Incidence of Diabetes in the Atomic Bomb Survivors: 1969-2015.

CONTEXT: Recent epidemiological studies have shown increased risk of diabetes among childhood cancer survivors who received high therapeutic doses of radiation, particularly to the total body or to the abdomen. However, the effect of low-to-moderate dose radiation (<4 Gy) on the risk of diabetes is still unknown.
OBJECTIVES: To investigate the radiation effect on diabetes incidence among atomic bomb (A-bomb) survivors, and whether the dose response is modified by other factors including city, sex, and age at time of bombing (ATB).
METHODS: 9131 participants without diabetes at baseline were observed through biennial clinical exams from 1969 to 2015. A Cox proportional hazards model was used to estimate hazard ratios (HR) to evaluate the dose response for diabetes incidence.
RESULTS: During the study period, 1417 incident diabetes cases were identified. The overall crude incidence rate was 7.01/103 person-years. Radiation dose was significantly associated with diabetes incidence, with effect modification by city and age ATB. In Hiroshima, at ages 10 and 30 ATB, the HRs at 1 Gy of pancreatic radiation dose were 1.47 (95% CI, 1.31-1.66) and 1.13 (95% CI, 0.97-1.31), respectively. However, no significant radiation dose response was observed at these ages in Nagasaki. The HR for radiation dose was higher among those who were younger ATB and decreased 1% for each additional year of age.
CONCLUSIONS: Among A-bomb survivors, a radiation association was suggested for incidence of diabetes. Results were inconsistent by city and age ATB, which could indicate potential confounding of the radiation association with diabetes.

In Japan, the prevalence of diabetes has remarkably increased with changes in lifestyle and environmental factors since World War II (1,2). Diabetes has now become a major public health problem and an important risk factor for various diseases and conditions including cardiovascular diseases (2-4). The Radiation Effects Research Foundation (RERF), formerly the Atomic Bomb Casualty Commission, has conducted a wealth of research into the long-term effects of atomic bomb (A-bomb) radiation exposure, including through biennial health examinations since 1958 (called the Adult Health Study [AHS]) (5). However, regarding long-term radiation effects on diabetes risk, thus far there have been only a small number of studies with limited observation periods (6-10). The first study conducted by Rudnick et al included 3581 A-bomb survivors who had participated in AHS health examinations from January through August of 1960 in Hiroshima (6). Differences in diabetes risk by radiation exposure groups were suggested for females only, but the relationship was not significant. Subsequently, there were 2 studies among AHS participants in both Hiroshima and Nagasaki between 1958 and1962 (7) and 1958 and 1978 (8), respectively, wherein no significant associations of A-bomb radiation exposure and prevalence or incidence of diabetes were observed. After the 1980s, 2 cross-sectional studies among AHS participants in Nagasaki were conducted. An association between radiation dose and prevalence of diabetes and impaired glucose tolerance was suggested for younger female participants between 1983 and 1987 (9), although no significant association was found for Nagasaki AHS participants between 1990 and 1992 (10). Thus far, there has been no clear evidence to suggest increased risk of diabetes as a result of A-bomb radiation exposure.

Diabetes has recently emerged as a late effect of radiation therapy among childhood cancer survivors who underwent total body irradiation or radiation to the head and the abdomen (11-17). For survivors exposed to abdominal radiation, linear or linear-exponential dose-response relationships have been suggested, with those at younger ages at cancer diagnosis or radiation therapy showing greater risk of diabetes (18-20). However, it remains unclear whether the observed radiation effects on diabetes risk are also seen in the low-to-moderate dose range such as that received by the A-bomb survivors in the AHS cohort (where estimated doses range from 0-4 Gy, with the vast majority below 1 Gy).

In the present study, we conducted a comprehensive review of all data available from participants in the AHS clinical program. We investigated the association of pancreatic dose with incidence of diabetes with about 50 years of follow-up and evaluated whether any observed radiation association was modified by other factors including city, sex, and age at time of bombing (ATB).

