Multilevel modeling for the analysis of longitudinal periodontal data

Objectives: To investigate the effects of time-varying independent variables and the time changing effects of the independent variables on the healing response in longitudinal periodontal data using multilevel modeling. Methods: Data from 40 patients, 20 smokers and 20 non-smokers, were used to investigate the effects of smoking on the reduction in probing pocket depth (PPD) after non-surgical periodontal treatment. The follow-up changes from baseline in PPD for the 594 initially diseased sites (baseline PPD≥5mm) were analyzed. In order to investigate the effects of time-varying independent variables (updated full mouth % plaque and bleeding on probing (BOP), presence of plaque and BOP at the sites along the study period) and the time changing effects of the independent variables (smoking status, location of the sites, etc.) on reduction in PPD, a 4-level regression model with subject as level-4, tooth as level-3, site as level-2 and time as level-1 was adopted. Results: The 4-level model showed that non-smokers had greater reduction in PPD than smokers (p<0.05), the differences at the 6- (0.77mm) and 12-month follow-ups (0.84mm) were similar but were greater than that at 3-month follow-up (0.41mm). Posterior teeth and sites at lingual surfaces had significantly smaller reductions in PPD (0.35mm and 0.44mm, respectively, p<0.05) constantly throughout the study period. None of the time-varying independent variables showed significant effect. Conclusions: A 4-level multilevel model can provide appropriate statistical inference on the effects of time-varying independent variables and the time changing effects of the independent variables on the healing responses in longitudinal periodontal data.