A quality-adjusted survival meta-analysis of adjuvant chemotherapy for premenopausal breast cancer


  • Bernard F. Cole,

    1. Department of Community Health and Division of Applied Mathematics, Brown University, Box F, Providence, Rhode Island 02912, U.S.A.
    Search for more papers by this author
  • Richard D. Gelber

    1. Department of Pediatrics, Harvard Medical School, Department of Biostatistics, Harvard School of Public Health and Division of Biostatics, Dana-Farber Cancer Institute, 44 Binney Street, Boston, Massachusetts 02115, U.S.A.
    Search for more papers by this author


The purpose of this paper is to develop and apply a meta-analysis methodology, that does not require patient-level data, for comparing treatments in terms of quality-of-life-adjusted survival. As a motivating example, we considered adjuvant chemotherapy for breast cancer. This therapy has been shown to offer an improvement in recurrence-free and overall survival, especially for youger women, but its acute toxic effects discourage some physicians from prescribing it. To determine whether the benefit of adjuvant chemotherapy treatment outweighs its costs in terms of toxic effects, we performed a meta-analysis of quality-adjusted survival based on data from 1229 patients, 49 years of age or younger, randomized in eight clinical trials that compared chemotherapy versus no adjuvant systemic therapy. We conducted the meta-analysis by performing a quality-adjusted survival analysis known as a Q-TWiST analysis on each trial. A Q-TWiST analysis allows one to make treatment comparisons that incorporate differences in quality of life associated with various health states. In this analysis, we define as health states the periods of time patients spend: (i) with subjective toxic effects of chemotherapy; (ii) without symptoms of recurrence and toxicity, and (iii) following disease recurrence. We assigned weights to each health state which reflect their relative value in terms of quality of life and allowed them to vary in a sensitivity analysis. We then combined the individual trial results in a meta-analysis, using a multivariate regression model, in such a way that we could easily perform an overall sensitivity analysis. Individual patient-level data are not required to perform this meta-analysis methodology if the individual Q-TWiST analysis results for each trial are available.