The use of analysis of covariance to increase precision wtih DMFS measures

Development and application of a prediction model for dental caries

The development and validation of a caries prediction model comprising 13 sociodemographic and dental examination variables on Grade 1 and Grade 5 children in the National Preventive Dentistry Demonstration Program are described. The objective was to derive a method of predicting children at high risk to caries early in order that preventive measures might be undertaken. True high risk children were defined in two ways: highest 25% of children based on their 4-yr DMFS increment, and their total DMFS score at the end of the study. In both cases, children predicted to be at high risk were defined as the 25% with the highest discriminant score. Discriminant function and logistic regression analyses were used to determine the extent to which the 13 variables collectively discriminated between true high risk and non-high risk children so defined. Sensitivity was approximately 0.50 and specificity around 0.82, using the 4-yr increment as the criterion for defining true high risk, and approximately 0.64 and 0.88, respectively, using the final DMFS score for defining true high risk.

Applicability of multiple covariance analysis in caries studies

The relationship between pre- and posttreatment DMFS measures in clinical trials of dentifrices may be quadratic rather than linear and, if linear, the slope may differ from one. Given either of these eventualities the traditional DMFS increment measurement does not utilize all the information available in the pre-DMFS baseline score. If nonlinearity is present, analysis of multiple covariance with parameters derived from a polynomial regression is indicated.

Why covariance analysis should replace increment analysis of variance in caries studies

Consider a caries study where experimental units are a) randomly assigned to groups, b) premeasured on DMFS, c) administered a specified treatment depending on group membership, and d) postmeasured on DMFS. Traditional analysis of these data consists of analysis of variance of the increment scores (increment ANOVA). In the place of increment ANOVA, others have suggested analysis of covariance with the postmeasure as criterion and the premeasure of covariate (ANOCOV). The present paper examines and documents the following: 1) Increment ANOVA and ANOCOV test the same null hypothesis. 2) Increment ANOVA and ANOCOV have exactly the same assumptions. 3) Increment ANOVA is usually less precise than ANOCOV. 4) The same concern for violations of assumptions must be expressed with increment ANOVA as with ANOCOV (see No. 2 above). 5) ANOCOV should replace increment ANOVA in caries studies (see points 1-4).