Cancer mortality following protracted low-level radiation exposure for early and contemporary workers in two large occupational cohorts in the U.S. Million Person Study

Abstract

An evaluation is presented of differences in radiation-related solid cancer mortality risk for early versus contemporary sub-groups of radiation workers in both of two of the constituent Million Person Study (MPS) cohorts. The two previously analysed MPS cohorts are 123,401 Industrial Radiographers monitored from 1939–2011 and followed through 2019 and 135,193 Nuclear Power Plant workers monitored 1957–1984 and followed through 2011. The rationale behind this extended new analysis is to investigate if these two MPS cohorts support recently published increased risks for contemporary workers in a different cohort, The International Nuclear Workers Study (INWORKS) (Richardson et al. 2023) with pooled US, French and UK nuclear worker data, particularly in the US component.

The US-INWORKS (Kelly-Reif et al. 2023) contributed about one-third of the workers to the full-INWORKS study based on 309,932 workers. For all solid cancer mortality, the US-INWORKS study reported a low and non-significant Excess Relative Risk (ERR) per Sv cumulative equivalent dose for the whole cohort of 0.19 (95%CI: −0.10; 0.52), whereas for contemporary workers the ERR per Sv was 2.23 (95% CI: 1.13, 3.49), approximately 10 times higher than the entire US-INWORKS cohort. The risk for the full-INWORKS cohort was 0.52 (90%CI: 0.27; 0.77) per Gy colon dose whereas for contemporary workers the risk was 1.44 (90%CI: 0.65, 2.32), nearly 3 times higher. These risks for contemporary workers are both larger than risks informing radiation protection and much higher (7.0 and 4.5 times) than the Japanese A-bomb survivor’s risk for males exposed acutely between ages of 20 and 60 years of 0.32 (95% CI: 0.01; 0.50) (Cardis et al. 2005). Limitations include missing information on organ doses from radionuclide-intake, neutrons and the absence of adjustment for non-radiation risk factors (notably asbestos exposure). The analysis of the MPS cohorts addresses these dosimetric- and asbestos-related limitations. For all solid cancer mortality, Industrial Radiographers showed equal Poisson ERRs per 100 mGy colon dose for early and contemporary workers: 0.06 (95%CI: 0.00; 0.12) and 0.07 (95%CI: 0.01; 0.13), respectively. The results for Nuclear Power Plant Workers were 0.10 (95%CI: −0.09; 0.29) and 0.02 (95%CI: −0.02; 0.06), respectively. It appears premature to conclude that there is generally a difference in excess risk between early and contemporary workers from radiation exposures.

INTRODUCTION

The Million Person Study (MPS) includes epidemiologic data on mortality outcomes and radiation exposures for over one million US radiation exposed workers and veterans (1). Extensive results for numerous mortality outcomes in relation to time-dependent, low-dose radiation exposures have been published for 13 of the MPS constituent cohorts and have been summarised recently (1,2). The purpose of this paper is to extend already published analyses to evaluate differences in all solid cancer mortality risk for early versus contemporary radiation workers within two large MPS cohorts, i.e., Industrial Radiographers (IR) and Nuclear Power Plant Workers (NPP), totalling 258,594 workers (3,4). The approach is to conduct similar analyses, by period of first employment, as those conducted in a different cohort, i.e.,The International Nuclear Workers Study (INWORKS) with pooled US, French and UK nuclear worker data, and particularly in the US component of INWORKS. As noted by Wakeford (5) referring to the INWORKS cohort: The MPS “offers an opportunity to examine cancer risks in other groups of US workers by period of first employment and monitoring for radionuclide intake.”

The US-INWORKS study (6) includes 101,363 workers, followed up between 1944 and 2016, and is part of the full-INWORKS study (7) which includes 309,932 workers, also followed up between 1944 and 2016. For all solid cancer mortality, the US-INWORKS study reported a low and nonsignificant Excess Relative Risk (ERR) per Sv cumulative equivalent dose for the whole cohort of 0.19 (95%CI: −0.10; 0.52), whereas for contemporary workers, first hired in 1960 or later, the ERR per Sv was 2.23 (95% CI: 1.13, 3.49) or approximately 10 times higher than the entire cohort. Similarly, the full-INWORKS cohort published an ERR of 0.52 (90%CI: 0.27; 0.77) per Gy colon dose whereas for contemporary workers, hired in 1965 or later, the ERR per Gy was 1.44 (90%CI: 0.65; 2.32) or nearly 3 times higher. These two risks for contemporary workers are both much larger than risks currently informing international radiation protection, and much higher (7.0 and 4.5 times) than the comparable Japanese A-bomb survivors risk for males exposed acutely between ages of 20 and 60 years old of 0.32 (95% CI: 0.01; 0.50) from Table 2 of (8). These results are contrary to expectation because older, early workers, on average, have higher doses and longer follow-up than contemporary, younger workers. Limitations of the US-INWORKS analysis include the apparent absence of information on organ doses from the intake of radionuclides, on organ doses from neutrons and on non-radiation risk factors such as smoking and asbestos.

