Efficacy of endoscopic therapy in patients with T1b gastric cancer and construction of a prognostic prediction model: a retrospective cohort study and multicenter validation study.

Materials and methods

Patient selection
This retrospective study amalgamated data from the Surveillance, Epidemiology, and End Results (SEER) database with clinical records from the Chinese multicenter database. The SEER database comprises 18 regional cancer registries and encapsulates millions of patient records, representing over 30% of the U.S. population.
For the period between 2010 and 2020, data on GC patients were extracted utilizing the SEER*Stat software version 8.4.1.2. The inclusion criteria were specified as follows: (i) Diagnosis under the International Classification of Diseases (ICD) code-O-3 for gastric cancer, encompassing 15 specific codes. (ii) Tumors located in the gastric cardia, greater curvature, lesser curvature, pylorus, fundus, body, or other regions of the stomach. (iii) Histologically confirmed as a primary tumor of the stomach, with no history of previous tumors. (iv) Comprehensive and validated data on survival status. Concurrently, clinical information for patients from the Chinese multicenter database was retrieved based on these inclusion criteria.

Definition of cohorts and variable recording
Patients from the SEER database were randomly stratified into two cohorts at a 7:3 ratio: the training cohort and Validation Cohort 1. Data from patients the Chinese multicentre validation set (diagnosed from three hospitals, namely Shandong Cancer Hospital, Tianjin Cancer Hospital and Renmin Hospital of Wuhan University since 2015) constituted Validation Cohort 2. Demographic variables included race, gender, age, year of diagnosis, and marital status. Tumor-specific variables comprised histological subtype, primary tumor location, tumor dimensions (tumor diameter determined by endoscopy, CT scan, and pathological biopsy), number of tumors, grade of differentiation (according to WHO classification), and therapeutic modality. Major treatment approaches included chemoradiotherapy, endoscopy, and gastrectomy. Codes 10–14 and 20–27 were allocated for endoscopic procedures, delineating whether pathological specimens were collected. Gastrectomy was denoted by codes 30–80. This study did not utilize a pre-registered protocol. The studies involving humans were approved by The Ethics Committee of Renmin Hospital of Wuhan University (Approval No. WDRY2022-K274).

Primary and secondary endpoints
The primary endpoint for this investigation was 6-year overall survival (OS), with a secondary emphasis on 6-year cancer-specific survival (CSS). OS was defined as mortality from any etiology, while CSS was specified as mortality attributable to the malignancy. These parameters were ascertained through the SEER database, relying on cancer registry data and death certificates. For patients diagnosed at the Chinese multicenter database, the cause of death was corroborated via death certificates and longitudinal follow-up (cut-off August 2023).

Statistical analyses
Continuous variables conforming to a normal distribution were expressed as either mean ± standard deviation or median (interquartile range, IQR), and subjected to statistical analysis via t-tests or Mann–Whitney U-tests. Categorical variables, characterized by frequencies and percentages, underwent evaluation using Chi-square tests or Fisher’s exact tests. Kaplan–Meier survival curves, coupled with log-rank tests, assessed both overall survival (OS) and cancer-specific survival (CSS) and identified variations in survival outcomes among different subgroups. Comparative analyses utilized three distinct statistical approaches: unadjusted models, inverse probability treatment weighting (IPTW), and propensity score matching (PSM). First, we used a logistic regression model to calculate each patient’s propensity score based on factors such as age, gender, race, primary site, pathological grade, and tumor size. Then, we performed propensity score matching (PSM) using the nearest neighbor matching method at a 1:1 ratio. IPTW assigns each individual a weight, which is calculated as the inverse of their propensity score for treatment selection. Specifically, we calculate each patient’s propensity score for receiving a particular treatment and weight each individual’s contribution based on the inverse of that propensity score. After matching, we examined the baseline characteristics of each group before and after matching to ensure that the matched samples were balanced on important covariates, thereby reducing the likelihood of confounding bias.6.
Least absolute shrinkage and selection operator (Lasso) regression analyses were conducted on all variables to identify those significantly correlated with survival outcomes.
The Harrell’s Consistency Index (C-index) served to quantify the discrepancy between observed and predicted outcomes, thus appraising the prognostic model’s predictive accuracy. Validation of the model’s discriminative and calibrative performance was performed through the area under the receiver operating characteristic curve (AUC) and calibration plots, respectively. A C-index and AUC value surpassing 0.7 signified robust predictive estimations.
Decision curve analysis (DCA) was implemented to evaluate the clinical utility and benefit-risk profiles of the predictive models. All P-values were two-tailed, and a threshold of P < 0.05 established statistical significance. All statistical analyses were executed using R software, version 4.2.2 (R Foundation for Statistical Computing, Vienna, Austria).

