- Email: [email protected]
A multivariate prediction model for high malignancy potential gastric GI stromal tumors before endoscopic resection
Journal Pre-proof A multivariate prediction model for high malignancy potential gastric GI stromal tumors before endoscopic resection Ze Yang, MD, Yuan Gao, MD, Xiaotong Fan, MD, Xin Zhao, MD, Shaohua Zhu, MD, Meng Guo, MD, Zhiguo Liu, MD, Xiaocui Yang, MD, Ying Han, MD PII:
S0016-5107(19)32299-0
DOI:
https://doi.org/10.1016/j.gie.2019.09.032
Reference:
YMGE 11773
To appear in:
Gastrointestinal Endoscopy
Received Date: 14 July 2019 Accepted Date: 21 September 2019
Please cite this article as: Yang Z, Gao Y, Fan X, Zhao X, Zhu S, Guo M, Liu Z, Yang X, Han Y, A multivariate prediction model for high malignancy potential gastric GI stromal tumors before endoscopic resection, Gastrointestinal Endoscopy (2019), doi: https://doi.org/10.1016/j.gie.2019.09.032. This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. Copyright © 2019 by the American Society for Gastrointestinal Endoscopy
A multivariate prediction model for high malignancy potential gastric GI stromal tumors before endoscopic resection Ze Yang MD1, Yuan Gao MD2, Xiaotong Fan MD1, Xin Zhao MD1, Shaohua Zhu MD1, Meng Guo MD1, Zhiguo Liu MD1, Xiaocui Yang MD2, Ying Han MD1 1
Xijing Hospital of Digestive Diseases, Air Force Medical University (Fourth Military
Medical University), Xi’an, China 2
Ankang Central Hospital, Shaanxi, China
Corresponding author: Zhiguo Liu MD and Ying Han MD, Xijing Hospital of Digestive Diseases, Air Force Medical University (Fourth Military Medical University), 127 Changle West Road, Xi’an, Shaanxi 710032, China. Telephone: +86-29-84771535; Fax: +86-29-82539041; Email:[email protected] and [email protected]
Running title: A prediction model for high malignancy potential GIST
Declaration of interest All authors declare no conflicts of interest.
Author contributions Study concept and design: Zhiguo Liu; acquisition of data: Ze Yang, Yuan Gao; analysis and interpretation of data: Ze Yang, Xin Zhao, Shaohua Zhu, Zhiguo Liu; drafting of the manuscript: Ze Yang, Zhiguo Liu; critical revision of the manuscript for important intellectual content: Xiaocui Yang, Meng Guo; statistical analysis: Xiaotong Fan; administrative support and final approval of the article: Ying Han.
Acknowledgments This work was supported in part by the National Natural Science Foundation of China (81572820) and Key Research and Development Program of Shaanxi Province (2018ZDXM-SF-050).
1
Abstract Background and Aims: Endoscopic resection is becoming an option in the management of gastric gastrointestinal stromal tumors (GISTs). Although no consensus has been reached, high malignancy potential GISTs are generally considered to be surgical candidates. However, no systematic preoperative evaluation strategy has yet been developed. The current study was performed to develop a preoperative multivariate model to predict the malignant potential of gastric GISTs. Methods: This study consisted of 2 stages. First, a multivariate prediction model for gastric GISTs smaller than 5 cm was developed using a multivariate logistic regression analysis in a retrospective cohort. Next, the prediction model was validated further in a validation cohort of gastric GISTs. Results: In the developing stage, 275 patients were included. The multivariate analysis demonstrated that independent risk factors for high malignancy potential gastric GISTs smaller than 5 cm were tumor size ≥2 cm (according to cutoff value), an irregular tumor shape and mucosal ulceration (P<0.05). Based on accordant regression coefficients, 3 risk factors were weighted with point values: one point for mucosal ulceration, 2 points for an irregular tumor shape and 3 points for tumor size ≥2 cm. In the validation stage, 186 patients were included. The area under the curve (AUC) of the prediction model was 0.80 (95% confidence interval [CI], 0.73–0.85), which was significantly higher than that of tumor size alone (P=0.034). Conclusions: The independent risk factors for high malignancy potential gastric GISTs smaller than 5 cm were tumor size larger than 2 cm, an irregular tumor shape 2
and mucosal ulceration; these factors could be used to predict malignancy potential of gastric GISTs in a simple combination.
Key words: Gastrointestinal stromal tumors; Preoperative multivariate prediction model; Risk factors; Malignant potential; Endoscopic resection
INTRODUCTION Gastrointestinal stromal tumors (GISTs) are common digestive mesenchymal tissue-derived neoplasms that can appear in any part of the digestive tract, although most are found in the stomach (0.60-0.70) with fewer found in the small intestine, colon, and esophagus.1-2 GISTs usually originate from the muscularis propria. It is necessary to consider immunohistochemical staining for CD117 and DOG-1 to obtain a pathologic diagnosis.3-4 GISTs vary from benign to malignant phenotypes based on their biological behavior, with those with high malignancy potential usually having a poor prognosis.5 According to the current guidelines, surgical resection was the preferred treatment for localized resectable gastric GISTs larger than 2 cm, and follow-up was recommended for those smaller than 2 cm when high-risk EUS features were absent.5 In recent years, endoscopy resection has rapidly emerged as a toll for the management of low malignancy potential GISTs and is generally considered a safe and effective treatment, although long-term outcome studies are still scarce.6-7 Currently, endoscopic resection of high malignancy potential GISTs should still be avoided because it can break the 3
pseudocapsule of the tumor. Therefore, it would be useful to be able to predict which GISTs have high malignancy potential before endoscopic treatment. Tumor size and high-risk EUS features are the 2 most common measures used to predict high malignancy potential GISTs.5 However, it is difficult to achieve a clinical decision as these features are not always consistent. In the current retrospective study, we attempted to establish and validate a simple model to predict high malignancy potential GISTs.
