Forest plots are commonly used in medical research publications, especially in meta-analysis. It can also be used to report the coefficients and confidence intervals (CIs) of the regression models.

There are lots of packages out there that can be used to draw a forest plot. The most popular one is forestplot. Packages specialized for the meta-analysis, like meta, metafor, and rmeta. Some other packages, like ggforestplot use ggplot2 to draw a forest plot, it is not available on the CRAN yet.

The main differences between the forestploter from the other packages are:

Focus on the forest plot.
Treat the forest plot as a table, elements are aligned in rows and columns. The user has full control over what and how to display the forest plot contents.
Graphical parameters are controlled with a theme.
Allowing post-hoc plot editing.
CIs in multiple columns and by groups.

Basic forest plot

The layout of the forest plot is determined by the dataset provided. Please check out the other vignette if you want to change the text or background, add or insert text, add borders to cells, and edit the color of the CI in some cells.

Simple forest plot

The first step is to prepare a data.frame to be used as a basic layout of the forest plot. Column names of the data will be drawn as the header, and the contents inside the data will be displayed in the forest plot. One or multiple blank columns without any content (blanks) should be provided to draw a confidence interval. The width of the box to draw the CI is determined by the width of this column. Increase the number of spaces in the column to give more space to draw CI.

First, we need to get the data ready to plot.

library(grid)
library(forestploter)

# Read provided sample example data
dt <- read.csv(system.file("extdata", "example_data.csv", package = "forestploter"))

# Keep needed columns
dt <- dt[,1:6]

# Indent the subgroup if there is a number in the placebo column
dt$Subgroup <- ifelse(is.na(dt$Placebo), 
                      dt$Subgroup,
                      paste0("   ", dt$Subgroup))

# NA to blank or NA will be transformed to carachter.
dt$Treatment <- ifelse(is.na(dt$Treatment), "", dt$Treatment)
dt$Placebo <- ifelse(is.na(dt$Placebo), "", dt$Placebo)
dt$se <- (log(dt$hi) - log(dt$est))/1.96

# Add a blank column for the forest plot to display CI.
# Adjust the column width with space, and increase the number of spaces below 
# to have a larger area to draw the CI. 
dt$` ` <- paste(rep(" ", 20), collapse = " ")

# Create a confidence interval column to display
dt$`HR (95% CI)` <- ifelse(is.na(dt$se), "",
                             sprintf("%.2f (%.2f to %.2f)",
                                     dt$est, dt$low, dt$hi))
head(dt)
#>          Subgroup Treatment Placebo      est        low       hi        se
#> 1    All Patients       781     780 1.869694 0.13245636 3.606932 0.3352463
#> 2             Sex                         NA         NA       NA        NA
#> 3            Male       535     548 1.449472 0.06834426 2.830600 0.3414741
#> 4          Female       246     232 2.275120 0.50768005 4.042560 0.2932884
#> 5             Age                         NA         NA       NA        NA
#> 6          <65 yr       297     333 1.509242 0.67029394 2.348190 0.2255292
#>                                                   HR (95% CI)
#> 1                                         1.87 (0.13 to 3.61)
#> 2                                                            
#> 3                                         1.45 (0.07 to 2.83)
#> 4                                         2.28 (0.51 to 4.04)
#> 5                                                            
#> 6                                         1.51 (0.67 to 2.35)

The data we have above will be used as the basic layout of the forest plot. The example below shows how to draw a simple forest plot. A footnote was added as a demonstration.

p <- forest(dt[,c(1:3, 8:9)],
            est = dt$est,
            lower = dt$low, 
            upper = dt$hi,
            sizes = dt$se,
            ci_column = 4,
            ref_line = 1,
            arrow_lab = c("Placebo Better", "Treatment Better"),
            xlim = c(0, 4),
            ticks_at = c(0.5, 1, 2, 3),
            footnote = "This is the demo data. Please feel free to change\nanything you want.")

