Skip to contents

Generate Sensitivity Analysis Plot for Bayesian Meta-Analysis

Source: R/sensitivity_plot.R
sensitivity_plot.Rd

Creates a sensitivity analysis visualisation showing how meta-analytic estimates vary across different model strategies and user-specified priors.

For RoBMA results, models must be pre-computed using run_robma_sensitivity() and attached to the bayesma object before calling this function.

Usage

sensitivity_plot(
  model,
  data,
  priors,
  estimand,
  prior_order = NULL,
  model_order = NULL,
  rob_var = NULL,
  exclude_high_rob = FALSE,
  incl_common_effect = FALSE,
  incl_random_effect = TRUE,
  incl_bias_corrected = FALSE,
  incl_selection_copas = FALSE,
  incl_selection_weight = FALSE,
  incl_pet_peese = FALSE,
  incl_robust = FALSE,
  incl_robma = FALSE,
  parallel = FALSE,
  workers = NULL,
  seed = TRUE,
  add_probs = FALSE,
  null_value = NULL,
  null_range = NULL,
  add_null_range = FALSE,
  color_null_range = "#77bb41",
  label_control = "Control",
  label_intervention = "Intervention",
  title = NULL,
  subtitle = NULL,
  title_align = "left",
  xlim = NULL,
  x_breaks = NULL,
  color_palette = NULL,
  color_overall_posterior = "dodgerblue",
  color_overall_posterior_outline = "blue",
  split_color_by_null = FALSE,
  color_favours_control = "firebrick",
  color_favours_intervention = "dodgerblue",
  plot_width = 4,
  font = NULL
)

Arguments

model

A fitted bayesma object. For RoBMA results, must have $robma_sensitivity attached via attach_robma_sensitivity().

data

A data frame containing the study data used to fit the model.

priors

A named list of prior specifications. Each element must be a list with at least mu_prior and (optionally) tau_prior, and may include name used for display.

estimand

Effect measure or marginal estimand. Relative-effect: "OR", "HR", "RR", "IRR", "MD", "SMD". Marginal: "RD"/"ARR", "ATE", "ATT", "CATE".

prior_order

Optional character vector specifying the display order of priors. Should contain the names (IDs) of the priors in the desired order, e.g., c("vague", "weak_reg", "informative"). If NULL (default), priors are displayed in the order they appear in the priors list.

model_order

Optional. Specifies the display order of model sections. Can be provided as unquoted names (using rlang) or as a character vector.

Valid values match the incl_* argument names (without the incl_ prefix): common_effect, random_effect, bias_corrected, selection_copas, selection_weight, pet_peese, robust, robma.

For RoBMA, use robma to include both conditional and model-averaged estimates (conditional first), or robma_conditional / robma_model_averaged separately.

Example: model_order = c(common_effect, random_effect, pet_peese, robma)

If NULL (default), sections are displayed in the default order.

rob_var

Optional. Name of the risk-of-bias variable (unquoted).

exclude_high_rob

Logical. If TRUE, runs an "Excluding High RoB" section.

incl_common_effect, incl_random_effect, incl_bias_corrected, incl_selection_copas, incl_selection_weight, incl_pet_peese, incl_robust

Logical. Which model strategy sections to include.

incl_robma

Logical. If TRUE, include RoBMA section. Requires model$robma_sensitivity to be present (see Details).

parallel

Logical. If TRUE, uses parallel processing for non-RoBMA refits.

workers

Optional integer. Number of parallel workers.

seed

Logical. If TRUE (default), uses parallel-safe seeding.

add_probs

Logical. Add probability columns to the results table.

null_value, null_range, add_null_range, color_null_range

ROPE settings.

label_control, label_intervention

Group labels for the plot.

title, subtitle, title_align

Title settings.

xlim, x_breaks

Density axis settings.

color_palette, color_overall_posterior, color_overall_posterior_outline

Colour settings.

split_color_by_null

Logical. If TRUE, colour the posterior split at the null value using color_favours_control and color_favours_intervention.

color_favours_control

Colour for the side of the posterior favouring control.

color_favours_intervention

Colour for the side favouring intervention.

plot_width

Width ratio for the density plot section.

font

Optional font family.

Value

A bayesma_sensitivity_plot object (patchwork combining tables and plots).

Details

RoBMA Results

To include RoBMA in the sensitivity plot, you must pre-compute the RoBMA fits:

# Step 1: Run RoBMA sensitivity analysis
robma_sens <- run_robma_sensitivity(
  data = my_data,
  priors = my_priors,
  robma_template = my_robma_fit,
  parallel = TRUE
)

# Step 2: Attach to bayesma model
model <- attach_robma_sensitivity(model, robma_sens)

# Step 3: Create sensitivity plot with RoBMA
sensitivity_plot(model, data, priors, estimand = "OR", incl_robma = TRUE)

This separation allows the computationally expensive RoBMA fitting to be done once and cached, rather than being re-run every time the plot is generated.