TMLE transporting a treatment effect to a target population

Transport is an extrapolation in the covariate distribution, not in the outcome model: the effect is identified in the target population only under S-admissibility. Relative to a subgroup analysis what changes is the weight – the source rows are reweighted by the SAMPLING ODDS, \(H = I(S = 1)/P(S = 0) * (1 - p(X))/p(X) * [D/g(X) - (1 - D)/(1 - g(X))]\) with p(X) = P(S = 1 | X). Target-population outcomes are never used; only its covariates enter.

Usage

Tmltrn(y, D, X, S)

Arguments

y

Outcome; entries with S = 0 are ignored.

D

Binary treatment; entries with S = 0 are ignored.

X

Covariates, in both populations.

S

1 for a source (trial) row, 0 for a target-population row.

Value

List with estimate, se, eps, n_source, n_target, n.

References

Rudolph, K. E. & van der Laan, M. J. (2017). JRSS B 79(5):1509-1525.

Examples

set.seed(1)
r <- Tmltrn(y = rnorm(10), D = rbinom(10, 1, 0.5), X = rnorm(10), S = rnorm(10)); TRUE
#> [1] TRUE