In CTU-Bern/accrualPlot: Accrual Plots and Predictions for Clinical Trials

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.width=7.5,
  fig.height=3
)

Following a trials recruitment is an important task for timing of analyses and ensuring that a trial will not run for too long (longer trials are more expensive). accrualPlot provides tools for easily creating recruitment plots and even for predicting when a trial will have successfully recruited all participants.

The package is loaded like any other:

library(accrualPlot)

The accrual_df

To work with accrualPlot, we need some data, specifically dates and, optionally, site identifiers. Here's some data that we will use in the following examples.

data(accrualdemo)
head(accrualdemo)

accrual_dfs are simply dataframes with the number of participants on each day participants are recruited (or site) began recruiting.

Monocentric trials

Monocentric trials obviously have only a single site, so we only need the x object we just created. We can pass this into the accrual_create_df function.

df <- accrual_create_df(accrualdemo$date)
print(df, head = TRUE)

In this case, the accrual_df has a single data frame.

Multicentric trials

For multicentric trials, we should also pass the site identifier to accrual_create_df in the by argument.

df2 <- accrual_create_df(accrualdemo$date, by = accrualdemo$site)
print(df2, head = TRUE)

In this case, the accrual_df is a list of dataframes, one for each site and an overall.

Start and end dates

By default, the start and end dates are defined based on the dates that you pass to accrual_create_df. You can override these via the start_date and current_date arguments. This is useful for when you have particularly slow recruiting trials (such as those with particularly strict inclusion criteria). For example, our fictitious example trial might have started recruiting on the 1st November. By adding this information, we modify other output

df3 <- accrual_create_df(accrualdemo$date, start_date = as.Date("2020-07-08"))

For multicentric trials where different sites started recruiting at different times, we can pass a vector to start_date

start_date<-as.Date(c("2020-07-09","2020-07-09","2020-08-01"))
names(start_date)<-c("Site 1","Site 2","Site 3")
df4 <- accrual_create_df(accrualdemo$date, by = accrualdemo$site, start_date = start_date)

Accrual plots

oldpar <- par(no.readonly = TRUE)

accrualPlot has three flavours of plots:
* Cumulative
* Absolute
* Prediction

and supplies both base graphics as well as ggplot2 graphics implementations (allowing easier modification).

Cumulative plots

Cumulative plots show a standard step function of the number of participants recruited up to a given point in time. The plots are produced via the plot method (which is a wrapper for the internal function accrual_plot_cum)

par(mfrow = c(1, 3))
plot(df)
plot(df2)
plot(df4)

For ggplot2 graphics, use the engine option:

library(patchwork)
library(ggplot2)
p1 <- plot(df, engine = "ggplot")
p2 <- plot(df2, engine = "ggplot") + 
   theme(axis.text.x = element_text(angle = 45, vjust = 1, hjust=1),
         axis.title.x = element_blank())
p3 <- plot(df4, engine = "ggplot") +
  labs(col = "Site") +
  theme_classic() +
  theme(legend.position = c(.35,.9),
     legend.key.height = unit(2, "mm"),
         legend.text=element_text(size=8),
         legend.title=element_blank(),
     axis.text.x = element_text(angle = 45, vjust = 1, hjust=1),
         axis.title.x = element_blank())
p1 + p2 + p3

Absolute recruitment

Recruitment plots per unit time can be obtained via the absolute method (specify which = "absolute" to plot)

par(mfrow = c(1, 3))
plot(df, which = "abs", unit = "week")
plot(df2, which = "abs", unit = "week", legend.list=list(x="topleft"), xlabsel=seq(1,20,by=2))
plot(df4, which = "abs", unit = "month",legend.list=list(x="topleft"))

Options for unit are year, month, week and day.

