Announcement

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

  • admetan: A new, comprehensive meta-analysis command

    With thanks as ever to Kit Baum (but particularly in this case, as I found a last-minute bug and had to ask him to re-upload...), I am very happy to introduce v3.0 of the admetan / ipdmetan meta-analysis command suite. The primary aim of this release was to separate out the functionalities of admetan and ipdmetan as far as possible. To that end, admetan can now be installed from SSC under its own name: that is, ssc install admetan works in the same way as ssc install ipdmetan; the package files are the same either way. Going forward, admetan can now directly be disseminated as an update to metan without any (potentially confusing) reference to individual participant data (IPD).

    For those of you who attended the London Stata Conference back in September: sorry for the wait
    Slides from that conference, including mine, can be found here: https://www.stata.com/meeting/uk18/


    What’s new in admetan:
    - a couple of new random-effects models, and a new syntax model() replacing re() [although the earlier syntax still works]
    - a couple of continuity-correction alternatives as proposed by Sweeting et al (2004)
    - the heterogeneity p-value can optionally be put back into the forest plot [I had quite a few aggrieved emails about that :-) ]
    - other information such as pooled effect p-values can also now be added to the forest plot. This isn’t done via an option, but instead admetan’s saved “results sets” (and forestplot itself) now leave behind a variable named _EFFECT which contains the string concatenation of effect size and confidence limits which appears in the forest plot. This can be edited to include p-values etc. before running forestplot. More info is in the help file (see e.g. the penultimate example for forestplot)
    - effect size, standard error, confidence limits, p-value, heterogeneity statistics etc. are now returned in matrices r(ovstat) [for overall] and r(bystats) [for subgroups]. These matrices are similar in structure to r(table) as returned by regression commands, and provide easy access to a much greater range of statistics
    - with cumulative and influence meta-analyses, the saving(filename) option now provides access to heterogeneity statistics for each iteration, as well as effect sizes, standard errors etc. Indirectly, this allows such information to be displayed in a forest plot
    - various minor bug fixes and improvements

    What’s new in forestplot:
    - a new option useopts which recalls forestplot options previously supplied to admetan, ipdmetan or ipdover and includes them in the current forestplot command line. The idea is that you can specify such options all in one go and not have to repeat yourself. In particular, you can use options such as counts, save and edit the “forestplot results set”, and run forestplot without needing to know how the “counts” information is stored
    - diamonds are now drawn as polygons instead of line segments, and hence can be filled!
    - various minor bug fixes and improvements

    What’s new in ipdover:
    - fixed the bug which meant no options could be specified to the command to the right of the colon

    What’s new in ipdmetan:
    - Nothing major (just the usual bug fixes and improvements) … but just to note that additional aggregate data [option ad() ] is now handled within ipdmetan.ado so that admetan.ado can stand alone.


    I hope these routines prove useful to the Stata community! It'd be great if you could help spread the word to anyone who uses Stata for meta-analysis or for creating (e.g. trial subgroup) forest plots.

    Many thanks,

    David.

    David Fisher
    Statistician
    MRC Clinical Trials Unit at UCL
    e-mail: [email protected]



    SSC TITLE

    'ADMETAN': module to provide comprehensive meta-analysis

    DESCRIPTION/AUTHOR(S)

    The main routine, admetan, is intended as an update of the
    popular Stata meta-analysis command ‘metan’, with greatly
    extended functionality including a wide range of random-effects
    models. The routine ‘forestplot’ is a stand-alone, re-written
    and extended version of the graphics routine within ‘metan’.
    ‘admetan’ can save data in a format which ‘forestplot’
    understands; together they allow extremely flexible and
    generalised forest plots to be produced. Also included is an
    “immediate” command ‘admetani’, which accepts numlists or
    matrices as input rather than variables in memory. Finally, there
    is ‘ipdmetan’, for two-stage individual participant data
    (IPD) meta-analysis; and an associated command ‘ipdover’ for
    creating forest plots of trial subgroups. For more information
    on these commands, type ssc describe ipdmetan.

  • #2
    A small update to the admetan and ipdmetan packages is now available, to fix the following bugs:

    -- in forestplot, unless using Stata 15 the program would fail due to the presence of lalign(center) as a default option

    -- in ipdover, the risk-difference method failed due to a typo

    -- in admetan, certain random-effects models (such as HKSJ) were not described correctly in the Results Window, although the results themselves were correct.

    With thanks as ever to Kit Baum.

    Enjoy!

    David.

    Comment


    • #3
      Statalist post 4th Feb 2019

      With thanks as ever to Kit Baum, the most recent update (v3.2) to the admetan and ipdmetan packages is now available via SSC.

      Main changes in this version are:

      -- Added the Sidik-Jonkman "robust" or "sandwich-like" variance estimator (Sidik & Jonkman CompStatDataAnalysis 2006)

      -- Added the Skovgaard second-order likelihood correction (Guolo Stat Med 2012) as an suboption to the Profile Likelihood model (note that the similar Bartlett correction was already available)

      -- Corrected the bug which caused continuity correction options to fail; continuity correction (if applicable) is also now explicitly shown on-screen

      -- Corrected inconsistencies for certain models between statistics returned on-screen and in the r() namespace

      -- Corrected errors causing some of the help-file examples not to run; plus other minor clarifications & improvements in the help-file texts

      Many thanks,

      David.
      Last edited by David Fisher; 04 Feb 2019, 03:28.

      Comment


      • #4
        Dear David,

        I've just already installed admetan and ipdmetan packages in Stata 15 and it seems something is wrong with the latest update (yesterday i performed the same command in another PC with no error at all) and Stata 15.

        I've tried with the examples you mention in the help documentation and it occurs the same, the message i obtain is:

        "
        option lalign() not allowed

        Error in forestplot
        "

        What's going on?

        Thanks in advance
        Ana

        Comment


        • #5
          Hi Ana,

          This bug should have been fixed back in December (see post #2 of this thread). Please type which admetan at the Stata command line; it should read "*! version 3.2 David Fisher 28jan2019". If it does not, then you are not using the most recent version.

          The easiest thing to try is: adoupdate admetan or adoupdate admetan, update. If this does not solve the problem, there may be other issues on your system e.g. having user-installed ado files in multiple locations. Let me know if I can help further!

          Thanks,

          David.

          Comment

          Working...
          X