|
Spring 2006
STATISTICS
COLLOQUIUM
Wednesday, March 8, 2006
3:30-3:45—Refreshments
3:45-4:45—Talk
Yost Hall, Room 101
Thomas Love
Center for Health Care Research and Policy, Case Western Reserve University
A Statistician’s Adventures in Public Health Research: Balance and Design
I demonstrate the importance of balance in [1] observational studies and [2] cluster randomized trials designed to estimate exposure effects. The talk is fairly non-technical, so as to appeal especially to students interested in challenging problems in applied biostatistics and study design. The work is in collaboration with several researchers at the Center for Health Care Research and Policy.
I begin with an example (from the SUPPORT study of right heart catheterization within the first 24 hours in an ICU) which evaluates the impact of propensity score matching for selection bias adjustment on the balance of covariates not included in the propensity model. We illustrate selection bias reduction in these new covariates and posit a general model relating the correlation of an “unmeasured” variate with propensity for exposure and the likely bias reduction implied by matching. The main conclusion here is that researchers selecting variables for purposes of propensity adjustment should take into account the presumed correlation of each new variable with the existing propensity score, in addition to the strength of the new variable’s relationship with outcome, and its current imbalance.
Second, I describe the role of balancing in the design of an ongoing cluster-randomized trial of electronic decision support to improve diabetes care and outcomes. In particular, I illustrate the importance of electronic medical record (EMR) data in executing well-balanced designs for health care intervention trials, and describe some of the logistical and statistical challenges. The main conclusions are that EMR systems provide new opportunities for state-of-the-art study design, and that our general approach could be effectively used to create study groups for a wide range of community-based therapeutic trials or health care delivery trials.
|
