Abstract
Background
Outcome assessment problems and errors that could lead to biased risk estimates in low-dose radiation epidemiological studies of cancer risks have not been systematically evaluated.
Methods
Incidence or mortality risks for all cancers or all solid cancers combined and for leukemia were examined in 26 studies published in 2006–2017 involving low-dose (mean dose ≤100 mGy) radiation from environmental, medical, or occupational sources. We evaluated the impact of loss to follow-up, under- or overascertainment, outcome misclassification, and changing classifications occurring similarly or differentially across radiation dose levels.
Results
Loss to follow-up was not reported in 62% of studies, but when reported it was generally small. Only one study critically evaluated the completeness of the sources of vital status. Underascertainment of cancers (“false negatives”) was a potential shortcoming for cohorts that could not be linked with high-quality population-based registries, particularly during early years of exposure in five studies, in two lacking complete residential history, and in one with substantial emigration. False positives may have occurred as a result of cancer ascertainment from self- or next-of-kin report in three studies or from enhanced medical surveillance of exposed patients that could lead to detection bias (eg, reporting precancer lesions as physician-diagnosed cancer) in one study. Most pediatric but few adult leukemia studies used expert hematopathology review or current classifications. Only a few studies recoded solid cancers to the latest International Classification of Diseases or International Classification of Diseases for Oncology codes. These outcome assessment shortcomings were generally nondifferential in relation to radiation exposure level except possibly in four studies.
Conclusion
The majority of studies lacked information to enable comprehensive evaluation of all major sources of outcome assessment errors, although reported data suggested that the outcome assessment limitations generally had little effect on risk or biased estimates towards the null except possibly in four studies.
The internal validity of observational cohort studies requires complete and accurate outcome assessment for both exposed and comparison persons. High-quality outcome assessment is critical for accurate cancer risk estimation in relation to low-dose radiation dose-response for both cohort and case-control studies. Comprehensive reviews of epidemiological studies of cancer and other health effects associated with low-dose radiation, including some of the 26 studies evaluated in this article (1–29), have identified several sources of uncertainty (30) and potential bias (31), but the reviews and textbooks (32) provide less description of methodological problems due to inaccurate outcome assessment and potential biases associated with these problems for risk estimates than those related to exposure assessment or confounding.
Major sources of error in outcome assessment that could result in potentially biased cancer risk estimates in epidemiological studies of low-dose radiation exposure include differential (between exposed and comparison populations or across radiation dose categories) loss to follow-up, under- or overascertainment of outcomes, misclassification, and application of changes in classification over time. The most common designs for epidemiological studies of ionizing radiation are retrospective record linkage-based cohort (or nested case-control) studies or mixtures of retrospective and prospective cohort follow-up (26–28,33). These study designs enable long-term follow-up of large populations but may have limitations for outcome assessment.
In this review, the major sources of methodological problems and errors and potential biases associated with such errors in outcome assessment of total cancer (all-cancer sites combined), total solid cancers, leukemia, and a few site-specific solid cancers in relation to low-dose radiation exposures (≤100 mGy) are systematically appraised for 26 epidemiological studies. The selected studies were published subsequent to the Biological Effects of Ionizing Radiation VII report (30) during the period 2006–2017.
Methods
Study Populations
The study eligibility requirements, exclusion criteria, features, and findings are provided in the Rationale and Overview article included in this monograph (34). In brief, eligible cohort or case-control epidemiological studies were those of human populations exposed to low-dose (mean dose ≤100 mGy), predominantly low Linear Energy Transfer radiation with individualized cumulative radiation dose estimates and dose-response analyses reporting cancer risks and associated confidence intervals per unit dose or at a given exposure level. A total of 26 eligible studies included eight populations with environmental, four with medical, and 14 with occupational radiation exposures characterized by mean estimated exposures ranging from 0.1 mSv (2) to 82 mGy (20) (see table 1 in Berrington de González et al in this monograph (34)).
Table 1.
Major sources of error in outcome assessment in radiation epidemiological studies
| Potential effect on dose response |
Strategies to reduce error in risk | |||
|---|---|---|---|---|
| Source of error | Description | Nondifferential | Differential | assessment of outcomes |
| Loss to follow-up |
|
Should not result in bias if same data sources, methods, and level of effort used to trace and/or link in both exposure groups but may result in loss of statistical power | Bias may occur if loss substantial and associated with both exposure and outcome (eg, used different data sources for tracing or linkage, different methods or levels of effort in exposed compared with unexposed) | Use standardized methods and all available sources to target 100% follow-up. If active tracing of all patients not feasible (best approach), investigators should aim for linkage with high-quality vital status registries and data sources. Investigators should characterize level of completeness of vital status databases. High-quality nationwide mortality registries may be better for outcome assessment when nationwide high-quality cancer registries unavailable. |
| Under- or over ascertainment of outcomes |
|
Should not result in bias if same data sources, methods, and classification system used | Bias may occur with use of different (by exposure status) sources of outcome data or methods for identifying cancer outcomes (eg, self-administered interviews, population-based cancer registry linkage, clinical assessment, cancer screening) or classification systems | Use of rigorous standardized methods, same data sources, and most recent classification systems (including reclassification where applicable or possible) should result in complete and accurate identification of all cancers. Cohort linkage with high-quality population-based cancer registries may still result in underascertainment for certain outcomes generally identified and treated in physicians’ practices or diagnosed solely by imaging. Use of multiple sources of data will likely yield high ascertainment. Screen-detected cancer outcomes should be excluded if identified in exposed groups only. |
| Misclassification, changes in classification over time, or indeterminate classification |
|
Could lead to under- or overestimation if substantial differences in outcome(s) misclassified between exposed vs unexposed populations | May result in biased cancer risk estimates. If efforts are made to reclassify cancer outcomes according to new criteria, bias could occur with usage of different sources of data, different approaches, or different levels of effort in exposed vs unexposed populations. |
|
Exposure Assessment
For persons with medical and occupational radiation exposure, doses were estimated for individuals. In most of the studies of background radiation exposure, doses of individuals were modelled with the same dose generally assigned to those persons considered to be similar in geographic location. Analyses estimated risk per unit dose or risk at a specified dose. The dosimetry methods for the 26 studies are described by Daniels et al. in this monograph (35).
Outcome Assessment: Sources of Error and Likely Impact on Risk Estimates
As described in the overview article in this monograph (34), this review focuses on the results for all solid cancers and for leukemia separately rather than site-specific solid cancer analyses, which may lack power, may be more prone to misclassification, and may be difficult to interpret due to multiple testing issues. Most of the studies included in this review involved whole-body uniform radiation exposure, although a few assessed more localized exposures. For studies of nonuniform exposures, however, the results are generally for specific cancer sites that were highly exposed and/or highly radiosensitive, for example, brain tumors after pediatric computed tomography scans and breast cancer after medical or occupational exposures. Table 1 and the text immediately below summarizes the major sources of methodological problems and errors in outcome assessment that could lead to potential biases associated with these problems.
Loss to Follow-Up
Complete follow-up to ensure identification of all cancer outcomes in exposed and comparison populations is critical to ensure accurate cancer risk estimates in exposure response analyses (36–38). The accuracy of risk estimates in nested case-control studies could also be affected by loss to follow-up in the target cohort (36). Follow-up in cohort studies ideally continues until the event (which may include one or multiple outcomes of specific interest), the date individuals are censored when they can no longer be followed, or the end of the follow-up period. At the end of the study, persons may be alive and diagnosed with the outcome(s) of interest, alive and free of the outcome(s) being assessed, deceased with known or unknown causes or dates of death, or lost to follow-up at some time before the end of the study period. Loss to follow-up can occur under two different scenarios. In the first instance, persons can move from the geographic area (a city, state, region, or country) where the investigator has been able to link the person with a population register or otherwise contact the person to ascertain the outcome(s) of interest to a geographic area where such linkage or follow-up is no longer possible and the date of the inability to follow-up the person is known. In this circumstance, analyses generally censor the person at the date when the person's status was last known. In the second instance, persons move from the geographic area where follow-up can be undertaken to another area where follow-up is not feasible, but the investigator does not know when the persons moved. In this second scenario, the “date of censoring” is unknown, and the time point at which accurate follow-up ceases is also unknown.
A key consideration affecting the level and interpretation of loss to follow-up is whether investigators have carried out “active” or “passive” individual follow-up. Active individual follow-up is defined as direct contact with persons and/or linkage of the person with different types of alternative registries or other sources that provide credible information about vital status to the end of the study or diagnosis of incident cancer or death. Passive individual follow-up is characterized by linkage or indirect follow-up such that the vital status of the person is “assumed alive” or “assumed deceased,” but rigorous confirmatory documentation may be lacking or is unclear. Active individual follow-up, if highly successful, can lead to improved ascertainment, whereas linkage with vital status, cancer, and death registries may be less complete if substantial fractions of the population are missing from registries or the persons are missing key information for linkage.
Under- or Overascertainment
Under- or overascertainment of cancer outcomes is another potential source of error. Loss to follow-up may result in underascertainment. Underascertainment may occur in the absence of active individual follow-up if not comprehensive or successful. Underascertainment may also occur if cohort members have incomplete identifiers that may result in problems with linkage, if the registers are incomplete and missing a substantial fraction of population members, if there is substantial emigration, or if registers are not established until after the exposure period of a study population commences. Overascertainment may occur if persons inaccurately self-report cancer outcomes that are incorrect or if screening or regular medical surveillance is undertaken such that preclinically cancerous lesions are identified. If the same level of under- or overascertainment occurs in exposed and unexposed individuals included in a study (or a low-exposure referent group), then risk estimates are unlikely to be biased, but if differential by radiation exposure level then there is the potential for biased risk estimates. The impact of outcome assessment errors on cancer risk estimates depends on the type of risk measure (eg, relative vs absolute excess risk). In this monograph, we focused on reviewing the relative risks and thus evaluated the biases relevant to relative risks. The data sources and methods used to identify and validate cancer outcomes are important determinants of the likelihood of under- or overascertainment.
Misclassification and Changes in Classification Systems.