# Print plot
plot(p)

Change theme

Now we will use the same data above, and add a summary point. We also want to change the graphical parameters for the confidence interval and other parts of the plot. The theme of the forest plot can be adjusted with forest_theme function, check out the manual for more details.

dt_tmp <- rbind(dt[-1, ], dt[1, ])
dt_tmp[nrow(dt_tmp), 1] <- "Overall"
dt_tmp <- dt_tmp[1:11, ]
# Define theme

tm <- forest_theme(base_size = 10,
                   # Confidence interval point shape, line type/color/width
                   ci_pch = 15,
                   ci_col = "#762a83",
                   ci_fill = "black",
                   ci_alpha = 0.8,
                   ci_lty = 1,
                   ci_lwd = 1.5,
                   ci_Theight = 0.2, # Set a T end at the end of CI 
                   # Reference line width/type/color
                   refline_lwd = gpar(lwd = 1, lty = "dashed", col = "grey20"),
                   # Vertical line width/type/color
                   vertline_lwd = 1,
                   vertline_lty = "dashed",
                   vertline_col = "grey20",
                   # Change summary color for filling and borders
                   summary_fill = "#4575b4",
                   summary_col = "#4575b4",
                   # Footnote font size/face/color
                   footnote_gp = gpar(cex = 0.6, fontface = "italic", col = "blue"))
#> refline_lwd will be deprecated, use refline_gp instead.
#> Warning in numnotnull("lwd"): NAs introduced by coercion
#> Warning in numnotnull("lwd"): NAs introduced by coercion


pt <- forest(dt_tmp[,c(1:3, 8:9)],
            est = dt_tmp$est,
            lower = dt_tmp$low, 
            upper = dt_tmp$hi,
            sizes = dt_tmp$se,
            is_summary = c(rep(FALSE, nrow(dt_tmp)-1), TRUE),
            ci_column = 4,
            ref_line = 1,
            arrow_lab = c("Placebo Better", "Treatment Better"),
            xlim = c(0, 4),
            ticks_at = c(0.5, 1, 2, 3),
            footnote = "This is the demo data. Please feel free to change\nanything you want.",
            theme = tm)

# Print plot
plot(pt)

Text justification and background with theme

By default, all cells are left aligned. But it is possible to justify any cells in the forest plot by setting parameters in forest_theme. Set core=list(fg_params=list(hjust=0, x=0)) to left align content, and rowhead=list(fg_params=list(hjust=0.5, x=0.5) to center header. Set hjust=1 and x=0.9 to right align text. You can also change the justification of text by with edit_plot, see details in another vignette.

The same rule applies to change the background color by setting core=list(bg_params=list(fill = c("#edf8e9", "#c7e9c0", "#a1d99b"))). Change settings in core if you want to change graphical parameters of contents, colhead for header. Change settings in fg_params to modify the text, see parameters for textGrob() in grid package. Change bg_params to modify settings for background graphical parameters, see gpar() in grid package. You should pass parameters as a list. More details can be found here.

Provide a single value if you want cells to have the same justification or vector for each cell. As you can notice the second example justifies text by row using the provided vector, and the vector will be recycled.

dt <- dt[1:4, ]

# Header center and content right
tm <- forest_theme(core=list(fg_params=list(hjust = 1, x = 0.9),
                             bg_params=list(fill = c("#edf8e9", "#c7e9c0", "#a1d99b"))),
                   colhead=list(fg_params=list(hjust=0.5, x=0.5)))

p <- forest(dt[,c(1:3, 8:9)],
            est = dt$est,
            lower = dt$low, 
            upper = dt$hi,
            sizes = dt$se,
            ci_column = 4,
            title = "Header center content right",
            theme = tm)

# Print plot
plot(p)


# Mixed justification
tm <- forest_theme(core=list(fg_params=list(hjust=c(1, 0, 0, 0.5),
                                            x=c(0.9, 0.1, 0, 0.5)),
                             bg_params=list(fill = c("#f6eff7", "#d0d1e6", "#a6bddb", "#67a9cf"))),
                   colhead=list(fg_params=list(hjust=c(1, 0, 0, 0, 0.5),
                                               x=c(0.9, 0.1, 0, 0, 0.5))))

p <- forest(dt[,c(1:3, 8:9)],
            est = dt$est,
            lower = dt$low, 
            upper = dt$hi,
            sizes = dt$se,
            ci_column = 4,
            title = "Mixed justification",
            theme = tm)
plot(p)

Text parsing

Similar to the text justification, you can parse text in any cells. Parsing all text will remove the blanks in the data, this will also applies to the blank columns to draw the whisker.

