Prognostic Value of the Circumferential Resection Margin in Esophageal Cancer Patients After Neoadjuvant Chemoradiotherapy

Background Circumferential resection margins (CRM) for esophageal cancer (EC), defined by the College of American Pathologists (CAP; >0 mm) or the Royal College of Pathologists (RCP; >1 mm) as tumor-free (R0), are based on a surgery-alone approach. We evaluated the usefulness of both definitions in current practice with neoadjuvant chemoradiotherapy (nCRT). Methods CRMs were measured in 209 patients (104 with nCRT) with locally advanced EC after transthoracic esophagectomy. Local recurrence and cancer related death were scored as events. Patients were followed for at least 2 years or until death. Prognostic factors (P < 0.1 in univariate analyses) for 2-year disease-free survival (DFS) and local recurrence-free survival (LRFS) were incorporated in multivariate Cox regression analyses. Both CRM measurements were analyzed separately and prognostic cutoff values (0–1.0 mm) were assessed in both groups. Results Independent prognostic factors (P < 0.05) for 2-year DFS were tumor length, lymph node ratio, angioinvasion, and CAP R0 in the surgery-alone group and pN stage (P < 0.01) in the nCRT group. Prognostic factors (P < 0.05) for 2-year LRFS were CAP, lymph node ratio, and tumor length in the surgery-alone group, and CAP and grade in the nCRT group. Optimal CRM cutoff values between 0.0 and 0.2 mm were prognostic for 2-year DFS in the surgery-alone and at 0.3 mm for the nCRT group. Conclusions nCRT affected the CRM cutoff values. After nCRT, the CRM R0 according to the CAP was only prognostic for 2-year LRFS. However, in the surgery-alone group, it was prognostic for both the 2-year DFS and LRFS.

surgery-alone control group. [13][14][15] Two studies, with only squamous cell carcinoma, showed a significant better survival rate in patients with a CRM [ 1 mm, whereas no survival benefit was observed in R0 resections according to the CAP and RCP in a study with only T3 stage adenocarcinomas. [13][14][15] We assessed the optimal CRM cutoff point and the prognostic value of R0 resections according to the CAP and RCP criteria in EC patients treated either with nCRT or surgery alone.

PATIENTS AND METHODS
Data collection of this explorative retrospective study was provided from a prospective maintained database of EC patients according to the national guidelines and the rules approved by the local ethical commission (www. ccmo.nl). We included only patients with a locally advanced curatively resectable EC (stage II-III) treated between 1997 and 2013, in whom the CRM was adequately assessed by our expert pathologists. Of the patients treated with nCRT (n = 127) between 2005 and 2013, 23 were excluded because of the following criteria: incomplete medical records (n = 0), postoperative mortality (death within 90 days or in-hospital, n = 10), progressive disease within 3 months after surgery or microscopic irradical (R1; tumor cells \1 mm) longitudinal margins (n = 0) or follow-up \24 months (n = 13). Based on these exclusion criteria, a reference group of surgery-alone treated patients (n = 105) was constructed. Patients and tumor-related factors were matched and were equally distributed between both groups (Table 1).
Tumors staged according to the 6th TNM edition were recoded into the 7th edition. 16,17 Before 2000 (n = 11), staging consisted of endoscopic ultrasonography (EUS) with fine-needle aspiration (FNA), computed tomography (CT) of the neck, thorax, and abdomen and occasionally 18-F-fluorodeoxyglucose positron emission tomography (FDG-PET, n = 8). After 2000, a standard FDG-PET was added, which was replaced by FDG-PET/CT after 2009. Two weeks after nCRT, patients were restaged with a CT thorax and abdomen.