Table 2.

Industrial Radiographers sub-cohorts, numbers of deaths from all solid cancers with dose category specific details, estimated baseline cases, estimated excess cases, estimated fitted number of cases and estimated attributable fractions based on the ERR model. The upper panel is for early workers with all monitoring done entirely before 1979 and the lower panel is for everyone else, referred to as contemporary workers.

dose category mGy	number of early workers	person years at risk	actual number of cases	estimated number of baseline cases	estimated number of excess cases	estimated fitted number of cases	estimated attributable fraction (%)
0 to <25	18596	705144	2757	2775	7.7	2782	0.3
25 to <50	1722	50462	309	284	7.2	292	2.5
50 to <100	1257	33750	245	226	11.4	238	4.8
100 to <250	963	23251	213	181	20.1	201	10.0
250+	288	5846	63	58	15.9	74	21.4
total	22826	818452	3587	3525	62.3	3587	1.7
ERR (95%CI) per 100 mGy (sub-group analysis), p-value for dose response							0.07 (0.01; 0.13), p = 0.02
EAR (95%CI) excess cases per 100 mGy and 10000 person-years (sub-group analysis)							7.61 (3.84; 11.38), p < 0.001
ERR (95%CI) per 100 mGy (full-cohort analysis with indicator variable)							0.06 (0.01; 0.12), p = 0.03
EAR (95%CI) excess cases per 100 mGy and 10000 person-years (full-cohort with indicator variable)							8.37 (4.72; 12.02), p < 0.001
	number of contemporary workers
0 to <25	86227	2321018	3179	3216	7.3	3224	0.2
25 to <50	5956	117359	344	317	7.8	324	2.4
50 to <100	4578	89811	307	266	13.0	279	4.6
100 to <250	3261	60055	268	229	24.0	253	9.5
250+	553	9614	52	55	14.4	70	20.7
total	100575	2597856	4150	4083	66.5	4150	1.6
ERR (95%CI) per 100 mGy (sub-group analysis), p-value for dose response							0.07 (0.01; 0.13), p = 0.02
EAR (95%CI) excess cases per 100 mGy and 10000 person-years (sub-group analysis)							1.34 (−0.02; 2.71), p = 0.05
ERR (95%CI) per 100 mGy (full-cohort analysis with indicator variable)							0.07 (0.01; 0.13), p = 0.01
EAR (95%CI) excess cases per 100 mGy and 10000 person-years (full-cohort with indicator variable)							1.11 (−0.22; 2.44), p = 0.10

The IR and NPP cohorts address the dosimetric and asbestos limitations noted above. The analysis of the MPS cohorts generally addresses these dosimetric- and asbestos- related limitations (9–13) and have similar exposure conditions to the INWORKS pooled cohort, i.e. protracted exposures mainly to low dose and low-dose rate external gamma radiation, providing an excellent opportunity for a comparative analysis.

MATERIALS AND METHODS

Approval for the study on human subjects was received from the Vanderbilt University Institutional Review Board. The cohort definitions for Industrial Radiographers (IR) and Nuclear power plant (NPP) workers and previous analyses have been published in full (3,4). Briefly, for both cohorts the methods in Mumma et al. (14) were used to ascertain cause of death and vital status, and published methodology (9, 15, 16) was applied to combine occupational dose contributions and calculate annual organ dose estimates. The final IR study population consisted of 123,401 industrial radiographers born between 1902 and 1993. These radiographers were involved in applying ionizing radiation to examine dense material for flaws and defects and monitored for exposures between 1939 and 2011. The IR cohort follow-up study period was from 1969 to the end of 2019, during which time there were 3.42 million person-years (PY) at risk and 7737 deaths from all solid cancer, of which 309 were females. The mean of the final accumulated radiation doses to the colon, lagged by 10 years, was 15.8 mGy (maximum 1.48 Gy; percent >100 mGy was 4.1%) mainly from external gamma radiation. The NPP study population consisted of 135,193 nuclear power plant workers born between 1902 and 1966. The NPP workers were monitored for ionizing radiation during their work at nuclear power plants and first monitored for exposures between 1957 and 1984. The NPP cohort follow-up study period was from 1957 to the end of 2011, during which time there were 4.08 million person-years at risk and 8445 deaths from all solid cancer, of which 158 were females. The mean of the final accumulated radiation doses to the colon, lagged by 10 years was 41.0 mGy (maximum 1.1 Gy; percent >100 mGy was 11.5 %) mainly from external gamma radiation.

