The information that can be derived from a dose–response analysis is always conditional upon assumptions about the functional relationship between radiation dose and exposure-related excess risk. In Fig. 2.1, the interval-based estimates are based on virtually no such assumptions; the different estimates are minimally correlated with each other because they share a common reference (i.e. the value for the zero dose interval is constrained to be zero); thus, observations at any given non-zero dose interval contribute information towards the estimated ERR at that interval. However, for each of the two fitted models used for trend tests (the plots of which differ because their assumed functional forms are different), the corresponding dose-specific estimates are all determined by the same estimated parameter, α in Eq. (2.1) or β in Eq. (2.2), and are therefore perfectly dependent on each other, conditionally on the estimated dose values. The confidence intervals (CIs) on the fitted curves are accordingly much narrower than those on estimates computed separately for individual dose intervals along the abscissa.

(25) Once existence of a dose–response relationship has been established, it makes sense to find a parametric dose–response model that is consistent with the epidemiological data and plausible in terms of radiobiology. Such a model provides a way to use all dose–response data to estimate radiation-related risk at various dose levels, and at low dose levels in particular.

(26) Of the two models used here to test for trend, the linear model [Eq. (2.1)] is biologically plausible in the sense that the primary mechanism by which IR exposure is thought to influence subsequent cancer risk is damage to cellular DNA from ionising events, and the frequency of such ionising events in a defined volume of tissue is proportional to the absorbed radiation dose. The log-linear model [Eq. (2.2)] is less plausible but is often mathematically convenient (e.g. in logistic model analyses).

(27) An experimentally and theoretically derived general radiation dose–response model, often cited in connection with cancer risk related to low-LET radiation (NAS/NRC, 1980; Upton, 1961) is:

(2.3)

(28) Here α , β, γ , and δ are unknown, positive parameters. The linear term, αD, dominates at low doses (where D² is small), and the term αβ D² dominates at doses somewhat greater than the so-called ‘crossover dose’ (D=1/β) at which the terms proportional to dose and dose-squared contribute equally to estimated risk. The exponential term, exp(–γD−δD²), represents the competing effect of ‘cell killing’ or cell reproductive death, observed experimentally, that would prevent a radiation-damaged cell from becoming cancerous; this term dominates at high doses, leading to a reduction in slope and eventually to a turnover and gradual decline in risk. For the present purposes, the contribution of the parameter δ is of minor importance and we will assume δ=0 in the following text. Like the other components of Eq. (2.3), the exponential cell-killing term is modelled as a continuous function of dose without threshold. Thus, cell killing is considered to be a stochastic effect, the probability of which increases with increasing dose, and not a deterministic effect, such as tissue injury, which becomes noticeable when the proportion of damaged cells exceeds a threshold level.

(29) The general dose–response function [Eq. (2.3)] is not often used in epidemiological research, mainly because the constrained parameters β and γ produce effects opposite in curvature that may, to some extent, cancel each other out. While the model is used successfully with very precise and numerous experimental data, most epidemiological dose–response data lack the statistical power needed to support estimates for a model of such complexity. Accordingly, the cancer risk estimates reported here are generally based on an assumed linear dose–response relationship. The exception to this is the leukaemia dose–response relationship from the Life Span Study (LSS), which is based on a linear-quadratic relationship. The problem of statistical power is illustrated here using the atomic bomb survivor data of Fig. 2.1 for total solid cancers following a whole-body exposure, among the most statistically powerful epidemiological radiation dose–response data in existence at the time they were published (Thompson et al., 1994). The general model fits these data reasonably well [Fig. 2.2 (dashed line) and Table 2.2], but is not significantly better than the linear model of Fig. 2.1 (P=0.11). The estimated ERR per Gy at low doses (i.e. the estimated value of α ), 0.52 (90% CI 0.16–0.83), does not differ markedly from that according to the linear model, 0.57 (90% CI 0.49–0.66); however, the CIs are substantially wider for the more complex model, reflecting the wide range of combinations of positive values of the parameters α, β, and γ consistent with the data. The analysis offers little evidence in support of a positive value of the (dose-squared) parameter β (P=0.28), but suggestive evidence in support of a non-zero value of the cell-killing parameter γ (P=0.07).

• 
  • View Large Image
  • Download to PowerPoint

• Fig. 2.2. 

  General dose–response model, excess relative risk (ERR)(D)=αD×(1+βD)×exp(–γD−δD²), fit to the dose–response data of Fig. 2.1, and linear dose–response model, ERR(D)=αD, fit to the data subset restricted to radiation doses between 0 and 2 Sv. Details of the parameter estimates are given in Table 2.2. ERR, excess relative risk.

Table 2.2. Parameter estimates corresponding to the general dose–response model, ERR(D)=αD×(1+βD)×exp(–γ D−δD²), where D is neutron-weighted (weight=10), reconstructed radiation dose to the colon from the atomic bombings, and ERR(D) is the dose-related excess relative risk of solid cancer morbidity, 1958–1987, among members of the Radiation Effects Research Foundation’s Life Span Study cohort of survivors of the bombings (available at www.rerf.or.jp)

CI, confidence interval.

(30) Less than 1% of the members of the LSS cohort for whom dose estimates have been calculated have estimates greater than 2 Gy, and there are reasons to believe that the dose estimates above 2 Gy may be biased upwards (Pierce and Preston, 2000). Restriction of the dose–response analysis to subjects with doses under 2 Gy yielded the linear-model parameter estimate α=0.64 (90% CI 0.54–0.74). Adding either the quadratic or the cell-killing terms to the model produced zero or minimal change, whereas adding both of them yielded parameter estimates so uncertain as to be of no predictive value (Table 2.2).

(31) In the remainder of this report, epidemiological risk estimates are based on linear dose–response analyses.

2.3. Inferences based on acute exposures in the moderate-to-high dose range 

(32) The information obtained from studies of the atomic bomb survivors and other populations mentioned above is rich in detail. For many cancer sites and groups of sites, we can estimate with some precision not only the dose-specific risk of radiation-related cancer, but also its variation by cancer site and by sex, age at exposure, attained age, and/or time following exposure. In general (but not always), radiation-related relative risk is higher among women and following exposure at young ages. The relationship to age at exposure is marked for thyroid cancer, acute leukaemia, and female breast cancer (Land et al., 2003, Preston et al., 1994, Preston et al., 2003, Ron et al., 1995). Risk decreases somewhat, in relative terms, with advancing age at observation, but increases in absolute terms because baseline cancer risk tends to increase as a power of age, and faster than dose-specific decreases in ERR (Pierce, 2002, Pierce and Vaeth, 2003, Thompson et al., 1994, UNSCEAR, 2000).

