The prognostic impact of tumor length on pathological stage IA-IC esophageal adenocarcinoma

Abstract

This study was completed to evaluate the relationship between tumor length and the prognosis of patients with pathological stage IA-IC esophageal adenocarcinoma (EAC). Patients were identified from the Surveillance, Epidemiology, and End Results Program database (United States, 2006–2015). X-tile software and ROC analysis were mainly used to explore the best threshold of tumor length for dividing patients into different groups, and then propensity score matching (PSM) was used to balance other variables between groups. The primary outcome assessed was overall survival (OS). A total of 762 patients were identified, and 500 patients were left after PSM. Twenty millimeters were used as the threshold of tumor length. Patients with longer tumor lengths showed worse OS (median: 93 vs. 128 months; P = 0.006). Multivariable Cox regression analysis showed that longer tumor length was an independent risk factor (hazard ratio 1.512, 95% confidence interval, 1.158–1.974, P = 0.002). Tumor length has an impact on patients with pathological stage IA-IC EAC who undergo surgery alone. The prognostic value of the pathological stage group may be improved after combining it with tumor length and age.

INTRODUCTION

Esophageal adenocarcinoma (EAC) is a highly lethal malignancy with an average 5-year survival of only 15%.¹ It is now the predominant subtype in many Western countries, and its incidence is rising.² The tumor node metastasis (TNM) staging system is applied to predict prognosis in the clinical setting. Within the same tumor stage, the clinical outcomes of patients can vary substantially,³ indicating the importance of identifying other factors that might improve prognosis prediction.

Between 1983 and 1987, the American Joint Committee on Cancer (AJCC) staging system included tumor length (T1, <5 cm; T2, >5 cm; T3, evidence of extraesophageal spread) and peripheral involvement to determine T stage.⁴ However, tumor length has been removed as a staging parameter. Studies have reported that increasing tumor size is related to the prognosis of solid tumors such as lung cancer,⁵ kidney cancer,⁶ breast cancer,⁷ and gastric cancer.⁸ However, the relationship between tumor length and the prognosis of EAC has been debated in recent years.

In an overall preliminary analysis of the Surveillance, Epidemiology, and End Results Program (SEER) database, we found that the prognosis of different tumor lengths in some stages of EAC was different. This study therefore sought to assess the relationship between tumor length and the prognosis of patients undergoing surgery alone for patients with pathological stage IA-IC EAC.

METHODS

Ethics statement

This study was approved by the Cancer Hospital of Zhengzhou University and Henan Cancer Hospital Ethics Committee. SEER database is a publicly available database containing deidentified data, and the need for informed consent was waived.

Study population

We extracted data from the SEER database. Patients whose first primary cancer was EC (International Classification of Diseases for Oncology tumor site codes 150–155 and 158–159) were identified between 2006 and 2015. Patients other than those with adenocarcinoma (SEER codes 8140–8389) were excluded. This was a retrospective analysis based on the final pathologically determined tumor length of the resected specimen, so we stratified the cohort using pathological stage rather than clinical stage and only included patients with pathological stage IA (pT1aN0G1, pT1aN0Gx), IB (pT1aN0G2, pT1bN0G1–2, pT1bN0Gx), and IC (pT1N0G3, pT2N0G1–2). Only patients who had malignant tumor behavior and active follow-up, were treated with surgery alone and survived >1 month after the surgery. We only included patients who underwent surgery alone involving reconstruction of the digestive tract (comprising partial esophagectomy, total esophagectomy, and esophagectomy with laryngectomy and/or gastrectomy). Patient demographics and tumor-related information were extracted, including sex, age at diagnosis, race, pathological T (pT) stage, pathological stage group, tumor grade, and tumor length (the sixth and seventh editions of the AJCC-TNM staging information were restaged by the eighth edition for EAC).

Study steps

1. Finding that tumor length is a prognostic factor.

In a pre-study, we identified tumor length (as a continuous variable) as an independent prognostic factor for pathological stage IA-IC EAC by multivariable analysis.

2. Exploring the threshold for tumor length.

The median of tumor length, the result of X-tile software⁹ (Yale University, New Haven, CT), and ROC analysis were used to explore the best threshold of tumor length (the principle of X-tile software was the minimum P-value method, but the results were cross-validated, which reduced the error of false positives to a certain extent).

3. Dividing the cohort of patients according to this threshold.

4. Performing PSM of these two cohorts.

Propensity score matching¹⁰ (PSM) analysis was used to reduce selection bias from an imbalance in the variables that may potentially influence the outcomes. The variables entered into the PSM were age, sex, race, tumor length, tumor grade, pT stage, and pathological stage group, and a 1:1 matching method with a caliper width of 0.001 times the standard deviation (SD) was performed.

