Association of Tumor-Associated Collagen Signature With Prognosis and Adjuvant Chemotherapy Benefits in Patients With Gastric Cancer

Key Points

Question  Can the accuracy of prognosis prediction be improved using a tumor-associated collagen signature of gastric cancer (TACS_GC) to detect patients who are likely to benefit from adjuvant chemotherapy for GC?

Findings  In this cohort study of 519 patients, a low TACS_GC level was associated with a better prognosis, and patients with stage II and III GC with a low TACS_GC level were more likely to benefit from adjuvant chemotherapy.

Meaning  The findings suggest that TACS_GC may improve the accuracy of prognosis prediction and provide a helpful reference for decision-making regarding adjuvant chemotherapy in GC.

Importance  The current TNM staging system provides limited information for prognosis prediction and adjuvant chemotherapy benefits for patients with gastric cancer (GC).

Objective  To develop a tumor-associated collagen signature of GC (TACS_GC) in the tumor microenvironment to predict prognosis and adjuvant chemotherapy benefits in patients with GC.

Design, Setting, and Participants  This retrospective cohort study included a training cohort of 294 consecutive patients treated between January 1, 2012, and December 31, 2013, from Nanfang Hospital, Southern Medical University, People's Republic of China, and a validation cohort of 225 consecutive patients treated between October 1, 2010, and December 31, 2012, from Fujian Provincial Cancer Hospital, Fujian Medical University, People's Republic of China. In total, 146 collagen features in the tumor microenvironment were extracted with multiphoton imaging. A TACS_GC was then constructed using the least absolute shrinkage and selection operator Cox proportional hazards regression model in the training cohort. Data analysis was conducted from October 1, 2020, to April 30, 2021.

Main Outcomes and Measures  The association of TACS_GC with disease-free survival (DFS) and overall survival (OS) was assessed. An independent external cohort was included to validate the results. Interactions between TACS_GC and adjuvant chemotherapy were calculated.

Results  This study included 519 patients (median age, 57 years [IQR, 49-65 years]; 360 [69.4%] male). A 9 feature–based TACS_GC was built. A higher TACS_GC level was significantly associated with worse DFS and OS in both the training (DFS: hazard ratio [HR], 3.57 [95% CI, 2.45-5.20]; OS: HR, 3.54 [95% CI, 2.41-5.20]) and validation (DFS: HR, 3.10 [95% CI, 2.26-4.27]; OS: HR, 3.24 [95% CI, 2.33-4.50]) cohorts (continuous variable, P < .001 for all comparisons). Multivariable analyses found that carbohydrate antigen 19-9, depth of invasion, lymph node metastasis, distant metastasis, and TACS_GC were independent prognostic predictors of GC, and 2 integrated nomograms that included the 5 predictors were established for predicting DFS and OS. Compared with clinicopathological models that included only the 4 clinicopathological predictors, the integrated nomograms yielded an improved discrimination for prognosis prediction in a C index comparison (training cohort: DFS, 0.80 [95% CI, 0.73-0.88] vs 0.78 [95% CI, 0.71-0.85], P = .03; OS, 0.81 [95% CI, 0.75-0.88] vs 0.80 [95% CI, 0.73-0.86], P = .03; validation cohort: DFS, 0.78 [95% CI, 0.70-0.87] vs 0.76 [95% CI, 0.67-0.84], P = .006; OS, 0.78 [95% CI, 0.69-0.86] vs 0.75 [95% CI, 0.67-0.84], P = .002). Patients with stage II and III GC and low TACS_GC levels rather than high TACS_GC levels had a favorable response to adjuvant chemotherapy (DFS: HR, 0.65 [95% CI, 0.43-0.96]; P = .03; OS: HR, 0.55 [95% CI, 0.36-0.82]; P = .004; dichotomized variable, P < .001 for interaction for both comparisons).

Conclusions and Relevance  The findings suggest that TACS_GC provides additional prognostic information for patients with GC and may distinguish patients with stage II and III disease who are more likely to derive benefits from adjuvant chemotherapy.

