5. What are the health effects of exposure to ionizing radiation?
The SCENIHR opinion states:
3.6 Health effects
3.6.1 Types of health effects
Deterministic effects (or tissue reactions) of ionising radiation are related directly to the absorbed radiation dose and the severity of the effect increases as the dose increases. A deterministic effect typically has a threshold (of the order of magnitude of 0.1 Gy or higher) below which the effect does not occur. Deterministic effects are based on tissue damage.
However, deterministic effects of ionising radiation do not need to be considered as a health hazard at the low doses delivered by X-ray scanners based on the threshold dose recommendation in ICRP publications 60 and 103 (reviewed in ICRP 2011 and Wrixon 2008) - that ‘in the absorbed dose range up to around 100 mGy (low LET or high LET) no tissues are judged to express clinically relevant functional impairment. This judgement applies to both single acute doses and to situations where these low doses are experienced in a protracted form as repeated annual exposures’. According to ICRP publication 103 (ICRP 2007), the threshold for deterministic effects following pre- and post-natal exposure is proposed to be >100 mGy and this judgement for acute doses has been ratified by ICRP (document in consultation 2012). Radiation-induced malformations are considered by ICRP 103 (ICRP 2007) to have a dose-threshold of ~100 mGy.
Lens opacities induced by ionising radiation have been traditionally regarded as a deterministic effect with a threshold exceeding 1 Gy. Recently, several studies have demonstrated lens opacities at dose levels around 100 mSv, but extrapolation down to microsievert dose level would not be meaningful.
Stochastic effects of ionising radiation are chance events, with the probability of the effect increasing with dose, but the severity of the effect is independent of the dose received. Stochastic effects are assumed to have no threshold. Primarily cancer risk, but also hereditary disorders are stochastic effects with a combined detriment of ~5%/Sv (ICRP publication 103 (ICRP 2007)). Hereditary effects of radiation (germline mutations induced by radiation that are transmitted to the offspring and may result in congenital anomalies or increased risk of common multifactorial disease) are not considered here, because they have not been observed in human populations with higher doses (and any theoretical risk would be obscured by the vastly higher spontaneous mutation rate).
3.6.2 Biological effects
DNA double strand breaks (DSB) play a critical role in the carcinogenesis process. DSB induced by low Linear Energy Transfer (LET) radiation in mammalian cells shows a linear dose dependence down to the lowest measured dose of 1 mGy (Leatherbarrow et al. 2006, Rothkamm and Löbrich 2003) and in vivo to doses as low as 100 mGy (Löbrich 2003). The shape of the dose-response curve is in support of the Linear No Threshold (LNT) model (most recent ICRP 103 (ICRP 2007)) down to the lowest experimental doses of 1 mGy. The shape of the dose dependence curve for DNA damage induction at doses much lower than 1 mGy is unknown. Therefore, the present assumptions on the use of LNT for radiation protection are based on the linear extrapolation and the inability to measure biological changes at doses of a few µGy.
The repair of DSB after a dose of 1 mGy in human fibroblasts and in tissue samples from 10 mGy irradiated mice is compromised even up to 24 h post-irradiation, whereas at higher doses DSB rejoining occurs (Grudzenski et al. 2010). At these low doses, a single DSB will only be formed on average in ~1 in 20-30 of the cells exposed, with even fewer formed (<1 DSB per 2-3 x 104 cells) at doses of a few µGy, unless as yet unknown processes occur at ultra-low doses. Based on the present knowledge and the inability to measure biological changes at doses of a few µGy, the present assumptions based on the LNT model for radiation risk estimate remain valid.
Experimental data (Nakano et al. 2007) indicating that chromosomal aberrations do not persist after in uteroirradiation with a dose of 1-2 Gy may also relate to the recent epidemiological data finding that pre-natal exposures are of lower risk.
Although the relative biological effectiveness (RBE) for various end points, based on in vitro cellular radiobiology, increases with decreasing photon energy relative to 60Co gamma rays at doses >1 Gy (Hill 2004), these findings question the recommended use of a wR value of 1 for all photon energies as recommended in ICRP publication 60 (ICRP 1991) [Missing from reference list – Judy Burns]. However, it was concluded from epidemiology studies that it is not statistically feasible to draw any conclusions of an underlying dependence of cancer risks for thyroid or breast on Linear Energy Transfer (LET) for radiation with photon energies less than that for60Co-radiation (Hunter and Muirhead 2009). The epidemiological findings are compatible with the use of a wR value of 1.
