Skip to contents

Generate Interactive AE Forest Plots with Drill Down to AE Listing

Source: vignettes/forestly.Rmd
forestly.Rmd

Introduction

The forestly R package creates interactive forest plots for clinical trial analysis and reporting. The interactive forest plots include: - AE-specific tables along with the AE proportion difference and its confidence interval visualized. - AE listing tables corresponding to a selected AE.

Through its interactive features, users can: - Search for the AE PT term or SOC term of interest. - Filter AEs with an incidence greater than a specified percentage in one or more arms. - Sort AEs from the smallest proportion or proportion difference to the largest. - Switch between different AE-specific tables. - Drill down to listings for subjects with the selected AE.

Procedure to generate interactive forest plot

Step 1: read the ADAM data

In the example presented in this vignette, we utilize the ADAM data from the forestly R package. We further factorize the actual arm variable, TRTA, to enable the forestly R package to identify which arm serves as the reference group.

library(metalite)
library(forestly)

adsl <- forestly_adsl
adae <- forestly_adae

adsl$TRTA <- factor(forestly_adsl$TRT01A, 
                    levels = c("Xanomeline Low Dose", "Placebo"), 
                    labels = c("Low Dose", "Placebo"))
adae$TRTA <- factor(forestly_adae$TRTA, 
                    levels = c("Xanomeline Low Dose", "Placebo"), 
                    labels = c("Low Dose", "Placebo"))

Step 2: build the metadata

We use the metalite R package to generate a minimal metadata for the generation of the interactive forest plot.

To begin, we input the adsl data as the population data and the adae data as the observation data.

meta <- meta_adam(population = adsl, observation = adae)

Without loss of generality, our plan is to generate the following four AE-specific tables:

  • Any AE (we give it a short name/nickname as "any")
  • Drug-related AE (we give it a short name/nickname as "drug-related")
  • Serious AE (we give it a short name/nickname as "serious")
  • Drug-related serious AE (we give it a short name/nickname as "drug-related-serious") However, users are allowed to add more AE-specific tables as needed. For the above 4 AE specific tables, we take the APaT for both observation and observation.
meta <- meta |>
  define_plan(plan = plan(analysis = "ae_forestly", 
                          population = "apat",
                          observation = "apat", 
                          parameter = "any;drug-related;serious;drug-related-serious"))

Following the definition of the analysis plan, we further specify the details of analysis = "ae_forestly" by providing its label for display.

meta <- meta |>
  define_analysis(name = "ae_forestly", label = "Interactive Forest Plot")

We then define the related population (population = "apat") and observation (observation = "apat") by specifying their grouping variable (group = ...), subject ID (id = ...), population flag (subset = ...) and label (label = ...). If users wish to use different population flags, grouping variables, or labels, they can make changes as needed.

meta <- meta |>
  define_population(name = "apat", group = "TRTA", id = "USUBJID",
                    subset = SAFFL == "Y", label = "All Patient as Treated") |>
  define_observation(name = "apat", group = "TRTA", 
                     subset = SAFFL == "Y", label = "All Patient as Treated")

Next, we specify the details of parameter = "any;drug-related;serious;drug-related-serious". For any AEs ("any"), their is no filter applied.

meta <- meta |>
  define_parameter(name = "any", 
                   subset = NULL,
                   label = "Any AEs", 
                   var = "AEDECOD", soc = "AEBODSYS")

For drug related AEs ("drug-related"), its filter is AEREL %in% c("PROBABLE", "POSSIBLE").

meta <- meta |>
  define_parameter(name = "drug-related", 
                   subset = toupper(AREL) == "RELATED",
                   label = "Drug-related AEs", 
                   var = "AEDECOD", soc = "AEBODSYS")

Similarly, we define serious AE and drug related serious AEs.

meta <- meta |>
  define_parameter(name = "serious", 
                   subset = AESER == "Y",
                   label = "Serious AEs", 
                   var = "AEDECOD", soc = "AEBODSYS") |>
  define_parameter(name = "drug-related-serious", 
                   subset = AESER == "Y" & AEREL %in% c("PROBABLE", "POSSIBLE"),
                   label = "Drug-related serious AEs", 
                   var = "AEDECOD", soc = "AEBODSYS")

Finally, we build the metadata by running the meta_build() function.

meta <- meta |>  meta_build()

Step 3: generate the interative AE forest plot

With the meta built as described above, users can generate the interactive AE forest seamlessly through a few simple function calls connected with the pipe operator (|>). The prepare_ae_forestly() prepares datasets (i.e., calculate all to-be-reported numbers). The format_ae_forestly() takes care of formating, and the ae_forestly() function realize the interactivity.