Effective doses and risks from medical diagnostic x-ray examinations for male and female patients from childhood to old age

The cancer risk models developed by ICRP for application in its 2007 Recommendations are based primarily on cancer incidence data from the life span study (LSS) of the Japanese atomic bomb survivors, with follow up from 1958 to 1998 for solid cancers and from 1950 to 2000 for leukaemia (ICRP 2007, 2022, Cléro et al 2019, 2022, Ban et al 2022). Models based on the excess absolute risk (EAR, the additional risk due to the exposure) and excess relative risk (ERR, the proportional increase in risk compared to the background risk) were developed that related EAR and ERR to organ/tissue-specific absorbed dose using the DS02 dosimetry system for the following ten specific organs/tissues: female breast, lung, stomach, colon, red bone marrow (RBM), bladder, liver, thyroid, oesophagus and ovary. The nominal risk estimates from the 1990 Recommendations (ICRP 1991) were used for bone and (non-melanoma) skin cancers because LSS data for these cancers offered only a limited opportunity for the derivation of risk models (ICRP 2007, 2022, Cléro et al 2019, Ban et al 2022). EAR and ERR models were developed from the LSS incidence data for the remaining solid cancers considered as a single group; risk estimates were not derived for the lymphomas or multiple myeloma, for which evidence of any radiation-related risk is limited. For solid cancers, the models are linear, no-threshold dose-responses, allowing for risk modification by sex, age-at-exposure and attained age. For leukaemia following irradiation of RBM, ICRP (2007) used an EAR model with an explicit linear-quadratic dose-response that allowed for the modifying effects of sex, age-at-exposure and time-since-exposure, similar to that described by Preston et al (1994); ERR estimates were derived from this EAR model using the LSS background leukaemia incidence rates (Cléro et al 2019, 2022, ICRP 2022, Ban et al 2022). Minimum latency periods were set at two years for leukaemia and five years for solid cancers (Cléro et al 2019) (although it later transpired that the latency period used for leukaemia was actually five years (Cléro et al 2022)).

EAR and ERR models lead to similar predictions of the excess risks of cancer in the population that was used to derive the risk models, but they can lead to markedly different excess risk estimates when applied to other populations with different baseline cancer rates. ICRP (2007) used baseline cancer incidence rates for two fixed composite populations (based on populations with well-established cancer registries) defined by averaging, respectively, Asian (Shanghai, Osaka, Hiroshima and Nagasaki) and Euro-American (Sweden, UK and USA) populations for specific periods; leukaemia baseline incidence rates excluded chronic lymphocytic leukaemia (CLL, now considered to be a form of non-Hodgkin lymphoma) (Cléro et al 2019). Age-averaged lifetime cancer incidence risks were calculated for each composite population and sex by applying the EAR and ERR models to the databases using a single dose of 100 mGy of low-LET radiation delivered to each organ/tissue, and a weighted average of the model estimates was then computed. The relative weights given to the EAR and ERR models varied for the different cancer sites, based on judgements concerning their relative applicability for risk transfer, as follows: EAR:ERR of 0.0:1.0 for thyroid and skin, 1.0:0.0 for breast and bone, 0.7:0.3 for lung, and 0.5:0.5 for all other cancers. For solid cancers, ICRP (2007) applied a dose and dose rate effectiveness factor (DDREF) of 2 to reduce risk estimates for normal applications at low doses (<∼100 mGy) or low dose rates (<∼5 mGy h⁻¹) ⁴ to organs and tissues, but no DDREF was used for leukaemia. The resulting lifetime risk estimates were further averaged over composite population and sex to produce final nominal risk coefficients consisting of sex-, age-, population- and model-averaged (weighted as appropriate) lifetime excess absolute risks of cancer incidence (LEAR, the EAR of cancer incidence over the remaining lifetime) per Gy for the ten sites, and one grouping of sites (other solid), of cancer; nominal risk coefficients were additionally presented for bone and skin cancers (ICRP 2007). Details may be found in Cléro et al (2019, 2022), Ban et al (2022) and ICRP Publication 152 (ICRP 2022). ICRP (2007) defines 'nominal' in this context as 'sex-averaged and age-at-exposure-averaged lifetime risk estimates for a representative population'.