Poisson Excess Relative Risk (ERR) and Excess Absolute Risk (EAR) models were fitted to the data using the AMFIT module of the EPICURE software (17), and the 95% confidence level was applied for reporting limits. The Poisson models included the continuous covariates of time dependent, cumulative colon organ dose, year of birth, age attained and categorical variables for sex and socio-economic status (SES) fitted to individual level input data, where each data record was one Poisson cell with only one person-year at risk or a fraction of a person-year (PY) at risk at the start and end of each person’s follow-up period (see (18) for more details). ERR per 100 mGy cumulative colon dose and EAR in excess cases per 10,000 PY/100 mGy were fitted according to Equations 1–3 assuming a linear risk to dose response, based on a parametric baseline with male and female, sex-specific (M, F) log-linear-quadratic in the logarithm of age attained, age, in years with a spline-knot applied above age 70 years, a log-linear birth cohort effect using year of birth, by, and a category variable for area-level education (as an indicator of Socio-Economic Status (SES)) (19), educat, which had 3 categories for both the NPP and IR cohorts.

$$\text{Total}\text{all}\text{solid}\text{cancer}\text{rate}=\text{Baseline}\text{rate}*\left(1+\left({\text{ERR}}_{\text{per}\text{unit}\text{dose}}*\mathit{\text{dose}}\right)\right)$$ (Eq.1)

$$\text{Total}\text{all}\text{solid}\text{cancer}\text{rate}=\text{Baseline}\text{all}\text{solid}\text{cancer}\text{rate}+\left({\text{EAR}}_{\text{per}\text{unit:}\text{dose},\text{PY}}*\mathit{\text{dose}}\right)$$ (Eq. 2)

$$\text{Baseline}\text{all}\text{solid}\text{cancer}\text{rate}=\text{EXP}\left\{{ß}_{1-3}*\mathit{\text{educat}}+{ß}_4*\left(by-1950\right)+{ß}_{\text{M},5}*\left(\text{ln}\left(age/70\right)\right)+{ß}_{\text{F},5}*\left(\text{ln}\left(age/70\right)\right)+{ß}_{\text{M},6}*{\left(\text{ln}\left(age/70\right)\right)}^2+{ß}_{\text{F},6}*{\left(\text{ln}\left(age/70\right)\right)}^2+{ß}_{\text{M},7}*{\left(\text{ln}\left(age/70\right)\right)}^2,\text{if}\mathit{\text{age}}>70+{ß}_{\text{F},7}*{\left(\text{ln}\left(age/70\right)\right)}^2,\text{if}\mathit{\text{age}}>70\right\}$$ (Eq. 3)

The fit parameters are either ERR_{per unit dose} or ERR_{per unit: dose, py,} referred to just as ERR and EAR henceforth, and those for the baseline, parameters ß₁ to ß₇ which can be sex specific as indicated in the subscripts of M (male) or F (female). The doses were fitted in units of 100 mGy time-dependent cumulative colon dose so that the ERR and EAR fit parameters were directly obtained for this unit. The person-year (PY) unit for EAR was 10,000.

Attributable fractions were calculated from the numbers of radiation related excess deaths and the corresponding total numbers of deaths from all solid cancer using either the linear ERR or EAR models. Excess number of deaths and attributable fractions and confidence intervals were computed following the approach applied by Ozasa et al (20), which involves fixing the ERR or EAR dose-response parameters at their upper and lower CI values and re-optimising the models to obtain the numbers of excess and fitted cases in the AMFIT module of EPICURE. ERR and EAR models, excess cases and attributable fractions were fitted to the whole cohort and, for consistency in methodology with the comparisons of risk from non-MPS cohorts, fitted to sub-cohorts of early versus contemporary radiation workers. Furthermore, to achieve robust estimates for baseline risks, the model above was fitted to the full cohort with an indicator variable for early versus contemporary radiation workers, used as a multiplicative interaction term on the dose response parameter. The early workers sub-cohort was defined to have all monitoring done entirely before 1979 and the remaining sub-cohort, corresponding to everyone else, was therefore by definition more contemporary, with monitoring occurring for an individual either before and after 1979 or first monitoring occurred in 1979 and thereafter. The Akaike Information Criterion (AIC) (21) and the likelihood ratio test were applied to compare relative goodness of fit to the data between the different models.

