Introduction

Ionizing radiation can cause cancer [1, 2]. Therapeutic doses of ionizing radiation to the heart and large arteries are associated with various types of circulatory disease [3–6]. More recently, and controversially, studies on several groups exposed to low-to-moderate doses of radiation have reported excess mortality and morbidity from circulatory diseases, in particular the Life Span Study (LSS) of Japanese atomic bomb survivors [7] and several occupationally exposed cohorts [8]. There is biological data suggesting there might be a variant response for circulatory disease below vs above about 0.5 Gy [9]. However, the complicated, multifactorial nature of circulatory disease, possible contributions from unmeasured confounders and errors in dose estimates inevitably raise concerns about whether the observed associations are causal [8].

Individuals receiving fluoroscopic X-rays as part of treatment for tuberculosis in Canada and Massachusetts have been studied in relationship to cancer [10–14], but noncancerous diseases have not been so extensively examined. A recent analysis of the Canadian fluoroscopy cohort study indicated small radiation-associated excess relative risks (ERR) of ischemic heart disease (IHD) mortality, with the highest risk for those with the most prolonged period over which the fluoroscopies took place [15]. Radiation-related risks of IHD also decreased significantly with increasing time since first exposure and age at first exposure [15].

We therefore analyzed the Massachusetts tuberculosis fluoroscopy cohort study to assess circulatory disease mortality. We decided a priori to concentrate on the relationship between cumulative lung and thyroid tissue dose (surrogates for dose to the heart and carotid artery, respectively) and death from several circulatory diseases and on possible dose-fractionation associations and modifications by age at exposure and time since exposure. The dose response overall and under 0.5 Gy will be assessed.

Materials and methods

Cohort characteristics and follow-up

The methods used to assemble the Massachusetts tuberculosis fluoroscopy cohort are detailed elsewhere [10, 14, 16]. Briefly, data collected from the medical records of patients with a primary diagnosis of pulmonary tuberculosis between 1915 and 1968 and discharged alive from 12 Massachusetts hospitals were identified, and their medical records were abstracted (Table 1). Cohort entry was defined as the date of admission to one of the participating institutions for treatment of tuberculosis. Of the 13,716 members of the full cohort, 144 were excluded for lack of adequate follow-up information, and another 4 for missing last exposure date, leaving an analysis dataset of 13,568 persons. This dataset is a slightly larger cohort than that considered by Davis et al. [14], because we were more successful at tracing the cohort members originally assembled by Boice [16] and Davis et al. [14]. Data were obtained on pneumothorax treatments, fluoroscopic X-ray exposures (which took place between 1916 and 1961), smoking and alcohol use, and information to assist in locating study subjects. The vital status was determined as of December 31, 2002. Deaths were retrospectively ascertained from the Vital Statistics Offices in the state of last known residence by linking to the mortality files of the Social Security Administration and the National Death Index and by contacting relatives and friends [16]. Vital status was also confirmed through records from the post office, motor vehicle departments, credit bureaus, and other sources [14].

Table 1 Causes of death for 13,568 patients in the Massachusetts tuberculosis fluoroscopy cohort more than 5 years after entry
Full size table

All causes of death on death certificates were coded again using the ninth revision of the International Classification of Diseases (ICD-9). The current analysis describes mortality from all circulatory diseases (ICD-9 codes 390–459), cerebrovascular diseases (CeVD) (ICD-9 430–438), IHD (ICD-9 codes 410–414), hypertensive heart disease (ICD9 401–405), all heart disease (ICD-9 390–429), and other cardiovascular (non-CeVD, non-heart) diseases (ICD-9 439–459) (see Table 1). These endpoints were chosen a priori because they might be radiogenic [8].

Dosimetry

Dosimetry is described elsewhere [14]. Briefly, exposure groups were defined by receipt of air-collapse therapy (pneumothorax/pneumoperitoneum) as indicated on treatment records. Air-collapse therapy was standard treatment for tuberculosis in the 1920s–1940s and involved injecting air into the pleural cavity to force lung tissue away from the chest wall. Typically this procedure was repeated, with the aid of a fluoroscopic examination, 2–3 times per month for over 2 years, and up to 5 years for patients with advanced disease. The radiation dose absorbed by several organs adjacent to the lung and exposed during the fluoroscopic procedures was estimated [17, 18]. This dosimetry method accounted for the number of fluoroscopies, calendar year of exposure, sex, age at treatment (<18, ≥18 years of age), and phantom studies of organ-specific doses using contemporary machine exposure settings to the extent possible.

