Nothing
R/metabind.R
In meta: General Package for Meta-Analysis
Defines functions
metabind
Documented in
metabind
#' Combine and summarize meta-analysis objects
#'
#' @description
#' This function can be used to combine meta-analysis objects and is,
#' for example, useful to summarize results of various meta-analysis
#' methods or to generate a forest plot with results of several
#' subgroup analyses.
#'
#' @param ... Any number of meta-analysis objects or a single list
#' with meta-analyses.
#' @param name An optional character vector providing descriptive
#' names for the meta-analysis objects.
#' @param subgroup An optional variable to generate a forest plot with
#' subgroups.
#' @param common A logical vector indicating whether results of common
#' effect model should be considered.
#' @param random A logical vector indicating whether results of random
#' effects model should be considered.
#' @param prediction A logical vector indicating whether results of
#' prediction intervals should be considered.
#' @param backtransf A logical indicating whether results should be
#' back transformed in printouts and plots. If
#' \code{backtransf=TRUE} (default), results for \code{sm="OR"} are
#' printed as odds ratios rather than log odds ratios, for example.
#' @param outclab Outcome label for all meta-analyis objects.
#' @param pooled Deprecated argument (replaced by \code{common} and
#' \code{random}.
#' @param warn.deprecated A logical indicating whether warnings should
#' be printed if deprecated arguments are used.
#'
#' @details
#' This function can be used to combine any number of meta-analysis
#' objects which is useful, for example, to summarize results of
#' various meta-analysis methods or to generate a forest plot with
#' results of several subgroup analyses (see Examples).
#'
#' Individual study results are not retained with \code{metabind} as
#' the function allows to combine meta-analyses from different data
#' sets (e.g., with randomised or observational studies). Individual
#' study results are retained with R function \code{\link{metamerge}}
#' which can be used to combine results of meta-analyses of the
#' same dataset.
#'
#' @return
#' An object of class \code{c("metabind", "meta")} with corresponding
#' generic functions (see \code{\link{meta-object}}).
#'
#' @author Guido Schwarzer \email{guido.schwarzer@@uniklinik-freiburg.de}
#'
#' @seealso \code{\link{metagen}}, \code{\link{forest.metabind}},
#' \code{\link{metamerge}}
#'
#' @examples
#' data(Fleiss1993cont)
#'
#' # Add some (fictitious) grouping variables:
#' #
#' Fleiss1993cont$age <- c(55, 65, 55, 65, 55)
#' Fleiss1993cont$region <- c("Europe", "Europe", "Asia", "Asia", "Europe")
#'
#' ma <- metacont(n.psyc, mean.psyc, sd.psyc, n.cont, mean.cont, sd.cont,
#' data = Fleiss1993cont, sm = "SMD")
#'
#' # Conduct two subgroup analyses
#' #
#' ma1 <- update(ma, subgroup = age, subgroup.name = "Age group")
#' ma2 <- update(ma, subgroup = region, subgroup.name = "Region")
#'
#' # Combine random effects subgroup meta-analyses and show forest
#' # plot with subgroup results
#' #
#' mb <- metabind(ma1, ma2, common = FALSE)
#' mb
#' forest(mb)
#'
#' # Use various estimation methods for between-study heterogeneity
#' # variance
#' #
#' ma.pm <- update(ma, method.tau = "PM")
#' ma.dl <- update(ma, method.tau = "DL")
#' ma.ml <- update(ma, method.tau = "ML")
#' ma.hs <- update(ma, method.tau = "HS")
#' ma.sj <- update(ma, method.tau = "SJ")
#' ma.he <- update(ma, method.tau = "HE")
#' ma.eb <- update(ma, method.tau = "EB")
#'
#' # Combine meta-analyses and show results
#' #
#' taus <- c("Restricted maximum-likelihood estimator",
#' "Paule-Mandel estimator",
#' "DerSimonian-Laird estimator",
#' "Maximum-likelihood estimator",
#' "Hunter-Schmidt estimator",
#' "Sidik-Jonkman estimator",
#' "Hedges estimator",
#' "Empirical Bayes estimator")
#' #
#' ma.taus <- metabind(ma, ma.pm, ma.dl, ma.ml, ma.hs, ma.sj, ma.he, ma.eb,
#' name = taus, common = FALSE)
#' ma.taus
#' forest(ma.taus)
#'
#' @export metabind
metabind <- function(..., subgroup = NULL,
name = NULL,
common = NULL, random = NULL, prediction = NULL,
backtransf = NULL, outclab = NULL,
pooled = NULL,
warn.deprecated = gs("warn.deprecated")) {
#
#
# (1) Check and set arguments
#
#
missing.subgroup <- missing(subgroup)
subgroup.meta <- subgroup
missing.name <- missing(name)
missing.pooled <- missing(pooled)
missing.backtransf <- missing(backtransf)
#
chklogical(warn.deprecated)
#
args <- list(...)
#
n.meta <- length(args)
seq.meta <- seq_len(n.meta)
#
if (n.meta == 1) {
if (inherits(args[[1]], "meta.rm5")) {
args <- args[[1]]
#
if (missing.name) {
name <- unlist(lapply(args, "[[" , "outclab"))
missing.name <- FALSE
}
}
#
else if (inherits(args[[1]], "copas"))
return(metamerge(args[[1]]))
#
else if (inherits(args[[1]], "limitmeta"))
return(metamerge(args[[1]]))
#
else if (inherits(args[[1]], "netpairwise"))
stop("Elements of argument '...' may not be of class 'netpairwise'.",
call. = FALSE)
#
else if (inherits(args[[1]], "meta"))
return(args[[1]])
#
else if (!is.list(args[[1]]))
stop("All elements of argument '...' must be of class 'meta', ",
"'limitmeta', or 'copas'.",
call. = FALSE)
else {
n.meta <- length(args[[1]])
seq.meta <- seq_len(n.meta)
#
args2 <- list()
for (i in seq.meta)
args2[[i]] <- args[[1]][[i]]
#
args <- args2
}
}
#
if (!missing.pooled) {
pooled <- setchar(pooled, c("common", "random", "fixed"))
pooled[pooled == "fixed"] <- "common"
#
pooled.common <- pooled == "common"
pooled.random <- pooled == "random"
}
#
common <-
deprecated2(common, missing(common), pooled.common, missing.pooled,
warn.deprecated, "pooled")
#
if (!is.null(common)) {
chklogical(common[1], "common")
#
if (length(common) == 1)
common <- rep_len(common, n.meta)
else
chklength(common, n.meta,
text = paste("Length of argument 'common' differs from",
"number of meta-analyses."))
