Establishment and validation of a prognostic nomogram for postoperative patients with gastric cardia adenocarcinoma: A study based on the Surveillance, Epidemiology, and End Results database and a Chinese cohort
胃贲门腺癌术后患者预后列线图的建立和验证：一项基于监测、流行病学和最终结果数据库和中国队列的研究

Abstract 抽象

Short abstract 简短摘要

1. INTRODUCTION 1. 引言

Gastric cardia adenocarcinoma (GCA) is a commonly diagnosed malignant tumor of the digestive tract. ¹ Although the overall incidence of gastric cancer (GC) has recently declined globally, the incidence of GCA is still increasing. ^{1 , 2} GCA originates at the independent zone of the esophagogastric junction (EGJ), which differs from GC that originates at other sites in terms of pathophysiology. ³ GCA remains a challenging disease, and radical surgery is currently the only available management method. Early diagnosis and timely surgery can have a positive impact on the prognosis of patients with GCA. In addition, despite advances in treatment, the prognosis for patients with GCA remains suboptimal due to factors such as tumor recurrence. ^{4 , 5} However, the introduction of personalized treatment has directed renewed attention to the prognostic factors that impact patients with cancer. Therefore, it is significant to analyze the prognostic risk factors and construct a survival prediction model for patients with GCA after radical surgery.
胃贲门腺癌（GCA）是一种常见诊断的消化道恶性肿瘤。 ¹ 尽管最近全球胃癌 （GC） 的总体发病率有所下降，但 GCA 的发病率仍在上升。 ^{1 , 2} GCA 起源于食管胃交界处 （EGJ） 的独立区，在病理生理学方面与起源于其他部位的 GC 不同。 ³ GCA仍然是一种具有挑战性的疾病，根治性手术是目前唯一可用的治疗方法。早期诊断和及时手术可以对GCA患者的预后产生积极影响。此外，尽管治疗取得了进展，但由于肿瘤复发等因素，GCA 患者的预后仍然不理想。 ^{4 , 5} 然而，个性化治疗的引入重新引起了人们对影响癌症患者的预后因素的关注。因此，分析根治性手术后GCA患者的预后危险因素并构建生存预测模型具有重要意义。

The TNM staging system is currently the primary method used to evaluate patient prognosis and guide clinical treatment approaches. ⁶ However, relying solely on the TNM staging system has obvious limitations, as it lacks important variables such as age, differentiation grade, and other influential factors. Therefore, it is not suitable for individualized analysis. ^{7 , 8} Multiple studies have demonstrated that the log odds of positive lymph nodes (LODDS) are superior to the N stage alone in determining lymph node metastasis and providing accurate prognostic assessments for patients with cancer. ^{9 , 10} In line with this, a nomogram is a graphical calculation instrument based on statistical models that can predict the likely outcomes. ¹¹ Each variable is assigned a score based on its degree of risk, and the final sum of all scores corresponds to the predicted survival probability. Prior research has demonstrated that nomograms have a better prediction performance compared to that of TNM staging, leading to the construction of various nomograms to predict prognosis in different types of cancer. ^{12 , 13 , 14 , 15}
TNM分期系统是目前用于评估患者预后和指导临床治疗方法的主要方法。 ⁶ 然而，仅依靠TNM分期系统具有明显的局限性，因为它缺乏年龄、分化等级和其他影响因素等重要变量。因此，它不适合进行个体化分析。 ^{7 , 8} 多项研究表明，在确定淋巴结转移和为癌症患者提供准确的预后评估方面，阳性淋巴结 （LODDS） 的对数几率优于单独的 N 期。 ^{9 , 10} 与此一致，列线图是一种基于统计模型的图形计算工具，可以预测可能的结果。 ¹¹ 每个变量都根据其风险程度分配一个分数，所有分数的最终总和对应于预测的生存概率。先前的研究表明，与TNM分期相比，列线图具有更好的预测性能，导致构建各种列线图来预测不同类型癌症的预后。 ^{12 , 13 , 14 , 15}

The Surveillance, Epidemiology, and End Results (SEER) database covers over 28% of the population of the United States and includes information on patient demographics, stage of diagnosis, treatment process, tumor morphology, primary tumor site, vital status follow‐up, and causes of death, thus providing an effective tool for tumor epidemiological research. ¹⁶ The objective of this study was to obtain clinical data from the SEER database and use it to analyze the prognostic risk factors in postoperative patients with GCA. The study also aimed to construct a nomogram model that could accurately predict cancer‐specific survival (CSS) in patients with GCA and evaluate the nomogram through internal and external validation.