Misclassification of outcomes in cohort or case-control studies may be another source of inaccurate risk estimation if the misclassification differs by radiation exposure level (39,40). Incorrect outcome information may be derived from inaccurate diagnoses provided by self-reports or listed in death certificates, medical records, physician reports, cancer registries, or other sources. Misclassification errors may occur if a nonmalignant condition is incorrectly reported as a malignancy, a malignancy is wrongly reported as a nonmalignant condition, or a specific type of malignancy is reported as another type of malignancy. Disorders that include benign and malignant variants (eg, meningioma, ovarian tumors) may be misclassified. Misclassification can also occur if a preclinical or in situ lesion identified in screening tests, self-reports, or medical records is incorrectly designated as a frank malignancy because such lesions may not transform into frank malignancies or may regress. Changes in cancer classification over time may lead to misclassification if there is increased specificity of histological or molecular subtypes; new International Classification of Diseases (ICD) codes added (eg, mesothelioma); reclassification from benign to malignant (eg, myelodysplastic syndromes [MDS], certain myeloproliferative disorders) or malignant to benign (eg, pilocytic astrocytoma); reclassification to another category of neoplasm (eg, chronic lymphocytic leukemia [CLL] changed from “leukemia” to “lymphoma”); and indeterminate with designation as borderline malignant (eg, certain ovarian tumors, meningioma). Cancer screening may be more likely to be undertaken in radiation-exposed than in unexposed populations or in populations with higher radiation exposure levels. Risk estimation may be biased upward or downward if the level of outcome ascertainment or misclassification differs between the radiation-exposed and comparison populations or across radiation exposure categories (32). A detailed description of qualitative and quantitative effects of nondifferential or differential misclassification by exposure level can be found in Greenland et al. (32). Statistical methods to assess the effect of misclassification on cancer and noncancer mortality risk estimates and to determine whether an observed noncancer mortality radiation dose response could be attributed to misclassification of cancer as noncancer outcomes on death certificates are described by Sposto et al. (41).
Secular Changes in Classification and Revisions in Classification Systems
Changes in classification of cancers over time and revisions in cancer classification during the course of long-term follow-up of radiation-exposed populations may affect the completeness of outcome ascertainment but are unlikely to result in bias unless occurring differentially in exposed and unexposed study individuals. Examples of such changes include the World Health Organization 2000 (42) reclassification of MDS and myeloproliferative disorders from “benign” conditions to malignancies, and of lymphoma subtypes (43–45); lack of ICD/International Classification of Diseases for Oncology codes for certain cancers (eg, mesothelioma before ICD-10); or cancers for which there is long-standing controversy about whether they should be considered “benign” or “malignant” (eg, pilocytic astrocytoma and borderline ovarian tumors) (46). Ideally, cohort and case-control studies in which internal comparisons are emphasized should employ the most recent internationally recognized classification scheme for outcomes and should reclassify outcomes using earlier classifications. However, reclassification should be conducted with great care (ideally with assistance from expert pathologists), because inadvertent errors may result.
Combining Heterogeneous Disorders for Analysis to Increase Statistical Power
Grouping of heterogeneous disorders (eg, “all solid tumors combined,” “gastrointestinal cancers not otherwise specified,” “total leukemia”) for analysis may be problematic. Although such grouping is often done to increase statistical power (47), radiation risks may be diluted if disease entities with known differences in radiogenic sensitivity are grouped and considered as a single “homogeneous” entity. Differences in methods of ascertainment by radiation exposure level (eg, screening of radiation-exposed individuals using ultrasound or radiologic imaging procedures vs identification based on physical symptoms in the comparison group, or more detailed expert review of histopathologic or molecular characteristics of tumors in exposed vs use of death certificates only in the comparison group) could further exacerbate dissimilarity with respect to homogeneity between exposed study individuals and the comparison group.
Outcome Evaluation Methods
The 26 studies were reviewed for information relevant to cancer outcome assessment (see Table 2). The criteria evaluated included study design; specific outcomes evaluated; numbers of study cases and study controls or numbers of outcomes and cohort size; source and method of cancer ascertainment; internationally recognized quality rating of the national or regional vital status, cancer, and mortality registries by the International Agency for Research on Cancer (IARC) (48); and exposure and follow-up periods.
Table 2.
Populations and study design, outcomes and ascertainment methods, IARC classification of registry quality, exposure and follow-up periods in 26 low-dose radiation epidemiological studies
| Study name and reference | Design, censoring, matching | Specific outcome evaluated | Numbers of cases and numbers of controls; outcomes and cohort size, | Cancer ascertainment method | IARC classification; mortality: 1–6; incidence: A-G* | Exposure period | Follow-up period |
|---|---|---|---|---|---|---|---|
| Environmental | |||||||
| Chernobyl-exposed children (1) | Case-control; matched regional clinic controls |
|
421 acute leukemia cases vs 842 controls | Hospital records, population-based cancer registries | A: Belarus (after 1997) and Ukraine; NA: Russian Federation | 1986–2000 | 1986–2000 |
| TMI residents (2) | Cohort study; TMI population registry enrolled 93% residents living within 5 miles; study end or cancer diagnosis | Incidence of all and selected malignancies, including leukemia | 1651 total incident malignancies, 56 incident leukemias in 22 069 TMI cohort members | TMI population registry matched annually with PA cancer registry | A: PA state cancer registry designated gold-level standard | 1979 | 1982–1995 |
| Chinese background (3) | Cohort study in high background and low background regions; in 1979, village household registry established to ascertain mortality | Mortality for all cancers, excluding leukemia and for leukemia | 956 total cancer deaths (of 6005 total deaths), 15 leukemias in 736 942 p-y follow-up | Household surveys, medical records of deceased, village doctors, family and next of kin | 4: Incomplete vital registration | 1905–1998 | 1979–1998 |
| GB background (4) | Case-control; matched controls from same birth registry | Childhood cancer incidence | 27 447 total incident cancers (9058 leukemias) vs 36 793 controls with all cancers (11 912 leukemias) | National cancer registry | A: High-quality national | 1991–1996 | 1980––2006 |
| Swiss children background (5) | Cohort study from persons in 1990 and 2000 censuses; death, study end, age 16 y, emigration | Childhood cancer incidence | 1782 cancers, 530 leukemias among 2 093 660 children | National cancer registry | B: High-quality regional | 1974–2008 | 1990–2008 |
| Techa River residents (6) | Cohort study of residents near Techa River or in Chelyabinsk City during some or all of 1956–2007; death, study end, emigration | Solid cancer incidence | 1933 solid cancers among 12 759 cohort members | Medical records from oncology dispensaries, regional oncology clinics, health centers, and death certificates; death certificate-only diagnoses were 19% (range 4–56%) by subtype | 2: Medium-quality vital registration | 1951–2007 | 1956–2007 |
| Finnish background (7) | Nested case-control study; matched population register controls | Childhood leukemia incidence | 1039 cases vs 3279 controls | National cancer registry | A: High-quality national | 1990–2011 | 1990–2011 |
| Taiwanese residents (8) | Cohort study; cancer diagnosis, death, study end | Cancer incidence | 111 cancers excluding leukemia (6 leukemias identified in 6242 persons with residential radiation exposure history who were followed-up) | National cancer registry | A: Currently; population cancer registry established in 1979 with notable improvement in quality over time | 1982–1992 | 1983–2012 |
| Medical | |||||||
| Cardiovascular imaging patients (9) | Cohort study: constructed from hospital discharge summary Quebec province-wide database | Cancer incidence; groupings by general anatomic sites but not type-specific | 12 020 incident cancers in 82 861 patients (77% underwent ≥ 1 cardiac imaging or therapeutic procedure) | Hospital admission and outpatient visits; review of diagnostic codes | Not applicable because hospital and outpatient clinic records were source of outcomes | 1996–2006 | 1996–2006 |
| French pediatric CT (10) | Cohort study: children who had first CT scan ≤10 y old identified from 23 radiology departments nationwide; cancer diagnosis, death, study end | Cancer incidence in children focusing on CNS, leukemia, lymphoma | 27 CNS, 25 leukemia, 21 lymphomas among 67 274 in cohort | National registry of childhood cancers | A: High-quality national; 2 = medium quality for vital status | 2000–2010 | 2000–2011 |
| UK pediatric CT (11) | Cohort study: children with first CT scan ≤22 y old identified from electronic radiology information systems at 81 hospitals throughout GB within National Health Service; cancer diagnosis, death, study end | Cancer incidence in children focusing on leukemia, MDS, CNS | 65 leukemias, 9 MDS, 135 CNS among 178 604 in cohort | National cancer registry, national mortality data, pathology reports from cancer registries and hospitals | A: High-quality national | 1980–2002 | 1980–2008 |
| PIRATES (low dose) (12) | Pooled data from 9 cohort studies of those undergoing irradiation in childhood: incident thyroid cancer, death, loss to follow-up | Thyroid cancer incidence | 252 cases in 2 588 559 p-y of follow-up in irradiated (184 cases in 2 114 683 p-y of follow-up for <0.1 mSv), 142 cases in 1 865 957 p-y of follow-up in nonirradiated cases | National cancer registries for 5 Nordic countries, hospital based for other 4 studies | Variable: national, regional, some not available | 1926–2000 | 1935–2009 |
| Occupational | |||||||
