Postoperative perforation risk after endoscopic removal of gastric gastrointestinal stromal tumors: Development and validation of a predictive nomogram

Abstract

BACKGROUND

Gastrointestinal stromal tumors are the most common mesenchymal tumors of the gastrointestinal tract, and gastric gastrointestinal stromal tumors (gGISTs) account for the majority of these tumors. Currently, endoscopic removal (ER) is increasingly adopted as a minimally invasive treatment. However, postoperative perforation remains a critical complication, necessitating robust prediction tools.

AIM

To identify the risk factors and develop a validated nomogram for predicting perforation after ER of gGISTs.

METHODS

This retrospective study analyzed the patients undergoing ER at Fuyang People’s Hospital from 2019 to 2024. Clinical data, including tumor size, location, and procedural details, were collected and analyzed. The risk factors were identified via univariate and multivariate logistic regression, and a nomogram was developed. Both internal and external validations were performed, and the model performance was assessed by receiver operating characteristic curves and calibration plots.

RESULTS

Among 301 patients, the perforation rate was 6.3%. Multivariate analysis identified tumor size (odds ratio = 4.699, 95% confidence interval: 2.382-9.267, P = 0.001) and cardia/fundus location (odds ratio = 3.492, 95% confidence interval: 1.121-10.875, P = 0.031) as independent predictors. A nomogram was constructed and achieved good predictive performance in both the training (area under the curve = 0.881) and validation sets (area under the curve = 0.878).

CONCLUSION

This study identified that tumor size and location were independent risk factors, and provides a clinically actionable nomogram for evaluating and predicting postoperative perforation risk in gGISTs treated with ER, facilitating preprocedural planning and risk monitoring.

Key Words: Gastric gastrointestinal stromal tumors; Endoscopic removal; Perforation; Nomogram; Predicting model; Logistic regression

Core Tip: Gastric gastrointestinal stromal tumors are the most common mesenchymal tumors of the gastrointestinal tract, with endoscopic resection being a preferred minimally invasive treatment. Postoperative perforation is a severe complication, necessitating a robust predictive tool. In this retrospective study, we enrolled 301 patients who underwent endoscopic removal for gastric gastrointestinal stromal tumors at Fuyang People’s Hospital. Risk factors were identified via univariate and multivariate logistic regression. A nomogram was then constructed and validated. Multivariate analysis identified tumor size and cardia/fundus location as independent risk factors. The nomogram showed good predictive performance, with an area under the receiver operating characteristic curve of 0.881 in the training cohort and 0.878 in the validation cohort, with calibration curves demonstrating consistency between predicted and observed perforation rates.

INTRODUCTION

Gastrointestinal stromal tumors (GISTs) are a group of tumors originating from the mesenchymal tissue of the gastrointestinal tract, and the estimated annual incidence of GISTs is approximately 10-20 cases per 1 million population[1]. Among them, 20%-30% of GISTs have malignant potential[2,3]. The stomach is the most common site of occurrence, accounting for as high as 60%-65% of cases, imposing a significant health burden on patients[4]. As the most common subtype of GISTs, gastric GISTs (gGISTs) have garnered significant attention due to their unique biological behaviors and clinical characteristics[5,6]. With the continuous advancement of endoscopic techniques, endoscopic removal (ER) has emerged as a crucial therapeutic modality for gGISTs, leveraging advantages such as minimal invasiveness, rapid recovery, and minimal impact on patients’ physiological functions[7,8]. Techniques such as endoscopic submucosal dissection (ESD), endoscopic submucosal excavation, and endoscopic full-thickness resection (EFTR) are suitable for patients with gGISTs across different disease profiles[9,10].

