QUERI – Quality Enhancement Research Initiative

Go to the HSRD website

Quality Improvement Methods

11. Design of Experiments

a. Definition: Design of Experiments (DOE) is a statistical approach to efficiently plan tests so that when changes are made, such as clinical interventions in existing processes, the effect on individual factors associated with changes can be determined. This approach to gain knowledge from the outcome of tests or experiments is to plan tests in such a way that information can be gathered with a minimum number of trials or experiments.

b. Literature:

  • Telford, Jacqueline K. "A brief introduction to design of experiments." Johns Hopkins apl technical digest 27.3 (2007): 224-232.
  • Neuhauser, Duncan. "Why design of experiments just may transform health care." Quality Management in Healthcare 14.4 (2005): 217-218.
  • Taguchi, Genichi, Subir Chowdhury, and Yuin Wu. Taguchi's quality engineering handbook. Wiley-Interscience, 2005. (Taguchi and Fisher are generally considered founders of DOE)
  • Fisher, Ronald A. (1971) [1935]. The Design of Experiments (9th ed.). Macmillan. ISBN 0-02-844690-9.
  • Barrentine, Larry B. An introduction to design of experiments: a simplified approach. ASQ Press, 1999.

c. Example: A hospital system was considering changes to a current clinical function. The changes were an increased use of a particular medication and also the reduced use of a certain therapy, which could both be done to varying degrees. Thus they were faced with a number of trials of possible change combinations. The hospital has various outcome measures with regards to evaluating the procedure. They want to determine a minimal set of trials. DOE provided a way to minimize the necessary set of trials (or experiments) so they could determine the best procedure as quickly as possible.

d. Steps: The use of the DOE idea is to determine what outcome you are interested in, what changes are to be tested and then to use DOE algorithms to determine an efficient plan to determine the best combination of changes. In order to utilize DOE there must first be some sort of model that relates the inputs to the outputs in a system to be studied.