644870 research-article2016
TAM0010.1177/1758834016644870Therapeutic Advances in Medical OncologyO. Dizdar et al.
Therapeutic Advances in Medical Oncology
Letter to the Editor
Interpretation of survival data: which graphical analysis is appropriate?
2016, Vol. 8(4) 312 DOI: 10.1177/ 1758834016644870 © The Author(s), 2016. Reprints and permissions: http://www.sagepub.co.uk/ journalsPermissions.nav
Omer Dizdar, Serkan Akin and Mutlu Hayran
We read with great interest the comprehensive review on the treatment of unresectable metastatic colorectal cancer by Chibaudel and colleagues [Chibaudel et al. 2015]. The authors analyzed current data on the initial treatment, maintenance therapy, second-line therapy and beyond, in a concise fashion. As mentioned in the manuscript, the choice of the biologic agent in the first-line treatment of RAS wild-type metastatic colorectal cancer is debatable because of the inconsistent results of the two large randomized trials. We have concerns on the interpretation of the combined information from CALGB/SWOG 80405 and AIO FIRE-3 studies in this manuscript. In our opinion, for the two studies in their Figure 2, the analytical approach, the inference, and the graphical presentation (Figure 2C) derived from this approach are imprecise. One key assumption in the Cox model is proportional hazards, which means that the survival curves for two strata must have hazard functions that are proportional over time [Cox, 1972]. When the proportional hazards assumption is violated, various approaches may help overcome the problem, by providing insight into the nature of the nonproportionality and how to modify the Cox model to best fit the data. However, the graphical analysis performed by the authors involves some degree of subjectivity in interpretation. In Graphs A and B, the boundaries of ‘superiority’, ‘uncertain’ and ‘no superiority’ are subjective, considering the effect size and the missing confidence intervals. The percentages on the combined analysis in Graph C is produced simply by finding the average of the values of the corresponding percentages in Graphs A and B. Such a descriptive approach ignores any methodological improvement which
Ther Adv Med Oncol
followed the introduction of the meta-analytical thinking by iconic statisticians such as Pearson and Fisher at the beginning of the 20th century [O’Rourke, 2007]. Therefore, the conclusion statement proposing the ‘probably unquestionable superiority of EGFR monoclonal antibodies over bevacizumab in 20% of all patients’ is below the standards of statistical inference. The extended exploration of graphics might be helpful tool in describing time segments with nonproportional hazards, but should not be the basis of simple mathematical calculations upon which a treatment decision is made.
Correspondence to: Omer Dizdar, MD Department of Preventive Oncology, Hacettepe University Cancer Institute, Altındag, 06100, Ankara, Turkey [email protected] Serkan Akin, MD Department of Medical Oncology, Hacettepe University Cancer Institute, Ankara, Turkey Mutlu Hayran, MD, PhD Department of Preventive Oncology, Hacettepe University Cancer Institute, Ankara, Turkey
Funding The author(s) received no financial support for the research, authorship, and/or publication of this article. Conflict of interest statement The author(s) declared that there is no conflict of interest.
References Chibaudel, B., Tournigand, C., Bonnetain, F., Richa, H., Benetkiewicz, M., André, T. et al. (2015) Therapeutic strategy in unresectable metastatic colorectal cancer: an updated review. Ther Adv Med Oncol 7: 153–169. Cox, D. (1972) Regression models and life tables (with discussion). J R Statist Soc B 34: 187–220. O’Rourke, K. (2007) An historical perspective on meta-analysis: dealing quantitatively with varying study results. J R Soc Med 100: 579–582.
Visit SAGE journals online http://tam.sagepub.com
SAGE journals
312 http://tam.sagepub.com
TAM0010.1177/1758834016644870Therapeutic Advances in Medical OncologyO. Dizdar et al.
Therapeutic Advances in Medical Oncology
Letter to the Editor
Interpretation of survival data: which graphical analysis is appropriate?
2016, Vol. 8(4) 312 DOI: 10.1177/ 1758834016644870 © The Author(s), 2016. Reprints and permissions: http://www.sagepub.co.uk/ journalsPermissions.nav
Omer Dizdar, Serkan Akin and Mutlu Hayran
We read with great interest the comprehensive review on the treatment of unresectable metastatic colorectal cancer by Chibaudel and colleagues [Chibaudel et al. 2015]. The authors analyzed current data on the initial treatment, maintenance therapy, second-line therapy and beyond, in a concise fashion. As mentioned in the manuscript, the choice of the biologic agent in the first-line treatment of RAS wild-type metastatic colorectal cancer is debatable because of the inconsistent results of the two large randomized trials. We have concerns on the interpretation of the combined information from CALGB/SWOG 80405 and AIO FIRE-3 studies in this manuscript. In our opinion, for the two studies in their Figure 2, the analytical approach, the inference, and the graphical presentation (Figure 2C) derived from this approach are imprecise. One key assumption in the Cox model is proportional hazards, which means that the survival curves for two strata must have hazard functions that are proportional over time [Cox, 1972]. When the proportional hazards assumption is violated, various approaches may help overcome the problem, by providing insight into the nature of the nonproportionality and how to modify the Cox model to best fit the data. However, the graphical analysis performed by the authors involves some degree of subjectivity in interpretation. In Graphs A and B, the boundaries of ‘superiority’, ‘uncertain’ and ‘no superiority’ are subjective, considering the effect size and the missing confidence intervals. The percentages on the combined analysis in Graph C is produced simply by finding the average of the values of the corresponding percentages in Graphs A and B. Such a descriptive approach ignores any methodological improvement which
Ther Adv Med Oncol
followed the introduction of the meta-analytical thinking by iconic statisticians such as Pearson and Fisher at the beginning of the 20th century [O’Rourke, 2007]. Therefore, the conclusion statement proposing the ‘probably unquestionable superiority of EGFR monoclonal antibodies over bevacizumab in 20% of all patients’ is below the standards of statistical inference. The extended exploration of graphics might be helpful tool in describing time segments with nonproportional hazards, but should not be the basis of simple mathematical calculations upon which a treatment decision is made.
Correspondence to: Omer Dizdar, MD Department of Preventive Oncology, Hacettepe University Cancer Institute, Altındag, 06100, Ankara, Turkey [email protected] Serkan Akin, MD Department of Medical Oncology, Hacettepe University Cancer Institute, Ankara, Turkey Mutlu Hayran, MD, PhD Department of Preventive Oncology, Hacettepe University Cancer Institute, Ankara, Turkey
Funding The author(s) received no financial support for the research, authorship, and/or publication of this article. Conflict of interest statement The author(s) declared that there is no conflict of interest.
References Chibaudel, B., Tournigand, C., Bonnetain, F., Richa, H., Benetkiewicz, M., André, T. et al. (2015) Therapeutic strategy in unresectable metastatic colorectal cancer: an updated review. Ther Adv Med Oncol 7: 153–169. Cox, D. (1972) Regression models and life tables (with discussion). J R Statist Soc B 34: 187–220. O’Rourke, K. (2007) An historical perspective on meta-analysis: dealing quantitatively with varying study results. J R Soc Med 100: 579–582.
Visit SAGE journals online http://tam.sagepub.com
SAGE journals
312 http://tam.sagepub.com
