Patterns of recurrence after trimodality therapy for esophageal cancer


  • Presented at the 2012 Gastrointestinal Cancers Symposium; January 19-21, 2012; San Francisco, CA.



Patterns of failure after neoadjuvant chemoradiotherapy and surgery for esophageal cancer are poorly defined.


All patients in the current study were treated with trimodality therapy for nonmetastatic esophageal cancer from 1995 to 2009. Locoregional failure included lymph node failure (NF), anastomotic failure, or both. Abdominal paraaortic failure (PAF) was defined as disease recurrence at or below the superior mesenteric artery.


Among 155 patients, the primary tumor location was the upper/middle esophagus in 18%, the lower esophagus in 32%, and the gastroesophageal junction in 50% (adenocarcinoma in 79% and squamous cell carcinoma in 21%) of patients. Staging methods included endoscopic ultrasound (73%), computed tomography (46%), and positron emission tomography/computed tomography (54%). Approximately 40% of patients had American Joint Committee on Cancer stage II disease and 60% had stage III disease. The median follow-up was 1.3 years. The 2-year locoregional control, event-free survival, and overall survival rates were 86%, 36%, and 48%, respectively. The 2-year NF rate was 14%, the isolated NF rate was 3%, and the anastomotic failure rate was 6%. The 2-year PAF rate was 9% and the isolated PAF rate was 5%. PAF was found to be increased among patients with gastroesophageal junction tumors (12% vs 6%), especially for the subset with ≥ 2 clinically involved lymph nodes at the time of diagnosis (19% vs 4%).


Few patients experience isolated NF or PAF as their first disease recurrence. Therefore, it is unlikely that targeting additional regional lymph node basins with radiotherapy would significantly improve clinical outcomes. Cancer 2014;120:2099–2105. © 2014 American Cancer Society.


In patients who undergo surgery for esophageal cancer, the addition of chemoradiotherapy (CRT) has been reported to improve overall survival.[1-3] In turn, the addition of surgery to CRT improves locoregional control, with reported 2-year local control rates of approximately 60% to 70%.[4, 5] Locoregional failure can occur at the surgical anastomosis or in regional lymph nodes.

To the best of our knowledge, there is no international consensus regarding which regional lymph nodes should be included in the RT clinical target volume (CTV) for esophageal cancer. Many studies, including the ongoing Radiation Therapy Oncology Group trial 1010,[6] designed the CTV to include the macroscopic (gross) tumor volume (GTV) plus a 4-cm to 5-cm proximal and distal and a 1-cm to 2-cm lateral margin, in addition to the celiac lymph node axis for distal or gastroesophageal junction (GEJ) tumors.[2, 7] Meanwhile, a randomized trial from Germany in addition included the left gastric artery, lesser curvature, splenic artery, and hepatic artery lymph node basins in the CTV for patients with GEJ tumors.[8]

To our knowledge, there currently are few published studies that characterize patterns of failure and their spatial relationship to RT fields to evaluate the RT CTV. As such, we reviewed patterns of failure among patients with esophageal cancer who underwent trimodality therapy.


This Institutional Review Board-approved study reviewed all patients who were treated with CRT followed by surgery for M0 esophageal cancer from 1995 to 2009 at the study institution. Patients who did not undergo surgery due to progressive disease during CRT, unresectable disease, or medical comorbidities were excluded.

The diagnosis of adenocarcinoma or squamous cell carcinoma (SCC) of the esophagus was confirmed by pathologists at the study institution. All patients underwent routine blood studies, endoscopy, and imaging. Many patients completed positron emission tomography/computed tomography (PET/CT) and endoscopic ultrasound (EUS). Staging was defined based on the 7th edition of the American Joint Committee on Cancer (AJCC).[9]

