Analysis of randomized controlled trials

Epidemiol Rev. 2002;24(1):26-38. doi: 10.1093/epirev/24.1.26.


Although the sophistication and flexibility of the statistical technology available to the data analyst have increased, some durable, simple principles remain valid. Hypothesis-driven analyses, which were anticipated and specified in the protocol, must still be kept separate and privileged relative to the important, but risky data mining made possible by modern computers. Analyses that have a firm basis in the randomization are interpreted more easily than those that rely heavily on statistical models. Outcomes--such as quality of life, symptoms, and behaviors--that require the cooperation of subjects to be measured will come to be more and more important as trials move away from mortality as the main outcome. Inevitably, such trials will have to deal with more missing data, especially because of dropout and noncompliance. There are fundamental limits on the ability of statistical methods to compensate for such problems, so they must be considered when studies are designed. Finally, it must be emphasized that the availability of software is not a substitute for experience and statistical expertise.

Publication types

  • Research Support, U.S. Gov't, Non-P.H.S.
  • Review

MeSH terms

  • Data Interpretation, Statistical
  • Humans
  • Linear Models
  • Multicenter Studies as Topic / statistics & numerical data*
  • Outcome and Process Assessment, Health Care / statistics & numerical data
  • Randomized Controlled Trials as Topic / statistics & numerical data*
  • Research Design / statistics & numerical data