Federated Average Treatment Effect Estimation Across Studies

Client-side 'DataSHIELD' package that orchestrates the federated estimation of the average treatment effect (ATE) across multiple independent randomised controlled trials (RCTs), each held on a separate data server. Implements in a federated way the g-formula estimator from Khellaf, Bellet and Josse (2025) "Federated Causal Inference: Multi-Study ATE Estimation beyond Meta-Analysis" , including variance estimation and optional adjustment for study-specific intercept shifts.

dsFedAteClient

Client-side 'DataSHIELD' package for federated estimation of the average treatment effect (ATE) across multiple independent studies, each held on a separate data server. dsFedAteClient runs on the analyst's machine; the companion server-side package dsFedAte must be installed on each data server.

Installation

install.packages("dsFedAteClient")

Example

library(dsFedAteClient)

conns <- DSI::datashield.login(logins = logindata, assign = TRUE,
                               symbol = "base")

result <- ds.EstimateAte(
  datasources     = conns,
  base_name       = "base",
  outcome_name    = "Y",
  treatment_name  = "W",
  covariate_names = c("X1", "X2", "X3"),
  heterogeneous   = TRUE
)

result$ate
result$ci_lower; result$ci_upper

A full worked example — including arm-specific variance and study-effects adjustment, run locally with DSLite — is provided in the vignette:

vignette("federated_rct", package = "dsFedAteClient")