Multilevel Analysis of Periodontal Treatment Response in Male Smokers

To account for the hierarchical structure of periodontal disease measurements, i.e. sites measurements clustered within teeth and then teeth clustered within individuals applied to follow up outcomes of periodontal therapy, analysis using a multilevel approach is adopted. Objectives: To investigate baseline factors which may predict non-surgical periodontal treatment response using multilevel multiple regression in Chinese male moderate-to-severe periodontitis smokers (test) and never-smokers (control), who were matched for age and periodontal disease severity. Methods: 40 patients (mean age 45.6 ± 6.3 years) were recruited. 20 were smokers (≥ 10 cigarettes/day). They received non-surgical periodontal therapy provided by experienced dental hygienists. 12-month reduction in probing pocket depth (PPD) of 5814 sites distributed on 969 teeth in these 40 patients was analyzed by a multilevel approach. A 3-level model was considered: site at level-1, tooth at level-2 and subject at level-3. 14 independent predictor variables, 7 on subject-level, 2 on tooth-level and 5 on site-level were included in the multilevel multiple regression. The analysis was performed using the software MLwiN version 2.1. Results: Significant variations existed at all three levels of the multilevel structure (p<0.001). Multilevel multiple regression showed that 1 predictor on subject-level, 2 on tooth-level and 4 on site-level were significantly associated with 12-month reduction in PPD (p<0.001) Non-smokers, lower teeth, anterior teeth, buccal surface, sites with absence of plaque, deeper PPD and higher PAL at baseline were associated with greater 12-month reduction in PPD. The variations at each level were reduced markedly with the inclusion of the 7 predictors in the multilevel multiple regression (subject: 76%, tooth: 52%, site: 56%). Conclusions: The use of multilevel analysis enables researchers to incorporate predictor variables measured at different levels in the same model. Multilevel analysis appears to be a useful statistical tool for the analysis of periodontal treatment data.