(33) The relative importance of exposure age and attained age as modifiers of the radiation dose–response relationship is uncertain because, in any epidemiological follow-up study, the two quantities are highly correlated and their effects are difficult to separate. With additional follow-up, as the major exposed populations are followed to the end of their life spans, the importance of this question for lifetime risk will become moot because projection to the end of life will no longer be required for subgroups exposed at young ages. However, the dependence of radiation-related risk on exposure age and attained age are likely to remain complicated. One consideration is the presence of secular trends in baseline risk in Japan during the period of follow-up for the atomic bomb survivors over the past half century, the reasons for which are not entirely clear (Parkin et al., 2002, Preston et al., 2003).

(34) Statistically stable descriptions can be obtained of the dependence of dose-specific risk on sex, age, and time for aggregations of cancer sites such as all cancers combined, all solid cancers, all leukaemia types, and other groupings. This is useful because radiation protection is concerned with the totality of possible adverse consequences of exposure, but also because overall patterns of dependence may emerge from such analyses that can be incorporated into site-specific estimates, resulting in greater statistical precision (NAS/NRC, 2000, NCI/CDC, 2003, Pierce and Preston, 1993).

(35) There is a relatively small but growing amount of epidemiological information (Table 2.3) on modification of radiation-related risk by history of lifestyle factors, such as tobacco smoking in the case of lung cancer (Kopecky et al., 1986, Lubin and Steindorf, 1995, NAS/NRC, 1999, Pierce et al., 2003, Prentice et al., 1983), childbearing and breastfeeding in the case of breast cancer (Boice and Stone, 1978, Land et al., 1994, Shore et al., 1980), ultraviolet light in the case of basal cell and squamous cell skin cancer (Ron et al., 1998, Shore, 2001, Shore et al., 2002), and disease history in the case of type C hepatitis infection and liver cancer (Sharp et al., 2003). Much more needs to be learned about interactions of IR exposure with lifestyle factors and with exposures to other agents. It is not unlikely that some of our current inferences about the dependence of radiation-related risk on exposure age, attained age, and sex may reflect secular changes in lifestyle, and in exposure to environmental agents, that have been associated with changes over time (and with successive birth cohorts) in both baseline and radiation-related risk. Preston et al. (2000) noted concerns about the difficulties in interpreting radiation age at exposure effects in the LSS cohort.

Table 2.3. Modification of radiation-related risk by individual and lifestyle factors, and by other exposures

LSS, Life Span Study; NS, not significant.

(36) There does not appear to be an obvious, consistent relationship between baseline and radiation-related cancer risk, either across cancer sites within a single population or across populations for a single cancer site. In the female Japanese population, age-standardised (world) rates per 100,000 per year are generally similar, at about 31 for gastric cancer and 34 for breast cancer (Parkin et al., 2002), whereas in the USA, they are about 3 and 90, respectively. Among atomic bomb survivors, the radiation-related ERR at 1 Gy is 0.32 for gastric cancer and 1.6 for breast cancer (Thompson et al., 1994). Gastric cancer contributes a substantial proportion of total radiation-related risk, but that proportion is considerably less than the proportion of risk of baseline gastric cancer to total baseline cancer risk (about 22%) among atomic bomb survivors (Thompson et al., 1994) and among Japanese people generally (Parkin et al., 2002). In the USA, the ratio is 2% for males and 1% for females. For female breast cancer, the opposite is true; the baseline rate in Japan is among the lowest in the world for developed countries, whereas the total cancer rate is not much different from that in most other countries (Parkin et al., 2002). Among the atomic bomb survivors, breast cancer contributes a disproportionately large fraction of the total radiation-related cancer burden (Thompson et al., 1994). In the USA, by contrast, baseline breast cancer rates are high but the radiation-related excess risk (in absolute terms) per unit dose among medically exposed women is similar to that among the atomic bomb survivors (Preston et al., 2002). That is, the dose-specific, radiation-related component of total breast cancer risk is likely to be similar in absolute magnitude for exposed Japanese and Western populations, but is likely to be smaller in Western populations as a proportion of total breast cancer risk. For gastric cancer, on the other hand, the US baseline rate is an order of magnitude lower than that in Japan, whereas the limited information on dose-specific, radiation-related excess risk suggests that, as a multiple of baseline risk, it may be comparable to that in the atomic bomb survivors (Carr et al., 2002, Griem et al., 1994).

(37) The above information suggests that, for breast cancer, radiation-related ERR per Gy (excess risk per Gy expressed as a multiple of the Japanese baseline risk) based on atomic bomb survivor data would overestimate the risk for an exposed US population while, for gastric cancer, radiation-related excess absolute risk (EAR; the difference between risk following exposure and the Japanese baseline risk) would result in an overestimate of risk for the US population. In addition, data are available for leukaemia and thyroid cancer from the atomic bomb survivors and medically and environmentally irradiated cohorts (Ron et al., 1995; UNSCEAR, 2000, Annex I, Table 21). For most other cancers, information of a similar nature is limited or non-existent (Table 2.3). This is not a trivial matter because any transfer of a risk estimate from one population to another requires making an assumption, explicit or implicit, about the relationship between excess and baseline risk. Moreover, for some sites (e.g. stomach, liver, and oesophagus), baseline rates can differ markedly between populations (Parkin et al., 2002).

(38) It should not be surprising that the relationship between radiation-related and baseline risk in different populations is not consistent for different cancer sites. There are reasons, as yet poorly understood, why baseline breast cancer rates are high in the USA, and why baseline gastric cancer rates are high in Japan. These reasons are almost surely related to differences in lifestyle, since the descendants of immigrants to the USA, for example, have tended to develop cancer rates that are typical of the general US population (Haenszel and Kurihara, 1968, Ziegler et al., 1993) and different from those of their countries of ancestral origin. The lifestyle factors affecting the rates for breast and stomach cancer are probably different, at least in part, and probably interact differently with radiation dose.

(39) Much of environmental, nutritional, and occupational cancer epidemiology is concerned with identifying cancer risk factors that may account for some part of the variation of site-specific baseline rates among populations. While there has been much progress, the problem is vast and, as discussed in Section 2.3.2, there is only limited information on interaction between radiation dose and lifestyle factors in terms of cancer risk. Thus, it is likely that, for the foreseeable future, the most useful information relevant to the transfer of radiation-related risk coefficients from one population to another will come from multinational comparisons of site-specific radiation-related risk, rather than from investigations of underlying cancer risk factors and their interactions with radiation dose.

