Meta-Analysis Of Associations

Use associate$meta() when the question is:

Can I combine compatible association evidence across cohorts, studies, or analysis batches?

Meta-analysis consumes summary-mode association objects and returns another PolyGeniusAssociations object.

Basic Pattern

cohort1 <- associate$regression(...)
cohort2 <- associate$regression(...)

pooled <- associate$meta(
  cohort1 = cohort1,
  cohort2 = cohort2,
  method = "random"
)

The input objects should represent compatible estimands: the same outcome, predictor, term, model family, and effect scale.

What Is Pooled

Meta-analysis pools inferential summary rows:

• effect estimates;
• standard errors;
• confidence intervals;
• p-values;
• sample sizes and study counts where available.

Artifacts are not pooled. For example, cohort-specific survival curves remain cohort-specific.

Returned Object

The meta-analysis result uses the meta schema. Because the result is still a PolyGeniusAssociations object, it can be used by association plotting helpers when the rows are coefficient-like.

Schema

Column	Meaning
source.schema	Source association schema pooled by the meta-analysis.
group.id	Grouping key identifying the pooled estimand.
estimate, se, lower, upper	Pooled effect estimate and uncertainty.
statistic, p.value, adj.p.value	Meta-analytic test result.
n.studies	Number of contributing studies or cohorts.
method, model	Pooling method and fixed/random-effects model label.

Added Artifacts

Meta-analysis pools inferential summary rows only. Source artifacts such as curves, risk tables, and prediction grids remain cohort-specific and are not merged into pooled artifacts.

visualize$associations$forest(pooled)

What Is Meta-Analyzable

Currently appropriate:

• regression rows with finite estimate and se;
• mediation rows with finite estimate and se;
• comparison or contrast rows with coefficient-like estimates and standard errors.

Not appropriate:

• Kaplan-Meier omnibus rows without coefficient-scale standard errors;
• raw artifacts such as curves, risk tables, or prediction grids;
• incompatible effect scales;
• rows that compare different scientific estimands.

Practical Workflow

Fit the same association workflow in each cohort, inspect the cohort-level objects, and pool only compatible rows:

rosmap <- associate$regression(
  rosmap.data,
  outcome = demented,
  predictors = PRS,
  covariates = c(age, sex, PCA)
)

act <- associate$regression(
  act.data,
  outcome = demented,
  predictors = PRS,
  covariates = c(age, sex, PCA)
)

pooled <- associate$meta(ROSMAP = rosmap, ACT = act)

The meta-analysis result is for pooled inference. It does not replace cohort-level diagnostics and plots.