| Korean radiation workers (13) | Cohort study: assembled from central registry for radiation workers (legally mandated medical surveillance for radiation workers) | Cancer incidence (total and 6 specific types) and mortality | 935 deaths (256 total cancers, 9 leukemias) and 564 total incident cancers (14 leukemias) in 79 679 radiation workers vs 206 deaths (53 cancers) and 254 incident cancers in 190 816 auto manufacturing workers | Deaths: Korean National Statistical Office registry (>95% registration); incident cancers: national health insurance claim data (>99% complete) |
|
1984–2004 |
|
| Belarus, Russia, and Baltic CL (14) | Nested case-control study within 66 000 Belarus, 65 000 Russian, and 15 000 liquidators from Baltic countries; each case matched on age to 4 controls from liquidator population | Hematologic neoplasm incidence | 117 incident hematologic neoplasms (69 leukemias, 34 non-Hodgkin lymphoma and 14 others); analyses restricted to 70 cases (40 leukemias, 20 NHL, and 10 others) with reliable work information; outcomes reviewed by expert international pathologists’ panel | National cancer registries in Belarus and Baltic countries, National Medical Dosimetry Registry in Russia, and Registry of Hematologic Disorders in Belarus | A: Belarus and Baltic countries; not applicable in Russia | 1986–1987 | 1986–2006 |
| UK NRRW NW (15) | Cohort study: population identified from UKNRRW; cancer diagnosis, death, and end of follow-up | Total cancer incidence and total, excluding leukemia, and most specific neoplasms; mortality from all causes and most specific neoplasms | 26 731 deaths (8107 cancers, 216 leukemias) in 3 301 400 p-y of follow-up; 11 113 incident cancers, 267 leukemias | National cancer registry |
|
1946–2001 |
mortality: 1965–2001; incidence: 1971-2001 |
| Korean male NP NW (16) | Cohort study: assembled from nuclear workers at 20 plants who were issued dosimeters during 1978–2005; comparison group worked at same facilities but not issued a dosimeter; cohort members invited to participate in survey and clinical evaluation during 1992–2005 | Total cancer incidence and total excluding leukemia, leukemia and 4 other types of neoplasms | 96 cancers excluding leukemia (3 leukemias) identified in 8429 radiation workers and 101 cancers excluding leukemia (3 leukemias) in 7807 unmonitored workers | Korean Central Cancer Registry | A: High-quality national cancer registry | 1978–2005 | 1992–2005 |
| Rocketdyne NW (17) | Cohort study: identified from work history cards, electronic files, and radiation monitoring files from Radiation and Health and Safety Department; compared with workers at same facilities not monitored for radiation; death, age 95 y, loss to follow-up or end of study | Mortality from all causes, all cancers, leukemia, and most specific cancer types | 2354 deaths (672 total cancers, 33 leukemias) among 5743 radiation workers with 194 731 p-y of follow-up compared with 14 854 deaths (4163 cancers) among 41 169 unmonitored workers with 1 392 648 p-y of follow-up | Linkage of cohort with NDI, California Death statistical Master File, SSA death master file, other SSA files and commercial information service providers | 1: High-quality national vital status information | 1948–1999 | 1948–2008 |
| Japanese male NW (18) | Cohort study: nuclear workers identified from radiation worker registry; death, study end, date last known alive | Mortality from all causes, total cancers excluding leukemia, leukemia and some other selected cancer types | 2703 total cancers (2636 total cancers excluding leukemia, 80 leukemias) in 200 583 male workers with 1 373 000 p-y of follow-up | Deaths identified from residence registration cards (vital status) and linked with vital statistics death records (cause of death) | 3: Low-quality complete vital status registration | 1957–2002 | 1991–2002 |
| Canadian NW (19) | Cohort study; nuclear workers identified from National Dose Registry of all radiation-exposed workers | Mortality from all solid cancers and all excluding leukemia, leukemia | 437 solid cancers and 21 leukemias in 45 656 nuclear workers | Deaths identified from historic summary tax file and Canadian Mortality Data Base | 1: High-quality national vital status information | 1956–1994 | 1956–1994 |
| Ukrainian CL (20) | Nested case-control study: identified from 110 645 Ukrainian liquidators in Chernobyl State Registry of Ukraine; controls matched 5:1 (on place of residence, alive at corresponding case diagnosis, and year of birth using incidence density sampling | Incidence of leukemia and other hematologic neoplasms | 162 leukemias | Potential leukemia cases in worker cohort identified from all health care institutions in study area during 1986–2000; cases diagnosed 2001–2006 identified from linkage of worker cohort with Ukrainian Cancer Registry | A: High-quality national level since 1997 | 1986–1987 | 1986–2006 |
| German NP NW (21) | Cohort study: identified from comprehensive accident insurance and prevention list; death, study end, last available information date | Mortality from all cancers, total cancers excluding leukemia, solid cancers and cancer groupings related to exposure | 126 total cancer deaths; 7 deaths from leukemia | Causes of death from federal and state vital statistics offices | 2: Medium-quality vital registration | 1966–2008 | 1991–2008 |
| US NW (22) | Pooled cohort: assembled from 5 previously studied cohorts (employed >30 d, ever monitored, hired between facility start-up or when work history records first available); death, study end for those alive ≥1979 or last employment date when last known alive <1979 | Mortality from all causes, all cancers excluding leukemia, leukemia, several specified cancers, and cancer groupings related to exposures | 10 389 (9979 excluding leukemia and 410 leukemias) in 119 196 workers with 2 664 782 p-y of follow-up | Mortality and cause-of-death from previous studies used and extended through 2005 with linkage to SSA Death Master File and national tax records to confirm vital status; cause of death from NDI | 1: High-quality vital registration | 1944–2005 | 1944–2005 |
| INWORKS NW (23, 24) | Pooled cohort: assembled from updated cohorts of nuclear workers from France, UK, and US; date of death, study end, loss to follow-up | Mortality from all cancer, all cancer other than leukemia, and all solid cancer; mortality from leukemia excluding CLL | 19 748 all cancers (19 064 cancers other than leukemia, 17 957 solid cancers) among 308 297 workers with 8.2 million p-y of follow-up; 531 leukemias excluding CLL | Linkage with national and regional death registries, employer records, and SSA and NDI | 1: UK and USA; 2: France | 1944–2005 | 1944–2005 |
| US atomic veterans (25) | Cohort study: assembled from Nuclear Test Review Program Information System; included military members from previous follow-up study; date of death, study end, loss to follow-up | Mortality from all causes of death, all cancers, leukemia and some other specific cancers | 236 total cancers (19 leukemias) among 12 219 participants with 264 664 p-y of follow-up | Linkage of cohort with SSA Death Master ile and NDI; other sources included Internal Revenue Service, Department of Veterans Affairs Beneficiary Identification Record Location System | 1: High-quality vital registration | 1945–1963 | 1957–2010 |
| USRT (26–28) | Cohort study: assembled from American Registry of Radiologic Technologists | Incidence and mortality of breast cancer, incidence of basal cell skin carcinoma, and mortality of brain cancer | 1922 incident breast cancers among 66 915 female techs; 586 breast cancer deaths among 83 538 female techs; 3615 incident basal cell carcinomas of the skin among 65 719 Caucasian techs; 193 brain cancer deaths among 110 297 technologists | Self-report in 4 surveys with validation by medical records; linkage of cohort with SSA Death Master File and NDI |
|
1916–1997 | 1983–2008 |
| French NW (29) | Cohort study: pooled from 2 nuclear worker cohorts assembled in 1990s that were previously followed-up; death, study end, loss to follow up | Mortality from all causes, all solid cancers, leukemia and many other specific cancers | 2356 solid cancers and 57 leukemias excluding CLL among 59 004 workers with 1 469 949 p-y follow-up | National death registry and self-report | 2: Medium-quality vital status registration | 1950–2004 | 1968–2004 |
*International Agency for Research on Cancer quality of estimates for mortality and incidence data by country. Mortality: 1 = high-quality vital registration; 2 = medium-quality vital registration; 3 = low-quality vital registration; 4 = incomplete vital registration; 5 = other sources; 6 = no data. Incidence: A = high-quality national or high-quality regional data (coverage >50%); B = high-quality regional (coverage >10%); C = high-quality regional data (coverage <10%); D = national data (population-based cancer registries); E = regional data (population-based cancer registries); F = frequency data (hospital based or pathology based); G = no data. GLOBOCAN 2012 Classification (http://globocan.iarc.fr/Pages/DataSource_and_methods.aspx. ATA = at the time of the accident; CL = Chernobyl Liquidator; CLL = chronic lymphocytic leukemia; CNS = central nervous system; CT = computed tomography; GB = Great Britain; I = incidence; IARC = International Agency for Research on Cancer; INWORKS = International Nuclear Workers Study; M = mortality; MDS = myelodysplastic syndromes; NDI = National Death Index; NP = nuclear power; NW = nuclear worker; PIRATES = Pooled International Radiation and Thyroid Cancer Epidemiology Studies; p-y = person-years; RBM = red bone marrow; RT = radiologic technologist; SSA = Social Security Administration; TMI = Three Mile Island; UKNRRW = United Kingdom National Registry for Radiation Workers; US = United States of America; USRT = United States Radiologic Technologists.
In Table 3 the methods used for follow-up (active individual follow-up vs linkage with registries and other databases), loss to follow-up, the likelihood of major sources of error in cancer outcome assessment, and other classification issues were described. The potential for bias was determined based on the evaluation of the major sources of error listed in Table 1 and described for each study in Table 3. We concluded that the potential for bias was clear (with the designation of “yes”) and that there was no potential for bias (with the designation of “no”) or that we could not rule out that there was a potential for bias (with the designation of “possible”), but these designations were somewhat subjective. The designation of “possible” was usually due to an absence of sufficient detail about the methods employed to reach a conclusion about the potential for bias.
Table 3.
