Variability in clinical data is often more useful than the mean: illustration of concept and simple methods of assessment

BACKGROUND: Clinical investigators, although they are generally familiar with testing differences between averages, have difficulty testing differences between variabilities. OBJECTIVE: To give examples of situations where variability is more relevant than averages and to describe simple methods for testing such data. RESULTS: Examples include: (1) testing drugs with small therapeutic indices, (2) testing variability in drug response, (3) assessing pill diameters or pill weights, (4) comparing patient groups for variability in patient characteristics, (5) assessing the variability in duration of clinical treatment, (6) finding the best method for patient assessment. Various fields of research, particularly in clinical pharmacology, make use of test procedures that implicitly, address the variability in the data. Tests specially designed for testing variability in data include the chi2-test for one sample, the F-test for 2 samples and Bartlett's or Levene's test for 3 or more samples. Additional methods include (1) the comparison of confidence intervals, and (2) testing confidence intervals against prior defined intervals of therapeutic tolerance or equivalence. Many of these tests are available in Excel and other statistical software programs and one such program is described. CONCLUSIONS: In the analysis of clinical data the variability in the data is often more important than averages. Eight simple methods for assessment variability are described to illustrate the value and importance of putting more emphasis on and this parameter