Skip to Content
Interactive Textbook on Clinical Symptom Research Logo


Home button

Trial Design: Pain Sections
Author Bio
Introduction
Placebo Effects
Single Dose Trials
Repeated Dose Trials
Explanatory Versus Pragmatic
Dose-Response
Currently selected section: Parallel Group Versus Crossover
Conclusion
 

 

Chapter 1: Clinical Trials of Pain Treatment: Parallel Group Versus Crossover Designs

 
          

Limitations of Crossover Studies (continued)

In the past two decades, the major concern with crossover studies has been the possibility of bias produced by unequal "carryover effects". Carryover effects are changes in the efficacy of treatments resulting from treatments given in earlier periods; they may be mediated by persistence of drug or metabolites, changes in brain or peripheral tissues caused by the treatment, or behavioral or psychological factors. Statisticians have most energetically attacked the two-treatment, two-period design ("2 x 2"; Figure 7.1, below). Critics claim that results may be difficult to interpret whenever the treatment effect differs for the two periods. In this event, one cannot distinguish with any certainty whether this is due to:

(a)a carryover effect (persistence of a pharmacological or psychological effect of the first treatment into the second period);
(b)a "treatment x period interaction" (the passage of time affects the relative efficacy of the treatments; e.g. by the second period, patients who initially received placebo might be too discouraged to respond to any subsequent treatment), or
(c)

a difference between the groups of patients assigned the two different orders of treatment.

Figure 7.1: Left-hand side is a 2 by 2 design with 2 rows: the first row is treatment A followed by treatment B; the second row is treatment B followed by treatment A. Right-hand side contains some Latin square designs of studies with 3 or 4 treatments. The 3 treatment design contains 3 rows: the first row is treatment A followed by treatment B followed by treatment C; the second row is treatment B followed by treatment C followed by treatment A; and the third row is treatment C followed by treatment A followed by treatment B. The 4 treatment design contains 4 rows: the first row is treatment A followed by treatment B followed by treatment C followed by treatment D; the second row is treatment B followed by treatment A followed by treatment D followed by treatment C; the third row is treatment C followed by treatment D followed by treatment B followed by treatment A; and the fourth row is treatment D followed by treatment C followed by treatment A followed by treatment B.

Because of these concerns, regulatory agencies have been reluctant to rely upon data from such designs.

Fortunately, these statistical difficulties are largely limited to the 2 x 2 case (and Senn (1993) argues that these difficulties have been exaggerated). For studies involving three or more treatments, there are a variety of designs that allow these effects to be distinguished (Jones and Kenward, 1989; Ratkowsky, 1993). The most commonly used are Latin square designs (Figure 7.1, above). My current view is that the relative brevity, simplicity, and superior power of the 2 x 2 design makes it attractive for single-center studies in which previous experience suggests that there is no significant carryover effect. However, if one is doing studies for regulatory review, one may wish to seek expert advice about the regulators’ current statistical thinking.

 

Page 48 of 52
      Previous Page