Follow-up methods and results of loss to follow-up, ascertainment type and completeness, classification issues, and potential for bias in 26 low-dose radiation epidemiological studies
| Study name (reference) | Method of follow-up: active vs linkage | Loss to follow-up* | Likelihood of
|
Other classification issues | Comment | Potential for bias (Y/N possible) |
|---|---|---|---|---|---|---|
| Environmental | ||||||
| Chernobyl-exposed children (1) | NA* | NA |
|
Expert hematologists reviewed each case | Authors state: high degree of completeness of case ascertainment but no further details; Ukraine component had higher percent of controls than cases from less contaminated regions | No |
| TMI residents (2) | Linkage | 1.6% lost to follow-up; annual follow-up via post office address change |
|
TMI population registry matched annually with PA state cancer registry from 1982 onward | Population-based cancer registry not established until 3 y after accident: possible incomplete ascertainment of leukemia | No |
| Chinese background (3) | Active | Villages visited annually to assess cancer mortality; 1.6% lost to follow-up in higher background radiation and 2.8% in lower background radiation regions, both mostly due to emigration |
|
Basis of diagnosis of variable quality | No national mortality registers. Some diagnoses from hospital records, but others from interviews of village doctors, family, or next of kin; thus, ascertainment of cancer incidence and mortality, particularly for leukemia, likely incomplete, potentially greater in the lower background radiation region. | Possible for both solid cancers and leukemia. Uncertain direction of bias because false negatives could be substantial in both higher and lower background radiation regions. |
| GB background (4) | Linkage | NA |
|
Specialized pediatric cancer registry used; cases classified using ICCC | No | |
| Swiss children background (5) | Linkage | NK*†; 1.7% excluded due to uncertain residence |
|
<4% Potential false positives; >400 cancers not linked that are potentially false negatives; ≤7% linkages resulted in false matches | Authors state: no evidence that false negatives differed from linked cases in radiation exposure | No |
| Techa River residents (6) | Active: case ascertainment from oncology clinics, medical records, and death certificates (latter percent declined over time) | NK‡; outcome available on 73% of cohort; 21% periodically lived outside study region; 5.7% nonmigrants lost to follow-up |
|
20% incident cancers were death certificate diagnoses | Authors state: no indication that migration confounds results, although other studies suggest that healthier persons more likely to emigrate | No |
| Finnish background (7) | Linkage | NA; very small numbers of exclusions (0.0036% prohibited use of data; residence data incomplete for 3.3-4.4%) |
|
No | ||
| Taiwanese residents (8) | Linkage | NK§; residential history based on interview; loss to follow-up not stated |
|
Improving quality of cancer registry over time; ICD-O first edition used | Incomplete and missing occupancy data | No |
| Medical | ||||||
| Cardiovascular imaging patients (9) | Linkage | NK‖; loss to follow-up not reported |
|
No estimates provided for total cancers, total solid cancers, or leukemia. Nonstandard anatomic site groupings used to report outcomes. | Possible. Increased risk of solid cancer may have been observed in those exposed due to increased imaging | |
| French pediatric CT (10) | Linkage | NK¶; loss to follow-up not reported |
|
No | ||
| UK pediatric CT (11) | Linkage and medical record review | Unlikely to bias results |
|
15.7% of potentially eligible patients identified in Radiology Information Systems database were excluded because could not be traced using National Health Services Central Register, but unlikely to differ by exposure status | No | |
| PIRATES (low dose) (12) | Pooled analysis: outcomes for 4 cohort outcomes identified via linkage; 4 cohorts through active follow-up; 1 cohort (a-bomb) via linkage from 1958 onward | NK**; loss to follow-up not reported |
|
No major changes in classification of total thyroid cancer over lengthy follow-up period; histologic confirmation required for cases | No | |
| Occupational | ||||||
| Korean radiation workers (13) | Linkage | NK††; loss to follow-up not reported |
|
Use of a modification of ICD-10 that is closer to ICD-9 for hematopoietic neoplasms | Follow-up (mortality: 1992–2004; incidence: 2000–2005) began several years after dose monitoring was initiated | Possible. Increased solid cancer and leukemia risks due to required medical surveillance of those exposed, whereas not required in unexposed |
| Belarus, Russia, and Baltic CL (14) | Linkage | NA |
|
Expert hematopathologists reviewed each case | No | |
| UKNRRW NW (15) | Linkage | Unlikely to bias results |
|
Refusal rate for participation approximately 1% | No | |
| Korean male NP NW (16) | Linkage | NK‡‡; loss to follow-up not reported |
|
None; ICD10 used | Follow-up (1992–2005) began 14 y after dose monitoring initiated | No |
| Rocketdyne NW (17) | Linkage | Unlikely to bias results; 0.6% loss to follow-up |
|
ICD9 | No | |
| Japanese male NW (18) | Linkage | NK§§; 27.4% loss to follow-up for period before prospective follow-up (1986–1991) but 0.3% subsequently (1991–1997) |
|
ICD9 | Loss to follow-up was higher in early years when doses were higher | Possible. Incomplete assessment of risks of solid cancers and leukemia due to loss to follow-up of early workers |
| Canadian NW (19) | Linkage | 97.6% vital status ascertainment |
|
ICD9 used to recode underlying cause of death from original ICD code in use at time of death | Cause known for 99.9% of deaths; a subset of workers (AECL) had problematic dosimetry and were therefore excluded. | No |
| Ukrainian CL (20) | Medical facility follow-up before 2001, linkage after | NA |
|
Expert hematopathologists reviewed each case | Reflects high level of medical surveillance: unexpected similar radiation-related risks for CLL and non-CLL leukemias. Unexpectedly high proportion of CLL cases in study (58%) compared with 40–44% in general populations of European descent in other countries | No |
| German NP NW (21) | Follow-up through linkage of last known residence with vital status | NK‖‖; low loss to follow-up (0.7%) |
|
ICD9 and ICD10 | Follow-up (1991–2008) began 25 y after dose monitoring initiated | No |
| US NW (22) | Linkage | 1.1% lost to follow-up |
|
Used mortality/cause of death ascertained in earlier follow-up. Confirmed vital status with SSA Death Master File, cause of death from the NDI, and newly identified deaths coded to the latest ICD revision. | No | |
| INWORKS NW (23, 24) | Linkage | Loss to follow-up not described for pooled cohorts |
|
See French nuclear workers below. Ascertained vital status through linkage of participating French, UK, and US cohorts with employer and SSA records. Identified cause of death through linkage with national and regional cancer registries. | Certain outcomes (smoking- and asbestos-related cancers) may have reflected confounding of radiation-related dose response by these exposures | No |
| US atomic veterans (25) | Linkage | Vital status determined for 95.3% |
|
Follow-up (1957–2010) began 12 y after first exposure; deaths occurring pre 1957 not identified. Findings for certain malignant and nonmalignant outcomes and lack of healthy worker effect observed among participants in 1 test (designated SMOKY) but not others. | No | |
| USRT (26–28) | Linkage to national mortality register; questionnaire-based ascertainment of incidence | NK¶¶; <0.2% loss to follow-up |
|
|
No | |
| French NW (29) | Linkage | NK***; 0.2% of workers lost to follow-up |
|
Causes of death were coded to ICD revision in effect at time of death. | French National Death Registry established1968; prior deaths not identified. Potential confounding by socioeconomic status. Many contractor workers not included because of poor follow-up, incomplete company records, problematic identifying information, and incomplete dosimetry reconstruction. | No |
Notes: In the fourth column (differential/incomplete ascertainment and false negatives or positives), definitions of the terms used include: likely = evidence there was a problem; unlikely = evidence there was no problem; possible = could not rule out a problem; downward arrows = evidence of notable reduced ascertainment; upward arrows = evidence of notably increased ascertainment of clinically nonapparent or precancer neoplasms. The basis of these qualitative assessments is described in the sixth column (comments). In the seventh column (potential for bias), definitions of the terms used include: yes = potential for bias is evident; no = potential for bias is not evident; possible = could not rule out the potential for bias. Estimates of differential or incomplete ascertainment, false negative, or false positive outcomes are qualitative and somewhat subjective. Quantification of these elements is not possible for almost all of the studies due to absence of critical information about methods.
AECL = Atomic Energy of Canada Limited; CL = Chernobyl liquidators; CLL = chronic lymphocytic leukemia; GB = Great Britain; ICCC = International Childhood Cancer Classification; ICD = International Classification of Diseases; ICD-O = International Classification of Diseases for Oncology; INWORKS = International Nuclear Workers Study; NK = not known; NP = nuclear power; NW = nuclear workers; TMI = Three Mile Island; UK = United Kingdom; UKNRRW = United Kingdom National Registry for Radiation Workers; US = United States; USRT = US Radiologic Technologists.
NK: Swiss children background cohort identified and residence determined from 1990 and 2000 national censuses and linked with the Swiss Childhood Cancer Registry (the latter estimated >90% complete). Unclear residential history between censuses because active follow-up was not undertaken, but the Swiss National Cancer Registry includes all residences from diagnosis back to birth and is therefore potentially more complete.
NK: Techa River residents study had loss to follow-up of 15% in 2007 publication that declined to 5.7% in the current publication; results from studies of migrants suggest that migration wqas more likely by healthier persons, although the authors explicitly indicated it was unlikely that migration acts as a confounding factor.
NK: Taiwan residents’ cohort was established in 1992, interviewed in 1995–2000 to reconstruct residential history from 1983 onward; incomplete and missing occupancy data resulted in exclusion of 1020 cohort members from 7262 registered as living in the building.
NK: Quebec (Canada) patients newly diagnosed with myocardial infarction (77% undergoing diagnostic or therapeutic procedures involving radiation exposure) were not actively followed-up for cancer and deaths; linkage for outcomes was limited to Quebec insurance databases for which no information was provided about level of completeness; no information about loss to follow-up or emigration.
NK: French pediatric CT population followed-up using vital status register designated medium quality; loss to follow-up not stated.
NK: PIRATES, likely highly complete identification of thyroid cancer incidence from four cohorts linked with high-quality nationwide population-based and cancer registries and the a-bomb survivors from 1958 onward when high-quality cancer registries were available; likely underascertainment for four cohorts with other methods of follow-up and before 1958 for the a-bomb survivors.
NK: Korean radiation workers were defined as all workers under medical surveillance because of ionizing radaition exposures as required by law, but there was little information about the source and completeness of this cohort. Follow-up began at the date of first exposure surveillance (as early as January 1984 or later) or January 1992, whichever was later. Hence, outcomes that ocurred before 1992 were not considered.
NK: Korean nuclear power industry workers radiation workers were defined as those who were issued with a dosimeter at nuclear power facilities until 2005. Follow-up for cancer incidence was undertaken for male nuclear power industry workers who were compared with unmonitored workers from the same facilities during the period 1992–2005. Outcomes that occurred before 1992 were not considered.
NK: Japanese nuclear workers were characterized by 27.4% loss to follow-up during 1986–1991 due to incorrect address information provided by nuclear facility or lack of availability of residential addresses, which were only maintained for 5 years; only 0.3% loss to follow-up during 1991–2002 because follow-up was initiated in 1991 and residential addresses were fully available during the 1991–2002 follow-up.
NK: German nuclear workers were followed-up by search of the regional population registry based on the last known address to determine vital status. No information about loss to follow-up or inability to identify workers in the regional population registries. Dose monitoring was initiated 25 years before follow-up was launched. Deaths occurring before 1991 were not identified.
NK: For the USRT, annual individual dose estimates were reconstructed for the period 1926–1997, although follow-up did not commence until completion of a baseline questionnaire (eg, 1983–1989 or 1994–1998, depending on which “baseline” questionnaire patients completed). The historical dose reconstruction required incorporation of questionnaire (work history, work procedures, radiation protection) information because badge doses were missing for a substantial fraction of technologists, particularly for the period before 1977.
NK: In the French nuclear workers, dose monitoring was initiated 18 years before follow-up was possible because a national death register was only available from 1968 onward. Deaths before 1968 were not systematically identified, but an earlier analysis revealed that the standardized mortality ratio for the period 1946–2004 was similar to that for the period 1968–2004 (49).
Results
Study Design and Major Category of Outcome
As shown in Table 2, 21 of the 26 studies employed a cohort design and generally followed-up persons for incident cancers, deaths from cancer and other causes of death, or study end, although a few ended the cohort follow-up at a predetermined age (3,5,10,17) or at the time of emigration (5,6).
Cancer incidence was the outcome for all studies of medical radiation exposure and all but one study of environmental radiation exposure (Table 2). Six cancer incidence studies focused primarily on pediatric radiation exposures and pediatric cancer outcomes (1,4,5,7,10,11) (although follow-up of study individuals in the UK computed tomography study extended into adulthood) (11), one on pediatric radiation and thyroid cancer at any age (12), a few on single adult cancer outcomes including two on leukemia (14,20), one each on breast (26) and skin cancers (27), and one on poorly defined cancer incidence outcomes not specified by recognized cancer categories (9). In contrast, seven (17–19,21,23,25,29) of the 14 occupational radiation studies assessed only mortality risks. Cancer incidence is a preferred outcome particularly for cancers with long survival and/or indolent clinical course, but the absence of national population-based cancer registries may preclude the feasibility of assessing incidence rather than mortality.