METHODS 1 Study populations This was a retrospective cohort based on consecutive patients with pathologically confirmed GISTs treated in Xijing Hospital and Ankang Central Hospital. The patients were divided into 2 groups. The first group (the developing cohort of the multivariate prediction model) included patients with gastric GISTs smaller than 5 cm treated in Xijing Hospital between 2008 and 2018. The second group (the validation cohort of the multivariate prediction model) included a separate cohort of patients with pathologically confirmed gastric GISTs smaller than 5 cm treated in Xijing Hospital between 2018-2019 and Ankang Central Hospital between 2014 and 2019. The inclusion criteria were as follows: (1) GISTs confirmed by postoperative pathology and immunohistochemistry examinations and (2) primary GIST patients who underwent surgical resection or endoscopic therapy with curative intent. The exclusion criteria were as follows: (1) the presence of metastases at diagnosis; (2) nongastric GISTs; (3) lack of complete clinicopathological data; (4) tumor size larger 4
than 5 cm; and (5) patients in neoadjuvant treatment before resection. The study protocol was approved by the Institutional Review Board of the 2 participating institutions. 2 Data collection Preoperative clinicopathological parameters were retrieved from the hospital electronic record system and included age, sex, tumor size (mainly based on computed tomography (CT) measurements; those small GISTs that could not be detected on CT were estimated by measuring postoperative pathological specimens in vitro), tumor shape (round/oval or irregular) (Fig. 1A), mucosal ulceration of the tumor surface (Fig. 1B), tumor location (cardia/gastric fundus, gastric body/gastric angle/gastric antrum), high-risk EUS features (heterogeneity, hyperechoic foci, or cystic spaces),5, 8-10 and risk groups according to the modified National Institutes of Health (NIH) classification system (Table 1).11 Patients were divided into a very low/low risk group with low malignancy potential and a moderate/high risk group with high malignancy potential. The data were collected by 2 independent endoscopists (Z.Y. and Y.G.) in a blinded fashion, and a third endoscopist (Z.L.) made a judgement when there was conflict. 3 Study design This study included 2 stages: the first stage was performed to establish the multivariate prediction model used to predict the malignant potential of gastric GISTs. We investigated the risk factors for high malignancy potential gastric GISTs to establish the multivariate prediction model for the developing cohort as described in 5
the study populations section. To develop the model, the potential risk factors for high malignancy potential GISTs reported in previous studies were evaluated in this study and included sex, age, tumor location, tumor size, tumor shape, high-risk ultrasonography features and mucosal ulceration.8-10, 12-15 In the second stage, the model was further validated with the validation cohort, and the AUC was compared with that of tumor size. 4 Statistical analysis The sample size of the developing cohort was estimated according to the number of independent variables included in the multivariate logistic regression.16 In our study of the developing cohort, seven independent variables were included in the multivariate logistic regression. Thus, the sample size of the developing cohort was estimated to be at least 140 cases. The sample size of the validation cohort was calculated for a diagnostic test of the multivariate prediction model for malignant potential of gastric GISTs using a 2-sided binomial test.17 Based on the developing cohort, the incidence rates of high malignancy potential cases in gastric GISTs, the sensitivity and the specificity were assumed to be 0.23, 0.59 and 0.81, respectively. We assumed a type I error of 5% and a power of 90%. The sample size estimation of the validation cohort was therefore at least 32 cases. Quantitative variables were presented as medians with interquartile ranges (IQRs) unless otherwise stated and were analyzed with the nonparametric Mann-Whitney U test. The chi-square test or the Fisher exact test was used for the univariate analysis, and logistic regression was used for the multivariate analysis. 6
In the first stage, univariate and multivariate logistic regression analyses were performed in the developing cohort. To establish the model, we assigned weighted points proportional to the β regression coefficient value for the factors determined in the multivariate analysis. The AUC value of the model might be slightly improved if it includes variables without significant differences. However, given our purpose of developing a simple-to-use prediction model, we selected 3 variables with significant differences (P<0.05) for the final model. The regression coefficients of the 3 independent risk factors were scored by linear transformation, and all regression coefficients were divided by 0.6. The purpose of dividing by 0.6 was to thoroughly discriminate the original coefficients so that their differences were reflected as much as possible. The obtained values were rounded to integers. A model score was then calculated for each patient and assessed by receiver operating characteristic (ROC) curve. The patients were divided into 2 groups, a low-grade risk group and a high-grade risk group, according to the cutoff value of the model score (the largest Youden index). Tumors size, which was categorized at 1-cm intervals and the factors identified in the multivariate prediction model were used as the test variables; the modified NIH classification system was used as the state variable to plot the ROC curve. The Delong method was used to compare the 2 AUCs. In the second stage, the multivariate prediction model was applied to the validation cohort. Patients were classified as low-grade risk or high-grade risk group according to the model we established. The incidence rates of high malignancy potential gastric GISTs in the 2 groups were evaluated using the chi-square test. 7
Moreover, calibration of the model was evaluated using the Hosmer-Lemeshow goodness-of-fit test. The data were analyzed using SPSS for Windows software (version 24.0), MedCalc statistical software (version 18.2.1), and PASS software (version 19.0.2). A 2-sided P value of <0.05 was considered statistically significant.
Results Baseline characteristics of the developing and validation cohorts The baseline characteristics of the developing and validation cohorts are shown in Table 2. According to the study criteria, a total of 275 and 186 patients with GISTs were enrolled in the developing and validation cohorts, respectively (Fig. 2). In the developing cohort, there were 211 patients (0.77) who were categorized as having low malignancy potential and 64 patients (0.23) who were categorized as having high malignancy potential; in the validation cohort, there were 152 patients (0.82) and 34 patients (0.18), respectively, in the corresponding groups. There were no significant differences in the distributions of low and high malignancy potential GISTs between the 2 cohorts (P =0.199). Establishing a multivariate prediction model for predicting the malignant potential of gastric GISTs In the developing cohort, the actual tumor size was used as the test variable, and the modified NIH classification system was used as the state variable to plot the ROC curve; the optimal cutoff value was calculated as 2 cm (Fig. 3).
8
The univariate analysis indicated that there were no significant differences between the high and low malignancy potential groups with regard for sex, age, lesion location or high-risk ultrasonography features (P>0.05), but there were significant differences when patients were stratified by tumor size, tumor shape, and mucosal ulceration (P<0.05, Table 3). The multivariate logistic regression analysis indicated that a tumor size larger than 2 cm, an irregular tumor shape, and mucosal ulceration were independent risk factors (P<0.05, Table 4). The regression coefficients of the 3 independent risk factors were scored by linear transformation. Therefore, mucosal ulceration, an irregular tumor shape, and a tumor size larger than 2 cm were given scores of 1, 2, and 3, respectively (Table 4). The multivariate prediction model was used as the test variable and the modified NIH classification system was used as the state variable to plot the ROC curve. The AUC value was 0.79 (95% CI, 0.74–0.84). In contrast, when tumor size, which was categorized at 1-cm intervals, was used as the test variable, the AUC value was 0.73 (95% CI, 0.67–0.78), which was significantly lower than the value achieved using the multivariate prediction model as the test variable (P=0.008, Fig. 4). The model score was assessed by ROC curve, and the optimal cutoff value was 4 (Fig. 5). According to this cutoff value, patients with scores of 0 to 3 were categorized into the low-grade risk group, and patients with scores of 4 to 6 were categorized into the high-grade risk group. In addition, this model was well-calibrated with the Hosmer-Lemeshow goodness-of-fit test (χ 2 =0.88, df=4, P =0.928), indicating that there was no significant difference between the predicted value of the model and the 9
actually observed value. The prediction model showed good calibration. The incidence rate of high malignancy potential GISTs was 0.13 in the low-grade risk group and 0.48 in the high-grade risk group. There was a significant difference between the 2 groups according to the chi-square test (χ 2 =38.27 and P<0.001). The corresponding sensitivity, positive predictive value, specificity, and negative predictive value were 0.59, 0.48, 0.81 and 0.87, respectively. Validation of the multivariate prediction model for predicting the malignant potential of gastric GISTs In the validation stage, the multivariate prediction model was used as the test variable, and the modified NIH classification system was used as the state variable to plot the ROC curve. The AUC value was 0.80 (95% CI, 0.73–0.85). As a control, when tumor size was used as the test variable, the AUC value was 0.73 (95% CI, 0.67–0.80), which was significantly lower than that achieved when the model was used as the test variable (P=0.034, Fig. 6). The incidence rate of high malignancy potential GISTs was 0.10 in the low-grade risk group (0-3 points) and 0.42 in the high-grade risk group (4-6 points) (χ 2 =23.69 and P<0.001); the OR was 6.33 (95% CI, 2.85 -14.03). Based on the aforementioned cutoff value, the sensitivity, positive predictive value, specificity and negative predictive value were 0.59, 0.42, 0.82 and 0.90, respectively. In addition, this model was well-calibrated with the Hosmer-Lemeshow goodness-of-fit test (χ 2 =1.09, df=3, P =0.779).
