Generate Sensitivity Analysis Plot for Bayesian Meta-Analysis

Creates a comprehensive sensitivity analysis visualization showing how meta-analytic estimates vary across different prior specifications and analysis approaches. The plot combines forest-plot style density distributions with tabulated results.

Usage

sensitivity_plot(
  model,
  data,
  priors,
  measure,
  study_var = NULL,
  rob_var = NULL,
  exclude_high_rob = FALSE,
  incl_pet_peese = FALSE,
  pet_peese_direction = "negative",
  pet_peese_threshold = 0.1,
  incl_mixture = FALSE,
  incl_bma = FALSE,
  model_bma = NULL,
  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 brmsfit object from a Bayesian meta-analysis
data: A data frame containing the study data used to fit the model
priors: A named list of brms prior specifications (created with brms::prior()) with names typically including "vague", "weakreg", and "informative"
measure: Character string specifying the effect measure. One of "OR" (odds ratio), "RR" (risk ratio), "HR" (hazard ratio), "IRR" (incidence rate ratio), "MD" (mean difference), or "SMD" (standardized mean difference)
study_var: Optional. Name of the study identifier variable (unquoted). Required if incl_pet_peese = TRUE or incl_mixture = TRUE
rob_var: Optional. Name of the risk of bias variable (unquoted). Should contain values like "Low", "Some concerns", "High"
exclude_high_rob: Logical. If TRUE, performs sensitivity analysis excluding studies with high risk of bias. Default is FALSE
incl_pet_peese: Logical. If TRUE, includes PET-PEESE bias adjustment analysis. Default is FALSE
pet_peese_direction: Character string. Direction of expected bias for PET-PEESE. Either "negative" or "positive". Default is "negative"
pet_peese_threshold: Numeric. Threshold for choosing between PET and PEESE (based on proportion of studies showing effect). Default is 0.10
incl_mixture: Logical. If TRUE, includes robust mixture model analysis. Default is FALSE
incl_bma: Logical. If TRUE, includes Bayesian model averaging results. Default is FALSE
model_bma: Named list of RoBMA model fits (from RoBMA_brms()) to include in model averaging. Names should be "vague", "weakreg", or "informative"
add_probs: Logical. If TRUE, adds probability columns to the results table. Default is FALSE
null_value: Numeric. The null value for the effect measure. If NULL, uses default based on measure (1 for ratio measures, 0 for difference measures)
null_range: Numeric vector of length 2 or single numeric value. Defines the range of practical equivalence around the null value. If single value, creates symmetric range around null_value
add_null_range: Logical. If TRUE, displays the null range region on the plot. Default is FALSE
color_null_range: Character string. Color for the null range region. Default is "#77bb41" (green)
label_control: Character string. Label for the control/reference group. Default is "Control"
label_intervention: Character string. Label for the intervention group. Default is "Intervention"
title: Optional character string. Main title for the plot
subtitle: Optional character string. Subtitle for the plot
title_align: Character string. Alignment for title and subtitle. One of "left", "center", or "right". Default is "left"
xlim: Optional numeric vector of length 2. X-axis limits for the density plot
x_breaks: Optional numeric vector. Custom x-axis break points
color_palette: Optional color palette for the plot
color_overall_posterior: Character string. Fill color for posterior densities. Default is "dodgerblue"
color_overall_posterior_outline: Character string. Outline color for posterior densities. Default is "blue"
split_color_by_null: Logical. If TRUE, colors densities based on which side of null they favor. Default is FALSE
color_favours_control: Character string. Color when effect favors control. Default is "firebrick"
color_favours_intervention: Character string. Color when effect favors intervention. Default is "dodgerblue"
plot_width: Numeric. Width ratio for the density plot section. Default is 4
font: Optional character string. Font family to use for the plot

Value

A patchwork object combining the sensitivity analysis tables and density plots

Details

This function performs comprehensive sensitivity analyses including:

Prior sensitivity: Re-fits the model with different prior specifications
Risk of bias sensitivity: Excludes high risk of bias studies
PET-PEESE: Publication bias adjustment using precision-based methods
Mixture models: Robust analysis using mixture distributions
Bayesian model averaging: Combines results across model specifications

The output combines three elements using patchwork:

Left table: Prior specifications
Center plot: Density distributions of posterior estimates
Right table: Numerical summaries and probabilities

Examples

if (FALSE) { # \dontrun{
# Basic sensitivity plot
sensitivity_plot(
  model = my_brms_model,
  data = my_data,
  priors = list(
    vague = prior("normal(0, 10)", class = "Intercept"),
    weakreg = prior("normal(0, 1)", class = "Intercept"),
    informative = prior("normal(0.5, 0.5)", class = "Intercept")
  ),
  measure = "SMD"
)

# Advanced sensitivity with all options
sensitivity_plot(
  model = my_model,
  data = my_data,
  priors = my_priors,
  measure = "OR",
  study_var = study_id,
  rob_var = rob,
  exclude_high_rob = TRUE,
  incl_pet_peese = TRUE,
  incl_mixture = TRUE,
  incl_bma = TRUE,
  model_bma = my_robma_models,
  add_probs = TRUE,
  null_range = c(0.9, 1.1),
  add_null_range = TRUE
)
} # }