Diabetes and breast cancer in Taiwanese women: a detection bias?

Purpose To evaluate whether diabetes is a risk factor for breast cancer considering confounders and potential detection examinations. Methods National Health Insurance data on 501 747 women without breast cancer were retrieved. Three-year cumulative incidence (2003-2005) and risk ratios (RRs) between diabetic and nondiabetic women were calculated. Potential detection examinations were compared between diabetic and nondiabetic women by chi-square test. Odds ratios (ORs) were estimated by logistic regression for diabetes status/duration with and without adjustment for potential detection examinations and confounders. Results The crude RR (95% confidence interval [CI]) for all ages, and age groups < 50, 50-64 and >= 65 years, was 2.62 (2.31-2.91), 2.69 (2.11-3.44), 1.39 (1.15-1.68) and 1.37 (1.03-1.84), respectively. Patients with diabetes more frequently received potential detection examinations than nondiabetes (17.5% vs. 7.4%, P-value < 0.001). The unadjusted OR (95% CI) for breast cancer for diabetes status (yes vs. no) was 2.63 (2.31-2.98) and was significant for any diabetes duration. The OR for diabetes status was 1.81 (95% CI: 1.59-2.06) after adjustment for potential detection examinations. In models adjusted for potential detection examinations, age, living region, occupation, comorbidities and used medications, OR for diabetes status attenuated to 1.13 (95% CI 0.96-1.32, P-value =0.14) and none was significant for any diabetes duration. Potential detection examinations were associated with a fivefold to sevenfold higher risk in various models, indicating a strong impact of detection bias. Conclusions An association between diabetes and breast cancer is observed, but this can be due to potential detection bias and confounders.