Validation of a clinical scoring model and proposal of a classification system to guide endoscopic resection strategies for cardial submucosal tumors

Abstract

Background and aims:

A scoring model was proposed to support endoscopic decision-making for cardial submucosal tumors (SMTs). The aim of this study is to perform a multicenter validation of the clinical scoring model and to introduce a new clinical classification system for cardial SMTs.

Methods:

A multicenter analysis of endoscopic decision-making for cardial SMTs was conducted. Individual data on patient characteristics, lesion features, and resection outcomes were collected to validate a clinical scoring model designed to guide endoscopic resection strategies. In addition, a novel classification system for cardial SMTs is proposed to support standardized assessment and treatment planning.

Results:

The scoring model and classification system achieved AUCs of 0.815 (95% CI: 0.748–0.881) and 0.809 (95% CI: 0.748–0.870), respectively. DeLong’s test showed no significant difference between them (P = 0.885).

Conclusions:

The scoring model and the classification system for cardial SMTs may provide clinicians with a practical reference for making appropriate endoscopic resection strategies.

Introduction

Endoscopic techniques, including submucosal tunneling endoscopic resection (STER) and non-tunneling methods, are minimally invasive treatments for cardial submucosal tumors (SMTs) that preserve curability^[1–3]. However, the cardia, as part of the esophagogastric junction (EGJ), remains a technically challenging site for endoscopic resection due to its narrow lumen and sharp angulation^[4].

HIGHLIGHTS

• This multicenter study validated the scoring system using real-world data, offering clinicians a practical tool for determining appropriate endoscopic resection strategies.

• We proposed a novel clinical classification system for cardial submucosal tumors (SMTs), which categorizes lesions into five distinct types. This system demonstrates predictive performance comparable to the scoring model and similarly serves to guide endoscopists in selecting optimal resection approaches for cardial SMTs.

Although prior studies have compared STER with non-tunneling techniques, no standardized clinical guidelines currently exist to assist in procedural selection for cardial SMTs^[5,6]. To address this gap, we previously developed a clinical scoring model to support endoscopic decision-making^[7]. This model assigns weighted points based on lesion features, including −2 for irregular morphology, 2 for ulceration, 1 for reversing direction, −2 for entering direction, and −2 for SMTs originating from the muscularis propria (MP) layer. A total score was calculated for each patient by summing the assigned points for all factors. Patients were stratified into low (< −4), intermediate (−4 to −3), and high (> −3) score groups.

To date, no standardized clinical guidelines exist to assist endoscopists in selecting optimal resection strategies for cardial SMTs, and existing decision-making tools have not undergone external multicenter validation. Moreover, while some reports have compared tunneling and non-tunneling techniques, no prior study has proposed a classification framework integrating lesion location, morphology, and technical feasibility. Therefore, this study aims to validate a previously developed scoring model across multiple centers and introduce a novel classification system that offers a standardized, anatomy-based approach to guide procedural selection in clinical practice.

This study has been reported in accordance with the STROCSS guidelines and the TITAN Guidelines 2025^[8,9].

Methods

Patients

A multicenter cohort study was conducted at six hospitals between March 2021 and March 2025. Patients with cardial SMTs were enrolled if they had complete clinical data, no history of upper gastrointestinal surgery, and were scheduled to undergo endoscopic resection. Data on patient demographics, lesion characteristics, procedural details, and clinical outcomes were collected. The study was approved by the institutional ethics committee (B-2018-222) and conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants.

Endoscopic procedure

The procedures included in our study were performed by experienced endoscopists, each having completed over 300 therapeutic cases prior to study initiation, and all participating centers are recognized tertiary referral centers for complex endoscopic resections.

To perform tumor dissection, a standard forward-viewing endoscope (GIF-Q260J; Olympus, Tokyo, Japan) fitted with a transparent distal cap (D-201-11804; Olympus) was employed. A combination of electrosurgical knives – including an I-type hybrid knife (ERBE; Erbe Elektromedizin GmbH, Tübingen, Germany), a hook knife (KD-620LR; Olympus), and an insulated-tip knife (IT knife; KD-611 L, Olympus) — was utilized based on lesion characteristics and anatomical location. Additional instruments included an argon plasma coagulation unit (APC300; Erbe), hot biopsy forceps (FD-410LR; Olympus), and a standard polypectomy snare.

