What can we learn from the ZOOM trial? - Authors' reply.

We thank Stefanie Hayoz and colleagues for their thoughts on our ZOOM trial.1 We acknowledge that the predefined non-inferiority margin was affected by some degree of subjectivity, which is a weakness of our study. However, this is the case with non-inferiority trials in general. Furthermore, at the time the study was designed, no formal guidelines for non-inferiority margins existed. When mature study data became available, the study team realised that the original limit was no longer appropriate and the margin was therefore adjusted (a formal interim analysis was not done) because the estimated variance seemed to be much lower than our initial assumptions. The new margin proposed by Hayoz and colleagues (0·15) is very close to the one used to this study (0·19). We are aware that if the alternative limit proposed by Hayoz and colleagues had been used, the study would not have provided formal proof of non-inferiority. However, our study showed that the maximum possible difference between the two treatments in terms of skeletal morbidity rate was 0·17 events per year. Whether this difference is clinically relevant or not is questionable (as it would be if the original limit had been used), but even if the margin suggested by Hayoz and colleagues were used, the difference would be very close to the non-inferiority margin. Hayoz and colleagues question the use of the skeletal morbidity rate as our primary endpoint. As in Rosen and colleagues' study,2 events that occurred in 21-day event windows in the ZOOM study counted as one event. Additionally, an analysis done with a negative-binomial model to protect against the assumptions of the ANCOVA model not being met confirmed that the difference between the two groups was very small. The upper limit of the 95% CI indicates that an increase of 57% cannot be excluded. In relative terms, this figure might seem to be a large difference, but in absolute terms, given a reference rate of 0·22 events per year, this translates into an increase of about 0·13 events per year, which is within the non-inferiority margin. A per-protocol analysis was done, which excluded 35 patients (17 patients in the 12-week group vs 18 in the 4-week group). In the per-protocol population, the overall skeletal morbidity rate was 0·24 (95% CI 0·15–0·34) in the 12-week group and 0·20 (0·13–0·27) in the 4-week group. The between-group difference was 0·04 and the upper limit of the one-tailed 97·5% CI was 0·16. Thus, intention-to- treat and per-protocol analyses led to the same conclusions.