}
#
random <-
deprecated2(random, missing(random), pooled.random, missing.pooled,
warn.deprecated, "pooled")
#
if (!is.null(random)) {
chklogical(random[1], "random")
#
if (length(random) == 1)
random <- rep_len(random, n.meta)
else
chklength(random, n.meta,
text = paste("Length of argument 'random' differs from",
"number of meta-analyses."))
}
#
if (!missing(prediction)) {
if (length(prediction) == 1)
prediction <- rep_len(prediction, n.meta)
#
if (!is.logical(prediction))
stop("Argument 'prediction' must contain logical values.",
call. = TRUE)
#
chklength(prediction, n.meta,
text = paste("Length of argument 'prediction' differs from",
"number of meta-analyses."))
}
#
if (!missing.backtransf)
chklogical(backtransf)
#
# Act on limitmeta and copas objects
#
name.i <- vector("character", n.meta)
#
is.limit <- is.copas <- is.trimfill <- is.subgroup <- rep(FALSE, n.meta)
samedata <- rep(NA, n.meta)
#
for (i in seq.meta) {
if (inherits(args[[i]], "metabind"))
stop("Elements of argument '...' may not be of class 'metabind'.",
call. = FALSE)
#
if (inherits(args[[i]], "netpairwise"))
stop("Elements of argument '...' may not be of class 'netpairwise'.",
call. = FALSE)
#
if (inherits(args[[i]], "meta")) {
if (inherits(args[[i]], "metamerge"))
stop("Meta-analysis objects created with metamerge() cannot ",
"be used in metabind().",
call. = FALSE)
#
if (!inherits(args[[i]], "metaadd"))
args[[i]] <- updateversion(args[[i]])
#
if (missing.name) {
if (inherits(args[[i]], "trimfill")) {
is.trimfill[i] <- TRUE
#
name.i[i] <- "trimfill"
}
else
name.i[i] <- replaceNULL(args[[i]]$subgroup.name, "")
#
if (name.i[i] == "")
name.i[i] <- class(args[[i]])[1]
}
}
#
else if (inherits(args[[i]], "copas")) {
args[[i]] <-
metamerge(update(args[[i]]$x, common = FALSE, random = FALSE),
args[[i]], label2 = "copas")
#
is.copas[i] <- TRUE
#
if (missing.name)
name.i[i] <- "copas"
}
#
else if (inherits(args[[i]], "limitmeta")) {
args[[i]] <-
metamerge(update(args[[i]]$x, common = FALSE, random = FALSE),
args[[i]], label2 = "limit")
#
is.limit[i] <- TRUE
#
if (missing.name)
name.i[i] <- "limitmeta"
}
#
else
stop("All elements of argument '...' must be of class 'meta', ",
"'limitmeta', or 'copas'.",
call. = FALSE)
#
is.subgroup[i] <- !is.null(args[[i]]$subgroup)
samedata[i] <- !samedata(args[[1]], args[[i]], stop = FALSE)
}
#
# Meta-analyses must all contain subgroups or none
#
if (length(unique(is.subgroup)) != 1)
stop("All or none meta-analyses must contain subgroups.",
call. = FALSE)
#
with.subgroups <- any(is.subgroup)
is.limit.copas <- is.limit | is.copas
#
samedata[is.limit.copas] <- TRUE
samedata <- all(samedata)
#
#
# (2) Extract meta-analytical methods
#
#
meth.list <- vector("list", n.meta)
#
for (i in seq.meta)
meth.list[[i]] <-
meta2meth(args[[i]], outclab)
#
meth <- list()
#
for (i in names(meth.list[[1]]))
meth[[i]] <- condense(meth.list, i)
#
meth$common <- replaceNULL(common, meth$common)
meth$common[is.limit.copas] <- FALSE
meth$random <- replaceNULL(random, meth$random)
meth$prediction <- replaceNULL(prediction, meth$prediction)
#
# Use common effect estimate if no result is selected
#
meth$common[!(meth$common | meth$random | meth$prediction)] <- TRUE
#
# Check whether settings are unique
#
meth2 <- meth[c("sm", "level.ma", "level.predict", "null.effect")]
n.meth <- lapply(meth2, function(x) length(unique(x)))
#
if (any(n.meth != 1))
stop("Setting for the following argument",
if (sum(n.meth != 1) > 1) "s",
" must be the same for all meta-analyses: ",
paste0(paste0("'", names(meth2)[n.meth != 1], "'"),
collapse = " - "))
#
# Unify settings
#
meth$sm <- makeunique(meth$sm)
meth$method <- unique(meth$method)
meth$method.random <- unique(meth$method.random)
meth$three.level <- makeunique(meth$three.level, FALSE)
#
meth$level <- meth$level.ma <- makeunique(meth$level.ma)
meth$level.predict <- makeunique(meth$level.predict)
#
common <- meth$common
random <- meth$random
prediction <- meth$prediction
#
meth$overall <- with.subgroups & samedata
meth$overall.hetstat <- with.subgroups & samedata
#
if (missing.pooled & all(meth$common) & all(!meth$random))
meth$common <- pooled == "onlycommon"
else
meth$common <- any(meth$common)
#
meth$random <- any(meth$random)
#
meth$prediction <- any(meth$prediction)
#
meth$method.common.ci <- unique(meth$method.common.ci)
meth$method.random.ci <- unique(meth$method.random.ci)
meth$adhoc.hakn.ci <- unique(meth$adhoc.hakn.ci)
meth$method.tau <- unique(meth$method.tau)
meth$tau.preset <- unique(meth$tau.preset)
meth$TE.tau <- unique(meth$TE.tau)
meth$tau.common <-
if (!with.subgroups)
FALSE
else
unique(meth$tau.common)
#
meth$method.I2 <- unique(meth$method.I2)
#
meth$prediction.subgroup <-
if (!with.subgroups)
FALSE