RESULTS

Industrial Radiographers

For all solid cancers, 123,401 Industrial Radiographers had full cohort ERRs per 100 mGy colon dose of 0.07 (95%CI: 0.03; 0.11) (AIC = 94160.5) and EAR per 100 mGy colon dose and 10,000 person-years of 2.61 (95%CI: 1.24; 3.99) (AIC=94157.6), where the change in AIC indicated that the EAR model fitted the data only slightly better than the ERR model (p=0.19). There were 125 (95%CI: 49; 199) estimated excess solid cancer deaths, and the cohort estimated attributable (to radiation) fraction of deaths was 1.6% (95%CI: 0.6; 2.6%), based on the ERR model. Table 1 gives dose category specific details of the estimated number of baseline cases, estimated excess cases, estimated fitted number of cases and estimated attributable fractions based on the ERR model for the full cohort. If the outcome was changed to all solid cancer with lung cancer and mesothelioma removed, so as to investigate the possible effects of asbestos and smoking exposure on the overall risk, the ERR per 100 mGy colon dose was 0.02 (95%CI: −0.03; 0.07) (AIC = 62435.1) and EAR per 100 mGy colon dose and 10,000 person-years was 1.17 (95%CI: 0.11; 2.24) (AIC=62431.3), and the number of estimated excess cases was 25.

Table 1.

Industrial Radiographers full cohort, numbers of deaths from all solid cancers with dose category specific details, estimated baseline cases, estimated excess cases, estimated fitted number of cases and estimated attributable fractions based on the ERR model. Note: a finer dose categorisation for the IR full cohort could be justified than for the IR subcohorts.

dose category mGy	actual number of cases	estimated number of baseline cases	estimated number of excess cases	estimated fitted number of cases	estimated attributable fraction (%)
0 to <5	4374	4597	3	4599	0.1
5 to <50	2215	1995	26	2022	1.3
50 to <75	359	306	13	318	4.0
75 to <100	193	189	11	200	5.5
100 to <150	260	228	19	247	7.6
150 to <200	133	109	13	121	10.4
200+	203	189	41	230	17.7
total	7737	7612	125	7737	1.6
ERR (95%CI) per 100 mGy					0.07 (0.03; 0.11)
EAR (95%CI) excess cases per 100 mGy and 10000 person-years					2.61 (1.24; 3.99)

Results for the subsidiary analysis using models for the full cohort with indicator variables for early versus contemporary radiation workers on the dose response parameter are given in Table 2. The ERR model with one full cohort dose response (AIC = 94160.5) fitted slightly better (ΔAIC =2, p=0.27) than the model with the indicator variable (AIC = 94162.5). However, the EAR with one full cohort dose response (AIC=94157.6) provided a statistically significantly worse fit (ΔAIC =12.5, p = 0.002) to the data than the model with the indicator variable on monitoring period (AIC=94145.1).

There were 40,073 workers first monitored before 1979, of whom 43% (or 17,247) also were monitored in 1979 or later. Thus, there were 22,826 workers only monitored before 1979 considered as early radiographers. The IR workers sub-cohort with all radiation monitoring done entirely before 1979 (mean of the final accumulated colon doses, lagged by 10 years, 21.1 mGy) had ERR per 100 mGy colon dose of 0.07 (95%CI: 0.01; 0.13) (AIC = 41337.8) and EAR per 100 mGy colon dose and 10,000 person-years of 7.61 (95%CI: 3.84; 11.38) (AIC=41325.7), where the change in AIC indicated that the EAR model fitted the data statistically significantly better (ΔAIC = 12.2, p=0.002) than the ERR model. There were 62 (95%CI: 10; 113) estimated excess all solid cancer deaths, and the sub-cohort estimated attributable fraction of deaths was 1.7% (95%CI: 0.3; 3.2%) based on the ERR model. The number of estimated excess all solid cancer deaths based on the EAR model increased to 92, but for consistency of reporting, the ERR based number of 62 was reported in the tables. The upper panel of Table 2 (for early workers with all monitoring done entirely before 1979) gives dose category specific details of the estimated: number of baseline cases; excess cases; fitted number of cases; and attributable fractions; based on the ERR model, for consistency. The results for the subsidiary analysis using the full cohort data with an indicator variable for monitoring period, were very similar to the main subgroup analysis (Table 2).