However, endoscopic treatment of gGISTs is not without risks, including bleeding, infection, and perforation[5,11,12]. Postoperative perforation - defined as unplanned, accidental perforation that occurs after ER, excluding iatrogenic perforations intentionally induced during procedures such as EFTR for therapeutic purposes - is one of the most severe complications of this procedure. This complication is distinct from intraoperative iatrogenic perforation, as it arises unexpectedly during the postoperative period rather than being a controlled part of the surgical technique. Although its incidence varies across studies and treatment centers, when it occurs, it can lead to serious consequences such as peritonitis, intra-abdominal infection, and hemorrhage, significantly prolonging hospital stays, increasing patient suffering and medical costs, and even threatening patients’ lives[11,13]. Statistics show that approximately 2%-8% of patients develop postoperative perforation after endoscopic treatment of gGISTs, which has become a key issue affecting the clinical efficacy and safety of endoscopic therapy[11,14,15]. Therefore, identifying risk factors for postoperative perforation in gGISTs and constructing effective predictive models are of profound significance for clinical decision-making.

A nomogram, a visual predictive tool, is typically constructed based on logistic regression or other statistical models[16-18]. Characterized by high readability and cost-effectiveness, it can be directly printed or included in clinical work manuals, enabling clinicians to use it readily for real-time clinical decision-making without relying on complex statistical software. This accessibility has made nomograms highly popular among clinicians. This study retrospectively collected clinical and pathological data on patients with gastric stromal tumors treated endoscopically. We aimed to identify risk factors for postoperative perforation, develop a nomogram predictive model, and provide clinicians with a precise risk-assessment tool to enhance treatment safety, efficacy, and prognosis.

MATERIALS AND METHODS

Study design

This retrospective study enrolled gGISTs patients admitted to the Department of Gastroenterology, Fuyang People’s Hospital Affiliated to Anhui Medical University from January 2019 to December 2024. The research design and flowchart are shown in Figure 1. Patients were included in the study if they met the following criteria: (1) Diagnosed with gGISTs and treated at the Department of Gastroenterology, Fuyang People’s Hospital; (2) Aged between 18 and 80 years old; (3) Had relatively complete clinical data, including basic demographics, laboratory test results, enhanced computed tomography (CT) imaging characteristics, gastroscopy images, operation time, surgical approach; (4) Underwent ER of gGISTs (e.g., ESD, endoscopic submucosal excavation, EFTR); (5) Provided written informed consent; and (6) Were pathologically confirmed to have stromal tumors postoperatively. Patients were excluded if any of the following criteria was met: (1) Lack of pathological confirmation of a gastric stromal tumor; (2) Combined with other tumors; (3) Incomplete or missing clinical data; (4) Severe comorbidities or poor physical condition; and (5) History of immune system diseases, severe cardiovascular diseases, hepatic/renal failure, or other serious underlying conditions. Postoperative perforation was diagnosed based on a combination of clinical and radiological evidence. Clinical signs: Presence of persistent abdominal pain, tenderness, rebound tenderness, fever (body temperature > 38.5 °C), or leukocytosis (white blood cell count > 10 × 10⁹/L) after ER. Imaging findings: Abdominal CT showing pneumoperitoneum (free air in the peritoneal cavity), extraluminal contrast extravasation, or signs of peritoneal inflammation (e.g., fat stranding). Representative CT images of perforation cases are provided in Supplementary Figure 1. This study was approved by the Ethics Committee of Fuyang People’s Hospital.

Figure 1 A flowchart of the study. ROC: Receiver operating characteristic; AUC: Area under the curve.

Data collection

Clinical parameters, including sex, age, height, weight, body mass index, laboratory biochemical indices, CT imaging characteristics, gastroscopy images and information, operation time, surgical approach, and occurrence of postoperative perforation, were collected. Active intraoperative perforation (such as in EFTR) was not defined as accidental perforation and was excluded from this study.

Risk analysis and logistic regression analysis

All clinical parameters [such as tumor size, hemoglobin (HGB) level, and tumor location] were subjected to statistical analysis, and potential risk factors were preliminarily identified. Univariate logistic regression analysis was then performed on the potential risk factors to identify those factors associated with postoperative perforation. Variables with a P-value < 0.05 were retained for subsequent multivariate analysis to identify independent predictors of perforation.