(40) Risk estimates for low-LET radiation protection purposes are based mainly on epidemiological studies of populations exposed to substantial doses of medical x rays, or to mixed gamma and neutron radiation from the Hiroshima and Nagasaki atomic bombs. According to the DS86 dose reconstruction algorithm (Roesch, 1987) as represented by public-use RERF data sets (RERF, 2003), the correlation between neutron and gamma doses within each city is greater than 95%, and the proportion of total absorbed bone marrow dose contributed by neutrons is only 0.7–2.7% in Hiroshima and 0.3–0.7% in Nagasaki, depending upon shielding and exposure distance. According to the as yet unpublished DS02 dose reconstruction system, the neutron component is reduced slightly, compared with DS86, in both Hiroshima and Nagasaki. In particular, an anticipated large increase of the neutron component for low-dose survivors in Hiroshima did not materialise (Preston et al., 2004). Due to the relatively small contribution from neutrons, there is minimal statistical power for estimating the relative biological effectiveness of the two radiation types based on the atomic bomb survivor data. Moreover, there are essentially no useful data on cancer risks in populations exposed mainly to neutron radiation (IARC, 2001). Therefore, the relative biological effectiveness of neutron vs gamma-ray doses can only be estimated from experimental data. Risk coefficients for gamma-ray doses are obtained from the atomic bomb survivor data through the use of a nominal weighting factor of 10 for the neutron component of dose (Thompson et al., 1994). However, Preston et al. (2004) noted that with the advent of the DS02 atomic bomb survivor dosimetry system, the estimated neutron doses are so small that the question of variation in the estimated gamma-ray dose–response relationship due to uncertainty in the choice of the neutron-weighting factor is essentially moot.

(41) Cancer risks associated with alpha-radiation exposure have been studied for lung cancer among uranium miners exposed to inhaled radon decay products (NAS/NRC, 1999) and in populations exposed to lower radon levels in residential settings, for bone cancer associated with ingested ²²⁶Ra and ²²⁸Ra among former radium dial painters (Carnes et al., 1997, Fry, 1998, Stebbings et al., 1984) and with injected ²²⁴Ra in patients treated for benign disease (Nekolla et al., 1999, Nekolla et al., 2000, Spiess and Mays, 1970), and for cancers of the liver and other organs in patients injected with x-ray contrast media containing thorium (Travis et al., 2003). Thus, estimates of cancer risk associated with exposure to alpha-particle radiation have a basis in direct observations, while estimation of risk associated with neutron exposure is indirect, relying on scaled estimates of risk from low-LET radiation using experimentally derived estimates of the effectiveness of neutrons compared with low-LET radiation.

(42) Epidemiological risk estimates based on exposure to gamma rays (photons with energies of >250 keV) and most medical x radiation (photons with energies in the 30–250 keV range) are often treated as interchangeable quantities (ICRP, 1991). However, it has long been considered, based on biophysical considerations, that medical x rays are more effective biologically than higher-energy gamma rays. This consideration has been cited as a factor that may complicate inferences based on comparisons of cancer risk associated with fractionated x-ray exposures and acute gamma-ray exposures (Brenner, 1999). Kocher et al., 2002, NCI/CDC, 2003 estimated uncertain radiation effectiveness factors (REF), compared with gamma radiation, for 30–250 keV and soft (<30 keV) x rays, assigning subjective uncertainty distributions with mean REF values of 2 and 2.7, respectively, and 95% uncertainty limits of 1–4.7 and 1.1–6.4, respectively, for the two x-ray energy ranges. Electrons at energies like those of secondary electron tracks induced by gamma-ray photons, i.e. above 30 keV, were assigned an REF value of 1, while lower-energy electrons were assigned an uncertain REF with a mean of 2.6 and 95% uncertainty limits of 1.2–5.0.

2.4. Estimation of risk at low doses and low dose rates 

(43) Except for radiation therapy, where there is a recognised benefit from the radiation dose itself, very few people are exposed to radiation effective doses of 0.2 Sv and above. Most public concern is with exposures to less than 50 mSv, the historical annual limit for radiation workers before a reduced level (20 mSv) was recommended in ICRP Publication 60 (1991); that concern extends to effective doses well below 1 mSv, the annual population limit recommended by both the ICRP (1991) and the NCRP (1993), as well as the annual dose from natural background radiation for most tissues other than the lung. As mentioned previously, a chest x ray delivers about 0.1 mGy to lung tissue, the dose to breast tissue from a two-view mammography examination is about 3 mGy, and an astronaut may get about 2.4 mSv tissue-weighted effective dose on a typical 3-day space shuttle mission (NCRP, 2000).

(44) Although such low-dose exposures (except, of course, the astronaut’s) are very common, it is extremely difficult to estimate the associated excess cancer risks by studying populations with exposures limited to the low-dose range. This is because, at low doses, the radiation-related excess risk, which is thought to be proportional to dose or perhaps somewhat less when compared with risks at higher doses, tends to be dwarfed by statistical and other variation in the background risk level in the absence of exposure. As a result, truly enormous sample sizes (e.g. millions) would theoretically be required to obtain a statistically stable estimate of radiation-related risk, and even then the estimate would be untrustworthy because we do not understand, and therefore cannot control or adjust for, all of the sources of variation in baseline levels of risk (Land, 1980). At higher dose levels, there are fewer problems because the excess risk tends to be large relative to statistical variation in baseline risk, and we are more likely to understand the causes of any substantial variation in baseline risk that may be confounded with radiation dose.

(45) Suppose that: (1) baseline cancer risk in a given population, for a certain (unspecified) subset of cancer sites, was known to be 10%; (2) exposure to a whole-body effective dose of 1 Sv would double the risk (i.e. add another 10%) of that same subset of cancers; and (3) excess risk was strictly proportional to radiation dose over the interval 0–1 Gy. Suppose also that it was possible to find large study populations with baseline risks known to be 10% and with uniform exposures to 1 Gy, 100 mGy, 10 mGy, or 1 mGy. (This is a simplified, and unrealistic, version of a study in which observed cancer frequencies in an exposed population are compared with expected frequencies calculated on the basis of published population rates. Note also that the assumed baseline rate, doubling dose, etc. were chosen to simplify the arithmetic and not to describe any actual population or subset of cancers.) The estimated excess cancer rate in such a population would be the number of cancers divided by the population size, less the known baseline rate of 10%. The estimated excess would be distributed approximately as a normal random variable with the mean equal to the baseline rate, 10%, times the dose D, in Gy, and variance equal to 10% times (1+D) divided by the population size, N. The population size needed to be able to detect the excess risk associated with dose D with a probability of 80% using a one-sided test at the 5% significance level is shown in Table 2.4. The calculation is, in fact, unrealistically optimistic since, as illustrated in a later example, one can never be that sure of the baseline rate in any exposed population. If, as is almost always the case, we had to estimate the baseline rate by including an equal number of non-exposed subjects, more than twice as many total (exposed plus non-exposed) subjects would be required to have equal power for detecting the difference. Moreover, one could still not be sure, particularly at the lower dose levels, that the exposed and unexposed populations were truly comparable in terms of baseline rates at the level of resolution required.

