This proposal involves a secondary data analysis of 186 subjects from a NIDCR-funded (DE013967) clinical trial to determine if sensory alteration following a Bilateral Sagittal Split Osteotomy (BSSO) procedure differed for those patients who were given opening exercises only following surgery and those who received sensory re-training exercises in conjunction with standard opening exercises. The primary outcome, an average of five questionnaire items rated on a scale from (1) """"""""no problem"""""""" to (7) """"""""serious problem"""""""" did not significantly differ between the two treatment groups when aggregated over time. An item-specific supplementary analysis based upon longitudinal proportional odds cumulative logit models, however, found significant treatment differences for some of the items at some visits.
The aims of this proposal are to definitively assess the efficacy of sensory re-training by: (1) testing and relaxing the proportional odds assumption in cumulative logit models via partial proportional odds whereby proportional odds are assumed for only some of the predictors; this is achieved by applying alternating logistic regressions (ALR) procedure, an implementation of generalized estimating equations (GEE) useful for modeling the marginal probability that an ordinal response is above a certain cut-point and the within-cluster association structure of the correlated binary data resulting from multiple logits and visits; (2) conducting a simulation study to determine whether the statistical methods applied to the sensory re-training data are valid for the particular models and sample sizes considered in terms of coverage rates of 95% confidence intervals and test sizes; (3) developing, evaluating, and applying to the sensory re-training data a modified ALR procedure that has better small sample properties than standard ALR as assessed in the simulation study; and (4) writing comprehensive statistical software implementing modified and standard ALR as user-friendly options. The simulation study results should provide general guidance on sample size requirements for use of ALR and modified ALR as general tools for the analysis of correlated binary data. Given that complex correlation structures are common in dental research, the modified ALR method should increase the scope of application of GEE-based methodologies. The project proposes secondary data analyses of 186 subjects from a clinical trial to examine whether the addition of sensory re-training exercises following jaw surgery improves the recovery of facial sensation above that achieved with standard opening exercises. New statistical methodology and models are applied to the data to examine more thoroughly the treatment comparisons studied in the original NIDCR-funded clinical trial. ? ? ?
Preisser, J S; Phillips, C; Perin, J et al. (2011) Regression models for patient-reported measures having ordered categories recorded on multiple occasions. Community Dent Oral Epidemiol 39:154-63 |