Statistical methods
Statistical analysis was performed with a multi-level logistic
regression model for patency when data for patency was given in 2-by-2
tables stratified by type of technique (on-pump or off-pump) and
harvesting technique. The main predictor variable for patency was the
two different techniques, on- or off-pump. Most patients had more than
one graft, with a maximum number of three grafts per patient. Multiple
grafts per patient might introduce dependencies between the
observations. We corrected for clustered observations in the calculation
of confidence intervals (CIs), with each patient forming a separate
cluster.
With the aim of investigating a possible equivalence between on- and
off-pump, the graphical approach recommended by Fleming11 and by Blackwelder 12 was used.
An essential point in this approach is to define, based on experience,
limits of what can be regarded as equivalent outcomes with respect to
patency. Therefore, no formal statistical hypothesis testing was
conducted, and instead results are shown in graphics with CIs for the
differences in patency rates between the two techniques. Through these
CIs, possible equivalences between on- and off-pump can be visualised.
When the CI does not cover the point of no difference in patency between
the methods (the ‘0-point’), equivalence between the techniques is
rejected. If, on the other hand, the CI covers the ‘0-point’ and in
addition the CI does not stretch outside the equivalence limits there is
an evidence for equivalence. Finally, if the ‘0-point’ is covered by the
CI but at least one of the two limits of CI stretches outside the
equivalence limits the result is inclusive, possibly due to lack of
power. Computations were performed with STATA version 15
(www.stata.com).