The Colorado Plateau cohort has been described previously.⁶ Briefly, the cohort was assembled by compiling records from a series of cross-sectional examinations held periodically between 1950 and 1960 among active uranium miners in southwestern USA.^1,2 The cohort consists of all male miners who participated in a medical examination and questionnaire at least once between 1950 and 1960 and worked underground in a uranium mine for at least one month (n=4137). Initially, miners were followed up triennially via interview and physical examination and later by record linkage to multiple sources. Since the previous reported follow-up,⁶ we linked records for all miners not known to be deceased with the National Death Index (NDI), death records of the Social Security Administration, and tax records of the Internal Revenue Service, which confirmed status as either alive or deceased. Causes of death for deceased miners were obtained from the NDI or death certificates, which were coded to the revision in effect at the time of death. Mortality follow-up was extended through 31 December 2016.

RDP exposure estimates are based on thousands of measurements collected from hundreds of mines between 1951 and 1969.^1,2 Exposures from both uranium mining and other hard rock mining are based on the historical unit “working level month” (WLM), equivalent to a potential alpha energy exposure of 3.54 mJ hour m^-3.¹⁵ In previous analyses,⁶ available data consisted of the dates on which each miner’s RDP exposure reached 9 pre-specified cumulative exposure levels. Here, we made extensive effort to link the cohort to original dosimetry records to verify employment data and exposure estimates by mine and year. Data are available at the level of the combination of individual miners’ work history data and the periodic exposure estimation (usually, annual) made for each mine. We interpolated daily rates of exposure between these dates for each worker. We included estimated exposure to RDP from previous non-uranium hard rock mining according to the dates estimated for each miner before beginning uranium-mine employment. Uranium mine exposure is subject to uncertainties, particularly before 1960, including exposure measurement and estimation error;¹⁶ however, consideration of these uncertainties is beyond our scope here.

Quantitative cigarette smoking histories were collected for nearly all cohort members using surveys conducted in the 1950s-1970s, and (for white miners) 1986.³ Data were available in two forms: smoking category at the last survey (never, former, current <1 pack per day, current 1 pack per day, current >1 pack per day), and cumulative number of cigarettes smoked (expressed in pack-years) for ever-smokers. To facilitate analyses based on pack-years, we estimated a quit date to reflect a more realistic lifetime smoking quantity, given the long follow-up period. We created a detailed smoking history for each miner, individually imputing the last smoking date for each current smoker at the time of last questionnaire (Supplemental Materials, Supplemental Tables 1–3), conditional on expected values for single-year age- and 5-year birth cohort-specific cessation probabilities for U.S. men.¹⁷

We examined associations between RDP exposure and mortality for the following cancer types with at least 10 cases: all cancers excluding lung and neoplasms of uncertain type overall or in the respiratory system (ICD-9 codes: 140–208, except 162, 165, 195, 199); lung (162); extrathoracic airways (140–148, 160–161, 164); digestive cancers as a group (150–159); esophagus (150); stomach (151); intestine (152–153); liver and gall bladder (155–156); pancreas (157); prostate (185); melanoma (172); all skin combined (172–173); lymphatic & hematopoietic (200–208, 273.3) and its subcategories non-Hodgkin lymphoma (200, 202, 273.3), multiple myeloma (203), and leukemia (204–208).

For Aim 1, we used the NIOSH Life Table Analysis System (LTAS)^18,19 to estimate standardized mortality ratios (SMRs) compared with the male population of Arizona, Colorado, New Mexico, and Utah (for white miners) and with those in Arizona and New Mexico (for miners of other races), as described previously.⁶ Person-time and event tabulation began at the later of date of first examination or 1/1/1960 (when American Indian miners’ follow-up⁵ and general population rates began) and ended at the earliest of date of emigration (n<5), loss to follow-up, death, or 31/12/2016. SMRs were indirectly standardized on race, age, and calendar period, stratifying by dose category and decade of follow-up for lung and other cancers. We also calculated internally adjusted standardized rate ratios (SRRs) directly standardized to the race, age, and calendar year distribution of the entire cohort. Using SRRs, we estimated EAR trends across RDP exposure in lung cancer mortality rates.¹⁹