126.96.36.199 Non-targeted effects
New findings on non-targeted effects such as bystander effects and genomic instability could affect the Linear No Threshold (LNT) model. The European Integrated Project NOTE (2006-2010) addressed whether the effects of ionising radiation, characteristically associated with the consequences of energy deposition in the cell nucleus, arise in non-irradiated cells and are relevant for the use of the LNT model in extrapolation to low dose to estimate risk (the final report is available from: https://ssl.note-ip.org/documentindex.asp?id=3089&type=1&show=1
). The majority of the studies were carried out at doses of 10 mGy or higher. Taking into account concerns relating to the LNT model of radiation protection recommended most recently in ICRP publication103 (ICRP 2007), it was concluded from the NOTE studies that no compelling evidence for non-targeted effects requires modification of the LNT model for risks to human health. Based on present knowledge of non-targeted phenomena, their incorporation into radiation protection for sparsely ionising radiation (such as X-rays used in both types of security scanners using backscattered or transmission X-rays) is premature in the absence of direct evidence of relevant health endpoints (Averbeck 2010, Goodhead 2010). The majority of non-targeted effects have only been seen at doses >1 mGy for low LET radiation.
188.8.131.52 In vivo models
Animal models are well established methods for improving understanding of ionising radiation induced carcinogenesis. To date the majority of findings on radiation carcinogenesis using mouse models have been obtained at high doses of low LET radiation with a few studies using doses extending down to around 50 mGy (Munley et al. 2011, Pazzaglia et al. 2009). The incidence of tumorigenesis on radiation dose is linear in the range 50-500 mGy (Pazzaglia et al. 2009, Shuryak et al. 2011). A direct dose rate effect was seen with reduced incidences for a dose rate of 0.01 Gy/day (Shuryak et al. 2011). The radiation doses used in this study are about 10 times greater than those estimated for a human scan with a backscatter scanner. As for the epidemiology data with animal models in the mGy range, a high number of studies/large sample sizes are required to obtain statistically significant data. Additionally, the incidence of carcinogenesis at the lowest doses around 50 mGy approaches the spontaneous levels in these mouse models. Data from animal models are not available in the µGy range.
Epidemiological evidence regarding the health effects of low-dose radiation has been obtained from numerous studies since the mid-20th century. Studies informing about health risks from radiation have covered various sources and circumstances of exposure, including environmental, medical and occupational radiation exposures. Population-level studies in humans have demonstrated a dose-dependent increase in cancer risk, with consistent findings in different populations. Ionising radiation can induce most, but not all cancer types. The latency from exposure to the occurrence of excess cancer is typically approximately one decade, but shorter (2-5 years) for leukaemia and thyroid cancer. The elevated risk appears to persist for several decades.
In ICRP publication 103 (ICRP 2007), the cancer risk for prenatal exposure was judged to be similar to that following irradiation in early childhood. However, recent evidence (Preston et al. 2008) on in uteroexposure to ionising radiation indicates that the lifetime risks of solid cancers (but not leukaemia) are considerably lower than those for exposure in childhood, i.e. at most, about three times that of the population as a whole.
The disease risk caused by radiation can be expressed in terms of relative risk (RR), i.e. as a multiple of the underlying disease risk. A RR of 1 indicates a similar occurrence in the study population as in the reference group, while for instance an RR of 1.5 shows a relative increase by 50% (i.e. 1.5-fold occurrence). Such a relative risk model inherently assumes that the health effect is proportional to the baseline risk of the population, i.e. whether the disease risk due to other factors such as age is low or high, the exposure would always increase the risk as a multiple of the baseline. Alternatively, an absolute risk model can be used to depict the effect of an exposure with a given increase in occurrence. Here, a constant absolute increase in risk is assumed, independent of the baseline risk. As an example, a cohort study might report cancer incidence of 150 per 100,000 person-years in an unexposed group and 200 per 100,000 among subjects with radiation exposure. The relative risk (rate ratio) for the exposed cohort is then 1.33 (200/150), while the absolute risk (excess incidence) is 50/100,000 (200/100,000-150/100,000). Adopting a relative risk model would imply that the effect of a similar exposure in any other population would result in 1.3-fold increase, whereas extrapolation using an absolute risk (or absolute effect) model would predict an increase by 50/100,000.