Cumulative lagged doses to the lung, red bone marrow (RBM) and thyroid were estimated. We regard thyroid dose as a surrogate for dose to the carotid artery, and lung dose as a surrogate for dose to the heart; RBM dose was used because of suggestions of immunologic effects in circulatory disease [19, 20]. Therefore we used thyroid dose to analyze CeVD, and lung dose for all other circulatory disease endpoints, with RBM dose being used for certain sensitivity analyses (Table 6). For most analyses, cumulative dose was lagged by 5 years, as in most previous analyses of these endpoints [8].

Statistical methods

Each patient contributed person-years at risk from 5 years after starting treatment (or entry into the study for those unexposed) to December 31, 2002, or the date of death or last date contacted, whichever occurred earlier. In sensitivity analysis (not shown) we varied the exclusion period from 5 years to between 0 and 10 years. The fitted model assumed that the expected number of deaths in stratum i with cumulative lung/thyroid/RBM dose \( D_{i} \) (in Gy), lung dose rate \( DR_{i} \) (Gy y−1), age at first exposure \( a_{i} \), time since last exposure \( t_{i} \), and associated other covariates \( \left( {X_{ij} } \right)_{j = 1}^{n} \) is given by:

$$ PY_{i} \exp \left[ {\sum\limits_{j = 1}^{n} {X_{ij} \beta_{j} } } \right]\left[ {1 + \alpha D_{i} \exp \left( {\delta_{1} (a_{i} - 26.20) + \delta_{2} (t_{i} - 25.01) + \delta_{3} (DR_{i} - 10.44)} \right)} \right] $$
(1)

where \( PY_{i} \) is the number of person-years of follow-up. Age at first exposure, years since last exposure and dose rate (defined as the total dose, multiplied by 365.24, and divided by the number of days of irradiation) were centered by subtracting their person-year weighted mean values over the exposed part of the cohort, 26.20, 25.01 years, and 10.44 Gy year−1, respectively. For the purposes of the interaction analysis in Tables 3 and 4 we fitted a slight variant of this model in which for a given factor variable \( X_{m} \) taking values \( 1, \ldots ,M_{m} \) the expected number of deaths is:

$$ PY_{i} \exp \left[ {\sum\limits_{j = 1}^{n} {X_{ij} \beta_{j} } } \right]\left[ {1 + \sum\limits_{k = 1}^{{M_{m} }} {1_{{X_{im} = k}} } \alpha_{k} D_{i} \exp \left( {\delta_{1} (a_{i} - 26.20) + \delta_{2} (t_{i} - 25.01) + \delta_{3} (DR_{i} - 10.44)} \right)} \right] $$
(1′)

It should be emphasized that the models used all incorporated the variables to be used (alcohol consumption, cigarette smoking status, thoracoplasty status, pneumolobectomy status, tuberculous disease status, etc.) in the background model, so that we are testing specifically the adjustment to the radiation dose response. The only exception related to the radiation-specific variables (age at first exposure, years since last exposure, dose rate), which cannot be incorporated in the background model.

Maximum likelihood techniques [21] were used to fit the models with EPICURE [22] and thereby to estimate all the above model parameters, in particular the ERR/Gy, \( \alpha \). All tests were 2-sided with a specified type I error of 0.05, and unless otherwise stated all confidence intervals for risk estimates were derived from the profile likelihood [21]. A forward stepwise procedure [21] determined the form of the model of the underlying risk for each endpoint, in relation to all factors other than radiation dose. Terms were selected for inclusion in the model if a p value of 0.1 or less was achieved by their incorporation. We only evaluated interactions of sex × age and sex × calendar year, and evaluated interactions as groups of variables, namely sex × {ln[age], ln[age]2, …, ln[age]k}, and sex × {year, year2, …, yearm}); the maximal exponents of ln[age], year, namely \( k \), \( m \), were determined by the significance of the associated main effect. The criteria for selection of these groups of interaction variables was the same as for all other variables, namely a p value of 0.1 or less. The results of the analysis of background rates are given in Tables 8, 9, 10, 11, 12, 13 and 14 by endpoint. The forward-stepwise variable selection procedure we used to construct the background rate models was not automated. Automatic variable selection procedures can result in models in which higher polynomial powers of a variable are used, but not all lower order terms, and likewise can add interaction terms without both the associated main effect terms, both of them undesirable features of a model. Certain more automatic forms of the variable selection procedure use a mixture of forward (variable selection) and backward (variable elimination) methods, with different p value thresholds for selection and dropping of variables [23]; the choice of p values requires some care. We have used an alternative fully automatic method, described in “Appendix 2”, using Akaike’s information criterion (AIC) [24, 25] to choose an optimal background model. Minimizing AIC is a standard method of variable selection that avoids over-parameterised (and therefore over-fitted) models. AIC penalises against overfitting by adding 2 × [number of fitted parameters] to the model deviance. The selected variables for each endpoint are given in Table 15. The results of using this automatically selected set of models are shown in Table 16. A mixed forward–backward stepwise procedure was used, implemented in R [26].

