Division of General Internal Medicine

Department Home >

  • Base >
  • You are here
Division of General Internal Medicine

Presentations: Evidence-based Analysis for Economic Analysis: Sensible Response to Reduce Noise or Dangerous Violation of the Randomization Principle?

Issues Panel #6

ISPOR 9th Annual European Congress
Copenhagen, Denmark
10:30 - 11:30, May 24, 2006
Moderator: Mark Sculpher, Ph.D., University of York, York, UK


Economic analyses of patient level data are typically hampered by the large amounts of noise in the data.  In particular, it is well known that there are high levels of variability in cost and health related quality of life data.  This can lead to problems of power, particularly in randomised controlled trials, where the sample size calculations are rarely based on economic outcomes.  As a consequence, standard approaches to economic evaluation alongside trials often result in high levels of uncertainty reflected by wide confidence intervals on cost-effectiveness.  One solution to this problem has been to explicitly design studies as 'event based' analyses.  This approach is based on the assumption that it is events that drive the cost and quality of life effects of treatment interventions.  These events could be the primary efficacy endpoints in a clinical trial, but might also include unintended effects (adverse events) associated with treatment.  Use of regression models with these events as explanatory for cost and quality of life offer the potential for the precise estimation of cost and disutility associated with these events, while disregarding the background variability or 'noise'.

Some commentators may argue that such use of post randomisation variables violates the randomisation principle.   While others may counter that through the use of a conditional independence assumption (that treatment only affects cost or quality of life through the impact on event rates) the validity of randomisation is preserved.

The purpose of this panel will be to use recent examples of event based analyses from the cardiovascular and respiratory disease areas to compare and contrast the two approaches to engage the audience in a debate about the appropriateness of the two approaches - in particular whether the benefits of event based analyses in terms of increased precision are worth the additional structural assumptions imposed.