DISCUSSION 10
GISTs smaller than 2 cm in diameter are referred to as small GISTs. According to the modified NIH classification system, gastric GISTs smaller than 2 cm and with <5 mitoses/50 high-power fields (HPFs) were considered very low risk.11 Thus, surgical resection was the treatment of choice for primary, localized gastric GISTs larger than 2 cm, whereas conservative follow-up was suggested for lesions smaller than 2 cm.5 However, a small portion of small gastric GISTs have also been reported to have malignant potential.18 In addition, small gastric GISTs have also been reported to undergo early metastasis to the liver.19 Thus, tumor size alone might not reliably evaluate the malignant potential of gastric GISTs. In the current study, we established a novel, simple-to-use preoperative multivariate prediction model for gastric GISTs smaller than 5 cm. This model efficiently predicted high malignancy potential gastric GISTs based on easily obtained clinical features. High-risk ultrasonography features were not found to be independent predictive factors in our multivariate analysis, although they are recommended by current guidelines to predict the malignant potential of gastric GISTs.5 EUS was previously shown to be valuable in preoperatively predicting the malignant potential of gastric GISTs.8-10, 13-14, 20 However, the criteria for identifying high-risk ultrasonography features largely rely on the expertise of endoscopists and have therefore remained subjective. There is a lack of objective executive criteria for the identification of hyperechoic foci, heterogeneity, and cystic spaces during the evaluation of EUS. There was no well-accepted consensus regarding which EUS features were the most 11
highly correlated with the malignancy of GISTs.21 We considered endoscopist variation might influence the significance of high-risk features in clinical practice. Another possible reason might be that previously identified high-risk features were obtained mainly from large-size GISTs, and those results could be different from those obtained in the cohort we included in this study.9-10, 13 However, we did not analyze the significance of single high-risk feature one by one due to limitations associated with the sample size, which could have led to false-negative evaluations of these features. EUS-guided fine-needle aspiration biopsy (EUS-FNA) is an excellent way to obtain tissue for preoperative analysis of GIST. However, the gastric GISTs in our study were relatively small, making it technically difficult to perform a standard EUS-FNA procedure. The specimen sample volume obtained by EUS-FNA was usually small; therefore, assessment of the mitotic index was difficult.21 Moreover, the diagnostic rate of EUS-FNA tended to decrease as the tumor diameter decreased,22-23 hence, EUS-FNA was recommended for relatively large tumors.24 To our knowledge, the current study is the first to reveal that an irregular tumor shape is an independent risk factor for high malignancy potential gastric GISTs. In fact, this factor has not previously been identified as an independent risk factor even though it has been included in previous studies.8, 14, 25 One possible reason might lie in the fact that previous studies had usually included a quite limited number of patients, especially those with GISTs less than 5 cm. In some studies, only factors with a P value < .05 in the univariate analysis were introduced into the multivariate 12
analysis.25-26 P values were influenced by the sample size and became relatively larger as the sample size decreased. Therefore, some factors could be missed if they were adopted or abandoned based only on the P value.27 Biologically diversity in the rapid growth rate of the internal structure could contribute to the formation of an irregular tumor shape, but the detailed mechanisms require further study. It has been suggested that ulceration is associated with high malignancy potential gastric GISTs,13-14, 25 in accordance with our research results. The National Comprehensive Cancer Network (NCCN) Guidelines also recommended that endoscopists should take ulceration as a risk factor for high malignancy potential gastric GISTs.5 There are several potential advantages related to our new 6-point multivariate prediction model in a clinical setting. First, this prediction model could be used to predict the malignant potential of gastric GISTs before resection. At present, various guidelines call for the malignant potential of gastric GISTs to be evaluated based on postoperative pathological calculation of the mitotic index, which can be acquired only after tumor resection.5, 11, 28 Our prediction model provides support for making appropriate treatment decisions for patients. Ideally, patients with low malignancy potential gastric GISTs can be treated safely endoscopically, whereas those with high malignancy potential gastric GISTs could choose a more-aggressive surgical treatment. Second, the predictive accuracy of the model was better than that achieved when using tumor size alone as the test variable. In the validation cohort, the model achieved satisfactory predictive accuracy when compared with tumor size alone. 13
Finally, the 3 factors that were included in the multivariate prediction model were simple and easy to obtain in clinical practice. There are also a few limitations to this study. First, in this study, we diagnosed the malignant potential of gastric GISTs based on the modified NIH classification system,11 which is a less-optimal parameter than patient outcomes. Low-risk gastric GISTs evaluated using the modified NIH classification system were reported to recur locally or to metastasize after R0 resection.29 Further study is needed to confirm the relationship between the model and patient outcomes. Second, this study was retrospective in nature and therefore had inherent potential for bias. A large-scale prospective randomized controlled trial is required for verification of this multivariate prediction model. Third, the events per variable value observed in this study were slightly low for a proper logistic regression analysis. In this study, the number of estimated variables in the logistic regression analysis was 7, whereas that of high malignancy potential gastric GISTs was 64, resulting in an event per variable value of 9.1, which was slightly lower than the preferable value of ≥10.30 Thus, a type II error might have occurred in this analysis. In summary, in this study, we demonstrate that a tumor size larger than 2 cm, mucosal ulceration and an irregular tumor shape are independent risk factors for patients with high malignancy potential gastric GISTs smaller than 5 cm. Moreover, these 3 factors can be applied to establish a simple-to-use preoperative predictive model that could help clinicians predict the malignant potential of gastric GISTs smaller than 5 cm. 14
REFERENCES 1. Soreide K, Sandvik OM, Soreide JA, et al. Global epidemiology of gastrointestinal stromal tumours (GIST): A systematic review of population-based cohort studies. Cancer Epidemiol 2016;40:39-46. 2. Chandrasekhara V, Ginsberg GG. Endoscopic management of gastrointestinal stromal tumors. Curr Gastroenterol Rep 2011;13:532-9. 3. Sarlomo-Rikala M, Kovatich AJ, Barusevicius A, et al. CD117: a sensitive marker for gastrointestinal stromal tumors that is more specific than CD34. Mod Pathol 1998;11:728-34. 4. West RB, Corless CL, Chen X, et al. The novel marker, DOG1, is expressed ubiquitously in gastrointestinal stromal tumors irrespective of KIT or PDGFRA mutation status. Am J Pathol 2004;165:107-13. 5. von Mehren M, Randall RL, Benjamin RS, et al. Soft Tissue Sarcoma, Version 2.2018, NCCN Clinical Practice Guidelines in Oncology. J Natl Compr Canc Netw 2018;16:536-63. 6. Meng Y, Li W, Han L, et al. Long-term outcomes of endoscopic submucosal dissection versus laparoscopic resection for gastric stromal tumors less than 2 cm. J Gastroenterol Hepatol 2017;32:1693-7. 7. Feng F, Liu Z, Zhang X, et al. Comparison of Endoscopic and Open Resection for Small Gastric Gastrointestinal Stromal Tumor. Transl Oncol 2015;8:504-8. 8. Chen T, Xu L, Dong X, et al. The roles of CT and EUS in the preoperative evaluation of gastric gastrointestinal stromal tumors larger than 2 cm. Eur Radiol 15
2019;29:2481-9. 9. Chak A, Canto MI, Rosch T, et al. Endosonographic differentiation of benign and malignant stromal cell tumors. Gastrointest Endosc 1997;45:468-73. 10. Shah P, Gao F, Edmundowicz SA, et al. Predicting malignant potential of gastrointestinal stromal tumors using endoscopic ultrasound. Dig Dis Sci 2009;54:1265-9. 11. Joensuu H. Risk stratification of patients diagnosed with gastrointestinal stromal tumor. Hum Pathol 2008;39:1411-9. 12. Kim MY, Park YS, Choi KD, et al. Predictors of recurrence after resection of small gastric gastrointestinal stromal tumors of 5 cm or less. J Clin Gastroenterol 2012;46:130-7. 13. Chen TH, Hsu CM, Chu YY, et al. Association of endoscopic ultrasonographic parameters and gastrointestinal stromal tumors (GISTs): can endoscopic ultrasonography be used to screen gastric
GISTs for potential malignancy?
Scand J Gastroenterol 2016;51:374-7. 14. Jeon SW, Park YD, Chung YJ, et al. Gastrointestinal stromal tumors of the stomach: endosonographic differentiation in relation to histological risk. J Gastroenterol Hepatol 2007;22:2069-75. 15. Joensuu H, Vehtari A, Riihimaki J, et al. Risk of recurrence of gastrointestinal stromal tumour after surgery: an analysis
of pooled population-based cohorts.
Lancet Oncol 2012;13:265-74. 16. Sun ZQ, Xu YY, Liang WJ, et al. Medical Statistics, 4th edition. Beijing: People’s 16
Medical Publishing House 2014; pp. 295. 17. Li J, Fine J. On sample size for sensitivity and specificity in prospective diagnostic accuracy studies. Stat Med 2004;23:2537-50. 18. Yang J, Feng F, Li M, et al. Surgical resection should be taken into consideration for the treatment of small gastric gastrointestinal stromal tumors. World J Surg Oncol 2013;11:273. 19. Tanaka J, Oshima T, Hori K, et al. Small gastrointestinal stromal tumor of the stomach showing rapid growth and early metastasis to the liver. Dig Endosc 2010;22:354-6. 20. Palazzo L, Landi B, Cellier C, et al. Endosonographic features predictive of benign and malignant gastrointestinal stromal cell tumours. Gut 2000;46:88-92. 21. Faulx AL, Kothari S, Acosta RD, et al. The role of endoscopy in subepithelial lesions of the GI tract. Gastrointest Endosc 2017;85:1117-32. 22. Attila T, Aydin O. Lesion size determines diagnostic yield of EUS-FNA with onsite cytopathologic evaluation for upper gastrointestinal subepithelial lesions. Turk J Gastroenterol 2018;29:436-41. 23. Akahoshi K, Sumida Y, Matsui N, et al. Preoperative diagnosis of gastrointestinal stromal tumor by endoscopic ultrasound-guided fine needle aspiration. World J Gastroenterol 2007;13:2077-82. 24. Akahoshi K, Oya M, Koga T, et al. Clinical usefulness of endoscopic ultrasound-guided fine needle aspiration for gastric subepithelial lesions smaller than 2 cm. J Gastrointestin Liver Dis 2014;23:405-12. 17
25. Lv GR, Zhou YH, Zhong JW, et al. Comparison of endoscopic ultrasonography image characteristics and analysis of risk factors for different invasive risk of gastric stromal tumors. China Journal of Endoscopy 2016;1:1-4. 26. Zhou C, Duan X, Zhang X, et al. Predictive features of CT for risk stratifications in
patients
with
primary gastrointestinal
stromal
tumour.
