Crossover Trials in Diabetes Research

Abstract

Crossover designs are efficient alternatives to parallel designs. Sample size reduction is approximately 4 to 1 in diabetes research with hemoglobin A1c as the primary variable. Challenge occurs on understanding and handling potential carry-over effect. This presentation starts with fundamentals: design, model and analyses with SAS programs. One of the main focuses is to understand the nature of the carry-over effect. The well accepted two stage approach by Grizzle (1965) possesses serious flaws. Summarization of critiques on Grizzle's approach and different views of handling the carry-over effect from Freeman (1989), Lehmacher (1991) and Senn (1993, 1997) are discussed. Lastly, we assess the influence of carry-over effect from results of several clinical trials to evaluate the legitimacy on the use of crossover designs in diabetes research. Recommendation to the conduct and analysis of crossover trials are also provided.