else
unique(meth$prediction.subgroup)
meth$method.predict <- unique(meth$method.predict)
meth$adhoc.hakn.pi <- unique(meth$adhoc.hakn.pi)
#
meth$method.bias <- unique(meth$method.bias)
meth$null.effect <- makeunique(meth$null.effect)
#
meth$title <- makeunique(meth$title, "")
meth$complab <- makeunique(meth$complab, "")
meth$outclab <- makeunique(meth$outclab, "")
#
meth$label.e <- makeunique(meth$label.e, "")
meth$label.c <- makeunique(meth$label.c, "")
meth$label.left <- makeunique(meth$label.left, "")
meth$label.right <- makeunique(meth$label.right, "")
#
meth$print.subgroup.name <- makeunique(meth$print.subgroup.name, FALSE)
meth$sep.subgroup <- makeunique(meth$sep.subgroup, "")
meth$warn <- makeunique(meth$warn, FALSE)
#
meth$backtransf <-
if (missing.backtransf)
any(meth$backtransf)
else
backtransf
meth$pscale <- makeunique(meth$pscale, 1)
meth$irscale <- makeunique(meth$irscale, 1)
meth$irunit <- makeunique(meth$irunit, "")
#
if (!(any(common) | any(random) | any(prediction))) {
warning("No results to bind.", call. = FALSE)
return(NULL)
}
#
# Name of meta-analysis object
#
if (missing.name) {
name <- name.i
name[name == ""] <- paste0("meta", seq.meta[name == ""])
}
else
chklength(name, length(is.subgroup),
text =
paste("Number of meta-analyses and names provided in",
"argument 'name' differ."))
#
# Names for meta-analyses must be unique
#
if (length(unique(name)) != length(name) & missing.subgroup) {
for (i in seq.meta)
if (name[i] %in% c("metabin", "metainc", "metaprop", "metarate") &
!is.trimfill[i])
name[i] <- paste(name[i], args[[i]]$method, sep = ".")
}
#
if (length(unique(name)) != length(name) & missing.subgroup) {
for (i in seq.meta)
if (random[i])
name[i] <- paste(name[i], args[[i]]$method.tau, sep = ".")
}
#
if (length(unique(name)) != length(name) & missing.subgroup)
name <- paste0("meta", seq.meta)
#
# Check if more than one common effect / random effects CI or PI is
# provided
#
for (i in seq.meta) {
if (length(args[[i]]$lower.common) > 1) {
ith <- min(i, 4)
stop("More than one result for common effect model provided in ",
i, switch(ith, "st", "nd", "rd", "th"),
" meta-analysis.",
call. = FALSE)
}
}
#
for (i in seq.meta) {
if (length(args[[i]]$lower.random) > 1) {
ith <- min(i, 4)
stop("More than one result for random effects model provided in ",
i, switch(ith, "st", "nd", "rd", "th"),
" meta-analysis.",
call. = FALSE)
}
}
#
for (i in seq.meta) {
if (length(args[[i]]$lower.predict) > 1) {
ith <- min(i, 4)
stop("More than one prediction interval provided in ",
i, switch(ith, "st", "nd", "rd", "th"),
" meta-analysis.",
call. = FALSE)
}
}
# #
# # Show individual results
# #
show.studies <- TRUE
overall.hetstat <- TRUE
# #
# if (length(unique(meth$method)) != 1 |
# length(unique(meth$method.tau)) != 1) {
# show.studies <- FALSE
# overall.hetstat <- FALSE
# }
#
#
# (3) Extract meta-analysis data
#
#
meta.list <- vector("list", n.meta)
#
if (with.subgroups)
for (i in seq.meta)
meta.list[[i]] <-
subgr2meta(args[[i]], common[i], random[i], prediction[i], name[i])
else {
for (i in seq.meta)
meta.list[[i]] <-
overall2meta(args[[i]], common[i], random[i], prediction[i], name[i])
}
#
meta <- list()
#
for (i in names(meta.list[[1]]))
meta[[i]] <- condense(meta.list, i)
#
#
# (4) Extract meta-analysis results and store them in subgroups
#
#
subgroup.list <- vector("list", n.meta)
#
for (i in seq.meta)
subgroup.list[[i]] <- overall2subgr(args[[i]])
#
subgroup <- list()
#
for (i in names(subgroup.list[[1]]))
subgroup[[i]] <- condense(subgroup.list, i)
#
if (!with.subgroups) {
subgroup$subgroup.levels <- name
}
#
#
# (5) Extract study data
#
#
data.list <- vector("list", n.meta)
#
if (with.subgroups) {
for (i in seq.meta)
data.list[[i]] <-
subgr2data(args[[i]], common[i], random[i], prediction[i], name[i])
}
else {
if (!is.null(subgroup.meta)) {
if (length(subgroup.meta == 1))
subgroup.meta <- rep_len(subgroup.meta, n.meta)
else if (length(subgroup.meta) != n.meta)
stop("Argument 'subgroup' must be of same length as ",
"number of meta-analysis results.",
call. = FALSE)
}
else
subgroup.meta <- rep("", n.meta)
#
for (i in seq.meta)
data.list[[i]] <-
overall2data(args[[i]], common[i], random[i], prediction[i], name[i],
subgroup.meta[[i]])
}
#
data <- do.call("rbind", data.list)
#
if (!with.subgroups & all(subgroup.meta == ""))
data <- data[, names(data) != "subgroup"]
#
#
# (6) Generate meta-analysis object
#
#
res <- c(meta, meth, subgroup)
#
res$data <- data
#
res$common.meta <- common
res$random.meta <- random
res$prediction.meta <- prediction
#
if (!with.subgroups) {
res$n.harmonic.mean <- res$n.harmonic.mean.ma
res$t.harmonic.mean <- res$t.harmonic.mean.ma
}
else {
res$n.harmonic.mean.ma <- unique(res$n.harmonic.mean.ma)