Eur Radiol
2016;26:3086-93. 27. Chen CZ, Chen XL, Wei C. Epidemiological Data Analysis and EmpowerStats Software Application, 1th edition. Shanghai: Shanghai Scientific & Technical Publishers 2016; pp. 19. 28. Fletcher CDM, Bridge JA, Hogendoorn PCW, et al. WHO Classfication of Tumours of Soft Tissue and Bone. 4th edition. Lyon: IARC Press 2013; pp. 164-7. 29. Lee S, Kim YN, Son T, et al. Oncologic Safety of Laparoscopic Wedge Resection with Gastrotomy for Gastric Gastrointestinal Stromal Tumor: Comparison with Conventional Laparoscopic Wedge Resection. J Gastric Cancer 2015;15:231-7. 30. Peduzzi P, Concato J, Kemper E, et al. A simulation study of the number of events per variable in logistic regression analysis. J Clin Epidemiol 1996;49:1373-9.
18
Table 1. Modified NIH classification system Risk category Tumor size (cm) Mitotic index (per 50 HPFs) Very-low risk ＜2.0 ≤5 Low risk 2.1-5.0 ≤5 Intermediate risk 2.1-5.0 ＞5 ＜5.0 6-10 5.1-10.0 ≤5 High risk Any Any ＞10.0 Any Any ＞10 ＞5.0 ＞5 2.1-5.0 ＞5 ≤5 5.1-10.0 NIH, National Institutes of Health.
Primary tumor site Any Any Gastric Any Gastric Tumor rupture Any Any Any Nongastric Nongastric
19
Table 2. Baseline characteristics of the developing (n=275) and validation (n=186) cohorts. Developing cohort Validation cohort P value (n=275) (n=186) a 57 (50-64） 56 (50-63) 0.469 Age, y, median (IQR) 0.895 Sex, n (%) Male 121 (44.0） 83 (44.6) Female 154 (56.0） 103 (55.4) 0.308 Location, n (%) Cardia/Gastric fundus 166 (60.4） 121 (65.1) Gastric body/angle/antrum 109 (39.6） 65 (34.9) 2.5 (1.3-3.6) 0.896 2.4 (1.2-3.5） Tumor size (cm), median (IQR) 0.070 Tumor shape, n (%) Round/oval 215 (78.2） 158 (84.9) irregular 60 (21.8) 28 (15.1) 0.676 Mucosal ulceration, n (%) Presence 49 (17.8) 36 (19.4) Absence 226 (82.2) 150 (80.6) High-risk ultrasonography b features , n(%) Presence 150 (54.5) Absence 125 (45.5) a. IQR, interquartile range. b. High-risk ultrasonography features: heterogeneity, hyperechoic foci, or cystic spaces.
20
Table 3. Univariate analysis of the risk factors for high malignancy potential gastric GISTs smaller than 5 cm (n [%]) in developing cohort. Factor Low malignancy High malignancy P value a a potential group potential group (n=64) (n=211) 0.000 Tumor size (cm)b <2 95 (45.0) 5 (7.8) 2-5 116 (55.0) 59 (92.2) 0.535 Sex Male 95 (45.0) 26 (40.6) Female 116 (55.0) 38 (59.4) 0.533 Age (years) ≤60 126 (59.7) 41 (64.1) >60 85 (40.3) 23 (35.9) 0.000 Mucosal ulceration Absence 187 (88.6) 39 (60.9) Presence 24 (11.4) 25 (39.1) 0.376 High-risk ultrasonography c features Absence 99 (46.9) 26 (40.6) Presence 112 (53.1) 38 (59.4) 0.000 Tumor shape Round/oval 182 (86.3) 33 (51.6) Irregular 29 (13.7) 31 (48.4) 0.490 Location Cardia/gastric fundus 125 (59.2) 41 (64.1) Gastric body/angle/antrum 86 (40.7) 23 (35.9) a. Low malignancy potential group, very-low-/low-risk groups based on the modified NIH classification system; high malignancy potential group, moderate-/high-risk groups based on the modified NIH classification system.11 b. Mainly based on CT measurements (those small GISTs that could not be detected in CT were estimated by measuring postoperative pathological specimen in vitro). c. High-risk ultrasonography features: heterogeneity, hyperechoic foci, or cystic spaces.
21
Table 4. Multivariate logistic regression analysis of the risk factors for high malignancy potential gastric GISTs smaller than 5 cm in developing cohort and the prediction modela. Factor Standard Wald P Odds ratio (95% CI) β error value Tumor size (cm)b <2 1 (reference) 2-5 0.520 12.888 0.000 6.474 (2.335-17.948) 1.868 Sex Male 1 (reference) Female 0.338 1.600 0.206 1.534 (0.791-2.976) 0.428 Age (years) ≤60 1 (reference) >60 0.340 0.457 0.499 0.794 (0.408-1.548) -0.230 Mucosal ulceration Absence 1 (reference) Presence 0.376 5.479 0.019 2.412 (1.154-5.041) 0.880 High-risk ultrasonography featuresc Absence 1 (reference) Presence 0.342 0.680 0.410 0.754 (0.386-1.474) -0.282 Tumor shape Round/oval 1 (reference) Irregular 0.362 12.227 0.000 3.548 (1.745 -7.216) 1.266 Location Cardia/Gastric fundus 1 (reference) Gastric body/angle/antrum 0.338 2.514 0.113 0.585 (0.302-1.135) -0.536 a. Low malignancy potential group, very-low risk/low-risk groups based on the modified NIH classification system; high malignancy potential group, moderate-/high-risk groups based on the modified NIH classification system.11 b. Mainly based on CT measurements (those small GISTs that could not be detected in CT were estimated by measuring postoperative pathological specimen in vitro). c. High-risk ultrasonography features: heterogeneity, hyperechoic foci, or cystic spaces. d. The regression coefficients of the 3 independent risk factors were scored by linear transformation, and all regression coefficients were divided by 0.6. The purpose of dividing by 0.6 was to thoroughly discriminate the original coefficients so that their difference was reflected as much as possible. The obtained values were rounded to integers.
22
Pointsd
3
-
-
1
-
2
-
FIGURE LEGENDS Figure 1. Endoscopic images of high malignancy potential gastric GISTs. A, Irregular tumor shape. B, Mucosal ulceration. Figure 2. Flowchart of patient enrollment. GISTs, gastrointestinal stromal tumors. Figure 3. ROC curves of actual tumor size in the developing cohort. The actual tumor size was used as the test variable, whereas the modified NIH classification system was used as the state variable to plot the ROC curve (very-low risk/low-risk groups as the low malignancy potential group and medium-/high-risk groups as the high malignancy potential group). Figure 4. ROC curves for tumor size categorized at 1-cm intervals and the prediction model for the developing cohort. Tumor size and the multivariate prediction model were used as the test variables, whereas the modified NIH classification system was used as the state variable to plot the ROC curve. Figure 5. ROC curves of the model score in the developing cohort. The model score was used as the test variable, whereas the modified NIH classification system was used as the state variable to plot the ROC curve. Figure 6. ROC curves of tumor size categorized at 1-cm intervals and the prediction model in the validation cohort. Tumor size and the multivariate prediction model were used as the test variables, whereas the modified NIH classification system was used as the state variable to plot the ROC curve.
