These results disagree with the publication from
These results disagree with the 1997 publication from the same cohort reporting a higher breast incidence among women with diabetes , but the analysis was based on few cases, a short follow-up time, and was not adjusted for BMI. The presence of diabetes predicted breast cancer mortality (HR 1.72, 95%CI 1.34–2.19) within the Asia Cancer Consortium  and breast cancer incidence (HR 1.71, 95%CI 1.15–2.54) among Asian Americans in California . In contrast, the results agree with null findings in studies conducted in Japan [, , ] and among Japanese Americans . Among six Japanese cohorts (182,542 women and 1,380 breast cancer patients) , the risk estimate for breast cancer associated with diabetes was 0.98 (95%CI 0.69–1.38) despite a significantly elevated risk for all cancers (HR 1.19, 95%CI 1.12–1.25). Similarly, no association was seen among 28,012 Japanese American women in the MEC, many of whom are descendants from migrants around Hiroshima [35,36], whereas obesity predicted breast cancer incidence to a similar degree as in the current study . Studies from China [37,38], Korea , Taiwan , Thailand , and Pakistan  reported significantly elevated risks of developing breast cancer, ranging from 1.5 to 8.4-fold, in women with a history of diabetes than in those without diabetes, but many studies did not include BMI [, , , , ] or had a small sample size . The findings among Asian women disagree with meta-analyses among primarily white women reporting a 20% higher breast cancer incidence in women with diabetes [9,13] and relative risks of 1.16 (95%CI 1.08–1.24) for BMI-adjusted studies and 1.33 (95%CI 1.18–1.51) in studies without BMI . The current findings related to BMI agree with previous reports of an elevated breast cancer incidence associated with overweight/obesity among Asian women despite having lower mean BMI levels than white women [1,16]. It has been proposed that higher proportions of visceral fat in Asians than whites are partly responsible for this Z-Guggulsterone observation . A worldwide report recently estimated that having a high BMI was responsible for a larger proportion of the breast cancer burden (6.9%) than for diabetes (2.2%) with a combined attributable risk of 8.9% . Strengths of the current analysis include the long-term follow-up of the cohort, the complete follow-up information, the repeated questionnaires, and the availability of many covariates. Changes in lifestyle risk factors and breast cancer risk we addressed by including birth cohort into the models. However, the assessment of BMI and diabetes status had a number of problems. Information was collected at different ages from the participants and varying frequency (1–3 times). New diabetes cases after 1991, the last survey date, were not captured resulting in possible under-ascertainment. Self-reports without information on glucose status may have underestimated the true prevalence of diabetes although the validity of self-reports on diabetes has been shown to be fairly high . The term “diabetes” in the questionnaires does not allow differentiation between the wide spectrum of clinical conditions, including type 1 and 2 diabetes. However, the number of participants with type 1 diabetes would be extremely small as insulin was not available during the childhood of the first two birth cohorts, i.e., 1871–1925, and survival for children with type 1 diabetes born during 1926-45 would remain low. Misclassification of diabetes status was even more likely for the 7,146 NIC cohort members who did not complete surveys after 1969. Yet, the findings in models using only the first data collection as compared to the time-varying models were very similar. With changing diabetes criteria over time and increased screening efforts leading to rising diabetes incidence , undiagnosed diabetes cases likely introduced misclassification bias. However, as part of a clinical study within the AHS, diabetes prevalence was similar to those derived from the surveys (1978: 5.9% and 1991: 7.6%). In addition, the fact that some participants may have moved out of the catchment area of the cancer registries might have introduced misclassification bias as the relatively large migration rates especially in younger generations may have been associated with diabetes status [46,47]. Information on hormone treatment, was not available, but the low use of in Japan makes it unlikely that major bias was introduced. Finally, we were not able to consider the impact of treatment for diabetes although the study participants (certified atomic bomb survivors) were eligible for free health check-ups and medical care.