Table 2.4. Statistical power calculations for a hypothetical study in which baseline cancer risk, for an (unspecified) subset of cancer sites, is known to be 10%, and the unknown radiation-related excess risk is 10% at 1 Gy and proportional to dose between 0 and 1 Gy

(46) If an enormous study population is required to detect any excess risk associated with exposure to a small radiation dose, it follows that the implications are unexciting if we use a much smaller population and fail to detect any excess risk. A result predictable under both of two opposing hypotheses supports neither of them against the other. Thus, for example, failure of epidemiological studies to demonstrate a statistically significant excess cancer risk associated with exposures of the order of 1 mGy does not imply that there is no risk, although it does suggest that any such risk is small relative to baseline cancer rates.

(47) At low and very low radiation doses, statistical and other variations in baseline risk tend to be the dominant sources of error in both epidemiological and experimental carcinogenesis studies, and estimates of radiation-related risk tend to be highly uncertain because of a weak signal-to-noise ratio and because it is difficult to recognise or to control for subtle confounding factors. At such dose levels, and with the absence of bias from uncontrolled variation in baseline rates, positive and negative estimates of radiation-related risk tend to be almost equally likely on statistical grounds, even under the LNT theory. Also, by definition, statistically significant positive or negative findings can be expected in about one in 20 independent studies when the underlying true excess risk is close to zero. Thus, even under the LNT theory, the smaller the dose, the more likely it is that any statistically significant finding will be a purely chance occurrence, and that it will be consistent with either beneficial effects of radiation (hormesis) or a grossly exaggerated risk (Land, 1980). Such estimates tend to be only a small fraction of the total, but when selectively presented, they can give the appearance of a substantial and even overwhelming body of evidence in one direction or the other.

Medical studies 

(48) There is, in fact, some direct epidemiological evidence of excess cancer risk associated with radiation exposures of the order of a few tens of mGy. One example is a relative risk of approximately 1.4 for mortality from leukaemia and solid cancer by 15 years of age, which has been observed in several case–control studies (Bithell and Stiller, 1988, Harvey et al., 1985, ICRP, 2003; MacMahon, 1962; Monson and MacMahon, 1984, Stewart et al., 1956) among children exposed in utero to radiation from x-ray pelvimetry. There has been considerable debate on the interpretation of these in-utero studies. Difficulties cited have included the absence of an association in any cohort study, including the atomic bomb LSS, discrepancies between radiation-related failure to find increased childhood cancer rates among twins compared with singletons despite presumably higher exposure frequency among twins, and possible biases in the case–control approach based on the argument that similar estimated relative risks for leukaemia and solid cancers is an implausible finding (Boice and Miller, 1999, Doll and Wakeford, 1997, ICRP, 2003).

(49) In a comprehensive review paper, Doll and Wakeford (1997) concluded that, on the balance of evidence, ‘irradiation of the fetus in utero increases the risk of childhood cancer, that an increase in risk is produced by doses of the order of 10 mGy, and that in these circumstances the excess risk is approximately 6%/Gy’. They discussed four different grounds for controversy which have been raised to suggest that the estimates derived from the case–control studies may be unreliable. Three of these grounds [including the evidence for bias in the case–control studies emphasised by Boice and Miller (1999)] were considered by Doll and Wakeford (1997) to be probably or possibly invalid, and they judged that only the remaining one would appear to be serious, ‘namely, the lack of any comparable excess in cohorts known to have been irradiated in utero, most notably in those exposed to radiation from the explosion of the atomic bombs in Japan’.

(50) The atomic bomb survivor study is discussed below. Regarding the possibility of a cohort study of childhood cancer following pelvimetry, however, childhood leukaemia and solid cancer are very rare, and the sample size requirement for a cohort study of adequate statistical power would be unmanageably large. For example, according to current SEER statistics, in the USA, the likelihood of dying from any cancer between birth and 10 years of age is 0.026%, and the likelihood of being diagnosed with any cancer is 0.164% (use ‘FastStats’ under http://seer.cancer.gov/statistics). If it were possible to select a cohort with equal numbers of exposed and non-exposed children, it would require a total sample size of over 630,000 children in order to detect, with 80% statistical power at the 5% level of significance, a 1.4 relative risk for cancer mortality among the exposed, and a total sample size of about 100,000 to detect a similar increase in cancer morbidity risk. If we could not easily select exposed and non-exposed children in advance, and if there were 10 times as many non-exposed as exposed children in the population [approximately the ratio observed by Stewart et al. (1956)], the required numbers to detect a relative risk of 1.4 would be about 2 million for mortality and 320,000 for morbidity. Thus, the absence of a positive cohort study based on pelvimetry exposure would not appear to be a strong reason to question the findings of the case–control studies of childhood cancer risk following in-utero exposure to x-ray pelvimetry.

