We have wandered a considerable distance from the question at the start: did PACE demonstrate safety of GET as claimed?
The dispute over subjective versus objective measures is actually important here.
We are dealing with two primary kinds of subjectivity: patient self-reports and doctor/researcher assessments. A considerable effort is made in material distributed for doctors associated with the study to discourage patients from making negative self-reports. This ought to bring those patient reports into question as any kind of objective evidence. Studies
on bias introduced by peers or authority figures illustrate just how badly this can distort perceptions and reports.
Doctor/researcher assessments are another problem. We have a number of mere assertions that patients either did not experience harms, or that any setbacks they might have suffered were not caused by treatment. To call these opinions much in the way of scientific evidence is ludicrous.
We are forced to rely on researcher assertions in many aspects of this research simply because data have not been released, assuming it was even collected in the first place. We do, however, have some interesting assertions to work with.
One which was not in the formal publication, as far as I recall, was the distinction between "responders" and "nonresponders". It scarcely matters whether the actual numbers were 60/40 or 65/35, or some similar split.
What this reveals, in a way outside evaluators cannot check, is that a substantial number of patients went through a year of treatment without receiving any benefit -- in the judgment of the researchers themselves.
There was an effort to remove patients who were clinically depressed from the cohort in advance, but I see no way to estimate the number of participants who might have become clinically depressed during the trial. This is important because clinical depression, especially with suicidal ideation, is a life-threatening medical problem.
Now, I have to ask, is it reasonable to expect people who experience the collapse of careers, finances and relationships as a result of illness which wrecks both mental and physical performance to spend a year undergoing ineffective treatment without becoming depressed? Are we recommending a therapy only for saints and epic heroes?
Once this pattern of long-term illness is established, with or without depression, it is all too easy for practitioners to adopt the implicit attitude that they are dealing with "inferior people". No evidence of performance or accomplishment prior to onset is likely to change this perception. This is a serious problem in treating any chronic disease, and doubly important when causes are obscure. This has a strong impact on patient self respect.
Now, back to the question of possible patient deterioration. No group mean is going to tell you that all patients studied showed improving scores unless you have some other assurance that all results after therapy were better than those before. When group means scarcely move it is more likely than not that close to half the patients moved up and half down.
So, the assurance that the treatment is safe comes down to the assertion that any deterioration experienced by some number of unidentified patients was not clinically significant. This causes me to revise the question above. The researchers are not terribly bothered by a substantial percentage of patients going a year without perceived benefits, and are even willing to assume that some sizable number undergo a decline over the period without counting this as harm. Furthermore, the researchers are not concerned about possible additional effort patients must put into participating in therapy. What happens to those who don't have this reserve of activity?
(Actually, I don't have to ask. This is where locked wards and feeding tubes appear in a number of accounts. I can point to cases in the U.K. news where the prospect of such a transfer precipitated suicide.)
If we knew what effort patients were making to participate, and how that compared to previous capacity, we might have a better idea of how serious a burden treatment imposes on patients. We simply don't know, based on PACE, though we might have if the proposed measures of activity had been used. Nor do we have a clue about the size of possible declines in performance the researchers assure us are not significant.
If you are dealing with a range of psychiatric problems (recall that the researchers tend to class this with psychiatric problems) the period of days or weeks before treatment becomes effective is especially important. I've already mentioned the week patients on Prozac may expect to suffer in the best cases, and the danger of suicide there. What we see in this study seems to me like an absence of measures at this time granularity. Not just suicide, but even pure accidents are regularly associated with cognitive or physical deterioration on this time scale. Did such deterioration take place? We don't know, and it is quite possible the researchers don't either.
We have assertions that CFS patients, as defined in PACE, don't suffer from PEM, but no measures which might show these claims were wrong. We have anecdotal reports of longer periods of relapse in similar patients, which are simply absent here.
Were there any measures which I am missing?
My point is that if you don't test in a way which could expose harms you cannot make any assurances about absence of harms. You must simply assume the light goes off when the refrigerator door is closed.
See the logical fallacy?