Skip to content

Meta-analyze a comparative effect estimate across databases (evidence synthesis)

Source: R/analysis.R
ds.omop.meta.effect_estimate.Rd

The CLIENT half of OHDSI evidence synthesis: run a per-site fitted comparative effect estimate on every server, then INVERSE-VARIANCE meta-analyze the per-site log-estimates into ONE pooled estimate + 95% CI (the metafor::rma pattern by hand — no new dependency). A single site cannot compute a cross-database pooled estimate, so the server-side dsomop:cm.effect_estimate (CohortMethod HR/RR; the es_cm_result delegate) and dsomop:sccs.incidence_rate_ratio (SCCS IRR; the es_sccs_result delegate) each emit only the disclosure-safe per-site log_estimate + SE; this function pools them.

Usage

ds.omop.meta.effect_estimate(
  name = "dsomop:cm.effect_estimate",
  params = list(),
  cohort = NULL,
  tables = NULL,
  combine = "union",
  pooling_policy = "strict",
  symbol = "omop",
  conns = NULL
)

Arguments

name

Character; the per-site effect-estimate analysis id. Default "dsomop:cm.effect_estimate" (CohortMethod). Use "dsomop:sccs.incidence_rate_ratio" for SCCS, or the es_cm_result / es_sccs_result evidence-synthesis ids.

params

Named list of analysis params (e.g. outcome_concept_id, model_type); passed through to the per-site analysis unchanged.

cohort

For CohortMethod, the two-population target+comparator scope (a length-2 set of cohort handles / ids / table names); for SCCS, the scoped case population.

tables

Optional omop.table symbol scope (see ds.omop.analysis.run).

combine

Character; "union" (default) or "intersect" for multi-source scope folding.

pooling_policy

Character; "strict" (default) or "pooled_only_ok".

symbol

Character; the session symbol (default "omop").

conns

DSI connection object(s) or NULL to use the session default.

Value

A dsomop_result: per_site holds each server's gated per-site effect-estimate frame; pooled holds the one-row meta-analysis (pooled HR/RR + CI under both models, n_databases, i2, tau2).

Details

Both a FIXED-effect and a random-effects (DerSimonian-Laird) pooled estimate are returned, with Cochran's Q, \(I^2\), and \(\tau^2\) heterogeneity. No patient data crosses sites — only the already-gated per-site sufficient statistics. A site whose per-site estimate the server suppressed (small/empty arm) is ABSENT from the pool: under pooling_policy = "strict" (default) any suppressed site aborts the pool fail-closed; "pooled_only_ok" pools the remaining sites and warns.

See also

ds.omop.analysis.run

Examples

if (FALSE) { # \dontrun{
# Pool a CohortMethod hazard ratio across databases.
res <- ds.omop.meta.effect_estimate(
  params = list(outcome_concept_id = 4329847),
  cohort = c(target_cohort, comparator_cohort))
res$pooled   # estimate_random, ci_lo_random, ci_hi_random, i2, ...
} # }