More PACE trial data released

This is all completely beyond me but i wanted to say thank you & well done for all the hard work involved in obtaining it & sorting through it.

 

CSRI summary(ish):

Other info:

Ref: Beecham J K & Knapp MRJ: Costing mental health interventions. London: Gaskell; 2001.

Measured at Baseline visit 2, 24 weeks (end of therapy), 52 weeks (trial end)

 

Reading through the questionnaires in the Protocol again is putting me in a Very Bad Mood. So apols if I'm being a bit stroppy elsewhere on here.

It's shameful that they collected so much information on participants, and gave no thought as to how they were going to use it. Combining such complex info into simple metrics makes much of the data uninterpretable. Shocking use of participants' time and effort. Makes me so cross.

 

That's the way the money goes. Pop! goes the weasel.

 

Most of the "power" of the trial came from its cost. It could have been done as is for barely 1/10 the price tag, but that would have been much less impressive. It created a sunk cost that motivated people to justify the huge expense, regardless of the fact that it wholly lacked substance or anything meaningful.

So it seems expected that they'd have wasted much of it on things they didn't even have a use for. It was busywork, an exercise in confirmation that was going to show "success" no matter what. Same for the alleged expensive training. It's a bullshit treatment with a fictitious narrative model, what training could it have even involved? There's no specific expertise required, no novel technology or anything even justifying doing any training beyond a basic information session.

The whole point was to spend money so decisions could be justified as "we spent a lot of money confirming that so you're going to use it".

 

Haha. Yeah. Totally. Implausible deniability, what a hoot.

 

I don't want to put a dampener on this, but I have a nasty feeling about these assumptions... Can anyone confirm? I'm unable to replicate the secondary outcome summary stats.

 

Here's the summary data from White et al. 2011 Table 6 - Secondary Outcomes (truncated to just include relevant variables):



I've also attached a text version (includes Jenkins, just because...).

 

How off were you? If they were adjusting for variables we don't have (age, sex, etc) couldn't there be very minor differences without it being much of a problem?

 

The summary data (means, SDs) in the table are unadjusted. It's only the comparisons that are adjusted.
The data should match exactly.

 

There were no common variables.

All I was able to do was merge the columns, on the assumption that the participant order was the same in each case. I added a fictitious participant ID number, so that if anyone sorts the file for some reason they will be able to unsort it again, but that's it. We either have to trust that the records are in the same order for every file, or go back to the researchers and ask them to provide some way to be sure about this.

 

Could we look at something like missing data to check if they line up, for example if all of the 52 week scores are missing from the first set there should be a reasonably high likelihood that the 52 week scores would also be missing from the second set. Giving it a quick glance over that doesn't seem to be the case.

 

Without confirmation, it is not safe to trust that the records are in the same order for every file. Without knowing which groups the data were in, they are pretty much useless, apart from providing a summary of the entire cohort.

 

I have found some time to start building some code this evening. I reproduced some of the values from Table 3, where we have fatigue and physical functioning scores at baseline and 52 weeks. (Table 3 has these two side by side, but I don't have room on the page, so they are one above the other here).

The moderately encouraging news is that all of the means, SDs, and Ns seem to match the table in the article. That suggests that we have the correct treatment arm for each participant, at least for the four variables in question (but I think that these arrived on a single spreadsheet, so that would be expected).

The less encouraging news is that the mean difference values don't match the published table. The differences are small for the difference between the "active" treatment and SMC, but quite a lot larger for the difference between the "active" treatment and APT.

Now of course this could be because of how I calculated the mean differences, but the only way I can see to do that is to take the difference in the means. If it was just the CI where the discrepancy was occurring then I would assume I was using a different method from the one used by the authors to calculate the standard error, but for the means I don't know.

Edit: Does this board have a "code" mode, to preserve the spacing of a piece of text? I had this table nicely formatted, but then all the spaces got eaten.

Fatigue
TX APT CBT GET SMC
Baseline 28.5 (4.0) n=159 27.7 (3.7) n=161 28.2 (3.8) n=160 28.3 (3.6) n=160
52 weeks 23.1 (7.3) n=153 20.3 (8.0) n=148 20.6 (7.5) n=154 23.8 (6.6) n=152
MdiffSMC -0.7 (-2.2,0.9) -3.6 (-5.2,-1.9) -3.3 (-4.8,-1.7)
MdiffAPT -2.9 (-4.6,-1.2) -2.6 (-4.2,-0.9)

Phys.function
TX APT CBT GET SMC
Baseline 37.2 (16.9) n=159 39.0 (15.3) n=161 36.7 (15.4) n=160 39.2 (15.4) n=160
52 weeks 45.9 (24.9) n=153 58.2 (24.1) n=148 57.7 (26.5) n=154 50.8 (24.7) n=152
MdiffSMC -4.9 (-10.4,0.7) 7.4 (1.8,12.9) 6.9 (1.2,12.7)
MdiffAPT 12.2 (6.7,17.8) 11.8 (6.0,17.5)