Nomogram construction and validation

The patients were randomly divided into a training cohort and a validation cohort at a ratio of 2:1 (200:101). Based on the results of multivariate logistic regression analysis, a nomogram for predicting the probability of postoperative perforation was constructed using the training cohort. The nomogram was internally validated using 1000 resampling bootstrapping to assess its discriminative ability and calibration. Discriminative ability was evaluated by the receiver operating characteristic (ROC) curve and the area under the curve (AUC). Calibration was assessed by comparing the predicted probabilities with the observed outcomes through calibration curves. Subsequently, external validation of the nomogram was performed using the validation cohort, where ROC, AUC, and calibration curves were similarly calculated to confirm the generalizability of the model.

Statistical analysis

All statistical analyses were conducted using GraphPad Prism 8 and R.4.2.1 software. The measurement data were analyzed using the Student’s t-test or Mann-Whitney U test, while enumeration data were analyzed using the χ² or Fisher’s exact tests. The logistic regression model was constructed using the “glm” function, and the nomogram was plotted using the “rms” package in R. All statistical tests were two-sided, and a P-value < 0.05 was considered statistically significant.

RESULTS

Baseline characteristics

As shown in Figure 1, this study retrospectively included 344 patients with gGISTs who visited the Department of Gastroenterology, Fuyang People’s Hospital from January 2019 to December 2024. Among them, 12 cases were transferred to the Department of Surgical for surgical procedures, 11 cases only received drug treatment, 5 cases received combined dual-endoscopy treatment, and 15 cases lacked relevant clinical data; these patients were excluded. Finally, 301 patients with gGISTs who received ER were included. Among the included patients with gGISTs, 19 cases (6.3%) had postoperative perforation, and 282 cases (93.7%) did not. Of the patients with postoperative perforation, 14 cases improved after conservative treatment, and 5 cases, for whom conservative treatment was ineffective, received laparoscopic perforation repair. Baseline data of the relevant parameters are shown in Table 1.

Table 1 Baseline characteristics of patients with gastric gastrointestinal stromal tumors.

Characteristics	Perforation group (n = 19)	Non-perforation group (n = 282)	t/χ2	P value
Age (years)	59.89 ± 9.50	59.96 ± 11.07	0.024	0.981
Gender			0.806	0.369
Male	5	103
Female	14	179
History of hypertension			0.228	0.633
Yes	5	89
No	14	193
History of diabetes			/	0.141
Yes	4	29
No	15	253
History of stroke			/	0.169
Yes	3	20
No	16	262
Height (m)	1.62 ± 0.04	1.63 ± 0.08	0.017	0.987
Weight (kg)	62.61 ± 7.27	61.49 ± 10.50	0.453	0.651
BMI (kg/m2)	23.57 ± 2.12	23.126 ± 3.21	0.584	0.553
HGB (g/L)	117.47 ± 15.17	125.27 ± 16.40	2.008	0.046
WBC (× 109)	5.43 ± 1.34	5.04 ± 1.56	1.066	0.287
ALB (g/L)	39.07 ± 4.69	39.80 ± 3.26	0.915	0.361
Glu (mmol/L)	5.16 ± 0.51	5.30 ± 1.20	0.503	0.616
Tumor location			4.830	0.028
Cardia/fundus	13	120
Antrum/body	6	162
Tumor size (cm)	3.53 ± 1.15	1.56 ± 0.85	10.73	0.001
Growth pattern			5.879	0.015
Exophytic type	13	113
Non-exophytic type	6	169
Operative approach			1.458	0.483
ESD	5	47
ESE	4	85
EFTR	10	150
Operation duration (minutes)	135.12 ± 79.80	73.55 ± 51.19	4.752	0.001
Risk assessment			80.35	0.001
Very low risk	1	210
Low risk	10	63
Intermediate risk	1	3
High risk	7	6

BMI: Body mass index; HGB: Hemoglobin; WBC: White blood cell; ALB: Albumin; Glu: Glucose; ESD: Endoscopic submucosal dissection; ESE: Endoscopic submucosal excavation; EFTR: Endoscopic full-thickness resection.