2. METHODS

2.1. Patient selection

The specific operational process for extracting information from the SEER database is outlined below: (1) Register for a personal account on the SEER database official website and install the SEER*Stat software; (2) Extract clinical data of patients based on the inclusion and exclusion criteria determined in this study; (3) Export the extracted data as a spreadsheet and proceed to the next step of organizing and analyzing the data.

The data for patients with GCA between 2010 and 2015 were downloaded from the SEER*Stat 8.4.0 software using a private ID (13914‐Nov2021), based on the 2021 release of the SEER database. According to the SEER codes, patients with adenocarcinoma (ICD‐O‐3 codes: 8140–8145,8147, 8210, 8211, 8214, 8220, 8221, 8230, 8231, 8255, 8260–8263, 8310, 8480, 8481, 8490, 8510, 8560, 8562, 8570–8576) and tumors located in the cardia (site code: C16.0) were included. All patients underwent radical surgery (surgery encode 30–80); however, those with stage IV GCA were excluded due to the controversial nature of their operation. In addition, patients with missing information on relevant variables were excluded from this study. The detailed screening process and selection criteria are shown in Figure 1. The patients included in this study were randomly divided into training (70%) and internal validation (30%) groups.

Open in a separate window

FIGURE 1

Flowchart for the selection of the patients.

In addition to the above data, the data of patients with GCA after surgery at the First Affiliated Hospital of Zhengzhou University between 2012 and 2018 were collected for validation. Patients underwent thorough preoperative assessments, including physical exams, medical history recording, hematological testing, and imaging. Furthermore, tumor metastasis was evaluated using various imaging techniques. The study was approved by the hospital ethics committee, and patient records were kept confidential in accordance with ethical standards.

3. RESULTS

3.2. Independent prognostic factors

The univariate Cox regression analysis showed that race, age, marital status, grade, T stage, N stage, tumor size, and LODDS score were related to CSS. To avoid overfitting, LASSO regression was performed using those eight variables. All eight variables were included in the model, as their coefficients were non‐zero (Figure 2). The variables that showed significance in the univariate and LASSO regression analyses were included in the multivariate analysis. In the multifactorial Cox analysis, age (65–71 and >71 years), grade (poor), T stage (T1–3), and LODDS (LODDS2 and LODDS2) were independent prognostic risk factors in the patients with GCA at stages I–III. Moreover, race (other) and marital status (married) were protective factors that showed better patient prognosis. Table 2 displays the outcomes of both univariate and multivariate Cox analyses conducted on the training set.

Open in a separate window

FIGURE 2

LASSO coefficients of eight features (A) and the selection of tuning parameter (λ) for the LASSO model (B).

TABLE 2

Univariate and multivariate cox regression analyses of the prognostic factors for CSS.

