Bayesian Meta-Analysis for Few Studies

Specialized Bayesian meta-analysis for situations with few studies (typically < 5-10 studies). Uses conservative priors and includes sensitivity analysis.

Usage

bayesian_meta_analysis_few(
  yi,
  sei,
  study_labels = NULL,
  effect_measure = c("hr", "or", "rr", "rd", "md", "smd"),
  prior_mu = list(mean = 0, sd = 2),
  prior_tau = list(type = "half_cauchy", scale = 0.25),
  prior_sensitivity = TRUE,
  chains = 4,
  iter = 6000,
  warmup = 3000,
  seed = NULL,
  adapt_delta = 0.99,
  max_treedepth = 15,
  backend = c("auto", "cmdstanr", "rstan"),
  ...
)

Arguments

yi

Numeric vector of effect estimates

sei

Numeric vector of standard errors

study_labels

Character vector of study names

effect_measure

Character. Effect type: "hr" (hazard ratio), "or" (odds ratio), "rr" (risk ratio), "rd" (risk difference), "md" (mean difference), "smd" (standardized mean difference)

prior_mu

Prior for overall effect mean (default: 0, very wide)

prior_tau

Prior for heterogeneity. For few studies, uses more informative priors that pull tau towards smaller values

prior_sensitivity

Logical. Whether to perform prior sensitivity analysis

chains

Integer. Number of MCMC chains (default: 4)

iter

Integer. Total iterations per chain (default: 6000)

warmup

Integer. Warmup iterations (default: 3000)

seed

Integer. Random seed for reproducibility

adapt_delta

Numeric. MCMC sampler tuning parameter (0-1)

max_treedepth

Integer. Maximum tree depth for NUTS sampler

backend

Character. Stan backend: "auto", "cmdstanr", "rstan"

...

Additional arguments passed to brms::brm

Value

A BayesianMetaFewResult object

Details

This function is specifically designed for meta-analyses with few studies and implements several adjustments:

1. Conservative priors: Uses more informative priors that are appropriate when data is sparse

2. Regularization: Stronger regularization of between-study variance (tau^2)

3. Sensitivity analysis: Includes prior sensitivity analysis by default

4. Few studies warnings: Provides explicit warnings about limitations

Recommended for meta-analyses with 2-10 studies. For larger meta-analyses, use the standard bayesian_meta_analysis() function.

References

Spiegelhalter, D.J. et al. (2004). Bayesian approaches to clinical trials and health-care evaluation. Wiley.

Examples

if (FALSE) { # \dontrun{
# Meta-analysis with only 3 studies
yi <- log(c(0.75, 0.82, 0.68))  # log(HR) from 3 studies
sei <- c(0.12, 0.15, 0.18)       # standard errors

result <- bayesian_meta_analysis_few(
  yi = yi,
  sei = sei,
  effect_measure = "hr",
  chains = 2,
  iter = 4000
)

# With prior sensitivity analysis
result <- bayesian_meta_analysis_few(
  yi = yi,
  sei = sei,
  effect_measure = "hr",
  prior_sensitivity = TRUE,
  chains = 2,
  iter = 4000
)
} # }