%\VignetteEngine{knitr::knitr} %\VignetteIndexEntry{The forest.plot and table.plot functions} %\VignetteDepends{tidyverse, figuRes2}
The forest.plot, dot.plot and table.plot functions share some similarities. These are simple figures to describe, but labeling, idiosyncrasies in data structure, and aesthetic considerations complicate their construction. To get these figures to look appealing, one must have a good understanding of the incoming data structure and iterate towards a final product. These can be very time consuming in terms of design and execution.
This first example will demonstrate that pre-processing and post-processing are essential steps. First, start a session:
remove(list=ls())
require(figuRes2)
require(survival)
## Loading required package: survival
require(ggplot2)
require(scales)
require(stringr)
## Loading required package: stringr
require(plyr)
## Loading required package: plyr
require(grid)
require(gridExtra)
require(reshape2)
## Loading required package: reshape2
require(gtable)
## Loading required package: gtable
default.settings()
## my.path is set to:C:/Users/eri7441/AppData/Local/Temp/2/Rtmp8o65bH/Rbuild930512f5617/figuRes2/vignettes
## dd is set to:C:/Users/eri7441/AppData/Local/Temp/2/Rtmp8o65bH/Rbuild930512f5617/figuRes2/vignettes/dddata/
## cd is set to:C:/Users/eri7441/AppData/Local/Temp/2/Rtmp8o65bH/Rbuild930512f5617/figuRes2/vignettes/code/
## od is set to:C:/Users/eri7441/AppData/Local/Temp/2/Rtmp8o65bH/Rbuild930512f5617/figuRes2/vignettes/output/
## logd is set to:C:/Users/eri7441/AppData/Local/Temp/2/Rtmp8o65bH/Rbuild930512f5617/figuRes2/vignettes/log/
## The default page dimension is set to landscape: 11 inches wide by 8.5 inches tall.
## The default page left and right page margins: 0.75 inches and 0.75 inches, respectively.
## The default top an bottom margins: 0.9 inches and 1.25 inches, respectively.
## The region available for graphics/tables is 9.5 inches wide by 6.35 inches tall.
##
## The default theme: theme_bw
Suppose we are handed the following data set with the task of producing a 1x2 panel of 2 graphics - a forest plot on the left and a table plot on the left. Let's inspect the data:
data(forest.data)
working.df <- forest.data
head(working.df)
## n1 n2 e1 p1 e2 p2 hr low high intchi level
## 1 3252 3253 278 8.548585 308 9.468183 1.11 0.95 1.31 0.382 No
## 2 4656 4667 493 10.588488 502 10.756375 1.02 0.90 1.15 0.382 Yes
## 4 2002 1963 240 11.988012 232 11.818645 0.99 0.82 1.18 0.393 No
## 5 5906 5957 531 8.990857 578 9.702871 1.08 0.96 1.22 0.393 Yes
## 7 6706 6732 625 9.320012 663 9.848485 1.06 0.95 1.19 0.668 No
## 8 1202 1188 146 12.146423 147 12.373737 1.01 0.80 1.26 0.668 Yes
## subgroup
## 1 Qual. diag.: Prior MI
## 2 Qual. diag.: Prior MI
## 4 Qual. diag.: Prior Coronary Revas.
## 5 Qual. diag.: Prior Coronary Revas.
## 7 Qual. diag.: Multivessel CHD
## 8 Qual. diag.: Multivessel CHD
dim(working.df)
## [1] 89 12
We have a data.frame with 89 rows and 35 different subgroup analyses. Clearly, 89 line segments on a single page would be a bit too much. Suppose we plan to have 15-20 rows displayed on figures and therefore will ultimately need to partition this dataset into 5 or 6 smaller data.frames. For the present example then, we work with the first 16 rows. Determining how to divide up the remaining rows is an exercise left to the reader.
