Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/namakala/inappropriate-acid-suppressor-agent-use/llms.txt

Use this file to discover all available pages before exploring further.

Subgroup analysis lets you examine whether pooled prevalence differs across categories of study-level characteristics and whether heterogeneity is reduced within strata compared with the overall model. The pipeline fits a separate random-effects model for each level of five pre-specified variables, using the same fitMetaprop() configuration as the overall model. All subgroup targets are generated automatically using tarchetypes::tar_map(), so adding or removing a variable requires only a change to the uni_vars vector in _targets.R.

​
Subgroup variables

The five variables used for subgroup analysis are defined in _targets.R: 
uni_vars <- c(
  "Age", "Continent", "Setting", "JBI_Classification", "use_guideline"
)
VariableCategories
AgePatient age group (as reported by study)
ContinentNorth America, Asia, Europe, Other
SettingHospital Setting vs Other
JBI_ClassificationMethodological quality tier per JBI appraisal
use_guidelineFollowed Guideline(s) vs No Guideline

​
How subgroup analysis works

fitSubMetaprop() wraps fitMetaprop() to handle the subgroup-specific data filtering and passes the grouping column as the subgroup argument: 
fitSubMetaprop <- function(tbl, varname, ...) {
  idx     <- !is.na(tbl[[rlang::as_name(varname)]])
  sub_tbl <- tbl |> dplyr::filter(idx)

  mod <- sub_tbl %>% fitMetaprop(
    subgroup = dplyr::pull(., {{ varname }}), keepdata = TRUE
  )

  return(mod)
}
The function:
  1. Removes rows where the grouping variable is NA, so studies that did not report that characteristic are excluded from the subgroup model (but remain in the overall model).
  2. Passes the non-missing subset to fitMetaprop(), forwarding the grouping column as subgroup.
  3. Sets keepdata = TRUE so that meta retains the raw study data alongside the fitted object — required for downstream Copas modeling.

​
Iterating over all subgroup variables with tar_map()

Rather than writing one target per variable, _targets.R uses tar_map() to expand a single template across all entries in uni_vars: 
tar_map(
  values = list("colname" = uni_vars),
  unlist = FALSE,
  tar_target(mod_subgroup, fitSubMetaprop(tbl_clean, varname = rlang::sym(colname))),
  tar_target(mod_copas_subgroup, applyCopas(tbl_clean, varname = colname)),
  tar_target(
    plt_forest_subgroup,
    vizForest(
      mod_subgroup,
      file = sprintf("docs/figures/subgroup-meta-analysis-%s.pdf", colname),
      print.subgroup.name = FALSE
    )
  ),
  tar_target(plt_funnel_subgroup, vizFunnel(mod_subgroup))
)
This produces targets named mod_subgroup_Age, mod_subgroup_Continent, mod_subgroup_Setting, mod_subgroup_JBI_Classification, and mod_subgroup_use_guideline, along with paired Copas, forest-plot, and funnel-plot targets for each.

​
Accessing subgroup results

Load any subgroup model by its generated target name: 
targets::tar_load(mod_subgroup_Continent)
summary(mod_subgroup_Continent)
The returned object is a standard meta metaprop object with an additional bylevs slot listing the subgroup labels. You can inspect per-stratum estimates through mod_subgroup_Continent$TE.random.w, mod_subgroup_Continent$lower.random.w, and mod_subgroup_Continent$upper.random.w.

​
Interpreting heterogeneity within subgroups

Each stratum in the subgroup model has its own set of heterogeneity statistics. You should examine:
  • Within-subgroup I² — the proportion of variability within a stratum attributable to true heterogeneity rather than sampling error.
  • Within-subgroup τ² — the between-study variance inside the stratum; a reduction relative to the overall τ² suggests the variable explains some heterogeneity.
  • Within-subgroup Q — the Wald statistic for the test of heterogeneity within that stratum.
A variable is considered a meaningful moderator when the between-subgroup test is significant (reported as Q_b in the summary() output) and within-stratum heterogeneity is lower than the overall model.
Use tar_objects() with a dplyr selector to list every subgroup target at once without typing each name manually:
targets::tar_objects(dplyr::contains("mod_subgroup"))

Build docs developers (and LLMs) love

Get started for freeTalk to us