Annals of Oncology 14:1154-1155, 2003
© 2003 European Society for Medical Oncology
Letters to the Editor |
Reply to Letter to the Editor "Whose ‘wish bias’?", by M. Velasco-Garrido & R. Busse (Ann Oncol 2003; 14: 1154)
Istituto di Ricerche Farmacologiche "Mario Negri", Department of Oncology, Milan, Italy
*E-mail: fossati@marionegri.it
Velasco-Garrido and Busse [1] maintain that the response to chemotherapy of patients with advanced breast cancer (ABC) could have been decreasing over the last two decades because the latest trials entered patients with worse prognosis. In support of this hypothesis they produce a graph suggesting a response rate tendency over time resembling that shown in our figure 1 [2], and report a significant positive correlation between the overall response rate in doxorubicin-containing and doxorubicin-free arms. Since the response to doxorubicin-based chemotherapy was shown to have a negative trend over time, the authors conclude that the other chemotherapy regimens paralleled that trend. We disagree with such a conclusion, because we deem both the database and the analytical approach inadequate.
We think that the vast array of chemotherapy regimens used in the doxorubicin-free arm makes this database unsuitable for exploring the existence of a pattern in response rate. In fact, when we identified the temporal trend we applied a multivariate analysis, and we had at least one drug as a common denominator through out the 29 trials considered, i.e. doxorubicin. If we evaluate overall response rate over time in the doxorubicin-free arms (2.617 patients) with an univariate logistic analysis we do find a significant odds ratio (OR) for the ‘time’ covariate [OR = 0.90; 95% confidence interval (CI) = 0.83–0.97], but when we analyse the data set with the same parsimonious multivariate model used in our paper, introducing the covariates that take into consideration previous adjuvant chemotherapy and the number of drugs in the chemotherapy regimen, the ‘time’ covariate is no longer significant (OR = 0.96; 95% CI = 0.87–1.06).
Moreover, we suppose that the coefficient of correlation reported by the authors is a parametric estimator (Pearson’s correlation coefficient) but since the doxorubicin response rate is not normally distributed (Shapiro-Wilk test, P = 0.046) this analytical approach might not be correct. Using the non-parametric Spearman’s correlation coefficient we obtained a value of 0.481 (P = 0.008) but, again, the coefficient flattened to 0.348, losing its statistical significance (P = 0.0814), when is was adjusted for the number of drugs and the previous use of adjuvant chemotherapy. Therefore, we conclude that the suggested progressive widening of enrolment to patients with less responsive ABC, taking place in more recent trials, can not be proved with this database and analytical approach.
R. Fossati* C. Confalonieri G. Apolone S. Cavuto & S. GarattiniIstituto di Ricerche Farmacologiche "Mario Negri", Department of Oncology, Milan, Italy (*E-mail: fossati@marionegri.it)
References
1. Velasco-Garrido M, Busse R. Whose ‘wish bias’? Ann Oncol 2003; 14: 000–000.
2. Fossati R, Confalonieri C, Apolone G et al. Does a drug do better when it is new? Ann Oncol 2002; 13: 470–473.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||