All patients received concurrent CRT that primarily consisted of a platinum in combination with 5-fluorouracil, a taxane, or both. RT treatment planning was performed using fluoroscopic simulation or 3-dimensional conformal techniques. RT typically consisted of 45 to 50.4 gray (Gy) delivered with megavoltage radiation in 1.8-Gy to 2-Gy fractions. Earlier in the study period (1995-2002), patients were treated with anterior-posterior/posterior-anterior fields to 45 Gy, with or without off-cord opposed lateral boost fields to 50.4 Gy. After 2002, patients were treated with anterior-posterior/posterior-anterior fields to 30 Gy to 36 Gy, followed by opposed oblique fields (avoiding cardiac tissues) to 45 Gy, with or without off-cord opposed lateral boost fields to 50.4 Gy. Primary treatment fields encompassed the macroscopic tumor with an approximate 5-cm proximal/distal margin and 2-cm lateral margin to include the esophagus and periesophageal lymph nodes. For GEJ tumors, a distal gastric margin of 3 cm to 5 cm was generally added to treat the proximal perigastric lymph node basins. The celiac axis lymph node basin was usually treated for distal esophageal and GEJ tumors, as were the supraclavicular (SCV) lymph nodes for cervical/proximal esophageal tumors. Other lymph node stations in the neck, thorax, and abdomen were not routinely treated unless clinically involved. RT boost fields encompassed macroscopic disease with an approximate 2-cm margin. However, given the long study period and differing physicians involved, there was some variability in RT field design.

After neoadjuvant CRT, patients were restaged with CT or PET/CT to exclude disease progression. Our practice was to wait at least 6 weeks after CRT, and longer if necessary, to perform surgery once patients recover approximately 90% of their pre-CRT performance status. Acceptable surgical candidates then underwent 1 of 3 procedures: transhiatal during the first one-half of the study compared with Ivor-Lewis or triincisional esophagogastrectomy during the second one-half, using the stomach as the conduit. The choice of procedure was based on surgeon preference and the location of the tumor.

Patients were usually followed in RT, surgical, and medical oncology clinics every 3 months to 4 months in the first 2 to 3 years and every 6 months to 12 months thereafter. Generally, follow-up imaging, including CT scans, was performed at least every 3 months to 6 months for the first 2 years and at the time of clinical suspicion of disease recurrence. Patterns of first failure were analyzed based on reviewing images and/or reports for radiographic studies. Failures were defined as the appearance of new and enlarging masses, some of which were confirmed by biopsy. Imaging studies were ordered as part of routine follow-up or for clinical suspicion of disease recurrence. Locoregional failure included regional lymph node failure (NF) and anastomotic failure. Regional lymph nodes were defined based on the 7th edition of the AJCC (regional lymph nodes: SCV, mediastinal, and celiac axis).[9] Mediastinal lymph nodes were also classified according to the Mountain and Dresler stations.[10] Abdominal paraaortic failure (PAF) at and/or below the superior mesenteric artery, but above the aortic bifurcation, was scored separately. Isolated NF or PAF were defined as occurring without other sites of local or distant failure.

Statistical Analysis

Locoregional failure (LRF), NF, PAF, event-free survival (EFS), and overall survival (OS) were defined from the time of surgery. LRF events were characterized as failure within the anastomosis or within regional lymph nodes. NF events included failure within the regional lymph nodes. PAF events included failure within the abdominal paraaortic region. All other failures were considered distant failures. Furthermore, event times were censored if any other event, such as distant failure or death, occurred first or if the patient was event free at the time of last follow-up. EFS events included esophageal cancer recurrence or death as a result of any cause, whichever occurred first; event time was censored for patients who were alive or free of disease progression at the end of the study. OS included death as a result of any cause; event time was censored at the date of last follow-up for patients who were alive at the end of the study. For the above endpoints, 2-year estimates with 80% confidence intervals (80% CI) were calculated by the Kaplan-Meier method. All P values were given for 2-tailed tests, and all analyses were performed using SAS statistical software (version 9.3; SAS Institute Inc, Cary, NC).


A total of 230 patients were identified who were treated with CRT for nonmetastatic esophageal cancer at the study institution from 1995 through 2009. Patients who did not undergo surgery due to progressive disease during CRT (27 patients) and those who were unresectable or medically inoperable (42 patients) were excluded from the analysis. In addition, 3 patients died during CRT and 3 patients died during the postoperative period. Therefore, 155 patients were identified who met inclusion criteria.