# Check out the `plotmath` function for math expression.
dt <- data.frame(
  Study = c("Study ~1^a", "Study ~2^b", "NO[2]"),
  low = c(0.2, -0.03, 1.11),
  est = c(0.71, 0.35, 1.79),
  hi = c(1.22, 0.74, 2.47)
)

dt$SMD <- sprintf("%.2f (%.2f, %.2f)", dt$est, dt$low, dt$hi)
dt$` ` <- paste(rep(" ", 20), collapse = " ")

fig_dt <- dt[,c(1,5:6)]

# Get a matrix of which row and columns to parse
parse_mat <- matrix(FALSE, 
                    nrow = nrow(fig_dt),
                    ncol = ncol(fig_dt))

# Here we want to parse the first column only, you can amend this to whatever you want.
parse_mat[,1] <- TRUE  

# Remove this fi you don't want parse the column head.
tm <- forest_theme(colhead=list(fg_params = list(parse=TRUE)), 
                   core=list(fg_params = list(parse=parse_mat)))

p <- forest(fig_dt,
       est = dt$est,
       lower = dt$low,
       upper = dt$hi,
       ci_column = 3,
       theme = tm)

# Add customised footnote.
# Due to the limitation of the textGrob, passing a parsed text with linebreak 
# has some issue. We use different approach here.
txt <- "<sup>a</sup> This is study A<br><sup>b</sup> This is study B"

add_grob(p, 
         row = 4, 
         col = 1:2,
         order = "background",
         gb_fn = gridtext::richtext_grob,
         text = txt,
         gp = gpar(fontsize = 8),
         hjust = 0, vjust = 1, halign = 0, valign = 1,
         x = unit(0, "npc"), y = unit(1, "npc"))

Multiple CI columns

Sometimes one may want to have multiple CI columns, each column may represent a different outcome. If this is the case, one only needs to provide a vector of the position of the columns to be drawn in the data. If the number of columns provided to draw the CI columns are same as the number of est, one CI will be drawn into each CI column. If the number of columns provided is less than the number of est, the extra est will be considered as a group and will be drawn to the CI columns sequentially. In the latter case, the group number equals to number of est divided by the number of ci_column and multiple columns will be drawn into one cell. As seen in the example below, the CI will be drawn in columns 3 and 5. The first and second elements in est, lower, and upper will be drawn in columns 3 and column 5.

In a multiple groups example, two or more CI in one cell. The solution is simple, provide all the values sequentially to est, lower, and upper. This means that the first n elements in the est, lower, and upper are considered as the same group, same for the next n elements. The n is determined by the number of ci_column. As it is shown in the example below, est_gp1 and est_gp2 will be drawn in column 3 and column 5, considered as group 1. The est_gp3 and est_gp4 will be drawn in column 3 and column 5, considered as group 2.

This is an example of multiple CI columns and groups:

dt <- read.csv(system.file("extdata", "example_data.csv", package = "forestploter"))
dt <- dt[1:7, ]
# Indent the subgroup if there is a number in the placebo column
dt$Subgroup <- ifelse(is.na(dt$Placebo), 
                      dt$Subgroup,
                      paste0("   ", dt$Subgroup))

# NA to blank or NA will be transformed to carachter.
dt$n1 <- ifelse(is.na(dt$Treatment), "", dt$Treatment)
dt$n2 <- ifelse(is.na(dt$Placebo), "", dt$Placebo)

# Add two blank columns for CI
dt$`CVD outcome` <- paste(rep(" ", 20), collapse = " ")
dt$`COPD outcome` <- paste(rep(" ", 20), collapse = " ")

