A Bayesian approach facilitates interpretation of results in meta-analysis: the example of periodontal disease

Tags: Oral
Brignardello-Petersen R1, Carrasco-Labra A2, Pintor F3, Ulloa C4, Tomlinson G5
1Faculty of Dentistry, University of Chile, Chile. Institute of Health Policy, Management and Evaluation, University of Toronto, Canada, 2Faculty of Dentistry, University of Chile, Chile. Department of Clinical Epidemiology and Biostatistics, McMaster University. Canada, 3Faculty of Dentistry, Universidad Mayor, Chile, 4School of Dentistry, Faculty of Medicine, Universidad Austral de Chile. Faculty of Dentistry, University of Chile, Chile, 5Institute of Health Policy, Management and Evaluation, University of Toronto. Department of Medicine, University Health Network. Canada

Background: Systematic reviews and meta-analysis about the treatment for periodontal disease have demonstrated statistically significant benefits of antibiotics as adjuvant therapy. However, the magnitude of the treatment effect has a clinical significance that is questionable. In this context, the more natural inferences provided by the Bayesian approach seem particularly helpful. As a result of a Bayesian meta-analysis, questions with direct application to clinical decision-making can be answered.

Objectives: To illustrate the benefits of the use of Bayesian meta-analysis, and how it enhances the interpretation of the systematic review results, through the use of an example in dentistry (periodontal disease treatment).

Methods: We performed a systematic review of the literature. We conducted searches in PubMED and EMBASE up to January 2013 to retrieve all randomized clinical trials studying the effectiveness of metronidazole as an adjunctive therapy in patients with chronic periodontitis. Two independent reviewers screened abstracts, full-text articles and extracted data. The results of the eligible trials were combined using a Bayesian Meta-analysis, using the software WinBUGS.

Results: We will present the main findings of the meta-analysis. The point estimate and credible interval of the treatment effect, and inferences such as the probability of specified treatment effects or the probability of a range of treatment effects will be used to illustrate how Bayesian meta-analysis enhances the interpretation of results. Finally, we will contrast the applicability of these inferences with those from the traditional meta-analysis.

Conclusions: Bayesian meta-analysis provides inferences more compatible with decision-making. These inferences can have an even higher value in settings in which the clinical significance of the results is in question.