Members of University of Rochester School of Medicine and Dentistry Foregut Group: Jaffer A. Ajani, MD (Department of Gastrointestinal Medical Oncology, The University of Texas M. D. Anderson Cancer Center, Houston, Texas); Jeremy Erasmus, Jr., MD (Department of Diagnostic Radiology, The University of Texas M. D. Anderson Cancer Center, Houston, Texas); Ritsuko Komaki, MD (Department of Radiation Oncology, The University of Texas M. D. Anderson Cancer Center, Houston, Texas); Jeffrey H. Lee, MD (Gastrointestinal Hepatology and Nutrition, The University of Texas M. D. Anderson Cancer Center, Houston, Texas); Dipen Maru, MD (Department of Pathology, The University of Texas M. D. Anderson Cancer Center, Houston, Texas); Reza Mehran, MD, David C. Rice, MD, Jack A. Roth, MD, Ara A. Vaporciyan, MD, and Garrett L. Walsh, MD (Department of Thoracic and Cardiovascular Surgery, The University of Texas M. D. Anderson Cancer Center, Houston, Texas); and Jeffrey H. Peters, MD (Department of Surgery, University of Rochester School of Medicine and Dentistry, Rochester, New York).
Pathologic esophageal tumor length (pL) is an independent predictor of long-term survival. However, whether patients with longer (high-risk) tumors can be identified by endoscopy before surgery has not been established. The objective of the current study was to determine the value of endoscopically measured tumor length (cL) in predicting overall survival in patients with esophageal adenocarcinoma.
All patients with esophageal adenocarcinoma who had undergone resection without neoadjuvant therapy and who had documented preoperative endoscopy findings were identified retrospectively by using prospectively collected databases at 2 institutions: The University of Texas M. D. Anderson Cancer Center (n = 164; training set) and University of Rochester Medical Center (n = 109; validation set). Esophageal tumors were assessed preoperatively by endoscopy for cL, depth (cT), and lymph node involvement (cN). Univariate and multivariate analyses of cL and other standard prognostic factors were performed.
In the training set, cL was correlated directly with pL (Pearson correlation [r] = 0.683; P < .001). Regression tree analyses suggested an optimum cutoff point of cL >2 cm to identify patients with decreased long-term survival (5-year survival rate: cL >2 cm, 29%; cL ≤2 cm, 78%; P < .001). Multivariate Cox regression analysis demonstrated that cL >2 cm was an independent risk factor for long-term survival (hazard ratio, 2.3; 95% confidence interval, 1.1-4.4; P = .02) even after controlling for age, cT, and cN. Validation with the validation dataset confirmed that cL was correlated directly with pL (r = 0.657; P < .001) and predicted long-term survival using a cL cutoff point of >2 cm (hazard ratio, 2.8; 95% confidence interval, 1.4-5.8; P = .004; univariate analysis).
The incidence of esophageal cancer in the United States has continued to rise over the past 2 decades.1-3 With the advent of neoadjuvant therapy, we have started to make significant progress in terms of response rates and overall prognosis in patients with esophageal cancer.4, 5 However, the selection of patients to receive neoadjuvant treatment is based strictly on TNM staging, which does not take tumor length into account. In the current American Joint Committee on Cancer (AJCC) staging system for esophageal tumors, stage depends on the depth of the tumor (T classification), lymph node involvement (N classification), and distant metastasis (M classification). Before 1987, the AJCC staging system used tumor length (T1, <5 cm; T2, >5 cm; and T3, evidence of extraesophageal spread) and circumferential involvement to predict patient prognosis. However, at the 1987 AJCC annual meeting, the current TNM staging system was adopted, in which tumor length is not a staging criterion.
We previously demonstrated that pathologic tumor length determined after surgical resection is an independent predictor of survival in patients with esophageal cancer, even after controlling for other standard prognostic factors, such as pathologic tumor depth, pathologic lymph node involvement, and tumor grade.6, 7 However, to our knowledge, whether patients with longer tumors can be identified by endoscopy before surgical resection has not been fully evaluated. In the current study, we sought to determine the value of endoscopically measured tumor length in predicting overall patient survival and, thus, identifying those high-risk patients with esophageal cancer who may benefit from neoadjuvant therapy.