5. Assessing the survival of the two groups.

Survival estimates were determined by the Kaplan–Meier method, and the log-rank test was used to evaluate survival differences. The Cox regression model was used to perform univariate and multivariable analysis to evaluate whether tumor length was an independent prognostic factor. Proportional hazards assumptions for each covariate were assessed graphically or using the test for proportional hazards based on Schoenfeld residuals to make judgements. Variables were eliminated in a stepwise manner until all remaining values were < 0.05.

6. Model comparisons.

The accuracy of the model was preliminarily evaluated by Harrell’s concordance index (C-index¹¹), wherein a C-index of 1 indicated perfect prediction accuracy, while a C-index of 0.5 indicated that a model was not better than random chance.

Statistical analysis

Continuous variables conforming to a normal distribution are expressed as the mean ± SD, and those with a skewed distribution are expressed as the median ± interquartile range. Categorical variables are expressed as the number of cases and percentages. Categorical data between groups were compared using the chi-square test or Fisher’s exact test. An independent samples t-test was used to analyze the differences between the groups. All-cause mortality was the primary endpoint, and all other events were considered censored. All statistical analysis were performed with R4.1.3 software (Institute for Statistics and Mathematics, Vienna, Austria) and IBM SPSS 26 (IBM Corp., Armonk, NY), with P < 0.05 deemed to indicate statistical significance.

RESULTS

Patient Clinicopathological characteristics

We identified 762 eligible patients with pathological stage IA-IC EAC who underwent surgery alone and 500 patients remained after PSM (the process is shown in Fig. 1). The demographic and tumor characteristics are presented in Table 1. Of the 762 EAC patients, 53.5% were aged 65 years and older, 88.5% were male, and 96.2% were white.

Finding that tumor length is a prognostic factor

The findings of our initial analysis, in which the multivariable Cox regression model showed that tumor length as a continuous variable was a risk factor (hazard ratio [HR] of 1.008, 95% confidence interval [CI], 1.002–1.013, P = 0.011) (Supplemental Material Fig. 1, Table 1).

Exploring the threshold for tumor length

The median tumor lengths were 15 mm (8–25 mm) before PSM and 19.5 mm (10–28 mm) after PSM. Before PSM, the best threshold for tumor length was between 20 mm, according to the X-tile software, and 19.5 mm, obtained by the ROC analysis (Fig. 2). After PSM, they were 20 and 22.5 mm (Fig. 2), respectively. From the results, we found that the results were mostly distributed approximately 2 cm, plus the integer as the threshold could facilitate the memory and application of clinical practitioners, so we took 20 mm as the threshold for subsequent analysis.

Dividing the cohort and performing PSM

Before PSM, the two subgroups divided according to tumor length (<20 mm and tumor length ≥ 20 mm) differed significantly in differentiation, pT stage, and pathological stage. There were no significant differences in sex, age, or race between the groups (Table 1). However, after PSM, there were no significant differences in any variables. The median follow-up times were 73 months (42–104) before PSM and 75 months (42–104) after PSM, the numbers of patients who died during the last follow-up were 292 (38.3%) and 220 (44%), and the overall 5-year survival rates were 73.1 and 69.9%, respectively.

Assessing the survival of the two groups

The Kaplan–Meier survival curves (Fig. 3) showed that patients with a longer tumor length were associated with worse prognosis after PSM (P < 0.001). The median survival times of patients whose tumor length was longer than or equal to 20 mm were 98 months before PSM and 93 months after PSM. In the shorter tumor length subgroup, it was not reached before PSM, and it was 128 months after PSM.

Before PSM, multivariable Cox regression models showed that age over 65 years was a risk factor (HR of 2.097, 95% CI, 1.643–2.676, p < 0.001), as was tumour length longer than or equal to 20 mm (HR of 1.495, 95% CI, 1.182–1.892, p = 0.001). However, after PSM, age over 65 years was a risk factor (HR of 2.012, 95% CI, 1.505–2.691, p < 0.001), as was tumour length longer than or equal to 20 mm (HR of 1.512, 95% CI, 1.158–1.974, p = 0.002) (Table 2).

Model comparisons

Three prediction models were established based on patients after PSM. The first was the basic model of the AJCC pathological stage group. The second was the first model combining tumor length. The third was the model with the highest C-index. The C-index of the model based on the pathological stage group was 55.35%; it was 57.7% after adding tumor length and 63.7% for the third model (including age, tumor length, and pathological stage group).