Variables	Univariate analysis		Multivariate analysis
	HR (95% CI)	p‐value	HR (95% CI)	p‐value
Sex
Female	1
Male	0.989 (0.796–1.229)	0.922
Race
White	1		1
Black	1.121 (0.765–1.641)	0.558	0.910 (0.616–1.346)	0.638
Other	0.621 (0.440–0.875)	0.007	0.621 (0.439–0.877)	0.007
Age, years
<65	1		1
65–71	1.240 (1.006–1.527)	0.043	1.348 (1.091–1.666)	0.006
>71	1.445 (1.168–1.788)	<0.001	1.472 (1.184–1.830)	<0.001
Marital status
Unmarried	1		1
Married	0.787 (0.654–0.946)	0.011	0.742 (0.614–0.897)	0.002
Histological type
Adenocarcinoma	1		1
Signet ring carcinoma	1.096 (0.832–1.443)	0.514
Tumor grade
Well	1		1
Poor	1.809 (1.513–2.163)	<0.001	1.405 (1.168–1.689)	<0.001
T stage
T1	1		1
T2	2.044 (1.402–2.980)	<0.001	1.613 (1.091–2.383)	0.016
T3	3.569 (2.553–4.800)	<0.001	2.304 (1.649–3.218)	<0.001
T4	6.029 (4.025–9.032)	<0.001	2.868 (1.826–4.505)	<0.001
N stage
N0	1		1
N1	2.183 (1.735–2.746)	<0.001	1.004 (0.758–1.329)	0.979
N2	3.742 (2.907–4.816)	<0.001	1.152 (0.820–1.620)	0.415
N3	4.885 (3.703–6.443)	<0.001	0.983 (0.661–1.461)	0.931
Size, cm
<2.6	1		1
2.6–4.8	1.736 (1.374–2.194)	<0.001	1.101 (0.855–1.420)	0.456
>4.8	2.445 (1.943–3.076)	<0.001	1.282 (0.993–1.656)	0.056
LODDS
LODDS1	1		1
LODDS2	2.420 (1.935–3.028)	<0.001	2.266 (1.740–2.950)	<0.001
LODDS3	5.222 (4.198–6.496)	<0.001	3.869 (2.843–5.266)	<0.001

Open in a separate window

The bold font represents p < 0.05, indicating a statistically significant variable.

3.3. Establishment of the nomogram

Based on the results of the multivariate Cox regression analysis, a nomogram was established and is presented in Figure 3. The C‐index values for the training, internal validation, and external validation sets were 0.737 (95% confidence interval [CI], 0.717–0.757), 0.733 (95% CI, 0.702–0.764), and 0.746 (95% CI, 0.689–0.803), respectively. The calibration curves in Figure 4 show that the predicted results are consistent with the observed results. As shown in Figure 5, the tdROCs showed that the AUC values of the nomogram in the 3‐ and 5‐year groups were 0.802 (95% CI, 0.774–0.830) and 0.818 (95% CI, 0.789–0.847) in the training set, 0.796 (95% CI, 0.753–0.839) and 0.804 (95% CI, 0.755–0.853) in the internal validation set, and 0.805 (95% CI, 0.732–0.877) and 0.836 (95% CI, 0.746–0.925) in the external validation set, respectively. In comparison, the AUC values of TNM stage in the 3‐ and 5‐year groups were 0.692 (95% CI, 0.663–0.722) and 0.717 (95% CI, 0.683–0.751) in the training set, 0.715 (95% CI, 0.671–0.758) and 0.744 (95% CI, 0.693–0.796) in the internal validation set, and 0.712 (95% CI, 0.646–0.778) and 0.758 (95% CI, 0.655–0.862) in the external validation set, respectively. The nomogram had superior predictive ability compared to that of the TNM stage alone (p < 0.05). Furthermore, the DCA curves in Figure 6 show that the nomogram for cancer‐CSS provides significant clinical benefits. In summary, the predictive model based on the aforementioned factors had a strong predictive power for CSS with high accuracy and clinical utility in patients who underwent radical surgery for GCA. This was supported by the results of various statistical analyses such as the –index, calibration curve, ROC, and DCA.

Open in a separate window

FIGURE 3

Nomogram predicting CSS of the postoperative patients with GCA.

Open in a separate window

FIGURE 4

Calibration curves of the nomogram for predicting 3‐year CSS (A) and 5‐year CSS (B) in the training set; the calibration curves of the nomogram for predicting 3‐year CSS (C) and 5‐year CSS (D) in the internal validation set; and the calibration curves of the nomogram for predicting 3‐year CSS (E) and 5‐year CSS (F) in the external validation set.

Open in a separate window

FIGURE 5

Time‐dependent ROC curves were used to test the predictive power of the 3‐year CSS and 5‐year CSS in the training se (A and D); the internal validation set (B and E); and the external validation set (C and F), respectively.

Open in a separate window

FIGURE 6

DCA of the nomogram and TNM stage 3‐year CSS and 5‐year CSS of the training (A and D), internal (B and E), and external cohorts (C and F), respectively.