Materials and Methods

AHS Cohort

The AHS is a clinical program that aims to investigate late effects of radiation exposure on A-bomb survivors in Hiroshima and Nagasaki through biennial health examinations. Each AHS biennial health examination consists of history taking, physical examination, laboratory tests, and an interview conducted by trained nurses to obtain medical and lifestyle information.

The original AHS cohort was established in 1958 and consists of a core group and 3 city-, age- and sex-matched comparison groups. The core group included all individuals who were exposed within 2 km of the hypocenters and who reported symptoms of acute radiation exposure (about 5000 survivors). Each of the comparison groups were approximately the same size as the core group and consisted of (1) individuals without acute radiation symptoms who were within 2 km of the hypocenters; (2) individuals who were exposed more than 3 km from the hypocenter; and (3) individuals who were not in city (NIC) ATB. Follow-up of the NIC group was discontinued in the 1980s due to suspected differences in socioeconomic status between NIC members and other groups. In 1977, another approximately 2400 individuals with relatively high doses and their controls were added to the AHS cohort (expansion cohort). A total of 18 610 AHS participants have visited RERF at least once since 1958. The participation rate for residents in the contact area (both cities and their neighboring towns) was higher than 75% until around 2005 and subsequently has decreased with aging. A detailed description of this program has been published elsewhere (5).

Measurements and Assessment of Incident Diabetes

Selective measurement of serum glucose by autoanalyzer was started in 1969 in the Hiroshima laboratory using the Technicon Autoanalyzer N-2b method with a modification of Hoffman’s potassium ferricyanide reduction method. From 1979, measurement of serum glucose was conducted separately in each city’s laboratory, and the glucose oxidase method was used for Nagasaki participants only. Prior to 1986, measurement of serum glucose levels or an oral glucose tolerance test (OGTT) was performed only for participants whose initial screening of urine glucose by nonfasting urinalysis was positive.

After 1986 (1987 in Nagasaki), serum glucose was routinely measured for all participants at every health examination using the glucose dehydrogenase method and later the hexokinase/glucose-6-phosphate dehydrogenase method (from 1997 onward). The machines in both laboratories were calibrated every month with control blood samples to maintain reproducibility and consistency.

Diabetes was diagnosed in accordance with the Japanese Diabetes Association criteria in 1982, 1999, and 2010 (21-23). The diagnostic criteria for diabetes included any of the following: (1) a fasting blood glucose level ≥ 7.8 mmol/L (140 mg/dL) before April 30, 1999 or ≥ 7.0 mmol/L (126 mg/dL) after May 1, 1999; (2) a nonfasting blood glucose level ≥ 11.1 mmol/L (200 mg/dL); (3) a 2-hour glucose level ≥ 10.0 mmol/L (180 mg/dL) or ≥ 11.1 mmol/L (200 mg/dL) for participants who underwent 50-g OGTT or 75-g OGTT, respectively; (4) a self-report of a new diabetes diagnosis; or (5) initiation of medical treatment for diabetes during the follow-up period.

Study Subjects and Observation Period

In the current study, observation began at the first AHS health examination after January 1, 1969 when measurement of serum glucose level by autoanalyzer was started. Participants were followed until either a diagnosis of diabetes (using the midpoint date between the AHS exam at which the diabetes was detected and the prior AHS visit) or their last AHS visit date prior to the end of 2015 (at which time the participant was censored).

Of the original AHS cohort and its initial expansion comprising 18 610 participants, this study excluded 4087 participants who were NIC, 1863 participants with unknown radiation dose, 3005 participants who did not attend at least 2 examinations during the observation period between 1969 and 2015, and 524 participants with a diabetes diagnosis at baseline (prior to or during their first clinical visit after 1969). The remaining 9131 participants were analyzed in the present study. All participants provided informed consent, and the protocol was approved by the RERF’s institutional review board.