Treatment
All patients underwent a transthoracic esophagectomy with en bloc dissection of regional mediastinal and abdominal (including the celiac trunk region) lymph nodes. Patients with nCRT were treated according to the Dutch Chemoradiotherapy for Oesophageal Cancer Followed by Surgery Study (CROSS) regimen, consisting of intravenous paclitaxel (50 mg/m 2 ) and carboplatin (AUC: 2 ml/min), administered five times during a 5-week concurrent radiation period (41.4 Gy/23 fractions of 1.8 Gy). 1 Before 2009,  patients received nCRT based on their participation in the  CROSS trial, from 2009 onwards nCRT became standard  of care for locally advanced EC patients (T1-4aN1-3, T2-4aN0-3; n = 75).

Pathology
Resected specimens were examined according to a standardized protocol by two specialized gastrointestinal pathologists. The resected specimen was pinned on a Styrofoam plate by the surgeon, enabling accurate pathological assessment of the marked Clinical Tumor Volume and Gross Tumor Volume areas in patients treated with nCRT. 18 CRM was measured according to the method of Quirke; the specimens were inked with Indian ink and fixed in formalin during 24 h. 6 The specimens were sliced into transverse cross-sections of 0.5 cm for macroscopic assessment and sampling of at least two sections with the smallest CRM. 2 The CRM was microscopically assessed on hematoxylin and eosin stained samples in tenths of millimetres. Furthermore, the pT-stage, pN-stage, the lymph-node ratio ([0.2 metastatic lymph node ratio), number of positive lymph nodes ([4), histological tumor type, tumor grade, angioinvasion, and perineural tumor growth were assessed.

Follow-up
Patients were followed for at least 2 years or until death, every 3 months during the first year after surgery, every 6 months in the second year, and every year thereafter for the next 10 years. Tumor recurrence was defined as histo/cytologically proven, suspected radiological imaging, or clinically evident recurrence. Local recurrence included recurrent disease at the anastomotic site or in the original tumor/mediastinal bed.

Statistical Analysis
Distribution of continuous patient characteristics was reported as median [interquartile range] and categorical variables were reported in numbers and percentages. The patients groups were compared with the Mann-Whitney test for continuous variables and v 2 or Fisher exact test for categorical response variables. Kaplan-Meier curves and log-rank test determine the 5-year disease-free survival (DFS) and local recurrence-free survival (LRFS) of both CRM definitions. Prognostic values of all variables for 2year DFS were assessed with univariate Cox regression analysis. Factors within the univariate analysis were: age, tumor type, and grade (G1-2 vs. G3-4), clinical T and N stage, tumor length ([5 cm, measured endoscopic or with CT), treatment type (nCRT or surgery alone), and pathologic outcome: T and N stage, number of LN metastases ([4), and metastatic lymph node ratio ([0.2), perineural growth, and angioinvasion. Multivariate Cox regression was performed by incorporating all variables with a P value \0.1 on univariate analysis. Both, the CAP (CRM [0 mm) year LRFS was assessed with multivariate Cox regression analyses in both treatment groups. To assess the optimal cutoff value of the CRM on 2-year DFS, an explorative analysis was performed in both groups. Univariate analyses were undertaken to assess the prognostic value of all cutoff values (from 0.0 to 1.0 mm). The observed interval is based on the assumption that the expected optimal CRM cutoff should be between 0.0 and 1.0 mm. The Akaike Information Criterion (AIC), which quantifies the quality of a statistical model for a set of data was used to indirectly compare the prognostic value of the CAP and RCP model. 19 It penalizes the number of explanatory variables by adding twice the number of variables in the model to the -2 log likelihood; in a formula AIC = -2 log likelihood ?2 k, in which k is the number of explanatory variables in the model. The model with the lowest AIC was considered to be most prognostic. The backwards likelihood ratio method was used in the Cox regression analysis. Analyses were performed with SPSS version 22.

