Maraca Plots - Validation

In maraca: The Maraca Plot: Visualization of Hierarchical Composite Endpoints in Clinical Trials

This vignette is largely based on the PharmaSUG 2023 conference paper.

knitr::opts_chunk$set(echo = TRUE, collapse = TRUE)
library(dplyr)
library(ggplot2)
library(maraca)

When using a maraca plot in a regulatory setting, for example to visualize clinical study results for regulatory submissions, there will be strict validation demands. For example, the results might need to be double programmed for validation purposes.

In order to facilitate the validation of the graphic output, the maraca package includes the function validate_maraca_plot() that allows the user to extract important metrics from the plot itself. This allows to programmatically compare the results of a plot produced using the maraca package with other programmatic approaches. The paper linked above walks through a validation example where the double programming was done in SAS.

To use the validation functionality, we need to first create a maraca object.

library(maraca)

data(hce_scenario_a)

maraca_dat <- maraca(
  data = hce_scenario_a,
  step_outcomes = c("Outcome I", "Outcome II", "Outcome III", "Outcome IV"),
  last_outcome = "Continuous outcome",
  fixed_followup_days = 3 * 365,
  column_names = c(outcome = "GROUP", arm = "TRTP", value = "AVAL0"),
  arm_levels = c(active = "Active", control = "Control"),
  compute_win_odds = TRUE
)

We then create a maraca plot and save the actual plot as an object.

# Save plot as its own object
maraca_plot <- plot(maraca_dat)
# The plot has its own class called "maracaPlot"
class(maraca_plot)
# Display plot
maraca_plot

Now we can validate the plot using the validate_maraca_plot() function.

validation_list <- validate_maraca_plot(maraca_plot) 
# Display which metrics are included
str(validation_list)

Running the validate_maraca_plot() function on a maraca plot object returns a list with the following items:

1. plot_type: depending on which density_plot_type was selected for the plot either GeomPoint, GeomViolin and/or GeomBoxplot
2. proportions: the proportions of the HCE components
3. tte_data: time-to-event data if part of the step outcomes has type tte, otherwise NULL
4. binary_step_data: binary data if part of the step outcomes has type binary, otherwise NULL
5. binary_step_data: if last endpoint was binary then contains the data for the minimum, maximum and middle point x values displayed in the ellipsis, otherwise NULL
6. scatter_data: if last endpoint was continuous and plot was created with density_plot_type = "scatter" then contains dataset that was plotted in scatter plot, otherwise NULL
7. boxstat_data: if last endpoint was continuous and if plot was created with density_plot_type = "box" or density_plot_type = "default" then contains the boxplot statistics, otherwise NULL
8. violin_data: if last endpoint was continuous and if plot was created with density_plot_type = "violin" or density_plot_type = "default" then contains the violin distribution data, otherwise NULL
9. wo_stats: if maraca object was created with compute_win_odds = TRUE then contains the win odds statistics, otherwise NULL

These can then be converted to a convenient format for validation, such as as individual data.frames.

library(dplyr)
library(tidyr)

validation_list$proportions %>% 
  as.data.frame() %>%
  rename("proportion" = ".")

head(validation_list$tte_data)

validation_list$boxstat_data %>%
 unnest_wider(outliers, names_sep = "") %>%
 pivot_longer(., cols = -group, names_to = "stat_name", values_to = "values") %>%
 filter(!is.na(values)) %>% 
  as.data.frame()

head(validation_list$violin_data)

validation_list$wo_stats

maraca documentation built on May 29, 2024, 8:08 a.m.