Radiation Dosimetry

Weighted, absorbed pancreas dose estimates, calculated as the sum of the gamma ray dose and 10 times the neutron dose, were used from the newest dosimetry system (called DS02R1) (24). DS02R1 doses were calculated by DS02 based on improved information on survivor’s location and shielding. Doses were corrected for dose uncertainty and truncated at 4 Gy.

Covariates

Participants’ sex, city, age ATB, and proximal-distal location ATB (defined at 3 km ground distance from the hypocenter) were obtained from the AHS cohort source data. Smoking and alcohol drinking histories were compiled as time-varying covariates based on information collected from mail surveys, self-reported questionnaires, and clinical interviews throughout the follow-up period. Due to the previously mentioned changes in diagnostic criteria for diabetes and procedures for measuring blood glucose over the course of follow-up, we identified and accounted for 5 calendar time periods (1969-1982, 1982-1986, 1986-1999, 1999-2010, and 2010-2015) in the analysis. The presence of any family history of diabetes was only ascertained via interview at the first AHS clinical exam, while parental history of diabetes was asked once during 1986 to 1999. These limited sources were combined into a single indicator of whether a history of diabetes was reported for any family member. Family history information was missing from both sources for only 7 subjects, who were treated as having no reported family history.

Statistical Analysis

Regression models for event-time outcomes were fit with attained age as the primary time scale, accounting for left truncation of the original AHS cohort and the expansion cohort. We used a Cox proportional hazards model to estimate the cause-specific hazard ratio (HR) of radiation on diabetes incidence. The baseline hazard was an unspecified function of attained age, stratified by categories of age ATB (<15, 15-29, ≥30) and calendar time (with breaks in 1982, 1986, 1999, and 2010) to flexibly account for cohort and period differences in background incidence. All models adjusted for sex, city, AHS subcohort (original cohort, expansion cohort), and time-varying smoking and alcohol status (classified as never, current, former, or unknown). To investigate potential effect modification of the radiation dose effect, we tested as a set its interactions with sex, city, and continuous linear age ATB. Interaction terms were only retained when significant. In all primary analyses, the nominal level of significance was 0.05. We did not adjust for multiple comparisons in analyses of dose effect to avoid overlooking potentially important determinants of radiation effect (25). Also, we included known potentially important factors in the model of the baseline diabetes incidence regardless of their significance because we thought it was more important to include factors in the baseline that may be influential rather than exclude them and possibly introduce bias in our assessment of dose response.

We performed secondary analyses evaluating models with linear, linear-quadratic, and categorical doses (with breaks at 0.005, 0.1, 0.25, 0.5, 0.75, 1, and 2 Gy), which inform the shape of the radiation dose-response. Furthermore, we explored a range of 2-parameter fractional polynomials of various functional forms against the simple linear model to assess the adequacy of the fit to the data.

Several modeling assumptions were assessed to ensure adequacy and robustness of the final fitted model, including proportional hazards of the predictors and switching the time scale from attained age to time since exposure. Furthermore, sensitivity analyses were used to investigate some potential confounders. An indicator for family history of diabetes was added to the model to assess any impact on the radiation risk estimate. Due to the limitations of the information we have available regarding family history of diabetes, it was not retained in the final model. Given that the hypocenter of the A-bomb was over the urban city center in Hiroshima but more rural in Nagasaki (26), there is potential for confounding between radiation dose and distance due to background differences in diabetes incidence associated with urbanicity. We adjusted for proximal/distal category (defined at 3 km), and its interaction with city to assess whether there was any impact on the risk estimates when redefining the baseline reference group as proximal survivors in each city. We also calculated E-values to assess the potential impact of unmeasured confounding on any observed association between radiation exposure and diabetes incidence (27,28).

Results

Among the 9131 eligible participants of this study, approximately 70% (n = 6331) were Hiroshima residents and 65% (n = 5937) were women. Table 1 shows the distribution of participants by radiation dose and city. About 11% of participants had high doses (≥1 Gy), while 40% had estimated doses below 0.005 Gy. Ages ATB tended to be higher in Hiroshima than Nagasaki.

Table 1.

