|Home | About | Journals | Submit | Contact Us | Français|
We are grateful to Campbell and Freeman for raising issues relating to the presentation of data in our paper.1 Interestingly, their comments do not change our findings or their interpretation. We would like to respond to the points they raise in turn.
We acknowledge that survival data are positively skewed and therefore reporting a mean survival time is not always the most helpful statistic. We do not necessarily agree that this is best called ‘mean follow-up time’ as they suggest, but feel that a median survival time may do more justice to the data.
We agree that the four figures showing survival have different starting points for the y-axis, which can cause confusion. However the axes were clearly labelled and should therefore be easy to interpret. It is a question of style for a particular journal as to whether this is the norm or not. We originally chose to start the axes at different points in order to demonstrate the data as clearly as possible and because we did not directly compare differences between the four cancers. We are not convinced that there is consensus within the statistical community that cumulative mortality curves are better as they suggest.
We are grateful for their diligence in spotting the absence of open boxes on the figures. These appear to have been lost in final production of the paper, but their absence does not detract from the main messages from these data.
Again we are grateful for their suggestion of including the number of patients along the x-axis, and agree that in some circumstances this can add clarity to survival curves. However we do not believe that it has become routine practice. A quick look through recent similar papers has confirmed these beliefs. Perhaps it is time for journals to lay down explicit guidelines about the presentation of such data?
Lastly, they raise the issue of when a non-significant result is indicative of no difference. Certainly it is possible to calculate a hazard ratio and a confidence interval, but this should not detract from the more important question of when a statistical difference equates to a clinically meaningful difference.