The contemporary IR workers (i.e. everyone in the IR cohort excluding those with all monitoring entirely before 1979) had a mean of the final accumulated colon dose, lagged by 10 years of 14.6 mGy. This subset of the IR workers had an ERR per 100 mGy colon dose of 0.07 (95%CI: 0.01; 0.13) (AIC = 52810.3) and EAR per 100 mGy colon dose and 10,000 person-years of 1.34 (95%CI: −0.02; 2.71) (AIC=52812.4), where the change in AIC of 2.1 indicated that the EAR model fitted the data only slightly worse than the ERR model. There were 67 (95%CI: 10; 121) estimated excess all solid cancer deaths, and the sub-cohort estimated attributable fraction of deaths was 1.6% (95%CI: 0.2; 2.9%) based on the ERR model. The lower panel of Table 2 (for everyone else, referred to as contemporary workers) gives dose category specific details of the estimated: number of baseline cases; excess cases; fitted number of cases; and attributable fractions; based on the ERR model, for consistency. The results for the subsidiary analysis using the full cohort with an indicator variable for monitoring period, were very similar to the main subgroup analysis (Table 2).

Nuclear Power Plant Workers

For all solid cancers, 135,193 Nuclear Power plant workers had a full cohort ERR per 100 mGy colon dose of 0.01 (95%CI: −0.03; 0.05) (AIC = 105773.2) and EAR per 100 mGy colon dose and 10,000 person-years of 0.27 (95%CI: −0.42; 0.96) (AIC=105773.0), where the change in AIC of only 0.2 indicated that the EAR model had a similar quality of fit to the ERR model. There were 32 (95%CI: −80; 140) estimated excess all solid cancer deaths, and the cohort estimated attributable fraction of deaths was 0.4% (95%CI: −1.0; 1.7%). Table 3 gives dose category specific details of the estimated: number of baseline cases; excess cases; fitted number of cases; and attributable fractions; based on the ERR model for the full cohort. If the outcome was changed to all solid cancer with lung cancer and mesothelioma removed, to investigate the effects of these latter two smoking related cancer types on the overall risk, the ERR per 100 mGy colon dose was very similar to the all solid cancer risk given above at 0.02 (95%CI: −0.03; 0.07) (AIC = 66514.5). The EAR per 100 mGy colon dose and 10,000 person-years was also similar to the all solid cancer risk given above at 0.37 (95%CI: −0.24; 0.97) (AIC=66513.8), and the number of estimated excess deaths for this outcome was 35, based on the ERR model. Results for the subsidiary analysis using models for the full cohort with indicator variables, for early versus contemporary radiation workers, on the dose response parameter are given in table 4. The ERR model with one full cohort dose response (AIC = 105773.2) fitted worse (ΔAIC = 5.5, p= 0.06) than the model with the indicator variable on monitoring period (AIC = 105767.7). However the EAR with one full cohort dose response (AIC=105773.0) provided a statistically significantly worse fit (ΔAIC = 9.4, p = 0.009) to the data than the model with the indicator variable on monitoring period (AIC=105763.6).

Table 3.

Nuclear Power Plant full cohort numbers of deaths from all solid cancers with dose category specific details, estimated baseline cases, estimated excess cases, estimated fitted number of cases and estimated attributable fractions based on the ERR model.

dose category mGy	actual number of cases	estimated number of baseline cases	estimated number of excess cases	estimated fitted number of cases	estimated attributable fraction (%)
0 to <25	5275	5399	5	5403	0.1
25 to <50	1591	1415	6	1420	0.4
50 to <100	904	926	7	933	0.8
100 to <250	569	553	9	563	1.6
250+	106	121	5	126	3.8
total	8445	8413	32	8445	0.4
ERR (95%CI) per 100 mGy					0.01 (−0.03; 0.05)
EAR (95%CI) excess cases per 100 mGy and 10000 person-years					0.27 (−0.42; 0.96)

Table 4.

Nuclear Power Plant sub-cohorts numbers of deaths from all solid cancers with dose category specific details, estimated baseline cases, estimated excess cases, estimated fitted number of cases and estimated attributable fractions based on the ERR model. The upper panel is for early workers with all monitoring done entirely before 1979 and the lower panel is for everyone else, referred to as contemporary workers.

