Aernoud Fiolet

142 Chapter 5 with an end point, time‑to‑event was calculated as the difference between the date of occurrence of the end point and the date of randomization plus 1. In participants without an end point, time to censoring was calculated as the difference between the end of trial date and the date of randomization plus 1. The frequencies of end point occurrences were expressed using percentages and incidence rate per 100 patient-years. For the primary end point, the null hypothesis of no treatment benefit of colchicine (vis-à-vis placebo) versus the alternative hypothesis of treatment benefit was tested using a 2‑sided log-rank test at an alpha level of 0.05. The corresponding 2‑sided P value (for superiority) was calculated as two times the 1‑sided P value for superiority of colchicine vis-à-vis placebo. The testing of the secondary end points was assessed in a closed testing procedure to preserve the alpha level at 0.05. If statistical significance for the primary end point was obtained, the secondary end points were tested in a hierarchical fashion according to the ranking that was pre-specified in the protocol and in the Statistical Analysis Plan. If the hierarchical testing procedure allowed formal hypothesis testing of a secondary end point, the 2‑sided P value for superiority was calculated with the 2‑sided log-rank test, as for the primary end point. No formal hypothesis testing was performed outside the predefined hierarchy of primary and secondary end points. The additional time‑to‑event end points were analysed by a Cox proportional- hazards model in a similar fashion to those for the primary and secondary end points, without formal hypothesis testing. For new onset diabetes the cumulative incidence ratio and the corresponding 95%‑confidence interval was reported, since time‑to‑event was not available for all participants. The proportionality assumption was assessed by visually comparing the plot of the log of cumulative hazard between treatments and by additionally adding a treatment by logarithm-transformed time interaction into the Cox model. There was no violation of the proportionality assumption for the primary, secondary and additional time-to-event end points. An on-treatment analysis of the primary end pointwas performed as an exploratory sensitivity analysis. Premature permanent discontinuation of trial medication was determined to have occurred if the trial medication had been permanently discontinued more than 30 days prior to the occurrence of the primary end point,