• © 2007 by American Society of Clinical Oncology

In Reply

Dr Ruhstaller et al suggest that our conclusion¹ is not supported by our data and that the trial design has some major limitations. Their first remark is that chemoradiotherapy was not the same in both arms—this is not a bias and simply reflects the essential difference between strategic trials and drug-based trials. When comparing two drugs, everything must be identical in both arms except the study drug, though when comparing two strategies, an experimental strategy is compared with the therapeutic strategy recognized as the reference. Chemoradiotherapy and surgery can be considered as the reference strategy on the basis of three consistent meta-analyses.² The experimental strategy may differ considering the doses of chemotherapy or radiotherapy. Indeed, it is not recommended to apply high doses of radiation before surgery, and postoperative chemotherapy or chemoradiotherapy is seldom possible in these patients. Moreover, no randomized studies showed any survival benefit for postoperative adjuvant treatment. As an example, the dissymmetry of treatment schemes between arms was present in the trial by Al-Sarraf et al,³ which proved the superiority of chemoradiotherapy on radiation alone in inoperable esophageal cancer. The dose of radiation was 64 Gy in the radiotherapy alone arm versus 50 Gy in the chemoradiotherapy arm.

The second remark puts forward the problem of inaccurate staging, based on the figure of 75% of the patients having R0 resection in the surgery arm. Staging of esophageal cancer is indeed difficult, and the prediction of R0 resectability is a challenge. However, in the published trials of surgery versus preoperative chemoradiotherapy in patients deemed operable, R0 resection rates vary between 69% and 91% in the surgery arm and between 69% and 96% in the preoperative chemoradiotherapy arm, in intent to treat, which is the only nonbiased way of analyzing a randomized trial.^4-7 The percentage of R0 resectability in the Fédération Francophone de Cancérologie Digestive 9102 was 75% in intent to treat, and is in agreement with the published data, considering that only locally advanced tumors were included in the trial. The R0 resectability rate raised up to 85% if the patients refusing surgery were not taken into account. This figure compares well with the 87% R0 resectability rate in the series at the M.D. Anderson Cancer Center between 1986 and 1996, the period during which FFCD 9102 was conducted.⁸

In FFCD 9102, we supported an academic and public health point of view with the aim of reflecting the impact of strategy in clinical practice. It included taking into account (1) the insufficiency of staging by clinicians when they have to decide which treatment to apply and (2) the patient's choice regarding therapeutic decision. Indeed, the interest of randomizing large numbers of patients is to offset possible imbalance resulting from the difficulty of precisely staging esophageal cancer.

However, to strengthen our conclusion, we provided in Figure 3 the survival curves in per protocol (ie, removing the patients who refused surgery from the surgery arm and including them in the chemoradiotherapy arm). Overall survivals remained similar. Thus, our result cannot be explained by any bias related to misstaging or patient refusal.

The third remark is based on a misinterpretation of the published results. The real percentage of eligible patients qualified as nonresponders, and thus not randomized, was 26% and not 46%, as stated by Dr Ruhstaller et al. However, we agree on the fact that there is a part of incertitude concerning the actual response soon after the end of chemoradiotherapy. In a previous phase II trial—in which preoperative evaluation after chemoradiotherapy included esophagoscopy with biopsies, an esophagogram, and a thoracoabdominal computed tomography (CT) scan—we found that clinical response overestimated pathological response in 30% and underestimated it in 30% of the cases, though the imprecision was in distinguishing between complete or partial response.⁹ In no cases was a progression or a stabilization classified as a partial or complete response. Other authors reported similar findings.¹⁰ Consequently, we simplified the work-up, with the advantage of shortening the delay before random assignment. Before 2000, no investigation, including CT scan, could define precisely the response to chemoradiotherapy, whereas nowadays the fluorodeoxyglucose positron-emission tomography (FDG-PET) predicts objective response with an 86% accuracy.¹¹ Moreover, we consider that not randomly assigning patients who did not present at least a partial response on esophagogram and a symptomatic improvement was not detrimental for the patients, as the recommended therapy in this case was surgery. However, in agreement with the suggestion of Dr Ruhstaller et al's, as a dedicated study is under way to precise the treatment, evolution, and survival of the nonrandomized patients, taking into account the reasons of nonrandomization. In the meantime, it is not possible to state that patients' nonrandomized for lack of efficacy of the induction treatment had a specially bad prognosis. The hazard ratio (HR) was only 1.22 compared with randomized patients. The HRs for the other causes of nonrandomization were superior: 1.39 in case of contraindication to either treatment, and 1.63 in case of refusal to be randomized.

We do not understand the fourth criticism. It is indeed because of the mortality and morbidity due to surgery in esophageal cancer, and not rewarded by a high survival rate, that this trial was undertaken. The mortality rate in the surgery arm, calculated during the first 3 months after registration, was 9% (not 11%) versus 1%. It explains the similitude of the number of deaths, though the percentage of cancer deaths was lower in the surgery arm (45% versus 55%, not 64% versus 79%, as calculated by Dr Ruhstaller et al). This is not surprising, as a patient dying postoperatively cannot die eventually from cancer.

We are afraid we do not support the conclusion of Dr Ruhstaller et al about the insufficient methodology of our trial. First, randomization was made to offset the difficulty of accurately staging esophageal cancer, which is the principal criticism. Obviously, the future trials should include in the work-up endoscopic ultrasonography and FDG-PET, but these tools were not available when the trial was conducted. Second, the strongest end point to evaluate the efficacy of a treatment in a highly lethal disease is overall survival, which was used in our trial. Third, we did not conclude that surgery was to be abandoned, but that in responding patients, overall survival was equivalent to that with chemoradiotherapy alone, though resection improved local recurrence-free survival. It is clear that we need new trials to better discriminate the respective indications of surgery and nonoperative therapies in esophageal cancer, including biologic predictive factors, but we consider that, with its limitations, our study, supported by the randomized study of Stahl et al,¹² provides a valuable basis for further research.