This proposal describes training and research activities whose goal is to further the development of Dr. Hernan as an independent investigator in the area of causal inference from complex longitudinal HIV data. Dr. Hernan is a physician with graduate degrees in epidemiology and biostatistics. He is currently a research associate at the Harvard School of Public Health, where he works on the application and development of causal methods under the supervision of professor James Robins. His career goals are to become an academic investigator with a faculty appointment and to develop his research career on causal inference from longitudinal data. To accomplish his goals, Dr. Hernan will devote a large proportion of the first two years of the award to strengthening his background in causal methods. The proposed research career development plan includes specialized tutorials, independent study, coursework, seminars, and participation in working groups and scientific meetings. During the first two years he will also conduct supervised research project of increasing complexity. The last three years of the award will be research-intensive. Inferring valid causal inferences from longitudinal HIV studies requires the use of the appropriate methods for confounding and selection bias adjustment. Causal methods (such as marginal structural models and structural nested models) are the best methods available to estimate the effect of a treatment in or a cofactor on an outcome of interest from observational data and to adjust for dependent censoring, non-random non-compliance, treatment cross-over or termination, and the concurrent effect of additional non-randomized treatments in randomized clinical trials. On the other hand, standard methods, such as regression (e.g, logistic, Poison, Cox proportional hazards) models or stratification may produce effect estimates that cannot be endowed with a causal interpretation. A substantial part of the research is aimed at conducting a systematic evaluation of the relative advantages and disadvantages of causal and standard methods in HIV epidemiology. This evaluation will include the application of causal methods to answer substantive questions using actual HIV data sets (e.g., the MACS). Also, the research plan describes an evaluation of several common problems in longitudinal analysis and how they can be handled using causal methods. These problems include dependent censoring missing data, model misspecification, and unmeasured confounding.
Showing the most recent 10 out of 14 publications
