Ionising radiation as a risk factor for lymphoma: a review

The ability of ionising radiation to induce lymphoma is unclear. Here, we present a narrative review of epidemiological evidence of the risk of lymphoma, including chronic lymphocytic leukaemia (CLL) and multiple myeloma (MM), among various exposed populations including atomic bombing survivors, industrial and medical radiation workers, and individuals exposed for medical purposes. Overall, there is a suggestion of a positive dose-dependent association between radiation exposure and lymphoma. The magnitude of this association is highly imprecise, however, with wide confidence intervals frequently including zero risk. External comparisons tend to show similar incidence and mortality rates to the general population. Currently, there is insufficient information on the impact of age at exposure, high versus low linear energy transfer radiation, external versus internal or acute versus chronic exposures. Associations are stronger for males than females, and stronger for non-Hodgkin lymphoma and MM than for Hodgkin lymphoma, while the risk of radiation-induced CLL may be non-existent. This broad grouping of diverse diseases could potentially obscure stronger associations for certain subtypes, each with a different cell of origin. Additionally, the classification of malignancies as leukaemia or lymphoma may result in similar diseases being analysed separately, while distinct diseases are analysed in the same category. Uncertainty in cell of origin means the appropriate organ for dose response analysis is unclear. Further uncertainties arise from potential confounding or bias due to infectious causes and immunosuppression. The potential interaction between radiation and other risk factors is unknown. Combined, these uncertainties make lymphoma perhaps the most challenging malignancy to study in radiation epidemiology.


Introduction
Lymphomas comprise a diverse group of malignancies involving cells of the immune system. These diseases include (1) precursor cell lymphoid malignancies involving proliferation of immature lymphoblasts, and (2) mature lymphoid malignancies involving differentiated Band T-cells (figure 1, table 1) [1,2]. Lymphomas are classed as Hodgkin lymphoma (HL) or non-Hodgkin lymphoma (NHL) according to the presence or absence of Reed-Sternberg cells, respectively [3,4]. Lymphoma belongs to the wider disease category of lymphoid malignancies. These diseases may present as leukaemia, with proliferation predominantly occurring in the bone marrow and blood, or as lymphoma, in which proliferating cells form extra-marrow mass lesions. Solid tumours are typically found in lymph nodes, though can occur anywhere in the body where lymphoid tissue is found (e.g. [5,6]). The distinction between leukaemia and lymphoma is now regarded as artificial [7,8], as these malignancies have considerable histological overlap, with many diseases involving both solid and circulating phases [9]. For example, chronic lymphocytic leukaemia (CLL) and small lymphocytic lymphoma (SLL) are considered to be histologically the same disease [10], with the former involving proliferation in the blood and marrow, and the latter forming solid tumours. These two disease manifestations are given separate International Classification of Diseases (ICD) and ICD Oncology (ICD-O) [11,12] codes, however. Likewise, acute lymphoblastic leukaemia (ALL) and lymphoblastic lymphoma (LBL) are regarded as a single disease entity [1], though are coded differently. Multiple myeloma (MM) and hairy cell leukaemia (HCL) are malignancies of mature B-lymphocytes, and thus are considered sub-types of NHL [1], though again they have distinct ICD and ICD-O codes and are almost invariably analysed separately in epidemiological studies and recorded as separate disease entities in cancer statistics (e.g. [13]).

Epidemiology
Around 120 000 cases of HL and NHL, including MM and CLL, are diagnosed each year in the USA [14], while around 25 000 cases are diagnosed in the UK, representing around 7% of total cancer cases [15][16][17][18]. Overall, incidence of NHL exceeds HL by a factor of around ten [3,4], though between the ages of 10 and 30 years, HL tends to be the more common form, especially in Europe and North America [13]. In teenagers, lymphomas represent around a quarter of all cancers. All major mature lymphoid malignancy subtypes are more common in males [15][16][17][18]. Both CLL and MM are extremely rare before age 50 years and almost unknown in childhood [19,20].

