Hi Bob. If you are rounding up for the NNT of CBT and GET for a clinical response in both fatigue and physical function, it should be 7.14 (or 8) for CBT and 6.25 (or 7) for GET. You are correct that the average for both is 7 when rounded up (even when accounting for slightly different total group numbers between CBT and GET).
Thanks for that, biophile. T'was an oversight. Have corrected.
Also, note that the NNT for a "positive change" in CGI is also 6.25 (or 7).
I'm not sure what you mean by "CGI"?
The primary outcomes for fatigue and physical function were changed, from dichotomous scores ie meeting 50% relative improvement or meeting an absolute threshold score (which were stricter than the post-hoc thresholds of CUD), to continuous scores as shown in Figure 2. Everything else is additional or "secondary". Also, as you would be aware, fatigue scoring on the CFQ was changed from bimodal to Likert scoring, but note that the latter was already a secondary outcome in the 2007 protocol and became primary.
From the wording of the paper, it seems that the Likert scoring method for Chalder is at least
intended to be a primary outcome measure.
It
can be legitimate to change the protocol or design of a study, after the protocol is published.
Whether it is a legitimate change, or not, I think depends on the contractual rules of the MRC.
For an MRC-funded study, there are certain procedures that should be carried out, in order for changes to be legitimate, and I don't know if these were met or not.
The 'improvement rates', based on the CUD, were included in Table 3, under the heading "primary outcomes."
But I agree that it does not state that it is a primary outcome, except where it is included as a measure in Table 3. But it does not state that it is a secondary outcome measure either, whereas it does clarify this for at least one other post-hoc measure.
But, studying it further, I think I agree that the primary outcome measures were just the participant scores for Chalder Fatigue and SF-36 physical function. (And in the protocol, "positive outcomes" were also included as primary measures.)
The post-hoc definition of a CUD was used for the
main analyses in the published paper, such as the 'mean difference from SMC', and the 'improvement rates'.
So it would have been better for me to describe it as the 'main primary outcome analysis'.
I will change the wording in my revision when I get around to revising it.
There are lots of question marks hanging over the methodology used to determine the CUD.
But my opening post is just intended to clearly demonstrate the main published results, as published.
It's not a critique of the methodology.
The authors should still publish the results based on the original goalposts, it would literally only taken up one paragraph.
Yes, agreed. We might need more FOI requests to get the info out of them.
The authors predicted response rates of 10% for SMC, 60% for CBT and 50% for GET. However, their definition of a "response" was watered-down. CBT and GET could only approach those rates of response when the threshold was low, but this inflated the SMC response rate. Conversely, the predicted response rate for SMC was probably accurate when using the original thresholds, but would demonstrate that CBT and GET performed relatively poorly compared to expectations. The authors cannot have it both ways, and the original thresholds would not have allowed news articles to parrot the supposed "60%" response rate for CBT/GET while failing to mention the 45% response rate for SMC.
Good points.