(51) The atomic bomb survivor studies are a different matter. The total number of 3018 in-utero exposed subjects who survived until 1 October 1950 includes 313 with estimated doses between 0.1 and 0.5 Gy (mean 0.23 Gy), 88 with estimated doses between 0.5 and 1 Gy (mean 0.72 Gy), and 54 with estimated doses of 1 Gy or more (mean 2.4 Gy) (Delongchamp et al., 1997), and there would appear to be sufficient power to detect a linear dose–response relationship at an ERR per Gy consistent with the results of the case–control studies of children exposed through x-ray pelvimetry, given efficient detection of cancer cases at under 15 years of age. An examination of published studies (Delongchamp et al., 1997, Yoshimoto et al., 1988, Yoshimoto et al., 1994) of morbidity and mortality following in-utero exposure from the Hiroshima and Nagasaki bombs indicates that ascertainment of cancer among exposed and non-exposed children during the period 1945–1955 and, especially, 1945–1950 may have been incomplete. The most recent reports (Delongchamp et al., 1997, Yoshimoto et al., 1988, Yoshimoto et al., 1994) are confined to cancer mortality or morbidity among people alive as of 1 October 1950, excluding 198 deaths occurring before that date. Most of the early deaths were infant deaths that occurred shortly after the bombings, with little documentation of cause (Delongchamp et al., 1997). According to current SEER rates (http://canques.seer.cancer.gov), 5.0 incident baseline cancer diagnoses would be anticipated to occur before 10 years of age in a population of 3018; of these, nearly two-thirds would be expected to occur before 5 years of age. Thus, there is a definite possibility that, among members of the in-utero cohort and/or people who would have been included in the cohort if they had survived until 1 October 1950, radiation-related and/or baseline cases of childhood leukaemia occurred but were not detected for various reasons during the first 5 or more years following the bombings. However, an increase is observed when the lifetime risk is studied in this cohort (Delongchamp et al., 1997).

(52) A less direct example is increased breast cancer risk among young women exposed to high cumulative doses from multiple thoracic x-ray exposures, delivered in fractions that were, on average, of the order of 10 mGy (Boice et al., 1991, Davis et al., 1987, Doody et al., 2000, Howe and McLaughlin, 1996). In the case of fluoroscopy examinations during lung collapse therapy for tuberculosis where anterior–posterior exposure occurred, individual fractions could sometimes exceed 100 mGy and, in one such study, it was assumed, for dose reconstruction purposes, that 25% of the fluoroscopic examinations involved direct, frontal exposures for which the breast dose per examination was 54 mGy, and 75% were directed from the back, for which the average breast dose was 1.8 mGy (Boice et al., 1978). Successive exposures were separated by a week or more, but were repeated often enough to yield cumulative doses of hundreds or even thousands of mGy. Excess (absolute) risks per unit of total dose [about 10 excess cases/10,000 women/year/Gy at 50 years of age, following exposure at 25 years of age (Preston et al., 2002)] were comparable with those associated with acute doses among atomic bomb survivors (Boice et al., 1979, Land et al., 1980, Little and Boice, 1999, Preston et al., 2002). A similar relationship for excess risk of lung cancer, compared with estimates based on high-dose, acute exposures, was not observed among fluoroscopy patients, even though lung doses were comparable with breast doses (Davis et al., 1987, Howe, 1995). Although excess lung cancer risk per unit dose of acute radiation is, in general, less than for breast cancer (Thompson et al., 1994), the difference between the breast and lung cancer findings among fluoroscopy patients suggests that there may be variation among cancer sites in terms of fractionation effects. It should be kept in mind, however, that exposure to tobacco smoke is by far the dominant risk factor for lung cancer. Among, for example, tuberculosis patients who underwent lengthy courses of lung collapse therapy associated with high cumulative radiation doses from fluoroscopic examinations, below-average exposure to tobacco smoke may mask a radiation-related increase in lung cancer risk, although attempts were made to control for smoking in the analyses.

(53) It is difficult to make inferences about protraction and fractionation effects from studies of thyroid cancer among irradiated populations, largely because risk estimates vary among populations for reasons that are poorly understood. A highly significant, dose-related excess risk of thyroid cancer was observed among 10,834 Israeli patients treated as children by x-ray depilation for ringworm of the scalp (Tinea capitis), with estimated (fractionated) dose to the thyroid gland averaging 90 mGy (range 40–500 mGy), compared with 16,226 non-exposed comparison subjects (Ron et al., 1995). The estimated linear model ERR per Gy was 32.5 (95% CI 14–57), based on 44 cases among the exposed and 16 cases among the non-exposed. On the other hand, no significant excess was observed among 2224 patients given similar treatment (average thyroid dose 60 mGy) in the USA compared with 1380 patients given topical ointment treatment alone; two thyroid cancers were found in the x-ray group, consistent with general population rates, and none were found in the non-irradiated group. However, the between-study difference in risk estimates was not statistically significant (Shore et al., 2003). More generally, a pooled analysis of data from five studies of thyroid cancer following irradiation during childhood (Ron et al., 1995), including the Israeli T. capitis patients, the youngest atomic bomb survivors, two populations treated by x ray for enlarged tonsils or lymphoid hyperplasia, and a population treated for supposedly enlarged thymus, obtained an overall ERR per Gy of 7.7 (95% CI 2.1–28.7). Two Swedish studies of skin haemangioma patients with low-dose-rate exposures from ²²⁶Ra obtained similar estimates: ERR/Gy=7.5 (95% CI 0.4–18.1) based on an estimated mean thyroid dose of 120 mGy (Lindberg et al., 1995), and ERR/Gy=4.9 (95% CI 1.3–10.2) based on a mean dose of 260 mGy (Lundell et al., 1994).

Occupational studies 

(54) Except for (mainly historical) worker populations with fairly high levels of exposure, such as uranium miners (NAS/NRC, 1999), radium dial painters (Stebbings et al., 1984), Russian plutonium workers (Gilbert, 2002), and early radiologists (Matanoski et al., 1975; Smith and Doll, 1981), most occupational studies can be classified as low dose and, therefore, of low statistical power. Their main utility is to validate generally accepted estimates in the sense that they are consistent with estimated radiation-related risks among regulated radiation workers. For example, a large, combined analysis of cancer mortality among nuclear workers in the USA, the UK, and Canada found a statistically significant dose–response relationship for leukaemia and a non-significant dose–response relationship for all solid cancers (Cardis et al., 1995). Occupational radiation exposure and cancer mortality in the UK National Registry for Radiation Workers were similarly associated, and consistent with estimates based on the atomic bomb survivor studies (see below) (Muirhead et al., 1999). Patterns of cancer mortality were inversely related to year of first employment among US radiological technicians, consistent with a radiation aetiology given higher occupational exposures to radiation in earlier compared with more recent times (Mohan et al., 2002, Mohan et al., 2003).

Atomic bomb survivor studies 

(55) It is sometimes forgotten that the vast majority of the exposed (as distinguished from people not present at the time of the bombings) LSS cohort of atomic bomb survivors received radiation doses under 100 mGy (Table 2.5). For solid cancer mortality between 1950 and 1997 (Preston et al., 2003), direct assessment of risks at low doses obtained a statistically significant dose–response relationship when the analysis was restricted to survivors with dose estimates less than approximately 120 mGy. The estimated ERR per Gy over this range was 0.74 (90% CI 0.1–1.5). There was no indication that the slope of the fitted dose–response curve differed significantly (P>0.5) from the estimate over the full dose range (ERR/Gy=0.47), and no evidence of a threshold. As discussed below, a similar result was obtained from analyses of the same epidemiological data using the DS02 dose estimates (Preston et al., 2004).

