Abstract: When testing for independence among exposures and clinical outcomes, confounders such as age and gender must be considered. Since tests such as χ², binomial, and Fisher’s exact assume a constant per-patient prior risk, investigators either stratify or pair cases and controls for a cohort with similar properties. These approaches sacrifice statistical power by limiting sample size. We present a straightforward approach to modeling the effects of confounding variables, removing false positives, and retaining statistical power.

Describe the new knowledge and additional skills the participant will gain after attending your presentation.: Participants will learn a straightforward approach to modeling the affects of confounding variables without limiting statistical power. The approach couples logistic regression modeling with the Poisson binomial distribution.

Participants will be given access to CoDe, the Comorbidity Discovery Engine. CoDe provides a GUI for navigating the network of connections among medical diagnoses, procedures, medications, and lab results in an effort to understand disease progression.

Authors:

Gordon Lemmon (Presenter)
University of Utah

Sergiusz Wesolowski, University of Utah
Alex Henrie, University of Utah
Mark Yandell, University of Utah