Announcement

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

  • meta analysis of proportions

    I am conducting a meta analysis of prevalence studies in Stata. Stata displays proportions in table and forest plot. Is there any way to get % on both table and forest plot? I ma using metan cmd.

  • #2
    Originally posted by Parveen Akhtar View Post
    I am conducting a meta analysis of prevalence studies in Stata. Stata displays proportions in table and forest plot. Is there any way to get % on both table and forest plot? I ma using metan cmd.
    Hi Parveen,
    Welcome to the Statalist.

    Also have a go with the user written metaprop command for meta-analysing proportions. You can then get the forest plot to display the % by calling
    Code:
    lcols(_ES)
    Good luck.






    Many thanks,
    Alexander
    (Stata v14.2 IC for Mac)

    Comment


    • #3
      Thanks @
      Alexander
      .Does Stata allows to do a mixed effect subgroup analysis meaning random-effects-model within subgroups, fixed-effects-model between subgroups?

      Comment


      • #4
        Originally posted by Parveen Akhtar View Post
        Thanks @
        Alexander
        .Does Stata allows to do a mixed effect subgroup analysis meaning random-effects-model within subgroups, fixed-effects-model between subgroups?
        Hi Parveen,
        I'm not quite sure what you are asking but you can certainly ask metaprop to fit fixed and random effects models to your data. As for sub-groups, again a quick glance at the help file answers some of these questions.

        Here is an example. I hope this demonstrates that what manipulations and additions to the plot you can make.






        Attached Files
        Many thanks,
        Alexander
        (Stata v14.2 IC for Mac)

        Comment


        • #5
          Many thanks. I want to do a subgroup analysis for prevalence of depression among university students.
          For subgroup analysis, need to use mixed effects random analysis and need Q-value and its it corresponding p-value for difference between subgroups. . But stata metan cmd does not provide it. How can I do that?

          Comment


          • #6
            Originally posted by Alexander Rodriguez View Post

            Hi Parveen,
            I'm not quite sure what you are asking but you can certainly ask metaprop to fit fixed and random effects models to your data. As for sub-groups, again a quick glance at the help file answers some of these questions.

            Here is an example. I hope this demonstrates that what manipulations and additions to the plot you can make.





            Hi Alexander,

            May I ask what command you used to add 'study population' and 'number of success' onto your forest plot please?

            Comment


            • #7
              I would like to know how I could conduct a meta-analysis of prevalence studies when I don't have the number of cases and sample size, which are the variables requested by metaprop. I have many papers that only provide the estimated prevalence and its 95%CI. Thanks for your attention.

              Comment


              • #8
                Unless do not know already the standard error of your effect sizes you can't because the essence of meta-analysis is to give effects of different studies a weight depending on their standard errors. In case of proportions (or percentages) you can calculate the sample size if you know the standard error -- if the confidence intervals are symmetric (which need not be the case for proportions) using the 95%-CIs instead of the standard errors will do, as well. See some introduction into meta analysis such as Lipsey & Wilson (see here) or the Stata manual [META] (the manuals are neglegted far too often).

                BTW: The link to the effec size calculator in the link above is broken, use this link instead: https://www.campbellcollaboration.or...lator-Home.php
                Last edited by Dirk Enzmann; 29 May 2021, 16:25. Reason: Adding a note to the link of Wilson's effect size calculator

                Comment


                • #9
                  Here an example that shows how to get the number of cases from a proportion and its lower 95%-CI only -- assuming normal (Wald) confidence intervals (otherwise the estimated number of cases will not be valid):
                  Code:
                  /* Estimate sample size from proportion and lower 95%-CI (Wald-Test) */
                  
                  program define est_nprop, rclass
                     syntax, Prop(real) CIL(real) [ ITeration(int 100000)]
                     local vp = `prop'*(1-`prop')
                     local p_cil = `prop'-`cil'
                     local n = 1
                     while `n' < `iteration'+1 {
                        local ++n
                        local e_cil = sqrt(`vp'/`n')*invt(`=`n'-1',.975)
                        if `p_cil' >= `e_cil' {
                           di _n as txt "Sample size = " as res `n'
                           return scalar n = `n'
                           return scalar se = sqrt(`vp'/`n')
                           return scalar prop = `prop'
                           exit
                        }
                        if `n'==`iteration' & `p_cil' < `e_cil' {
                           di _n "convergence not achieved after `n' iterations"
                           exit
                        }
                     }
                  end
                  
                  * ------------------------------------------------------------------------------
                  // Example:
                  
                  sysuse auto, clear
                  
                  proportion foreign, citype(normal)
                  local p = r(table)[1,1]
                  local cil = r(table)[5,1]
                  
                  // Forget sample size, use proportion and lower 95%-CI only:
                  est_nprop, p(`p') cil(`cil')
                  Result:
                  Code:
                  . proportion foreign, citype(normal)
                  
                  Proportion estimation             Number of obs   =         74
                  
                  --------------------------------------------------------------
                               |                                   Normal
                               | Proportion   Std. Err.     [95% Conf. Interval]
                  -------------+------------------------------------------------
                       foreign |
                     Domestic  |   .7027027   .0531331      .5968085    .8085969
                      Foreign  |   .2972973   .0531331      .1914031    .4031915
                  --------------------------------------------------------------
                  
                  . local p = r(table)[1,1]
                  
                  . local cil = r(table)[5,1]
                  
                  .
                  . // Forget sample size, use proportion and lower 95%-CI only:
                  . est_nprop, p(`p') cil(`cil')
                  
                  Sample size = 74

                  Comment


                  • #10
                    Hi Dirk Enzmann

                    Many thanks for your help.

                    Comment


                    • #11
                      Hi Everyone. Is there a chance to include also studies with just one case using meta prop? if so how? Thx

                      Comment

                      Working...
                      X