Balance
is expected beforehand with respect to observed and unobserved
patient characteristics. However, after the trial is completed,
clinically relevant imbalances may be noted on one or more predictors,
especially when the size of the trial is relatively small. Such
imbalances are due to pure chance when a proper randomization
procedure has been followed. Imbalance may be corrected by including
predictors and treatment as independent variables in a regression
model ("adjusted analysis"). Note that the adjusted analysis
should generally be pre-specified, and not determined by the
actual imbalance that is observed.