Use of the limits of agreement approach in periodontology

PURPOSE: To discuss the statistical approaches that have been traditionally used to compare measures in periodontal research, highlighting its strengths and weaknesses and, finally, to suggest the use of the limits of agreement method of Altman and Bland (1983) as an alternative method to address this question.

MATERIALS AND METHODS: Using a sample dataset of clinical periodontal measures as a background, the different possible approaches for agreement assessment are discussed and statistical and clinical points are considered. Eight hundred and forty repeated measures, belonging to the training phase of a clinical study, were performed in five individuals presenting different severities of periodontal conditions. The use of correlation coefficient, comparison of means, linear regression technique, Kappa coefficient, intra-class correlation coefficient and means versus differences plot is demonstrated.

RESULTS: Most of the methods are applied without the appropriate care, resulting in misleading interpretations. The information that arises from some of the methods used so far is poorly informative and adds little understanding to the operational characteristics of the raters or instruments. Some of the resulting information from the correlation coefficient and kappa coefficient may even be false or not applicable for the entire range of possible values.

CONCLUSIONS: The graphical approach that plots differences against means, including the 95% limits of agreement estimated by the mean difference +/- 1.96 standard deviation of the differences is the most informative approach and its application should be considered for continuous clinical periodontal measures.