Cancer Incidence Studies
Population-based cancer registries were the primary source of cancer incidence ascertainment in 13 of 17 populations; 12 (1,4,5,7,8,10–12,14–16,20) were linked with national and one (2) with regional cancer registries. The remaining four identified cancer outcomes from national health insurance data (13), hospital discharge and outpatient clinic information (9), self-administered questionnaires (26,27), or multiple sources (6); two (6,26,27) of the four studies were carried out in geographic regions lacking national cancer registries. IARC rated cancer registries as high quality in all but two studies (5,12) based on coverage greater than 50%, although the high-quality designation for some registries to which the cohorts were linked (8,14) applied only to the latter part of follow-up.
Among the 17 cancer incidence studies, 13 reported leukemia incidence (1,2,4,5,7,8,10,11,13–16,20). The six pediatric leukemia studies included expert pathology review (1) and use of the International Childhood Cancer Classification (4,5) or the current International Classification of Diseases for Oncology (7,10,11), whereas five (2,8,13,15,16) of the seven studies reporting leukemia incidence in adults used older classifications and only two used expert pathology review (14,20).
Exposure periods for the 26 populations spanned from 1926 through 2011. However, follow-up periods were generally within the period of 1974 to 2011 or earlier for all but four studies (6,12,15,26–28). Follow-up for cancer incidence generally began later than for mortality because population-based cancer registries were established more recently than mortality registries [e.g., in the UK radiation worker study (15), mortality and cancer incidence ascertainment began in 1955 and 1971, respectively] (50). Follow-up was 20 years or less for all but five (6,8,11,12,15) of the 12 cancer incidence cohort studies.
Cancer Mortality Studies
Eleven of 12 cancer mortality studies [except for the China background study (3)] linked the study populations with national death registries. The cancer mortality registries were mostly characterized as high quality by IARC except for two designated as medium (21,29) and one low quality (18). Although active individual follow-up was employed in the Chinese background study (3), the absence of population-based cancer incidence and mortality registries, variable cancer outcome ascertainment sources, and lack of histopathologic confirmation likely led to incomplete vital status registration. In addition, the lack of blinding of members of the Chinese data collection team to natural background radiation exposure status may have led to differential cancer ascertainment in the geographic region with higher background radiation vs that in the region with lower background radiation.
Eleven of the 12 cancer mortality studies reported leukemia risks (3,13,15,17–19,21,22,24,25,29). Death certificates may lack altogether or provide incompletely specified leukemia subtype, particularly in earlier decades. None of the leukemia mortality studies used newer classifications for myeloid or lymphoid neoplasms, although most reported risks for leukemia excluding CLL and a few reported CLL risks separately (data not shown). For cancer outcomes, the emphasis of the current analysis was on total cancer or all solid cancers, and thus there should be little misclassification resulting from lack of information about type of solid cancer and level of specification of specific solid cancers. Estimates of radiation dose response for total cancer would include leukemia, but the rarity of leukemias compared with total solid cancer would result in relatively little effect of inclusion vs exclusion of leukemias. Clinico-pathologic characteristics of leukemias differ notably from nonhematopoietic solid cancers, so misclassification of these two major cancer groupings is highly unlikely.
Exposure periods for the cancer mortality studies ranged from 1916 to 2008. A majority included persons first exposed as early as the 1940s (15,22–29) and 1950s (16,18,19). Follow-up periods were 36 years or more for six cohorts and 11–25 years for the others.
Follow-Up, Ascertainment, and Classification Issues and Potential for Bias
Follow-Up Issues
Only two cancer incidence studies (6,12) and two cancer mortality studies (3,22) carried out active follow-up of individual persons. Information about loss to follow-up was not reported in the majority (N = 16) of the 26 studies (Table 3). For those incidence studies that did report loss to follow-up, most reported loss to follow-up of less than 5%. Exceptions included the 21% emigration in Techa River residents (6) and 38% loss-to-follow-up among the Taiwanese residents, many of whom were students who only resided a few years in the contaminated buildings (8). Most cancer mortality studies reported loss to follow-up of less than 5% (3,17,19,21,22,25–28); however, loss to follow-up was reported to be “low” but not specified in the French nuclear workers (29), and the Japanese nuclear worker study reported loss to follow-up of 27.4% (18). In the Chinese background study (3), the methods for follow-up of mortality changed part-way through the study.
Close to one-half of the studies including incidence (5,6,8–10,12,13,16,26,27) and mortality outcomes (13,18,21,26,28,29) had methodological features that potentially precluded accurate quantification of loss to follow-up (see Table 3). Loss to follow-up will be greater if the approach used is “passive individual follow-up” rather than active individual follow-up as described above. Loss to follow-up was likely to be greater in studies when exposure periods were evaluated before registers were established to enable systematic follow-up of incidence (13) and mortality outcomes (13,16,18,21,23,24,26,28). Validity would only be affected if follow-up completeness was related to dose; control for calendar year period would be helpful in this situation. Incomplete information about residential history (5,8) may also have affected completeness of follow-up.
Ascertainment Issues
Incomplete ascertainment of cancer was likely in three cancer incidence studies (9,14,26) and two cancer mortality studies (3,18) and was possible in several other cancer incidence (5,6,8,12,13,20) and cancer mortality studies (13,21,28) (Table 3). Four studies appeared to be characterized by differential ascertainment associated with radiation level (3,9,13,18).
Classification Issues
False negatives (ie, underdiagnosis) were likely in two studies with cancer incidence (6,11) and two with cancer mortality outcomes (3,13) and possible in other cancer incidence (5,8,9,12,26) and cancer mortality studies (17,18,21–26,28,29) (Table 3). The occurrence of false negatives does not result in bias if this shortcoming is not related to radiation exposure status by radiation dose levels. False positives (ie, overdiagnosis) were likely in one cancer incidence study (9) and possible in other cancer incidence (5,6,12,26) and cancer mortality (28) studies. False positives in cancer mortality studies could occur if metastases to a site (eg, brain, liver, or other) are reported as the primary site. False positives could result in bias if occurring differentially according to radiation exposure status, but false positives are less likely to be a problem in mortality studies of “all cancer” as an outcome compared with studies focusing on specific cancer outcomes. Only the three case-control studies of the Chernobyl residents and liquidators’ populations (1,14,20) included expert hematopathologists’ review. Several long-term studies reclassified solid cancers using the latest ICD classification (19,22,23,26). Two studies had potential variability in the accuracy of diagnoses, mostly due to different sources of cancer outcome data (3,6), and one study reported risks for anatomic site groupings not routinely used (9).
Potential for Bias
Four studies were judged to be possibly biased in cancer outcome assessment (3,9,13,18). Three of the studies were also possibly biased in outcome assessment due to the potential for differential levels in medical surveillance in the radiation-exposed vs the comparison populations in these studies (3,9,13) (Table 3). In the China background study, differential levels of surveillance used in the geographic regions with higher vs lower background radiation along with lack of population-based mortality registers may have resulted in overestimation of the excess relative risk for leukemia (3). Use of diagnostic or therapeutic imaging procedures along with cancer outcome reporting by nonstandard anatomic site groupings in the cardiovascular imaging study would have been more likely to result in differential ascertainment and/or misclassification in the exposed population but not the comparison population (9). For the study of Korean radiation workers, medical surveillance was required for radiation workers but not for the comparison group of motor vehicle manufacturing workers (13). Akiba and Mizuno refer to the article describing the earlier follow-up of the Japanese nuclear workers for details about the methods (51). As noted by Iwasaki et al. (51), at the start of cohort follow-up of the Japanese nuclear workers, more precise information on residential address (the primary method of follow-up) could be obtained for workers who were alive and currently working compared with those who were deceased. The investigators stated that “since currently employed workers were included more frequently in the higher-dose group than the lower-dose group, differential follow-up rates were introduced in the ‘retrospective’ (1986–1990) component of the study” (51). Although difficult to estimate, the net effect is likely to be attenuation of the excess relative risk per unit radiation dose through early loss to follow-up during periods with higher radiation exposures. The investigators indicated that the prospective component of the follow-up was less affected by this selection bias; the more recent report of the extended follow-up excluded the retrospective (1986–1990) component, restricting follow-up to 1991 and later (18).
Discussion
The systematic evaluation of cancer outcomes in 26 studies of low-dose or low-dose rate external radiation revealed that the majority of studies lacked information to enable comprehensive evaluation of major sources of cancer outcome assessment. For most of the studies, the outcome information was generally ascertained from high-quality, population-based cancer registries (76%) or mortality registries (67%). Follow-up rates were not described in the majority (62%) of the 26 studies, and only one of the studies using registry linkage provided a critical assessment of the quality, completeness, or accuracy of the vital status, cancer, or mortality registries used (22). Underascertainment of outcomes (false negatives) was likely if populations could not be linked with high-quality national registries that were lacking altogether, for example, in the United States (52,53), or not established during early exposure years (13).
Linkage may not have been feasible in studies lacking detailed individual residential history (5,8) or those with substantial emigration (6), thus affecting tracing (eg, obtaining detailed personal identifier information to enable registry linkage and locating and contacting a person to verify that personal identifier information is correct and, for some studies, to request participation in a questionnaire survey). Incomplete ascertainment of myeloid neoplasms, such as MDS and myeloproliferative disorders, due to misclassification or changing classification was likely in early years as discussed below. False positives may have occurred as a result of cancer ascertainment from self- or next-of-kin report (3,26,27), through misclassification of benign entities as malignant, or inclusion of screen-identified cancer precursors or borderline neoplastic disorders in studies with enhanced medical surveillance (9). All pediatric leukemia studies had expert hematopathology review or used the most recent International Classification of Diseases for Oncology or International Childhood Cancer Classifications (54), whereas the adult studies of leukemia incidence and mortality often used older classifications. Some, but not all, long-term studies recoded solid cancers to the latest ICD classification in use. These outcome ascertainment shortcomings were mostly nondifferential by radiation exposure level and either had little effect or reduced risk estimates towards the null. We concluded that the Chinese high background radiation-exposed residents (particularly for leukemia mortality ascertainment) (3), cardiovascular imaging patients (9), Korean radiation workers (13), and Japanese nuclear workers (18) may have had cancer ascertainment that was possibly differential by radiation exposure level.
Linkage of a large US cohort with high-quality, population-based cancer registries found close to 90% of medical record-verified cancers to have been identified in the registries (55). Comparison of questionnaire-derived self-report with registry report revealed 21–39% underreporting (56–59), particularly noting this problem for myeloid neoplasms (60) that may have led to underascertainment of myeloid leukemias and the related MDS. The reported international variation in underreporting as summarized by Navarro et al. (61) and particularly poor reporting of primary liver and uterine cervical cancer (62) should not have had a major impact on the studies evaluating nonleukemia cancers in the current review because the primary outcomes focused on total cancers or total solid cancers. Death certificates may be inaccurate and incomplete for some causes of death based on comparisons with autopsy studies (63), but comparison of breast cancer on death certificates with hospital discharge records revealed only a modest level of misclassification and no statistically significant differences between screen-detected and nonscreen-detected cases (64). All-cancer sites combined (or those representing broad site groupings) have been found to have relatively high sensitivity and specificity on death certificates (63,65,66).
