More PACE trial data released

JohnTheJack said:
I have now received a new data file.

I requested that the data be aligned with previously released data. QMUL say that they are providing me with the information as requested. I cannot verify that to be the case. I believe, however, that these data are aligned with the individual results in the data released after Alem Matthees's successful tribunal hearing.

https://www.dropbox.com/s/ur8h533uio0x9ej/PACE_July2019.xlsx?dl=0
Click to expand...
In future cases, would it be allowed under the FOI rules to include in any new request at least one of the previously released pieces of data, so that an alignment verification check could be run? Or even request all of the original data together with the new, so you would have a fully coherent superset of the original data set?
 
Barry said:
In future cases, would it be allowed under the FOI rules to include in any new request at least one of the previously released pieces of data, so that an alignment verification check could be run? Or even request all of the original data together with the new, so you would have a fully coherent superset of the original data set?
Click to expand...

I'm not sure how that could be worded, but it's worth thinking about.
 
Barry said:
In future cases, would it be allowed under the FOI rules to include in any new request at least one of the previously released pieces of data, so that an alignment verification check could be run? Or even request all of the original data together with the new, so you would have a fully coherent superset of the original data set?
Click to expand...
Not only should it be allowed, it should be mandatory. But in fact it should be mandatory for the authors not to play silly buggers and pretend that they don't have, somewhere, a single file with all the data in it, which is how everyone works, because it's simple and obvious. (People who have never seen scientific data are often surprised by just how simple it looks, unless it's coming from particle detectors at CERN, and even then, the main complexity is the number of rows.)

Of course, it's just about theoretically possible that there is no single-file copy of the data left, but at that point, the entire research team should hand back their PhDs and get jobs in McDonald's.
 
sTeamTraen said:
Not only should it be allowed, it should be mandatory. But in fact it should be mandatory for the authors not to play silly buggers and pretend that they don't have, somewhere, a single file with all the data in it, which is how everyone works, because it's simple and obvious. (People who have never seen scientific data are often surprised by just how simple it looks, unless it's coming from particle detectors at CERN, and even then, the main complexity is the number of rows.)

Of course, it's just about theoretically possible that there is no single-file copy of the data left, but at that point, the entire research team should hand back their PhDs and get jobs in McDonald's.
Click to expand...
visualising the combination of uniform and customer interface skills ! lol
 
sTeamTraen said:
Not only should it be allowed, it should be mandatory. But in fact it should be mandatory for the authors not to play silly buggers and pretend that they don't have, somewhere, a single file with all the data in it, which is how everyone works, because it's simple and obvious. (People who have never seen scientific data are often surprised by just how simple it looks, unless it's coming from particle detectors at CERN, and even then, the main complexity is the number of rows.)

Of course, it's just about theoretically possible that there is no single-file copy of the data left, but at that point, the entire research team should hand back their PhDs and get jobs in McDonald's.
Click to expand...
£5M. Biggest trial of its kind. Trial meeting notes show they planned for it. No excuses.
 
JohnTheJack said:
Thanks to @sTeamTraen who says released data are probably in the correct order & has merged them with #Matthees data.

New data https://www.dropbox.com/s/hqp4drejh6e3lue/PACE_July2019.xlsx?dl=0 …

Merged files https://www.dropbox.com/s/hvwl4dayvnzgqch/PACE-AugMerged.xls?dl=0 …

CSV https://www.dropbox.com/s/tzevyxbu0o6ucnj/PACE-AugMerged.csv?dl=0 …

Folder link https://www.dropbox.com/sh/f3nfolkh1hlw9kg/AACb78M_jA3Q_NoBbnbQsWvXa?dl=0
Click to expand...
I assume missing cells are dropouts? Reasons for dropping out are still withheld?

Walking tests are abysmal. No wonder they dropped them entirely and pretended it wasn't important. There is so much variation in the initial walking tests. Is it walk as fast as you can? Or walk normally? If it's the latter then you have a natural ceiling of about 600360 and that makes it very misleading. A walk-as-fast-as-you-can could easily go over 1,000700. And it still fails. Pathetic.

Bit puzzled about the Oxford-caseness binary. Some participants are rated as no longer meeting Oxford criteria yet actually lowered on their walking test. So they must no longer have had any "unspecified fatigue" in the prior 6 months yet are no more functional. Shows how useless the criteria are.

What a joke this trial is. Sorry but something is massively wrong that this is actually considered serious research. A typical high-school science fair project is more serious than this. Smaller, cheaper, but definitely more rigorous. I can hardly express my disappointment in the entirety of the medical profession that this is actually taken seriously and not laughed out of the room. What tripe.

Screw it. Time to flex out some programming and do a bit of analysis.
 
Last edited:
I've had a very quick look.

The differences look alarming, I guess I'm going to have to do my own checks, but a few points: It does mostly seem to match up. The vast array of orange in the comparisons with SMC will be because the papers state the adjusted analyses and not the unadjusted analyses that Nick has done. We don't have access to the variables that would enable us to replicate their analyses unfortunately.