Table 2.5. Distribution of subjects, solid cancers, and estimated radiation-associated, excess solid cancers among 79,901 exposed members of the Life Span Study cohort of Hiroshima and Nagasaki atomic bomb survivors (Pierce and Preston, 2000)

(56) An earlier analysis of solid cancer incidence data from the LSS Tumor Registry for 1958–1994 (Pierce and Preston, 2000) focused on people exposed at distances under 3000 m, of whom approximately 10,000 had estimated neutron-weighted doses under 5 mGy and 41,000 had doses between 5 and 500 mGy. An analysis restricted to people exposed at distances less than 3000 m found a statistically significant linear dose–response relationship that was not overestimated by linear-model risk estimates computed over the wider dose ranges 0–2 Gy or 0–4 Gy (Fig. 2.3). A statistically significant estimate was obtained from an analysis restricted to the 0–120 mGy dose range; another finding was that any threshold over 60 mGy would be statistically inconsistent with the data.

• 
  • View Large Image
  • Download to PowerPoint

• Fig. 2.3. 

  Estimated low-dose relative risks. Dose-specific cancer rates over the 1958—1994 follow-up period relative to those for an otherwise similar exposed person, averaged over the follow-up, for sex, and for 30 years of age at exposure. The dotted lines represent 1 standard error limits for the smoothed curve. The straight line is the estimated linear dose–response relationship for 0–2 Sv (see inset). The unity baseline corresponds to zero-dose survivors exposed within 3 km of the bombs. The horizontal dotted line represents the alternative baseline if survivors exposed beyond 3 km had been included. Source: Pierce, D.A., Preston, D.L., 2000. Radiat. Res. 154, 178–186.

(57) When cohort members exposed beyond 3000 m were included in the analysis, the estimated slope of the fitted dose–response relationship was reduced slightly (by 3%), and the statistical significance of the fitted linear dose–response relationship in the range 0–120 mGy was reduced. Fig. 2.3 shows a moving-average plot of dose-specific cancer rates over the 0–500-mGy range, with uncertainty bounds corresponding to ±1 standard deviation (SD). At 100 mGy, the moving-average estimate of relative risk is about 3.7 SD units above 1 for an analysis restricted to survivors exposed at distances under 3000 m, and about 2 SD units above the redefined baseline (represented by the dotted horizontal line at a relative risk of approximately 1.04) using the less restricted data set.

(58) Fig. 2.4 is based on the same data as Fig. 2.3 but shows linear regression estimates of the ERR per Gy over dose intervals that are progressively trimmed of high-dose data. Moving from right to left, the rightmost estimate and its standard error (SE) are based on observations over the dose range 0–2 Gy, the next on 0–1.5 Gy, and so on, while the leftmost estimate is based on data at 0–0.05 Gy. There is more variation between consecutive estimates on the left-hand side of each graph than there is on the right-hand side, and the ±SE limits become progressively wider towards the left-hand side of each panel as the dose range is further restricted at the high end (Donald Pierce, personal communication).

• 
  • View Large Image
  • Download to PowerPoint

• Fig. 2.4. 

  Linear regression estimates of the excess relative risk (ERR) per Gy (points and connecting line, with error bounds of ± one standard error) for solid cancer incidence, based on Poisson regression over dose intervals of differing ranges from zero to the horizontal co-ordinate of the plotted point. The analysis is limited to proximal survivors exposed at distances under 3 km.

(59) The reference population used in the analyses of Fig. 2.3, Fig. 2.4 is the group of ‘proximal’ survivors (exposed within 3 km) in Hiroshima and Nagasaki with neutron-weighted dose estimates less than 5 mGy. This choice was justified on the basis that the ‘distal’ population exposed beyond 3 km was more rural and may have experienced different cancer risk factors other than radiation compared with the more urban proximal survivors. The horizontal line in Fig. 2.3, corresponding to a relative risk of 1.04, represents the baseline if the distal survivors had been included in the analysis. Fig. 2.5 repeats the analysis of Fig. 2.4 with the distal survivors included. While estimates on the ERR per Gy based on higher-dose data are little affected by the change, the estimates at the left-hand side of Fig. 2.5 are substantially lower than those on the left-hand side of Fig. 2.4, with similarly wide error bounds. Comparison of Fig. 2.4, Fig. 2.5 demonstrates the sensitivity of estimates, if based on low-dose data alone, to the influence of minor, and largely unknown or poorly understood, confounding factors.

• 
  • View Large Image
  • Download to PowerPoint

• Fig. 2.5. 

  Linear regression estimates of the excess relative risk (ERR) per Gy (points and connecting line, with error bounds of ± one standard error) for solid cancer incidence, based on Poisson regression over dose intervals of differing ranges from zero to the horizontal co-ordinate of the plotted point. The analysis is based on all exposed survivors with estimated radiation doses less than 2 Gy.

(60) The same overall patterns are seen in Fig. 2.6, an analysis similar to Fig. 2.5 (in that data for distal survivors contribute to the estimates) for LSS breast cancer incidence, 1950–1990 (Land et al., 2003). Together, Fig. 2.4, Fig. 2.5, Fig. 2.6 demonstrate that regression estimates of dose-specific cancer risk for combined sites and for some single sites are highly consistent with linearity, depend substantially on excess risk observed among survivors with estimated doses under 200 mGy, and are statistically unstable when based solely on data pertaining to doses under approximately 100 mGy. These analyses provide no strong evidence that excess risks per unit dose are substantially different at very low doses than at doses up to 4 Gy.

• 
  • View Large Image
  • Download to PowerPoint

• Fig. 2.6. 

  All-age linear regression estimates of excess relative risk (ERR) per Gy for female breast cancer assuming a 12-year minimum latent period, with dose-specific data trimmed from the right. Horizontal placement corresponds to the mean breast tissue dose for the highest neutron-weighted kerma interval included in the regression. Thus, the rightmost point corresponds to the full dose range, the next point to the left to doses under 4 Gy, the next to doses under 3 Gy, and so on.

