A New Decade of ME Research: The 11th Invest in ME International ME Conference 2016
Mark Berry presents the first in a series of articles on the 11th Invest in ME International ME Conference in London ...
Discuss the article on the Forums.

Carlisle’s statistics bombshell names and shames rigged clinical trials

Discussion in 'Other Health News and Research' started by BurnA, Jun 6, 2017.

  1. BurnA

    BurnA Senior Member

    Messages:
    2,086
    Likes:
    9,863
  2. Ysabelle-S

    Ysabelle-S Highly Vexatious

    Messages:
    523
    Likes:
    4,041
    Carlisle.JPG

    Someone posted that link on Carlisle on Twitter earlier and I thought it was interesting. I particularly liked the above.
     
  3. A.B.

    A.B. Senior Member

    Messages:
    3,751
    Likes:
    23,191
    Someone apply this technique to CBT/GET studies please.
     
  4. trishrhymes

    trishrhymes Senior Member

    Messages:
    2,153
    Likes:
    17,871
    I suspect this particular technique would not show up any problems with ME studies. It only looks for falsification in patient's basic measured data at the beginning of the study. It doesn't look at the problems we see in PACE of outcome switching etc.
     
    Woolie, Never Give Up and Esther12 like this.
  5. A.B.

    A.B. Senior Member

    Messages:
    3,751
    Likes:
    23,191
    You never know what they might find. At the very least, in the case of PACE, it would be interesting to see just how much the selected patients deviated from the norm.
     
  6. RogerBlack

    RogerBlack Senior Member

    Messages:
    891
    Likes:
    2,889
    AIUI, it relies on understanding of the statistics.
    So if you don't understand the correlates of severity of disease, and the statistics of that versus the normal population, it may be less than useful.
     
  7. BurnA

    BurnA Senior Member

    Messages:
    2,086
    Likes:
    9,863
    I am not sure if I have understood it correctly so hoping others might explain it.
    But to me he analysed the "randomised" baseline data in the subgroups to see if it was truly random or not. If the baseline data of the randomised subgroups was 'too' normal or if it was not normal enough then this was raised as potential fraud because the chances of the data being randomised was too small.
     
  8. Woolie

    Woolie Senior Member

    Messages:
    1,930
    Likes:
    14,558
    Yes. There is no evidence the PACE authors falsified anything. All their data checks out. Its just they presented it in a way that was designed to maximise the apparent success of the trial. But that's not the same as falsification.
     
    Last edited: Jun 7, 2017
    Esther12, Valentijn and trishrhymes like this.
  9. BurnA

    BurnA Senior Member

    Messages:
    2,086
    Likes:
    9,863
    It's more than likely you are correct but...
    When you say all their data checks out - has PACE been analysed using this technique?

    Remember, for the trials listed in this report, there was no evidence of fraud before this analysis took place, so prior absence of evidence of fraud, is not a predictor of whether this technique will find anything or not.

    I think this technique will find if someone has played with the figures, speculating that it might or might not reveal something in an individual trial ( whatever that trial might be) is kind of pointless. Nobody knows until the analysis is done.

    ETA : Do we have enough data to do this on PACE. ?
     
  10. BurnA

    BurnA Senior Member

    Messages:
    2,086
    Likes:
    9,863
  11. RogerBlack

    RogerBlack Senior Member

    Messages:
    891
    Likes:
    2,889
    To sum up the important bits of the blog above it would probably be fair to say 'The technique is interesting, but risks falsely flagging papers as problematic when there are significant correlations between the reported values at baseline'. For example, if height looks a bit unlikely, and gender looks a bit unlikely, you can't simply multiply the likelyhoods together, as an unlikely gender mix is likely to imply an unlikely height mix, as gender and height are strongly correlated.
     
    ScottTriGuy and BurnA like this.

See more popular forum discussions.

Share This Page