One thing I would like to know though is whether the APT group were scoring significantly higher on the depression rating for HADS. That they "mistranscribed" the data for this in the paper seems a tiny bit suspicious.

One of my initial criticisms of the original PACE paper was the lack of heterogeneity measures on the baseline data. So maybe the groups were less balanced than it appears.
 
And here's my summary of the variables, from earlier in this thread:
EQ_index is EuroQol-5D-3L (EQ_index in dataset).
The questionnaire asks about 5 items of health (Mobility, Pain/discomfort, Self-care, Anxiety/depression, Usual activities) scored on 3 levels (score 1, 2, 3) + a VAS of current health state (scored 0-100). EQ asks about your health today. The questionnaire used in PACE has also added a question about comparing how you feel now with 1 year ago.
The data in the dataset is the summary index score, and not the health state score (which replicates the individual item scores - eg 12132). The summary index score is normally used for cost-utility analyses. Euroqol summary index scores [in this dataset] appear to be <1.0 (although some are negative) - each country requires its own method of calculation based on age and sex.
The Euroqol user guide (section 4, page 11) provides more info on its calculation from health states obtained from the questionnaire itself: https://euroqol.org/wp-content/uploads/2016/09/EQ-5D-3L_UserGuide_2015.pdf
I assume higher scores on the index indicate better quality of life.

Hospital Anxiety and Depression Scale is a 14-item scale and asks how you feel over the past week.
Each item scores 0-3. Overall scores between 0 and 21 - presumably because anxiety and depression qs are split (7 questions each).
(HAANT and HADET in dataset).

Work and Social Adjustment Scale (WSAT). 5 items scored on 0-8 scale.
Scores between 0 and 40. No time specified - about how you feel now.

Borg scale is scored between 6 and 20. On the laminated sheet given to pts, 7 is scored "Very, very light", 11="Fairly light", 13="Somewhat hard", 19="Very, very hard".
Measures effort perception after the step test.
(STBOR in dataset)
[ETA: This webpage provides more context: https://www.hsph.harvard.edu/nutritionsource/borg-scale/]

Patient Health Questionnaire records physical (somatic) symptoms in past 4 weeks.
15-item scale. Each item scores 0-2. Scored out of 30.
(PSTOT in dataset)

PACE trial protocol has copies of the questionnaires used, but provides no info on how the scores are calculated.
Click to expand...
 
sTeamTraen said:
Not only should it be allowed, it should be mandatory. But in fact it should be mandatory for the authors not to play silly buggers and pretend that they don't have, somewhere, a single file with all the data in it, which is how everyone works, because it's simple and obvious. (People who have never seen scientific data are often surprised by just how simple it looks, unless it's coming from particle detectors at CERN, and even then, the main complexity is the number of rows.)

Of course, it's just about theoretically possible that there is no single-file copy of the data left, but at that point, the entire research team should hand back their PhDs and get jobs in McDonald's.
Click to expand...

But they were struggling to convert from Stata formats even though its very easy and there are plenty of examples of how to convert in places like stack exchange.

Its they way they suggest the data sets are huge when they are quite small. You don't need a server cluster to perform analytics on this data.
 
Even if they have provided the data in the same order as the Matthees data release, it is shockingly bad data management not to provide any fields on which to do an external check of any merge, such as trial arm. It shows they really don't care about the data. Makes me so cross. :mad:
 
Last edited:
Lucibee said:
Even if they have provided the data in the same order as the Matthees data release, it is shockingly bad data management not to provide any fields on which to do an external check of any merge, such as trial arm. It shows they really don't care about the data. Makes me so cross. :mad:
Click to expand...
Zero chance that the data provided to peer reviewers had those problems.

When there are no consequences to acting like an asshole, assholes will act to the maximum extent of being an asshole, and then some.

And so far the consequence to acting like assholes for the PACE gang has been in the form of awards, promotions, generous research funding and having their personal grievances amplified in international news coverage. So why stop now? The standard has been set that we are sub-human, so the people who should care about those things don't bother here.
 
rvallee said:
Zero chance that the data provided to peer reviewers had those problems.
Click to expand...

I very much doubt that the peer reviewers saw any more data than was in the published study. It was fast-tracked - which means they had even less time to scrutinise it than normal (generally an overnight turnaround). And given that those reviewers were likely to be friends of the trial authors anyway...
 
Lucibee said:
Even if they have provided the data in the same order as the Matthees data release, it is shockingly bad data management not to provide any fields on which to do an external check of any merge, such as trial arm. It shows they really don't care about the data. Makes me so cross. :mad:
Click to expand...

They should be embarassed about some of their claims around data management which suggest that QMUL is simply not fit to run a trial. There are levels of data management that you would not expect a professional organization to fall below and from what I can see they fell below these levels.

But it wouldn't surprise me if they did this on purpose to make it hard to validate the data.
 
You must log in or register to reply here.
Back
Top Bottom