dose category mGy	number of early workers	person years at risk	actual number of cases	estimated number of baseline cases	estimated number of excess cases	estimated fitted number of cases	estimated attributable fraction (%)
0 to <25	9,844	325792	1134	1139	8	1147	0.7
25 to <50	2,150	47937	248	212	6	218	2.9
50 to <100	665	15243	60	71	4	75	5.5
100 to <250	195	4597	29	27	4	31	11.6
250+	22	422	4	4	1	5	22.3
total	12,876	393990	1475	1452	23	1475	1.5
ERR (95%CI) per 100 mGy (sub-group analysis), p-value for dose response							0.09 (−0.10; 0.27), p = 0.36
EAR (95%CI) excess cases per 100 mGy and 10000 person-years (sub-group analysis)							−0.06 (−1.08: 0.96), p > 0.5
ERR (95%CI) per 100 mGy (full-cohort analysis with indicator variable)							0.26 (0.08; 0.44), p = 0.005
EAR (95%CI) excess cases per 100 mGy and 10000 person-years (full-cohort with indicator variable)							9.66 (4.34; 14.98), p < 0.001
	number of contemporary workers
0 to <25	65,373	2697968	4141	4252	6	4258	0.1
25 to <50	24,835	461590	1343	1199	7	1206	0.6
50 to <100	18325	318126	844	845	10	855	1.1
100 to <250	11966	181106	540	517	13	530	2.4
250+	1818	26461	102	115	7	122	5.6
total	122,317	3685251	6970	6928	42	6970	0.6
ERR (95%CI) per 100 mGy (sub-group analysis)							0.02 (−0.02; 0.06), p = 0.43
EAR (95%CI) excess cases per 100 mGy and 10000 person-years (sub-group analysis), p-value							0.54 (−0.18; 1.25), p = 0.14
ERR (95%CI) per 100 mGy (full-cohort analysis with indicator variable)							0.004 (−0.035; 0.043), p > 0.5
EAR (95%CI) excess cases per 100 mGy and 10000 person-years (full-cohort with indicator variable)							0.20 (−0.48; 0.88), p > 0.5

There were 60,837 workers first monitored before 1979 of whom 47,961 (79% of the full cohort) also were monitored in 1979 or later. Thus, there were 12,876 workers (9.5% of the full cohort) only monitored before 1979 considered ‘early’ NPP workers (i.e. the sub-cohort with all radiation monitoring done entirely before 1979). These workers had a mean of the final accumulated colon dose, lagged by 10 years, of 19.1 mGy. The ERR per 100 mGy colon dose was 0.09 (95%CI: −0.10; 0.27) (AIC = 17395.2) while the EAR was −0.06 (−1.08; 0.96) (AIC = 17396.2). There were 23 (95%CI: −27; 69) estimated excess all solid cancer deaths, and the sub-cohort estimated attributable fraction of deaths was 1.5% (95%CI: −1.8; 4.7%) based on the ERR model. The upper panel of Table 4 (for early workers with all monitoring done entirely before 1979) gives dose category specific details of the number of estimated baseline cases, estimated excess cases, estimated fitted number of cases and estimated attributable fractions based on the ERR model, for consistency. The results for the subsidiary analysis using the full cohort with an indicator variable for monitoring period, were similar, with overlapping confidence intervals, to the main subgroup analysis for the ERR models, but the EAR model for the full cohort with indicator variable, estimated a much higher EAR of 9.66 (95%CI: 4.34; 14.98) than the EAR from the sub-group analysis of −0.06 (95%CI: −1.08; 0.96) (Table 4).

In the Contemporary NPP workers (i.e., the subcohort defined by excluding those with all monitoring entirely before 1979), the mean of the final accumulated colon dose, lagged by 10 years, was 43.3 mGy. For the Contemporary NPP workers, the ERR per 100 mGy colon dose was 0.02 (95%CI: −0.02; 0.06) (AIC = 88325.7) and EAR per 100 mGy colon dose and 10,000 person-years was 0.54 (95%CI: −0.18; 1.25) (AIC=88324.3), where the change in AIC of 1.4 indicated that the ERR model fitted the data slightly worse than the EAR model. There were 42 (95%CI: −63; 144) estimated excess all solid cancer deaths, and the sub-cohort estimated attributable fraction of deaths was 0.6% (95%CI: −0.9; 2.1%) based on the ERR model. The lower panel of Table 4 (for everyone else, referred to as contemporary workers) gives dose category specific details of the number of estimated: baseline cases; excess cases; fitted number of cases; and attributable fractions; based on the ERR model, for consistency. The results for the subsidiary analysis using the full cohort with an indicator variable for monitoring period, were similar, with overlapping confidence intervals, to the main subgroup analysis (Table 4).

Overall comparisons of risks.

