Dr. Ridker responded to these concerns on the NEJM website: 

If the "protective" effect on diabetes incidence reported in WOSCOPS is treated as hypothesis-generating, then a summary of published hypothesis-testing trials demonstrates that all statins modestly increase the risk of diabetes, with no heterogeneity according to potency. In our study, many of the patients in whom diabetes developed were obese or had an impaired fasting glucose level, groups in which large reductions in vascular events were associated with rosuvastatin.

The independent data and safety monitoring board for our trial followed rigorous principles³ in its prespecification that early termination of the study because of an observed benefit would require proof beyond a reasonable doubt. Members of the board were experienced in monitoring publicly and privately funded trials and viewed the trial's prespecified statistical boundary as only one component required for proof. Although the formal statistical boundary was conservative and evaluated only after accrual of ample data, the board elected to continue the trial for an additional 6 months after the boundary was crossed. Data that were accrued thereafter independently confirmed both the magnitude and statistical significance of the apparent benefit. We thus respectfully disagree with Pierard and Davis. The board appropriately protected the interests of society and the trial participants and provided a valid estimate of the treatment effect.⁴

The evaluation by Koller et al. ignores the significant reduction in death from any cause that we observed. If death from any cause is added to our primary composite outcome (a standard approach to account for competing risks), then the absolute risk difference increases and the number needed to treat declines.