Box 98 General guidelines for the analysis and reporting of QL data

♦ A small number of hypotheses should be pre-specified to allow definitive analysis to be performed on them (see Chapter 6). All other analyses should be regarded as exploratory and hypothesis generating.

♦ The statistical methods of analysis must be described in sufficient detail so that other researchers could repeat the analysis.

♦ In all analyses all patients must be accounted for.

♦ It is often useful to analyse the data in more than one way to confirm any differences observed are not model dependent.

♦ It is important to specify how patients who died before reaching the endpoint were dealt with in the analysis.

♦ Graphical presentations can be helpful. In such presentations, where possible, it is important to specify the number of patients at risk (or contributing to each section of the plot) by treatment group beneath the time axis in the plot, similar to that for a Kaplan-Meier plot introduced in Section 9.3.4.

♦ Non-parametric tests, such as the Wilcoxon or the Mann-Whitney test may be more appropriate for QL data as the data are often skewed with ceiling or floor effects (for example, patients with no symptoms cannot get better, while patients with the worst category for a symptom cannot get worse).

0 0

Post a comment