Figure 1 presents the solid cancer ERRs per 100 mGy colon dose presented here for the IR and NPP cohorts compared with the risks from the INWORKS full cohort that show an increased risk in contemporary workers. Results are also shown for the US component of the INWORKS cohort that show an increased risk in contemporary workers. Another component of INWORKS, was mortality data from the UK National Registry for Radiation Workers (NRRW) study. Figure 1 also shows the risks of all solid cancer incidence between early and contemporary workers from a recently published paper on workers in the UK NRRW study (22). Comparisons made in Figure 1 are facilitated by all cohorts having early exposed workers, i.e., from 1939 for IR, from 1957 for NPP, and according to (23) for INWORKS, their U.K. and US components had monitoring commencing in the nineteen-forties and -thirties, respectively.

Figure 1.

ERRs per 100 mGy colon dose and number of all solid cancers (n), for all solid cancer mortality presented here for the IR and NPP cohorts (top 6 rows) compared with the risks from the INWORKS cohorts that show and increased risk in contemporary workers, i.e. INWORKS (7) (workers first hired in 1965 or later, risks given with 90% CIs) and the US component of INWORKS (6) (workers first hired in 1960 or later). Also shown (bottom 3 rows) are all solid cancer incidence risks for the full cohort, early (hired before 1960) and contemporary (hired after 1960) workers from the UK NRRW study (22), where the sub-cohort sizes were only reported as percentages of the total. The dotted vertical line gives the Japanese A-bomb survivors risk for males exposed between ages of 20 and 60 years of 0.03 (95% CI: 0.001; 0.050) (8). Note: the INWORKS cohorts portrayed in this figure (intermediate rows 7–10 from the top) did not provide the corresponding risks for early workers. The rationale for the monitoring group definitions is based on dosimetric uncertainties, which are generally agreed to be higher in the earlier years of monitoring than in the later years of monitoring.

DISCUSSION

Previously published results (3) for all solid cancer mortality in 123,401 Industrial Radiographers, showed equal ERRs per 100 mGy colon dose for early and contemporary workers: 0.06 (95%CI: 0.00; 0.12) and 0.07 (95%CI: 0.01; 0.13), respectively from Cox regression. These were essentially the same as the Poisson regression ERRs per 100 mGy colon dose presented here, and in Figure 1, of 0.07 (95%CI: 0.01; 0.13) and 0.07 (95%CI: 0.01; 0.13) respectively, showing that the results do not differ appreciably by regression method. However, results obtained via Cox regression do not allow a thorough examination of the baseline rates, the number of estimated excess cases, excess absolute risk and the estimated attributable fraction. Similarly, the INWORKS results (7) were obtained using Poisson regression with a stratified baseline, which also does not allow a convenient examination of baseline rates, or the convenient computation of estimated excess cases and excess absolute risks. Therefore, no information is available on how many excess solid cancer cases the high risk published by INWORKS (7) for contemporary workers were based on.

Similarly, and based on an extended new analysis of the data used and methods applied in (4), the Cox regression ERRs per 100 mGy colon dose, for the early and contemporary Nuclear Power Plant workers were 0.10 (95%CI: −0.09; 0.29) and 0.02 (95%CI: −0.02; 0.06), respectively. These were essentially the same as the Poisson regression ERRs per 100 mGy colon dose presented here, and in Figure 1, of 0.09 (95%CI: −0.10; 0.27) and 0.02 (95%CI: −0.02; 0.06). The NPP mean colon dose for the sub-cohort with all monitoring before 1979 is lower than the mean dose for the sub-cohort with everyone else (see results section). This difference is due to the selection of these workers and opportunities in the nuclear industry for the “everyone else” sub-cohort to get higher doses. When the sub-cohort is re-defined and the split is made for any monitoring before 1979 (which includes the main sub-cohort analyzed here of all monitoring done entirely before 1979), the mean dose is higher than the “everyone else” sub-cohort as expected.

Although the NPP cohort is very large with 1,475 all solid cancer deaths in 12,876 workers for the early worker sub-cohort with all monitoring done before 1979, and 6,970 all solid cancer deaths in 122,317 workers for the more contemporary (everyone else) sub-cohort, none of the risks presented here are statistically significant. Such results, while being highly informative for radiation protection purposes, to the extent of indicating that such a large cohort occupationally exposed to low doses does not have a statistically significant risk of solid cancer, are not especially informative for the comparisons of risk between early workers and more contemporary workers.

Limitations of the work presented here include some cohort overlap, i.e., 5,071 industrial radiographers were also in the nuclear power worker study population, of whom 482 died from all solid cancer. At this stage in the work on building the full MPS cohort from all of the separate cohorts, the overlap in workers has not yet been accounted for in the risk analysis for the separate cohorts, but will eventually be fully accounted for in the ongoing current work to form the full MPS database. The outcome of “all solid cancer” is not ideal because asbestos-related pleural cancers and mesothelioma may be included as well as smoking and asbestos related lung cancer. In the MPS paper on IR (3) a significant correlation of Chronic Obstructive Pulmonary Disease (COPD) mortality with radiation dose was reported, which could indicate that smoking could have influenced the radiation risk to dose response.