We also fitted a simple generalized additive model (GAM) [27], in which the expected number of deaths is:

$$ PY_{i} \left\{ {\exp \left[ {\sum\limits_{j = 1}^{n} {X_{ij} \beta_{j} } } \right] + \kappa D_{i} \times \exp \left( \begin{aligned} &\theta_{1} (a_{i} - 26.20) + \theta_{2} (t_{i} - 25.01) \hfill \\ &+ \theta_{3} (DR_{i} - 10.44) \hfill \\ \end{aligned} \right)} \right\} $$
(2)

However, models other than that with the simplest possible radiation effect term, with \( \theta_{1} = \theta_{2} = \theta_{3} = 0 \), proved generally numerically unstable (results not shown). Therefore we present results only for this special case, the constant excess absolute risk (EAR) model:

$$ PY_{i} \left\{ {\exp \left[ {\sum\limits_{j = 1}^{n} {X_{ij} \beta_{j} } } \right] + \kappa D_{i} } \right\} $$
(2′)

Inference relates to the EAR coefficient \( \kappa \). Further details are given in Table 17. GAMs were fitted using EPICURE [22] and R [26].

Results

Among persons followed for 5 or more years (345,948 person-years of follow-up), 3221 died of circulatory diseases (Table 1). Overall, radiation had no marked effects on the circulatory system when adjusting for various lifestyle and environmental factors in the background (as per Tables 8, 9, 10, 11, 12, 13 and 14). For all circulatory disease, the ERR/Gy was −0.023 (95 % CI −0.067, 0.028, p = 0.3574, Table 2). There are stronger indications of excess risk for hypertensive heart disease (ERR/Gy = 0.357, 95 % CI −0.043, 1.030, p = 0.0907, Table 2). On the other hand, the dose–response for IHD was negative (ERR/Gy = −0.077, 95 % CI −0.130, −0.012, p = 0.0211, Table 2). The fits of the GAM (Table 17) were also generally non-significant, and some were numerically unstable.

Table 2 Excess relative risks for circulatory disease mortality in Massachusetts tuberculosis fluoroscopy cohort and modification by age at entry, years since entry, and dose rate
Full size table

Risk did not change significantly for any endpoint with continuous modification by dose fractionation, age at entry, or time since entry (Table 2). This lack of a marked effect was also the case when factor (grouped) modifications of the temporal and dose rate variables were employed (Table 3).

Table 3 Excess relative risks for all circulatory disease mortality in the Massachusetts tuberculosis fluoroscopy cohort and modification by groups of demographic (age at entry, years since entry, attained age) and lifestyle (cigarette smoking, alcohol consumption) variables
Full size table

Cigarette smoking did not significantly modify radiation risk, but the category of alcohol consumption did (p = 0.0075), with statistically significant excess radiation risk in the group whose alcohol consumption was unknown (Table 3).

Thoracoplasty, other surgery, and tuberculosis status did not significantly modify all circulatory disease radiation risk (p > 0.2), but pneumolobectomy did (p = 0.0319), with radiation risk highest (and statistically significant) among those reporting a pneumolobectomy (ERR/Gy = 0.252; 95 % CI 0.024, 0.579; Table 4).

Table 4 Excess relative risks for all circulatory disease mortality in the Massachusetts tuberculosis fluoroscopy cohort and modification by surgical status (thoracoplasty, pneumolobectomy, other surgery) or maximal tuberculosis disease status
Full size table

Risk in the low-dose region fluctuated considerably, with indications of excess risk for some endpoints (Fig. 1). If the dose range was restricted to less than 0.5 Gy, borderline significant elevations in ERR were associated with all circulatory disease (ERR/Gy = 0.345; 95 % CI −0.032, 0.764; p = 0.0743), IHD (ERR/Gy = 0.465; 95 % CI −0.032, 1.034; p = 0.0682), and for all heart disease (ERR/Gy = 0.352; 95 % CI −0.067, 0.824; p = 0.1032) (Table 5).

Fig. 1
figure 1

a Dose response for all circulatory disease, and all heart disease, with 95 % CI. b Dose response for cerebrovascular disease, and hypertensive heart disease, with 95 % CI. Lower panel in each graph is low dose (<0.5 Gy) part of upper graph

Full size image
Table 5 Excess relative risk (ERR) per Gy of various circulatory disease mortality endpoints by dose range examined
Full size table

Different organ doses were associated with a considerable range in risks (Table 6). Risks are particularly large in relation to RBM dose, the use of which increases the ERR/Gy for CeVD more than five-fold (to 0.676, compared with 0.132 for thyroid dose).