MATERIALS AND METHODS
Acquisition of Clinical Data
All patients who underwent esophagectomy for adenocarcinoma of the esophagus or gastroesophageal junction at The University of Texas M. D. Anderson Cancer Center (M. D. Anderson) between January 1995 and December 2007 were reviewed. Patients who had received neoadjuvant therapy in the past were excluded, as were those who underwent an emergent, salvage, or redo operation. Patients who did not undergo a curative resection and patients with incomplete data or perioperative mortality also were excluded. In total, 164 patients from M. D. Anderson ultimately were included, and these patients were used as a training set. By using the same inclusion and exclusion criteria, a second group of patients treated at the University of Rochester School of Medicine and Dentistry was identified; these 109 patients were used as a validation set. This review was approved by the institutional review boards of both institutions.
Before resection, all patients were staged clinically with esophagogastroduodenoscopy or endoscopic ultrasonography, computed tomography of the chest and abdomen, positron emission tomography, or fine-needle aspirations when available. Medical records were reviewed for demographic factors (age, sex, race, and functional status as determined by the American Society of Anesthesiologists scale8), tumor-related factors (histology, location, grade, endoscopically determined tumor length, and clinical T and N classification), type of resection, and other prognostic factors.
Endoscopic esophageal tumor length was determined on the basis of a review of each patient's preoperative esophagogastroduodenoscopy or endoscopic ultrasonography report. Endoscopic tumor length was determined by the longest craniocaudal axial length. If multiple nodules were identified, then the longest nodule length was recorded as the endoscopic tumor length. If the endoscopist did not identify a tumor, mass, lesion, or nodule, then the length of the polyp (the long axis of the greatest dimension of the polyp), ulcerated area, mucosal defect, nodular mucosa, or focal thickening of the esophageal wall/stricture was used. Pathologic tumor length was determined from each patient's pathology report. The pathologic esophageal tumor length was determined by a team of trained pathologists on examination of the resected esophageal specimen as described previously.6, 7 If adenocarcinoma was noted in 1 or more foci but no clear mass was identified, then a pathologic tumor length of 0.1 cm was assumed.
The Pearson chi-square test was used to assess correlation between endoscopically measured tumor length and pathologically measured tumor length. A regression tree survival analysis was used to determine the optimum cutoff point for endoscopic tumor length, which was 2 cm. Then, patients were divided into 2 groups—patients with tumors ≤2 cm in length and patients with tumors >2 cm in length—for survival analyses. Further analysis was performed using a univariate Cox proportional hazards regression model to examine the association between each potential predictor and survival.9 Hazard ratios (HRs) with 95% confidence intervals (CIs) were used to quantify the strength of the association between predictors and survival. Preoperative predictors with P values ≤.25 in univariate analysis were entered into a multivariate Cox proportional hazards regression model.9 By using Wald stepwise selection with P = .10 as entry and removal probability, the final model for the dataset was obtained. Overall survival was estimated by the Kaplan-Meier method using the date of esophageal resection as the starting point and the date of death or last follow-up as the endpoint.10 Kaplan-Meier survival curves were compared using the log-rank test. All statistical calculations were performed using S-PLUS 8.0 (rpart library; TIBCO Software Inc., Palo Alto, Calif), SPSS 15.0 (SPSS Inc., Chicago, Ill), and GraphPad Prism 5.0 (GraphPad Software, San Diego, Calif). Statistical significance was defined as P < .05.
Training Set Patients
For the 164 patients in the training set, the median follow-up for all patients who remained alive was 28 months. Tables 1 and 2 summarize the patient-related and tumor-related characteristics of patients in the training set. The median age was 65 years (range, 26-84 years), and the majority of patients were men (85%). The endoscopic tumor length ranged from 0 cm to 13 cm, and the median endoscopic tumor length for the population was 1.5 cm.
Table 1. Characteristics of Patients in the Training and Validation Sets
No. of Patients (%)
ASA indicates American Society of Anesthesiologists; NA, not available.