Non-radiogenic risk factors
Most forms of both HL and NHL are strongly associated with congenital, acquired or druginduced immunosuppression [21,22]. Greatly elevated rates have been observed among individuals with HIV/AIDS and those receiving immunosuppressant drugs following organ transplantation [21]. Certain infections are also associated with lymphoma, including Epstein-Barr virus [23], hepatitis C virus [24] and Helicobacter pylori [4]. Positive associations have been found between autoimmune conditions and several lymphoma subtypes [24]. Smoking has also been identified as a risk factor, especially for T-cell NHL and HL [25,26]. Obesity may be associated with increased risk of large B-cell NHL [27], though not CLL/SLL [28]. Alcohol consumption is associated with reduced rates of both NHL [22,29] and HL [30], while there is a possible negative association between HL and ultraviolet radiation exposure [31]. Elevated lymphoma rates have been observed in certain occupational groups, including crop farmers, painters, textile workers and women's hairdressers [32], suggesting that certain chemicals may be a risk factor [33]. There is no evidence of any impact of socio-economic status on risk of MM or CLL [34]. There is a suggestion of increased HL incidence for most deprived areas, but for males only [34].

Cellular origins of lymphoma
While most lymphomas are formed of mature lymphocytes, these cells do not necessarily represent the origin of the disease. Alternatives include lymphoblasts, multi-lymphoid progenitor R153 Table 1.
Interlymph classification [1] of lymphoid malignancies. The 'grouping' column represents the disease each malignancy is usually categorised as in epidemiological studies, e.g. follicular lymphoma is included within the category 'NHL' or the broader category of all lymphomas.
Diseases in bold are the subject of this review. cells and haematopoietic stem cells (figure 1) [35,36]. Each cell type may be found in different locations, including the bone marrow, blood, spleen, thymus or germinal centres in lymph nodes, according to stage of lymphopoiesis. T-lymphocytes, for example, initially develop in the marrow before being released into the blood, maturing in the thymus, then migrating to a lymph node or other lymph tissue. One or more radiation-induced genetic mutations may occur at any of these stages and locations, while the tumour itself may develop somewhere else. Evidence is emerging that at least some lymphomas develop from 'lymphoma stem cells' residing in the marrow [35]. Cases have been reported of allogenic bone marrow transplant donors and recipients both developing identical mature lymphoid malignancies, often at the same time, post-transplant [35]. Further evidence comes from cases of 'composite lymphomas', in which the patient develops multiple, histologically distinct lymphomas, derived from a common precursor cell [37]. This uncertainty in the cell of origin for lymphoma is almost unique in radiation epidemiology, where the site of a tumour is generally assumed to be the site of cell proliferation and the site of initial and subsequent genetic mutations, thus defining the tissue/organ for which dose must be estimated in dose response analysis. Consequently, there is no consensus on the appropriate target organ for lymphoma. Most epidemiological dose response analyses for lymphoma are based on bone marrow dose, although colon dose [38], personal dosemeter reading (in occupational studies) and mean lymphocyte dose [39] have also been used. An additional exposure pathway for circulating lymphocytes, relevant to inhaled alpha-emitting radionuclides such as radon, is via the transbronchial epithelium in the lungs [40]. While often similar, mean lymphocyte and bone marrow doses may differ by a factor of up to 2.6, depending on exposed region [41]. Along with potential misclassification and the profound impact of immunosuppression, the uncertainty in cellular origin complicates assessment of the relationship between radiation and lymphoma. The findings of the epidemiological studies discussed in the remainder of this review should be placed in the context of this uncertainty.

Epidemiological evidence of ionising radiation as a risk factor for lymphoma
We searched PubMed for English language epidemiological studies (cohort and case control) analysing lymphoma risks following radiation exposure. We searched using the MeSH terms 'Lymphoma' and the respective MeSH terms for each sub-type of lymphoma including broad categories 'Non-Hodgkin lymphoma' and 'Hodgkin lymphoma', and the MeSH terms for ionising radiation, background radiation, x-rays, gamma rays and alpha and beta radiation This search yielded 3156 items, of which 106 were selected. The reference lists of included papers and previous reviews of lymphoma risks [42,43] were also searched. No restriction was applied for publication period. As external analyses may be potentially biased, e.g. due to healthy worker effect, a greater weight was given to those publications reporting a dose response based on internal analysis.
As mentioned in the introduction, CLL and MM are regarded as subtypes of NHL [1], but are almost always analysed separately and will be reported as described in the cited papers. We assumed that papers reporting figures for 'lymphoma' without giving further details were restricted to the current (3rd edition, 1st revision) ICD-O-3 codes [12] 9590 to 9729, 'Hodgkin lymphoma' was restricted to codes 9650 to 9667 and 'non-Hodgkin lymphoma' implied codes 9670-9729 (thus not including CLL, MM or HCL). Dose units are written as reported in the cited papers. An equivalent dose of 1 sievert (Sv) corresponds to a dose of 1 gray (Gy), averaged over the whole organ, for photons and electrons. Where there is a neutron or alpha component, this equivalency no longer applies.
In a detailed analysis of lymphoma mortality among male atomic bombing survivors aged 15-64 years at the time of exposure, Richardson et al [38] reported positive associations for the periods 35-45 years (ERR = 2.23 Sv −1 , 90% CI: 0.09, 6.91) and 46-55 years (1.70, 95% CI: 0.16, 5.36) since exposure, but not for 5-25 years (0.08) and 26-35 years (−0.10). Confidence intervals were undefined in both cases. A similar, but stronger, pattern was observed for NHL alone. The authors also reported a higher ERR when analysis was restricted to survivors receiving doses below 0.5 Sv, compared to the whole dose range. Although this suggests nonlinearity of dose response, the introduction of a quadratic term did little to improve goodness of fit, compared to a purely linear model [38].