High background area studies 

(61) There are a number of published epidemiological studies of cancer rates in populations living in areas where natural background levels are several times greater than those experienced by the vast majority of the world’s population, e.g. the high background area in Yagjiang, China, where the estimated annual effective dose from natural background is about 6.4 mSv (three-fold higher than in most areas of the world). Studies of such populations generally find relative risks for cancer mortality or nodular thyroid disease (considered to be a biomarker for thyroid cancer risk) that are not significantly different from 1 (Tao et al., 1999, Tao et al., 2000, Wang et al., 1990). It is, of course, possible that no significant increase in risk is observed because there is none, but for reasons discussed in Sections 2.4.1, 2.4.2, problems of a low signal-to-noise ratio in the epidemiological data, insufficient sample size, and difficulties recognising and controlling for possible confounding factors that may have effects of similar magnitude on risk, pose problems of interpretation of findings that are difficult or impossible to overcome.

(62) Epidemiological data are informative about radiation-related risks at acute doses, on a logarithmic scale, in the moderately high (∼1 Gy), moderate (∼100 mGy), and, to some extent, low (∼10 mGy) dose ranges, but not in the very low (∼1 mGy) and extremely low (∼0.1 mGy) ranges. Arguably the most important single problem in radiation risk protection is how to extrapolate from statistically stable, and relatively unbiased, risk estimates that pertain to higher dose exposures down to the lower dose levels that are of greater concern in everyday life. The analyses of Fig. 2.3, Fig. 2.4, Fig. 2.5, Fig. 2.6 suggest that for the 1958–1987 LSS solid cancer incidence data at least, linear extrapolation over one order of magnitude, e.g. from 2 Gy to 200 mGy, is justified. Dose–response analyses for leukaemia risk, on the other hand, support a linear-quadratic dose–response relationship with approximate equivalence of the linear and dose-squared components of risk at bone marrow doses of approximately 1 Gy (Preston et al., 1994). Solid cancer mortality data (all sites combined) for 1950–1997 (Preston et al., 2003) suggest linearity even for doses in the 0–150-mGy range. A later analysis, using the DS02 dosimetry, found statistically significant upward curvature over the restricted dose range 0–2 Gy, but the authors noted that linear model dose–response analyses restricted to 0–1 Gy, 0–0.5 Gy, and 0–0.25 Gy, considered to be more relevant to risk at low and very low doses, gave substantially higher estimates of low-dose risk and they therefore did not recommend using the linear-quadratic model to estimate low-dose risk (Preston et al., 2004).

Dose and dose-rate effectiveness factor 

(63) The combined-site LSS solid cancer incidence data support linearity of the dose–response relationship down to low-LET radiation doses of the order of 200 and even 100 mGy. They provide no evidence that linearity does not continue down to zero dose, nor do they rule out the possibility of non-linearity at 10 mGy and lower. The in-utero pelvimetry studies and the fractionated fluoroscopy study breast cancer data suggest that radiation doses of the order of 10 mGy/fraction are associated with excess cancer risk. However, the same fluoroscopy cohort shows no evidence of increased lung cancer risk. The heterogeneity of dose distribution between patients has, however, been reported to be considerable (Boice et al., 1978), and these fluoroscopy data do not necessarily imply proportionality between radiation dose and excess cancer risk down to a few tens of mGy. The curvilinearity of the LSS leukaemia dose–response relationship is the main epidemiological evidence in support of a reduced risk per unit dose at low and very low doses [otherwise suggested by experimental observations (NCRP, 1980)]. Such curvilinearity is consistent with ICRP and UNSCEAR recommendations that extrapolated dose-specific risk estimates should be divided by a dose and dose-rate effectiveness factor (DDREF) of 2 for chronic exposures and for acute doses less than 200 mGy (ICRP, 1991, NCRP, 1993, UNSCEAR, 1993). A DDREF greater than 2 would, in the context of a linear-quadratic dose–response model, be statistically inconsistent with the 1958–1987 LSS solid cancer incidence data (Pierce and Preston, 2000), although not necessarily with the 1950–2000 LSS solid cancer mortality data (Preston et al., 2004).

(64) An independent analysis of the 1958–1987 tumour registry data by Little and Muirhead (2000) used a linear-quadratic model to assess possible overestimation of low-dose risk based on use of a linear dose–response model with these data, taking into account random errors in DS86 neutron and gamma dose estimates, and systematic errors in Hiroshima neutron dose estimates. They concluded that for all solid tumours combined, there was some indication of upward curvature over the 0–2-Gy dose range, but they expressed some caution about the interpretation of the data. As mentioned at the beginning of Section 2.4.4, the more recent analysis by Preston et al. (2004) of LSS solid cancer mortality data using the DS02 dosimetry found a statistically significant quadratic component of the dose–response relationship over the dose range 0–2 Gy, but the authors did not recommend using the linear-quadratic model for estimating risk at low doses. In particular, they noted that ongoing analyses of LSS incidence data did not indicate significant upward curvature, and that progressive trimming of the solid cancer mortality dose–response data from the right did not lead to decreased linear model estimates, as it did in the case of leukaemia.

(65) A DDREF would not be applied to the estimated linear-quadratic dose–response relationship for leukaemia, as it is already included in the model.

Site-specific differences 

(66) The analyses of Fig. 2.3, Fig. 2.4, Fig. 2.5 are based on the numerous data for all solid cancers combined, and that of Fig. 2.6 is based on female breast cancer for which the radiation-related signal-to-noise ratio is high in the sense that dose-specific, radiation-related risk tends to be high compared with the level of, and unexplained variation in, age-specific baseline breast cancer rates. Risk estimates for thyroid cancer and leukaemia are based on far fewer cases, but signal-to-noise ratios tend to be high on a dose-specific basis, especially for exposures at young ages. For these three cancer types, there is evidence of radiation-related excess risk at doses below 200 mGy, and for all except leukaemia, there is little evidence for departure of the dose–response relationship from linearity. For most other cancer sites, however, numbers of cases and/or radiation-related signal-to-noise ratios are too low to support strong statements about low-dose risk, although there is little or no evidence of departure from linearity (Thompson et al., 1994).

