Gastric submucosal tumors (SMTs) are occasionally found during routine gastroduodenoscopy (prevalence, 0.2%–3%) [1], most of which are gastrointestinal stromal tumors (GISTs) and leiomyomas. Based on the newly published European Society of Gastrointestinal Endoscopy (ESGE) guideline, the indications for the treatment of SMTs are as follows: risk of malignancy (GISTs, neuroendocrine neoplasms, and granular cell tumors); symptoms such as obstructions or bleeding; and lesions in specific locations for patients receiving bariatric surgery [2]. If there is an indication for resection, the ESGE suggests endoscopic resection (ER) as an alternative to laparoscopic surgery for gastric GISTs <35 mm in size that protrude into the gastric lumen. ER, which includes endoscopic submucosal excavation (ESE), endoscopic full-thickness resection (EFTR), and submucosal tunneling endoscopic resection (STER), can be performed to curatively resect tumors in an en bloc fashion [3,4], with a high R0 resection rate, a low serious complication rate, and a low recurrence rate [2].

The pooled estimate rate of delayed perforation was reported to be 0.14%–5.78%, and the pooled rate of delayed bleeding was 0.14%–1.01% [5,6]. ESE is an endoscopic submucosal dissection (ESD)–derived technology, and it is more suitable for the ER of SMTs originating from the muscularis propria (MP) [7]. For SMTs involving the serosa and extraluminal component, another ESD-derived technology known as EFTR is applied [8-10]. Although both of these procedures can curatively resect tumors in an en bloc fashion, perforation seems to result in more problems. Compared with EFTR, ESE tends to have advantages of earlier postoperative feeding, shorter postoperative hospital stays, lower hospitalization expenses, and a lower probability of infection [8,11]. Hence, attempts should be made to avoid perforation during ER if the circumstances are appropriate. Being able to predict the risk of perforation would help endoscopists be more cautious and be prepared ahead of time; moreover, appropriate patients could be selected for inexperienced endoscopists.

Endoscopic ultrasonography (EUS) is an important technique in the diagnosis of SMTs, as it can provide information on the layer from which the SMTs originate, as well as the size, borderline, and echoes to help evaluate the possible pathology of the SMTs [12]. Additionally, it can show information on the membrane integrity, spatial relationship between the tumor capsule and the serosal layer, and other factors that predict procedural complications and patients’ prognoses [13]. Zhang et al. [13] showed that SMTs located in the superficial MP with a length of the muscular connection <40% of the tumor capsular semi-circumference may have a lower risk for perforation.

In real-world clinical practice, it is sometimes difficult to differentiate the superficial and deep MP, due to a lack of a quantification index and the fact that the length of the connection is hard to measure, as it is usually a curved line rather than a straight line. The authors agree with the above-mentioned researchers that a deeper location of the SMT is associated with a higher risk of perforation, and a quantification index would help prove this hypothesis. Therefore, this study was conducted to determine important parameters in EUS images to predict perforation and to attempt to build predictive models to guide clinical practice.

In this study, the patients in the case group tended to be older, have lower ALB levels, have SMTs more commonly located at the gastric fundus, have a higher I/O ratio, and have a higher proportion of GISTs or schwannomas. The multivariate analysis further helped in building two models to predict perforation, and the AUCs of model 1 and model 2 were 0.836 and 0.755, respectively.

In the case group, a certain proportion (21.21%) of patients had abdominocentesis during ER; these data have not been published in previous studies, thus indicating the importance of avoiding perforation if the situation is appropriate. A literature search demonstrated that there was still a lack of articles investigating risk factors using EUS images to predict perforation. Zhang et al. analyzed 43 patients and found that more SMTs in the perforation group were located in the deep MP and had a length of the muscular connection ≥40% of the tumor capsular semi-circumference; however, they did not perform univariate or multivariate analyses [13]. Moreover, it was difficult to apply their findings in realistic situations. The authors’ experiences supported the proposal that deeper SMTs have a higher risk of perforation; hence, this study developed a new parameter known as the I/O ratio. The reason for the use of this ratio is that this was a retrospective study, meaning that it was not possible to know the absolute distance; however, it was possible to measure the relative distance, so the ratio was adopted, with a larger ratio indicating a deeper SMT. For the first time, these results proved that the I/O ratio was much higher in the case group, and the univariate and multivariate analyses further supported that the I/O ratio was a risk factor for perforation. This phenomenon is considered reasonable, as a deeper tumor location means that deeper layers were cut, and the higher chance it had to be perforated in ER.