(continued)
Study (year) Disease Grouping Sample size (exposed/un-  Other large occupational cohorts include Mayak workers [95] and nuclear workers in Japan [93] and Canada [85,89]. In each case, large central values of ERR are associated with extremely wide confidence intervals, preventing meaningful interpretation. The most recent analysis of the cohort of Rocketdyne workers [88] found no evidence of increased risk of lymphoma, leukaemia or all cancers combined (table 3). Raised risks for NHL [7], MM [64] and, interestingly, CLL [63], have been reported among Chernobyl clean-up workers. The ERR for CLL was 2.58 Gy −1 (95% CI: 0.02, 8.43) [63], which although unusual is statistically compatible with the INWORKS study due to overlapping confidence intervals.
Underground miners are exposed to increased levels of alpha-emitting radon progeny, which are known to be associated with lung cancer. Darby et al [86] found no evidence of increased mortality among miners, for either NHL (SMR = 0.80, 95% CI: 0.56, 1.10) or HL (SMR = 0.93, 95% CI: 0.54, 1.48). The SMR for all cancers other than lung was close to background (1.01, 95% CI: 0.95, 1.07). Leukaemia SMR was raised for miners with <10 years employment (1.93) but not ⩾10 years (0.99). The leukaemia SMR was reduced from 1.93 to 1.28 after excluding CLL. Studies of uranium miners and mill workers [47][48][49][50][51][52][53][54][55][56][57][58][59][60][61][62], exposed to both low LET gamma and high LET alpha radiation, are similarly inconclusive. As with other industry nuclear workers, overall cancer rates tend not to be raised relative to the general population, although there are exceptions [52]. An internal analysis of Eldorado uranium miners by Zablotska et al [60] was also inconclusive. Central ERR values, based on external gamma dose (mean whole body dose = 52.2 mSv), were raised for incidence but not mortality, or vice versa depending on disease subtype (table 4). Confidence intervals were either extremely wide or undefined.
Studies of nuclear weapons test participants [65][66][67][68][69][70][71] typically show similar rates of lymphoma, and all cancers combined, to the general population. The most informative study is a recent analysis of a pooled cohort of eight US testing programs [71]. The mean gamma dose was 6 mSv, with 55.3% receiving <5 mSv. The SMR was reduced for NHL and HL (table 3), though this appeared to be driven by a 'healthy soldier' effect in the early years of follow-up [71]. Mortality rates for MM were raised among UK test participants in an early study [65] (SMR = 1.17), based on six cases. In a more recent analysis [68], SMR was reduced to 0.96, based on 22 cases. There was little suggestion of raised MM rates in the 8-series US study [71]. The SMR was 0.98 (95% CI: 0.88, 1.09), while an internal analysis yielded an ERR of −0.16 per 100 mGy (95% CI: −1.03, 0.72).
Airline crew are exposed to elevated levels of cosmic radiation, including neutrons and muons, with annual effective doses ranging from 2 to 6 mSv [84]. There is limited evidence of increased incidence or mortality from lymphoma or overall cancer [79][80][81][82][83][84]. There are some suggestions of raised SMR for lymphoma among male cabin crew, although this appears to be associated with the extremely high rates of AIDS prior to 2000 in this group (mortality for R165 AIDS was raised 16-fold compared to the general population [83]). A decrease in AIDS prevalence in more recent years may explain the lower SMR for NHL in the most recent analysis of German airline crew [84], compared to previous [81].
The study by Eheman et al [101] is notable for the analysis of lymphoma risk for different subtypes and grades (disease aggressiveness) as part of the Selected Cancers Study. Odds ratios were higher for low-grade lymphoma (1.07, 95% CI: 0. 76 A number of studies have focussed on radiologists and radiographers (radiologic technologists), many of whom were exposed to particularly high doses of external low LET radiation in the early decades of medical x-ray imaging. In a recent analysis, Berrington de González et al [77]  There was little suggestion of raised risks for radiologists graduating after 1940, neither for lymphoma, nor for other cancer types. The authors noted lower rates of HIV among radiologists, compared with psychiatrists, potentially resulting in downward bias of risk estimates. A later analysis of physicians likely to perform x-ray guided interventional procedures [78] found no evidence of increased lymphoma risk (RR: 0.77, 95% CI: 0.58, 1.04). In contrast to the LSS, studies of radiographers provide no evidence of any sex difference in mortality or incidence of lymphoma [74,75].