Patient characteristics are presented in Table 1. The primary tumor location was the upper esophagus in 1% of patients, the middle esophagus in 17% of patients, the lower esophagus in 32% of patients, and the GEJ in 50% of patients (adenocarcinoma in 79% of patients and SCC in 21% of patients). Staging included EUS (73%), CT (46%), and PET/CT (54%). Approximately 4% of patients had AJCC stage I disease, 36% had AJCC stage II disease, and 60% had AJCC stage III disease. Concurrent chemotherapy consisted primarily of a platinum in combination with 5-fluorouracil, a taxane, or both. Primary RT fields targeted the tumor and regional lymph nodes (median dose, 45 Gy) and boost fields encompassed macroscopic disease (median total dose, 50.4 Gy [range, 36 Gy-66 Gy]). RT treatment planning was performed using fluoroscopic simulation in 56% of patients and using 3-dimensional conformal techniques in 44%. Surgical technique was transhiatal (28%), Ivor-Lewis (47%), or triincisional (25%), with a median of 8 lymph nodes dissected (interquartile range, 4 -12 lymph nodes).

Table 1. Patient Characteristics

All Patients (n = 155)

  1. Abbreviations: CT, computed tomography; Gy, gray; PET, positron emission tomography; RT, radiotherapy.

Age, y 
Median (range)60 (34-79)
Stage, no. (%) 
T2N0, grade 26 (4)
T2N0, grade 310 (6)
T1-2 N16 (4)
T3N041 (26)
T1-2N25 (3)
T3N159 (38)
T4N03 (2)
T3N221 (14)
T1-4N33 (2)
T4N1-31 (1)
Location, no. (%) 
Upper2 (1)
Middle27 (17)
Lower49 (32)
Gastroesophageal junction77 (50)
Histology, no. (%) 
Adenocarcinoma123 (79)
Squamous cell carcinoma32 (21)
Staging studies, no. (%) 
CT without PET71 (46)
PET/CT84 (54)
Endoscopic ultrasound113 (73)
Chemotherapy, no. (%) 
5-fluorouracil5 (3)
5-fluorouracil/platinum49 (32)
Platinum/taxane65 (42)
5-fluorouracil/platinum/taxane36 (23)
Total RT dose, Gy 
Median (range)50.4 (36-66)
Surgery, no. (%) 
Transhiatal43 (28)
Ivor-Lewis73 (47)
Triincisional39 (25)


The median follow-up was 1.3 years in all patients and 1.8 years in surviving patients. At the time of last follow-up of the 155 patients, 93 died and 62 were censored, 30 of whom had more than 2 years of follow-up. Causes of death included cancer recurrence (64 patients), complications occurring within 90 days of surgery (9 patients), unknown (11 patients), second cancers (3 patients), respiratory failure occurring > 90 days after surgery (2 patients), congestive heart failure (1 patient), sepsis occurring > 90 days after surgery (1 patient), complications of adjuvant chemotherapy (1 patient), and bleeding after an esophageal dilation (1 patient).

The 2-year LRC, EFS, and OS rates were 85% (80% CI, 79%-89%), 36% (80% CI, 31%-41%), and 48% (80% CI, 43%-54%), respectively (Fig. 1). The 2-year LRC was higher among 55 patients who achieved a pathologic complete response to neoadjuvant CRT compared with those who achieved less than a complete response (94% vs 80%; P = .02). Four patients with pathologic complete response experienced locoregional failure involving mediastinal lymph nodes and/or the anastomosis, 2 of which were isolated. There was no difference in the 2-year LRC rate among patients who were treated with a 2-dimensional versus a 3-dimensional RT technique (15% vs 16%, respectively) or those who underwent a transhiatal versus other surgical approaches (13% vs 17%).

Figure 1.

Locoregional control (LRC), event-free survival (EFS), and overall survival (OS) are shown.