(67) The latter category of cancers includes some sites for which there is little or no epidemiological evidence that radiation exposure either is or is not associated with increased risk; examples include small intestine, prostate gland, testes, female genital organs other than ovary, malignant melanoma and squamous cell skin cancer, and chronic lymphocytic leukaemia (NCI/CDC, 2003, UNSCEAR, 2000). In the most recent analysis of cancer mortality among the atomic bomb survivors (Preston et al., 2003), rectal cancer mortality was not associated with radiation dose among men, based on 172 deaths during 1950–1997 and linear model estimates of ERR/Gy=−0.25 (90% CI <-0.3–0.15) for exposure at 30 years of age in a model with no dependence upon attained age, but was positively and significantly associated with dose among women, based on 198 deaths [ERR/Gy=0.75 (CI 0.16–1.6), exposure at 30 years of age]. In addition, rectal cancer, bone cancer, and soft tissue sarcoma have been shown to be significantly associated with high-dose, partial-body exposure among patients given radiation therapy (Boice et al., 1988, UNSCEAR, 2000). Cancer of the small intestine, which is very rare in most populations (Parkin et al., 2002), can be induced in experimental animals by high-dose irradiation of exteriorised intestinal loops (Osborne et al., 1963, Watanabe et al., 1986), and the small intestine is therefore a susceptible organ. However, the small intestine appears to have characteristics that render it highly resistant to carcinogenesis at low-to-moderate levels of exposure to radiation and other environmental carcinogens (Cairns, 2002, Potten et al., 2002; see Section 5.2.1). Thus, inferences based on all cancers as a group, or on certain cancer sites for which there is substantial information about the dose-response relationship and its modification by other factors, need not necessarily apply to all site-specific cancers, or even to all histological subtypes of cancers of any given site. Nevertheless, for those cancers clearly inducible by radiation exposures under 5 Gy, there is evidence of some degree of commonality with respect to dose effects and their modification by sex and age (Pierce and Preston, 1993), and it is therefore useful and informative to examine radiation-related risk for certain groups of cancer sites.

2.5. Thresholds vs the linear, non-threshold theory 

(68) The LNT theory (Brenner and Raabe, 2001) is part of the current basis for risk-based radiation protection. The theory assumes proportionality between radiation dose and subsequent cancer risk, usually with allowance for a DDREF to reduce risk per unit dose of low-LET radiation at dose levels below 200 mGy (ICRP, 1991). However, at doses at which the DDREF applies fully, excess risk is assumed to be proportional to dose. A consequence of the LNT theory is that exposures resulting in very small average doses to very large populations are assumed to be associated with excess numbers of cancers that, although undetectable by epidemiological study, may be numerous.

(69) The threshold theory is a competing theory that, if generally accepted, may make it easier to ignore possible consequences of very-low-dose exposures. According to the theory, there is some ‘threshold’ dose below which there is either no radiation-related health detriment or a radiation-related health benefit that outweighs any detriment. If the threshold was a universal value for all individuals and all tissues, a consequence of the theory is that, at some point, a very low dose to any number of people would have no associated risk and could be ignored. Much, of course, depends upon the value of the assumed threshold dose, since even under the LNT theory, there must be a level of estimated risk so low that it is not worth the trouble to avoid. If, however, thresholds existed but were known or believed to differ widely among individuals and/or tissues, the effect of this knowledge on radiation practice and philosophy may be much less, and radiation protection may be even more complex than it is under the LNT theory.

(70) One argument made against the LNT theory is that there is little or no direct epidemiological evidence of excess cancer risk in populations exposed to less than 50 mGy or so. That is not quite true, as discussed above, but it is true that there is no direct, credible epidemiological evidence of a radiation-related risk associated with exposures of the order of 1 mGy, for example. Nevertheless, as also discussed above, the argument is specious; failure to detect a risk that (if it exists) is very small is not evidence that the risk is zero.

(71) A more subtle, and statistically more sophisticated, argument is to demonstrate that a dose–response model with a threshold, such as a linear model for dose-specific ERR with a fitted negative intercept at zero dose, can fit a data set as well as a linear or linear-quadratic model constrained to have a zero intercept (Hoel and Li, 1998; Little and Boice, 1999). The approach has the potential for showing disproportionality between excess risk and dose, consistent with a threshold (and usually, but not necessarily, also consistent with a linear-quadratic dose–response relationship), and could conceivably provide more substantial evidence of a threshold. That strong support for a threshold is hardly ever found in this way is more a reflection of low statistical power in the low-dose region than of statistical evidence against the existence of a threshold. In a more recent paper, Baker and Hoel (2003) modified the then-current DS86 atomic bomb doses for presumed systematic error in estimates of the neutron component of dose from the Hiroshima bomb, and a dose-dependent relative biological effectiveness for neutrons compared with gamma rays, finding that an improved fit to morbidity data for solid cancers and leukaemia was obtained by introducing a threshold. However, their assumptions about underestimation of the neutron dose for low-dose survivors of the Hiroshima bombing, on which their conclusions depended, have not been borne out by subsequent measurement data (Preston et al., 2004, Straume et al., 2003).

(72) It is clear that epidemiological studies are very unlikely to establish the presence or absence of a threshold at some low-dose level, although they can place limits on the likely value of any possible threshold (Pierce and Preston, 2000). Radiobiological evidence presented elsewhere in this report identifies the induction of DNA DSBs and more complex clustered DNA damage as (probably) the most important mechanism by which IR exposure contributes to radiation carcinogenesis. Such events have been demonstrated by calculation (Brenner and Ward, 1992, Goodhead, 1994) and by experiment (Boudaiffa et al., 2000a, Boudaiffa et al., 2000b) to result from a single low-energy electron track produced by an x-ray or photon interaction. At low doses and low dose rates, the occurrence of such events is proportional to radiation dose and to the number of cells irradiated (Kellerer, 1985). Current research on development of timely assays for the presence and repair of DSBs may lead to findings that resolve the question of low-dose thresholds vs the LNT theory. As discussed in Section 4.5, the answer is still very much in doubt.

2.6. Conclusions: implications for low-dose cancer risk 

(73) Epidemiological data from studies of human populations exposed to IR provide direct evidence that such exposure is associated with increased risk of cancer, and reason to believe that excess risk is not confined to people exposed to very high radiation doses. Our knowledge of radiation-related risk is highly quantified, more so than for any other common environmental carcinogen, and we have learned much about factors that modify that risk. Our understanding of risks associated with doses commonly encountered in daily life is not insignificant; we know, for example, that such risks are far lower than those observed in populations exposed to hundreds or thousands of mGy. However, the problem of quantifying risks that are so low as to be practically unobservable, and then recommending policies based on that quantification, is very difficult.

(74) It is highly likely that there will always be uncertainty about the risk of low doses, and that we will have to come to terms with that uncertainty. One way to do that is to quantify the uncertainty in a manner consistent with mainstream scientific information, and to evaluate actions and policies in terms of plausible probability distributions of risks associated with these actions and policies. An example of this type of approach is given in Chapter 6.