Environmental exposures
Studies of natural background radiation [8, 102-108] include populations exposed to both low LET gamma radiation and high LET alpha radiation originating from inhaled radon. Again, with the exception of that due to radon, exposures are reasonably uniform over the whole body. A summary of study findings is presented in table 5.
Tao et al [102] found no evidence of raised rates of lymphoma among residents of the Yangjiang region of China, exposed to elevated background radiation levels, estimated at 6.4 mSv (effective dose) per year. The RR for lymphoma, compared to a neighbouring region with normal background radiation levels, was 0.98 (95% CI: 0.26-3.71). A dose response analysis was attempted by categorising cohort members into low-, medium-and high-exposure groups; however, the results are too imprecise to be informative. Hwang et al [104] reported a raised risk of NHL among residents of buildings in Taiwan made using steel contaminated with cobalt-60 (SIR = 5.4, 95% CI: 1.8, 12.6), though based on just five cases.
No significant associations were detected by Kendall et al [105] between either childhood NHL or HL and gamma or radon background radiation. Positive associations were seen for leukaemia, particularly lymphoid leukaemia, but not for all cancers except leukaemia. Spycher et al [107] also found no evidence of an association between childhood lymphoma and background radiation in Switzerland. Hazard ratios were 1.08, 0.96 and 0.91 for estimated dose rates of 100-150, 150-200 and >200 nSv h −1 versus <100 nSv h −1 , respectively, while positive dose responses were observed for leukaemia, CNS tumours and all cancers combined.

R167
Other studies, including those of children living near nuclear facilities [8,108] are limited by low case numbers and limited dosimetry (e.g. based on residential distance from the facility).

