The Effect of Censoring on the Statistical Evaluation of Composite-to-Composite Bond Strength

Purpose: To determine the effect of Weibull analysis with censoring of cohesive failures on the evaluation of the effect of surface treatment and adhesive on the microtensile composite-composite bond strength. Materials and Methods: Surfaces of Filtek Z250 (3M Oral Care) specimens aged in distilled water for five months at 37 degrees C were ground with SiC paper P320 or air abraded with Al2O3 and built up with the same composite using Optibond FL (OPF, Kerr), Clearfil SE Bond (CSE, Kuraray Noritake) and Gluma Comfort Bond (GLU, Kulzer) adhesives. After sectioning and trimming, the specimens were stored at 37 degrees C in distilled water for 24 h and then tested in a microtensile bond strength setup. Complete datasets of all strength data and adhesive datasets containing adhesively failed specimens were analyzed with a one-way Welch ANOVA at alpha = 0.05 and compared with the results of Weibull analysis applied on complete and adhesive datasets as well as complete datasets right censored for cohesive failures. Weibull shape (m) and scale (sigma(o)) parameters were estimated using the maximum likelihood method and used to calculate bond strength (sigma(0.10)) at 10% probability of failure as a measure of system performance. The surface morphology was studied using light and scanning electron microscopy. Results: While ANOVA indicated a significant effect of surface treatment and non-significant effect of the adhesive, the Weibull analysis results depended on bond strength characteristics and censoring. The scale parameter of the complete dataset indicating a significant effect of surface treatment for all adhesives became statistically non-significant for CSE after censoring. When this effect was evaluated by sigma(0.10), the non-significant effect of surface treatment for OPF and GLU became statistically significant due to censoring. The effect of the adhesive evaluated by the scale parameter was non-significant both for surfaces as well as complete and censored datasets. On the other hand, s(0.10) suggested significant differences between adhesives on the ground surface for complete datasets which became statistically non-significant for the censored datasets. Conclusion: This study indicates that the results of statistical evaluation of bond strength can be affected not only by the experimental set-up and statistical model but also the bond strength characteristics and the manner in which the strengths of cohesively failed specimens are included in calculations.