Analysis of a Longitudinal Cluster-Randomized Comparative Trial with a Binomial Outcome

• Activity Number: 18
• Type: Contributed
• Date/Time: Sunday, August 11, 2002 : 2:00 PM to 3:50 PM
• Sponsor: General Methodology
• Author(s): Lynette Keyes-Elstein*+ and Peter Margolis and Leslie Loeding and Carole Lannon
• Affiliation(s): Rho, Inc. and University of North Carolina, Chapel Hill and University of North Carolina, Chapel Hill and University of North Carolina, Chapel Hill
• Keywords: cluster-randomized ; longitudinal ; binomial ; logistic

Abstract

For comparative trials using cluster-randomization, statistical analyses must target the appropriate unit of analysis while accounting for the correlation associated with nested data. We describe the analysis of a cluster-randomized longitudinal trial to evaluate an intervention designed to improve delivery of preventive services (e.g., immunizations) to two year olds by pediatric practices. Since the intervention targets "practice performance," "practice" is the unit of analysis. Random samples of thirty charts per practice were selected at intervals of six months for twelve intervention practices and twelve months for twelve control practices. To compare the pattern of change over time in the treatment groups,the proportion up-to-date on services is modeled using a logistic random regression approach. The fixed effects include separate intercepts and pre-intervention linear time effects for each treatment group. In addition, linear and quadratic post-intervention time effects are included for the intervention group (i.e., a regression spline). Random practice effects for intercept, and pre- and post-slopes are assumed to be independent and normally distributed.