Medical radiation-therapeutic
Therapeutic radiation is usually delivered to a localised region, with the remainder of the body exposed to a relatively low dose of scattered radiation. Careful selection of the suitable organ/tissue for dose response analysis is therefore required. Patients with malignant disease may also be treated with chemotherapy, which is known to increase subsequent cancer risk [109]. Some individuals have genetic disorders predisposing them to cancer development and/or increased radiosensitivity (e.g. [110,111]). Comparison of subsequent cancer rates for patients treated with or without radiotherapy is problematic as treatment choice reflects disease type. Furthermore, apparent associations may also be complicated by therapy-induced immunosuppression [112], especially when radiotherapy is combined with chemotherapy. Table 6 shows a summary of study findings. Increased rates of lymphoma have been observed among individuals treated with radiotherapy for a previous malignancy [112][113][114][115], ankylosing spondylitis [116][117][118] and peptic ulcer [119] and patients injected with radium-224 (MM) [120]. Studies of patients treated for non-malignant gynaecological conditions [121,122], otitis serosa (treated with nasopharyngeal radium irradiation) [123] or benign locomotor conditions [124] have found little or no evidence of raised lymphoma risk, though case numbers are small. No association between radioactive iodine-131 (RAI) treatment for hyperthyroidism and lymphoma was found by Holm et al [125] or Franklyn et al [126]. A very small increase in overall cancer incidence was reported in the former study (SIR = 1.06, 95% CI: 1.01, 1.11) but not the latter (SIR = 0.83, 95% CI: 0.77, 0.90). A recent update of a 1998 study [127] of cancer risks among patients treated with RAI for hyperthyroidism by Kitahara et al [128] found RRs, per 100 mGy, of 1.07 and 1.69 for NHL and MM respectively. In both cases, confidence intervals included unity.
A number of studies have made use of Surveillance, Epidemiology and End Results (SEER) data to study risk of lymphoma risk following radiotherapy. Kim et al [112] identified 5590 NHL second malignancies reported by nine SEER registries. RR of NHL was increased for primary malignancies of any type treated with radiotherapy compared to those treated without radiotherapy (RR = 1.13, 95% CI: 1.08, 2.17). When results were analysed by primary disease, positive associations were seen only for non-small cell lung cancer and prostate cancer. No significant difference in NHL risks was seen between males and females. Likewise, there was little evidence of variation in risk between NHL subtypes for all primary cancers combined. Significant differences in RR (versus treatment without radiotherapy) between subtypes were observed, however, for primary cancers of the rectosigmoid, in which diffuse large B-cell lymphoma risks were higher, and thyroid, in which follicular lymphoma risks were higher. Chaturvedi et al [115] identified 52 613 patients among SEER and Scandinavian cancer registries treated with pelvic radiotherapy for cervical cancer. The SIR was raised for NHL (1.20, 95% CI: 1.02, 1.40), based on 157 cases, but not for HL, MM or CLL. In comparison, the SIR for all cancers was 1.34 (95% CI: 1.31, 1.38). Radivoyevitch et al [129] compared CLL/SLL risk among patients treated for non-haematological malignancies using SEER data. RRs were raised among 4483 857 patients not treated using radiotherapy (1.2, 95% CI: 1.17, 1.23) but not raised among 1808 105 patients who were treated with radiotherapy (1.00, 95% CI: 0.96, 1.05). Again, these findings should be interpreted with caution as the types of primary cancer typically treated or not treated with radiotherapy are different and may have inherent different subsequent cancer risks or exposure to other carcinogenic agents (e.g. chemotherapy). Wright  [130] found no evidence of increased MM risk among 66 896 patients with pelvic malignancies treated with radiotherapy, compared to 132 372 patients treated without radiotherapy (hazard ratio: 1.08, 95% CI: 0.81, 1.44). Lymphoma is a relatively uncommon second malignancy among survivors of cancer in childhood or early adulthood (e.g. [131][132][133]). NHL second malignancies tend to follow a primary diagnosis of HL [134]. Underlying genetic factors and immunosuppression may therefore be the primary risk factor.

Medical radiation-diagnostic
Diagnostic x-ray exposures include general radiography [135,136], fluoroscopy [137,138], computed tomography (CT) [39,[139][140][141][142], a combination of these [143], or pre-natal x-rays [144,145]. As with radiotherapy, exposures are usually localised, rather than whole-body. Studies of individuals exposed for diagnostic purposes need to be interpreted with caution due to the potential for reverse causality, where patients are exposed to investigate early symptoms of a later diagnosed cancer [146][147][148]. Alternatively, some patients may have diseases predisposing them to cancer development. If these individuals undergo more medical imaging tests, the association between radiation and lymphoma may be confounded by indication. A summary of study findings for diagnostic exposures is presented in table 7.
Elevated rates of several forms of cancer have been observed among individuals injected with the alpha-emitting contrast agent Thorotrast [153][154][155]. Throughout their lifetime, these patients received cumulative absorbed doses of several gray to the bone marrow and several tens of gray to the spleen [154]. A suggestion of an increased risk of NHL was found for a German cohort [153], based on 15 cases (RR versus controls: 2.5), but not for HL or CLL (RR was 0.8 in both cases). The number of cases of lymphoma diagnosed within Swedish [155], Danish [153,155] and American [155] cohorts are too low to be informative.
Several recent studies have examined lymphoma incidence following diagnostic x-ray exposures before birth or in early childhood. One of the major findings of the Oxford Survey of Childhood Cancers was a raised risk of cancer among children exposed in utero during pelvic radiography [144]. RR (versus unexposed) was raised for lymphoma (1.35, 95% CI: 1.07, 1.65) and for all malignancies combined (1.47, 95% CI: 1.34, 1.62). Foetal doses from obstetric radiography were approximately 10-20 mGy per film [156], though subject to large uncertainties. A large odds ratio was reported by Rajaraman et al [149] for all lymphoma (5.14, 95% CI: 1.27, 20.80) and specifically NHL (6.85, 95% CI: 1.31, 35.70) following diagnostic x-ray exposure in infancy. These findings contrast with those of studies led by Hammer [135] and Baaken [136] in which no evidence of an association was found between lymphoma and post-natal exposure to diagnostic x-rays in a cohort of over 90 000 German children. It should be noted that the average dose received by members of the German cohort was exceptionally low (median estimated effective dose = 0.007 mSv, mean = 0.135 mSv).