# Generate point estimation and 95% CI. Paste two CIs together and separate by line break.
dt$ci1 <- paste(sprintf("%.1f (%.1f, %.1f)", dt$est_gp1, dt$low_gp1, dt$hi_gp1),
                sprintf("%.1f (%.1f, %.1f)", dt$est_gp3, dt$low_gp3, dt$hi_gp3),
                sep = "\n")
dt$ci1[grepl("NA", dt$ci1)] <- "" # Any NA to blank

dt$ci2 <- paste(sprintf("%.1f (%.1f, %.1f)", dt$est_gp2, dt$low_gp2, dt$hi_gp2),
                sprintf("%.1f (%.1f, %.1f)", dt$est_gp4, dt$low_gp4, dt$hi_gp4),
                sep = "\n")
dt$ci2[grepl("NA", dt$ci2)] <- ""

# Set-up theme
tm <- forest_theme(base_size = 10,
                   refline_lty = "solid",
                   ci_pch = c(15, 18),
                   ci_col = c("#377eb8", "#4daf4a"),
                   footnote_gp = gpar(col = "blue"),
                   legend_name = "Group",
                   legend_value = c("Trt 1", "Trt 2"),
                   vertline_lty = c("dashed", "dotted"),
                   vertline_col = c("#d6604d", "#bababa"),
                   # Table cell padding, width 4 and heights 3
                   core = list(padding = unit(c(4, 3), "mm")))
#> refline_lty will be deprecated, use refline_gp instead.

p <- forest(dt[,c(1, 19, 23, 21, 20, 24, 22)],
            est = list(dt$est_gp1,
                       dt$est_gp2,
                       dt$est_gp3,
                       dt$est_gp4),
            lower = list(dt$low_gp1,
                         dt$low_gp2,
                         dt$low_gp3,
                         dt$low_gp4), 
            upper = list(dt$hi_gp1,
                         dt$hi_gp2,
                         dt$hi_gp3,
                         dt$hi_gp4),
            ci_column = c(4, 7),
            ref_line = 1,
            vert_line = c(0.5, 2),
            nudge_y = 0.4,
            theme = tm)

plot(p)

It is obvious that the forest uses whatever you provided as the skeleton of the forest plot. You can use your imagination and put whatever you want in a cell, including linebreaks. Please check out the other vignette to modify the alignment of the text.

Different parameters for different CI columns

If the desired forest plot is multiple columns, some may want to have different settings for different columns. For example, different CI column has different xlim, x-axis ticks, x-axis labels, x_trans, reference line, vertical line, or arrow labels. This can be easily done by providing a list or vector. Provide a list for xlim, vert_line, arrow_lab, and ticks_at, atomic vector for xlab, x_trans and ref_line. See the example below.


dt$`HR (95% CI)` <- ifelse(is.na(dt$est_gp1), "",
                             sprintf("%.2f (%.2f to %.2f)",
                                     dt$est_gp1, dt$low_gp1, dt$hi_gp1))
dt$`Beta (95% CI)` <- ifelse(is.na(dt$est_gp2), "",
                             sprintf("%.2f (%.2f to %.2f)",
                                     dt$est_gp2, dt$low_gp2, dt$hi_gp2))

tm <- forest_theme(arrow_type = "closed",
                   arrow_label_just = "end")

p <- forest(dt[,c(1, 21, 23, 22, 24)],
            est = list(dt$est_gp1,
                       dt$est_gp2),
            lower = list(dt$low_gp1,
                         dt$low_gp2), 
            upper = list(dt$hi_gp1,
                         dt$hi_gp2),
            ci_column = c(2, 4),
            ref_line = c(1, 0),
            vert_line = list(c(0.3, 1.4), c(0.6, 2)),
            x_trans = c("log", "none"),
            arrow_lab = list(c("L1", "R1"), c("L2", "R2")),
            xlim = list(c(0, 3), c(-1, 3)),
            ticks_at = list(c(0.1, 0.5, 1, 2.5), c(-1, 0, 2)),
            xlab = c("OR", "Beta"),
            nudge_y = 0.2,
            theme = tm)

plot(p)

Custom CIs

It is possible to pass a custom CI drawing function to forest. The fn_ci accepts the CI drawing function for normal confidence interval and fn_summary for summary CI. Other parameters for those functions can be passed via forest. If you need to row values as est and lower to those functions, you need to define the name of the parameters you have passed via index_args. This is an advanced technique and not the purpose of this vignette is to show how to create a function to draw CI, but you can find some tutorial here if you are interested. Below is an example of the usage for a box plot CI with the built-in make_boxplot function.