3.4. Risk stratification based on the nomogram

The risk score was calculated using the nomogram, and patients were categorized into low‐risk and high‐risk groups based on the median value as the cutoff point. The Kaplan–Meier plot (Figure 7) showed that patients in the low‐risk group had a significantly better prognosis than those in the high‐risk group (p < 0.001). Further analysis (Figure 8) revealed that chemotherapy was beneficial only in the high‐risk group, as identified by our model. This suggests that our model can aid physicians in identifying high‐risk patients who may benefit from chemotherapy, allowing for personalized treatment plans.

Open in a separate window

FIGURE 7

Kaplan–Meier curves for the patients in the low‐ and high‐risk groups based on the risk scores. (A) training cohort; (B) internal validation set; and (C) external validation set.

Open in a separate window

FIGURE 8

Effect of chemotherapy on the survival in the total population (A), low‐risk group (B), and high‐risk group (C).

4. DISCUSSION

Currently, mainstream studies consider the EGJ to be a region that is separate from both the esophagus and the stomach. ³ Given that GCA is a type of malignant tumor occurring in the EGJ that has caused the death of millions of individuals, this disorder warrants further studies. Surgical resection is undoubtedly the most important treatment approach for GCA, with a 5‐year survival rate of 43–49%. ¹⁷ However, despite advancements in treatment, the survival rates for postoperative patients with GCA remain below desired levels, and many experience recurrence and death each year. While the TNM staging system is widely used as a prognostic assessment tool to guide postoperative treatment protocols, ¹⁸ it has limitations in terms of specificity and does not consider important patient factors such as age, sex, and marital status. As a result, the predictive ability of TNM staging may not be sufficiently accurate. In contrast, nomograms are one of the most widely used forecasting tools that can comprehensively consider multiple factors, including clinical pathology and demographic characteristics. This is why our research study focused on developing a nomogram for predicting CSS in postoperative patients with GCA. Previous studies have reported different prognostic models for adenocarcinomas of the EGJ. ^{5 , 19 , 20} However, most of the previously built models lacked external validation, which reduced their applicability. Therefore, we aimed to construct a predictive model for the prognosis of postoperative patients with GCA at stages I–III GCA and enhance the reliability of the model through validation in an external population.

Our study aimed to construct a nomogram that accurately predicts the prognosis of postoperative patients with GCA at stages I–III based on a multivariate analysis. The C‐index and AUC values were both greater than 0.71, indicating favorable discrimination of the nomogram. Compared to the TNM staging system alone, the nomogram model had a higher accuracy, as shown by the AUC value. The DCA curve was used to analyze the clinical benefits of the model, and the results suggested that the nomogram had a high net clinical benefit. Further statistical analyses showed that the nomogram model had more advantages than the TNM staging system. The use of our nomogram model aligns with the current principles of personalized treatment. Finally, the applicability of the model was verified using an external population.

Although chemotherapy is widely used in the treatment of GCA, our study found that it did not have a positive therapeutic effect on patients in this study. ²¹ However, the CLASSIC trial established the benefits of adjuvant capecitabine and oxaliplatin in patients with GC/AEG who underwent surgery. ²² Based on the nomogram score in this study, we stratified the study population into high‐risk and low‐risk groups. We observed that chemotherapy had a positive therapeutic effect only in the high‐risk group. This indicates that the nomogram‐based identification of high‐risk groups is accurate and can guide clinicians in formulating personalized chemotherapeutic regimens. As for the low‐risk population, the results of our study suggest that chemotherapy may not provide significant therapeutic benefits. Therefore, overtreatment should be avoided to prevent unnecessary adverse effects and improve the patient's quality of life.