All procedures were performed under general anesthesia with endotracheal intubation. Carbon dioxide (CO₂) insufflation was used to minimize the risk of gas-related adverse events. In cases of intraoperative pneumoperitoneum, decompression was achieved via needle puncture using a 20-gauge catheter inserted in the right lower quadrant.

In non-tunneling approaches such as endoscopic submucosal dissection (ESD) and endoscopic full-thickness resection (EFTR), the procedure involves direct mucosal incision followed by stepwise submucosal dissection and tumor removal after endoscopic localization^[1,2]. In contrast, the STER technique utilizes a submucosal tunnel to access and resect the tumor while maintaining mucosal integrity, enabling precise dissection under stable endoscopic visualization^[3,10].

ESD procedure

ESD was performed for lesions arising from the superficial muscularis propria with an intraluminal growth pattern^[11]. The procedure involved: (1) marking the mucosa approximately 5 mm beyond the lesion margin; (2) submucosal injection of saline mixed with indigo carmine; (3) mucosal and submucosal incision using a hybrid I-type knife; (4) progressive dissection around the tumor base using an IT knife; (5) lesion retrieval with a snare; (6) hemostasis using argon plasma coagulation or hot biopsy forceps; and (7) defect closure with endoscopic clips.

EFTR procedure

EFTR was used for tumors originating from the deep muscularis propria or serosa with an extraluminal growth pattern^[12]. The technique shared initial steps with ESD but involved intentional full-thickness perforation to achieve complete resection. When feasible, the serosal layer was preserved to minimize gas leakage and intra-abdominal contamination. Small to moderate defects were closed using standard clips. For larger defects (>15–20 mm), closure was achieved with a combination of clips and nylon loops, or by employing the omental patch technique, depending on lesion size and location.

STER procedures

STER offers the advantage of minimizing leakage risk due to mucosal preservation; however, its applicability is anatomically limited. The esophagus and cardia remain the most favorable sites for this technique. The procedure typically consists of the following key steps: (1) submucosal injection of saline mixed with indigo carmine approximately 2–4 cm proximal to the lesion; (2) mucosal incision to establish the tunnel entry; (3) creation of a submucosal tunnel via stepwise dissection between the submucosal and muscularis propria layers; (4) tumor resection using ESD or EFTR techniques; (5) en bloc retrieval with a snare; (6) hemostasis with argon plasma coagulation or hot biopsy forceps as needed; and (7) closure of the mucosal entry with endoscopic clips.

Following endoscopic resection, a nasogastric tube was placed to facilitate gastric decompression and allow early detection of delayed bleeding. Postoperative monitoring included assessment for clinical signs such as fever, hematemesis, melena, and abdominal pain. In cases of persistent or worsening symptoms, emergent endoscopy and abdominal computed tomography were performed to evaluate for adverse events. All patients received intravenous proton pump inhibitors to promote mucosal healing and reduce the risk of postoperative bleeding.

Statistical analysis

Continuous variables are presented as means ± standard deviations (SDs), and categorical variables as numbers (percentages). Baseline characteristics were compared using Student t- and χ²-tests. The model’s discrimination was assessed with receiver operating characteristic (ROC) curves. The DeLong test was performed to evaluate the statistical significance of the difference between the AUCs of the scoring model and the classification system. Calculations were performed with SPSS 26.0 and R version 4.0.2.

Results

The comparison between STER and non-tunneling techniques in the validation cohort

A total of 200 patients with cardial SMTs treated by endoscopic resection at six hospitals were included in this study, comprising 68 STER procedures and 132 non-tunneling techniques. As shown in Table 1, there were no significant differences between the two groups in terms of patient demographics (age and gender), tumor growth pattern, maximum diameter, pathological type, procedure time, en bloc resection rate, complete resection rate, hospital stay, complications, or recurrence (P > 0.05). SMTs resected by STER tended to have irregular morphology, originate from the muscularis propria layer, and were more often observed in the entering direction. In contrast, SMTs treated with non-tunneling techniques more frequently showed ulcerative changes on the mucosal surface and were more often visualized in the reversing direction.

Table 1.

