An Application of Several Analyses for a Group Randomized Trial to Improve Cardiovascular Disease Prevention in Ambulatory Care Practices

• Activity Number: 40
• Type: Contributed
• Date/Time: Sunday, August 11, 2002 : 4:00 PM to 5:50 PM
• Sponsor: Section on Health Policy Statistics*
• Author(s): Paul Nietert*+ and Ruth Jenkins and Steven Ornstein
• Affiliation(s): Medical University of South Carolina and Medical University of South Carolina and Medical University of South Carolina
• Keywords: health services research ; group randomized trials ; prevention ; mixed models ; primary care ; cardiovascular disease

Abstract

A two-year randomized trial in twenty-one primary care practices is underway to evaluate a practice-based intervention aimed at improving cardiovascular disease prevention. For this trial, statistical methods to examine twenty-two outcomes of interest need to account for the practice being the unit of intervention. Randomization tests, mixed models, and traditional methods have all been applied to year one data. The randomization test is excellent for practice-level analyses but cannot account for patient covariates or correlation among outcomes. Mixed models that treat the practice as a random effect allow for both practice- and patient-level covariates. However, using mixed models to determine which practices have made clinically significant improvements is not straightforward. Traditional non-parametric tests do not easily allow for any covariate adjustment, and fixed effects models do not effectively account for correlation among outcomes within practices. Applying these techniques to the year one analysis has demonstrated the pros and cons of each method, and the study investigators have a better understanding where to focus future efforts in making improvements within each practice.