Results

Baseline and clinical characteristics
In total, 2020 patients were extracted from the SEER database for this study. Among them, endoscopy was the treatment modality for 287 patients (14.21%), while 1613 patients (79.85%) underwent gastrectomy. Chemoradiotherapy was administered to 120 patients, constituting 5.94% of the study population. A significant proportion of gastric cancer (GC) patients, 85.84%, were aged 55 years or older. Additionally, males represented 64.11% of the GC patients, with whites comprising 66.78% and married individuals making up 59.85%. Regarding tumor localization, gastric cardia and antrum/body were the primary sites in 34.70% and 36.19% of cases, respectively. Adenocarcinoma was the histological classification in 98.17% of patients. Notably, only 13.02% of tumors exhibited Grade I differentiation. Over 70% of patients did not undergo either radiotherapy or chemotherapy. In 88.76% of patients, a single tumor was observed. Furthermore, the rate of cancer-specific mortality stood at 18.51% (Table 1).
The proportion of gastrectomy has shown a gradual decline, with 89.6% in 2010–2013, 74.1% in 2014–2017, and 73.3% in 2018–2020. Conversely, the proportion of endoscopy significantly increased in 2014–2017 and 2018–2020 compared to 2010–2013, while the usage of chemoradiotherapy remained relatively stable. These trends in treatment modalities for T1b GC patients in recent years are depicted in Fig. 1.

Survival analysis

Differences in survival outcomes prior to adjustment
During the 6-year follow-up period, 259 out of 287 in the endoscopic therapy group died, compared to only 1068 out of 1613 patients in the gastrectomy group. Patients undergoing gastrectomy demonstrated superior overall survival (OS) compared to those receiving endoscopic therapy and radiotherapy. Over the 6-year follow-up period, 220 of the 248 patients in the endoscopic therapy group died from gastric cancer, while 901 of the 1383 patients in the gastrectomy group died with gastric cancer. Cancer-specific survival rates (CSS) were significantly higher in patients with gastrectomy compared to those treated with chemoradiotherapy. This suggests that endoscopy is statistically comparable to gastrectomy in terms of survival outcomes (Fig. 2, Supplementary Fig. 1).

Differences in survival outcomes after PSM
To enhance the reliability of our findings, propensity score matching (PSM) was employed, 1:1 matching was achieve(Supplementary Table 1–4). For OS, there were 242 patients in both the endoscopy and gastrectomy groups. The gastrectomy group showed a better OS rate than the endoscopic therapy group, though the difference did not reach statistical significance (Fig. 3). For both OS and CSS, the survival benefits of endoscopy were significantly superior to those of chemoradiotherapy (p < 0.001, and p < 0.001, respectively) (Supplementary Fig. 2).

Differences in survival outcomes after IPTW
IPTW-adjusted baseline characteristics for the survival analysis are shown in Supplementary Table 5–8. For both OS and CSS, the survival curves of endoscopy and gastrectomy intersected (P = 0.0766, and P = 0.854, respectively) (Fig. 4), which were not statistically significant. However, the long-term survival benefits of endoscopy were markedly superior to those of chemoradiotherapy (p = 0.0002, and p < 0.0001, respectively) (Supplementary Fig. 3).

Differences in survival outcomes in an external dataset from the Chinese multicenter database
To further demonstrate the effect of treatment on survival, we collected clinical data on 116 patients from the Chinese multicenter database. The median age of the cohort was 58 years and the median follow-up was 66 months. Of these, 34 underwent endoscopy, and 82 underwent gastrectomy. Although the 6-year OS rate was lower for endoscopy than for gastrectomy (17.65% vs. 51.22%), the difference was not statistically significant (p = 0.096) (Fig. 5).