Among all patients, the 2-year NF rate was 14% (80% CI, 10%-19%; 16 patients) and the anastomotic failure rate was 6% (80% CI, 4%-10%; 10 patients). NFs included the SCV (5 patients), mediastinal (10 patients), and/or celiac (1 patient) region. Figure 2 depicts the anatomic location of NF and PAF for patients with tumors of the middle esophagus, lower esophagus, or GEJ. The 2-year isolated NF rate was 3% (80% CI, 1%-5%; 3 patients), the PAF rate was 9% (80% CI, 6%-13%; 10 patients), and the isolated PAF rate was 5% (80% CI, 3%-8%; 6 patients). Approximately 95% of the NF and PAF events occurred outside or near the margin of the primary RT fields. Although the 2-year NF rate was similar between patients with zero versus those with ≥ 1 clinically involved lymph nodes at the time of initial staging (15% vs 13%, respectively), it was higher among patients with ≥ 1 pathologically involved lymph nodes in the surgical specimen (8% vs 26%, respectively).

Figure 2.

Patterns of first failure are shown based on the initial tumor location with each number representing an individual patient with ≥ 1 site(s) of failure. The numbers with a black background indicate isolated failure(s) within the region(s) indicated, whereas those with a white background occurred simultaneously with a distant failure. The regions above correspond to the following Mountain and Dresler lymph node stations: superior mediastinum: 2R/L, 3 and 4 R/L; aortic: 5 and 6; and carina: 7. Paraaortic failures are defined as those at or below the superior mesenteric artery (SMA) but above the aortic bifurcation. GE indicates gastroesophageal; SCV, supraclavicular.

Patterns of failure differed based on the initial location of the esophageal tumor (Fig. 2). For example, the 2-year regional NF rate was higher for patients with middle to lower esophageal tumors at 19% (80% CI, 12%-28%), compared with 9% for patients with GEJ tumors (80% CI, 5%-16%), with the majority of failures occurring within the mediastinum and SCV fossa. Among 49 patients with lower esophageal tumors, 8 patients experienced regional lymph node failure, 2 of which were isolated in nature. Failures in the group of patients with lower esophageal tumors occurred in the following lymph node stations: right or left SCV (3 patients), 2R/L (2 patients), 3 (1 patient), 5 (2 patients), 7 (1 patient), and celiac (1 patient).

Among 77 patients with GEJ tumors, 5 patients experienced regional NF, all of which were simultaneous with distant failure (Fig. 2). These 5 patients failed in the following lymph node stations: right SCV (1 patient), 2R/L (2 patients), 5 (2 patients), and 7 (2 patients).

The 2-year PAF rate was higher for patients with GEJ tumors at 11% (80% CI, 7%-18%), compared with 6% for all other tumor locations (80% CI, 2%-16%). The PAF rate was highest for patients with GEJ tumors who had ≥ 2 clinically involved lymph nodes at the time of initial staging compared with those with 0 to 1 lymph nodes (19% vs 4%).

Among all patients with GEJ tumors, the 2-year isolated PAF rate was 10% (6 patients). Of the 6 patients who experienced isolated PAF, radiographic images of 4 patients taken at the time of first failure were available to review. In the first patient, the PAF was located 1 cm inferior to the superior mesenteric artery and measured 2 cm in superior-to-inferior extent (Fig. 3). In patients 2, 3, and 4, all the lymph node failures started superiorly at the left renal vein and ended inferiorly at 8 cm, 4 cm, or 2.5 cm superior to the aortic bifurcation, respectively. Lymph node masses 1, 2, and 3 were located to the left of the midline of the aorta, and lymph node volume 4 extended to left lateral edge of the inferior vena cava. From the left lateral edge of the aorta, all of the masses extended 2 cm to 2.5 cm laterally. Patients who experienced isolated PAF were treated with a variety of salvage therapies including chemotherapy and/or RT. Two of the 6 patients received salvage CRT, but both later developed other distant metastatic disease.

Figure 3.

Anatomic location of paraaortic failures is shown. Numbers 1 to 4 indicate the anatomic location of paraaortic lymph node recurrence in patients 1 to 4, respectively. SMA indicates superior mesenteric artery; L, left.