res$t.harmonic.mean.ma <- unique(res$t.harmonic.mean.ma)
}
#
res$call <- match.call()
res$version <- packageDescription("meta")$Version
#
res$show.studies <- show.studies
#
res$with.subgroups <- with.subgroups
res$samedata <- samedata
#
res$method <- unique(res$method)
res$method.random <- unique(res$method.random)
res$method.common.ci <- unique(res$method.common.ci)
res$method.random.ci <- unique(res$method.random.ci)
res$three.level <- unique(res$three.level)
res$adhoc.hakn.ci <- unique(res$adhoc.hakn.ci)
res$tau.common <- unique(res$tau.common)
#
if (with.subgroups & samedata) {
sel.c <-
!duplicated(
data.frame(res$TE.common,
res$lower.common,
res$upper.common))
#
res$TE.common <- res$TE.common[sel.c]
res$seTE.common <- res$seTE.common[sel.c]
res$lower.common <- res$lower.common[sel.c]
res$upper.common <- res$upper.common[sel.c]
res$statistic.common <- res$statistic.common[sel.c]
res$pval.common <- res$pval.common[sel.c]
#
res$text.common <-
if (is.null(res$text.common))
gs("text.common")
else
res$text.common <- res$text.common[sel.c]
#
sel.r <-
!duplicated(
data.frame(res$TE.random,
res$lower.random,
res$upper.random))
#
res$TE.random <- res$TE.random[sel.r]
res$seTE.random <- res$seTE.random[sel.r]
res$lower.random <- res$lower.random[sel.r]
res$upper.random <- res$upper.random[sel.r]
res$statistic.random <- res$statistic.random[sel.r]
res$pval.random <- res$pval.random[sel.r]
#
res$text.random <-
if (is.null(res$text.random))
gs("text.random")
else
res$text.random <- res$text.random[sel.r]
#
res$df.random <- unique(res$df.random)
res$df.hakn <- unique(res$df.hakn)
res$df.kero <- unique(res$df.kero)
#
sel.p <-
!duplicated(
data.frame(res$lower.predict,
res$upper.predict))
#
res$lower.predict <- res$lower.predict[sel.p]
res$upper.predict <- res$upper.predict[sel.p]
#
res$text.predict <-
if (is.null(res$text.predict))
gs("text.predict")
else
res$text.predict[sel.p]
#
res$n.e <- unique(res$n.e)
res$n.c <- unique(res$n.c)
#
res$k <- unique(res$k)
res$k.study <- unique(res$k.study)
res$k.all <- unique(res$k.all)
res$k.TE <- unique(res$k.TE)
#
res$tau2 <- unique(res$tau2)
res$tau <- unique(res$tau)
#
res$detail.tau <- ""
#
res$H <- unique(res$H)
res$lower.H <- unique(res$lower.H)
res$upper.H <- unique(res$upper.H)
#
res$I2 <- unique(res$I2)
res$lower.I2 <- unique(res$lower.I2)
res$upper.I2 <- unique(res$upper.I2)
#
res$Rb <- unique(res$Rb)
res$lower.Rb <- unique(res$lower.Rb)
res$upper.Rb <- unique(res$upper.Rb)
#
res$subgroup.name <- "meta-analysis"
res$w.common <- res$w.random <- rep(0, length(res$TE))
res$w.common.w <- rep(0, length(res$TE.common.w))
res$w.random.w <- rep(0, length(res$TE.random.w))
res$lower.predict.w <- rep(NA, length(res$w.random.w))
res$upper.predict.w <- rep(NA, length(res$w.random.w))
#
res$Q.b.common <- unique(res$Q.b.common)
res$Q.b.random <- unique(res$Q.b.random)
res$df.Q.b <- unique(res$df.Q.b)
res$pval.Q.b.common <- unique(res$pval.Q.b.common)
res$pval.Q.b.random <- unique(res$pval.Q.b.random)
}
else {
res$TE.common <- res$seTE.common <-
res$lower.common <- res$upper.common <-
res$statistic.common <- res$pval.common <- NA
#
res$TE.random <- res$seTE.random <-
res$lower.random <- res$upper.random <-
res$statistic.random <- res$pval.random <- NA
#
res$lower.predict <- res$upper.predict <- NA
#
if (!with.subgroups) {
res$studlab <- res$data$studlab
res$TE <- res$data$TE
res$seTE <- res$data$seTE
res$lower <- res$data$lower
res$upper <- res$data$upper
res$statistic <- res$data$statistic
res$pval <- res$data$pval
res$zval <- res$data$statistic
#
res$w.common <- res$w.random <- rep_len(NA, length(res$studlab))
#
res$text.w.common <- res$text.w.random <- ""
}
}
#
if (all(is.na(res$tau.preset)))
res$tau.preset <- NULL
#
res$Q.w.common <- NA
res$Q.w.random <- NA
res$df.Q.w <- NA
res$pval.Q.w.common <- NA
res$pval.Q.w.random <- NA
#
res$is.limit.copas <- is.limit.copas
#
res$classes <- unique(as.vector(sapply(args, function(x) class(x))))
res$classes <- res$classes[res$classes != "meta"]
#
# Backward compatibility
#
res <- backward(res)
#
class(res) <- c("metabind", "meta")
res$list <- list(meth = meth, meta = meta, subgroup = subgroup, data = data)
res
}
Try the meta package in your browser
Any scripts or data that you put into this service are public.
meta documentation built on April 2, 2026, 5:07 p.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Defines functions metabind
Documented in metabind
#' Combine and summarize meta-analysis objects
#'
#' @description
#' This function can be used to combine meta-analysis objects and is,
#' for example, useful to summarize results of various meta-analysis
#' methods or to generate a forest plot with results of several
#' subgroup analyses.