For Aims 2–4, to evaluate temporal patterns in RDP-associated lung cancer mortality rates and effect modification by smoking, we used full risk sets with attained age as the time scale, also caliper-matching on birth cohort within 2.5 years of the case.^20,21 Partial likelihoods when using the full risk set are identical to those from Cox proportional hazards regression. We estimated the ERR per 100 WLM using models commonly explored in other studies of uranium miners.^8,13 Analyses were conducted using the PHREG procedure in SAS (ver. 9.4).

To address Aim 2, we fitted cubic b-spline models, with two equally spaced knots,²² to flexibly examine the shape of exposure-response patterns and their variation with TSE, attained age, and smoking level. We fit the following model, based on the BEIR VI⁸ model considering effect modification by attained age, TSE, and exposure rate ERR=β·(w5-14+θ1·w15-24+θ2·w25-34+θ3·w35+)·φage·φdoserate

where w_5–14, w_15–24, w_25–34 and w₃₅₊ represent windows of cumulative exposure accrued 5–14, 15–24, 25–34 and 35+ years prior to attained age, respectively, and, therefore, their sum equals a worker’s cumulative exposure lagged 5 years. β represents the ERR of the weighted sum of cumulative exposures (weighted by parameters θ_i), φ_age represents a modifying factor of attained age in categories <55, 55–64, 65–74 and 75+ years old, and γ_{dose rate} represents modifying factors of the radon progeny alpha energy decay airborne concentration (mJ m^–3), in categories <10 and 10+ working levels (WL, calculated as WLM/number of months in uranium mining). Studies of RDP suggest that, at a given total exposure, low exposure rates are more effective than high exposure rates in causing cancer. In previous analyses of this cohort,⁴ a cutpoint of 10 WL was found to capture most of these effects. The above model does not control for smoking (which is not related to RDP exposure in this cohort; Pearson r = −0.00123, p=0.9370) but matches on age and birth date. The ability to differentiate attained age from TSE effects is limited at their lower and upper extremes, due to their strong correlation in this cohort (Supplemental Table 4).

We addressed Aim 3 using three approaches:

For Aim 4, we examined the other cancer types under consideration, matching on age and birth cohort. Because other cancers may have a different latency pattern than lung cancer, we identified an optimal lag through a grid search by fitting all yearly lags between 0 and 40 and choosing the model with the lowest −2 log-likelihood (as for studies of US nuclear workers²⁴).

With 695 deaths, the overall lung cancer SMR was 4.55 (95% CI: 4.21, 4.90). The highest SMRs were seen among miners exposed to more than 120 WLM, monotonically increasing above this level to nearly 10 above 1000 WLM (Supplemental Table 5a). Stratified by decade of follow-up, we observed a clearly decreasing pattern of SMRs over time, from 6.3 in the 1960s to 3.7 in the 2010s; however, within exposure categories the patterns were more equivocal. The EAR per 100 WLM increased across decades, from 2.8 to 16 cases per 10,000 person-years from the 1960s to the 2000s, declining in the 2010s (Supplemental Fig. 1).

Lung cancer SMRs increased substantially across cigarette pack-years, in the cohort overall and within categories of RDP exposure (Supplemental Table 5b). However, the relative increase in smoking-related SMRs was much greater for miners exposed to 0-<60 WLM than for miners exposed to more than 1000 WLM. Within each smoking category, a clear pattern of RDP-related exposure-response was seen above 120 WLM, with RDP-related SRR increases much higher among the 0-<10 pack-year smoking category than in the ≥50 pack-year category (Supplemental Table 5b). However, the EAR/100 WLM for lung cancer mortality increased monotonically across smoking categories, from 5 to 15 deaths per 10,000 person-years in the lightest to heaviest smoking groups.