The reported LRF rate is 20% to 40% among randomized trials of trimodality therapy for esophageal cancer.[4, 5, 8] This wide range in reported failure rates likely relates to differences in follow-up imaging practices and biologic differences between adenocarcinoma and SCC, as well as potential differences in era of treatment (with or without PET/EUS), RT field design, and surgical technique. The 2-year LRF rate of 15% noted among patients with predominantly adenocarcinoma in the current study compares favorably with other reported series.

The general RT field design in the current study is similar to that of the Cancer and Leukemia Group B trial 9781, as well as the ongoing Radiation Therapy Oncology Group trial 1010.[3, 6] However, to the best of our knowledge, there is no international consensus as to which elective lymph node regions should be treated with RT. Several trials that demonstrated improved outcomes with neoadjuvant CRT in addition to surgery did not electively treat the celiac axis.[2, 11] In contrast, a randomized German trial including patients with GEJ tumors electively treated the celiac axis, left and right cardia, left gastric artery, lesser curvature, splenic artery, and hepatic artery lymph node basins. Although it was closed early due to poor accrual, OS rates favored patients who received CRT versus those treated with chemotherapy alone before surgery (3-year OS rate of 47% vs 28%; P = .07), despite a low RT dose of 30 Gy.[8]

Analysis of the patterns of locoregional disease recurrence after trimodality therapy is important to define in order to evaluate the adequacy of RT field design. In the current study, the regional NF rate was 14%, with 11% of cases occurring simultaneous with a distant failure. Given that failures often occurred in multiple lymph node regions, each with low numbers, it is unlikely that targeting additional regional lymph node basins with RT would significantly improve clinical outcomes. Similarly, a recent preliminary analysis of patients (75% with adenocarcinoma and 25% with SCC) who were treated with neoadjuvant CRT to an elective lymph node region extending 4 cm superior/inferior from the GTV followed by surgery reported a 2-year local recurrence rate of 13%, with 6% of patients developing disease recurrence outside the RT field.[12] This study, along with the current study, suggests that there may be limited benefit to targeting additional lymph node regions with CRT.

Although the conclusions of our study mainly apply to esophageal adenocarcinoma that was diagnosed in 79% of the patients, other studies evaluated the elective lymph node target for patients with esophageal SCC. A retrospective study detailed patterns of disease recurrence among patients treated with trimodality therapy for esophageal SCC with or without elective lymph node irradiation (ENI) to the SCV or celiac axis for tumors of the upper or lower esophagus, respectively.[13] Indeed, many experts advocate targeting these sites with ENI.[3, 7] ENI decreased the elective lymph node failures from 11% to 3% (P = .05), but did not improve the OS or disease-free survival rates. In this study, elective lymph node failures were characterized as occurring either in the SCV/cervical or the celiac/paraaortic regions. This suggests that failures may have occurred in both regional lymph nodes (SCV and celiac axis) and their respective contiguous distant lymph node regions (cervical and paraaortic). Although only 1 patient in our study experienced a celiac lymph node failure, 9% of all patients failed in the contiguous paraaortic lymph node bed, possibly suggesting the presence of subclinical disease in this region at the time of diagnosis.

Although patterns of failure after trimodality therapy were previously poorly characterized, patterns of pathologic lymph node involvement in patients who undergo initial surgery can define areas at greatest risk for disease spread. There are different patterns of lymphatic spread according to the Siewert classification of the anatomic location of the tumor epicenter (Siewert type I tumor epicenters are located > 1 cm above the GEJ, type II are located within 1 cm above to 2 cm below the GEJ, and type III are located > 2 cm but < 5 cm below the GEJ).[14] Siewert type I tumors metastasize to the middle and lower paraesophageal lymph nodes in 38% and 55%, respectively, of cases. In turn, for patients with Siewert type II/III tumors, the rate of lower esophageal involvement is 47% with ≥ 1.5 cm invasion superior to the Z line and 16% with lesser invasion.[15] For all GEJ tumors, the most commonly (approximately 50%) involved abdominal lymph nodes include the paracardial, lesser curvature, and left gastric artery regions, followed by the celiac axis (20%-30%). For patients with T3 to T4 tumors, the reported incidence of splenic artery/hilum involvement is higher for those with Siewert type II/III tumors (41%) versus type I tumors (9%).