The major sources of uncertainty in these studies have included exposure assessment (dosimetry), exposures from other sources, and effects of other factors on disease risk (confounding). Typically, highest quality dose estimates have been available for studies assessing the effects of medical uses of radiation and lowest for studies on environmental exposures. On the other hand, other sources of uncertainty can play a large role.
Uncertainty can be divided into two types: stochastic and epistemic. Stochastic uncertainty is related to the observations available and can be reduced by increasing the amount of data. Statistical variability or random error is an example of stochastic uncertainty. Epistemic uncertainty pertains to models and assumptions involved in interpreting the data. It is not related to the amount of observations. For instance, the validity of an experimental model is typically epistemic uncertainty. If for instance DNA breaks are not essential for cancer risk from ionising radiation, conducting more studies using such an approach will not improve the knowledge due to epistemic knowledge being a limiting factor.
Stochastic uncertainty, i.e. random error, is a limitation of the ability of the epidemiological studies in particular when dealing with small effects. Random error is the variability unrelated to exposure of interest and can be thought of as the background noise against which the phenomenon of interest needs to be distinguished. The capacity to demonstrate either the presence or absence of an effect is called statistical power. It depends on the amount of information available, which is related to the study size, exposure distribution and disease risk. Roughly, the higher the number of events and the more evenly they are distributed across the compared groups, the higher the statistical power.
Besides random error, the quality of the results of a study depends on systematic error, i.e. bias and confounding. Bias is distortion of information that (unlike random error) is related to the phenomenon under study, either exposure (potential determinant of disease risk studies) or the outcome (disease status). The major types of bias are information bias and selection bias. The former has to do with the availability or quality of information (differences in extent or quality of exposure data between those with and without the health outcome, or differences in outcome data between exposure groups). Selection bias occurs when the inclusion in the study differs from the ideal or intended in such a way that it distorts the comparability of the groups within the study. Selection bias can occur if two groups differ from each other in terms not only of exposure being studied, but also other factors affecting the disease outcome. The so-called healthy worker effect is an example of selection bias. Healthy worker bias occurs due to the fact that employed people have generally better health than those who are not working (as some of them may have retired due to an illness, or their health may have deteriorated because of unemployment), which results in lower mortality in several occupational groups compared with the general population. Also, patients undergoing medical interventions such as diagnostic X-rays or radiotherapy may differ from the healthy subjects in terms of risk of cancer or other diseases (because they are selected for the intervention based on a suspected or diagnosed health condition).
Confounding is the distorting effect that other risk factors may cause on the exposure-outcome relation of the studies. For example, studies on areas with elevated rates of background radiation may suffer from confounding if the population in other nearby areas that they are compared with differs also in other respects such as lifestyle factors relevant for cancer risk (e.g. smoking, diet, physical activity, infections etc.).
The single most important source of epidemiological knowledge on health effects of radiation has been the Life Span Study of the atomic bomb survivors in Hiroshima and Nagasaki. A wide range of doses (from several Sv down to 5 mSv), good quality dose estimates, a large study population covering a wide age span and long follow-up with information on both cancer incidence and mortality increase the amount and quality of evidence from the study. Among atomic bomb survivors, a significant dose-response relationship is seen in the dose range 0-150 mGy for solid cancers and the existence of a threshold (below which no effect is seen) can be excluded at 85 mGy or higher (but not below).
