| Abstract: | It is most important that the proper design of a clinical trial be developed prior to the start of the study. Such a design is one that will not tend to find a treatment to be effective when it really is ineffective, and will not tend to miss a good treatment when it really is effective. The questions to be answered by a clinical trial must be decided in advance and will help determine the design. Trials should take into consideration how the agent is to be used: carefully controlled, ordinary use, subject or professionally applied, etc. The double-blind parallel randomized trial is preferred to the double blind cross-over design, except possibly for a professionally applied treatment which does not leach into the saliva or affect the oral microflora. Blocking or stratification tends to increase the efficiency of a clinical trial, and also may produce more information on the agent's effectiveness. Stratification is preferred to blocking, particularly when subjects enter the trial serially over time. Subjects should be stratified by age, sex and baseline level of plaque or gingivitis. Stratification permits a check on whether the effect of treatment is atypically strong or weak for certain types of patients. Where possible, the same examiner should examine the same patients throughout the study. Examiners should be self-calibrated to reduce intra-examiner error. Plaque and gingivitis trials should be at least of 6 months duration and may be preventive or therapeutic in nature. The preferred statistical analysis when 2 treatments are being compared is a comparison of mean changes in plaque or gingivitis scores by the independent sample t-test. An analysis of covariance using the baseline scores as the covariate may improve power somewhat. When pairwise t-tests are performed in comparing several treatments, multiple comparison criteria are indicated. Parametric rather than nonparametric tests are preferred since they usually have more power. The square root of whole mouth means in gingivitis clinical trials is the preferred transformation to obtain greater power and to produce approximately normal distributions. |