Age was another risk factor for perforation and was selected as a parameter in the model in this study. However, in previous studies, the researchers thought that age was not a risk factor for perforation, but was a possible risk factor for delayed bleeding [15,16]. The reason for this phenomenon may be that the present study regarded age as a continuous variable and did not divide the patients into different age groups, whereas in previous studies, the researchers classified the patients into different age groups. The reason why age was a risk factor for perforation was unknown, although a possible explanation is that older patients were more likely to have comorbid hypertension or other diseases involving vascular sclerosis.

The univariate analysis showed that SMTs located at the gastric fundus were also a risk factor for perforation. Previous studies have supported the idea that tumors located at the upper portion of the stomach were more likely to be perforated [17,18]. There may be several reasons for this. First, it was difficult to reach the fundus due to retroflexion of the endoscope. Second, the knife was vertically oriented to the fundus, which could also result in a prolonged procedure. Third, the gastric fundus is thinner than other parts of the stomach. Fourth, the submucosal arteries were larger in the upper stomach, and intraoperative bleeding may have impaired the field of view during ER [3,19-21]. Moreover, tumors located at the upper portion of the stomach were reported to be associated with incomplete resection [18]; additionally, the spleen and many blood vessels surrounding the gastric fundus increased the risk of other complications [3]. Hence, it was suggested that only experienced endoscopists should perform such procedures [3].

Another interesting finding of this study is that the pathology of the tumors was a risk factor for perforation, as leiomyoma was a protective factor, whereas GIST and schwannoma were risk factors. A reason for this phenomenon may be that, compared with other tumors, leiomyoma was softer, with greater elasticity and better compliance, whereas GIST and schwannoma were relatively harder and were more likely to be perforated. These results were slightly different from a previous study [13], as those researchers showed no difference for perforation regarding pathology; however, in their study, their sample size was smaller than that in this study. Moreover, Hsu et al. [22] showed that ESD procedures tended to be more likely to involve complications (perforation, delayed bleeding, and surgical interventions) for GISTs than for leiomyomas, which was in accordance with the present results in some ways. The predictive models were constructed with consideration of the fact that it was often not possible to know the pathology result of the SMT before ER; therefore, two models were built, differing in terms of whether pathology was included or not. If the pathology of the SMT is known, model 1 could be used. Of note, a study showed that artificial intelligence (AI) technology could help in differentiating GISTs and non-GISTs [23], especially for cases with sizes ≥20 mm. However, at the present time, the technology of AI has not been widely used in clinical practice; thus, if the pathology is not known in advance, model 2 could be used to make a prediction.

This study had some limitations. First, this was a single-centered study, and the predictive models that were built were not tested in other centers; external validation would be needed to apply the models to other centers. Second, this was a retrospective study, and some information was lacking; for example, the absolute distance of the tumor edge to the mucous layer and serosal layer may also be important factors, but it was not possible to evaluate this factor in the present study. Hence, a well-designed prospective study would be better for the assessment of inner and outer distances. Third, most of the tumors in this study were smaller than 20 mm, which has been reported to be a risk factor for perforation in gastric ESD in other studies [21,24]; thus, this factor was not regarded as an important parameter in the model. The inclusion of more patients with SMTs >20 mm could possibly enable a better model to be built. Fourth, the I/O ratio was measured in this study, although the distance may be influenced by several factors. For example, if the balloon-assisted method was used, the thickness of the layer seemed to change depending on the degree of pressing against the lesion, and if the tumor was adjacent to the liver or the spleen, their thickness would also be influenced. Additionally, various endoscopes and probes were used in this study, which could possibly influence the results; however, as both the inner distance and outer distance would be influenced, calculating their ratio might attenuate the impact of this issue. Thus, caution is warranted in these situations when using this method. Fifth, the function of EUS images in predicting perforation was evaluated in a preliminary sense, and the AUC values of model 1 and model 2 were only 0.836 and 0.755, respectively. However, both of these values were >0.7, which was not optimal, and models with greater AUC, sensitivity, and specificity are still needed.

In conclusion, EUS was useful in predicting perforation of ER for gastric SMTs originating from the MP, and the I/O ratio was an important factor. In addition, age, tumors located in the gastric fundus, and tumor pathology of GISTs or schwannomas were also possible risk factors. Overall, this study developed two models for predicting perforation, which may serve as a useful tool in clinical practice.