Send letter to journal:
Re: No discrepancy between protocol and trial report

We thank Professor Thornton for raising these issues and are grateful for the opportunity to clarify details of the trial protocol and analysis.

There is, in fact, no discrepancy between the description of outcomes in the trial protocol published on the trial web site, the protocol that was registered on the Australian New Zealand Clinical Trials Registry, and the report published in the British Journal of Sports Medicine. All three sources describe the same primary outcomes (risk of soreness and time to injury) and the same secondary outcomes (severity of soreness, time to ("preventable") muscle/ligament/tendon injuries, time to injuries for which professional care was sought, and perceptions of looseness during and after exercise). In addition, all three sources describe the same secondary analyses of the interactions between frequency of activity, age and strength of belief in affects of stretching and each of the two primary outcomes. Confusion may have arisen because in the trial report we refer to the analysis of the interactions as "outcomes", although the analyses of interactions were conducted on the same primary and secondary outcomes as listed above. And we may have added to the confusion by inconsistently referring to the perceptions of looseness during and after exercise as either one or two outcomes. We acknowledge that the wording may not have been clear but we reiterate that there was no inconsistency between the analyses described in the protocol, register and trial report.

The protocol, register and trial report describe an analysis of the subset of injuries which could plausibly be prevented by stretching. The protocol (which is more detailed than the registry entry) indicated that the classification of injuries into those that could and could not plausibly be prevented by stretching would be classified by an independent expert. We found that the data lacked sufficient detail to inform decisions about whether individual injuries were preventable so we decided simply to classify all muscle, ligament and tendon injuries as potentially preventable. This decision was made before the allocation code was broken without reference to the data. We did not know, at the time that decision was made, that there would be a significant effect of stretching on the subset of muscle, ligament and tendon injuries.

As the protocol indicated, no adjustment was made for multiple comparisons. We interpret frequentist analyses as Bayesian analyses with vague priors[1] and, from this perspective, the decision not to make adjustments for multiple comparisons is justified. At any rate, there were only two primary outcomes so adjustment for multiple comparisons would not have changed the conclusions from the primary analysis. We agree that the finding of an effect on the secondary outcome of muscle, ligament and tendon injuries is less robust than the finding of an effect on the primary outcome risk of soreness because muscle, ligament and tendon injuries were a secondary outcome. That is why we concluded that stretching "probably reduces the risk of some injuries and does reduce the risk of bothersome soreness".

Reference

1. Barnett V (1973). Comparative Statistical Inference. London: Wiley.

Send letter to journal:
Re: Discrepancies between protocol and trial report

Dear Editor

It is stated that “two primary outcomes and 12 secondary outcomes were specified a priori in the analysis plan” but only the two primary outcomes and five secondary outcomes appear in the paper. Four secondary endpoints (three reported in the paper) are listed in the trial protocol registered with the Australian and New Zealand Clinical Trial registry, accessible here http://www.anzctr.org.au/trial_view.aspx?ID=82520. Five secondary endpoints (three reported in the paper) are listed in the protocol available on the trial website, accessible here http://www.kunnskapssenteret.no/The%20Stretching%20Trial. In that protocol the second of the two primary endpoints (time to injury) is subdivided in a number of ways that, with the exception of whether the participant sought help from a professional, differ from the analyses reported.

The reported secondary analysis of “time to injuries to muscles, ligaments and tendons” differs from the pre-specified secondary outcome in both protocols of “time to injuries that might be considered could be preventable by stretching”? In the website protocol it is stated that this “preventability” classification would be done without knowledge of the trial group. Can the authors clarify whether this was done?

Can they also clarify which outcomes were pre-specified in the analysis plan before the trial allocation code was broken, can they report all these outcomes, and state what adjustment was made for multiple tests of statistical significance?

These questions may sound pedantic, but the primary outcome for injury was negative. The apparent effect on muscle ligament and tendon injuries was of only modest statistical significance (P=0.03), and might disappear if adjustment was made for multiple significance testing. The apparent effect on the “bothersome soreness” is unavoidably susceptible to reporting bias in an open trial. The higher rate of dropout from follow-up at all time points in the experimental group, which appears unlikely to be due to chance, might bias the results in either direction.

As a participant, I was impressed with the trial organisation and design. Without reassurance that analysis and reporting were of an equally high standard, I remain fearful that the authors’ conclusion that “stretching … probably reduces the risk of some injuries and does reduce the risk of bothersome soreness”, is too strong.

Jim Thornton Nottingham