RESULTS
Patient characteristics are summarized in Table 1. All nCRT patients with CAP-R1 resections (n = 9; 8.7 %) had stage pT3. Of the 27 (25.7 %) R1 resections in patients treated with surgery alone, 24 had stage pT3, 1 had pT2, and 2 had stage pT4a disease. The median CRM differed significantly with 3. Prognostic Value of the CAP and RCP Criteria Figure 1 displays the DFS of both treatment groups, with a R0 resection or involved CRM (R1 resection) according to CAP (Fig. 1a) and RCP (Fig. 1b). With the log-rank test, the CAP definition was prognostic for 5-year DFS in both the surgery (P = 0.008) and nCRT group (P \ 0.001) and the RCP definition was prognostic in the nCRT group (P \ 0.001) but not in the surgery group (P = 0.071). The 5-year DFS was not different (P = 0.131) between CAP R1 patients treated with or without nCRT but differed (P = 0.031) between patients with an RCP R1 resection in both groups. Table 2 displays all prognostic factors with a P \ 0.1 on univariate analysis and Table 3 shows the multivariate Cox regression models containing either the CAP or RCP for 2year DFS and LRFS in both groups.

DISCUSSION
The prognostic value of the circumferential margin (CRM) has been proven in EC patients after surgery alone, but its significance after neoadjuvant treatment is not well defined yet. This study conducted in stage II-III EC patients showed that both definitions of a free CRM were not prognostic for 2-year DFS in patients treated with nCRT. The CAP definition ([0 mm), however, was an independent prognostic factor for 2-year DFS in the surgery-alone group and for LRFS in the nCRT and surgery-alone group. The optimal CRM cutoff value for 2-year DFS was 0.3 and between 0.0 and 0.2 mm in the nCRT and surgery-alone group, respectively.
This study is one of the first to assess the optimal cutoff value of the CRM after nCRT; previously published studies used either the RCP or CAP criteria of a free CRM. Although neoadjuvant treatment decreases the rate of R1 resection by transversal and sagittal tumor reduction, the induced fibrosis may contain different amounts of undetectable viable tumor cells. 1 Therefore, the CRM assessment depends upon accurate histological examination of residual tumor, which might be related to tumor heterogeneity. CRM [1 mm showed to be prognostic, but    4 In our study, only one patient staged as ypT2 disease had a R1 resection, due to extensive angioinvasive tumor growth within the CRM, which depends more on biologic aggressiveness rather than poor surgery. Another factor that might influence the CRM is the used surgical  24 As in patients treated with nCRT, the optimal CRM definition in surgically treated patients also is unclear. Two recent meta-analyses showed that both CRM definitions were associated with a poor survival, although the CAP criteria differentiated higher-risk groups. 11,12 Moreover Chan et al. found that the CAP definition, based on the hazard ratio and subgroup analysis, had a prognostic advantage over the RCP criteria. 12 Concordant to these results, we found that the optimal CRM cutoff value in the surgery-alone group, analyzed with the Akaike Information Criterion, was the CAP.

Follow-up (months)
Beside the CRM, lymph node metastasis associated variables were important prognostic factors in this study; lymph node ratio[0.2 was independent prognostic for both 2-year DFS and LRFS in the surgery-alone group and pNstage was the only prognostic factor for 2-year DFS in the nCRT group. One meta-analysis, which underlined the importance of lymph node metastasis, indicated that nodal metastases appeared to negate the prognostic value of the CRM. 12 Moreover, the presence of lymph node metastases and an involved CRM indicated a more advanced-staged disease. 26 Another prognostic factor in surgery-alone patients was the tumor length, which is in correspondence with previously published data. 27 Pultrum et al. assessed the optimal CRM in surgically treated patients using the area under the curve (AUC) analysis on receiver operating curves (ROC, which does not incorporate the time factor. 2 A method that includes the time factor is the more complex time-dependent ROC method according to Heagerty et al. 28 For our limited explorative study, however, we prefer to use multivariate Cox regression analysis and suggest validating the results in a larger cohort.

CONCLUSIONS
This study showed that both definitions of a tumor-free CRM (CAP [ 0 mm, RCP [ 1 mm) were not prognostic for DFS in patients treated with nCRT. A free CRM according the CAP definition was prognostic for 2-year DFS in the surgery-alone group and an optimal CRM cutoff between 0.0 and 0.2 and at 0.3 mm in the surgeryalone and nCRT groups, respectively. These findings should be validated in a large, prospective study.