47 3 performance. A subject was asked to chew on this tablet 20 times in order to mix the two colors. The tablet is then flattened, pressed to a thickness of 2 mm, and scanned on both sides using a high-quality scanner (Epson® V750, Long Beach, CA, USA). The scanned images are then processed using Adobe Photoshop CS3 extended (Adobe, San Jose, CA, USA). The histograms of both sides of the flattened and scanned wax tablet are added to obtain red and blue intensity distributions. The spread of the color intensities is measured.12 Subjects were instructed to chew 20 times on two different tablets in order to test reliability. The interim period between the two tests was approximately 2 hours for patients with HNC, and 30 minutes for healthy controls, with the same testing conditions for all participants. Statistical Analyses Test-retest reproducibility of the MAT was tested by a two-way random, single measurement, absolute agreement, Intraclass Correlation Coefficient (ICC2,1), calculated as , in which MSR = mean square of rows; MSE = mean square for error; MSC = mean square for columns; k = number of measurements; and n = number of subjects. Cut-off points for the ICC were chosen as poor (<0.5), moderate (0.5 to 0.75), good (0.75 to 0.90), and excellent (>0.90).21,22 Standard Error of Measurement (SEM) was calculated as SEM = .23 For the SD, the Standard Deviation of the difference between the two MATs was used. SEM percent change was calculated as SEM%= , in which = the mean of all measurements of test and retest. Smallest detectable change (SDC) was calculated as SDC = !"# $ .24,25 The SDC percent change was calculated as SDC% = % , in which = the mean of all measurements of test and retest. In order to check for systematic bias, variability and agreement, Bland-Altman plots were constructed by plotting the test-retest difference versus the mean value of the test and retest. Agreement between test and retest was summarized using the mean difference and SD of the difference, and the 95% Limits of Agreement (LoA) were calculated as LoA = &'( ) !"# .26 A power analysis was conducted, in which an ICC of at least 0.7 was expected. A p1 value of 0.9 was chosen, therefore the sample size had to be at least 18.4.27 In addition, a comparable study in children with cerebral palsy showed a sample size of 25 to 30 patients,15 therefore it was chosen to include at least 30 subjects. Data were tested for normality using a Shapiro-Wilk test. Because data were not normally distributed, a Wilcoxon Signed Ranks test was conducted to examine differences between test and retest for both patients with HNC and healthy controls, and a Kruskal-Wallis test was conducted to examine differences in MAT scores according to age and sex. A Mann-