The effect of radiation on cancer risk is not uniform across the population; it is modified by some factors. First, cancer risk following radiation exposure at a young age is higher than for exposures later in life (although this effect can also be explained in terms of age attained). Exposure at a young age generally tends to result in a larger relative effect than at older ages. In the atomic bomb survivor studies, the excess relative risk per Gray (ERR/Gy) for all solid cancers decreased by 17% per decade of age at exposure (90% CI -25%, -7%). In terms of absolute excess risk, the decrease per one decade increment in age was -24% (90% CI -32, -16). Alternatively, the effect of age can be expressed in terms of attained age, i.e. in relation to the risk at a given age during follow-up (age at observation), with an equally good fit with the observations. The cancer types where this effect is very pronounced include thyroid cancer, leukaemia and breast cancer. Also, the risk coefficients tend to be slightly higher for women than men. Among atomic bomb survivors, the ERR coefficients at 1 Gy for incidence of all solid cancers for women have been larger by a factor of 1.6 compared with men (ERR of 0.35/Gy for men and 0.58/Gy for women) (Preston et al. 2007). This may reflect more the difference in background rates than sensitivity to radiation effects. A smaller difference (female:male ratio of 1.4) is found in absolute excess risk (43 versus 60 excess cases per 1,000 person-year-Gray), and it decreases further, when the gender-specific cancers (breast, prostate and gynecological) cancers are excluded.
Extensive research on cancer risk related to low doses of radiation (in the mSv range) received from occupational exposure, medical diagnostic procedures and in areas with an elevated natural background radiation has been conducted during the past decades. Some of the key findings are summarized in the following section.
A meta-analysis of leukaemia risk from low-dose exposures combined the results of 10 studies (mainly on occupational exposures) and showed a pooled risk estimate of ERR 0.19 (95% CI 0.07-0.32) per 100 mGy (Daniels and Schubauer-Berigan 2011).
A systematic review of cancer risk from diagnostic X-rays showed no clear excess from nine case-control studies of prenatal exposure published after 1990 (OR 0.99, 95% CI 0.87-1.13), though it did not include the early Oxford Survey (Schulze-Rath et al. 2008).
A recent large case-control study found no significant excess of all cancers (OR 1.14, 95% CI 0.90-1.45) or leukaemia (OR 1.36, 95% CI 0.91-2.02) associated with any diagnostic radiation in utero(Rajaraman et al. 2011). Also, a cohort study with 5,590 pregnant women who had been exposed to ionising radiation for diagnostic purposes showed no clear excess cancer incidence (HR 0.68, 95% CI 0.25-1.80 based on four childhood cancers) (Ray et al. 2010). A German cohort of more than 78,000 children who had undergone diagnostic radiographic examinations also showed no excess of childhood cancer (RR 0.97, 95% CI 0.75-1.23), or trend across dose categories (Hammer et al. 2011).
Studies in high natural background areas in India and China have not been able to show elevated cancer rates when comparing populations with annual doses of around 1 mSv versus 4 mSv (and cumulative doses up to several hundred mSv) (Nair et al. 2009, Tao et al. 2012).
The results of these studies do not of course exclude the existence of a health effect in the mSv dose levels. They are indeed compatible with risk estimates from studies of higher doses and mainly indicate that risks at low doses are not materially larger than predicted from high-dose studies.
Epidemiological studies have not found major differences in health risks from ionising radiation between subgroups of the population defined by hereditary factors. Among patients receiving radiotherapy for retinoblastoma, a childhood tumour of the eye, those with the hereditary bilateral form of the disease have a higher risk of secondary sarcoma. Breast cancer patients who are carriers of the rare missense variant form of the ataxia telangiectasia gene have been shown to be at an increased risk of contralateral breast cancer following radiotherapy compared with other patients receiving radiotherapy for their first breast cancer.
Epidemiological studies have not provided consistent evidence regarding a lower risk from radiation exposure occurring over an extended period of time compared with similar doses received at higher dose rates. A pooled analysis of 12 epidemiological studies of occupationally exposed groups (Jacob et al. 2009) [Missing reference not included in ref list – Judy Burns] did not find evidence of lower cancer risk related to protracted rather than acute exposure.
The effects of ionising radiation on the risk of cardiovascular disease have been shown in the past 20 years. Radiotherapy at high doses (>10 Gy) to the heart increases the risk of cardiac disease, with radiation-related heart disease (such as pericarditis, valvular disease or cardiomyopathy as direct result of radiation) from the dose level of several Gray upwards emerging after a minimal latency of 1-2 years (although acute pericarditis may develop as soon as some weeks after). Among atomic bomb survivors, there is a dose-response relationship in late cardiovascular disease mortality, including both heart disease and stroke after at least a decade (Shimizu et al. 2010). Such an effect could, however, be neither confirmed nor excluded at dose levels below 0.5 Gy. In some occupational cohorts an increased risk of cardiovascular disease in relation to radiation dose has also been suggested, but the possible effect of confounding has not been ruled out.