Demographic information, lesion characteristics and procedural outcomes of the STER and non-tunneling groups

	STER (n = 68)	Non-tunneling technique (n = 132)	P value
Demographic information
Male, n (%)	32 (47.1)	57 (43.2)	0.601
Age (years), mean ± SD	49.5 ± 12.8	53.2 ± 12.8	0.056
Lesion characteristics
Growth pattern, n (%)			0.269
Intraluminal growth	64 (94.1)	118 (89.4)
Extraluminal growth	4 (5.9)	14 (10.6)
Morphology, n (%)			0.001
Regular	24 (35.3)	79 (59.8)
Irregular	44 (64.7)	53 (40.2)
Mucosa, n (%)			0.024
Smooth	68 (100.0)	120 (90.9)
Ulcerative	0 (0.0)	12 (9.1)
Max diameter (mm), mean ± SD	27.1 ± 17.3	25.9 ± 18.2	0.671
Relationship to the cardia, n (%)			0.000
Type 1	18 (26.5)	8 (6.1)
Type 2	21 (30.9)	3 (2.3)
Type 3	6 (8.8)	36 (27.3)
Type 4	20 (29.4)	5 (3.8)
Type 5	3 (4.4)	80 (60.6)
Direction of the gastroscope, n (%)			0.000
Enter	25 (36.8)	7 (5.3)
Reverse	14 (20.6)	72 (54.5)
Both	29 (42.6)	53 (40.2)
Layer, n (%)			0.043
Muscularis mucosa and submucosa	2 (2.9)	15 (11.4)
Muscularis propria	66 (97.1)	117 (88.6)
Histopathologic evaluation, n (%)			0.132
Leiomyoma	61 (89.7)	103 (78.0)
GIST	5 (7.4)	20 (15.2)
Lipoma	0 (0.0)	5 (3.8)
Cyst	1 (1.5)	2 (1.5)
Schwannoma	1 (1.5)	0 (0.0)
Inflammatory mass	0 (0.0)	2 (1.5)
Procedural outcomes
Nasogastric tube	20 (29.4)	103 (78.0)	0.000
Surgery time (min), mean ± SD	66.4 ± 57.0	67.1 ± 45.3	0.927
En bloc resection rate, n (%)	68 (100.0)	129 (97.7)	0.523
Complete resection rate, n (%)	68 (100.0)	129 (97.7)	0.523
Hospital stay (day), mean ± SD	5.8 ± 9.0	6.0 ± 11.3	0.893
Complications, n (%)	1 (1.5)	11 (8.3)	0.105
Recurrence, n (%)	0 (0.0)	0 (0.0)	-

Values in bold type are statistically significant.

STER, Submucosal tunneling endoscopic resection; SD, Standard deviation.

A total of 12 patients experienced procedure-related complications. Among them, one case of post-STER tunnel infection occurred, which was successfully managed with adequate drainage via nasogastric tube placement and intravenous antibiotics. The remaining 11 complications occurred in the non-tunneling technique group. These included one case of delayed post-procedural bleeding, which was treated successfully with endoscopic hemostasis; five cases of delayed microperforation – these SMTs were located deep within the muscularis propria layer and were managed endoscopically using reinforced clipping; and five cases of localized infection, all of which resolved with adequate drainage and antibiotic therapy.

Performance assessment of the scoring model

We developed a scoring model to guide endoscopic decision-making for cardial SMTs. Points were assigned as follows: −2 for irregular morphology, 2 for ulceration, 1 for reversing direction, −2 for entering direction, and −2 for tumors originating from the MP layer. A total score was calculated for each patient by summing these values. Higher scores were associated with increased use of non-tunneling techniques (Table 2). Patients were stratified into low (< –4), intermediate (–4 to –3), and high (> –3) score groups. The corresponding STER rates were 100.0%, 45.1%, and 14.6%, respectively (Table 3). The scoring system demonstrated good discriminatory ability, with an AUC of 0.815 (95% CI: 0.748–0.881) (Fig. 1) ^[13].

Table 2.

Distribution of scores for surgical decision-making in the validation cohorts

Total points	Validation cohort
	Patients (n = 200)	STER (n = 68)	STER rate (%)
−6	13	13	100.0
−4	47	33	70.2
−3	44	8	18.2
−2	47	8	17.0
−1	35	6	17.1
0	6	0	0.0
1	5	0	0.0
2	1	0	0.0
3	2	0	0.0

STER, Submucosal tunneling endoscopic resection.

Table 3.

