Objective. To report methodological difficulties with restoration surv
ival data in controlled clinical trials on Class II amalgam restoratio
ns using a split-mouth design. The advantages and disadvantages of dif
ferent ways of handling these data are described. Methods. Three stati
stical methods (Kaplan-Meier estimation, logistic regression with rand
om component and Friedman's statistic) are compared using data from a
controlled clinical trial in which cavosurface angle (regular or non-s
tandard) and cavity wall finish (applied or not applied) determine fou
r treatment modalities of New True Dentalloy(R) restorations. Results.
In this study logistic regression with a random component yields the
best interpretable results. Cavity wall finish in combination with a r
egular cavosurface angle is indicated as the worst and cavity wall fin
ish in combination with a non-standard cavosurface angle as the best t
reatment when the criterion is replacement or not within 15 years. Con
clusions. The dependency between restorations within a patient needs t
o be taken into account. Logistic regression with a random component m
ay be a valuable alternative to very advanced statistical survival mod
elling when restricting the research question to replacement within a
certain time interval is not a major problem. (C) 1998 Elsevier Scienc
e Ltd. All rights reserved.