Risk factors identified

As shown in Table 1 and Figure 2, higher preoperative HGB levels were associated with reduced postoperative perforation risk (P = 0.046), and tumors located in the gastric fundus/body significantly reduced perforation risk compared to the cardia/antrum (P = 0.028). In addition, larger tumor size (P = 0.001), exophytic growth pattern (P = 0.015), and longer operation duration (P = 0.001) were associated with higher postoperative perforation risk.

Figure 2 Analysis of risk factors for postoperative perforation in gastric gastrointestinal stromal tumors after endoscopic removal. ^aP < 0.05; ^cP < 0.001.

Logistic regression analysis

As shown in Table 2, univariate analysis demonstrated that lower HGB level [odds ratio (OR) = 0.974, 95% confidence interval (CI): 0.948-0.999, P = 0.046], cardia/fundus location (OR = 2.925, 95%CI: 1.081-7.917, P = 0.035), larger tumor size (OR = 3.366, 95%CI: 2.173-5.211, P = 0.001), and longer operation duration (OR = 1.012, 95%CI: 1.006-1.018, P = 0.001) were associated with higher postoperative perforation risk. However, tumor growth pattern (exophytic type vs non-exophytic type) was not associated with postoperative perforation risk in this study. Multivariate analysis showed that cardia/fundus location (OR = 3.492, 95%CI: 1.121-10.875, P = 0.031) and larger tumor size (OR = 4.699, 95%CI: 2.382-9.267, P = 0.001) were independent risk factors.

Table 2 Logistic regression analysis.

Characteristics	Univariate logistic regression analysis			Multivariate logistic regression analysis
	OR	95%CI	P value	OR	95%CI	P value
HGB (g/L)	0.974	0.948-0.999	0.046	1.025	0.993-1.058	0.129
Tumor location
Antrum/body	1
Cardia/fundus	2.925	1.081-7.917	0.035	3.492	1.121-10.875	0.031
Tumor size (cm)	3.366	2.173-5.211	0.001	4.699	2.382-9.267	0.001
Growth pattern				/
Exophytic type	1
Non-exophytic type	0.390	0.149-1.021	0.055
Operation duration (minutes)	1.012	1.006-1.018	0.001	0.999	0.989-1.008	0.804

OR: Odds ratio; CI: Confidence interval; HGB: Hemoglobin.

Subgroup analysis

To further analyze the results of the logistic regression, we completed a corresponding subgroup analysis (Figure 3). The analysis showed that patients with anemia (HGB < 120 g/L) (P = 0.003), exophytic growth type (P < 0.001), and operation time ≥ 60 minutes (P < 0.001) had significantly larger gastric stromal tumors, with a demonstrated correlation between these factors and tumor size. In contrast, tumor location (cardia/fundus vs antrum/body) was not associated with tumor size (P = 0.556).

Figure 3 Tumor size distribution across clinical parameters in gastric gastrointestinal stromal tumors. A: Tumor size was larger in patients with lower hemoglobin levels compared to those with higher hemoglobin levels; B: Tumor size between different locations (antrum/body vs cardia/fundus) was not statistically significant; C: Tumor size was larger in the exophytic type compared to the non-exophytic type; D: Longer operation duration was related to tumor size. ^bP < 0.01; ^cP < 0.001. HGB: Hemoglobin.