Further methodological improvements could come by keeping all risk analyses based on full cohorts, rather than selecting sub-cohorts for separate analyses, as presented here as a subsidiary analysis. Using an indicator variable, defined for early and more contemporary workers (or other covariable groups), to multiply the full-cohort excess or absolute risk per unit dose parameter, gives the risks in each indicator variable group relative to the full-cohort baseline. This results in baseline cancer rates (and therefore the excess and absolute risks too) being better defined in the baseline for the full cohort, than in the separate baselines for the separate sub-cohorts, because some of the age and birth cohort periods will be common to both sub-cohorts defined in the indicator variable. For the analysis presented here for NPP, there are some large differences in estimated EAR risks for the early workers, using sub-group analysis versus full cohort analysis with a monitoring period indicator variables on the dose response.

The work presented here for the IR and NPP cohorts do not support recent findings (6,7) from the INWORKS cohorts of increased risks for contemporary workers (Figure 1). It therefore appears premature to conclude that there is generally a difference in excess risk from radiation exposures between early and contemporary workers. From current radiobiological knowledge, it would be difficult to explain how cohort risks per unit dose can increase with time, and how cohorts, over time can become several-times more radiosensitive. It seems possible that the INWORKS (7) high ERR per Gy risks for contemporary workers (first hired in 1965 or later) of 1.44 (90%CI: 0.65, 2.32) comes almost entirely from the US-INWORKS cohorts. Additional insight about what could be causing the observed discrepant results from the different worker groups and hiring periods is that the high ERRs could only be based on a few estimated excess cases, and this requires a thorough analysis to check, because it is possible to have high ERRs but low EARs, with a low number of estimated excess cases. The French workers analysis (24) by Laurent et. al. does not explicitly evaluate early workers risks in comparison with contemporary workers. In fact, since follow-up started in 1968, the French cohort is open to the interpretation of being “contemporary” in its entirety. For males, the risk from the simple linear ERR model was reported as ERR per Gy = 0.71 (95% CI: −0.28; 1.80). Since “contemporary workers” did not have a significantly increased risk, this is open to the interpretation that a higher risk in contemporary workers is not seen for the French cohort. Kelly-Reif et. al. (6) called those first hired in or after 1960 to be contemporary workers in the US study, so it seems fair to infer that the entire French study could be considered “contemporary” by the US definition. A recently published paper on workers in the UK NRRW study had similar results to the IR and NPP results (see Figure 1) in that no noticeable difference in risks of all solid cancer incidence between early and contemporary workers was found (22).

This work has shown the advantages, connected with more fully understanding and interpreting risks, of the more thorough analysis that goes beyond just presenting ERR as in (6,7). The especially helpful metrics, applied here in the IR and NPP analyses, include estimated excess number of deaths (by dose category and period of employment/monitoring); estimated attributable fractions (also by dose category and period of employment/monitoring); EAR models; comparisons of goodness of model fit to the data between ERR and EAR models. The additional information provided by these measures, and their uncertainties, will facilitate assessments by radiation protection committees when evaluating the general validity of apparently high values of published ERR risks per unit dose, e.g., to see if published high risks per unit dose are supported by more than just a few estimated excess cases or not. Further areas of future effort required to gain a more thorough understanding of occupational risks will come from more data on monitoring for radionuclide intakes (25), and information on potential confounders such as e.g., smoking and asbestos exposure.

Other advantages will eventually come from continued efforts to pool high-quality cohort data to gain statistical power, as planned and currently being worked on with the MPS data.

CONCLUSIONS

The work presented here for the IR and NPP cohorts do not support recent findings from the INWORKS cohorts of increased risks for contemporary workers. This work has demonstrated that there are advantages to be gained in more fully understanding and interpreting risks in future radiation epidemiology, particularly if a more thorough analysis that goes beyond just presenting ERR is conducted applying the especially helpful metrics described above. The additional information provided by these measures will enable radiation protection committees to better assess if apparently high ERR risks are supported by high EAR risks and are based on more than just a few estimated excess cases. Further advantages will come by pooling high-quality cohort data – as planned in the MPS - to gain statistical power. It may also be advantageous to consider whether more suitable outcomes may be identified than “all solid cancer” taken together as one group.

It appears premature to conclude that there is generally a difference in excess risk from radiation exposures between early and contemporary workers.