In the multifactor analysis, race, age, marital status, histological grade, T stage, and LODDS were determined to be independent prognostic factors. It has been widely reported that older age is a poor prognostic factor in cancer patients. ^{23 , 24} For instance, elderly patients often have functional impairments, malnutrition, and comorbidities that prompt physicians to choose less aggressive treatments or shorten the course of treatment, which in turn affects the treatment outcomes. ²⁵ In contrast, it has been suggested that younger patients may have a better tolerance to the adverse effects of treatment, including myelosuppression after chemotherapy, compared with that in elderly patients. Additionally, in this study, unmarried patients were more prone to anxiety and had greater stress burdens than married patients, thereby reducing their immune capacity and affecting their metabolic balance, resulting in decreased survival. ^{26 , 27} In addition, unmarried patients had poorer compliance with treatment and received the treatment later, which may also be a reason for their poor prognoses. ²⁸

Zhu et al. observed that White patients had the highest risk of GCA compared with that in the other ethnic groups, which is consistent with our findings. ⁷ In prior studies, race‐related differences in morbidity and outcomes have been attributed to obesity and unequal incomes. ^{29 , 30} In addition, lymph node metastasis has been found to have a significant impact on postoperative outcomes, including long‐term survival and postoperative adjuvant therapy administration. ³¹ The results of our study showed that LODDS was a more significant prognostic factor than the conventional N stage in assessing postoperative risk factors for GCA. This is consistent with results of previous studies, suggesting that more attention should be paid to the LODDS of postoperative patients rather than solely focusing on the number of positive lymph nodes. ^{32 , 33 , 34 , 35} Additionally, tumors with poor differentiation are generally more aggressive and have a higher likelihood of recurrence and distant metastasis, requiring close monitoring of this patient group. ³⁶

Compared with the study by Guo et al., ⁵ our study included more variables. Additionally, we conducted an external validation, which enhanced the reliability of our model. To the best of our knowledge, this is the first nomogram to predict the survival of postoperative patients with GCA with both internal and external population validation. However, our study has some limitations. First, important clinical information was lacking from the SEER database. Smoking, alcohol consumption, body mass index, diet, performance, and family history are of great significance in the prognostic evaluation of malignant tumors. In addition, this study did not include molecular or genetic information that is used in routine clinical treatments, such as EGFR mutations and Her‐2 expression. Additionally, the SEER database lacks information on specific regimens of chemoradiotherapy, targeted therapy, and immunotherapy, which could have affected the results of our study. Another limitation of our study is that it was retrospective in nature, and thus, a prospective study is necessary to further validate our findings. This will be the focus of our subsequent studies.

5. CONCLUSIONS

In conclusion, we utilized clinical data from the SEER database to identify factors associated with survival in postoperative patients with GCA at stages I–III. Subsequently, we developed a nomogram that accurately predicted CSS in patients with GCA who underwent radical surgery. Our findings indicate that the nomogram outperforms TNM staging in terms of predictive power and may provide greater clinical benefits for patients with GCA after radical surgery.

AUTHOR CONTRIBUTIONS

Lei Wang: Conceptualization (lead); methodology (lead); writing – original draft (lead). Jingjing Ge: Conceptualization (equal); methodology (equal); writing – original draft (equal). liwen feng: Data curation (supporting); visualization (supporting). Zehua Wang: Data curation (supporting); visualization (supporting). Wenjia Wang: Data curation (supporting); visualization (supporting). Huiqiong Han: Conceptualization (equal); writing – review and editing (equal). Yanru Qin: Conceptualization (lead); writing – review and editing (lead).

CONFLICT OF INTEREST STATEMENT

The authors have no conflict of interest to declare.

ETHICS APPROVAL STATEMENT

The First Affiliated Hospital of Zhengzhou University's Medical Ethics Committee examined and approved this study involving human participants (2023‐KY‐0019‐001).

PATIENT CONSENT STATEMENT

As this was a retrospective study, no written informed consent was required.

ACKNOWLEDGMENTS

Notes

Wang L, Ge J, Feng L, et al. Establishment and validation of a prognostic nomogram for postoperative patients with gastric cardia adenocarcinoma: A study based on the Surveillance, Epidemiology, and End Results database and a Chinese cohort. Cancer Med. 2023;12:13111‐13122. doi: 10.1002/cam4.5994 [PMC free article] [PubMed] [CrossRef] [Google Scholar]

Lei Wang and Jingjing Ge contributed equally as first authors.

Contributor Information

Huiqiong Han, Email: moc.361@32388qhh.

Yanru Qin, Email: nc.ude.uzz@niqurnay.

REFERENCES