Statistical Analysis
In most cases, the exact sample size for Dixon’s UDM could not be
determined in advance. When 6 crossovers (conversion from successful
block to unsuccessful block or vice versa) had occurred, we ceased to
recruit patient[16, 17]. We can see that at least
20–40 patients will be required to provide reliable estimates of the
target dose in our simulation studies in anesthesia trials using the
Dixon’s UDM[18]. Our study recruited 29 patients
and achieved this goal.
To explore the target dose ED50, four statistical approaches were used,
including 3 parametric estimates of dose responsive
curve[18]: linear, linear-logarithmic and
exponential regressions, and one nonparametric model: the centered
isotonic regression, which was only for assuming a nondecreasing dose
and response relationship[19].
Residual standard errors, a statistical tool to determine the goodness
of fit, which analyses how well a set of data points fit with the actual
model, were calculated for all four statistical approaches.