Detriment (to health) was calculated by ICRP (2007) from these values of the nominal LEAR of cancer incidence per Gy by adjusting for severity of each cancer type, applying weighting factors for lethality, morbidity associated with non-fatal cancers, and years of life lost. As discussed by Cléro et al (2019), Ban et al (2022) and ICRP (2022), these severity weighting factors were broad evaluations of available data, which involved expert judgement in interpreting limited data for the purpose of these adjustments; for example, lethality fractions were those that had been used for the ICRP 1990 Recommendations (ICRP 1991) and did not vary between the sexes or change when children and the elderly were removed from the age-at-exposure averaging (ICRP 2007). Total cancer detriment was obtained by summing the detriment values for each cancer site. Estimated risks of heritable effects from irradiation of the gonads, derived from animal experiments and knowledge of human genetics, were similarly averaged to derive a nominal value of LEAR per Gy, which was severity weighted to provide a detriment value that was then added to the cancer detriment to provide the overall detriment for stochastic health effects. Tissue weighting factors, w_T, are simplified values chosen to represent relative detriment from each organ/tissue when averaged over both sexes and all ages. For simplicity, only four different values of w_T are used: 0.12, 0.06, 0.04 and 0.01, and Σw_T = 1 when summed over all tissues.

Note that, while nominal risk coefficients are calculated as LEAR per Gy absorbed dose to organs/tissues for the effect resulting from this irradiation, the nominal risk coefficients presented by ICRP (2007) are expressed as LEAR per Sv effective dose by convention as the basis for the use of this protection quantity. Similarly, the detriment is expressed per effective dose. Two overall detriment values were presented in ICRP Publication 103: 5.7 × 10⁻² Sv⁻¹ effective dose for the whole population (both sexes, all ages), and 4.2 × 10⁻² Sv⁻¹ effective dose for a working age population (both sexes and 18–64 yr age-at-exposure). Cancer detriment, for a uniform exposure, dominates at 5.5 × 10⁻² Sv⁻¹ and 4.1 × 10⁻² Sv⁻¹, respectively.

Wall et al (2011) constructed 11 cancer EAR and ERR models from the appropriate LSS cancer incidence data, following the methodology described in ICRP Publication 103 (ICRP 2007) as closely as possible, although later information on this methodology has revealed various aspects of the description provided in Publication 103 that require revision (Cléro et al 2019, 2022, ICRP 2022, Ban et al 2022). The LEAR of cancer incidence estimates were calculated for each organ/tissue receiving a uniform absorbed dose of 10 mGy of low-LET radiation. Comparison of the results of applying the models of Wall et al (2011) with equivalent results presented in Publication 103 show good agreement for most cancer types. However, discrepancies were observed for thyroid cancer (−42%) and leukaemia (+50%), and less so for female breast cancer (−13%), which were largely unexplained (Wall et al 2011), although the differences have, at least in part, been clarified by further elucidation of what was done to derive the ICRP models (Cléro et al 2019, 2022, ICRP 2022, Ban et al 2022). Despite these differences, the overall LEAR of cancer incidence (excluding bone and skin cancers) was calculated by Wall et al (2011) as 665 × 10⁻⁴ Gy⁻¹, (i.e. 665 cases per 10 000 persons per Gy) compared with a value of 688 × 10⁻⁴ Gy⁻¹ presented in Publication 103 (ICRP 2007), results that are within 3% of each other, which is good overall agreement.

In ICRP Publication 103 (ICRP 2007, Cléro et al 2019) LEAR estimates of cancer incidence were calculated using the measure, risk of exposure-induced cancer (REIC) (Thomas et al 1992), which was also the measure used by Wall et al (2011). ICRP Publication 147 (ICRP 2021) presented results for the LEAR of cancer incidence measured as the lifetime attributable risk (LAR), and LAR is also used in this paper, although the difference between the REIC and LAR estimates are generally small, as noted later.