Distribution of participants by weighted pancreas dose and city

	Weighted absorbed pancreatic dose
	<0.005Gy	0.005 to <0.5Gy	0.5 to <1Gy	1 to <2Gy	≥2Gy	Total
Hiroshima	n = 2403	n = 2431	n = 802	n = 522	n = 173	n = 6331
Sex
Men	801	732	275	200	70	2078
Women	1602	1699	527	322	103	4253
AHS groups
Original cohort	2090	2336	594	404	144	5568
Expansion cohort	313	95	208	118	29	763
Age ATB, years	26.7 (14.2)	28.3 (13.7)	25.8 (14.0)	24.3 (14.1)	21.6 (13.4)
Nagasaki	n = 1300	n = 592	n = 579	n = 271	n = 58	n = 2800
Sex
Men	551	206	212	122	25	1116
Women	749	386	367	149	33	1684
AHS groups
Original cohort	953	516	430	222	49	2170
Expansion cohort	347	76	149	49	9	630
Age ATB, years	22.2 (12.7)	23.8 (14.0)	21.8 (12.2)	20.6 (12.6)	19.0 (12.3)

Data are expressed as n or mean (SD).

Abbreviations: AHS, Adult Health Study; ATB, at time of bombing.

The median follow-up time was 21 years [interquartile range (IQR) 10-34 years]. Among 6543 subjects who died over the course of the study, the median time from their last AHS visit to their death was 2 years (IQR 1-7 years). Among 2588 subjects who were alive at the end of the study period (December 31, 2015), the median time from their last AHS visit (at which they were censored) to the end of the study was 3 years (IQR 1-15 years).

We identified 1417 incident cases of diabetes (606 men and 811 women) during 202 071 person-years of follow-up. Incidence rates of diabetes tended to increase with calendar year. The crude incidence rate among men (9.37 per 1000 person-years) was higher than that among women (5.90 per 1000 person-years) (Table 2). The crude incidence rates in Hiroshima and Nagasaki were 7.24 and 6.58 per 1000 person-years, respectively.

Table 2.

Crude incidence rates of diabetes by sex, city, and radiation dose category

	Participants	Person-years	Cases	Rate per 103
Sex
Male	3194	64 686	606	9.37
Female	5937	137 385	811	5.90
City
Hiroshima	6331	133 504	966	7.24
Nagasaki	2800	68 566	451	6.58
Age ATB
<15	2268	66 192	485	7.33
15-29	3462	89 345	659	7.38
≥30	3401	46 533	273	5.87
Pancreas dose, Gy
0 to 0.005	3703	84 246	554	6.58
0.005 to <0.5	3023	65 922	443	6.72
0.5 to <1	1381	30 407	218	7.17
1 to <2	793	16 831	141	8.38
≥2	231	4665	61	13.08
Overall	9131	202 071	1417	7.01

In a simple linear dose-response model with adjustment for conventional background factors (sex, city, age ATB, AHS subcohort, smoking and alcohol use status, and diagnosis period) but no effect modification, radiation dose was significantly associated with diabetes incidence (HR at 1 Gy = 1.24; 95% CI 1.13-1.36). There were no significant associations between incidence of diabetes and background factors except for sex and city, wherein men and Hiroshima participants showed increased risk. When testing for effect modification of the linear dose response, city and age ATB showed significant interactions with radiation (Table 3). The interaction of sex with radiation was not significant (P = 0.48), and thus was not retained in the final model. Among Hiroshima participants who were age 10 ATB, the HR at 1 Gy was 1.47 (95% CI 1.31-1.66), while for those at age 30 ATB, the HR at 1 Gy was 1.13 (95% CI 0.97-1.31). The radiation-associated HR in Nagasaki was not significant (age 10 ATB: HR at 1 Gy = 1.13; 95% CI 0.95-1.35; age 30 ATB: HR at 1 Gy = 0.87, 95% CI 0.71-1.07). The HR at 1 Gy for radiation dose decreased 1% for each additional year of age ATB (multiplicative change in HR at 1 Gy per year of age ATB = 0.99; 95% CI 0.98-1.00). There was no evidence of a 3-way interaction among radiation dose, city, and age ATB (P = 0.76).