#'
#' @param ... Any number of meta-analysis objects or a single list
#' with meta-analyses.
#' @param name An optional character vector providing descriptive
#' names for the meta-analysis objects.
#' @param subgroup An optional variable to generate a forest plot with
#' subgroups.
#' @param common A logical vector indicating whether results of common
#' effect model should be considered.
#' @param random A logical vector indicating whether results of random
#' effects model should be considered.
#' @param prediction A logical vector indicating whether results of
#' prediction intervals should be considered.
#' @param backtransf A logical indicating whether results should be
#' back transformed in printouts and plots. If
#' \code{backtransf=TRUE} (default), results for \code{sm="OR"} are
#' printed as odds ratios rather than log odds ratios, for example.
#' @param outclab Outcome label for all meta-analyis objects.
#' @param pooled Deprecated argument (replaced by \code{common} and
#' \code{random}.
#' @param warn.deprecated A logical indicating whether warnings should
#' be printed if deprecated arguments are used.
#'
#' @details
#' This function can be used to combine any number of meta-analysis
#' objects which is useful, for example, to summarize results of
#' various meta-analysis methods or to generate a forest plot with
#' results of several subgroup analyses (see Examples).
#'
#' Individual study results are not retained with \code{metabind} as
#' the function allows to combine meta-analyses from different data
#' sets (e.g., with randomised or observational studies). Individual
#' study results are retained with R function \code{\link{metamerge}}
#' which can be used to combine results of meta-analyses of the
#' same dataset.
#'
#' @return
#' An object of class \code{c("metabind", "meta")} with corresponding
#' generic functions (see \code{\link{meta-object}}).
#'
#' @author Guido Schwarzer \email{guido.schwarzer@@uniklinik-freiburg.de}
#'
#' @seealso \code{\link{metagen}}, \code{\link{forest.metabind}},
#' \code{\link{metamerge}}
#'
#' @examples
#' data(Fleiss1993cont)
#'
#' # Add some (fictitious) grouping variables:
#' #
#' Fleiss1993cont$age <- c(55, 65, 55, 65, 55)
#' Fleiss1993cont$region <- c("Europe", "Europe", "Asia", "Asia", "Europe")
#'
#' ma <- metacont(n.psyc, mean.psyc, sd.psyc, n.cont, mean.cont, sd.cont,
#' data = Fleiss1993cont, sm = "SMD")
#'
#' # Conduct two subgroup analyses
#' #
#' ma1 <- update(ma, subgroup = age, subgroup.name = "Age group")
#' ma2 <- update(ma, subgroup = region, subgroup.name = "Region")
#'
#' # Combine random effects subgroup meta-analyses and show forest
#' # plot with subgroup results
#' #
#' mb <- metabind(ma1, ma2, common = FALSE)
#' mb
#' forest(mb)
#'
#' # Use various estimation methods for between-study heterogeneity
#' # variance
#' #
#' ma.pm <- update(ma, method.tau = "PM")
#' ma.dl <- update(ma, method.tau = "DL")
#' ma.ml <- update(ma, method.tau = "ML")
#' ma.hs <- update(ma, method.tau = "HS")
#' ma.sj <- update(ma, method.tau = "SJ")
#' ma.he <- update(ma, method.tau = "HE")
#' ma.eb <- update(ma, method.tau = "EB")
#'
#' # Combine meta-analyses and show results
#' #
#' taus <- c("Restricted maximum-likelihood estimator",
#' "Paule-Mandel estimator",
#' "DerSimonian-Laird estimator",
#' "Maximum-likelihood estimator",
#' "Hunter-Schmidt estimator",
#' "Sidik-Jonkman estimator",
#' "Hedges estimator",
#' "Empirical Bayes estimator")
#' #
#' ma.taus <- metabind(ma, ma.pm, ma.dl, ma.ml, ma.hs, ma.sj, ma.he, ma.eb,
#' name = taus, common = FALSE)
#' ma.taus
#' forest(ma.taus)
#'
#' @export metabind
metabind <- function(..., subgroup = NULL,
name = NULL,
common = NULL, random = NULL, prediction = NULL,
backtransf = NULL, outclab = NULL,
pooled = NULL,
warn.deprecated = gs("warn.deprecated")) {
#
#
# (1) Check and set arguments
#
#
missing.subgroup <- missing(subgroup)
subgroup.meta <- subgroup
missing.name <- missing(name)
missing.pooled <- missing(pooled)
missing.backtransf <- missing(backtransf)
#
chklogical(warn.deprecated)
#
args <- list(...)
#
n.meta <- length(args)
seq.meta <- seq_len(n.meta)
#
if (n.meta == 1) {
if (inherits(args[[1]], "meta.rm5")) {
args <- args[[1]]
#
if (missing.name) {
name <- unlist(lapply(args, "[[" , "outclab"))
missing.name <- FALSE
}
}
#
else if (inherits(args[[1]], "copas"))
return(metamerge(args[[1]]))
#
else if (inherits(args[[1]], "limitmeta"))
return(metamerge(args[[1]]))
#
else if (inherits(args[[1]], "netpairwise"))
stop("Elements of argument '...' may not be of class 'netpairwise'.",
call. = FALSE)
#
else if (inherits(args[[1]], "meta"))
return(args[[1]])
#
else if (!is.list(args[[1]]))
stop("All elements of argument '...' must be of class 'meta', ",
"'limitmeta', or 'copas'.",
call. = FALSE)
else {
n.meta <- length(args[[1]])
seq.meta <- seq_len(n.meta)
#
args2 <- list()
for (i in seq.meta)
args2[[i]] <- args[[1]][[i]]
#
args <- args2
}
}
#
if (!missing.pooled) {
pooled <- setchar(pooled, c("common", "random", "fixed"))
pooled[pooled == "fixed"] <- "common"
#
pooled.common <- pooled == "common"
pooled.random <- pooled == "random"
}
#
common <-
deprecated2(common, missing(common), pooled.common, missing.pooled,
warn.deprecated, "pooled")
#
if (!is.null(common)) {
chklogical(common[1], "common")
#
if (length(common) == 1)
common <- rep_len(common, n.meta)
else
chklength(common, n.meta,
text = paste("Length of argument 'common' differs from",
"number of meta-analyses."))