Table 1 shows results obtained for lifetime EARs of cancer incidence calculated as LAR, considering 11 different cancer types: female breast, lung, stomach, colon, bladder, liver, thyroid, oesophagus, ovary, non-CLL leukaemia, and all other solid cancer sites as a single grouping, excluding skin and bone cancers. The cumulative EAR of cancer incidence per organ/tissue absorbed dose up to an attained age of 100 years was calculated separately for males and females and by category of age-at-exposure (ten age-at-exposure 10 yr groups, from 0–9 yr to 90–99 yr). Calculations were performed for the ICRP Euro-American composite population. The equivalent results using REIC as a measure of LEAR are shown in table S1, and are similar to those shown in table 1.

The LAR values presented in table 1 show, when taking the 30–39 yr age-at-exposure group as a reference, LAR estimates about two to three times higher in the youngest group (0–9 yr at exposure) and about two to three times lower at an age-at-exposure of 60–69 yr. A very similar pattern of results is obtained using the ICRP Asian composite population, as illustrated in figure 1, which shows the variation of the LAR of cancer incidence by sex and age-at-exposure for Euro-American and Asian composite populations.

Zoom In Zoom Out Reset image size

However, the results show substantial differences between cancer types (table 1), as illustrated in figure 2 for lung and thyroid cancers. While the LAR of cancer incidence values for most cancer types are greatest at younger ages at exposure, this tendency is more pronounced for some (e.g. thyroid cancer) than others (e.g. oesophageal cancer); lung cancer is a notable exception for which risks peak for exposures in middle age (UNSCEAR 2013, Cahoon et al 2017). Because of these variations, the contribution of the different cancer types to overall lifetime risk varies substantially with age-at-exposure as well as between males and females, and the sex-specific differences in LAR estimates are particularly notable for lung and thyroid cancers. Note that variations with age-at-exposure reflect cumulative lifetime risks of cancer incidence, so that reduction of risk with increasing age-at-exposure reflects mainly the reduction in remaining lifetime after exposure during which the risk may be expressed rather than a variation of sensitivity with age-at-exposure, although the latter variation also plays a role, especially in combination with a decrease in risk with time-since-exposure (UNSCEAR 2013).

Zoom In Zoom Out Reset image size

Applying the methodology described in section 2 to calculate the sex- and age-at-exposure-specific LEAR of cancer incidence per organ/tissue absorbed dose for each of 11 organs/tissues (table 1), and using UK estimates of organ/tissue doses from a range of medical diagnostic procedures, as illustrated in section 3, Wall et al (2011) derived sex- and age-at-exposure-specific LEAR estimates for overall cancer incidence (excluding bone and skin cancers) for each procedure using the ICRP (2007) Euro-American composite population. For each procedure, Wall et al (2011) also calculated the effective dose and divided the sex- and age-specific LEAR estimates of overall cancer incidence by the effective dose to obtain the LEAR of cancer incidence per effective dose for each sex and age-at-exposure group. Wall et al (2011) used the LEAR of cancer incidence calculated as the REIC, as in table S1, with results shown in table S2, whereas ICRP Publication 147 (ICRP 2021) presented results using the corresponding LEAR estimates calculated as LAR, as in table 1, which are shown in table 5. Variations in the LEAR of cancer incidence per Sv effective dose reflect the combination of organ/tissue doses received during each procedure, which can vary considerably between different x-ray examinations. A chest x-ray examination, for example, results in doses to a number of organs/tissues, including the liver and stomach, as well as the lungs and breasts, while a head CT scan mainly delivers dose to the brain.

Figure 3 presents the LAR of cancer incidence for 18 different diagnostic x-ray examinations and the resulting distributions of organ/tissue doses, together with that for a uniform whole-body gamma dose of 10 mGy (i.e. an effective dose of 10 mSv). The variation of the LEAR/Sv with sex and age-at-exposure for uniform whole-body exposure is a consequence of the explicit application of sex- and age-specific risk models in contrast to the sex- and age-averaging of risk model results by ICRP (2007), which led to the relative detriment values that formed the basis of the broad groupings of tissue weighting factors used in the calculation of effective dose (see section 2). For most diagnostic procedures, the estimates of LEAR/Sv are within about ±50% of those for uniform whole-body irradiation for the particular sex and age-at-exposure group. The variation of LEAR/Sv estimates about those for a uniform whole-body exposure reflects the application of organ/tissue-specific risk models to the different organ/tissue absorbed doses delivered in the various procedures, and the extent to which tissue weighting factors represent the sex- and age-specific LEAR/Gy estimates for the variously exposed organs/tissues in particular diagnostic x-ray examinations.

