The R package metaviz is a collection of functions to create visually appealing and information-rich plots of meta-analytic data using ggplot2. Currently functions to create several variants of forest plots (viz_forest
) and funnel plots (viz_funnel
, viz_sunset
) are provided. See the function documentations for more details and relevant references.
This vignette is a tutorial for the use of metaviz to visualize meta-analytic data and presents some of its main features.
In the following, we use four different example datasets distributed with the package metaviz. The datasets mozart
and homeopath
include published meta-analyses using the Cohen d metric (Pietschnig, Voracek, & Formann, 2010; Mathie et al. 2017), brainvol
contains data for a meta-analysis using correlation coefficients (Pietschnig et al., 2015), and exrehab
contains data of a published meta-analysis with dichotomous outcome data (Anderson et al. 2016). More details can be found in the respective help files (help(mozart)
, help(homeopath)
, help(brainvol)
, help(exrehab)
).
The main input of functions in package metaviz is a data.frame or matrix with the study effect sizes in the first column and the respective standard errors in the second column. Alternatively, the output of the function rma.uni
from R package metafor can be supplied as input, then effect sizes and standard errors are taken from there.
First, the R package metaviz needs to be installed and attached within the R environment.
The viz_forest
function is the main function in metaviz to create forest plots and their variants. Many options to visually customize the forest plot are provided (see help(viz_forest)
)
viz_forest(x = mozart[1:10, c("d", "se")], study_labels = mozart[1:10, c("study_name")],
summary_label = "Summary effect", xlab = "Cohen d")
In addition to traditional forest plots, rainforest plots as well as thick forest plots can be created via the variant
argument. Alternatively, the wrapper functions viz_thickforest
and viz_rainforest
can be used with identical results. Rainforest and thick forest plots are two variants and enhancements of the classical forest plot recently proposed by Schild and Voracek (2015). Both variants visually emphasize large studies (with short confidence intervals and more weight in the meta-analysis), while small studies (with wide confidence intervals and less weight in the meta-analysis) are visually less dominant. For further details see help(viz_rainforest)
and help(viz_thickforest)
.
viz_forest(x = mozart[1:10, c("d", "se")], study_labels = mozart[1:10, c("study_name")],
summary_label = "Summary effect", xlab = "Cohen d", variant = "rain")
viz_forest(x = mozart[1:10, c("d", "se")], study_labels = mozart[1:10, c("study_name")],
summary_label = "Summary effect", xlab = "Cohen d", variant = "thick", method = "FE")
The method
argument controls the meta-analytic model (fixed effect or random effects model). Setting method
from a fixed effect to a random effects model changes the estimated summary effect and meta-analytic (inverse-variance) weights assigned to each study accordingly.
The viz_forest
function is able to use a categorical moderator variable to visualize a subgroup analysis. This is done via the group
argument, a factor which corresponds to the subgroup membership of each study. We use the dichotomous moderator rr_lab
to compute and visualize separate meta-analyses. In the case of subgroup analysis, the summary_label
argument can be a vector containing names for all subgroups, arranged in the order of the levels of group
.
Different aspects of meta-analytic data can be shown in forest plots. Within function viz_forest
the type
parameter controls which aspects are shown. Two examples are given below.
Argument type = "cumulative"
shows a cumulative meta-analysis, that is, meta-analytic summary effects are computed sequentially by adding each study one-by-one.
viz_forest(x = mozart[, c("d", "se")],
group = mozart[, "rr_lab"],
study_labels = mozart[, "study_name"],
summary_label = c("Summary (rr_lab = no)", "Summary (rr_lab = yes)"),
xlab = "Cohen d",
variant = "thick",
type = "cumulative")
Argument type = "sensitivity"
shows a leave-one-out analysis. That is, for each study the meta-analytic summary effect is shown if that particular study is not considered in the computation of the summary effect.
The viz_forest
function was developed to allow aligning a table to the forest plot containing study or summary information. First, textual annotations of effect sizes and confidence intervals can be optionally displayed with the argument annotate_CI = TRUE
.
viz_forest(x = mozart[1:10, c("d", "se")],
group = mozart[1:10, "rr_lab"],
study_labels = mozart[1:10, "study_name"],
summary_label = c("Summary (rr_lab = no)", "Summary (rr_lab = yes)"),
xlab = "Cohen d",
variant = "thick",
annotate_CI = TRUE)
Second, arbitrary study information can be supplied as data.frame with the argument study_table
. This data.frame can contain several variables to be aligned as columns and should contain one row for each study in the meta-analysis.
