Outcome measures in split mouth caries trials and their statistical evaluation

• Published in: Community dentistry and oral epidemiology Vol. 22; no. 3; p. 192
• Main Authors: Riordan, P J, FitzGerald, P E
• Format: Journal Article
• Language: English
• Published: Denmark 01.06.1994
• Subjects: Bias; Confidence Intervals; Dental Caries - etiology; Dental Caries - prevention & control; Dental Materials - therapeutic use; Humans; Incidence; Models, Statistical; Pit and Fissure Sealants - therapeutic use; Risk Factors; Treatment Outcome
• ISSN: 0301-5661
• DOI: 10.1111/j.1600-0528.1994.tb01839.x

The split mouth study design in trials of fissure sealants and restorative materials neatly controls for confounding by many of the variables associated with poor retention of sealants and occurrence of dental caries. Traditionally, the outcome measures used in split mouth trials have been material retention, (per cent) effectiveness and net gain. A survey of the literature revealed that a large proportion of split mouth studies report no statistical evaluation of outcome measures. In those studies in which statistical evaluation had been conducted, McNemar's X2 was the test most frequently used. This statistic is appropriate for comparing differences between "success" and "failure" tooth pairs (or "positives" and "negatives" in split mouth terminology) but it cannot evaluate directly effectiveness and net gain. The distributions of effectiveness and net gain are different and it would be desirable to estimate confidence intervals for them. In this paper, we consider these statistics, suggest methods by which confidence intervals may be calculated, and provide examples of the calculations. We demonstrate the close relationship between effectiveness (as used in split mouth trials) and relative effect and relative risk (as used in general epidemiological analysis) and recommend that relative risk should be the preferred outcome measure for split mouth trials. Whatever outcome measure is chosen in split mouth trials should always be subjected to statistical evaluation, preferably by the calculation of confidence intervals.