Box 99 Alternative approaches to analysis of QL data

♦ Graphical summaries

♦ Scores at a common specific assessment point (landmark analysis)

♦ Summary measures such as mean or worst score

♦ Complex models

A number of alternative methods for analysing the difference between treatment are available, and the general approaches are summarized in Box 9.9. We then go on to discuss each of these analyses in more detail.

In the absence of a clear rationale to use one analytical approach ahead of another, it is useful to conduct a number of types of analyses to ensure that any conclusions are not just a function of the (arbitrary) analytical approach adopted. In a detailed analysis of longitudinal data from a randomized clinical trial in lung cancer Qian and colleagues [19] employ a variety of summary scores and a complex model-based approach to assess the sensitivity of any conclusions to the analytical approach. They show, that for this dataset at least, the different approaches generally produce the same broad conclusions. They also emphasize that by examining the reasons for inconsistent results using different approaches allowed them to reject some conclusions, which they otherwise might have inappropriately emphasized.

Whichever approach is used a major problem that is always present is how to include individuals with missing data and in particular how we should allow for those individuals who have died during the course of collecting QL data. Approaches to missing data were discussed above. Hollen etal. [20] strongly encourage adjusting for death (scoring death as the worst possible score) and give an example where without such adjustment, QL erroneously appears to improve with time. However, while it is logical to put patients who have died in the worst category if you are assessing, say, response, it may not be logical for other symptoms such as cough.

0 0

Post a comment