Now that I've read the full text:
Background
This study uses data from the 2010 FINE Trial which presumed that CFS is perpetuated by deconditioning and secondary biological changes and can be reversed by gradually increasing activity levels - Pragmatic Rehabilitation (PR). It found that the therapy, compared with GP treatment as usual, made no significant difference to either fatigue of function at the primary endpoint of 70 weeks:
there were no statistically significant differences in fatigue or physical functioning between patients allocated to pragmatic rehabilitation and those on treatment as usual
However, in a
Rapid Response to their own paper, the authors reanalysed the Fatigue results using a different scoring method (Likert, as opposed to bimodal), concluding:
With the fatigue scale re-scored 0123, we are able to demonstrate a clinically modest, but statistically significant effect of PR compared with GPTAU at both outcome points. Given the chronicity of CFS/ME in our sample, we believe that this on average small improvement in fatigue is important to these individuals.
[Effect estimates [95% confidence intervals] for 70 week comparisons are: PR versus GPTAU -2.55 [-4.99,-0.11], SE 1.24, P=0.040]
Regardless of the debate over how much weight should be given to this post-hoc analysis, this means that they are looking at what baseline variables predict or moderate a rather small effect.
Note that a
Predictor is a factor that predicts outcome (in this case a change in fatigue score) across
all treatment groups, while a
Moderator (the focus of this paper) is a factor that is only relevant in
one (or some) types of treatment.
Moderators: Depression? (probably not)
The main conclusion of the paper was that, for Pragmatic Rehabilitation, patients with higher depression scores were likely to show (even) smaller gains than those with less depression. There is, though, one rather big caveat: the authors didn't correct for the large number of comparisons they made, increasing the chance that the finding is a false positive. As they put it:
Our findings need to be interpreted with caution: we investigated the potential moderating effects of over 20 baseline variables, although most of these were prespecified in our protocol,16 and our two positive findings may be type 1 errors [false positives].
Further research to replicate the finding that level of depressive symptoms moderates the effect of pragmatic rehabilitation treatment on fatigue is needed before we can have confidence in the result.
Note they found the Depression score was a signifiant moderator with p=0.022, which is within a 'standard' confidence level of p=0.05. With so many comparisons it would have been appropriate to correct for multiple comparisons with a much lower p value; there are arguments over the best method to use but the most common one,
Bonferroni, would have required a p value of 0.0025,
9x lower than achieved. The Bonferroni correction is very conservative, but other methods available would also almost certainly give a p value requirement much smaller than the 0.022 acheived. To add doubt to the findings, a depression
diagnosis (yes/no), as opposed to Depression
score (HADS), was
not a moderator.
Predictors. Myths busted? Membership of self-help group or receipt of benefits don't predict outcome
The study specifically looked at the possibility that belonging to a ME self-help group, or receiving benefits could moderate treatment. Others have argued that these factors could even help prevent recovery (@Dolphin, can you help here?). Instead the study found that they were neither predictors nor moderators: being on benefits or in a self-help group does not affect the prospects for improvement. However, it's probably wise not to read too much into this: as the overall improvement in fatigue was small, it would be hard to detect any kind of effect.
Number of co-morbid illnesses at baseline also didn't predict fatigue outcome.
Younger age and shorter illness improve prospects
Across both the PR and treatment as usual group both age and illness duration were predictors of fatigue outcome. The regression coefficients (Table 5) were given as:
Age = -0.1
Illness duration= -0.01
I don't really know anything about how to interpret the size of these effects, and haven't been able to reach the authors yet for clarification, but those effects look rather small to me.
Washing up: the authors did find other significant effects, but for tiny groups of patients so its impossible to generalise from these findings. They also reported moderator effect trends (i.e. p values just above p=0.05) but given the problem with multiple comparisons I haven't included those here.