}
#
random <-
deprecated2(random, missing(random), pooled.random, missing.pooled,
warn.deprecated, "pooled")
#
if (!is.null(random)) {
chklogical(random[1], "random")
#
if (length(random) == 1)
random <- rep_len(random, n.meta)
else
chklength(random, n.meta,
text = paste("Length of argument 'random' differs from",
"number of meta-analyses."))
}
#
if (!missing(prediction)) {
if (length(prediction) == 1)
prediction <- rep_len(prediction, n.meta)
#
if (!is.logical(prediction))
stop("Argument 'prediction' must contain logical values.",
call. = TRUE)
#
chklength(prediction, n.meta,
text = paste("Length of argument 'prediction' differs from",
"number of meta-analyses."))
}
#
if (!missing.backtransf)
chklogical(backtransf)
#
# Act on limitmeta and copas objects
#
name.i <- vector("character", n.meta)
#
is.limit <- is.copas <- is.trimfill <- is.subgroup <- rep(FALSE, n.meta)
samedata <- rep(NA, n.meta)
#
for (i in seq.meta) {
if (inherits(args[[i]], "metabind"))
stop("Elements of argument '...' may not be of class 'metabind'.",
call. = FALSE)
#
if (inherits(args[[i]], "netpairwise"))
stop("Elements of argument '...' may not be of class 'netpairwise'.",
call. = FALSE)
#
if (inherits(args[[i]], "meta")) {
if (inherits(args[[i]], "metamerge"))
stop("Meta-analysis objects created with metamerge() cannot ",
"be used in metabind().",
call. = FALSE)
#
if (!inherits(args[[i]], "metaadd"))
args[[i]] <- updateversion(args[[i]])
#
if (missing.name) {
if (inherits(args[[i]], "trimfill")) {
is.trimfill[i] <- TRUE
#
name.i[i] <- "trimfill"
}
else
name.i[i] <- replaceNULL(args[[i]]$subgroup.name, "")
#
if (name.i[i] == "")
name.i[i] <- class(args[[i]])[1]
}
}
#
else if (inherits(args[[i]], "copas")) {
args[[i]] <-
metamerge(update(args[[i]]$x, common = FALSE, random = FALSE),
args[[i]], label2 = "copas")
#
is.copas[i] <- TRUE
#
if (missing.name)
name.i[i] <- "copas"
}
#
else if (inherits(args[[i]], "limitmeta")) {
args[[i]] <-
metamerge(update(args[[i]]$x, common = FALSE, random = FALSE),
args[[i]], label2 = "limit")
#
is.limit[i] <- TRUE
#
if (missing.name)
name.i[i] <- "limitmeta"
}
#
else
stop("All elements of argument '...' must be of class 'meta', ",
"'limitmeta', or 'copas'.",
call. = FALSE)
#
is.subgroup[i] <- !is.null(args[[i]]$subgroup)
samedata[i] <- !samedata(args[[1]], args[[i]], stop = FALSE)
}
#
# Meta-analyses must all contain subgroups or none
#
if (length(unique(is.subgroup)) != 1)
stop("All or none meta-analyses must contain subgroups.",
call. = FALSE)
#
with.subgroups <- any(is.subgroup)
is.limit.copas <- is.limit | is.copas
#
samedata[is.limit.copas] <- TRUE
samedata <- all(samedata)
#
#
# (2) Extract meta-analytical methods
#
#
meth.list <- vector("list", n.meta)
#
for (i in seq.meta)
meth.list[[i]] <-
meta2meth(args[[i]], outclab)
#
meth <- list()
#
for (i in names(meth.list[[1]]))
meth[[i]] <- condense(meth.list, i)
#
meth$common <- replaceNULL(common, meth$common)
meth$common[is.limit.copas] <- FALSE
meth$random <- replaceNULL(random, meth$random)
meth$prediction <- replaceNULL(prediction, meth$prediction)
#
# Use common effect estimate if no result is selected
#
meth$common[!(meth$common | meth$random | meth$prediction)] <- TRUE
#
# Check whether settings are unique
#
meth2 <- meth[c("sm", "level.ma", "level.predict", "null.effect")]
n.meth <- lapply(meth2, function(x) length(unique(x)))
#
if (any(n.meth != 1))
stop("Setting for the following argument",
if (sum(n.meth != 1) > 1) "s",
" must be the same for all meta-analyses: ",
paste0(paste0("'", names(meth2)[n.meth != 1], "'"),
collapse = " - "))
#
# Unify settings
#
meth$sm <- makeunique(meth$sm)
meth$method <- unique(meth$method)
meth$method.random <- unique(meth$method.random)
meth$three.level <- makeunique(meth$three.level, FALSE)
#
meth$level <- meth$level.ma <- makeunique(meth$level.ma)
meth$level.predict <- makeunique(meth$level.predict)
#
common <- meth$common
random <- meth$random
prediction <- meth$prediction
#
meth$overall <- with.subgroups & samedata
meth$overall.hetstat <- with.subgroups & samedata
#
if (missing.pooled & all(meth$common) & all(!meth$random))
meth$common <- pooled == "onlycommon"
else
meth$common <- any(meth$common)
#
meth$random <- any(meth$random)
#
meth$prediction <- any(meth$prediction)
#
meth$method.common.ci <- unique(meth$method.common.ci)
meth$method.random.ci <- unique(meth$method.random.ci)
meth$adhoc.hakn.ci <- unique(meth$adhoc.hakn.ci)
meth$method.tau <- unique(meth$method.tau)
meth$tau.preset <- unique(meth$tau.preset)
meth$TE.tau <- unique(meth$TE.tau)
meth$tau.common <-
if (!with.subgroups)
