Much of the literature on the analysis of longitudinal data using the intermediate observations assumes that the data are complete, i.e., no missing observations in the repeated measurements of the same individual. Yet longitudinal studies typically have some missing data that make standard repeated measurement procedures inapplicable. Often dental data of this kind are analyzed by looking at only on interval, the longest available. Wei and Johnson (1985) have proposed a procedure which allows the investigator to determine whether a new treatment consistently maintains an improvement over the standard therapy for the entire study period. The test procedure allows different patterns of missing observations in the comparison groups. Data collected in a dental clinical trial to evaluate DMFS increment of two treatment groups is used to illustrate this method. The test is based on a linear combination of individual t- test statistics on increment data collected at three different time points. The weights in the linear combination are chosen to produce locally the most powerful test when the sample size is large. The weights depend on the patient attrition rate at each time point and the covariance structure of the t-test at different time points. By combining DMFS increment from multiple time points in this clinical trial, the method is able to show that one group is consistently better than the other group over the entire period (p=0.03). This result could not be shown statistically by analyzing the data in the coventional way, i.e., by testing the difference in increment between the baseline and final follow-up only (p=0.09).