Zoom In Zoom Out Reset image size

By way of comparison, figure S1 shows the results of an equivalent exercise using the ICRP (2007) Asian composite population (ICRP 2021, table 5.4). The patterns of LAR of cancer incidence values are broadly similar to those shown in figure 3 for the ICRP (2007) Euro-American composite population, with differences due to the differences in baseline cancer incidence rates in the two composite populations.

The sex- and age-specific LEAR of cancer incidence per Sv estimates presented in table 5 (as LAR) and table S2 (as REIC), and illustrated in figures 3 and S1, respectively, were obtained by summing the LEAR/Sv estimates for each of the 11 cancer types for which ICRP (2007) risk models are available (see table 1), and depend on the organ/tissue doses received in each procedure. These LEAR/Sv estimates for each cancer type are not weighted by severity, which was done by ICRP (2007) to obtain the detriment values that are the basis of the w_T values used in the calculation of effective dose. Ideally, to provide additional information on lifetime risks, the cancer incidence LEAR/Sv estimates derived here would be weighted by severity, but as noted in section 2, the severity adjustments carried out by ICRP (2007) were broad factors applied to nominal cancer incidence risk coefficients that were the result of sex-, age- and population-averaging for each cancer type. The severity adjustments cannot be disaggregated into specific weightings by sex, age-at-exposure and composite population (Cléro et al 2019, ICRP 2022, Ban et al 2022). As a consequence, the severity adjustment factors presented in ICRP Publication 103 provide little additional information on the variation of cancer risk by sex and age-at-exposure beyond that available from the cancer incidence LEAR/Sv estimates.

For illustrative purposes, for the 11 separate cancer models derived by ICRP (2007), the overall LEAR of cancer incidence per Sv reduces by 21.6% when weighted by severity to produce the equivalent detriment value (from 688 × 10⁻⁴ Sv⁻¹ to 540 × 10⁻⁴ Sv⁻¹). The largest proportional changes to the relative contributions to the overall nominal risk coefficient and detriment are reductions of ∼50% for bladder and thyroid cancers and an increase of ∼87% for leukaemia, but these cancer types each contribute only ∼5%–6% to the overall LEAR/Sv. Otherwise, the relative contributions of cancer types to the detriment are within 15% of the relative contributions to the nominal risk coefficient. Although differences will result from a particular combination of organ/tissue doses and associated severity weightings, substantial changes to the sex/age patterns of LEAR/Sv shown in figures 3 and S1 if cancer types were weighted by (sex- and age-averaged) severity would not, in general, be expected. Nonetheless, as emphasised above, severity weighting factors are broadly based values derived from limited data and applied across the whole population (ICRP 2022), and although severity adjustment by sex, age and population would be desirable, this is not possible from the information available in Publication 103 (ICRP 2007); improvements in the derivation of detriment are currently under consideration by ICRP (2022).

The exclusion of bone and (non-melanoma) skin cancers from the cancer incidence LEAR/Sv estimates of tables 5 and S2 (and figures 3 and S1) will have little impact on the overall conclusions. In Publication 103, bone cancer contributes just 0.4% of the total cancer incidence LEAR/Sv and 0.9% of the cancer detriment for the whole population. While skin cancer accounts for 59% of the total cancer incidence LEAR/Sv, non-melanoma skin cancer is easily treated (often as an outpatient), and once weighting for severity is incorporated, skin cancer contributes just 0.7% of the total cancer detriment. This is reflected in w_T values of 0.01 for both bone surface and skin so that doses to these tissues generally make a relatively small contribution to the effective dose. Further, as noted in ICRP Publication 103, the bone cancer risk coefficient was an overestimate based on average dose of radium-244 to bone volume rather than the appropriate higher dose (x9) to the bone surface.

