Comparison of statistical methods of analysis of data from clinical periodontal trials

Abstract

Data from clinical trials have commonly been analyzed by seeking differences between means of clinical measurements in groups treated in different ways. Such differences may be difficult to interpret clinically and fail to impart meaningful information to the research worker or clinician. Attachment level measurements taken from a group of well-maintained patients, initially treated by either scaling or Widman flap surgery, were artificially manipulated to impose known changes on the scaled group. Data from individual sites were analyzed by the t-test and Mann-Whitney test; and frequency distributions by the Kolmogorov-Smirnov 2-sample test and ridit analysis. Extreme changes were sought by the Moses test for extreme reactions and by creating contingency tables in which changes below arbitrary thresholds of change were eliminated. Examination of the results indicated that the mean was sensitive to small changes in large numbers of sites, but insensitive to major changes in a smaller number of sites. Since the latter changes may be of significance to the research worker and clinician, it is recommended that the mean (or median values) be supplemented with the presentation of the frequency distribution of changes and an analysis seeking large changes in few sites.