For the minority of studies reporting loss to follow-up, the fraction of persons “lost” was generally less than 5%. If this percent is an accurate estimate, such a relatively small loss should have little impact on true radiation-related risk estimates, whereas a loss to follow-up of 20% or greater could lead to bias in risk estimates based on simulation studies because persons lost to follow-up are generally not missing at random (67). Even if persons appear to be missing completely at random, investigators have found that as loss to follow-up increases, the precision of the risk estimate decreases (68). Nevertheless, even if reported loss to follow-up is very small, all incident cancers in the cohort may not be identified in the absence of active individual follow-up and/or linkage with highly complete vital status and population-based registries.
Because loss to follow-up is generally not random, those who are lost often have notably different profiles than those who are followed up (67). In addition to the healthy worker or person effect, comparison of “exposed” workers or “patients” with the general population may be problematic because the “exposed” population undergoes tracing (eg, follow-up of individual persons to identify vital status, diagnosis and date of diagnosis of incident cancers, and cause of death and date of death for deceased cases) whereas the general population does not (69). In addition to some persons missing from registries, emigration may also contribute to loss to follow-up, but this issue has been evaluated in relatively few studies (70–72). Although some studies have reported similarities between emigrants and nonemigrants in serious health outcomes, others have reported that emigrants may have higher levels of skills (70) and better health than those who remain behind (73) but that emigrants who are very ill may return to their country of origin to be with family close to death (73,74).
False negatives are largely due to loss to follow-up and other forms of incomplete ascertainment as described above but may result from changes in classification. Substantial changes were made in the landmark 2001 World Health Organization classification of hematopoietic and lymphoproliferative neoplasms that included reclassification of MDS (previously known as “refractory anemia”) and myeloproliferative neoplasms as malignancies (75). These myeloid neoplasms were formerly and even recently frequently diagnosed as benign hematological disorders and often worked up in physicians’ offices or misdiagnosed and not reported to population-based registries (60). Myelodysplastic syndromes and other nonleukemia myeloid neoplasms may transform to acute myeloid leukemia (AML); certain systemic chemotherapy agents, benzene, and ionizing radiation may cause both AML and MDS (76). Therefore, epidemiological studies have increasingly combined MDS with AML to evaluate these associations (77,78). Ascertainment of MDS and myeloproliferative neoplasms is a potential problem for both incidence and mortality studies but is a greater problem for the latter due to poor specificity of these entities on death certificates.
Radiation epidemiology mortality studies that rely on death certificates have struggled with the approach for handling indeterminate or incompletely specified leukemias, particularly in long-term studies due to major changes in leukemia classification during the past 50 years. Because leukemia latency is short and occupational radiation doses were higher in the middle decades of the 20th century than subsequently, it is possible that poor leukemia subtype classification in the early years of follow-up may cause bias in subtype-specific risk estimates. For example, the US pooled nuclear worker study (22), which followed a large cohort of workers for up to 60 years, found that the excess relative risk per Sievert for non-CLL leukemia (eg, used in the present systematic appraisal) increased by 50–80% when excluding leukemias of indeterminate subtype, all of which occurred in years covered by ICD 6th and 7th revision classifications (79). Although the risk per unit dose was higher following the exclusions, the small percent and numbers of cases excluded (22 of 206 total non-CLL leukemia cases) limited firm conclusions about potential bias.
False positives are a possibility if screening radiation-exposed persons, but not the comparison unexposed persons or other referent population, resulted in higher levels of reporting of “in situ” neoplasms or precursors that are classified as malignancies but might not otherwise be identified in an unscreened population and may undergo involution or might never come to clinical attention. Other screen-detected entities might be misclassified as malignancies (eg, certain types of colon adenomas or borderline ovarian tumors). If the category “all leukemias” is used, particularly before 2001, and there is no exclusion for CLL, there could be false positives because CLL has been classified as a form of non-Hodgkin lymphoma since 2001.
Strengths of this evaluation included the comprehensive nature of the outcome assessment focusing on reported and unreported aspects of loss to follow-up, under- and overascertainment, misclassification, and changing classification. Comprehensive reviews of the radiation epidemiology literature for adults (30,80–84) and children (85) have not previously evaluated critically the key elements of outcome assessment as described in this article.
Limitations derive primarily from lack of detailed description of methods in the reports for the 26 studies about the potential for missing vital status, absence of information about the methods used and the results of loss to follow-up for the majority of studies, incomplete cancer or mortality registry data for an unknown fraction of population members in linkage studies, unknown levels of outcome misclassification, or, for several cohorts, a description of specific methods for outcome ascertainment for exposure periods before establishment of registries. Lack of active follow-up in all but four studies and the resulting underascertainment of cancer outcomes affects statistical power. If the level of active follow-up is differential between radiation-exposed and comparison populations, such a difference in level of effort could limit several aspects of outcome assessment. The classification of cancer and mortality registries according to quality criteria by IARC was primarily driven by more recent performance; the IARC classification did not include a comprehensive historical evaluation since the start of the registries. It is therefore possible that the quality measure assigned may not reflect earlier quality levels. Substantial reporting of large and heterogeneous categories (all-cancer sites combined, all solid cancers, all leukemias) in most of the studies focusing on multiple outcomes, which were likely related to the radiation protection emphasis and to limited statistical power, provided little consideration of morphologic subtypes and virtually no consideration of molecularly characterized tumor subtypes, which may differ in radiation-related risks. In the absence of detailed information about key elements needed to comprehensively and quantitatively evaluate all major sources of potential outcome assessment errors, the characterization of the potential likelihood of differential or incomplete ascertainment or false negative or false positive outcomes and the potential for bias is necessarily qualitative and somewhat subjective in nature. From the data presented and missing information as described above, it was not possible to quantify the impact of some of the key shortcomings.
Validation methods and statistical approaches might in some cases be used to reduce the effect of the outcome assessment limitations and errors described above. Comparison of diagnoses from registries and other electronic databases with persons' medical records could provide information that would allow for correction of errors in outcome classification. In addition, completeness of accurate outcome ascertainment can be improved by including medical record-confirmed cases that are listed by nonspecific and secondary codes in registries or electronic databases that are the source of case ascertainment (86). For problems with incomplete follow-up, adjustment of person-years by censoring could be useful (87). In studies with the follow-up curtailed due to emigration, one should censor follow-up at the time of emigration if the date is known (6); if unknown, it may be possible to estimate period, sex, and exposure status–dependent person-year adjustment factors to use in the analysis (41). If there are variables that are related to both exposure levels and completeness of case ascertainment, then one might consider adjusting the baseline risk for these variables (88). Following implementation of substantial improvements in outcome assessment in ongoing and future epidemiological studies, it would be important to use quantitative bias analysis to provide more definitive estimates of the direction, magnitude, and uncertainty arising from systematic errors (89).
In conclusion, the evaluation of outcome assessment in 26 studies of low-dose or low dose–rate radiation exposure and cancer risks has identified several methodological issues. Despite the shortcomings in outcome assessment, only four studies (3,9,13,18) may have had the potential for bias, but evidence is not conclusive. The findings from this evaluation point to the need to implement more rigorous active follow-up of exposed and comparison persons to clearly establish vital status such as used in the cohort study of US nuclear workers (22). Cohort studies also need to report not only loss to follow-up but also the details of the completeness and accuracy of the population-based cancer and mortality registries used for linkage and determination of outcome status. In addition, efforts should be undertaken to implement expert pathology review, particularly for hematopoietic and lymphoproliferative neoplasms; this may not be possible for mortality studies with follow-up beginning decades ago. Ongoing follow-up is also needed to assess radiation-related cancer risks at older ages in many radiation-exposed populations, such as nuclear workers and medical radiation workers, particularly those performing newer diagnostic and therapeutic procedures using imaging technologies. Use of multiple sources of validation and state-of-the-art statistical approaches that may help to overcome outcome assessment problems and errors, as described above, is also encouraged.
Funding
This work was supported by funding from the Intramural Research Program of the National Institutes of Health, National Cancer Institute, and the US Public Health Service of the Department of Health and Human Services.
Human subjects
This project was conducted at the National Cancer Institute, National Institutes of Health, and US Department of Health and Human Services. All data were abstracted from the published literature; therefore, no approval from the NIH human subjects institutional review board was required.
Declaration of interests: All authors (MSL, MKSB, and ABdeG) declare no conflicts of interest. The views expressed are those of the authors and do not necesarily represent the decisions, policy, or views of the National Cancer Institute, National Institutes of Health, US Department of Health and Human Services, or the International Agency for Research on Cancer.
Notes
Affiliations of authors: Radiation Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, NIH, DHHS, Rockville, MD (MSL, ABdG); Monographs Programme, Evidence Synthesis and Classification Section, International Agency for Research on Cancer, Lyon, France (MKSB). We are grateful for the suggestions provided by Drs. Robert D. Daniels, Ethel S. Gilbert, Michael Hauptmann, Gerald Kendall, Mark P. Little, Jay Lubin, and Dale L. Preston. We thank Ka Lai Lou, Maura Kate Costello, and Justine Yu for technical assistance in manuscript preparation.