FALSE
else
unique(meth$tau.common)
#
meth$method.I2 <- unique(meth$method.I2)
#
meth$prediction.subgroup <-
if (!with.subgroups)
FALSE
else
unique(meth$prediction.subgroup)
meth$method.predict <- unique(meth$method.predict)
meth$adhoc.hakn.pi <- unique(meth$adhoc.hakn.pi)
#
meth$method.bias <- unique(meth$method.bias)
meth$null.effect <- makeunique(meth$null.effect)
#
meth$title <- makeunique(meth$title, "")
meth$complab <- makeunique(meth$complab, "")
meth$outclab <- makeunique(meth$outclab, "")
#
meth$label.e <- makeunique(meth$label.e, "")
meth$label.c <- makeunique(meth$label.c, "")
meth$label.left <- makeunique(meth$label.left, "")
meth$label.right <- makeunique(meth$label.right, "")
#
meth$print.subgroup.name <- makeunique(meth$print.subgroup.name, FALSE)
meth$sep.subgroup <- makeunique(meth$sep.subgroup, "")
meth$warn <- makeunique(meth$warn, FALSE)
#
meth$backtransf <-
if (missing.backtransf)
any(meth$backtransf)
else
backtransf
meth$pscale <- makeunique(meth$pscale, 1)
meth$irscale <- makeunique(meth$irscale, 1)
meth$irunit <- makeunique(meth$irunit, "")
#
if (!(any(common) | any(random) | any(prediction))) {
warning("No results to bind.", call. = FALSE)
return(NULL)
}
#
# Name of meta-analysis object
#
if (missing.name) {
name <- name.i
name[name == ""] <- paste0("meta", seq.meta[name == ""])
}
else
chklength(name, length(is.subgroup),
text =
paste("Number of meta-analyses and names provided in",
"argument 'name' differ."))
#
# Names for meta-analyses must be unique
#
if (length(unique(name)) != length(name) & missing.subgroup) {
for (i in seq.meta)
if (name[i] %in% c("metabin", "metainc", "metaprop", "metarate") &
!is.trimfill[i])
name[i] <- paste(name[i], args[[i]]$method, sep = ".")
}
#
if (length(unique(name)) != length(name) & missing.subgroup) {
for (i in seq.meta)
if (random[i])
name[i] <- paste(name[i], args[[i]]$method.tau, sep = ".")
}
#
if (length(unique(name)) != length(name) & missing.subgroup)
name <- paste0("meta", seq.meta)
#
# Check if more than one common effect / random effects CI or PI is
# provided
#
for (i in seq.meta) {
if (length(args[[i]]$lower.common) > 1) {
ith <- min(i, 4)
stop("More than one result for common effect model provided in ",
i, switch(ith, "st", "nd", "rd", "th"),
" meta-analysis.",
call. = FALSE)
}
}
#
for (i in seq.meta) {
if (length(args[[i]]$lower.random) > 1) {
ith <- min(i, 4)
stop("More than one result for random effects model provided in ",
i, switch(ith, "st", "nd", "rd", "th"),
" meta-analysis.",
call. = FALSE)
}
}
#
for (i in seq.meta) {
if (length(args[[i]]$lower.predict) > 1) {
ith <- min(i, 4)
stop("More than one prediction interval provided in ",
i, switch(ith, "st", "nd", "rd", "th"),
" meta-analysis.",
call. = FALSE)
}
}
# #
# # Show individual results
# #
show.studies <- TRUE
overall.hetstat <- TRUE
# #
# if (length(unique(meth$method)) != 1 |
# length(unique(meth$method.tau)) != 1) {
# show.studies <- FALSE
# overall.hetstat <- FALSE
# }
#
#
# (3) Extract meta-analysis data
#
#
meta.list <- vector("list", n.meta)
#
if (with.subgroups)
for (i in seq.meta)
meta.list[[i]] <-
subgr2meta(args[[i]], common[i], random[i], prediction[i], name[i])
else {
for (i in seq.meta)
meta.list[[i]] <-
overall2meta(args[[i]], common[i], random[i], prediction[i], name[i])
}
#
meta <- list()
#
for (i in names(meta.list[[1]]))
meta[[i]] <- condense(meta.list, i)
#
#
# (4) Extract meta-analysis results and store them in subgroups
#
#
subgroup.list <- vector("list", n.meta)
#
for (i in seq.meta)
subgroup.list[[i]] <- overall2subgr(args[[i]])
#
subgroup <- list()
#
for (i in names(subgroup.list[[1]]))
subgroup[[i]] <- condense(subgroup.list, i)
#
if (!with.subgroups) {
subgroup$subgroup.levels <- name
}
#
#
# (5) Extract study data
#
#
data.list <- vector("list", n.meta)
#
if (with.subgroups) {
for (i in seq.meta)
data.list[[i]] <-
subgr2data(args[[i]], common[i], random[i], prediction[i], name[i])
}
else {
if (!is.null(subgroup.meta)) {
if (length(subgroup.meta == 1))
subgroup.meta <- rep_len(subgroup.meta, n.meta)
else if (length(subgroup.meta) != n.meta)
stop("Argument 'subgroup' must be of same length as ",
"number of meta-analysis results.",
call. = FALSE)
}
else
subgroup.meta <- rep("", n.meta)
#
for (i in seq.meta)
data.list[[i]] <-
overall2data(args[[i]], common[i], random[i], prediction[i], name[i],
subgroup.meta[[i]])
}
#
data <- do.call("rbind", data.list)
#
if (!with.subgroups & all(subgroup.meta == ""))
data <- data[, names(data) != "subgroup"]
#
#
# (6) Generate meta-analysis object
#
#
res <- c(meta, meth, subgroup)
#
res$data <- data
