Implementing and comparing statistical models in a clinical setting
by
Teena West
The University of Auckland
Coauthors: Derrick Bennett (The University of Auckland)

In a hospital setting complex statistical models are often difficult to implement and interpret and offer little help with clinical decision making. Clinicians can feel overwhelmed by anything more sophisticated than simple logistic regression and are reluctant to purse more intricate statistical methods for journal publications or medical conferences. In this talk 3 methods (MANOVA, Generalised linear mixed models and Bayesian estimation) are used to investigate the effect of carvedilol, a beta-blocker drug, on left ventricular size and function.

The study data comes from the Australia/New Zealand Carvediol Heart Failure Research Collaborative Group Study of 415 patients with mild to moderate ischemic heart failure. Patients were randomised to carvedilol or placebo after a 2- to 3-week run in phase on open-label carvedilol. While the overall conclusions for the 3 methods were similar, attention is drawn to the estimates precision, ease of implementation and simplicity of presentation to a clinical audience.

Date received: August 30, 2001