| |
Quantitative light-induced fluorescence (QLF) images were taken of 20 human molar teeth that had been previously subjected to a demineralising solution (lactic acid buffer, pH 4.5, F- 0.05 ppm) to create artificial lesions. Following examination of the images, 16 were chosen to represent a range of lesion size and severity. Three copies were made of the images and each was allocated a different file name. Seven examiners in two centres were asked to analyse each of the 16 images on 3 occasions, with at least 7 days between each attempt. Simple instructions describing the analysis procedure were supplied and examiners asked to adhere to these directions. The examiners were asked to rate each of the 16 teeth on their first attempt both quantitatively (5-point scale) and qualitatively in terms of difficulty of analysis. Data reported were the .Q at 5% threshold for each tooth on each of 3 attempts. Using the alpha model of reliability analysis, the interexaminer agreement was excellent (A = 0.9884, interexaminer correlations 0.9315, variance 0.0015). This was confirmed by the failure of ANOVA to detect significant differences between any of the examiners' mean scores. The 3 attempts of each examiner were used to determine intra-examiner reliability. This ranged from A = 0.99 to A = 0.81, with a mean of A = 0.93. ANOVA confirmed the lack of significant difference between the 3 attempts on all but one examiner. Following pairwise t tests, this examiner had differences between attempts 1 and 2, 2 and 3 (p = 0.006, 0.015). The mean reported difficulty level was 2.5 (B 0.71). This study has demonstrated that the analysis stage of QLF is reliable between examiners and within multiple attempts by the same examiner, when analysing in vitro lesions. |