Analysis of Observational Studies for Emerging Health Care Applications
Epidemiology, the study of
patterns, causes, and effects of health and disease conditions in defined
populations, is a cornerstone of public health. By identifying risk factors for
disease and targets for preventive healthcare, it shapes policy decisions and
evidence-based practice. Due to the constraints of ethnics and costs, most
epidemiologic research focuses on observational (or non-experimental) studies.
Our approach is to combine the advances in the design of observational studies
in epidemiology with corresponding advances in statistical methodologies for
the analysis of observational data, controlling for confounding factors and
other sources of systematic errors. It is expected to yield novel efficient
study designs and analysis methods that can capitalize on the recent explosion
of digital data such as electronic health records, medical billings and on-line
recruitments, while providing valid inference that adjusts for possible
confounding and selection bias.
We review in this connection an ongoing observational
cohort study that uses structural equation modeling in optimizing the latent
exposure and outcome variables and account for the issue of negative
confounding. This is joint work with Philippe Grandjean at the Harvard School
of Public Health.