|search externally:||Google Scholar, Springer, CiteSeer, Microsoft Academic Search, Scirus , DBlife|
The principal focus of Dr. Robins' research has been the development of analytic methods appropriate for drawing causal inferences from complex observational and randomized studies with time-varying exposures or treatments. The new methods are to a large extent based on the estimation of the parameters of a new class of causal models - the structural nested models - using a new class of estimators - the G estimators. The usual approach to the estimation of the effect of a time-varying treatment or exposure on time to disease is to model the hazard incidence of failure at time t as a function of past treatment history using a time-dependent Cox proportional hazards model. Dr. Robins has shown the usual approach may be biased whether or not further adjusts for past confounder history in the analysis when:
(A1) there exists a time-dependent risk factor for or predictor of the event of interest that also predicts subsequent treatment, and (A2) past treatment history predicts subsequent risk factor level.
Conditions (A1) and (A2) will be true whenever there are time-dependent covariates that are simultaneously confounders and intermediate variables.
In contrast to previously proposed methods, Dr. Robins' methods can:
1. be used to estimate the effect of a treatment (e.g., prophylaxis for PCP) or exposure on a disease outcome in the presence of time-varying covariates (e.g., number of episodes of PCP) that are simultaneously confounders and intermediate variables on the causal pathway from exposure disease;
2. allow an analyst to adjust appropriately for the effects of concurrent non-randomized treatments or non-random non-compliance in a randomized clinical trial. For example, in the AIDS Clinical Trial Group (ACTG) trial 002 of the effects of high-dose versus low-dose AZT on the survival of AIDS patients, patients in the low-dose arm had improved survival, but they also took more aerosolized pentamidine (a non-randomized concurrent treatment);
3. allow an analyst to adequately incorporate information on the surrogate markers (e.g., CD4 count) in order to stop at the earliest possible moment, randomized trials to the effect of the treatment (e.g., AZT) on survival.
Dr. Robins has applied his methods to analyze the effect of a non-randomized treatment aerosolized pentamidine on the survival of AIDS patients in ACTG Trial 002; the effect of arsenic exposure on the mortality experience of a cohort of Montana copper smelter workers; the effect of formaldehyde on the respiratory disease mortality of a cohort of U.S. chemical workers; and the effect of smoking cessation on subsequent myocardial infarction and death within the MRFIT randomized trial.
Causal inferences from complex observational and randomized studies with time-varying exposures or treatments.
as author at NIPS Workshops, Lake Tahoe 2013,