23
Abbreviations CI, confidence interval; ESD, endoscopic submucosal dissection; OR, odds ratio; ER, endoscopic resection; GISTs, gastrointestinal stromal tumors; CT, computed tomography; NIH, National Institutes of Health; ROC, receiver operating characteristic; AUC, area under the curve; HPF, high-power field; EUS, Endoscopic ultrasonography; FNA, fine needle aspiration biopsy; IQR, interquartile range; NCCN, National Comprehensive Cancer Network.
1
S0016-5107(19)32299-0
DOI:
https://doi.org/10.1016/j.gie.2019.09.032
Reference:
YMGE 11773
To appear in:
Gastrointestinal Endoscopy
Received Date: 14 July 2019 Accepted Date: 21 September 2019
Please cite this article as: Yang Z, Gao Y, Fan X, Zhao X, Zhu S, Guo M, Liu Z, Yang X, Han Y, A multivariate prediction model for high malignancy potential gastric GI stromal tumors before endoscopic resection, Gastrointestinal Endoscopy (2019), doi: https://doi.org/10.1016/j.gie.2019.09.032. This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. Copyright © 2019 by the American Society for Gastrointestinal Endoscopy
A multivariate prediction model for high malignancy potential gastric GI stromal tumors before endoscopic resection Ze Yang MD1, Yuan Gao MD2, Xiaotong Fan MD1, Xin Zhao MD1, Shaohua Zhu MD1, Meng Guo MD1, Zhiguo Liu MD1, Xiaocui Yang MD2, Ying Han MD1 1
Xijing Hospital of Digestive Diseases, Air Force Medical University (Fourth Military
Medical University), Xi’an, China 2
Ankang Central Hospital, Shaanxi, China
Corresponding author: Zhiguo Liu MD and Ying Han MD, Xijing Hospital of Digestive Diseases, Air Force Medical University (Fourth Military Medical University), 127 Changle West Road, Xi’an, Shaanxi 710032, China. Telephone: +86-29-84771535; Fax: +86-29-82539041; Email:[email protected] and [email protected]
Running title: A prediction model for high malignancy potential GIST
Declaration of interest All authors declare no conflicts of interest.
Author contributions Study concept and design: Zhiguo Liu; acquisition of data: Ze Yang, Yuan Gao; analysis and interpretation of data: Ze Yang, Xin Zhao, Shaohua Zhu, Zhiguo Liu; drafting of the manuscript: Ze Yang, Zhiguo Liu; critical revision of the manuscript for important intellectual content: Xiaocui Yang, Meng Guo; statistical analysis: Xiaotong Fan; administrative support and final approval of the article: Ying Han.
Acknowledgments This work was supported in part by the National Natural Science Foundation of China (81572820) and Key Research and Development Program of Shaanxi Province (2018ZDXM-SF-050).
1
Abstract Background and Aims: Endoscopic resection is becoming an option in the management of gastric gastrointestinal stromal tumors (GISTs). Although no consensus has been reached, high malignancy potential GISTs are generally considered to be surgical candidates. However, no systematic preoperative evaluation strategy has yet been developed. The current study was performed to develop a preoperative multivariate model to predict the malignant potential of gastric GISTs. Methods: This study consisted of 2 stages. First, a multivariate prediction model for gastric GISTs smaller than 5 cm was developed using a multivariate logistic regression analysis in a retrospective cohort. Next, the prediction model was validated further in a validation cohort of gastric GISTs. Results: In the developing stage, 275 patients were included. The multivariate analysis demonstrated that independent risk factors for high malignancy potential gastric GISTs smaller than 5 cm were tumor size ≥2 cm (according to cutoff value), an irregular tumor shape and mucosal ulceration (P<0.05). Based on accordant regression coefficients, 3 risk factors were weighted with point values: one point for mucosal ulceration, 2 points for an irregular tumor shape and 3 points for tumor size ≥2 cm. In the validation stage, 186 patients were included. The area under the curve (AUC) of the prediction model was 0.80 (95% confidence interval [CI], 0.73–0.85), which was significantly higher than that of tumor size alone (P=0.034). Conclusions: The independent risk factors for high malignancy potential gastric GISTs smaller than 5 cm were tumor size larger than 2 cm, an irregular tumor shape 2
and mucosal ulceration; these factors could be used to predict malignancy potential of gastric GISTs in a simple combination.
Key words: Gastrointestinal stromal tumors; Preoperative multivariate prediction model; Risk factors; Malignant potential; Endoscopic resection
INTRODUCTION Gastrointestinal stromal tumors (GISTs) are common digestive mesenchymal tissue-derived neoplasms that can appear in any part of the digestive tract, although most are found in the stomach (0.60-0.70) with fewer found in the small intestine, colon, and esophagus.1-2 GISTs usually originate from the muscularis propria. It is necessary to consider immunohistochemical staining for CD117 and DOG-1 to obtain a pathologic diagnosis.3-4 GISTs vary from benign to malignant phenotypes based on their biological behavior, with those with high malignancy potential usually having a poor prognosis.5 According to the current guidelines, surgical resection was the preferred treatment for localized resectable gastric GISTs larger than 2 cm, and follow-up was recommended for those smaller than 2 cm when high-risk EUS features were absent.5 In recent years, endoscopy resection has rapidly emerged as a toll for the management of low malignancy potential GISTs and is generally considered a safe and effective treatment, although long-term outcome studies are still scarce.6-7 Currently, endoscopic resection of high malignancy potential GISTs should still be avoided because it can break the 3
pseudocapsule of the tumor. Therefore, it would be useful to be able to predict which GISTs have high malignancy potential before endoscopic treatment. Tumor size and high-risk EUS features are the 2 most common measures used to predict high malignancy potential GISTs.5 However, it is difficult to achieve a clinical decision as these features are not always consistent. In the current retrospective study, we attempted to establish and validate a simple model to predict high malignancy potential GISTs.
METHODS 1 Study populations This was a retrospective cohort based on consecutive patients with pathologically confirmed GISTs treated in Xijing Hospital and Ankang Central Hospital. The patients were divided into 2 groups. The first group (the developing cohort of the multivariate prediction model) included patients with gastric GISTs smaller than 5 cm treated in Xijing Hospital between 2008 and 2018. The second group (the validation cohort of the multivariate prediction model) included a separate cohort of patients with pathologically confirmed gastric GISTs smaller than 5 cm treated in Xijing Hospital between 2018-2019 and Ankang Central Hospital between 2014 and 2019. The inclusion criteria were as follows: (1) GISTs confirmed by postoperative pathology and immunohistochemistry examinations and (2) primary GIST patients who underwent surgical resection or endoscopic therapy with curative intent. The exclusion criteria were as follows: (1) the presence of metastases at diagnosis; (2) nongastric GISTs; (3) lack of complete clinicopathological data; (4) tumor size larger 4
than 5 cm; and (5) patients in neoadjuvant treatment before resection. The study protocol was approved by the Institutional Review Board of the 2 participating institutions. 2 Data collection Preoperative clinicopathological parameters were retrieved from the hospital electronic record system and included age, sex, tumor size (mainly based on computed tomography (CT) measurements; those small GISTs that could not be detected on CT were estimated by measuring postoperative pathological specimens in vitro), tumor shape (round/oval or irregular) (Fig. 1A), mucosal ulceration of the tumor surface (Fig. 1B), tumor location (cardia/gastric fundus, gastric body/gastric angle/gastric antrum), high-risk EUS features (heterogeneity, hyperechoic foci, or cystic spaces),5, 8-10 and risk groups according to the modified National Institutes of Health (NIH) classification system (Table 1).11 Patients were divided into a very low/low risk group with low malignancy potential and a moderate/high risk group with high malignancy potential. The data were collected by 2 independent endoscopists (Z.Y. and Y.G.) in a blinded fashion, and a third endoscopist (Z.L.) made a judgement when there was conflict. 3 Study design This study included 2 stages: the first stage was performed to establish the multivariate prediction model used to predict the malignant potential of gastric GISTs. We investigated the risk factors for high malignancy potential gastric GISTs to establish the multivariate prediction model for the developing cohort as described in 5
