# Using multiple software in a meta-analysis

I am currently performing a meta-analysis of diagnostic test accuracy (DTA). I am planning to perform a univariate meta-analysis to estimate the pooled diagnostic odds ratio (DOR) with R. Then, I intend to fit the summary receiver operating characteristic (SROC) curves with HSROC parameterization model. However, due to the limited utility of R in DTA reviews, I want to use Stata to visualize the SROC curves.

Generally, is it okay to use multiple software for analysis in a meta-analysis (or any general analysis)? I am aware that both software will produce similar outputs, but not exactly same, so I am quite confused as some researchers believe that the exact effect estimates are important, in addition to the effect trends. Insights from both statistical and publication (journal) perspectives are highly appreciated. Thank you very much in advance

UPDATE: Here's some data to illustrate my case

> STATA Estimates
> --------------------------------------------------------------------------
>                            |         es         se     95%.low    95%.high
> Sensitivity                |   .8116316   .0455623    .7061064    .8854156
> Specificity                |   .8258526   .0396756    .7341574    .8906317
> Diagnostic Odds Ratio      |    20.4332   10.59374    7.396481     56.4479
> LR+                        |   4.660602   1.243752    2.762396    7.863176
> LR-                        |   .2280896   .0626711    .1331143    .3908286
> 1/LR-                      |   4.384241   1.204638    2.558667    7.512339
> Sensitivity Single miRNA   |   .7803988   .0838116    .5767566    .9026051
> Sensitivity Multiple miRNA |   .8254414   .0516111    .7009058    .9051411
> Specificity Single miRNA   |   .7818269   .0711182    .6128164     .890272
> Specificity Multiple miRNA |   .8451906   .0401313    .7495517    .9087538
> LR+ Single miRNA           |   3.576971   1.472639    1.596157    8.015951
> LR+ Multiple miRNA         |   5.331986   1.631764      2.9268    9.713704
> LR- Single miRNA           |   .2808821   .1272398    .1155924    .6825256
> LR- Multiple miRNA         |   .2065316   .0682497    .1080691    .3947041

> R Estimates
> --------------------------------------------------------------------------
>                            |         es         se     95%.low    95%.high
> Sensitivity                |      0.804          -       0.683       0.887
> False positive rate        |      0.182          -       0.112       0.281
> Specificity                |      0.818          -       0.719       0.888
> Diagnostic Odds Ratio      |      21.90          -        5.97      57.400
> LR+                        |       4.61          -        2.51       7.780
> LR-                        |       0.25          -        0.13       0.429
> 1/LR-                      |       4.39          -        2.33       7.690
> Sensitivity Single miRNA   |      0.814          -       0.617       0.922
> Sensitivity Multiple miRNA |      0.773          -       0.529       0.912
> Specificity Single miRNA   |       0.84          -       0.762       0.896
> Specificity Multiple miRNA |      0.802          -       0.415       0.959
> LR+ Single miRNA           |      5.230          -      2.7700       8.620
> LR+ Multiple miRNA         |       5.73          -      0.9380       21.70
> LR- Single miRNA           |      0.239          -      0.0886       0.486
> LR- Multiple miRNA         |       0.36          -      0.0928        1.08


As seen in the table above, the sensitivity, specificity, DOR, LR+, LR-, and 1/LR- outputs between Stata and R are relatively comparable. However, the estimates for each subgroup is different. It is worth noting that, since I calculated the estimates for each subgroup manually in R (i.e., fitting meta-analysis model for each subgroup), while the estimates in Stata are calculated by an automated process, I am quite unsure whether my method of calculation in R is valid. That being said, I believe it is more favorable to present the results from STATA. However, I need R to perform the some other analyses. My point is: R can perform the analysis, but I am more in favor of the outputs produced by STATA.

#n.b.: Below is the reason why I performed separate meta-analyses for each subgroup

Adding covariate by using reitsma in R will perform meta-regression, while in fact I want to make subgroups i.e., sensitivity, specificity etc. for each subgroup. As an alternative, I fitted the bivariate meta-analysis model for each subgroup manually:

# Meta-regression model
fit.data.type.reml <- reitsma(data.DTA, correction.control = "single", formula = cbind(tsens, tfpr) ~ type)

# Subgroup analysis - Fitting bivariate meta-analysis for each subgroup
fit.data.single <- reitsma(subset(data.DTA, type == "single"), correction.control = "single")
fit.data.multiple <- reitsma(subset(data.DTA, type == "multiple"), correction.control = "single")

• When presented with a choice, consider preferring software that is open-source. All else being equal, works that use software that is easy to access and verify will be more reproducible in the long run. Aug 2, 2021 at 15:19
• Fitting separate models for each sub-group is different from doing a meta-regression using a binary covariate. Aug 3, 2021 at 12:17