The strength of the epidemiological studies is their direct relevance for risk assessment – they deal with actual disease and exposure to agents as it occurs in real life without the need for extrapolation from species, dose levels or outcomes to another. Direct inference counterbalances the uncertainties usually encountered in epidemiological studies, particularly non-randomised studies.
The ability of epidemiological studies to demonstrate (or exclude) small health effects is limited by the uncertainties and sources of error outlined above. Common non-infectious diseases such as cancer and cardiovascular disease result from long multi-factorial processes. Such complex diseases have multi-factorial etiology. A malignancy caused by exposure to ionising radiation cannot be distinguished from tumours due to other factors. For instance (long-term occupational) radiation exposure with a cumulative dose of 200 mSv may result in 1.1-fold cancer risk. However, it is impossible to tell which of the cancers occurring in such a population are attributable to radiation and which are caused by other factors. The effects of very low radiation doses, say below 100-200 mSv, are very difficult to demonstrate in epidemiological studies. In order to include such small incremental risks as evidence, very accurate information on exposure (with minimal random error and bias) would be needed. In addition, exposure from other sources including natural background radiation would need to be known. Furthermore, the baseline risk due to other factors (confounding factors) would need to be very well characterized. Finally, comprehensive information on all disease cases should be available. In practice such ideal circumstances are not possible. Even in a very large study of 100,000 subjects followed up for cancer incidence for 10 years (after the 10 year latency period), the expected number of cancer cases might be of the order of 2,000. The effect of 100 mSv could be expected to induce 100 additional cases. Such a small increment would be easily missed due to random error – it can be calculated that if 100 such studies were carried out, only just over half (approximately 60) would be able to show an effect assuming that a comparable cohort of unexposed people was available (not considering bias and confounding).
3.6.4 Extrapolation to low doses
Overall, the doses from X-ray scanners are so low that the biological effects both in cellular and in vivo models cannot be experimentally determined or quantified. The dose is in the range classified as a negligible individual dose by NCRP Commentary No. 16 (NCRP 2003). The cumulative effective dose from a whole body X-ray backscatter scanner to a person who uses air travel daily is small relative to the control level of 0.25 mSv y-1 recommended by NCRP Commentary No. 16 (NCRP 2003) .
In view of the low doses from security scanners there is no scientific basis to separately consider potentially vulnerable groups (e.g. pregnant women, children) in risk assessment. This is due to the much larger uncertainties in risk estimates relative to the variation of risk between subgroups of the population, i.e. even the potential of µSv-level doses to induce any health effects is uncertain, while the differences in risk between population subgroups are within one order of magnitude and demonstrable only at dose levels exceeding 100 mSv.
Recently, the United States National Academy of Sciences and National Research Council published an evaluation of health risks from low doses of ionising radiation (BEIR 2006) (Biological Effects of Ionising Radiation, BEIR Committee report VII, 2006). Based on atomic bomb survivor data on cancer incidence and mortality, complemented with data from medically exposed population for breast and thyroid cancer, risk predictions were made for doses below 100 millisievert. Based on the conclusion of a review of both biological and biophysical studies on mechanisms of radiation-induced cancer, it was concluded that the cancer risk is likely to occur in direct proportion to dose (with a linear dose-response curve/relationship) even at lowest doses without a threshold, even if the risks would be very low. The lifetime risk model developed predicts that for a radiation dose of 100 mSv, one additional cancer case (including both solid cancers and leukaemia) would be expected to occur per 100 exposed persons (against a background of 42 cases unrelated to radiation). Correspondingly, one additional cancer case would be expected per 1,000 people exposed to 10 millisievert during their remaining lifetime. The number of excess cancer deaths due to radiation would be approximately half of the incident cancer cases. Individual risk versus population risk has been proposed (Brenner 2011) as ‘one of the means of assessing the acceptability of a facility or practice’ NCRP Commentary No. 16 (NCRP 2003) [Query – should this be cited as a reference and included in the ref list? Judy Burns]. Probability of an adverse effect due to radiation exposure from whole body X-ray backscatter scans is likely to be of the same order as the negligible individual risk level (NIRL) of 10-7 -1 given in NCRP Report No. 91 (NCRP 1987), corresponding to an effective dose equivalent of 0.01 mSv. The recommendation in NCRP Report No. 91 (NCRP 1987) that assessments of increments of collective annual dose from any particular individual source or practice should exclude those individuals whose annual effective dose equivalent from such sources was ≤0.01 mSv was withdrawn by the NCRP and superseded by NCRP Report No. 116 (NCRP 1993). A negligible individual dose, defined as an annual dose value for a particular radiation source or set of sources is described in NCRP Report No. 116 (NCRP 1993). The negligible individual dose was set at 10 µSv, corresponding to the effective dose per scan from at least 50 whole body X-ray backscatter scans. NCRP Commentary No. 16 (NCRP 2003) recommended that the cumulative effective dose to an individual member of the public from such X-ray systems used in security screening of humans should not exceed a control level of 0.25 mSv y–1, and for an individual scan the effective dose should be ≤0.1 µSv (discussed in Schauer 2011).