the study populations section. To develop the model, the potential risk factors for high malignancy potential GISTs reported in previous studies were evaluated in this study and included sex, age, tumor location, tumor size, tumor shape, high-risk ultrasonography features and mucosal ulceration.8-10, 12-15 In the second stage, the model was further validated with the validation cohort, and the AUC was compared with that of tumor size. 4 Statistical analysis The sample size of the developing cohort was estimated according to the number of independent variables included in the multivariate logistic regression.16 In our study of the developing cohort, seven independent variables were included in the multivariate logistic regression. Thus, the sample size of the developing cohort was estimated to be at least 140 cases. The sample size of the validation cohort was calculated for a diagnostic test of the multivariate prediction model for malignant potential of gastric GISTs using a 2-sided binomial test.17 Based on the developing cohort, the incidence rates of high malignancy potential cases in gastric GISTs, the sensitivity and the specificity were assumed to be 0.23, 0.59 and 0.81, respectively. We assumed a type I error of 5% and a power of 90%. The sample size estimation of the validation cohort was therefore at least 32 cases. Quantitative variables were presented as medians with interquartile ranges (IQRs) unless otherwise stated and were analyzed with the nonparametric Mann-Whitney U test. The chi-square test or the Fisher exact test was used for the univariate analysis, and logistic regression was used for the multivariate analysis. 6
In the first stage, univariate and multivariate logistic regression analyses were performed in the developing cohort. To establish the model, we assigned weighted points proportional to the β regression coefficient value for the factors determined in the multivariate analysis. The AUC value of the model might be slightly improved if it includes variables without significant differences. However, given our purpose of developing a simple-to-use prediction model, we selected 3 variables with significant differences (P<0.05) for the final model. The regression coefficients of the 3 independent risk factors were scored by linear transformation, and all regression coefficients were divided by 0.6. The purpose of dividing by 0.6 was to thoroughly discriminate the original coefficients so that their differences were reflected as much as possible. The obtained values were rounded to integers. A model score was then calculated for each patient and assessed by receiver operating characteristic (ROC) curve. The patients were divided into 2 groups, a low-grade risk group and a high-grade risk group, according to the cutoff value of the model score (the largest Youden index). Tumors size, which was categorized at 1-cm intervals and the factors identified in the multivariate prediction model were used as the test variables; the modified NIH classification system was used as the state variable to plot the ROC curve. The Delong method was used to compare the 2 AUCs. In the second stage, the multivariate prediction model was applied to the validation cohort. Patients were classified as low-grade risk or high-grade risk group according to the model we established. The incidence rates of high malignancy potential gastric GISTs in the 2 groups were evaluated using the chi-square test. 7
Moreover, calibration of the model was evaluated using the Hosmer-Lemeshow goodness-of-fit test. The data were analyzed using SPSS for Windows software (version 24.0), MedCalc statistical software (version 18.2.1), and PASS software (version 19.0.2). A 2-sided P value of <0.05 was considered statistically significant.
Results Baseline characteristics of the developing and validation cohorts The baseline characteristics of the developing and validation cohorts are shown in Table 2. According to the study criteria, a total of 275 and 186 patients with GISTs were enrolled in the developing and validation cohorts, respectively (Fig. 2). In the developing cohort, there were 211 patients (0.77) who were categorized as having low malignancy potential and 64 patients (0.23) who were categorized as having high malignancy potential; in the validation cohort, there were 152 patients (0.82) and 34 patients (0.18), respectively, in the corresponding groups. There were no significant differences in the distributions of low and high malignancy potential GISTs between the 2 cohorts (P =0.199). Establishing a multivariate prediction model for predicting the malignant potential of gastric GISTs In the developing cohort, the actual tumor size was used as the test variable, and the modified NIH classification system was used as the state variable to plot the ROC curve; the optimal cutoff value was calculated as 2 cm (Fig. 3).
8
The univariate analysis indicated that there were no significant differences between the high and low malignancy potential groups with regard for sex, age, lesion location or high-risk ultrasonography features (P>0.05), but there were significant differences when patients were stratified by tumor size, tumor shape, and mucosal ulceration (P<0.05, Table 3). The multivariate logistic regression analysis indicated that a tumor size larger than 2 cm, an irregular tumor shape, and mucosal ulceration were independent risk factors (P<0.05, Table 4). The regression coefficients of the 3 independent risk factors were scored by linear transformation. Therefore, mucosal ulceration, an irregular tumor shape, and a tumor size larger than 2 cm were given scores of 1, 2, and 3, respectively (Table 4). The multivariate prediction model was used as the test variable and the modified NIH classification system was used as the state variable to plot the ROC curve. The AUC value was 0.79 (95% CI, 0.74–0.84). In contrast, when tumor size, which was categorized at 1-cm intervals, was used as the test variable, the AUC value was 0.73 (95% CI, 0.67–0.78), which was significantly lower than the value achieved using the multivariate prediction model as the test variable (P=0.008, Fig. 4). The model score was assessed by ROC curve, and the optimal cutoff value was 4 (Fig. 5). According to this cutoff value, patients with scores of 0 to 3 were categorized into the low-grade risk group, and patients with scores of 4 to 6 were categorized into the high-grade risk group. In addition, this model was well-calibrated with the Hosmer-Lemeshow goodness-of-fit test (χ 2 =0.88, df=4, P =0.928), indicating that there was no significant difference between the predicted value of the model and the 9
actually observed value. The prediction model showed good calibration. The incidence rate of high malignancy potential GISTs was 0.13 in the low-grade risk group and 0.48 in the high-grade risk group. There was a significant difference between the 2 groups according to the chi-square test (χ 2 =38.27 and P<0.001). The corresponding sensitivity, positive predictive value, specificity, and negative predictive value were 0.59, 0.48, 0.81 and 0.87, respectively. Validation of the multivariate prediction model for predicting the malignant potential of gastric GISTs In the validation stage, the multivariate prediction model was used as the test variable, and the modified NIH classification system was used as the state variable to plot the ROC curve. The AUC value was 0.80 (95% CI, 0.73–0.85). As a control, when tumor size was used as the test variable, the AUC value was 0.73 (95% CI, 0.67–0.80), which was significantly lower than that achieved when the model was used as the test variable (P=0.034, Fig. 6). The incidence rate of high malignancy potential GISTs was 0.10 in the low-grade risk group (0-3 points) and 0.42 in the high-grade risk group (4-6 points) (χ 2 =23.69 and P<0.001); the OR was 6.33 (95% CI, 2.85 -14.03). Based on the aforementioned cutoff value, the sensitivity, positive predictive value, specificity and negative predictive value were 0.59, 0.42, 0.82 and 0.90, respectively. In addition, this model was well-calibrated with the Hosmer-Lemeshow goodness-of-fit test (χ 2 =1.09, df=3, P =0.779).