#
res$common.meta <- common
res$random.meta <- random
res$prediction.meta <- prediction
#
if (!with.subgroups) {
res$n.harmonic.mean <- res$n.harmonic.mean.ma
res$t.harmonic.mean <- res$t.harmonic.mean.ma
}
else {
res$n.harmonic.mean.ma <- unique(res$n.harmonic.mean.ma)
res$t.harmonic.mean.ma <- unique(res$t.harmonic.mean.ma)
}
#
res$call <- match.call()
res$version <- packageDescription("meta")$Version
#
res$show.studies <- show.studies
#
res$with.subgroups <- with.subgroups
res$samedata <- samedata
#
res$method <- unique(res$method)
res$method.random <- unique(res$method.random)
res$method.common.ci <- unique(res$method.common.ci)
res$method.random.ci <- unique(res$method.random.ci)
res$three.level <- unique(res$three.level)
res$adhoc.hakn.ci <- unique(res$adhoc.hakn.ci)
res$tau.common <- unique(res$tau.common)
#
if (with.subgroups & samedata) {
sel.c <-
!duplicated(
data.frame(res$TE.common,
res$lower.common,
res$upper.common))
#
res$TE.common <- res$TE.common[sel.c]
res$seTE.common <- res$seTE.common[sel.c]
res$lower.common <- res$lower.common[sel.c]
res$upper.common <- res$upper.common[sel.c]
res$statistic.common <- res$statistic.common[sel.c]
res$pval.common <- res$pval.common[sel.c]
#
res$text.common <-
if (is.null(res$text.common))
gs("text.common")
else
res$text.common <- res$text.common[sel.c]
#
sel.r <-
!duplicated(
data.frame(res$TE.random,
res$lower.random,
res$upper.random))
#
res$TE.random <- res$TE.random[sel.r]
res$seTE.random <- res$seTE.random[sel.r]
res$lower.random <- res$lower.random[sel.r]
res$upper.random <- res$upper.random[sel.r]
res$statistic.random <- res$statistic.random[sel.r]
res$pval.random <- res$pval.random[sel.r]
#
res$text.random <-
if (is.null(res$text.random))
gs("text.random")
else
res$text.random <- res$text.random[sel.r]
#
res$df.random <- unique(res$df.random)
res$df.hakn <- unique(res$df.hakn)
res$df.kero <- unique(res$df.kero)
#
sel.p <-
!duplicated(
data.frame(res$lower.predict,
res$upper.predict))
#
res$lower.predict <- res$lower.predict[sel.p]
res$upper.predict <- res$upper.predict[sel.p]
#
res$text.predict <-
if (is.null(res$text.predict))
gs("text.predict")
else
res$text.predict[sel.p]
#
res$n.e <- unique(res$n.e)
res$n.c <- unique(res$n.c)
#
res$k <- unique(res$k)
res$k.study <- unique(res$k.study)
res$k.all <- unique(res$k.all)
res$k.TE <- unique(res$k.TE)
#
res$tau2 <- unique(res$tau2)
res$tau <- unique(res$tau)
#
res$detail.tau <- ""
#
res$H <- unique(res$H)
res$lower.H <- unique(res$lower.H)
res$upper.H <- unique(res$upper.H)
#
res$I2 <- unique(res$I2)
res$lower.I2 <- unique(res$lower.I2)
res$upper.I2 <- unique(res$upper.I2)
#
res$Rb <- unique(res$Rb)
res$lower.Rb <- unique(res$lower.Rb)
res$upper.Rb <- unique(res$upper.Rb)
#
res$subgroup.name <- "meta-analysis"
res$w.common <- res$w.random <- rep(0, length(res$TE))
res$w.common.w <- rep(0, length(res$TE.common.w))
res$w.random.w <- rep(0, length(res$TE.random.w))
res$lower.predict.w <- rep(NA, length(res$w.random.w))
res$upper.predict.w <- rep(NA, length(res$w.random.w))
#
res$Q.b.common <- unique(res$Q.b.common)
res$Q.b.random <- unique(res$Q.b.random)
res$df.Q.b <- unique(res$df.Q.b)
res$pval.Q.b.common <- unique(res$pval.Q.b.common)
res$pval.Q.b.random <- unique(res$pval.Q.b.random)
}
else {
res$TE.common <- res$seTE.common <-
res$lower.common <- res$upper.common <-
res$statistic.common <- res$pval.common <- NA
#
res$TE.random <- res$seTE.random <-
res$lower.random <- res$upper.random <-
res$statistic.random <- res$pval.random <- NA
#
res$lower.predict <- res$upper.predict <- NA
#
if (!with.subgroups) {
res$studlab <- res$data$studlab
res$TE <- res$data$TE
res$seTE <- res$data$seTE
res$lower <- res$data$lower
res$upper <- res$data$upper
res$statistic <- res$data$statistic
res$pval <- res$data$pval
res$zval <- res$data$statistic
#
res$w.common <- res$w.random <- rep_len(NA, length(res$studlab))
#
res$text.w.common <- res$text.w.random <- ""
}
}
#
if (all(is.na(res$tau.preset)))
res$tau.preset <- NULL
#
res$Q.w.common <- NA
res$Q.w.random <- NA
res$df.Q.w <- NA
res$pval.Q.w.common <- NA
res$pval.Q.w.random <- NA
#
res$is.limit.copas <- is.limit.copas
#
res$classes <- unique(as.vector(sapply(args, function(x) class(x))))
res$classes <- res$classes[res$classes != "meta"]
#
# Backward compatibility
#
res <- backward(res)
#
class(res) <- c("metabind", "meta")
res$list <- list(meth = meth, meta = meta, subgroup = subgroup, data = data)
res
}
Try the meta package in your browser
Any scripts or data that you put into this service are public.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.