References
- 1.Davis S , Day RW, Kopecky KJ, et al.; International Consortium for Research on the Health Effects of Radiation Writing Committee, Study Team. Childhood leukaemia in Belarus, Russia, and Ukraine following the Chernobyl power station accident: results from an international collaborative population-based case-control study. Int J Epidemiol. 2006;35(2):386–396. [DOI] [PubMed] [Google Scholar]
- 2.Han YY , Youk AO, Sasser H, et al. Cancer incidence among residents of the Three Mile Island accident area: 1982-1995. Environ Res. 2011;111(8):1230–1235. [DOI] [PubMed] [Google Scholar]
- 3.Tao Z , Akiba S, Zha Y, et al. Cancer and non-cancer mortality among inhabitants in the high background radiation area of Yangjiang, China (1979-1998). Health Phys. 2012;102(2):173–181. [DOI] [PubMed] [Google Scholar]
- 4.Kendall GM , Little MP, Wakeford R, et al. A record-based case-control study of natural background radiation and the incidence of childhood leukaemia and other cancers in Great Britain during 1980-2006. Leukemia. 2013;27(1):3–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Spycher BD , Lupatsch JE, Zwahlen M, et al. for the Swiss Pediatric Oncology Group. Background ionizing radiation and the risk of childhood cancer: a census-based nationwide cohort study. Environ Health Perspect. 2015;123(6):622–628. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Davis FG , Yu KL, Preston D, et al. Solid cancer incidence in the Techa River incidence cohort: 1956-2007. Radiat Res. 2015;184(1):56–65. [DOI] [PubMed] [Google Scholar]
- 7.Nikkila A , Erme S, Arvela H, et al. Background radiation and childhood leukemia: a nationwide register-based case-control study. Int J Cancer. 2016;139(9):1975–1982. [DOI] [PubMed] [Google Scholar]
- 8.Hsieh WH , Lin IF, Ho JC, et al. 30 years follow-up and increased risks of breast cancer and leukaemia after long-term low-dose-rate radiation exposure. Br J Cancer. 2017;117(12):1883–1887. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Eisenberg MJ , Afilalo J, Lawler PR, et al. Cancer risk related to low-dose ionizing radiation from cardiac imaging in patients after acute myocardial infarction. CMAJ. 2011;183(4):430–436. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Journy N , Rehel JL, Ducou Le Pointe H, et al. Are the studies on cancer risk from CT scans biased by indication? Elements of answer from a large-scale cohort study in France. Br J Cancer. 2015;112(1):185–193. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Berrington de Gonzalez A , Salotti JA, McHugh K, et al. Relationship between paediatric CT scans and subsequent risk of leukaemia and brain tumours: assessment of the impact of underlying conditions. Br J Cancer. 2016;114(4):388–394. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Lubin JH , Adams MJ, Shore R, et al. Thyroid cancer following childhood low-dose radiation exposure: a pooled analysis of nine cohorts. J Clin Endocrinol Metab. 2017;102(7):2575–2583. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Ahn YS , Park RM, Koh DH. Cancer admission and mortality in workers exposed to ionizing radiation in Korea. J Occup Environ Med. 2008;50(7):791–803. [DOI] [PubMed] [Google Scholar]
- 14.Kesminiene A , Evrard AS, Ivanov VK, et al. Risk of hematological malignancies among Chernobyl liquidators. Radiat Res. 2008;170(6):721–735. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Muirhead CR , O'Hagan JA, Haylock RG, et al. Mortality and cancer incidence following occupational radiation exposure: third analysis of the National Registry for Radiation Workers. Br J Cancer. 2009;100(1):206–212. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Jeong M , Jin YW, Yang KH, et al. Radiation exposure and cancer incidence in a cohort of nuclear power industry workers in the Republic of Korea, 1992-2005. Radiat Environ Biophys. 2010;49(1):47–55. [DOI] [PubMed] [Google Scholar]
- 17.Boice JD Jr , Cohen SS, Mumma MT, et al. Updated mortality analysis of radiation workers at Rocketdyne (Atomics International), 1948–2008. Radiat Res. 2011;176(2):244–258. [DOI] [PubMed] [Google Scholar]
- 18.Akiba S , Mizuno S. The third analysis of cancer mortality among Japanese nuclear workers, 1991-2002: estimation of excess relative risk per radiation dose. J Radiol Prot. 2012;32(1):73–83. [DOI] [PubMed] [Google Scholar]
- 19.Zablotska LB , Lane RS, Thompson PA. A reanalysis of cancer mortality in Canadian nuclear workers (1956-1994) based on revised exposure and cohort data. Br J Cancer. 2014;110(1):214–223. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Zablotska LB , Bazyka D, Lubin JH, et al. Radiation and the risk of chronic lymphocytic and other leukemias among Chornobyl cleanup workers. Environ Health Perspect. 2013;121(1):59–65. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Merzenich H , Hammer GP, Troltzsch K, et al. Mortality risk in a historical cohort of nuclear power plant workers in Germany: results from a second follow-up. Radiat Environ Biophys. 2014;53(2):405–416. [DOI] [PubMed] [Google Scholar]
- 22.Schubauer-Berigan MK , Daniels RD, Bertke SJ, et al. Cancer mortality through 2005 among a pooled cohort of U.S. nuclear workers exposed to external ionizing radiation. Radiat Res. 2015;183(6):620–631. [DOI] [PubMed] [Google Scholar]
- 23.Richardson DB , Cardis E, Daniels RD, et al. Risk of cancer from occupational exposure to ionising radiation: retrospective cohort study of workers in France, the United Kingdom, and the United States (INWORKS). BMJ. 2015;351:h5359. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Leuraud K , Richardson DB, Cardis E, et al. Ionising radiation and risk of death from leukaemia and lymphoma in radiation-monitored workers (INWORKS): an international cohort study. Lancet Haematol. 2015;2(7):e276–81. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Caldwell GG , Zack MM, Mumma MT, et al. Mortality among military participants at the 1957 PLUMBBOB nuclear weapons test series and from leukemia among participants at the SMOKY test. J Radiol Prot. 2016;36(3):474–489. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Preston DL , Kitahara CM, Freedman DM, et al. Breast cancer risk and protracted low-to-moderate dose occupational radiation exposure in the US Radiologic Technologists Cohort, 1983–2008. Br J Cancer. 2016;115(9):1105–1112. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Lee T , Sigurdson AJ, Preston DL, et al. Occupational ionising radiation and risk of basal cell carcinoma in US radiologic technologists (1983-2005). Occup Environ Med. 2015;72(12):862–869. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Kitahara CM , Linet MS, Balter S, et al. Occupational radiation exposure and deaths from malignant intracranial neoplasms of the brain and CNS in U.S. Radiologic Technologists, 1983–2012. AJR Am J Roentgenol. 2017;208(6):1278–1284. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Leuraud K , Fournier L, Samson E, et al. Mortality in the French cohort of nuclear workers. Radioprotection. 2017;52(3):199–210. [Google Scholar]
- 30.National Research Council (NRC). Health risks from exposure to low levels of ionizing radiation. In: Biological Effects of Ionizing Radiation (BEIR) VII. Phase 2. Washington, DC: National Research Council; 2006:265–266, 278, 284–286. [Google Scholar]
- 31.National Council on Radiation Protection and Measurements. NCRP Commentary 27. Implications of Recent Epidemiologic Studies for the Linear-Nonthreshold Model and Radiation Protection. Bethesda, MD: National Council on Radiation Protection and Measurements; 2018.
- 32.Greenland S , Lash TL. Bias analysis. In: Rothman KJ, Greenland S, Lash TL, eds. Modern Epidemiology. 3rd ed. Philadelphia: Wolters Kluwer/Lippincott Williams and Wilkins; 2008:358–359. [Google Scholar]
- 33.Sakata R , Grant EJ, Ozasa K. Long-term follow-up of atomic bomb survivors. Maturitas. 2012;72(2):99–103. [DOI] [PubMed] [Google Scholar]
- 34.Berrington de Gonzalez A , Daniels RD, Cardis E, et al. Epidemiological Studies of Low-Dose Ionizing Radiation and Cancer: Rationale and Framework for the Monograph and Overview of Eligible Studies. JNCI Monographs. 2020;2020(56):97--113. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Daniels RD , Kendall GM, Thierry-Chef I, et al. Strengths and Weaknesses of Dosimetry Used in Studies of Low-Dose Radiation Exposure and Cancer. JNCI Monographs. 2020;2020(56):114--132. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Greenland S. Response and follow-up bias in cohort studies. Am J Epidemiol. 1977;106(3):184–187. [DOI] [PubMed] [Google Scholar]
- 37.von Allmen RS , Weiss S, Tevaearai HT, et al. Completeness of follow-up determines validity of study findings: results of a prospective repeated measures cohort study. PLoS One. 2015;10(10):e0140817. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Xue X , Agalliu I, Kim MY, et al. New methods for estimating follow-up rates in cohort studies. BMC Med Res Methodol. 2017;17(1):155. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Edwards JK , Cole SR, Chu H, et al. Accounting for outcome misclassification in estimates of the effect of occupational asbestos exposure on lung cancer death. Am J Epidemiol. 2014;179(5):641–647. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Gilbert R , Martin RM, Donovan J, et al. Misclassification of outcome in case-control studies: methods for sensitivity analysis. Stat Methods Med Res. 2016;25(5):2377–2393. [DOI] [PubMed] [Google Scholar]
- 41.Sposto R , Preston DL, Shimizu Y, et al. The effect of diagnostic misclassification on non-cancer and cancer mortality dose response in A-bomb survivors. Biometrics. 1992;48(2):605–617. [PubMed] [Google Scholar]
- 42.World Health Organization (WHO). International Classification of Diseases for Oncology. 3rd ed. Geneva: World Health Organization; 2000. [Google Scholar]
- 43.Al-Hamadani M , Habermann TM, Cerhan JR, et al. Non-Hodgkin lymphoma subtype distribution, geodemographic patterns, and survival in the US: a longitudinal analysis of the National Cancer Data Base from 1998 to 2011. Am J Hematol. 2015;90(9):790–795. [DOI] [PubMed] [Google Scholar]
- 44.Bowen JM , Perry AM, Laurini JA, et al. Lymphoma diagnosis at an academic centre: rate of revision and impact on patient care. Br J Haematol. 2014;166(2):202–208. [DOI] [PubMed] [Google Scholar]
- 45.Glaser SL , Clarke CA, Keegan TH, et al. Time trends in rates of Hodgkin lymphoma histologic subtypes: true incidence changes or evolving diagnostic practice? Cancer Epidemiol Biomarkers Prev. 2015;24(10):1474–1488. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46.Hauptmann S , Friedrich K, Redline R, et al. Ovarian borderline tumors in the 2014 WHO classification: evolving concepts and diagnostic criteria. Virchows Arch. 2017;470(2):125–142. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 47.Gilbert ES , Little MP, Preston DL, et al. Issues in Interpreting Epidemiologic Studies of Populations Exposed to Low-Dose, High-Energy Photon Radiation. JNCI Monographs. 2020;2020(56):176--187. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48.Ferlay J , Soerjomataram I, Ervik M, et al. GLOBOCAN 2012: estimated cancer incidence, mortality and prevalence worldwide in 2012. V1.0. IARC CancerBase No. 11. http://globocan.iarc.fr/pages/datasource_and_methods. Accessed October 26, 2019.