Nomogram construction and validation

Guided by the multivariate logistic regression analysis results, tumor size and tumor location were selected as key predictors to develop the nomogram model in the training cohort (Figure 4A). The model exhibited robust discriminative capability, as evidenced by the training set ROC curve (Figure 4B), which yielded an AUC of 0.881, indicating a strong ability to distinguish perforation probability. The validation cohort ROC curve showed an AUC of 0.878, closely comparable to the training results, further validating the model’s generalizability (Figure 4C). Calibration curves in both the training (Figure 4D) and validation (Figure 4E) cohorts demonstrated good agreement between predicted and observed outcomes, highlighting the nomogram model’s robustness.

Figure 4 Development and validation of a nomogram for predicting postoperative perforation in gastric gastrointestinal stromal tumors. A: Nomogram depicting the contribution of key variables (tumor location, tumor size) to the predicted probability of perforation after endoscopic resection; B: Receiver operating characteristic curve for the training cohort. The area under the curve was 0.881; C: Receiver operating characteristic curve for the validation cohort (area under the curve = 0.878); D: Calibration curve for the training cohort; E: Calibration curve for the validation cohort. ROC: Receiver operating characteristic; AUC: Area under the curve.

DISCUSSION

According to the GISTs-related clinical guidelines, ER is the preferred minimally invasive approach for gGISTs[19,20]. However, postoperative perforation, as a severe complication, directly affects patient prognosis, increases medical burden, and may even lead to serious adverse events such as peritonitis and intra-abdominal infection. Through this single-center retrospective analysis, the study enrolled 301 patients with gGISTs who underwent endoscopic resection in our hospital, systematically explored the risk factors for postoperative perforation, and constructed a clinical prediction model to assess perforation risk. The results confirmed that tumor size and anatomical location (cardia/fundus) were independent high-risk factors for postoperative perforation. The constructed nomogram, validated by external verification and ROC curve analysis (training cohort: AUC = 0.881; validation cohort: AUC = 0.878), demonstrated good robustness and clinical utility in predicting perforation risk.

In this study, we confirmed that tumor size was significantly positively correlated with the incidence of postoperative perforation, a finding that is highly consistent with previous studies. A multicenter study analyzing ESD for early gastric cancer found that a tumor diameter > 20 mm was an independent risk factor for postoperative perforation, which is highly consistent with the results of this study[21]. In addition, a large-sample retrospective study has shown that a tumor diameter > 2 cm is an independent risk factor for perforation after ER of gGISTs[22]. When gastric stromal tumors are too large, the area of muscularis defect formed after resection increases correspondingly, making it difficult to achieve precise three-layer (mucosa-muscularis-serosa) apposition with endoscopic clips or purse-string sutures. Intraluminal pressure generated by gastric peristalsis can gradually expand the incompletely closed wound, eventually leading to perforation. Additionally, larger tumors are often associated with more complex blood supply systems, and incomplete electrocoagulation during surgery can easily induce secondary hemorrhage. Repeated hemostatic procedures further damage the muscularis tissue, weakening its healing capacity. Also, larger-diameter tumors typically require longer operative times, not only increasing operator fatigue-related errors but also elevating gastric wall tension due to continuous intra-luminal insufflation, exacerbating muscularis injury.

In this study, the antrum and body were grouped together, as were the cardia and fundus. This is because the antrum and body are similar in muscular layer thickness, blood supply, and difficulty of endoscopic procedures; while the cardia and fundus share common features such as a thinner muscularis propria and the need for retroflexion during endoscopic observation, all of which affect surgical risk. This study found that gGISTs located in the gastric fundus and cardia are more prone to postoperative perforation, a conclusion that is highly consistent with previous studies[22]. The muscularis layer of the gastric fundus is significantly thinner than that of other gastric regions, with a sparse submucosal vascular plexus and relatively poor blood supply. Even if perforation is sutured during the operation, the healing capacity of the muscularis defect remains poor and prone to delayed perforation. The cardia is located at the esophagogastric junction, requiring retroflexion for endoscopic surgery, which reduces the precision of clip deployment and increases suture difficulty. Additionally, the cardia’s rich submucosal nerve plexus is susceptible to vagal reflexes during electrocoagulation, causing abrupt heart rate drops or gastrointestinal motility disorders, with frequent intraoperative peristaltic interference further exacerbating precise suture challenges.

