If neither carry-over nor period effects are thought to be a problem, then an appropriate analysis of continuous data from a two-period, two-intervention cross-over trial is a paired t-test. This evaluates the value of ‘measurement on experimental intervention (E)’ minus ‘measurement on control intervention (C)’ separately for each participant. The mean and standard error of these difference measures are the building blocks of an effect estimate and a statistical test. The effect estimate may be included in a meta-analysis using the generic inverse-variance method in RevMan.
A paired analysis is possible if the data in any one of the following bullet points is available:
· individual participant data from the paper or by correspondence with the trialist;
· the mean and standard deviation (or standard error) of the participant-specific differences between experimental intervention (E) and control intervention (C) measurements;
· the mean difference and one of the following: (i) a t-statistic from a paired t-test; (ii) a P value from a paired t-test; (iii) a confidence interval from a paired analysis;
· a graph of measurements on experimental intervention (E) and control intervention (C) from which individual data values can be extracted, as long as matched measurements for each individual can be identified as such.
For details see Elbourne et al. (Elbourne 2002).
If results are available broken by the particular sequence each participant received, then analyses that adjust for period effects are straightforward (e.g. as outlined in Chapter 3 of Senn (Senn 2002)).