Forest plots

Forest plots are useful when summarizing the invidivual impact of studies on the pooled end points for meta analysis or illustrating the influence of coefficients in a regression model. More importantly, they visualize the direction and magnitude of the associations between the individual studies or coefficients on the outcomes. The most basic forest plots include a point that is shaped as a diamond, box, or circle with error bars or whiskers to indicate the 95% confidence intervals (CI).

Here is an example of forest plots that I used for one of my past papers. These forest plots illustrate the impact of each coefficient in a logistic regression model and the 95% CIs. However, these forest plots were built using Stata, which is not avaialble to everyone. So, I decided to learn how to build forest plots using R.

Example of a forest plot

Example of a forest plot

Elements of a forest plot

There are several elements that are used in forest plots. We will use an example forest plot to identify the major elements. The diamond \(\blacklozenge\) represents the odds ratio or point estimate (Note: this can be another symbol other than a diamond, such as a circle or square). The error bars or whiskers \(\vdash\) \(\dashv\) represent the 95% CI of the odds ratio. The labels on the Y-axis represents the comparisions. For example, if the results were OR = 3.0; 95% CI: 2.25, 3.75, this means that High income earners have a 3-times greater odds of having an event than Low income earners (95% CI: 2.25, 3.75), which is statistically significant. The pink vertical line bisects the X-axis where the odds ratio is equal to 1 and represents the null. If any of the error bars cross the null, then the association is not statistically significant.

I provide the R code that I used to create this example. I use two libraries, gridExtra and ggplot2, both of which you can install using the following command: install.packages("gridExtra").

## Load libraries
library(gridExtra)
library(ggplot2)

Data

Data for the forest plot are generated as a data frame. (These data are not based on any real experiment, and they are only used for educational purposes.)

## Example data frame
dat <- data.frame(
  Index = c(1, 2, 3, 4), ## This provides an order to the data
  label = c("Age (65 and older versus <65)", "Male versus Female", "High income versus Low income", "High school or higher versus No High school"),
  OR = c(1.00, 2.00, 3.00, 0.50),
  LL = c(0.25, 0.90, 2.25, 0.2),
  UL = c(1.75, 3.10, 3.75, 0.8),
  CI = c("0.25, 1.75", "0.90, 3.10", "2.25, 3.75", "0.20, 0.80")
)
dat
##   Index                                       label  OR   LL   UL         CI
## 1     1               Age (65 and older versus <65) 1.0 0.25 1.75 0.25, 1.75
## 2     2                          Male versus Female 2.0 0.90 3.10 0.90, 3.10
## 3     3               High income versus Low income 3.0 2.25 3.75 2.25, 3.75
## 4     4 High school or higher versus No High school 0.5 0.20 0.80 0.20, 0.80

Forest plot

Then the forest plot is created using ggplot2.

## Plot forest plot
plot1 <- ggplot(dat, aes(y = Index, x = OR)) +
  geom_point(shape = 18, size = 5) +  
  geom_errorbarh(aes(xmin = LL, xmax = UL), height = 0.25) +
  geom_vline(xintercept = 1, color = "red", linetype = "dashed", cex = 1, alpha = 0.5) +
  scale_y_continuous(name = "", breaks=1:4, labels = dat$label, trans = "reverse") +
  xlab("Odds Ratio (95% CI)") + 
  ylab(" ") + 
  theme_bw() +
  theme(panel.border = element_blank(),
        panel.background = element_blank(),
        panel.grid.major = element_blank(), 
        panel.grid.minor = element_blank(), 
        axis.line = element_line(colour = "black"),
        axis.text.y = element_text(size = 12, colour = "black"),
        axis.text.x.bottom = element_text(size = 12, colour = "black"),
        axis.title.x = element_text(size = 12, colour = "black"))
plot1

Adding a table to the forest plot

Now that the forest plot has been constructed, we can add the text such as the "OR" and "95% CI" to the chart. To do this, we need to create separate tables that we will combine using grid.arrange. This of these tables a section or parts of a grid that we will put together. Another term that is used to described these parts is the grid graphical object or "grob".

Here is an example of how we will combine the grobs. Think of these are parts of a bigger picture.

First, we create the table base, which we call table_base. Next, we create the table that contains the "OR" values and call it tab1. The second table contains the "95% CI" for the point estimates, which we call tab2.

This part is a little tricky. One of the issues with using this method is that the values on the "OR" and "95% CI" may not be aligned with the values on the plot. I added the following to the table_base to help with this alignment:

axis.text.x = element_text(color="white", hjust = -3, size = 25)

This creates a text beneath the table base with font size = 25, but it's hidden because the colour is "white". I used a horizontal justification hjust = -3 to move it away from the plot's X-axis so that there would not be any overlays of the text.

## Create the table-base pallete
table_base <- ggplot(dat, aes(y=label)) +
  ylab(NULL) + xlab("  ") + 
  theme(plot.title = element_text(hjust = 0.5, size=12), 
        axis.text.x = element_text(color="white", hjust = -3, size = 25), ## This is used to help with alignment
        axis.line = element_blank(),
        axis.text.y = element_blank(), 
        axis.ticks = element_blank(),
        axis.title.y = element_blank(), 
        legend.position = "none",
        panel.background = element_blank(), 
        panel.border = element_blank(), 
        panel.grid.major = element_blank(),
        panel.grid.minor = element_blank(), 
        plot.background = element_blank())

## OR point estimate table
tab1 <- table_base + 
  labs(title = "space") +
  geom_text(aes(y = rev(Index), x = 1, label = sprintf("%0.1f", round(OR, digits = 1))), size = 4) + ## decimal places
  ggtitle("OR")

## 95% CI table
tab2 <- table_base +
  geom_text(aes(y = rev(Index), x = 1, label = CI), size = 4) + 
  ggtitle("95% CI")

Combining the elements of the forest plot

After we've created all the tables, we create the matrix. The matrix size will be defined as matrix(c(1,1,1,1,1,1,1,2,3,3). Finally, we combine the tables using grid.arrange.

## Merge tables with plot
lay <-  matrix(c(1,1,1,1,1,1,1,1,1,1,2,3,3), nrow = 1)
grid.arrange(plot1, tab1, tab2, layout_matrix = lay)

Conclusions

This exercise provided me with insights on how best to generate a forest plot. I also realize that there are other ways to generate forest plots in R. For example, there is the forestplot package. But I wanted to build a forest plot using ggplot2 and then combine data elements (e.g., OR and 95% CI) into a single figure that is good enough for publication in a journal. There are additional aesthesics that could be done in the future such as adding alternate colours between the rows. However, I'll save that for another time.

References

Learning how to create forest plots using R involved searching through forums and reading posts. I'm amazed at the community of R coders who selflessly share their expeirences, wisdom, and codes. They were an indispensible aid in my pursuit for constructing forest plots in R. Here are some posts that I used and drew inspiration from.

Creating forest plots:
https://stackoverflow.com/questions/40265494/ggplot-grobs-align-with-tablegrob

https://stackoverflow.com/questions/15420621/reproduce-table-and-plot-from-journal

https://www.selfmindsociety.com/post/a-forest-plot-in-ggplot2

Modifying elements in ggplot2:
https://rpubs.com/Mentors_Ubiqum/ggplot_remove_elements

Adding decimals to values in a table:
https://stackoverflow.com/questions/38369855/how-to-put-exact-number-of-decimal-places-on-label-ggplot-bar-chart