Classification for surgical decision-making in the validation cohorts

Category	Total points	Validation cohort
		Patients (n = 200)	STER (n = 68)	STER rate (%)
Low	<−4	13	13	100.0
Medium	−4 to −3	91	41	45.1
High	>−3	96	14	14.6

STER, Submucosal tunneling endoscopic resection.

Figure 1.

The ROC curve of the clinical scoring model.

Clinical classification system

Drawing from multicenter experience, we proposed a novel clinical classification system for cardial SMTs (Fig. 2):

Figure 2.

A novel clinical classification system for cardial SMTs.

Type 1: Located in the distal esophagus, adjacent to but not crossing the cardia

Type 2: Crossing the cardia, with the majority of the lesion in the distal esophagus

Type 3: Crossing the cardia, with the majority of the lesion below the cardia

Type 4: Located on the lesser curvature below the cardia

Type 5: Located on the greater curvature below the cardia

Based on this classification, Types 1, 2, and 4 are considered suitable for STER, while Types 3 and 5 are better suited for non-tunneling techniques. The classification system demonstrated good diagnostic performance with an AUC of 0.809 (95% CI: 0.748–0.870) (Fig. 3). DeLong’s test showed no significant difference in AUC between the scoring model and the classification system (P = 0.885).

Figure 3.

The ROC curve of the classification system.

Discussion

With advances in endoscopic techniques, an increasing number of cardial SMTs are being treated endoscopically. According to published guidelines and consensus from the ESGE, such features warrant resection due to the potential for malignant transformation or growth-related complications (e.g., bleeding, obstruction, or ulceration)^[14]. Preoperative differentiation between low-risk and higher-risk SMTs remains limited with current imaging modalities. For many lesions, especially those with endoscopic features such as ulceration, irregular shape, or submucosal stiffness, a definitive histological diagnosis can only be achieved through complete resection. In rare cases, lesions ultimately diagnosed as lipomas were resected due to unclear preoperative imaging features, symptoms, or atypical endoscopic appearance. Improved preoperative diagnostic tools are needed to optimize patient selection and minimize unnecessary intervention.

Our findings suggest that endoscopic resection, including both STER and non-tunneling techniques, is a feasible and generally safe treatment option for cardial submucosal tumors when performed in high-volume centers with experienced endoscopists. Although procedure-related complications occurred in 12 cases (6.0%), all were successfully managed with conservative or endoscopic interventions, without the need for surgical conversion. These outcomes underscore the importance of appropriate case selection, technical expertise, and perioperative management in minimizing the risks of endoscopic resection in the cardia, a traditionally high-risk anatomical location. When conducted under optimal conditions, endoscopic resection offers a minimally invasive alternative to surgery with acceptable safety and efficacy profiles.

Due to the narrow lumen and sharp angle at the cardia, selecting the appropriate resection method is crucial for successful removal. In a previous study, we developed a scoring model to support endoscopic decision-making for cardial SMTs based on multiple objective parameters. However, the initial model was constructed using single center’s data, and multicenter studies are needed to validate its performance and predictive accuracy.

In this multicenter study, we validated our previous model. The area under the ROC curve in the prospective validation cohort was 0.815, and an increasing proportion of non-tunneling technique was observed with rising scores. Stratifying patients with the scoring system enables endoscopists to select the most appropriate endoscopic resection technique.

Based on multicenter experience, we classified cardial SMTs into five types. Type 1 and 2 cardial SMTs are primarily located in the distal esophagus, making tunnel creation relatively easy; thus, STER is the preferred approach. Type 3 cardial SMTs extend across the esophagogastric junction, with the bulk of the tumor located below the cardia, making tunnel establishment challenging – non-tunneling techniques are more suitable. Type 4 cardial SMTs are entirely located below the cardia on the lesser curvature side, where tunneling is relatively straightforward; therefore, STER is appropriate. In contrast, type 5 cardial SMTs are on the greater curvature side beneath the cardia, where the presence of the gastric fundus requires a sharp turn to complete the tunnel, making non-tunneling techniques more feasible.

