The statistical analysis plan is available in the Supplementary Appendix. Details regarding the sample-size estimates and statistical analyses have been published previously.10 (link) We estimated that the follow-up of 7500 patients for approximately 5 years with an assumed event rate of 2.1 per 100 patient-years of exposure would produce 633 events and hence a power of 91% to rule on the null hypothesis. A Cox proportional-hazards regression model was used to analyze the intention-to-treat population for the primary composite outcome to test for the noninferiority of degludec as compared with glargine. Noninferiority would be confirmed if the upper boundary of the 95% confidence interval was less than 1.3. If noninferiority was established, we then tested for superiority with respect to severe hypoglycemic episodes using a negative binomial-regression model that was adjusted for observation time and treatment group to test for the number of events and a logistic-regression model that was adjusted for treatment group to test for incidence. Superiority of these secondary outcomes would be confirmed if the upper boundary of the 95% confidence interval was less than 1.0. Selected sensitivity analyses, including the per-protocol analysis, were performed to address the robustness of the results. The rationale for the use of a noninferiority threshold of 1.3 in the primary analysis and a threshold of 1.8 in the interim analysis is described in the Supplementary Appendix.