R174
González et al [39], who found no association between HL and estimated bone marrow or lymphocyte dose from CT scans in childhood. This analysis was based on the same UK cohort in which positive association were found for leukaemia/myelodysplasia and brain tumours [158,159]. Studies based in France [140,160] and Germany [141,151] found little evidence of raised risk of lymphoma following CT scans in childhood after excluding individuals with predisposing syndromes. A recent case control study [152] found no evidence of an association between lymphoma risk and self-reported lifetime medical x-ray exposure, for 2362 cases compared to 2465 controls. Cumulative estimated bone marrow doses were very low, however (median: 2.25 mGy). Raised lymphoma rates have been reported among individuals who underwent fluoroscopically guided cardiac catheterisation procedures in childhood or early adulthood [137,138]. A recent analysis [138] found elevated rates of both NHL (SIR = 19.49 95% CI: 11.39, 31.10) and HL (SIR = 2.70 95% CI: 0.68, 7.07). Following transplant registry linkage, it was found that all malignant lymphoma cases and nine cases of lymphoproliferative disease developed post-transplant. The proportion of transplant recipients in this cohort (around 5%) is likely to be much higher than in other medical radiation exposure studies, given the high proportion of patients with serious cardiac abnormalities.

Discussion
Lymphoma is frequently included in site-specific analyses of cancer risks following radiation exposure. Yet this disease is, perhaps, the most challenging form of cancer to analyse in radiation epidemiology studies. This is due to (1) the high degree of heterogeneity among lymphoid malignancies, with potential for irregularities in grouping between epidemiological studies, (2) uncertainty in the site of initiation, and (3) the profound impact of infection and immune system compromise on lymphoma risk. Partly for these reasons, evidence of an association between lymphoma and ionising radiation exposure has been inconclusive [42,43]. Issues (2) and (3) suggest a greater weight should be placed on studies of healthy populations receiving approximately whole-body exposures, including atomic bombing survivors, radiation workers and those exposed to elevated background radiation levels. While external comparisons typically show lymphoma rates similar to the general population, internal analyses tend to suggest a dose-dependent excess lymphoma risk. Confidence intervals are wide, however, indicating highly imprecise risk estimates. There is no evidence that RRs for lymphoma are any higher than for other types of cancer and appear to be lower than for leukaemia.
Studies also suggest a small excess risk of lymphoma following radiotherapy for malignant or non-malignant conditions. This is hardly surprising as almost all cancer types appear to be inducible by radiation if the dose is sufficiently high. Lymphoma risk is increased by a factor of around 5-10 among adult transplant recipients and up to 100 in people with HIV/AIDS [21], i.e. much higher risks than observed among individuals treated with radiotherapy (<2-fold). This suggests the impact of radiation and radiation-induced immunosuppression is relatively minor compared to the long-term immunosuppression associated with transplantation (druginduced) or AIDS. Effect modification, or an interaction between immunosuppression and radiation remains a possibility, however.
There is little evidence of any difference in lymphoma risk for acute versus chronic radiation exposures, or for high versus low LET, or internal versus external exposures, for a given absorbed dose. Risks are similar for male atomic bombing survivors exposed at a very high rate, and nuclear workers receiving protracted exposures. There are insufficient data to determine the impact of age at exposure on lymphoma risks.
The higher lymphoma ERR among male atomic bombing survivors is puzzling and lacks a biological explanation. No such pattern was observed for MM in the same study [45], or in studies of medical radiographers [74,75]. In each case, confidence intervals for male and female ERR figures overlap, suggesting any apparent sex differences in the LSS could be a chance finding. Lymphoma mortality was reported to be higher among male airline cabin crew than female [80], although this may be largely explained by much higher rates of AIDS in the former group.
The classification of lymphoma and other lymphoid malignancies has evolved considerably over the last 50 years, in parallel with advancements in understanding of the differentiation of immune cells and cell of origin [2,[161][162][163][164]. Overviews of current and historical classifications, including compatibility between systems, are presented elsewhere, e.g. [12,165]. For the purposes of this review however, the important consideration is whether the distinction between lymphoma subtypes, or between lymphoma and lymphoid leukaemia, could influence apparent radiation-associated risks. For example, in the analysis of childhood cancer in the vicinity of nuclear facilities, some cases were initially classified as NHL before being reclassified as leukaemia [8]. Misclassification, or inconsistencies in classification, could potentially explain unusually high risks for lymphoma but not leukaemia (e.g. [149]), or vice versa. In this regard, there may be some justification for grouping all lymphoid malignancies together [8,166].