Where multiple sites exist, the different sites are indicated by different colours on the stacked bars.

p1 <- plot(df, which = "abs", unit = "week", engine = "ggplot")
p2 <- plot(df2, which = "abs", unit = "week", engine = "ggplot") + 
    theme(axis.text.x = element_text(angle = 45, vjust = 1, hjust=1),
         axis.title.x = element_blank())
p3 <- plot(df4, which = "abs", unit = "month", engine = "ggplot") +
  labs(fill = "Site") +
  theme_classic() +
  theme(legend.position = c(0.01,0.9),
     legend.justification = "left",
     legend.key.height = unit(2, "mm"),
     legend.key.width = unit(2, "mm"),
         legend.title=element_blank(),
     axis.text.x = element_text(angle = 45, vjust = 1, hjust = 1),
         axis.title.x = element_blank())
p1 + p2 + p3

It might be desirable to have panels for each site. This is easy to do with the ggplot implementation. The variable to use in this case is site, which is constructed in the appropriate plot function.

plot(df2, which = "abs", unit = "week", engine = "ggplot") + facet_wrap(~site)

Predictions

Date at which a target sample size is reached

In order predict the time point at which a certain number of participants has been recruited (for estimating when a study will be complete). If we want to recruit a total of 300 participants, we can put that in the target option.

par(mfrow = c(1, 3))
plot(df, which = "predict", target = 300, cex_prediction=0.9)
plot(df2, which = "predict", target = 300, cex_prediction=0.9)
plot(df4, which = "predict", target = 300, cex_prediction=0.9,  center_legend="strip")

We can also include site-specific targets:

plot(df4, which = "predict", target=c("Site 1"=160,"Site 2"=100,"Site 3"=40,"Overall"=300),
     show_center=FALSE)

Or with ggplot2.

p1 <- plot(df, which = "predict", target = 300, engine = "ggplot2") +
  theme(plot.title.position = "plot")
p2 <- plot(df2, which = "predict", target=c("Site 1"=160,"Site 2"=100,"Site 3"=40,"Overall"=300),
  engine = "ggplot2") +
  labs(col = NULL) +
  theme_classic() +
  theme(legend.position = c(.025,.9),
        legend.justification = "left",
        legend.key.height = unit(2, "mm"),
        legend.key.width = unit(2, "mm"),
        legend.background = element_rect(fill = NA),
        axis.text.x = element_text(angle = 45, vjust = 1, hjust = 1),
        axis.title.x = element_blank())
p1 + p2

In the second ggplot2 example above, we specify different targets for each site, plus a study-level target. The syntax is the same for base graphics.

It's not strictly necessary to use plot for any of the above figures. plot is a wrapper which selects one of 6 underlying functions depending on the value of the which and engine arguments. The underlying functions for base graphics are accrual_plot_cum for cumulative plots, accrual_plot_abs for absolute values, and accrual_plot_predict for the prediction plots. The ggplot equivalents just prepend those names with gg_, i.e. gg_accrual_plot_cum, gg_accrual_plot_abs and gg_accrual_plot_predict. For more clarity, it might be desirable to use those instead, e.g.

gg_accrual_plot_predict(df2, target=c("Site 1"=160,"Site 2"=100,"Site 3"=40,"Overall"=300))

Sample size at a specific time point

It is also possible to predict the expected sample size at a specific time point. If we want to know how many patients will be recruited at the end of the year, we can put the date in the target option.

par(mfrow = c(1, 2))
plot(df4, which = "predict", target = as.Date("2020-12-31"), cex_prediction=0.9,  center_legend="strip")
target<-rep(as.Date("2020-12-31"),4)
names(target)<-c("Site 1","Site 2","Site 3","Overall")
plot(df4, which = "predict", target=target,show_center=FALSE)

In the second example we get site-specific predicions using a target vector (with the same date for all sites). Please note that the site-specific estimates do not necessarily sum up to the overall because they are derived from separate models.

Recruitment tables

Tables of recruitment can also be generated using accrualPlot, via the summary method. As with absolute recruitment above, a unit of time can be specified.

# accrual_table(df) 
summary(df, unit = "day") 
summary(df2, unit = "day") 
summary(df3, unit = "day") 
summary(df3, unit = "day", header = FALSE) 

par(oldpar)

CTU-Bern/accrualPlot documentation built on Aug. 17, 2024, 8:20 p.m.