Run subgroup analyses

This function allows to simultaneously conduct different subgroup analyses using runMetaAnalysis objects.

Usage

subgroupAnalysis(.model,
                 ...,
                 .which.run = .model$which.run[1],
                 .round.digits = 2,
                 .nntCer = NULL,
                 .tau.common = FALSE,
                 .html = TRUE)

Arguments

.model

An object of class "runMetaAnalysis", created by runMetaAnalysis.

...

<dplyr_data_masking>. A number of subgroup variables included in the original dataset provided to runMetaAnalysis, separated by commas.

.which.run

The model in .model that should be used for the subgroup analyses. Uses the first model in .model if no value is specified by the user. Studies or effects excluded in a model (e.g. "outliers", "rob") are also excluded in the subgroup analysis. Must be one of "overall", "combined", "lowest", "highest", "outliers", "influence", "rob", "threelevel", "threelevel.che", "ccrem" or "ccrem.che".

.round.digits

numeric. Number of digits to round the (presented) results by. Default is 2.

.nntCer

numeric. Value between 0 and 1, indicating the assumed control group event rate to be used for calculating NNTs via the Furukawa-Leucht method. If set to NULL (default), the value saved in .model is (re-)used.

.tau.common

logical. Should a common (TRUE) or subgroup-specific (FALSE) estimate of the between-study heterogeneity be calculated when analyzing the subgroups? FALSE by default. Note that subgroup analyses based on "multilevel" models automatically assume common heterogeneity estimates.

.html

logical. Should an HTML table be created for the results? Default is TRUE.

Value

Returns an object of class "subgroupAnalysis". This object includes, among other things, a data.frame with the name summary, in which all subgroup analysis results are summarized. Other objects are the "raw" subgroup analysis model objects returned. This allows to conduct further operations on some subgroup analysis specifically.

Details

When a multilevel model is selected in .which.run, subgroup effects are estimated in a meta-regression with the same random effects structure and sampling (co-)variances as in the original model. The p-value tests the null hypothesis that the effects are equal across subgroups. Confidence intervals and tests in this meta-regression are model-based, even if robust variance estimation was used in the original model.

In the "metabin" models created when es.type = "raw", subgroup effects are pooled using the inverse-variance method, not the Mantel-Haenszel method used in the main analysis.

For more details see the Get Started vignette.

See also

runMetaAnalysis

Author

Mathias Harrer mathias.h.harrer@gmail.com, Paula Kuper paula.r.kuper@gmail.com, Pim Cuijpers p.cuijpers@vu.nl

Examples

if (FALSE) { # \dontrun{
data("depressionPsyCtr")

depressionPsyCtr %>%
  checkDataFormat() %>%
  checkConflicts() %>%
  calculateEffectSizes() %>% 
  filterPoolingData(condition_arm2 %in% 
                      c("wl", "other ctr")) -> data

# Run the meta-analyses
runMetaAnalysis(data) -> res

# Subgroup analysis
subgroupAnalysis(res, condition_arm2, country,
                 .which.run = "combined",
                 .tau.common = TRUE) -> sg
plot(sg, "condition_arm2")
plot(sg, "country")
} # }