However, there is known heterogeneity in non-radiogenic risk factors within the broad subgroups of lymphoid malignancies (e.g. [22]). NHL, for example, is almost invariably analysed as a whole, despite potential differences in cell of origin, proliferating cell type and disease aggressiveness. Given that ALL is strongly associated with radiation (e.g. [46]) it may be assumed that LBL, which is histologically the same disease in mass lesion form, is similarly sensitive to induction by radiation. Yet ALL and LBL have different ICD [11] and ICD-O [12] codes and are likely to be assigned as leukaemia or lymphoma, respectively, in epidemiological analyses. Likewise, SLL and CLL are regarded as the same disease in solid and circulating form, respectively. Again, partly due to the different ICD coding, it is possible these diseases have been analysed separately in epidemiology studies, with SLL likely assigned as NHL.
Many forms of lymphoma are associated with good prognosis, with 5-year survival rates exceeding 80% for HL and 70% for CLL [167]. Richardson et al [100] note that CLL is often not listed as the primary cause of death, or even mentioned on death certificates. The use of mortality data may, therefore, be less reliable than incidence data. This could potentially explain the apparent lack of association between radiation exposure and CLL [100].
There is some suggestion that lymphoma may have a long latency period, potentially further explaining negative CLL findings [100,168] and the relatively late appearance of raised risks in the LSS and Savannah River cohorts [38]. In contrast to other mature lymphoid malignancies, both CLL and myeloma are almost exclusively diseases of middle and old age, with around 99% of cases being diagnosed after age 50 years [16,18]. Many of the cohorts, especially those of medically exposed individuals, lack sufficiently long follow-up periods to provide meaningful information on these diseases. The reverse may also be true, however. Lymphomas are occasionally classed as 'solid tumours' in epidemiological studies, meaning exclusion periods and dose lagging for lymphoma analyses are the same as for other solid tumours such as lung cancer (typically 5 or 10 years). However, histologically, lymphoma has more in common with leukaemia, a disease known to occur relatively early following radiation exposure (CLL possibly being the exception). One may assume acute leukaemia-like forms of lymphoma such as LBL may also develop similarly early. Again, the standard grouping of diseases may obscure meaningful patterns.
Ongoing studies involving large pooled cohorts hold promise for improved information on lymphoma risks following radiation exposure. The Million Worker Study (MWS), coordinated by the National Council on Radiation Protection and Measurements, includes atomic weapons test veterans, US Department of Energy workers, nuclear power plant workers, industrial radiographers and medical radiation workers [169,170]. The MWS will provide unprecedented information on risks from protracted exposures, based on analysis of over 300 000 deaths.
In addition, a number of large cohorts will provide improved information on lymphoma risks from exposures in childhood. The MEDIRAD study (https://www.medirad-project.eu/) involves a pooled cohort of children and young adults who received CT scans in Europe, continuing from the EPI-CT study [171]. The newly launched HARMONIC study [172] (https://harmonicproject.eu/) will examine the long-term effects of cardiac catheterisations and proton beam therapy in children. Given the high rates of heart transplantation in the former group, obtaining information on transplants (ideally through registry linkage) will be essential for this study.

Conclusion
The association between ionising radiation exposure and lymphoma, including MM and CLL, is complex and subject to large uncertainties. Increased risks for certain lymphoma subtypes may be obscured by broad classification schemes or confounded by the impact of immunosuppression or infection. The available evidence suggests a positive dose-dependent association between radiation exposure and lymphoma risk. This association appears to be stronger for males and stronger for NHL, as opposed to Hodgkin's lymphoma, MM or CLL. The risk of radiation-induced lymphoma is unlikely to be especially large, certainly no higher than for other cancer types. For populations in which lymphoma rates are unusually high, other aetiologies should be considered first, especially if rates for more radiogenic cancers are not raised.