Based on this pattern of spread, modifications to standard RT fields (including those used in the current study) for GEJ tumors were suggested as: 1) superior border at the carina to include the middle and lower mediastinal lymph nodes for all Siewert type I tumors and type II/III tumors with ≥ 1.5 cm invasion superior to the Z line; 2) inferior/lateral border to include the splenic artery/hilum lymph nodes down to the left renal hilum for T3 to T4 and Siewert type II/III tumors; and 3) inferior/lateral border to include additional gastric lymph nodes (gastroepiploic and greater curvature) for all T3 to T4 tumors.[15]

Patterns of failure from the current study can be used to assess the adequacy of current RT fields, although we were limited by the lack of accurate endoscopic tumor measurements for all patients, precluding strict Siewert classification. When considering the above recommendations, for all GEJ tumors, we only saw 2 failures at the carina and a few in the superior mediastinum, and none of these mediastinal failures was isolated. At least in the patients in the current study, the superior border set at the GTV plus 5 cm appeared to be adequate. This must be interpreted within the context of many patients undergoing transthoracic dissection of middle and lower mediastinal lymph nodes. In addition, we were limited by the relatively small patient numbers, short follow-up time, and few locoregional failures. The short follow-up time of 1.3 years among all patients and 1.8 years among survivors was mainly due to the large percentage of patients who died (60%).

For GEJ tumors, we did not observe any failures specifically within the splenic or gastric lymph node regions. However, we did observe 7 failures in the adjacent paraaortic/left renal hilum region, 6 of which were isolated. In 4 of the 7 cases, images were available for review and lymph node masses were located left and lateral to the aorta/vena cava. According to the lymph node classification of the Japanese Gastric Cancer Association, 1 PAF involved 16a2 (celiac trunk superiorly to left renal vein inferiorly), 1 involved 16b1 (left renal vein to inferior mesenteric artery), and 2 involved both 16b1 and 16b2 (inferior mesenteric artery to the aortic bifurcation).[16] This same pattern of retroperitoneal lymph node spread from GEJ tumors was reported in a small surgical series in which select patients underwent a D4 lymphadenectomy (including dissection of the paraaortic region) at the time of upfront gastrectomy. In this series, 4 of 15 patients with stage T2 to T4, Siewert type III adenocarcinoma had positive paraaortic lymph nodes.[17]

Another study of trimodality therapy for patients with esophageal cancer reported that 4% of patients had first sites of distant failure within the paraaortic lymph nodes.[18] Our 2-year PAF rate was higher at 9%, which was likely related to the finding that a greater percentage of patients in the current study had adenocarcinoma (79% vs 68%), with the majority involving the GEJ. We found the highest rates of PAF (19%) among patients with adenocarcinoma of the GEJ who had ≥ 2 clinically involved lymph nodes at the time of diagnosis. Therefore, these patients may be likely to have a greater risk of subclinical paraaortic involvement at the time of diagnosis.

To the best of our knowledge, the issue of whether it would be feasible or efficacious to include the paraaortic region with CRT for patients with esophageal cancer has not been addressed. Although there is no consensus regarding the ideal cutoff for the risk of lymphatic spread to include a region in the CTV, some have proposed a value of 10% to 20%.[15] This decision involves careful consideration of the increased toxicity that may result from covering additional regions, such as bowel, kidney, and bone marrow toxicity with paraaortic RT. In addition, there may be diminishing value for increasing RT field size if patients with subclinical paraaortic involvement are less likely to be cured. The current study was limited by its retrospective nature and the finding that patients were treated over 14 years by different physicians, and there was short follow-up time. Additional study is required to confirm the high risk of PAF among patients with GEJ tumors with multiple involved lymph nodes and the potential impact of paraaortic RT.


No specific funding was disclosed.


Dr. D'Amico has acted as a paid consultant for Scanlan for work performed outside of the current study.