Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • #16
    Hi David Fisher,
    I think I found a bug - when I do a proportion meta-analysis using the options 'model' and 'user_spec', in the forest plot, the diamond does not appear and ‘(Insufficient data)’ appears instead of the pooled estimate.
    These options seem to work fine with standard (non-proportion) meta-analysis.

    Attached is the dataset and the command I used
    Code:
    metan mild n, pr model(re \ 0.50 0.40 0.60) transform(ftukey, iv) counts

    Also, would it be possible to active the option 'sa [, isq(real) tausq(real)]' for non-standard models or modifications to standard models (at the moment it only works with standard inverse-variance random-effects model).

    Regards,
    Luis
    Attached Files

    Comment


    • #17
      Luis Furuya-Kanamori and David Fisher I don't think we should open up user defined tau2 for other models especially QE or IVhet as they only work with DL values and thus would open this up to wrong implementation and then another series of problems....

      Regards
      Suhail
      Regards
      Suhail Doi

      Comment


      • #18
        Dear all,

        With thanks as ever to Kit Baum, an updated version of metan (v4.03 28apr2021) is now available via SSC.

        The following bugs have (hopefully) been fixed:Thanks to all users who have reported issues; please continue to do so!

        Best wishes,

        David.


        P.S. Please note that, as a result of this update, users of ipdmetan will also need to update that package (via SSC). The latest version of ipdmetan is now v4.02 20apr2021.
        Last edited by David Fisher; 04 May 2021, 04:43.

        Comment


        • #19
          Luis Furuya-Kanamori : Unfortunately, upon opening this thread to post the above, I realise that I completely missed your post #16. Really sorry about that; this will be top of the list for the next update! In the meantime, here is a work-around which I hope is acceptable:

          Code:
          preserve
              metan mild n, pr model(re \ 0.50 0.40 0.60) transform(ftukey, iv) counts clear nograph
              foreach x in ES LCI UCI {
                  replace _Prop_`x' = _`x' if _LABELS=="Overall, User"
              }
              forestplot, useopts
          restore
          Transformed proportion data is currently handled as a unique case in metan, because the Freeman-Tukey transform and back-transform are complex. The exponential/log transform is simple, so I only need to store the log-transformed values and exponentiate them as and when needed. But with transformed proportion data I need to maintain two different sets of effect sizes. The bug you have found is a result of metan looking in the wrong place for the user-defined values. The code above copies these values and pastes them in the correct place.

          Apologies again, and best wishes,

          David.

          Comment


          • #20
            Hi David,

            Many thanks for the update. Something has changed with the "range" option in forestplot (only for proportions). For example if I run the below, it used to work perfectly in the previous version but now I cannot use range to get a well fitted plot no matter what I try and even if I remove the denom option

            Code:
            input str3 studyname long n int cases byte qi
            "S1"  217154 422 1
            "S10"  16557  32 1
            "S13"    676   1 1
            "S18"     44   1 1
            "S26"     29   1 1
            end
            
            
            metan cases n  , pr qe(qi) transform(ftukey, iv) study( studyname ) counts forestplot(astext(85) boxscale(35) spacing(0.99) leftjustify range(0.1 20) dp(3)) denom(1000)
            Please advise
            Regards
            Suhail
            Last edited by Suhail Doi; 06 May 2021, 15:33.
            Regards
            Suhail Doi

            Comment


            • #21
              Hi David,

              Another small suggestion for a future update - for the forestplot, if the het stats are by default on the next line after overall then it makes formatting easier. Currently they are on the same line and need adding a column before the CI plot to move them to the next line

              Regards
              Suhail
              Regards
              Suhail Doi

              Comment


              • #22
                Hi Suhail,
                I think the metan option extraline(yes) should do what you need.
                Thanks,
                David.

                Comment


                • #23
                  Hi David, that works great, thanks - seems I missed this in the help file!

                  Regards
                  Suhail
                  Regards
                  Suhail Doi

                  Comment

                  Working...
                  X