*2.3. Statistical Analysis*

Characteristics of men at the time of cancer diagnosis were described and compared among cancer treatment groups using chi-square test. Competing risk analyses were conducted using cause-specific hazard models, with deaths from all other causes besides CVD considered as competing risk events. Because the incidence of MBC and CVD mortality rates increase with advancing age, attained age in years was used as the time scale for all time-to-event analyses. Thus, estimates from such model are age-adjusted [22]. The validity of the proportional hazards assumption was tested and confirmed using weighted Schoenfeld residuals as well as using formal statistical test of non-proportionality.

Covariate selection for multivariable models was based on variables that were significant in bivariable analyses at an alpha of 0.2. Variables evaluated in bivariate models were year of diagnosis, race and ethnicity, geographic region, location, marital status, income, disease stage, estrogen and progesterone receptor status, tumor grade, tumor size, laterality, and number of regional lymph nodes examined. Because the impact of radiation on overall survival among MBC patients is not the same between breast conservation surgery and mastectomy [23], and radiation to the left side of the breast is associated with a higher risk of CVD than on the right breast [24], additional analyses were performed to evaluate the role of type of surgery and tumor laterality on the association of radiotherapy with CVD mortality among men who did not receive chemotherapy.

Finally, interaction between race and ethnicity with cancer therapy was tested. The SEER\*Stat version 8.4.0.1 software (Information Management Systems, Rockville, MD, USA) and the SAS software version 9.4 (SAS Institute, Inc., Cary, NC, USA) were used to conduct the statistical analyses with statistical significance determined with a two-tailed test *p* value of less than 0.05.