Table 6 Excess relative risk (ERR) per Gy of various circulatory disease mortality endpoints by organ dose used
Full size table

The main results used a follow-up period starting 5 years after entry (for those not exposed) or 5 years after last exposure (for those exposed), with cumulative doses lagged by 5 years. The results were essentially unchanged when these exclusion and lagging periods were varied between 0 and 10 years (results not shown). The results of using an automatically selected set of background models, selected to minimize AIC, are shown in Table 16. Comparison of this table with Table 2 indicates that very similar inference results from using this alternative set of background models.

Discussion

We found no strong evidence of radiation-associated excess risks for the all-circulatory disease mortality endpoint. Over the full dose range, there were borderline significant (p ≈ 0.1) indications of an excess risk for hypertensive heart disease. Borderline significant (0.05 < p ≤ 0.10) increasing trends were found for all circulatory disease, IHD, and all heart disease when dose was restricted to <0.5 Gy. Significant excess risk was found for pneumolobectomy. Dose fractionation, age at entry, and time since entry, did not modify radiation risk for circulatory mortality.

The absence of any fractionation effect in the present data contrasts with the inverse fractionation effect observed in the Canadian tuberculosis data [15]. However, the cohorts and analytical methods of these two studies differ in several ways. The significant dose-fractionation effect observed in the Canadian study was estimated for 10-year lagged lung doses, whereas we used 5-year lagged doses. When Canadian data were re-analyzed with the 5-year lag, the dose-fractionation was attenuated and no longer significant [15]. Whereas the Canadian study used time-dependent annual lung doses [15], we relied on cumulative lung and thyroid doses. We also defined dose rate differently. The Canadian study used actual days under treatment and fluoroscopy screening [15], and we defined duration of exposure as the difference between the dates of the first and last fluoroscopy. The two populations also differ, e.g., the Canadian cohort has different calendar times of exposure (1930–1952 vs 1901–1962 in our study). However, risks in the present cohort are entirely consistent with the overall pattern of risk (without adjusting for fractionation) in the Canadian data (Table 7).

Table 7 Estimated excess relative risks of circulatory disease in the present study and in various other studies of moderate- and low-dose radiation exposure
Full size table

Several authors and committees have reviewed evidence for excess risk of circulatory disease in groups exposed to low and moderate doses of radiation (mean dose < 0.5 Gy) [8, 9, 28]. For example, a recent systematic review and meta-analysis [8] documented statistically significant excess risk for three of the four major subtypes of circulatory disease. The risks in the present study, when evaluated over the full dose range or when restricted to less than 0.5 Gy, are similar to results in most other radiation-exposed groups (Table 7).

The candidate biological mechanisms for the circulatory disease effects of radiation have been recently reviewed [9, 28, 29]. At high radiotherapeutic doses (>5 Gy), the cell-killing effect on capillaries and endothelial cells plausibly explains effects on the heart and other parts of the circulatory system [29]. At lower doses (0.5–5 Gy), in humans and in in vivo and in vitro experiments, many inflammatory markers are upregulated long after exposure to radiation, although for exposures less than about 0.5 Gy, the balance shifts toward anti-inflammatory effects [9, 28, 30], implying that the initiating mechanisms for adverse effects in this dose range would not directly result from inflammation. A recent analysis of death from renal failure in the Life Span Study suggests that radiation-induced renal dysfunction may be a factor in increasing the risk of circulatory disease [31], and some experimental data support this suggestion [32].

We used thyroid dose (a surrogate for dose to the carotid artery) to analyze CeVD, and (as in the Canadian tuberculosis analysis [15]) lung dose (a surrogate for heart dose) to analyze all other endpoints. One would expect carotid artery dose to be higher than thyroid dose, but that lung dose is probably lower than heart dose; estimates of both the heart and carotid dose may be wrong by a factor of 2 [33].

