Inverse probability weighted estimation in survival analysis

Modern epidemiologic and clinical studies aimed at analyzing a time to an event endpoint T routinely collect, in addition to (possibly censored) information on T, high dimensional data often in the form of baseline (i.e. timeindependent covariates V (0)) and time-varying covariates V (t) , t > 0, measured at frequent intervals. Scientific interest, however, often focuses on a low dimensional functional β = β (FX) of the distribution FX of the (intended) full data X = ¡ T, V (T ) ¢ where V (t) ≡ {V (u) : 0 ≤ u ≤ t} . Inverse probability weighted augmented (AIPW) estimators of β meet the analytic challenge posed by these high dimensional data because they are consistent and asymptotically normal (CAN) under models that do not make assumptions about the parts of FX that are of little scientific interest. As such, they are not subject to biases induced by misspecification of models for these secondary parts of FX .