They say:
They used a self-report measure of depression, that looks at things like sadness/loss of joy, negative thoughts about self or future and nonspecific physical complaints like sleep problems, fatigue, poor concentration.
Now see how their statement looks when you substitute the actual thing they did measure whenever you see the word "depression":
Doesn't sound all that astounding now, does it?
I think there is a problem when abstracting from answers to a set of questions and scoring and calling it a scale and assigning a diagnosis as anything with a score greater than a threshold. As people do this they forget to think about the actual questions asked and what they may mean in the context that they are being used in.
The recovery criteria for PACE are another example. If they had tried to map a score of 60 back to possible question answers they would realize they were talking crap. But if you view SF36 as a scale and forget the statistical distrubitions what they've done could look reasonable.
Working in computer security one of the big reasons that software fails and becomes attackable is because people don't understand the lower-level systems they link with an use them in bad ways that open the system up to attacks. If feels like something similar happens in this world where people abstract away details and forget what they actually mean - from that they form unsafe conclusions.