The shape of the exposure-response curve for lung cancer mortality with RDP appears linear in the lower-exposure region, with curvilinearity above approximately 1200 WLM (Supplemental Fig. 2a). In the BEIR VI exposure-age-concentration model, at 5–14 years since exposure the estimated ERR/100 WLM at age <55 was 4.04 for dose rates <10 WL and 2.14 for higher dose rates (Table 2). ERR/100 WLM decreased notably with TSE: stratified on dose rate at 10 WL, exposures received 15–24, 25–34, and ≥35 years before age-at-risk were 87%, 49%, and 36% as effective, respectively, as those received in the 5–14-year window. Spline models confirm this pattern (Fig. 1a), with ERR/100 WLM stabilizing by about 40 years post-exposure.

ERR/100 WLM did not vary greatly by age-at-exposure up to age 40; however, above this age, rate per unit exposure increased, although with widening confidence intervals (Fig. 1b). By contrast, strong heterogeneity (p<0.001) was observed by attained age: compared to age <55, the ERRs/100 WLM for ages 55–64, 65–74 and ≥75 were 61%, 30%, and 16% as high, respectively (Table 2). Spline models suggest this decline in ERR with attained age levelled off by about age 70 (Fig. 1c). Examining ERR/100 WLM by smoking category, Supplemental Fig. 3 illustrates the diminishment of RDP-associated risk after about 25 years post-exposure for the heaviest smokers, with rates declining less steeply among lower smoking categories.

We used three approaches to evaluate the interaction between RDP exposure and cigarette smoking. In the first approach, the interaction term for pack-years and WLM exposure was negative and marginally significant (p=0.042) in a multiplicative model adjusting only for age, birth cohort, and race (Table 3). However, in a model that included the interactions of attained age with both WLM and smoking pack-years, interaction for pack-years and WLM exposure was not significant (deviance χ²=3.5, 3 df, p=0.321). Supplemental Fig. 2b indicates steeper RDP exposure-response curves for lower compared to higher smoking pack-years, and Fig. 1d shows the relatively modest decrease in ERR/100 WLM across increasing smoking pack-years.

In the second approach, the estimate of λ in a geometric mixture model was 0.288 (95% CI: 0.0250, 0.551), indicating a supra-additive but sub-multiplicative interaction between WLM and pack-year exposure. At cumulative exposure levels below 400 WLM, λ was 0.438 (95% CI: −0.183, 1.058), moving towards multiplicative.

In the third approach, the interaction appeared supra-additive (Supplemental Table 6). In an examination of departures from additive joint effects, only one combination of WLM and pack-year categories had an observed joint absolute risk (AR) less than the expected joint AR. The mean ratio of the observed to expected joint AR was 1.9, with no clear pattern across smoking categories. However, the observed-to-expected joint AR values tended to decrease across WLM exposure categories, suggesting an interaction closer to additive at higher RDP exposures. This was confirmed in the examination of joint multiplicative effects (Supplemental Table 7): overall, the interaction appeared sub-multiplicative, in that the mean of the observed-to-expected joint rate ratio (RR) across all dose and smoking categories was 0.74. However, at RDP exposures >400 WLM, the interaction was clearly sub-multiplicative (observed-to-expected ratio of joint RRs ranged from 0.29 to 0.60), while below 260 WLM, the interaction appeared super-multiplicative (observed-to-expected joint RRs ranged from 1.1 to 1.9). As smoking increased, the interaction moved away from multiplicative to additive across WLM categories (Supplemental Table 7).

Pancreas was the only individual cancer site other than lung that was significantly positively associated with RDP exposure (Table 4; Supplemental Table 8): under the best-fitting lag of two years, the ERR/100 WLM was 0.07 (95% CI: 0.01, 0.3). Estimated exposure-response coefficients were weakly and imprecisely positive for extrathoracic airways, intestine, and prostate cancers and were significantly negative for leukaemia and multiple myeloma (Table 4).