Subgroup analysis further revealed significant correlations between tumor size and factors including HGB level, tumor growth pattern, and operation duration[23]. Studies have shown that larger tumors are often associated with intratumoral vascular malformations or hemorrhage, leading to lower HGB levels[24]. Exophytic growth patterns, unrestricted by gastric lumen space, tend to develop into larger volumes; conversely, excessively large tumors often penetrate the serosa to grow extraluminally, forming an exophytic phenotype, indicating a bidirectional correlation[25]. Additionally, larger tumors require more complex resection techniques (such as EFTR) and extensive, difficult suturing, significantly prolonging operation time.

The final model in this study retained only two variables: Tumor size and location. While this enhances the simplicity of the model, potentially predictive variables such as operator experience, tumor growth pattern, and HGB level were not included. Regarding operator experience, the exclusion was due to the standardized training system implemented in our center for the ER of gGISTs. All doctors have over 5 years of endoscopic operation experience and hold the title of associate chief physician, with complex cases (e.g., EFTR) performed by a fixed team. This setup minimizes the impact of differences in experience and technical proficiency among operators on outcomes, thus weakening their independent predictive value. As for tumor growth pattern and HGB level, they were not included in the final model because subgroup analysis revealed their significant correlation with tumor size: Larger tumors are more likely to present with exophytic growth and be associated with clinical manifestations such as anemia. To avoid multicollinearity and prevent the inclusion of excessively weakly correlated variables from reducing the model’s accuracy, these variables were excluded. However, from a clinical perspective, these factors may be related to perforation. Future multicenter studies with large sample sizes, incorporating standardized recording of variables like operator experience, are needed to further validate their predictive value in larger datasets.

The nomogram model constructed in this study included tumor size and location as independent predictors, with an AUC of 0.881 in the training set and 0.878 in the validation set, indicating good discriminative ability. The nomogram model established in this study uses tumor size as the core predictor, with a stepwise risk assignment based on tumor size. The anatomic tumor location is also incorporated as a key auxiliary predictor, enabling precise quantification of risk assessment. Calibration curves showed high consistency between predicted probabilities and actual perforation rates. However, single-center data may have bias and are more susceptible to confounding factors, and molecular markers were not included. Multicenter studies and integration of gene testing are needed in the future to optimize the model.

This study has certain limitations. Firstly, as a single-center retrospective study, the data were derived from cases at Fuyang People’s Hospital between 2019 and 2024, with obvious regional characteristics and an insufficient proportion of perforation cases, which may affect the generalizability of the model. Secondly, some clinical data were missing, variables such as surgeons’ experience were not recorded in a standardized manner, and the development of endoscopic techniques during the study period may affect the stability of the results. In addition, genetic risk factors were not explored, the measurement of tumor volume based on preoperative enhanced CT had certain errors, the classification of tumor growth patterns was rough, and more detailed radiomics indicators were not included, affecting the accuracy of risk assessment. Future research can be optimized in multiple aspects. Multicenter studies should be conducted to include more cases for model validation. Genetic testing data and radiomics data should be integrated to construct a more accurate model. Prospective cohort studies should be performed to develop practical clinical tools. Mechanistic studies on high-risk factors should be carried out, and randomized controlled trials should be designed to explore intervention strategies to reduce the perforation rate.

CONCLUSION

This retrospective study identified tumor size and location as independent risk factors for postoperative perforation after ER of gGISTs. In addition, a nomogram was developed for predicting postoperative perforation risk after ER of gGISTs. The nomogram model showed good predictive performance, facilitating clinical risk assessment.