Development of predictive models
In the training set, 1412 patients from the SEER database constituted the training cohort, with 608 in validation cohort 1 and 120 from the Chinese multicenter database in validation cohort 2. The baseline characteristics of these cohorts are summarized in Table 2.
Binomial deviation curves were plotted against log(λ), where λ represents the tuning hyperparameter. The vertical solid line indicated the binomial deviation ± standard error (SE), and the vertical dashed line represented the optimal λ value as determined by the minimum criterion and the 1-SE criterion. The optimal λ was selected for the LASSO model through tenfold cross-validation based on the minimum criterion (Supplementary Fig. 4). The final model included the variables: age, sex, race, grade, size, and therapy. The corresponding risk score formula was as follows: Risk score = (0.301376362958108 × Age) + (0.101877469673232 × Sex) + (-0.109775482790134 × Race expression) + (0.00338072596258515 × Grade) + (0.0522943050231388 × Size) + (-0.533754928022369 × Therapy).
Subsequently, we incorporated these selected variables into a predictive model and constructed a dynamic nomogram (Fig. 6) capable of delineating survival zones and predicting survival probabilities. The dynamic nomogram is accessible at the following URL: https://doctoryyds.shinyapps.io/DynNomapp/.

Validation of the prediction model
The area under the curve (AUC) consistently exceeded 0.7 in both the training and validation cohorts (Fig. 7). The concordance index (C-index) for the training cohort was 0.732 (95% CI 0.721–0.745), for validation cohort 1 was 0.737 (95% CI 0.719–0.748), and for validation cohort 2 was 0.843 (95% CI 0.824–0.865). To further assess model performance and external validity, calibration curves were plotted. These indicated minimal deviation, suggesting that the predicted outcomes were largely congruent with actual results across cohorts and attesting to the model’s accuracy (Supplementary Fig. 5). The decision curve analysis (DCA) further substantiated the model’s predictive utility (Fig. 8).