The findings of this study contribute several important advancements to the literature. First, this is the first multicenter validation of a clinical scoring model to guide endoscopic resection strategies for cardial SMTs, demonstrating consistent performance across diverse patient populations and institutional practices. Second, we propose a novel five-type classification system that stratifies lesions based on anatomical location relative to the esophagogastric junction and technical considerations influencing procedural feasibility. This classification system achieved predictive accuracy comparable to the scoring model and provides a practical framework to standardize preoperative assessment and procedural planning. Together, these tools offer clinicians objective criteria to inform decision-making, potentially reducing technical complexity and improving patient outcomes in this anatomically challenging region. While intraoperative adjustment remains necessary, structured preoperative scoring supports case planning and facilitates safe, tailored treatment in anatomically complex sites such as the cardia.

Despite the strengths of this multicenter analysis, we recognize that heterogeneity among participating centers and patient populations represents an important limitation. Variations in operator experience, procedural techniques, and patient selection criteria could contribute to differences in outcomes and complicate direct comparisons. To mitigate this, we applied standardized inclusion criteria, uniform definitions for lesion characteristics and procedural endpoints, and consistent data collection protocols across sites. The AUCs consistently exceeded 0.80 across cohorts, supporting the robustness and potential generalizability of these tools in diverse clinical settings. Nonetheless, we acknowledge that unmeasured confounders and procedural nuances may have influenced the results. Future large-scale prospective studies incorporating centralized training and validation will be essential to further confirm these findings and refine decision-making frameworks for cardial SMTs. While we acknowledge that such case volume and experience may not be generalizable to all settings, we believe this study offers valuable insights into technical feasibility and selection strategy in centers where such procedures are performed routinely.

In conclusion, the validation of the score model yielded fair results using the multicenter data. Besides, the proposed classification system demonstrates comparable performance to the scoring model in guiding endoscopic resection strategies for cardial SMTs.

Key learning points

1. Multicenter validation confirmed the robust discriminatory performance of the scoring model (AUC 0.815).

2. The novel classification system demonstrated similar predictive value (AUC 0.809) and provides a practical, anatomy-based framework.

3. Stratification based on lesion location and morphology can guide appropriate procedural choice and optimize outcomes.

4. The integration of these tools may improve decision-making and reduce procedural complexity in endoscopic resection of cardial SMTs.

Ethical approval

This study was approved by the Ethics Committee of Zhongshan Hospital in accordance with the Declaration of Helsinki (B-2018-222).

Consent

Written consent from all patients before the procedures were obtained.

Sources of funding

This study was supported by the National Natural Science Foundation of China (82170555), Shanghai Academic/Technology Research Leader (22XD1422400), Shuguang Program of Shanghai Education Development Foundation and Shanghai Municipal Education Commission (2022SG06), Shanghai “Rising Stars of Medical Talent” Youth Development Program (20224Z0005), Shanghai Municipal Basic Research Program (Natural Science Foundation) (25ZR1402063) and the 74th General Support of China Postdoctoral Science Foundation (2023M740675), and Outstanding Resident Clinical Postdoctoral Program of Zhongshan Hospital Affiliated to Fudan University.

Author contributions

Z.-H.G. (Conceptualization: Equal; Data curation: Equal; Formal analysis: Equal; Investigation: Equal; Methodology: Equal; Software: Equal; Validation: Equal; Visualization: Equal; Writing – original draft: Lead; Writing – review & editing: Lead). M.-Y.W. (Conceptualization: Equal; Software: Equal; Writing – original draft: Equal; Writing – review & editing: Equal). M.-J.H. (Conceptualization: Equal; Software: Equal; Writing – original draft: Equal; Writing – review & editing: Equal). P.-Y.F. (Conceptualization: Equal). Y.Z. (Conceptualization: Equal). Z.-Z.L. (Conceptualization: Equal). L.H. (Conceptualization: Equal). Z.-B.J. (Conceptualization: Equal). J.-N.C. (Conceptualization: Equal). Y.-G.L. (Conceptualization: Equal). H.-B.D. (Conceptualization: Equal). Y.-J.Y. (Conceptualization: Equal). Z.W. (Conceptualization: Equal). L.N. (Conceptualization: Equal). Q.-L.L. (Conceptualization: Equal; Supervision: Equal). P.-H.Z. (Conceptualization: Equal; Supervision: Equal). J.-Z.L. (Conceptualization: Equal; Supervision: Equal).

Conflicts of interest disclosure

All authors have no conflicts of interest or financial ties to disclose.

Guarantor

Quan-Lin Li

Research registration unique identifying number (UIN)

ChiCTR2000035640

Provenance and peer review

Not commissioned, externally peer-reviewed

Data availability statement

Data are not available