A new graphic for quality adjusted life years (Q-TWiST) survival analysis: the Q-TWiST plot

One of the challenges of interpreting a Quality-adjusted time without symptoms of disease and toxicity (Q-TWiST) analysis is examining the sensitivity of conclusions that may be drawn to varying values of the utility coefficients for days with toxicity and days after disease progression. We present a graphic that parsimoniously displays the impact on median Q-TWiST survival across treatment groups of varying values of the utility coefficients. The goal of the graphic is to present a concise Q-TWiST analysis. We use Zhao and Tsiatis (Biometrika 1997; 84(2): 339-348) to adjust for the bias in Kaplan-Meier (K-M) estimates. The graphic contains bounds that approximate points for which statistical significance would be achieved by comparing the median Q-TWiST survival between treatment alternatives for each value of the utility coefficients. The plot may be generalized to compare Q-TWiST means, medians or percentiles across treatment groups. We demonstrate the application of the Q-TWiST plot through a re-analysis of a randomized phase III North Central Cancer Treatment Group (NCCTG) clinical trial of recombinant Interferon-2alpha in patients with malignant melanoma. We explore alternative options to customize the graphic representation for other data sets drawn from several NCCTG clinical trials.