Median age [range], y
1: No functional limitation
2: Definite functional limitation
3: Constant threat to life
Type of esophageal resection
Minimally invasive esophagectomy
Transabdominal vagal-sparing esophagectomy
Table 2. Characteristics of Tumors in the Training and Validation Sets
No. of Patients (%)
Clinical staging data were available for only 40 patients in the validation set.
A regression tree model was constructed based on the ability of endoscopic tumor length to predict survival in the training set (Fig. 1). A cutoff point of 2 cm was chosen for further survival analyses based on the initial split in the regression tree model (1.95 cm) to predict survival hazard rates. There was a direct relation between endoscopic tumor length and pathologic tumor length in the training set (Pearson correlation [r] = 0.683; P < .001) (Fig. 2).
Kaplan-Meier curves indicated that patients who had an endoscopically measured esophageal tumor length >2 cm were a high-risk group (Fig. 3, top). The overall 3-year and 5-year survival rates were 36% and 29%, respectively, for patients with endoscopic tumor length >2 cm compared with 85% and 78%, respectively, for patients with endoscopic tumor length ≤2 cm (P < .001). Endoscopically determined esophageal tumor depth and lymph node involvement also were associated with long-term survival in the training set (data not shown). Both continuous and categorical analyses of endoscopic tumor length revealed that this variable was associated with long-term survival in univariate analyses of the training set (continuous: HR, 1.4; 95% CI, 1.3-1.5; P < .01; categorical: HR, 5.81; 95% CI, 3.3-10.3; P < .01) (Table 3). Multivariate analysis also suggested that endoscopic tumor length was a significant independent risk factor for long-term survival (HR, 2.3; 95% CI, 1.1-4.4; P = .019) even after controlling for endoscopically determined tumor depth and lymph node involvement.
Table 3. Univariate and Multivariate Cox Regression Models for the Training Set
No. of Patients
95% CI for HR
95% CI for HR
CI indicates confidence interval; HR, hazard ratio; Ref, reference category; GEJ, gastroesophageal junction; Tis, tumor in situ.
Cervical/upper/middle esophagus (Ref)
Clinical tumor classification
Clinical lymph node classification
Clinical tumor length (continuous)
Clinical tumor length, cm
Survival Analysis of the Validation Set
The validation dataset, with 109 patients, was used to validate the ability of endoscopic tumor length to predict overall survival. The median follow-up of all patients who remained alive in the validation dataset was 21 months. Patient-related and tumor-related factors in the validation dataset were similar to those in the training dataset (data not shown). Analysis with the validation dataset confirmed that endoscopic esophageal tumor length was correlated directly with pathologic tumor length (r = 0.657; P < .001). Survival analysis of the validation dataset also confirmed the significance of endoscopic tumor length with a cutoff point of 2 cm (univariate analysis: HR, 2.8; 95% CI, 1.4-5.8; P = .004). The optimal cutoff point for endoscopic tumor length, as determined by regression tree survival analysis for the validation dataset, was 1.25 cm, in contrast to the optimal cutoff point for endoscopic tumor length in the training dataset, which was 1.95 cm. Finally, Kaplan-Meier curves demonstrated a significant association with survival for endoscopic tumor length >2 cm in the validation set (Fig. 3, bottom).
To our knowledge, our study is the first study to evaluate the value of preoperatively measured esophageal tumor length in predicting patient prognosis and, thus, predicting which patients may benefit from neoadjuvant therapy. In this study, we focused on patients with esophageal adenocarcinoma. We observed that endoscopic esophageal tumor length was correlated directly with pathologic esophageal tumor length and was closely associated with long-term survival even after controlling for other standard prognostic factors, such as endoscopically determined tumor depth and lymph node involvement. Our results suggest that endoscopically determined tumor length may help to identify a high-risk group of patients with esophageal cancer before surgical resection who may be served better with more aggressive neoadjuvant strategies. Our study also suggests that endoscopic esophageal tumor length may be a valuable prognostic factor to incorporate into future esophageal cancer staging systems, because it can be measured reliably before any cancer-directed treatment and allows the grouping of patients into different prognostic groups.