Our smaller working data.frame becomes:
working.df.1 <- working.df[1:16,]
working.df.1
## n1 n2 e1 p1 e2 p2 hr low high intchi level
## 1 3252 3253 278 8.548585 308 9.468183 1.11 0.95 1.31 0.382 No
## 2 4656 4667 493 10.588488 502 10.756375 1.02 0.90 1.15 0.382 Yes
## 4 2002 1963 240 11.988012 232 11.818645 0.99 0.82 1.18 0.393 No
## 5 5906 5957 531 8.990857 578 9.702871 1.08 0.96 1.22 0.393 Yes
## 7 6706 6732 625 9.320012 663 9.848485 1.06 0.95 1.19 0.668 No
## 8 1202 1188 146 12.146423 147 12.373737 1.01 0.80 1.26 0.668 Yes
## 10 1914 1906 178 9.299896 185 9.706191 1.05 0.85 1.29 0.972 Recent
## 11 5972 5999 592 9.912927 624 10.401734 1.05 0.94 1.18 0.972 Remote
## 13 2140 2131 200 9.345794 169 7.930549 0.84 0.69 1.03 0.014 No
## 14 5768 5789 571 9.899445 641 11.072724 1.13 1.01 1.26 0.014 Yes
## 16 5279 5198 441 8.353855 468 9.003463 1.08 0.95 1.23 0.522 No
## 17 2629 2722 330 12.552301 342 12.564291 1.01 0.87 1.18 0.522 Yes
## 19 5173 5218 481 9.298280 503 9.639709 1.04 0.91 1.17 0.664 No
## 20 2734 2698 289 10.570593 305 11.304670 1.08 0.92 1.27 0.664 Yes
## 22 6239 6286 601 9.632954 637 10.133630 1.06 0.95 1.18 0.708 No
## 23 1635 1593 166 10.152905 165 10.357815 1.01 0.82 1.25 0.708 Yes
## subgroup
## 1 Qual. diag.: Prior MI
## 2 Qual. diag.: Prior MI
## 4 Qual. diag.: Prior Coronary Revas.
## 5 Qual. diag.: Prior Coronary Revas.
## 7 Qual. diag.: Multivessel CHD
## 8 Qual. diag.: Multivessel CHD
## 10 Time from CHD event to randomization
## 11 Time from CHD event to randomization
## 13 CV risk factor: Age>=60 years
## 14 CV risk factor: Age>=60 years
## 16 CV risk factor: Diabetes req. pharm.
## 17 CV risk factor: Diabetes req. pharm.
## 19 CV risk factor: HDL-C <40 mg/dL
## 20 CV risk factor: HDL-C <40 mg/dL
## 22 CV risk factor: Current or previous smoker
## 23 CV risk factor: Current or previous smoker
Suppose this data.frame is sorted as we'd like to see it in the forest plot. (If not, accomplish this with additional pre-processing!) Namely, we'd like the top rows reporting line segments associated with Qual. diag.: Prior MI and the bottoms rows reporting line segments for CV risk factor: Current or previous smoker. The lower and upper endpoints of the line segments are associated with columns low and high containing the endpoints of 95% confidence intervals for the hazard ratio. The following items need to be added to the data.frame in order to make use of the forest.plot, table.plot and dot.plot functions. Columns need to be created for the following aspects of the graph:
First, we assign ranks for the line segments.
working.df.1$rank <- rev(1:16)
Next we assign a column for color. In this example, the color of all line segments will be the same, so we are creating a dummy column holding a factor with a single value. (In the next example, we'll see multiple colors.)
working.df.1$category <- factor(0)
In this example, each line segment will have a label associated with it. As such, the following step is superfluous; we could just as well reuse the rank column.
working.df.1$label.rank <- rev(1:16)
The actual labels to be used can be deduced from the data.frame. These will need to be a combination of values from subgroup and level columns.
working.df.1$labels <- paste(working.df.1$subgroup, working.df.1$level)
working.df.1
## n1 n2 e1 p1 e2 p2 hr low high intchi level
## 1 3252 3253 278 8.548585 308 9.468183 1.11 0.95 1.31 0.382 No
## 2 4656 4667 493 10.588488 502 10.756375 1.02 0.90 1.15 0.382 Yes
## 4 2002 1963 240 11.988012 232 11.818645 0.99 0.82 1.18 0.393 No
## 5 5906 5957 531 8.990857 578 9.702871 1.08 0.96 1.22 0.393 Yes
## 7 6706 6732 625 9.320012 663 9.848485 1.06 0.95 1.19 0.668 No
## 8 1202 1188 146 12.146423 147 12.373737 1.01 0.80 1.26 0.668 Yes
## 10 1914 1906 178 9.299896 185 9.706191 1.05 0.85 1.29 0.972 Recent
## 11 5972 5999 592 9.912927 624 10.401734 1.05 0.94 1.18 0.972 Remote
## 13 2140 2131 200 9.345794 169 7.930549 0.84 0.69 1.03 0.014 No
## 14 5768 5789 571 9.899445 641 11.072724 1.13 1.01 1.26 0.014 Yes
## 16 5279 5198 441 8.353855 468 9.003463 1.08 0.95 1.23 0.522 No
## 17 2629 2722 330 12.552301 342 12.564291 1.01 0.87 1.18 0.522 Yes
## 19 5173 5218 481 9.298280 503 9.639709 1.04 0.91 1.17 0.664 No
## 20 2734 2698 289 10.570593 305 11.304670 1.08 0.92 1.27 0.664 Yes
## 22 6239 6286 601 9.632954 637 10.133630 1.06 0.95 1.18 0.708 No
## 23 1635 1593 166 10.152905 165 10.357815 1.01 0.82 1.25 0.708 Yes
## subgroup rank category label.rank
## 1 Qual. diag.: Prior MI 16 0 16
## 2 Qual. diag.: Prior MI 15 0 15
## 4 Qual. diag.: Prior Coronary Revas. 