Table 3.

Multivariable-adjusted hazard ratios (95% CIs) for incidence of diabetes based on a linear dose model

Variables	Hazard ratio (95% CI)	P-value
Women (vs men)	0.68 (0.59-0.79)	<0.001
Nagasaki (vs Hiroshima)	0.94 (0.82-1.09)	0.42
AHS groups
Expansion cohort (vs original)	1.01 (0.87-1.18)	0.86
Smoking statusa
Current (vs never)	1.16 (0.98-1.36)	0.09
Former (vs never)	1.10 (0.93-1.30)	0.28
Unknown (vs never)	0.93 (0.67-1.31)	0.69
Alcohol drinking statusa
Current (vs never)	1.01 (0.88-1.15)	0.90
Former (vs never)	1.18 (0.90-1.54)	0.24
Unknown (vs never)	0.76 (0.56-1.02)	0.07
Radiation dose of 1Gyb	1.24 (1.10-1.39)	<0.001
Effect modification of radiation
Radiation dose × Nagasaki	0.77 (0.63-0.94)	0.01
Radiation dose × age ATB	0.99 (0.98-1.00)	0.002

The model is adjusted for all variables listed in the table.

Abbreviation: ATB, at time of bombing.

aTime-varying covariates.

bEstimated for Hiroshima residents at the median age at time of bombing (23 years old).

To assess patterns across the dose range within each city and evaluate the fit of the simple linear model, the city-specific linear HRs with 95% confidence bounds for ages ATB of 10 and 30 are shown in Figure 1, along with HRs estimated separately by city and dose categories. No dose response was apparent in Nagasaki, while there was a gradual increase seen in Hiroshima. Effect modification of the radiation dose response by age ATB is shown both linearly and categorically for each of the cities in Figure 2. In both cities, those at younger ages tended to have higher HRs than those who were older, although the pattern was less pronounced in Nagasaki (where an overall radiation effect was not observed).

Figure 1.

Estimated dose response for incidence of diabetes in Hiroshima and Nagasaki, 1969 to 2015. Linear radiation dose response (black) and 95% confidence bounds (grey) at age at time of bombing (ATB) of individuals who were 10 (solid lines) and 30 (dashed lines) years old are shown for the incidence of diabetes. The estimated relative risks and 95% CIs at age ATB of those who were 10 (● ) and 30 (▲) years old are shown for 4 dose categories. Background was stratified by categories of age ATB (<15, 15-29, and ≥ 30) and calendar time (with breaks in 1982, 1986, 1999, and 2010). The model also adjusted for sex, Adult Health Study group, and smoking and alcohol drinking status.

Figure 2.

Effects of age at time of bombing (ATB) and city on the radiation dose response for diabetes incidence. The curves display the trend for age at exposure in hazard ratio at 1 Gy based on the linear dose model with effect modification by city and continuous age ATB. The points are hazard ratios at 1 Gy for each combination of city and age ATB category (<15, 15-29, and ≥ 30) with 95% CIs.

To assess nonlinearity in the radiation dose response, we added a simple quadratic term to the linear model, which showed marginal but not significant evidence of upward curvature (P = 0.06). When we fit a range of flexible, 2-parameter fractional polynomials for the radiation dose response, there was no evidence of significantly improved fit compared to a simple linear model. Therefore, the linear dose-response model appeared to be adequate.

There were no indications of violation of the proportional hazards assumption, and the observed results were robust to other modeling variations, such as stratifying all background variables and using time since exposure as the primary time scale. Family history of diabetes was reported by 686 participants (7.5%). The family history indicator was a strong background predictor of diabetes incidence (HR 1.69, 95% CI 1.44-1.98), but its inclusion in the model did not impact the radiation risk estimate or its effect modifiers. Sensitivity analyses confirmed that adjustment for the distal (>3 km) category by city, adjustment for the lowest-dose (<0.005 Gy) category, or exclusion of the low-dose participants altogether (n = 3703) resulted in no substantial change in the overall linear dose estimate nor its effect modification by city and age ATB. The E-value for the association of radiation dose with diabetes incidence was 1.59 (lower limit 1.34).