To illustrate, we might want to align a table with the study identifiers and number of events observed in each study of data set exrehab
.
study_table <- data.frame(
name = exrehab[, "study_name"],
eventsT = paste(exrehab$ai, "/", exrehab$ai + exrehab$bi, sep = ""),
eventsC = paste(exrehab$ci, "/", exrehab$ci + exrehab$di, sep = ""))
head(study_table)
## name eventsT eventsC
## 1 Shaw (NEDHP) 81 109/323 113/328
## 2 Lewin 92 9/58 18/58
## 3 Haskell 94 62/145 72/155
## 4 Englom 96 26/102 34/91
## 5 Hofman-Bang 99 19/46 21/41
## 6 Belardinelli 01 11/59 21/59
We might also want to include the sum of all events as summary information. This can be done with the summary_table
argument. summary_table
should be a data.frame with the number of rows equal to the number of subgroups.
summary_table <- data.frame(
name = "Summary",
eventsT = paste(sum(exrehab$ai), "/", sum(exrehab$ai + exrehab$bi), sep = ""),
eventsC = paste(sum(exrehab$ci), "/", sum(exrehab$ci + exrehab$di), sep = ""))
head(summary_table)
## name eventsT eventsC
## 1 Summary 407/1471 453/1388
The table_headers of study_table
and summary_table
can be specified with table_headers
.
viz_forest(x = exrehab[, c("logrr", "logrr_se")], variant = "classic",
col = "Greys", xlab = "logRR", annotate_CI = T,
study_table = study_table,
summary_table = summary_table,
table_headers = c("ID", "Events (T)", "Events (C)"))
The spacing of the forest plot and aligned tables can be customized by supplying a layout matrix via table_layout
.
viz_forest(x = exrehab[, c("logrr", "logrr_se")], variant = "classic",
col = "Greys", xlab = "logRR", x_limit = c(-0.35, 0.05),
annotate_CI = T, type = "sensitivity",
study_table = data.frame(left_out = exrehab[, "study_name"],
remaining_N = sum(exrehab[, "n1i"] + exrehab[, "n2i"]) -
(exrehab[, "n1i"] + exrehab[, "n2i"])),
summary_table = "None",
table_headers = c("Study left out", "N remaining", "log Risk Ratio [95% CI]"),
table_layout = matrix(c(1, 2, 3), nrow = 1))
For some transformed effect sizes (e.g., log odds ratios, log risk ratios, or Fisher’s z) it is good practice to transform the labels of the x-axis, such that they display the effect sizes on their original scale (e.g., odds, ratios, risk ratios, or correlations). This can be conveniently done with the argument x_trans_function
.
For forest plots and thick forest plots it is possible to individually customize the color of studies by supplying a vector of colors to the col
argument. The color of the summary effect(s) can be customized individually as well with the summary_col
argument.
viz_forest(x = mozart[1:10, c("d", "se")],
group = mozart[1:10, "rr_lab"],
study_labels = mozart[1:10, "study_name"],
summary_label = c("Summary (rr_lab = no)", "Summary (rr_lab = yes)"),
xlab = "Cohen d",
col = c("firebrick", "steelblue4")[mozart[1:10, "rr_lab"]],
summary_col = c("firebrick", "steelblue4"))
The function viz_funnel
is capable to create a large set of different funnel plot variants. Options for several graphical augmentations (e.g., confidence, significance, and additional evidence contours; choice of the ordinate; showing study subgroups), and different statistical information displayed are provided (Egger’s regression line, and imputed studies by, as well as the adjusted summary effect from, the trim-and-fill method). See help(viz_funnel)
for further details and relevant references.
By default, significance contours (for the 5% and 1% level) and 95% confidence contours (fixed effect model) are shown.
In the brainvol
meta-analysis Fisher’s z values are used. It is convenient to show Fisher’s z values on their original scale (correlations) using the x_trans_function
. The meta-analytic model can be changed to a random effects model using method
. Note that for a random effects meta-analysis, confidence contours do not necessarily converge to a single point for a standard error of zero.