DISCUSSION 10
GISTs smaller than 2 cm in diameter are referred to as small GISTs. According to the modified NIH classification system, gastric GISTs smaller than 2 cm and with <5 mitoses/50 high-power fields (HPFs) were considered very low risk.11 Thus, surgical resection was the treatment of choice for primary, localized gastric GISTs larger than 2 cm, whereas conservative follow-up was suggested for lesions smaller than 2 cm.5 However, a small portion of small gastric GISTs have also been reported to have malignant potential.18 In addition, small gastric GISTs have also been reported to undergo early metastasis to the liver.19 Thus, tumor size alone might not reliably evaluate the malignant potential of gastric GISTs. In the current study, we established a novel, simple-to-use preoperative multivariate prediction model for gastric GISTs smaller than 5 cm. This model efficiently predicted high malignancy potential gastric GISTs based on easily obtained clinical features. High-risk ultrasonography features were not found to be independent predictive factors in our multivariate analysis, although they are recommended by current guidelines to predict the malignant potential of gastric GISTs.5 EUS was previously shown to be valuable in preoperatively predicting the malignant potential of gastric GISTs.8-10, 13-14, 20 However, the criteria for identifying high-risk ultrasonography features largely rely on the expertise of endoscopists and have therefore remained subjective. There is a lack of objective executive criteria for the identification of hyperechoic foci, heterogeneity, and cystic spaces during the evaluation of EUS. There was no well-accepted consensus regarding which EUS features were the most 11
highly correlated with the malignancy of GISTs.21 We considered endoscopist variation might influence the significance of high-risk features in clinical practice. Another possible reason might be that previously identified high-risk features were obtained mainly from large-size GISTs, and those results could be different from those obtained in the cohort we included in this study.9-10, 13 However, we did not analyze the significance of single high-risk feature one by one due to limitations associated with the sample size, which could have led to false-negative evaluations of these features. EUS-guided fine-needle aspiration biopsy (EUS-FNA) is an excellent way to obtain tissue for preoperative analysis of GIST. However, the gastric GISTs in our study were relatively small, making it technically difficult to perform a standard EUS-FNA procedure. The specimen sample volume obtained by EUS-FNA was usually small; therefore, assessment of the mitotic index was difficult.21 Moreover, the diagnostic rate of EUS-FNA tended to decrease as the tumor diameter decreased,22-23 hence, EUS-FNA was recommended for relatively large tumors.24 To our knowledge, the current study is the first to reveal that an irregular tumor shape is an independent risk factor for high malignancy potential gastric GISTs. In fact, this factor has not previously been identified as an independent risk factor even though it has been included in previous studies.8, 14, 25 One possible reason might lie in the fact that previous studies had usually included a quite limited number of patients, especially those with GISTs less than 5 cm. In some studies, only factors with a P value < .05 in the univariate analysis were introduced into the multivariate 12
analysis.25-26 P values were influenced by the sample size and became relatively larger as the sample size decreased. Therefore, some factors could be missed if they were adopted or abandoned based only on the P value.27 Biologically diversity in the rapid growth rate of the internal structure could contribute to the formation of an irregular tumor shape, but the detailed mechanisms require further study. It has been suggested that ulceration is associated with high malignancy potential gastric GISTs,13-14, 25 in accordance with our research results. The National Comprehensive Cancer Network (NCCN) Guidelines also recommended that endoscopists should take ulceration as a risk factor for high malignancy potential gastric GISTs.5 There are several potential advantages related to our new 6-point multivariate prediction model in a clinical setting. First, this prediction model could be used to predict the malignant potential of gastric GISTs before resection. At present, various guidelines call for the malignant potential of gastric GISTs to be evaluated based on postoperative pathological calculation of the mitotic index, which can be acquired only after tumor resection.5, 11, 28 Our prediction model provides support for making appropriate treatment decisions for patients. Ideally, patients with low malignancy potential gastric GISTs can be treated safely endoscopically, whereas those with high malignancy potential gastric GISTs could choose a more-aggressive surgical treatment. Second, the predictive accuracy of the model was better than that achieved when using tumor size alone as the test variable. In the validation cohort, the model achieved satisfactory predictive accuracy when compared with tumor size alone. 13
Finally, the 3 factors that were included in the multivariate prediction model were simple and easy to obtain in clinical practice. There are also a few limitations to this study. First, in this study, we diagnosed the malignant potential of gastric GISTs based on the modified NIH classification system,11 which is a less-optimal parameter than patient outcomes. Low-risk gastric GISTs evaluated using the modified NIH classification system were reported to recur locally or to metastasize after R0 resection.29 Further study is needed to confirm the relationship between the model and patient outcomes. Second, this study was retrospective in nature and therefore had inherent potential for bias. A large-scale prospective randomized controlled trial is required for verification of this multivariate prediction model. Third, the events per variable value observed in this study were slightly low for a proper logistic regression analysis. In this study, the number of estimated variables in the logistic regression analysis was 7, whereas that of high malignancy potential gastric GISTs was 64, resulting in an event per variable value of 9.1, which was slightly lower than the preferable value of ≥10.30 Thus, a type II error might have occurred in this analysis. In summary, in this study, we demonstrate that a tumor size larger than 2 cm, mucosal ulceration and an irregular tumor shape are independent risk factors for patients with high malignancy potential gastric GISTs smaller than 5 cm. Moreover, these 3 factors can be applied to establish a simple-to-use preoperative predictive model that could help clinicians predict the malignant potential of gastric GISTs smaller than 5 cm. 14
REFERENCES 1. Soreide K, Sandvik OM, Soreide JA, et al. Global epidemiology of gastrointestinal stromal tumours (GIST): A systematic review of population-based cohort studies. Cancer Epidemiol 2016;40:39-46. 2. Chandrasekhara V, Ginsberg GG. Endoscopic management of gastrointestinal stromal tumors. Curr Gastroenterol Rep 2011;13:532-9. 3. Sarlomo-Rikala M, Kovatich AJ, Barusevicius A, et al. CD117: a sensitive marker for gastrointestinal stromal tumors that is more specific than CD34. Mod Pathol 1998;11:728-34. 4. West RB, Corless CL, Chen X, et al. The novel marker, DOG1, is expressed ubiquitously in gastrointestinal stromal tumors irrespective of KIT or PDGFRA mutation status. Am J Pathol 2004;165:107-13. 5. von Mehren M, Randall RL, Benjamin RS, et al. Soft Tissue Sarcoma, Version 2.2018, NCCN Clinical Practice Guidelines in Oncology. J Natl Compr Canc Netw 2018;16:536-63. 6. Meng Y, Li W, Han L, et al. Long-term outcomes of endoscopic submucosal dissection versus laparoscopic resection for gastric stromal tumors less than 2 cm. J Gastroenterol Hepatol 2017;32:1693-7. 7. Feng F, Liu Z, Zhang X, et al. Comparison of Endoscopic and Open Resection for Small Gastric Gastrointestinal Stromal Tumor. Transl Oncol 2015;8:504-8. 8. Chen T, Xu L, Dong X, et al. The roles of CT and EUS in the preoperative evaluation of gastric gastrointestinal stromal tumors larger than 2 cm. Eur Radiol 15
2019;29:2481-9. 9. Chak A, Canto MI, Rosch T, et al. Endosonographic differentiation of benign and malignant stromal cell tumors. Gastrointest Endosc 1997;45:468-73. 10. Shah P, Gao F, Edmundowicz SA, et al. Predicting malignant potential of gastrointestinal stromal tumors using endoscopic ultrasound. Dig Dis Sci 2009;54:1265-9. 11. Joensuu H. Risk stratification of patients diagnosed with gastrointestinal stromal tumor. Hum Pathol 2008;39:1411-9. 12. Kim MY, Park YS, Choi KD, et al. Predictors of recurrence after resection of small gastric gastrointestinal stromal tumors of 5 cm or less. J Clin Gastroenterol 2012;46:130-7. 13. Chen TH, Hsu CM, Chu YY, et al. Association of endoscopic ultrasonographic parameters and gastrointestinal stromal tumors (GISTs): can endoscopic ultrasonography be used to screen gastric
GISTs for potential malignancy?
Scand J Gastroenterol 2016;51:374-7. 14. Jeon SW, Park YD, Chung YJ, et al. Gastrointestinal stromal tumors of the stomach: endosonographic differentiation in relation to histological risk. J Gastroenterol Hepatol 2007;22:2069-75. 15. Joensuu H, Vehtari A, Riihimaki J, et al. Risk of recurrence of gastrointestinal stromal tumour after surgery: an analysis
of pooled population-based cohorts.
Lancet Oncol 2012;13:265-74. 16. Sun ZQ, Xu YY, Liang WJ, et al. Medical Statistics, 4th edition. Beijing: People’s 16
Medical Publishing House 2014; pp. 295. 17. Li J, Fine J. On sample size for sensitivity and specificity in prospective diagnostic accuracy studies. Stat Med 2004;23:2537-50. 18. Yang J, Feng F, Li M, et al. Surgical resection should be taken into consideration for the treatment of small gastric gastrointestinal stromal tumors. World J Surg Oncol 2013;11:273. 19. Tanaka J, Oshima T, Hori K, et al. Small gastrointestinal stromal tumor of the stomach showing rapid growth and early metastasis to the liver. Dig Endosc 2010;22:354-6. 20. Palazzo L, Landi B, Cellier C, et al. Endosonographic features predictive of benign and malignant gastrointestinal stromal cell tumours. Gut 2000;46:88-92. 21. Faulx AL, Kothari S, Acosta RD, et al. The role of endoscopy in subepithelial lesions of the GI tract. Gastrointest Endosc 2017;85:1117-32. 22. Attila T, Aydin O. Lesion size determines diagnostic yield of EUS-FNA with onsite cytopathologic evaluation for upper gastrointestinal subepithelial lesions. Turk J Gastroenterol 2018;29:436-41. 23. Akahoshi K, Sumida Y, Matsui N, et al. Preoperative diagnosis of gastrointestinal stromal tumor by endoscopic ultrasound-guided fine needle aspiration. World J Gastroenterol 2007;13:2077-82. 24. Akahoshi K, Oya M, Koga T, et al. Clinical usefulness of endoscopic ultrasound-guided fine needle aspiration for gastric subepithelial lesions smaller than 2 cm. J Gastrointestin Liver Dis 2014;23:405-12. 17
25. Lv GR, Zhou YH, Zhong JW, et al. Comparison of endoscopic ultrasonography image characteristics and analysis of risk factors for different invasive risk of gastric stromal tumors. China Journal of Endoscopy 2016;1:1-4. 26. Zhou C, Duan X, Zhang X, et al. Predictive features of CT for risk stratifications in
patients
with
primary gastrointestinal
stromal
tumour.