Discussion

The Atomic Bomb Casualty Commission-RERF has conducted a limited number of studies regarding radiation exposure and risk of diabetes, and the findings have been inconclusive (6-10). AHS studies conducted throughout the 1970s reported no significant dose response for the prevalence or incidence of diabetes in both cities (6-8). While an association between radiation dose and prevalence of diabetes and impaired glucose tolerance was suggested for younger female Nagasaki participants between 1983 and 1987 (9), no significant association was found for Nagasaki participants between 1990 and 1992 (10). We evaluated the radiation dose-response relationship for incidence of diabetes and observed increased incidence of diabetes. This is the first longitudinal, comprehensive study in both cities to investigate the radiation effect on diabetes incidence, with consideration of effect modification and time-varying covariates.

Recent large-scale cohort studies of childhood cancer survivors have revealed late effects on diabetes. Increased risks of diabetes were observed for childhood cancer survivors who underwent total body, cranial, and abdominal irradiation, compared to those who did not receive radiation therapy or their siblings (12-17). Several studies have also been conducted using estimated radiation dose to the pancreas (18-20). A more recent analysis from the Childhood Cancer Survivor Study in the United States suggested a linear dose-response relationship between radiation to the pancreatic tail and risk of diabetes (18). A retrospective cohort study of childhood cancer survivors in France and the United Kingdom showed that the risk of diabetes increased by increasing radiation dose to the tail of pancreas up to doses of 20 to 29 Gy and then plateaued (19). Another cohort of Hodgkin lymphoma survivors who were diagnosed before the age of 51 years also showed increased risk of diabetes by increasing radiation dose to the tail of pancreas without a plateau, although the increased risk was only significant among those treated with ≥36 Gy (20). The present study suggested a linear relationship between radiation dose and incidence of diabetes, although the radiation effect in the relatively low dose range remains unclear.

The radiation dose-response relationship was modified by age ATB. Participants who were exposed in childhood were at higher risk of diabetes incidence than those exposed in adulthood. Past studies of A-bomb survivors have also shown that risks of cancer and some noncancer diseases, such as thyroid benign nodule and hyperparathyroidism, were higher among participants exposed in childhood than those in adulthood (29-31). Consistent with the present study, studies of childhood cancer survivors have shown that survivors who were diagnosed or treated at a younger age had greater risk of diabetes (12,18,19).

The present study also observed a difference in the radiation dose effect by city. So far, no clear differences in the dose response between Hiroshima and Nagasaki have been observed for various noncancer diseases such as hypertension and thyroid benign nodule among A-bomb survivors (5,29). It is notable that follow-up time in this study was predominated by Hiroshima and those who were relatively young ATB, and thus we may have had reduced power to assess radiation effects among Nagasaki and elderly ATB participants. We should also consider other potential factors such as geographic differences in background associated with diabetes in each city. Compared to Hiroshima survivors, Nagasaki survivors had smaller radiation-related HRs and wider CIs in all age ATB categories. In particular, the estimated HR of the eldest age ATB group (≥30 years old) in Nagasaki was less than 1 and substantially different compared to Hiroshima and the other age ATB groups in Nagasaki (<30 years old). Unlike Hiroshima, the hypocenter of Nagasaki was away from the urban area (26). Therefore, the high-radiation-dose group in Nagasaki resided in a relatively rural area along with a possibly more rural lifestyle. A longitudinal analysis of 1958 to 1986 data of total cholesterol levels showed that mean total cholesterol levels among elderly age ATB subjects from Nagasaki were lower in the exposed group than those in the group estimated to have a 0 dose (32), suggesting a difference in dietary habit between groups. The low HR observed in the Nagasaki elderly group may also reflect such difference in dietary habits by exposure area.