- 49.Metz-Flamant C , Laurent O, Samson E, et al. Mortality associated with chronic external radiation exposure in the French combined cohort of nuclear workers. Occup Environ Med. 2013;70(9):630–638. [DOI] [PubMed] [Google Scholar]
- 50.Muirhead CR , Hagan JA, Haylock RGE, et al. HPA-RPD-062 - Third analysis of the National Registry for Radiation Workers: Occupational Exposure to Ionising Radiation in Relation to Mortality and Caner Incidence. Didcot, Oxfordshire, UK: Health Protection Agency, Centre for Radition, Chemical, and Environmental Hazards, Radiation Protection Division; 2009:14–15. [Google Scholar]
- 51.Iwasaki T , Murata M, Ohshima S, et al. Second analysis of mortality of nuclear industry workers in Japan, 1986-1997. Radiat Res. 2003;159(2):228–238. [DOI] [PubMed] [Google Scholar]
- 52.Bender TJ , Beall C, Cheng H, et al. Methodologic issues in follow-up studies of cancer incidence among occupational groups in the United States. Ann Epidemiol. 2006;16(3):170–179. [DOI] [PubMed] [Google Scholar]
- 53.Ruder AM , Hein MJ, Hopf NB, et al. Cancer incidence among capacitor manufacturing workers exposed to polychlorinated biphenyls. Am J Ind Med. 2017;60(2):198–207. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 54.Steliarova-Foucher E , Stiller C, Lacour B, et al. International Classification of Childhood Cancer. 3rd ed. Cancer. 2005;103(7):1457–1467. [DOI] [PubMed] [Google Scholar]
- 55.Michaud A , Midthune D, Hermansen S, et al. Comparison of cancer registry case ascertainment with SEER estimates and self‐reporting in a subset of the NIH‐AARP Diet and Health Study. J Regist Manag. 2005;32(2):70–75. [Google Scholar]
- 56.Bergmann MM , Calle EE, Mervis CA, et al. Validity of self-reported cancers in a prospective cohort study in comparison with data from state cancer registries. Am J Epidemiol. 1998;147(6):556–562. [DOI] [PubMed] [Google Scholar]
- 57.Desai MM , Bruce ML, Desai RA, et al. Validity of self-reported cancer history: a comparison of health interview data and cancer registry records. Am J Epidemiol. 2001;153(3):299–306. [DOI] [PubMed] [Google Scholar]
- 58.Parikh-Patel A , Allen M, Wright WE, et al. Validation of self-reported cancers in the California Teachers Study. Am J Epidemiol. 2003;157(6):539–545. [DOI] [PubMed] [Google Scholar]
- 59.Freedman DM , Sigurdson AJ, Doody MM, et al. Comparison between cancers identified by state cancer registry, self-report, and death certificate in a prospective cohort study of US radiologic technologists. Int J Epidemiol. 2006;35(2):495–497. [DOI] [PubMed] [Google Scholar]
- 60.Jacobs EJ , Briggs PJ, Deka A, et al. Follow-up of a large prospective cohort in the United States using linkage with multiple state cancer registries. Am J Epidemiol. 2017;186(7):876–884. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 61.Navarro C , Chirlaque MD, Tormo MJ, et al. Validity of self reported diagnoses of cancer in a major Spanish prospective cohort study. J Epidemiol Community Health. 2006;60(7):593–599. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 62.Cho S , Shin A, Song D, et al. Validity of self-reported cancer history in the health examinees (HEXA) study: a comparison of self-report and cancer registry records. Cancer Epidemiol. 2017;50(Pt A):16–21. [DOI] [PubMed] [Google Scholar]
- 63.Mieno MN , Tanaka N, Arai T, et al. Accuracy of death certificates and assessment of factors for misclassification of underlying cause of death. J Epidemiol. 2016;26(4):191–198. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 64.Goldoni CA , Bonora K, Ciatto S, et al. The IMPACT Working Group. Misclassification of breast cancer as cause of death in a service screening area. Cancer Causes Control. 2009;20(5):533–538. [DOI] [PubMed] [Google Scholar]
- 65.Fink AK , German RR, Heron M, et al. Impact of using multiple causes of death codes to compute site-specific, death certificate-based cancer mortality statistics in the United States. Cancer Epidemiol. 2012;36(1):22–28. [DOI] [PubMed] [Google Scholar]
- 66.Percy C , Stanek E 3rd, Gloeckler L. Accuracy of cancer death certificates and its effect on cancer mortality statistics. Am J Public Health. 1981;71(3):242–250. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 67.Kristman V , Manno M, Cote P. Loss to follow-up in cohort studies: how much is too much? Eur J Epidemiol. 2003;19(8):751–760. [DOI] [PubMed] [Google Scholar]
- 68.Twisk J , de Vente W. Attrition in longitudinal studies. How to deal with missing data. J Clin Epidemiol. 2002;55(4):329–337. [DOI] [PubMed] [Google Scholar]
- 69.Johnson ES. Bias on withdrawing lost subjects from the analysis at the time of loss, in cohort mortality studies, and in follow-up methods. J Occup Med. 1990;32(3):250–254. [DOI] [PubMed] [Google Scholar]
- 70.Wadsworth ME , Mann SL, Rodgers B, et al. Loss and representativeness in a 43 year follow up of a national birth cohort. J Epidemiol Community Health. 1992;46(3):300–304. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 71.Kristensen P , Bjerkedal T. Dealing with emigration in cohort studies: follow-up of mortality and cancer incidence among Norwegians born between 1967 and 1976. Eur J Epidemiol. 2010;25(3):155–161. [DOI] [PubMed] [Google Scholar]
- 72.Gushulak BD , MacPherson DW. The basic principles of migration health: population mobility and gaps in disease prevalence. Emerg Themes Epidemiol. 2006;3(1):3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 73.Davey-Smith G , Chaturvedi N, Harding S, et al. Ethnic inequalities in health: a review of UK epidemiological evidence. Critical Public Health. 2000;10(4):375–408. [Google Scholar]
- 74.Abraido-Lanza AF , Dohrenwend BP, Ng-Mak DS, et al. The Latino mortality paradox: a test of the “salmon bias” and healthy migrant hypotheses. Am J Public Health. 1999;89(10):1543–1548. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 75.Jaffe ES , Harris NL, Stein H, et al. World Health Organization classification of tumours. In: Pathology and Genetics of Tumours of Haematopoietic and Lymphoid Tissue. Lyon, France: IARC Press; 2001:45–72. [Google Scholar]
- 76.Linet MS , Morton LM, Devesa SS, et al. Leukemias. In: Thun MJ, Linet MS, Cerhan JR, Haiman C, Schottenfeld D, eds. Schottenfeld and Fraumeni Cancer Epidemiology and Prevention. 4th ed. New York, NY: Oxford University Press; 2018:715–744. [Google Scholar]
- 77.Linet MS , Gilbert ES, Vermeulen R, et al. Benzene exposure response and risk of myeloid neoplasms in Chinese workers: a multicenter case-cohort study. J Natl Cancer Inst. 2019;111(5):465–474. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 78.Morton LM , Dores GM, Schonfeld SJ, et al. Association of chemotherapy for solid tumors with development of therapy-related myelodysplastic syndrome or acute myeloid leukemia in the modern era. JAMA Oncol. 2018;10–1001. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 79.Schubauer-Berigan MK , Daniels RD, Fleming DA, et al. Risk of chronic myeloid and acute leukemia mortality after exposure to ionizing radiation among workers at four U.S. nuclear weapons facilities and a nuclear naval shipyard. Radiat Res. 2007;167(2):222–232. [DOI] [PubMed] [Google Scholar]
- 80.United Nations Scientific Committee on the Effects of Atomic Radiation. Sources and Effects of Ionizing Radiation. Vol I. New York: United Nations Scientific Committee on the Effects of Atomic Radiation; 2008. https://www.unscear.org/docs/publications/2008/UNSCEAR_2008_report_Vol1.pdfhttps://www.unscear.org/docs/publications/2008/UNSCEAR_2008_report_VolII.pdf. [Google Scholar]
- 81.United Nations Scientific Committee on the Effects of Atomic Radiation. Sources and Effects of Ionizing Radiation. Vol. I. New York: United Nations Scientific Committee on the Effects of Atomic Radiation; 2000. https://www.unsear.org/docs/publications/2000/UNSCEAR_2000_report.pdf. [Google Scholar]
- 82.United Nations Scientific Committee on the Effects of Atomic Radiation (UNSCEAR). Effects and Risk of Ionizing Radiation. Report to the General Assembly with Scientific Annexes. Annex A: Attributing health effects to ionizing radiation exposure and inferring risks. New York, NY: United Nations Scientific Committee on the Effects of Ionizing Radiation; 2012. https://www.unscear.org/docs/publications/2012/UNSCEAR_2012_Report.pdf.
- 83.United Nations Scientific Committee on the Effects of Atomic Radiation (UNSCEAR). Effects and Risk of Ionizing Radiation. Report to the General Assembly with Scientific Annexes. Annex A: Principles and criteria for ensuring the quality of the committee’s reviews of epidemiologic studies of radiation exposure. In. New York, NY: United Nations Scientific Committee on the Effects of Ionizing Radiation; 2017. https://www.unscear.org/docs/publications/2017/UNSCEAR_2017_Report.pdf.
- 84.Berrington de Gonzalez A , Bouville A, Rajaraman P, et al. Ionizing radiation. In: Thun MJ, Linet MS, Cerhan JR, Haiman C, Schottenfeld D, eds. Schottenfeld and Fraumeni Cancer Epidemiology and Prevention. 4th ed. New York, NY: Oxford University Press; 2018:227–248. [Google Scholar]
- 85.United Nations Scientific Committee on the Effects of Atomic Radiation. Report to the General Assembly with Scientific Annexes, Annex B: Effects of radiation exposure of children. In: Sources, Effects, and Risks of Ionizing Radiation. In. New York, NY: United Nations Scientific Committee on the Effects of Atomic Radiation; 2013.
- 86.Pisa F , Castellsague J, Drigo D, et al. Accuracy of International Classification of Diseases, 9th Revision, Clinical Modification codes for upper gastrointestinal complications varied by position and age: a validation study in a cohort of nonsteroidal anti-inflammatory drugs users in Friuli Venezia Giulia, Italy. Pharmacoepidemiol Drug Saf. 2013;22(11):1195–1204. [DOI] [PubMed] [Google Scholar]
- 87.Greenland S , Applications of stratified analysis methods. In: Rothman KJ, Greenland S, Lash TL, eds. Modern Epidemiology. 3rd ed. Philadelphia: Wolters Kluwer/Lippincott Williams and Wilkins; 2008:289–293. [Google Scholar]
- 88.Sokolnikov M , Preston D, Gilbert E, et al. Radiation effects on mortality from solid cancers other than lung, liver, and bone cancer in the Mayak worker cohort: 1948-2008. PLoS One. 2015;10(2):e0117784. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 89.Lash TL , Fox MP, MacLehose RF, et al. Good practices for quantitative bias analysis. Int J Epidemiol. 2014;43(6):1969–1985. [DOI] [PubMed] [Google Scholar]