Analysis:

key points

 

i)                    Use of confidence intervals
The assessment of the difference between the treatment groups is greatly enhanced by the use of confidence intervals.  The confidence interval for the difference in treatment means contains the mean differences that are compatible with the data from the trial, where compatible is interpreted as not being rejected at the 5% level (for a 95% interval) by a hypothesis test.  This is particularly valuable when the difference does not reach a high level of significance, such as P < 0.05.

ii)                   Uses and abuses of baseline values
Baseline values, obtained just prior to randomization, can be useful in the analysis of a trial.  Comparing changes from baseline between the groups can provide a more precise comparison than just using outcomes, provided that the outcome and baseline have a sufficiently large correlation.  However, comparisons of P-values between groups can be found in the literature and is a flawed analysis.

iii)                 Use of analysis of covariance
If there is a baseline imbalance between groups, then it is reasonable to assume that this will result in an outcome imbalance, even if there is no treatment effect.  Allowing for this in the analysis requires the analysis of covariance, which will always give a more precise analysis than either analysing outcomes alone or changes from baseline.

 

 

 

Return to Main Outline document