In undertaking further analyses of the FIELD study,1 we recently detected that the reported effects of fenofibrate adjusted for on-study use of other lipid-lowering therapy are not time-dependent as was stated. The aim of the analysis was to account for the higher rate of use of other lipid-lowering therapy (mostly statins) during the study in the placebo group (average 17%) than in the fenofibrate group (average 8%). A programming mis-specification meant that this secondary analysis used drop-in to other lipid-lowering therapy as a fixed, rather than as a time-dependent, covariate in the Cox regression model.

The net effect is that, for patients who started lipid-lowering treatment during the trial without a prior on-study event, all their trial experience from the date of randomisation was grouped separately in the analysis for the comparisons of fenofibrate versus placebo (fixed covariate model). In this sense, subsequent lipid-lowering therapy was modelled like a baseline covariate. By contrast, the time-dependent model groups only the patients' experience after drop-in separately in the analysis.