DISCUSSION

In this study, we found that tumor length longer than or equal to 20 mm was an independent risk factor in patients with pathological stage IA-IC EAC who underwent surgery alone. In addition, age and pathological stage were also independent prognostic factors in this study.

In recent years, most studies^12–15 have suggested that tumor length is an important prognostic predictor for EC, but there has been no consensus on the threshold for tumor length, especially for EAC. The choice of threshold for continuous variables in previous studies had generally been as follows: simply using the median as a threshold ^16,17; using historical literature data¹⁸; and selecting the corresponding threshold using the survival analysis and log-rank test for the best P value.^19,20 The optimal threshold of tumor length was defined as 20 mm in this study, which was obtained by combining the results of some of the above approaches and considering the convenience of clinical application. Studies published in recent years have shown that the thresholds for tumor length were predominantly in the range of 30–40 mm.^20–25 In this study, we identified a lower threshold, possibly because the subjects of this study were patients with early-stage EAC.

Exploiting the large sample size of the SEER database, we adopted a very strict calliper width to balance the main variables of this study between groups. Previous studies suggested^26–28 that age was an independent factor. A similar result was obtained in this study using a patient age of 65 years. The C-index is an indicator that reflects the prediction accuracy of the model. As a result of the lack of original data, the TNM staging information transformed from the old version does not contain more precise staging information for some patients, which is why the values of the C-index appear lower in this study. However, the outcomes still show certain tendencies. The prediction accuracy was better than before after adding tumor length and age one at a time.

For early-stage EAC, endoscopic resection is an alternative treatment option,²⁹ which has the advantages of less surgical trauma, lower risk of perioperative complications, accelerated postoperative recovery, and higher medical economic benefits compared with conventional surgery, but its application is currently limited to very early-stage patients. The inclusion of tumor length as well as the depth of tumor infiltration may permit more accurate assessment of the severity of the primary tumor. Anatomically, the submucosal layer of the oesophagus is rich in lymphatic vessels, and a longer tumor length may be an important factor leading to regional lymph node metastasis. Tumor length may play a more important role both in the stratification of the prognosis of the current endoscopic resection-adapted population and in the exploration of the boundaries of the endoscopic resection audience. On the other hand, there are different options for endoscopic resection commonly used today, such as endoscopic mucosal resection, multiloop ligation mucosal resection, and endoscopic submucosal dissection, with each approach carrying distinct advantages and disadvantages. Considering the tumor length before making a specific choice of resection modality may allow patients to minimize complications while providing appropriate oncological clearance. Therefore, it is necessary to further investigate the relationship between tumor length and endoscopic resection.

This study has several inherent limitations. First, this was a retrospective study, limited by the study design. Second, the SEER database does not contain detailed information about the treatment performed on each patient. There was no surgical margin status information in the SEER database, which may affect the results. Third, multifocal tumors were not considered, which may affect the observational results.³⁰ Finally, the study was based on ‘pathological staging’ and did not demonstrate that the same clinical correlation would still exist if ‘clinical staging’ were used.

CONCLUSIONS

In this report, we observed that tumor length is a key factor influencing the prognosis of patients with pathological stage IA-IC EAC, with a possible threshold of 20 mm. The predictive ability of the AJCC pathological stage group may be improved to a certain extent by combining tumor length and age to identify high-risk patients for individualized therapies.

Financial support

This study was funded by the Health Committee of Henan Province (SB201901108 and YXKC2020022), Department of Science and Technology of Henan Province (222102310170) and Wu Jieping Medical Foundation (320.6799.15061).

Authors’ contributions

Writing—original draft: Sen Yan; Writing—review & editing: Xianben Liu, Wenqun Xing, Andrew Chang, Hai-Bo Sun; Study concept and design: Sen Yan, Xianben Liu, Wenqun Xing, Hai-Bo Sun; Acquisition of data: Sen Yan, Xianben Liu, Duo Jiang, Shao-Kang Feng, Hai-Bo Sun; Methodology: Sen Yan, Duo Jiang, Shao-Kang Feng, Hai-Bo Sun; Funding acquisition: Haibo Sun; Software: Sen Yan, Duo Jiang, Shao-Kang Feng, Hai-Bo Sun. Validation: Sen Yan, Xianben Liu, Wenqun Xing, Hai-Bo Sun, Duo Jiang, Shao-Kang Feng.

Data availability statement

The data for the SEER database can be downloaded from the official website (https://seer.cancer.gov/data-software).

Competing interests: The authors declare that they have no competing interests.

References