A multilevel analysis of factors affecting pocket probing depth in patients responding differently to periodontal...

3 distinct levels are involved in the periodontal inflammatory process: site, tooth, and individual. By focusing attention on the levels in the population, multilevel or hierarchical modelling (MLM) enables the researcher to understand where and how the effects at the levels involved are occurring. The aim of this paper is therefore to analyse the progression of periodontal disease using analytical models that consider the level hierarchy. 22 patients with periodontitis, in previous reports described as either non-responsive or responsive to periodontal treatment, were investigated. In the multilevel modelling method (MLM), the site pocket probing depth (PPD) is summarised in 3 parameters: the overall mean, the between-individual variance, and the within-individual, between-site variance. The model can readily be extended to include independent variables for sites, teeth and individuals. If these variables are important determinants of PPD, their inclusion in the model will lead to a reduction in residual variances between sites, teeth and individuals. The PPDs were used for construction of a PPD change variable (cPPD). This variable, together with the final registrations of PPD (fPPD) alone, were used as dependent variables in the MLM. Independent predictor variables, 12 on site-level, 3 on tooth-level, and 19 on individual-level, were constructed. The total number of sites assessed was 2236 distributed on 559 teeth in 22 subjects. Initially, a fixed, fully unconditional model (models A and E) was assessed, where no predictor variables were specified at any level. Different random-intercept models (B-D, F-H) were then calculated where the independent variables were inserted in blocks relating to each level. The variance components at all 3 levels were significantly larger than zero. This indicates that MLM is recommended for analysing the present data. The inserted predictors showed 100% sensitivity relating to the subject-level variance. Subsequent testing of the patient with disease or at high risk of disease would have to focus on diagnostic tests aimed at the individual teeth and sites. These tests would need to have a balance of sensitivity and specificity. Thus, by using multilevel modelling, the theoretical understanding of important factors in the pathogenesis of periodontitis is stimulated.