Members / Participants Membership Information Clinical Trials Meetings Publications Correlative Science / Tumour Bank Notices Fellowships Committees Mango

ANALYSIS OF ORDINAL QUALITY OF LIFE RESPONSE DATA:WHITHER SIMPLE CHI-SQUARE TESTS?

Dongsheng Tu, Jianhua Liu and Joe Pater

NCIC Clinical Trials Group, Queen’s University, 10 Stuart St., Kingston, Ontario, Canada

Recently, the Quality of Life Committee of the NCIC Clinical Trials Group (NCIC CTG) proposed the following method as a standard protocol for the analysis of the longitudinal quality of life (QoL) data: For all the patients who had a baseline QoL measurement and at least one measurement post the baseline, the change score from the baseline is first calculated for each post-baseline measurement. The QoL response of the patients are then defined based on all of their change scores and a clinically significant difference D for the change score: as improved, if there is at least one post-baseline change score which is higher than D; as worsened, if all the post-baseline change scores are lower than or equal to D but there is at least one post-baseline change score which is lower than –D; and as stable, if all the post-baseline change scores are between –D and D. A simple chi-square test is then applied to compare the difference in the QoL response profiles between treatment arms.

One might argue, though, that since the QoL response is by nature ordinal, a statistical test taking this into account might be more powerful. In this presentation, using the QoL data from several of the clinical trials conducted by the NCIC CTG, we compare the results of the analyses for the QoL response data based on the following tests: simple chi-square test, Fisher’s exact test, Wilcoxon exact rank sum test, exact rank test with Savage score, and the convex hull test as proposed by Berger et al. (1998, Biometrics, 54, 1541-1550).