Other than SAS are there any packages for implementing this model in statistical software? I was told since it is non-linear it cant be done in R or STATA.
1 Like
What about the mada package for R? The package description says:
Next to basic analysis and visualization the bivariate Model of Reitsma et al. (2005) that is equivalent to the HSROC of Rutter & Gatsonis (2001) can be fitted.
Does this fit your needs?
Unfortunately that model is only valid when all studies use the same threshold for classification, mine do not. I was told if it is used combining studies with mixed thresholds the summary point estimate is meaningless.
Hi, We wrote an R package for this which has recently become available on CRAN again after a spell. You can find the manual as well at this link:
https://cran.r-project.org/web/packages/HSROC/index.html
The Cochrane handbook suggests using this model if the positivity threshold is not the same across studies. See example 2 here: https://methods.cochrane.org/sites/methods.cochrane.org.sdt/files/public/uploads/Chapter%2010%20-%20Version%201.0.pdf
1 Like
PS: We have also written WinBUGS and SAS programs for Bayesian estimation of the HSROC model:
https://www.nandinidendukuri.com/software/diagnostic-meta-analysis/hsroc-programs
Thank you very much. This is very useful information.
Can covariates be added to this model to evaluate causes of heterogeneity?
Yes, covariates can be added when using the WinBUGS or SAS programs.
The R package HSROC does not allow for adding covariates.
There are two packages in Stata (metandi and midas) that use the bivariate method but this is equivalent to the HSROC approach and results are identical.
They are installed in the standard way (ssc install metandi)
However, I wouldn’t recommend HSROC / bivariate approaches and instead recommend the Stata package diagma
To install: ssc install diagma
The diagma package flips DTA meta-analysis on its head by synthesizing the odds ratio and then splitting it into Se and Sp and other measures associated with DTA. This method is called the split component synthesis method and is also available as a shiny app in R called SCSMeta. Simulation evidence suggests that this outperforms bivariate and HSROC methods.