14 0 14
## 5 Qual. diag.: Prior Coronary Revas. 13 0 13
## 7 Qual. diag.: Multivessel CHD 12 0 12
## 8 Qual. diag.: Multivessel CHD 11 0 11
## 10 Time from CHD event to randomization 10 0 10
## 11 Time from CHD event to randomization 9 0 9
## 13 CV risk factor: Age>=60 years 8 0 8
## 14 CV risk factor: Age>=60 years 7 0 7
## 16 CV risk factor: Diabetes req. pharm. 6 0 6
## 17 CV risk factor: Diabetes req. pharm. 5 0 5
## 19 CV risk factor: HDL-C <40 mg/dL 4 0 4
## 20 CV risk factor: HDL-C <40 mg/dL 3 0 3
## 22 CV risk factor: Current or previous smoker 2 0 2
## 23 CV risk factor: Current or previous smoker 1 0 1
## labels
## 1 Qual. diag.: Prior MI No
## 2 Qual. diag.: Prior MI Yes
## 4 Qual. diag.: Prior Coronary Revas. No
## 5 Qual. diag.: Prior Coronary Revas. Yes
## 7 Qual. diag.: Multivessel CHD No
## 8 Qual. diag.: Multivessel CHD Yes
## 10 Time from CHD event to randomization Recent
## 11 Time from CHD event to randomization Remote
## 13 CV risk factor: Age>=60 years No
## 14 CV risk factor: Age>=60 years Yes
## 16 CV risk factor: Diabetes req. pharm. No
## 17 CV risk factor: Diabetes req. pharm. Yes
## 19 CV risk factor: HDL-C <40 mg/dL No
## 20 CV risk factor: HDL-C <40 mg/dL Yes
## 22 CV risk factor: Current or previous smoker No
## 23 CV risk factor: Current or previous smoker Yes
We will return to fine tuning these labels in post-processing because of the need for mathematical symbol for less than or equal to; in the absence of this issue, an alternative attack would be to coerce the labels column into a factor and rename the levels at this stage.
p1 <- forest.plot(parent.df = working.df.1,
y.rank.col = "rank", # line segment's y-axis rank
Point.Est = "hr", # line segment's dot
lower.lim = "low", # line segment's lower endpoint
upper.lim = "high", # line segment's upper endpoint
y.label.rank.col = "label.rank", # label's y-axis rank
y.label.col = "labels", # label's text value
x.label = "Estimate",
y.label = NULL,
log.trans = TRUE,
x.limits = c(0.21, 5),
x.ticks = 2^(-2:2),
category.color = "category", # This colors the points and line segments
background.palette = c("red", "blue"),
category.palette = c("red", "blue"),
shape.palette = c(16, 16),
flip.palette = FALSE)
## y.limits are set to NULL; defaults are used.
## Warning: Ignoring unknown aesthetics: x
print(p1)
The following is a necessarily manual task.
p2 <- p1 + scale_y_continuous(
breaks = p1$data$LABEL.RANKS,
labels = c(
"Prior Myocardial Infaction: No",
"Yes",
"Prior coronary revasc.: No",
"Yes",
"Multivessel CHD: No",
"Yes",
"CHD Event Relative to Randomization: Recent",
"Remote",
expression(paste("Age ", phantom() >= 60,": No")),
"Yes",
"Diabetes req. pharm.: No",
"Yes",
"HDL-C < 40 mg/dL: No",
"Yes",
"Current or previous smoker: Yes",
"No"))
## Scale for 'y' is already present. Adding another scale for 'y', which will
## replace the existing scale.
print(p2)
We turn to the corresponding table plot.
t1 <- table.plot(
parent.df = working.df.1,
y.rank.col= "rank",
category.color= "category",
text.col1 = "hr",
text.col2 = "low",
text.col3 = "high",
text.col4 = NULL,
text.size = 3,
xtick.labs = c("Estimate", "LCI", "UCI"),
x.label= "Text",
y.label= "Item",
y.label.rank.col = "label.rank", # this identifies the y-axis values for labels
y.label.col = "subgroup",
category.palette = c("red", "blue"))
## y.limits are set to NULL; defaults are used.
## x.limits are set to NULL; defaults are used.
print(t1)
Since we're planning to juxtapose the table plot and the forest plot, we can suppress the labels here. In practice, it is worth verifying that labels in the forest and table plots agree before suppressing the labels. Note that arguments associated with y.label and y.label.rank.col are simply set to NULL. In addition teh x.label is set to white space. (A good exercise is to see what results when you step through the remainder of this exercise with NULL used in place of white space in the x.label argument.) Finally, the category.palette's first argument is changed from red to grey40.