In the 1958 to 1978 study of AHS participants by Brodsky et al (8), the prevalence of diabetes was clearly higher in Hiroshima than in Nagasaki. In the present study, the number of prevalent cases of diabetes prior to the start of follow-up was somewhat higher in Hiroshima than Nagasaki, although there was almost no difference over the course of the study period. Therefore, differences in socioeconomic factors between proximal and distal exposures and lower prevalence of diabetes during the early study period might be related to the observed difference in the eldest age ATB group in Nagasaki. Although we attempted to account for these background differences with crude adjustment for location, residual confounding may remain.

The results were not different after including family history of diabetes;, however, the frequency of family history is likely underestimated because, unfortunately, information about family history of diabetes is limited in this cohort. Almost all survivors with relatively high doses such as proximally exposed survivors with acute radiation symptoms were selected as cohort members. Therefore, it is possible that multiple members of the same family were included more frequently in the high-dose group, and that multiple cases of diabetes were detected within a family. Given the limitations of the information we have available for exploring this issue, it is possible that the higher radiation-associated HRs in Hiroshima compared to Nagasaki could be partially attributable to the inclusion of relatives with shared genetic or environmental risk factors for diabetes.

Potential mechanisms for radiation dose effects on risk of diabetes have been considered. High-dose radiation to the pancreas tail induced damage to the pancreatic β cells or to microvascular tissues around the pancreas and led to reduced insulin secretion later (17,20,33,34). However, estimated doses of A-bomb radiation are lower than therapeutic doses for radiation therapy and islet cells are considered more resistant to radiation than acinar cells of the pancreas, the basic functional unit of exocrine pancreas (34). Insulin resistance has also been suggested as a mechanism underlying the development of diabetes. Hormonal deficiencies such as growth hormone and damage to the normal subcutaneous depot of adipose tissue by radiation exposure may result in visceral fat accumulation and thereby lead to insulin resistance (14-16,35). In our cohort, a dose response for decreased standing height and body weight was reported for younger A-bomb survivors (36), and it is possible that factors such as poor nutrition due to radiation-induced gastrointestinal damage and socioeconomic status caused by the A-bomb were responsible. Indirect effects of A-bomb radiation such as changes to the gut microbiota (17,37) and famine exposure in early life (38,39) might be other possibilities.

There are some limitations in this study. Routine blood glucose level testing was started in July 1986 for all AHS participants. Before that, the initial screening test for diabetes was urine glucose. Therefore, diabetes may have been overlooked in urine sugar-negative cases. However, we confirmed that the results did not change when shifting the start of observation to the first AHS clinical visit after July 1986. In the present study, we did not exclude diabetes cases related to specific causes such as liver cirrhosis because the International Classification of Diseases diagnosis codes did not differentiate between liver cirrhosis and other liver diseases such as fatty liver disease until 1986. Past studies have suggested that A-bomb radiation increased mortality from liver cirrhosis (40,41); however, we do not anticipate a large impact on our results because the frequency of liver cirrhosis is relatively low. In the present study, we did not adjust for body weight or body mass index (BMI) because there was evidence that A-bomb radiation was associated with growth reduction and decreased BMI levels among AHS participants (36,42). However, past studies of childhood cancer survivors have indicated that increased risk of diabetes caused by radiation exposure persisted after adjustment for BMI levels (12,14,15,19).

The present study indicated that exposure to radiation might be associated with an increased risk of diabetes, especially among those exposed at younger ages. However, the results were inconsistent between cities for reasons that are unclear. Although we considered some potential confounders in the analyses, residual confounding by other unmeasured factors such as socioeconomic status may remain. The E-value of 1.59 (lower limit 1.34) indicates that a residual confounder with associations of that magnitude with both radiation dose and diabetes incidence, above and beyond the measured confounders, could explain away the observed association. We need to conduct further studies to elucidate the mechanisms behind radiation exposure’s involvement and other insulin resistance related conditions.