Eur Radiol
2016;26:3086-93. 27. Chen CZ, Chen XL, Wei C. Epidemiological Data Analysis and EmpowerStats Software Application, 1th edition. Shanghai: Shanghai Scientific & Technical Publishers 2016; pp. 19. 28. Fletcher CDM, Bridge JA, Hogendoorn PCW, et al. WHO Classfication of Tumours of Soft Tissue and Bone. 4th edition. Lyon: IARC Press 2013; pp. 164-7. 29. Lee S, Kim YN, Son T, et al. Oncologic Safety of Laparoscopic Wedge Resection with Gastrotomy for Gastric Gastrointestinal Stromal Tumor: Comparison with Conventional Laparoscopic Wedge Resection. J Gastric Cancer 2015;15:231-7. 30. Peduzzi P, Concato J, Kemper E, et al. A simulation study of the number of events per variable in logistic regression analysis. J Clin Epidemiol 1996;49:1373-9.
18
Table 1. Modified NIH classification system Risk category Tumor size (cm) Mitotic index (per 50 HPFs) Very-low risk ＜2.0 ≤5 Low risk 2.1-5.0 ≤5 Intermediate risk 2.1-5.0 ＞5 ＜5.0 6-10 5.1-10.0 ≤5 High risk Any Any ＞10.0 Any Any ＞10 ＞5.0 ＞5 2.1-5.0 ＞5 ≤5 5.1-10.0 NIH, National Institutes of Health.
Primary tumor site Any Any Gastric Any Gastric Tumor rupture Any Any Any Nongastric Nongastric
19
Table 2. Baseline characteristics of the developing (n=275) and validation (n=186) cohorts. Developing cohort Validation cohort P value (n=275) (n=186) a 57 (50-64） 56 (50-63) 0.469 Age, y, median (IQR) 0.895 Sex, n (%) Male 121 (44.0） 83 (44.6) Female 154 (56.0） 103 (55.4) 0.308 Location, n (%) Cardia/Gastric fundus 166 (60.4） 121 (65.1) Gastric body/angle/antrum 109 (39.6） 65 (34.9) 2.5 (1.3-3.6) 0.896 2.4 (1.2-3.5） Tumor size (cm), median (IQR) 0.070 Tumor shape, n (%) Round/oval 215 (78.2） 158 (84.9) irregular 60 (21.8) 28 (15.1) 0.676 Mucosal ulceration, n (%) Presence 49 (17.8) 36 (19.4) Absence 226 (82.2) 150 (80.6) High-risk ultrasonography b features , n(%) Presence 150 (54.5) Absence 125 (45.5) a. IQR, interquartile range. b. High-risk ultrasonography features: heterogeneity, hyperechoic foci, or cystic spaces.
20
Table 3. Univariate analysis of the risk factors for high malignancy potential gastric GISTs smaller than 5 cm (n [%]) in developing cohort. Factor Low malignancy High malignancy P value a a potential group potential group (n=64) (n=211) 0.000 Tumor size (cm)b <2 95 (45.0) 5 (7.8) 2-5 116 (55.0) 59 (92.2) 0.535 Sex Male 95 (45.0) 26 (40.6) Female 116 (55.0) 38 (59.4) 0.533 Age (years) ≤60 126 (59.7) 41 (64.1) >60 85 (40.3) 23 (35.9) 0.000 Mucosal ulceration Absence 187 (88.6) 39 (60.9) Presence 24 (11.4) 25 (39.1) 0.376 High-risk ultrasonography c features Absence 99 (46.9) 26 (40.6) Presence 112 (53.1) 38 (59.4) 0.000 Tumor shape Round/oval 182 (86.3) 33 (51.6) Irregular 29 (13.7) 31 (48.4) 0.490 Location Cardia/gastric fundus 125 (59.2) 41 (64.1) Gastric body/angle/antrum 86 (40.7) 23 (35.9) a. Low malignancy potential group, very-low-/low-risk groups based on the modified NIH classification system; high malignancy potential group, moderate-/high-risk groups based on the modified NIH classification system.11 b. Mainly based on CT measurements (those small GISTs that could not be detected in CT were estimated by measuring postoperative pathological specimen in vitro). c. High-risk ultrasonography features: heterogeneity, hyperechoic foci, or cystic spaces.
21
Table 4. Multivariate logistic regression analysis of the risk factors for high malignancy potential gastric GISTs smaller than 5 cm in developing cohort and the prediction modela. Factor Standard Wald P Odds ratio (95% CI) β error value Tumor size (cm)b <2 1 (reference) 2-5 0.520 12.888 0.000 6.474 (2.335-17.948) 1.868 Sex Male 1 (reference) Female 0.338 1.600 0.206 1.534 (0.791-2.976) 0.428 Age (years) ≤60 1 (reference) >60 0.340 0.457 0.499 0.794 (0.408-1.548) -0.230 Mucosal ulceration Absence 1 (reference) Presence 0.376 5.479 0.019 2.412 (1.154-5.041) 0.880 High-risk ultrasonography featuresc Absence 1 (reference) Presence 0.342 0.680 0.410 0.754 (0.386-1.474) -0.282 Tumor shape Round/oval 1 (reference) Irregular 0.362 12.227 0.000 3.548 (1.745 -7.216) 1.266 Location Cardia/Gastric fundus 1 (reference) Gastric body/angle/antrum 0.338 2.514 0.113 0.585 (0.302-1.135) -0.536 a. Low malignancy potential group, very-low risk/low-risk groups based on the modified NIH classification system; high malignancy potential group, moderate-/high-risk groups based on the modified NIH classification system.11 b. Mainly based on CT measurements (those small GISTs that could not be detected in CT were estimated by measuring postoperative pathological specimen in vitro). c. High-risk ultrasonography features: heterogeneity, hyperechoic foci, or cystic spaces. d. The regression coefficients of the 3 independent risk factors were scored by linear transformation, and all regression coefficients were divided by 0.6. The purpose of dividing by 0.6 was to thoroughly discriminate the original coefficients so that their difference was reflected as much as possible. The obtained values were rounded to integers.
22
Pointsd
3
-
-
1
-
2
-
FIGURE LEGENDS Figure 1. Endoscopic images of high malignancy potential gastric GISTs. A, Irregular tumor shape. B, Mucosal ulceration. Figure 2. Flowchart of patient enrollment. GISTs, gastrointestinal stromal tumors. Figure 3. ROC curves of actual tumor size in the developing cohort. The actual tumor size was used as the test variable, whereas the modified NIH classification system was used as the state variable to plot the ROC curve (very-low risk/low-risk groups as the low malignancy potential group and medium-/high-risk groups as the high malignancy potential group). Figure 4. ROC curves for tumor size categorized at 1-cm intervals and the prediction model for the developing cohort. Tumor size and the multivariate prediction model were used as the test variables, whereas the modified NIH classification system was used as the state variable to plot the ROC curve. Figure 5. ROC curves of the model score in the developing cohort. The model score was used as the test variable, whereas the modified NIH classification system was used as the state variable to plot the ROC curve. Figure 6. ROC curves of tumor size categorized at 1-cm intervals and the prediction model in the validation cohort. Tumor size and the multivariate prediction model were used as the test variables, whereas the modified NIH classification system was used as the state variable to plot the ROC curve.
23
Abbreviations CI, confidence interval; ESD, endoscopic submucosal dissection; OR, odds ratio; ER, endoscopic resection; GISTs, gastrointestinal stromal tumors; CT, computed tomography; NIH, National Institutes of Health; ROC, receiver operating characteristic; AUC, area under the curve; HPF, high-power field; EUS, Endoscopic ultrasonography; FNA, fine needle aspiration biopsy; IQR, interquartile range; NCCN, National Comprehensive Cancer Network.
1