For an extremely small individual risk, Brenner (2011) suggested the population risk is also negligible but not zero, although it was acknowledged that many uncertainties exist in the estimation of individual risk. ICRP publication 103 (ICRP 2007) does not recommend the use of collective effective doses (population doses) over long time periods as an appropriate management approach to make decisions and in particular, the calculation of the number of cancer deaths based on collective effective doses from trivial individual doses should be avoided.
Some examples of health risk assessment related to low doses of radiation below the level which can be directly observed in epidemiological studies are described below, even though these projections utilising theoretical calculations pertain to exposure levels higher than those received from body scanners. They suggest that radiation doses in the mSv range could be expected to increase the occurrence of cancer by an order of magnitude of 1%. As the radiation doses from body scanners are several orders of magnitude lower, the risks can also be assumed to be smaller.
To assess cancer risk attributable to radiology, Berrington de Gonzáles and Darby (2004) used linear excess absolute and relative risk models based on Japanese atomic bomb survivors. They took into account age at exposure (for breast cancer and leukaemia also age attained), frequency of diagnostic X-rays (nine types of radiographic, eight fluoroscopic and 10 CT examinations in 1991-96, with older British data on age and sex distribution of patients), organ doses and cancer incidence (specifically, leukaemia, and oesophagus, stomach, colon, liver, lung, bladder, and thyroid cancers) in 15 countries. Risks were projected for all other cancer sites (excluding lymphoma, multiple myeloma, and chronic lymphocytic leukaemia) assuming a similar risk coefficient. The estimated annual average radiation doses to various organs were below 1 mGy and projection carried out assuming that there is no threshold below which cancer risk would disappear. The results indicated that roughly 0.6% of the lifetime cancer risk in the UK might be attributable to diagnostic radiological examinations, with higher estimates for most other countries. More conservative assumptions regarding the duration of effect, mortality in the exposed patients relative to the general population and using a dose and dose-rate effectiveness factor (DDREF of) 2 decreased the effects by 10-50%. The long-term impact of the Chernobyl fallout on cancer incidence was predicted by Cardis and co-workers (Cardis et al. 2006). They used both excess relative risk (ERR) and excess absolute risk (EAR) models to project risks from doses of the order of 0.5 mSv received over 20 years time. The overall estimate was 0.01% increase in overall cancer incidence in Europe and 1.5% excess of thyroid cancer. The goal of the theoretical calculation was mainly to provide an indication of the order of magnitude of possible effect (2,400 cases in a population of 572 million over eight decades). The authors noted the need for caution when applying risk models developed based on different populations exposed to single high doses to circumstances of very low cumulative doses delivered over decades.
The contribution of background radiation to leukaemia risk was recently estimated. Based on annual doses of 1.2 mSv and the risk model derived by UNSCEAR from atomic bomb survivors, it was estimated that approximately 4% of all cases of leukaemia could be attributable to natural background radiation, and the proportion would be larger for childhood leukaemia (5-19%, depending on the risk model) (Kendall et al. 2011).