t2 <- table.plot(
parent.df = working.df.1,
y.rank.col= "rank",
category.color= "category",
text.col1 = "hr",
text.col2 = "low",
text.col3 = "high",
text.col4 = NULL,
text.size=3,
xtick.labs = c("Estimate", "LCI", "UCI"),
x.label= "",
y.label=NULL,
y.label.rank.col = "label.rank", # this identifies the y-axis values for labels
y.label.col = NULL,
category.palette = c("grey40", "blue"))
## y.limits are set to NULL; defaults are used.
## x.limits are set to NULL; defaults are used.
print(t2)
Here's a first pass at assembling the forest plot figure, allocating 50% of available width to the forest plot graphic and table plot raphic.
build.page(interior.h = c(1),
interior.w = c(1/2, 1/2),
ncol=2, nrow=1, interior=list(p2+ggtitle(""), t2+ggtitle("")) )
annotate.page(override = "", title=list("Title Line 1", "","","",""))
Perhaps allocating more space for the figure and less space for the table would look better:
build.page(interior.h = c(1),
interior.w = c(.6, .4),
ncol=2, nrow=1, interior=list(p2+ggtitle(""), t2+ggtitle("")) )
annotate.page(override = "", title=list("Title Line 1", "","","",""))
Pushing to an extreme.
build.page(interior.h = c(1),
interior.w = c(.8, .2),
ncol=2, nrow=1, interior=list(p2+ggtitle(""), t2+ggtitle("")) )
annotate.page(override = "", title=list("Title Line 1", "","","",""))
Recall comments in previous sections about altering plot.margins to decrease the padding between p2 and t2.
Suppose we want to separate the subgroups a bit better. The user has control over this when defining the rank columns.
working.df.1$rank2 <- working.df.1$rank + duplicated(working.df.1$subgroup)*.5
p3 <- forest.plot(parent.df = working.df.1,
y.rank.col = "rank2", # line segment's y-axis rank
Point.Est = "hr", # line segment's dot
lower.lim = "low", # line segment's lower endpoint
upper.lim = "high", # line segment's upper endpoint
y.label.rank.col = "rank2", # label's y-axis rank
y.label.col = "labels", # label's text value
x.label = "Estimate",
y.label = NULL,
log.trans = TRUE,
x.limits = c(0.21, 5),
x.ticks = 2^(-2:2),
category.color = "category", # This colors the points and line segments
background.palette = c("red", "blue"),
category.palette = c("red", "blue"),
shape.palette = c(16, 16),
flip.palette = FALSE)
## y.limits are set to NULL; defaults are used.
## Warning: Ignoring unknown aesthetics: x
# This step is same as before, with swap in the breaks argument
p4 <- p3 + scale_y_continuous(
breaks = p3$data$RANK,
labels = c(
"Prior MI: No",
"Yes",
"Prior Coronary Revasc.: No",
"Yes",
"Multivessel CHD: No",
"Yes",
"CHD Event Relative to Randomization: Recent",
"Remote",
expression(paste("Age ", phantom() >= 60,": No")),
"Yes",
"Diabetes req. pharm.: No",
"Yes",
"HDL-C < 40 mg/dL: No",
"Yes",
"Current or previous smoker: No",
"Yes"))
## Scale for 'y' is already present. Adding another scale for 'y', which will
## replace the existing scale.
# This is same as t2, save swap of rank for rank2
t3 <- table.plot(
parent.df = working.df.1,
y.rank.col = "rank2",
category.color = "category",
text.col1 = "hr",
text.col2 = "low",
text.col3 = "high",
text.col4 = NULL,
text.size=3,
xtick.labs = c("Estimate", "LCI", "UCI"),
x.label= "",
y.label =NULL,
y.label.rank.col = "rank2", # this identifies the y-axis values for labels
y.label.col = NULL,
category.palette = c("grey40", "blue"))
## y.limits are set to NULL; defaults are used.
## x.limits are set to NULL; defaults are used.
build.page(interior.h = c(1),
interior.w = c(.8, .2),
ncol = 2, nrow = 1, interior=list(p4 + ggtitle(""), t3+ggtitle("")) )
annotate.page(override = "", title=list("Title Line 1", "","","",""))