An additional component of stochastic health risk considered by ICRP (2007) is heritable effects following irradiation of the gonads. For the whole population (both sexes, all ages) ICRP (2007) derived a nominal risk coefficient of 20 × 10⁻⁴ Sv⁻¹ for heritable effects, by averaging a risk coefficient of 54 × 10⁻⁴ Gy⁻¹ gonadal dose for the reproductive population over the population of all ages. Wall et al (2011) adopted a risk coefficient of 50 × 10⁻⁴ Gy⁻¹ for those of reproductive capacity and zero otherwise. ICRP (2007) derived a detriment for heritable effects of 25.4 × 10⁻⁴ Sv⁻¹ for the whole population (an increase of 27% over the nominal heritable risk coefficient) to give a relative detriment of 0.044 (leading to a w_T for the gonads of 0.08, which also included a somatic component for cancer in the exposed individual). Even taking a severity-weighted risk coefficient for a reproductive population of 70 × 10⁻⁴ Gy⁻¹ (i.e. one predominantly of relevance to younger ages at exposure) the contribution of the risk of heritable effects to the total stochastic effects risk is comparatively small when compared to the cancer detriment of 549 × 10⁻⁴ Sv⁻¹ (both sexes, all ages at exposure). However, the contribution would be somewhat greater for diagnostic procedures giving higher gonadal doses, such as CT examinations of the abdomen and pelvis of young females.

It is important that the point estimates presented in table 5 (and table 1) do not impart a false impression of the precision of estimates of cancer risk from low-level radiation exposures. Point estimates are presented here to provide the broad overall pattern of sex and age-at-exposure differences in estimated risks. By way of illustration, other expert groups used essentially the same database as ICRP (the LSS cancer incidence data available in the mid-2000s) to generate cancer risk models and lifetime excess cancer incidence risks. One of these expert groups was the BEIR VII Committee of the US National Academy of Sciences (NRC 2006). The BEIR VII risk models have been modified by the US National Cancer Institute (NCI) for the purposes of producing the online adiation risk assessment tool (RadRAT) (Berrington de Gonzalez et al 2012).

4.5.1. RadRAT

RadRAT (https://radiationcalculators.cancer.gov/radrat/) has here been used to generate cancer incidence LAR/Gy estimates for comparison with the LAR/Gy estimates shown in table 1 that were obtained using the ICRP (2007) cancer risk models. This entailed, for both the ICRP and RadRAT models, using a 10 mGy gamma radiation dose received uniformly by all organs/tissues. For the ICRP models, the Euro-American composite population was used, and for RadRAT the US population during 2000–2005 was selected as an input option. Cancer incidence LAR estimates exclude non-melanoma skin cancer, and the ICRP estimates also exclude bone cancer. Figure 4 presents the results of this comparison, showing that the BEIR VII/NCI cancer incidence LAR/Gy estimates are greater than the ICRP estimates by a factor that approaches 2.5 for the youngest ages at exposure, for both females and males. Important factors in this difference are that RadRAT uses a central estimate of the DDREF for solid cancers of 1.5 whereas the ICRP calculations use a fixed DDREF of 2, and the combination of EAR/ERR model results for most cancers is 0.3/0.7 for BEIR VII/NCI in contrast to 0.5/0.5 for ICRP, but differences also arise from the forms of the risk models used, such as the treatment of modifying factors. This comparison exercise illustrates the considerable uncertainties that arise from modelling and associated decisions, such as the value of the DDREF and the EAR/ERR model ratios used, which must be borne in mind when viewing any one set of risk estimates. Although bone cancer is excluded from the ICRP cancer incidence estimates this will have little impact in comparing the BEIR VII/NCI and ICRP results because bone cancer accounts for only 1% of the overall LEAR/Sv of cancer incidence (excluding skin cancer) in ICRP Publication 103 (ICRP 2007).