Discussion
Operative resection remains the cornerstone of treatment for patients with gastric cancer (GC) (22). Classical surgical approaches for GC include total and subtotal gastrectomy, the choice of which depends on tumor location. To ensure complete tumor excision and reliable staging, gastrectomy is commonly accompanied by D2 lymph node dissection, currently the primary therapeutic strategy for GC eradication [22]. However, advancements in medical technology have catalyzed the evolution of therapeutic instruments, leading to the maturation of minimally invasive techniques. As a result, abdominal surgery has progressively transitioned from traditional open procedures to laparoscopy. Initially, laparoscopy was restricted to the treatment of early-stage distal GC, obviating the need for total gastrectomy or extended lymph node dissection [23]. With therapeutic outcomes comparable to traditional methods [24], laparoscopic interventions have demonstrated significant reductions in both procedural damage and associated side effects [25–27]. Besides, endoscopic techniques have advanced substantially over recent decades [28]. For differentiated gastric adenocarcinomas without ulcerative features, endoscopic mucosal resection (EMR) and endoscopic submucosal dissection (ESD) are available [29], both of which provide good long-term outcomes. In this study, we can also observe that the proportion of endoscopy has approximately tripled in comparison to 2010–2013, which further suggests that the status of endoscopic treatment is increasing. And we found that for patients with T1b GC, the efficacy of endoscopic treatment was similar to that of surgical resection.
Early GC confined to the mucosal layer constitutes an unequivocal indication of ESD [30]. Moreover, the presence of lymph node metastasis in early GC—particularly in cases of submucosal invasion—is gradually being recognized as an extended indication for ESD [31, 32], a premise corroborated by the study of Li et al. [33]. A retrospective analysis with a longitudinal follow-up exceeding five years evaluated Endoscopic submucosal dissection (ESD) versus surgical resection in key metrics: operative duration, length of hospitalization, postoperative complications, and rates of recurrence. The ESD cohort exhibited substantial reductions in operative time (79.68 min vs. 262 min, P < 0.001), duration of fasting (3.01 days vs. 4.49 days, P < 0.001), and length of hospital stay (7 days vs. 13 days, P < 0.001). This cohort also reported diminished rates of early postoperative complications (12.3% vs. 23.6%, P = 0.041) as well as late complications, which included hemorrhage, ileitis, ascites, and dumping syndrome (3.7% vs. 9.7%, P = 0.101). Despite these advantages, the ESD group revealed a reduced complete resection rate (92.6% vs. 100%), a lower 5-year disease-free survival rate (85% vs. 97%, P = 0.001), and a heightened 5-year cancer recurrence rate (12.3% vs. 2.1%, P = 0.001) [34]. In a parallel study conducted by Cho et al., no statistically significant disparities were found between ESD and surgical resection groups concerning overall survival rates (P = 0.565) and recurrence-free survival rates (P = 0.252). Nevertheless, the metastasis-free survival rate in the ESD group was significantly inferior compared to the surgical cohort (P = 0.002). Post-propensity score matching (PSM) analysis revealed no significant variances between the ESD and surgical cohorts in terms of overall survival (P = 0.691), recurrence-free survival (P = 0.073), and advanced cancer-free survival (P = 0.070). Conversely, the investigation led by Fukunaga et al. disclosed a superior 5-year overall survival (OS) rate for the ESD cohort when compared to the surgical group (97.1% vs. 85.8%, P = 0.01) [35]. Additionally, the ESD group experienced significantly fewer adverse events than the surgical group (6.8% vs. 28.4%, P < 0.01) [36]. To assess therapeutic disparities between these modalities, both pre-adjusted and post-adjusted (IPTW, PSM) Kaplan–Meier survival curves were plotted, demonstrating comparable efficacy between endoscopy and surgery. External data, separate from the SEER database, were also collected for further validation, corroborating the initial results. Not only does ESD reduce fasting duration, operative time, and hospital stay, but it also leads to fewer complications and, crucially, preserves the stomach, thereby enhancing patients’ quality of life. However, ESD treatments may fail due to margin invasion and lymph node involvement [29]. Reports indicate that the overall incidence of postoperative lymph node metastasis in early GC is approximately 20% [37]. ESD can effectively manage metastatic GC or local recurrence when lesions are detected early, necessitating vigilant surveillance endoscopy post-ESD. Previous studies have shown that endoscopy is as practical as a surgical intervention in treating early esophageal cancer [38]. In our study, both pre-adjusted and post-adjusted (IPTW, PSM) Kaplan–Meier survival analyses did not reveal any statistically significant differences between the ESD group and the surgical resection group. Compared to earlier research, our study cohort was more extensive, had a longer follow-up duration, and incorporated data from diverse sources to substantiate our findings.
Radiotherapy employs high-energy rays or particles to eradicate cancer cells and is occasionally used in treating GC, usually in conjunction with chemotherapy. Both neoadjuvant chemoradiotherapy and neoadjuvant chemotherapy markedly enhance clinical outcomes in patients with resectable GC [39]. Our study further compared the efficacy of endoscopy and radiotherapy, indicating a superior performance by the former.
Additionally, we developed a prognostic model to predict OS in patients with T1b GC. This model incorporates six variables: race, age, gender, degree of differentiation, tumor size, and treatment modality. By inputting these six variables, clinicians can ascertain the 1-, 3-, and 6-year OS rates for patients. Multiple validations confirmed the model’s reliability, and it was subsequently uploaded to a public website for clinical utilization. It is noteworthy that while the predictive model remains stable over time, patient prognosis could vary with changes in treatment modalities, potentially impacting the model’s accuracy. Further validation by clinical data is needed to determine its effectiveness in enhancing both patient and physician satisfaction.
This study is retrospective, and its conclusions warrant validation through a large-scale prospective study. It has several limitations that should be acknowledged. First, the SEER database lacks detailed treatment information, including: (1) the purpose and specifics of chemotherapy or radiotherapy, (2) documentation of underlying comorbidities, and (3) economic data. This makes it impossible to reliably compare long-term outcomes among patients receiving chemotherapy or radiotherapy. Second, the database provides only survival outcomes without information on postoperative recurrence or other potential complications. Third, the SEER database lacks data regarding risk factors for lymph node metastasis like submucosal and lymphovascular infiltration. Besides, the NCCN guidelines recommend endoscopic treatment only for patients with stage T1a disease and endoscopic treatment in patients with stage T1b disease is primarily intended for those who are not suitable for surgery, contributing to heterogeneity among patients. These limitations highlight the need for prospective studies incorporating comprehensive clinicopathological data to validate our findings.

Conclusion
Endoscopic therapy for T1b gastric cancer has shown an increasing trend in recent years, while gastrectomy rates have slightly declined. Long-term survival outcomes following endoscopic therapy for T1b gastric cancer are promising and suggest that it may be comparable to gastrectomy. We have also developed a robust network calculator that can be used to predict OS rates in patients with T1b GC. However, it is important to note that these results provide indications rather than definitive proof of equivalence between treatments.

Supplementary Information
Below is the link to the electronic supplementary material.