Detection of longitudinal change in periodontitis.

In any longitudinal study of periodontitis, it is important to determine when change in probing depth or clinical attachment level occurs as soon as possible. In this study, simulation was used to compare three statistical methods for detecting change in linear periodontal measurement under a variety of conditions including type of change, number of follow-up visits, magnitude of linear measurement and examiner variance. The statistical methods included: 1) testing the slope of a regression line, 2) a comparison of running medians and 3) the cumulative sum (cusum) method. Three types of change and random examiner error were imposed on baseline linear measurements. The types of change included a gradual change, a burst, and rapid loss followed by regeneration. The results indicated that all three methods yielded similar results when the change was gradual. However, the regression line was less sensitive to bursts of change than the other methods. Overestimation of examiner error for the running median and cusum methods increased specificity and decreased sensitivity. Underestimation of examiner variance decreased specificity and increased sensitivity of the cusum and running median methods, while the regression was unaffected by estimates of examiner variance. If preventive or therapeutic action does not need to be taken immediately, either the running median or the cusum would be a good choice for detecting change. If immediate intervention is needed, the cusum appears to be the method of choice for detecting change at specific sites in longitudinal periodontal studies.