Zoom In Zoom Out Reset image size

Further, RadRAT output includes '90% uncertainty ranges' for the cancer incidence LAR estimates, which permits an assessment of the precision of these estimates produced by the BEIR VII/NCI models and associated assumptions. The 90% uncertainty ranges output by RadRAT account not only for sampling uncertainty in the model parameters, produced by the limited data available for the generation of the risk models, but also for modelled uncertainties in the values of the DDREF, minimum latency period and transfer of risk between populations (Berrington de Gonzalez et al 2012). Table 6 presents the cancer incidence LAR/Gy point and interval estimates produced by RadRAT for a uniform whole-body acute dose of 10 mGy of high-energy low-LET radiation received at 5, 35 and 65 yr of age by typical females and males from the US population for 2000–2005. An equivalent exercise for the Japanese population (for 2010) showed similar results (see table S3 and figure S2). The point LAR estimates in table 6 are for the BEIR/NCI models shown in figure 4, but what is striking is the considerable absolute widths of the interval estimates of LAR/Gy at an age-at-exposure of 5 yr. Of note is that at this age-at-exposure, the point LAR/Gy estimates for cancer incidence generated by the ICRP (2007) models (table 1) are less than the lower 90% uncertainty limits for the estimates obtained using the BEIR VII/NCI models (table 6). The wide interval estimates are a reflection of the imprecision of LAR/Gy estimates based on models derived from the LSS data (and associated assumptions, such as the value of the DDREF) for those exposed at young ages. When comparing the point estimates of the LAR/Gy of cancer incidence predicted by the ICRP and BEIR VII/NCI risk models, as illustrated in figure 4, the uncertainty ranges shown in table 6 for the BEIR VII/NCI risk models and associated assumptions, as generated by RadRAT, must be recognised;. Moreover, these indicators of uncertainty in risk estimates are most unlikely to take account of all the sources of uncertainty that are inherent in the ICRP and BEIR VII/NCI risk modelling and associated assumptions.

Table 6. LAR of cancer incidence (×10⁻⁴ Gy⁻¹), the lifetime baseline risk (LBR) of cancer incidence (×10⁻⁴), and the proportional increase in risk expressed as a fraction of the baseline risk (LAR/LBR) (Gy⁻¹) for the US population for 2000–2005, for each sex and three ages-at-exposure (5, 35 and 65 yr), following a uniform whole-body acute dose of 10 mGy of high-energy low-LET radiation (received in 2005). The calculations were performed using the online radiation risk assessment tool, RadRAT (Berrington de Gonzalez et al 2012), which employs BEIR VII cancer risk models and associated assumptions (NRC 2006) as modified by the US National Cancer Institute, https://radiationcalculators.cancer.gov/radrat/. Cancer incidence excludes non-melanoma skin cancer.

Age-at-exposure (years)	Sex	LAR of cancer incidence a , b , ×10−4 Gy−1 (90% uncertainty range) d	LBR of cancer incidence from time of exposure a , ×10−4	Proportional increase in cancer incidence risk (LAR/LBR) a , c , Gy−1 (90% uncertainty range) d
5	Male	2530 (1280–4060)	4862	0.52 (0.26–0.84)
	Female	4090 (2300–7050)	4144	0.99 (0.56–1.70)
35	Male	964 (474–1650)	4923	0.20 (0.10–0.34)
	Female	1160 (663–1970)	4090	0.28 (0.16–0.48)
65	Male	526 (274–863)	4293	0.12 (0.06–0.20)
	Female	548 (317–906)	2990	0.18 (0.11–0.30)

^a Excluding non-melanoma skin cancer. ^b Lifetime excess absolute risk, LEAR, the probability of developing a cancer over the remaining lifetime as a result of the exposure. ^c Lifetime excess relative risk, LERR, the fractional increase in the probability of developing a cancer over the remaining lifetime as a result of the exposure in comparison to the baseline probability of developing a cancer over the remaining lifetime in the absence of the exposure. ^d A '90% uncertainty range' associated with a particular point estimate of the LAR of cancer incidence is output by RadRAT (Berrington de Gonzalez et al 2012). RadRAT also provides a 90% uncertainty range for a LBR point estimate, but proportionally is small in comparison to that for the LAR and so is ignored in the calculation of the 90% uncertainty range for LAR/LBR.

4.5.2. Lifetime baseline risk (LBR) of cancer incidence

4.5.3. Other sources of uncertainty