Dose-related variations in T cell and B-cell populations in Japanese atomic-bomb survivors suggest that radiation may harm the immune system [34] at doses > 1.5 Gy, implying that whole-body or RBM dose might be the most relevant to the radiation effects of the associated systems. Although other evidence implicates infections and the immune system in cardiovascular disease [19, 35, 36], the negative findings of two randomized-controlled trials of antibiotic administration [37, 38] suggest that bacterial infection is not likely involved in circulatory disease. The somewhat high (albeit non-significant) risks for hypertensive heart disease and CeVD if RBM dose is used (Table 6) (weakly) suggest that dose to this tissue may not be relevant for these endpoints. There is biological data suggesting radiation-associated senescence of monocytes [39], and a somewhat similar mechanism based on monocyte cell killing in the arterial intima suggests that the arterial intima may be causally associated with initiating atheroma in the arterial wall [40] (although there are many other stages between that point and plaque rupture [41, 42]), so that mean arterial dose might be the most relevant organ or tissue dose for studying circulatory disease.

Several recent reviews [8, 9, 28, 43] describe the abundant radiobiological reasons for considering the studies of moderate and low doses separately from studies of high doses. The mechanisms relevant for lower doses are likely to differ from those relevant at higher (e.g., radiotherapeutic) doses. However, risks in studies of medically-exposed groups, with relevant organ doses usually well above 0.5 Gy, are generally consistent with those in populations exposed at the much lower doses and dose rates discussed above [3–6, 44], suggesting that mechanisms operating at high doses and high dose rates may be similar to those at low doses and dose rates. The fact that the IHD risks using mean heart dose in these high-dose/partial-body exposed groups are similar to the risks in the generally uniformly whole-body-exposed groups using whole-body dose discussed above (Table 7) also suggests that mean dose to the heart is the most relevant metric for predicting radiation-associated IHD [44]. In the current analysis, we used lung dose as a surrogate for heart dose.

Epidemiological research has identified specific hereditary and lifestyle risk factors for circulatory disease, including male sex, family history of heart disease, cigarette smoking, diabetes, high blood pressure, obesity, increased low-density lipoprotein cholesterol, and decreased high-density lipoprotein cholesterol plasma concentrations [45–47]. Many studies lack this information on lifestyle factors. Of the studies considered in Table 7 only those of the Japanese atomic-bomb survivors [7, 48], Mayak workers [49, 50], and Canadian fluoroscopy patients [15] had such information. Some lifestyle factors were included in the Nordic breast cancer case–control study [4], and specific medical factors (surgery, thoracoplasty, pneumolobectomy), alcohol consumption, and cigarette smoking were included in the cohort considered here. Cigarette smoking did not modify the dose response in the present cohort, although unknown alcohol consumption and pneumolobectomy did (Tables 3, 4). However, the importance of these findings is unclear, and they may best be interpreted as the effects of chance. In all other radiation-exposed groups with such information there is no evidence that lifestyle factors interacted with radiation risk [4, 7, 48–50].

Strengths of the study include the fact that results are based on a long-term follow-up of a large cohort of subjects of both sexes exposed at different ages. Risks could be evaluated from low-to-moderate radiation doses protracted over time. Dose was evaluated to a number of organs, in particular to the lung, which should be a reasonable surrogate to dose to the heart (as discussed above). The outcome and exposure information are both register-based, so most biases (e.g., due to misclassification of exposure or outcome) are unlikely. As noted above we have information on certain lifestyle and medical variables. A weakness of the study is that there are many other lifestyle and medical risk factors for circulatory disease that we lack information on. These include diabetes, hypertension, and obesity (and related to that exercise). It is possible that these may confound the radiation dose response that we observe. However, as discussed above, there is little information in other studies to suggest interactions of such variables with radiation risk.

The International Commission on Radiological Protection has classified circulatory disease as a tissue reaction effect [51], with a threshold dose of 0.5 Gy. The threshold was derived by fitting a linear model to epidemiologic data and selecting the dose below which there was less than a 1 % chance of an effect. As such this does not represent a true no-effect dose threshold. Schöllnberger et al. [52], analyzing somewhat older Japanese atomic bomb survivor data, concluded that for CeVD and cardiovascular disease, risk estimates are compatible with no risk below threshold doses of 0.62 and 2.19 Gy respectively. However, this analysis is controversial [53]. The analysis of Table 5 suggests that a threshold of the order of 0.5 Gy is marginally inconsistent with the pattern of radiogenic excess risk observed in the Massachusetts tuberculosis fluoroscopy sub-cohort.

In summary, we found no strong evidence of radiation-associated excess risks for the circulatory disease overall. In contrast to the findings in the generally similar (although somewhat larger) Canadian TB fluoroscopy cohort, there was no indication of an inverse fractionation effect. However, borderline significant increasing trends were observed for all circulatory disease, ischemic heart disease, and all heart disease when dose was restricted to <0.5 Gy. The magnitude of the trends both overall and <0.5 Gy are consistent with those in other groups exposed at moderate and low doses. However, the indications of a much steeper low dose slope are unexpected, and should be tested against other data.