# Function to calculate Box plot values
box_func <- function(x){
  iqr <- IQR(x)
  q3 <- quantile(x, probs = c(0.25, 0.5, 0.75), names = FALSE)
  c("min" = q3[1] - 1.5*iqr, "q1" = q3[1], "med" = q3[2],
    "q3" = q3[3], "max" = q3[3] + 1.5*iqr)
}
# Prepare data
val <- split(ToothGrowth$len, list(ToothGrowth$supp, ToothGrowth$dose))
val <- lapply(val, box_func)

dat <- do.call(rbind, val)
dat <- data.frame(Dose = row.names(dat),
                  dat, row.names = NULL)

dat$Box <- paste(rep(" ", 20), collapse = " ")

# Draw a single group box plot
tm <- forest_theme(ci_Theight = 0.2)

p <- forest(dat[,c(1, 7)],
            est = dat$med,
            lower = dat$min,
            upper = dat$max,
            # sizes = sizes,
            fn_ci = make_boxplot,
            ci_column = 2,
            lowhinge = dat$q1, 
            uphinge = dat$q3,
            hinge_height = 0.2,
            # values of the lowhinge and uphinge will be used as row values
            index_args = c("lowhinge", "uphinge"), 
            gp_box = gpar(fill = "black", alpha = 0.4),
            theme = tm
)
p

Saving plot

One can use the base method or use ggsave function to save the plot. For the ggsave function, please don’t ignore the plot parameter. The width and height should be tuned to get the desired plot. You can also set autofit=TRUE in the print or plot function to auto-fit the plot, but this may change and not be as compact as it should be.

# Base method
png('rplot.png', res = 300, width = 7.5, height = 7.5, units = "in")
p
dev.off()

# ggsave function
ggplot2::ggsave(filename = "rplot.png", plot = p,
                dpi = 300,
                width = 7.5, height = 7.5, units = "in")

Or you can get the width and height of the forestplot with get_wh, and use this width and height for saving.

# Get width and height
p_wh <- get_wh(plot = p, unit = "in")
png('rplot.png', res = 300, width = p_wh[1], height = p_wh[2], units = "in")
p
dev.off()

# Or get scale
get_scale <- function(plot,
                      width_wanted,
                      height_wanted,
                      unit = "in"){
  h <- convertHeight(sum(plot$heights), unit, TRUE)
  w <- convertWidth(sum(plot$widths), unit, TRUE)
  max(c(w/width_wanted,  h/height_wanted))
}
p_sc <- get_scale(plot = p, width_wanted = 6, height_wanted = 4, unit = "in")
ggplot2::ggsave(filename = "rplot.png", 
                plot = p,
                dpi = 300,
                width = 6, 
                height = 4,
                units = "in",
                scale = p_sc)

FAQs

Q: The whisker/CI plot area is too narrow, please help!

A: I have to admit that the vignettes were not well written, but you should be able to get the idea if you look at the vignette carefully and check the examples. Increase the widths by having more blank space in the column where the CI is to be drawn. Please check out the first example for how to do this.

Q: Can I modify the width and height of each row and column?

A: Yes, although the content of the data decides the heights and widths of the rows and columns, you can also modify these after you have finished plotting. See this here for details. You can also use core = list(padding = unit(c(4, 3), "mm")) in the forest_theme to add some padding to the width and height of each cell.

Q: How should I use weight for sizes?

A: The forest function will not transform the size thus it will be used as it is. If you want to weigh the size on your own, check here for some options

Q: How can I plot grouped forestplot?

A: You can leave a few blank lines to indicate the group break. You can also use arrangeGrob from gridExtra package or wrap_elements from patchwork to combine two or more forestplot.

Introduction to forestploter