Until 1987, when the current AJCC esophageal cancer staging system was adopted, esophageal tumor length was a factor in determining the T classification.11, 12 However, in 1987, tumor length was removed from the staging system. Recent publications by our group and others have suggested that pathologic esophageal tumor length is correlated directly with long-term survival.6, 7, 13, 14 We have observed that the 5-year survival rate for patients who have pathologic tumor length >3 cm is approximately 10% compared with 68% for patients who have pathologic tumor length ≤3 cm (P < .001).6, 7 We also have demonstrated that pathologic tumor length is a significant predictor of long-term survival (HR, 2.13; 95% CI, 1.26-3.63; P < .01) even after controlling for pathologically determined tumor depth, lymph node involvement, and the presence of metastases,7 factors that are known to portend a poor prognosis.15-18 All of these studies suggest that the incorporation of pathologic tumor length into pathologic TNM staging would significantly improve the ability to predict long-term survival and would help identify a group of patients with esophageal cancer at high risk for recurrence and decreased survival.6, 7, 13, 14
The ability to identify such high-risk patients with esophageal cancer before surgical resection also may be of value because it might allow the better selection of patients for neoadjuvant therapy. Over the last few years, upper gastrointestinal endoscopies have been used to accurately locate esophageal tumors and to accurately measure tumor lengths, depths, and lymph node involvement.19-22 Two small-scale studies have suggested that endoscopically measured tumor length may be able to predict survival before surgical resection. Bhutani et al studied 35 patients to determine the relation between esophageal tumor length as measured on endoscopy and pathologic T classificaton.23 Those investigators observed that tumors ≥5 cm in length had an 89% chance of being T3 or greater, whereas tumors <5 cm in length had a 92% chance of being T1 or T2.23 Shinkai et al assessed tumor volume (as calculated from tumor length and cross-sectional area on endoscopy) and observed a significant correlation between endoscopic tumor length measurements and 5-year survival in patients with squamous cell carcinoma of the esophagus.24 Both of those studies suggested that increasing endoscopically measured tumor length might help identify high-risk esophageal cancer patients before surgical resection. Similarly, our results demonstrated that longer esophageal tumors tend to be greater in depth (data not shown). Indeed, our results suggest that, currently, tumor length measured on endoscopy can be used reliably at centers where endoscopic ultrasound is not readily available.
We recognize that our study has several limitations. First and foremost, this was a retrospective study with all the limitations that such a study design entails. In addition, because the study used data from 2 different institutions with different endoscopists, pathologists, and teams of surgeons, our study may have suffered from lack of uniformity in measurement methods. Thus, carefully designed prospective studies are needed to investigate the role of endoscopically measured tumor length in guiding neoadjuvant therapy for patients with locally controlled tumors. Furthermore, we used an independent dataset from a second institution to validate the prognostic utility of endoscopic esophageal tumor length. The validation dataset confirmed that endoscopic tumor length was correlated directly with pathologic tumor length and that endoscopic tumor length may be a valuable preoperative predictor of long-term survival. However, whereas, in the training dataset, the optimal cutoff point was 2 cm, in the validation dataset, the optimal cutoff point was 1.25 cm according to regression tree analyses. These disparate results may be because of the small number of patients in both datasets, different populations of patients (ie, patients with higher stage disease), and different methods of measurement that were operator-dependent (different team of endoscopists, surgeons, and pathologists). This finding suggests the need for larger prospective datasets to generate an optimal cutoff point for endoscopic tumor length that is applicable for all populations.
In summary, our study suggests that endoscopic esophageal tumor length may be used as a prognostic factor to help identify high-risk patients with esophageal cancer before surgical resection. This prognostic factor appears to be important even when controlling for endoscopically determined esophageal tumor depth and lymph node status. The optimal cutoff point for endoscopic tumor length, however, still may need to be refined, and this will require further assessment with larger prospective datasets.
CONFLICT OF INTEREST DISCLOSURES
Supported by NIH-T32 grant CA009599 (P.G.). This work also was supported by the Homer Flower Gene Therapy Research Fund, the Charles Rogers Gene Therapy Fund, the Charles B. Swank Memorial Fund for Esophageal Cancer Research, the George O. Sweeney Fund for Esophageal Cancer Research, the Phalan Thoracic Gene Therapy Fund, and the M. W. Elkins Endowed Fund for Thoracic Surgical Oncology.