Despite remarkable improvements in the diagnosis and treatment of gastric cancer (GC), it remains a major global health burden.¹ The pathological staging system is the gold standard for treatment planning and prognosis prediction for patients with GC.² Patients with advanced GC are advised to receive 5-fluorouracil–based adjuvant chemotherapy after radical surgery.^3-5 However, significant variations in survival outcomes have been observed even among patients with the same pathological stage who receive similar treatment regimens.⁶ These findings indicate that the pathological staging system provides inadequate prognostic information and fails to accurately identify which patients might benefit from adjuvant chemotherapy. Thus, biomarkers that could improve the prognosis prediction and determine adjuvant chemotherapy benefits are urgently needed.

The extracellular matrix is important for regulating neoplastic progression and response to chemotherapy.⁷ As the main component of the extracellular matrix, collagen accounts for most of its functions.^8-11 Increased collagen density enhances the invasiveness of tumor cells.^12,13 Oriented collagen around tumor cells is also an indicator of disease progression.^14,15 Moreover, stiff and dense collagen compresses intratumoral blood vessels, which induces hypoxia and impedes anticancer drug delivery.⁸ Thus, collagen alterations in the tumor microenvironment might provide information on prognosis and response to chemotherapy.

Multiphoton imaging has been widely applied in biological research.¹⁶ Multiphoton imaging, which combines 2-photon excitation fluorescence and second harmonic generation, is a stain-free method that shows comparable results to hematoxylin and eosin staining.¹⁷ Because of its physical origins, multiphoton imaging is sensitive for visualizing the collagen microstructure.¹⁸ High-dimensional quantitative collagen features, including morphologic and textural features, are therefore acquired from multiphoton images to describe collagen alterations.¹⁹

Integration of multiple features into a single signature shows a better performance than that of a single feature.^20,21 Cox proportional hazards regression with the least absolute shrinkage and selection operator (LASSO) is a state-of-the-art method that is used for the regression of high-dimensional data for survival analysis.^22-24 We used Cox proportional hazards regression with LASSO to construct a multiple collagen feature–based signature (ie, the tumor-associated collagen signature of GC [TACS_GC]) to predict disease-free survival (DFS) and overall survival (OS) among patients with GC. Moreover, we investigated whether TACS_GC could distinguish patients who might benefit from adjuvant chemotherapy.

The study design is shown in eFigure 1 in the Supplement. In this retrospective cohort study, we included 519 patients with resected GC from 2 medical centers. For training purposes, 294 consecutive patients were included from Nanfang Hospital, Southern Medical University, People's Republic of China, between January 1, 2012, and December 31, 2013. For validation purposes, 225 consecutive patients were included from Fujian Provincial Cancer Hospital, Fujian Medical University, People's Republic of China, between October 1, 2010, and December 31, 2012. The inclusion criteria were as follows: histologically diagnosed GC, radical gastrectomy with at least 15 lymph nodes harvested, availability of clinicopathological data, and complete postoperative follow-up. Patients treated with neoadjuvant chemotherapy, radiotherapy, or chemoradiotherapy were excluded. Data analysis was conducted from October 1, 2020, to April 30, 2021. This study was approved by the institutional review boards of Nanfang Hospital of Southern Medical University and Fujian Provincial Cancer Hospital of Fujian Medical University. Written informed consent was obtained from all patients before surgery, which contained a statement on the formalin-fixed, paraffin-embedded samples and clinicopathological data for scientific research. All data were deidentified. All procedures that involved human participants were performed in accordance with the Declaration of Helsinki.²⁵ This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline and the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) reporting guideline.

The baseline information, including age, sex, carcinoembryonic antigen level, cancer antigen 19-9 (CA 19-9) level, tumor location, tumor size, tumor differentiation, Lauren type, depth of invasion (T stage), lymph node metastasis (N stage), distant metastasis (M stage), TNM stage, and postoperative chemotherapy, was obtained from the archived data. The TNM stage was reclassified according to the eighth edition of the AJCC Cancer Staging Manual of the American Joint Committee on Cancer.²⁶ Routine 5-fluorouracil–based adjuvant chemotherapy was initiated after surgery if patient conditions were suitable according to the National Comprehensive Cancer Network guidelines.

The primary outcomes were 5-year DFS and OS. The patients were followed up once every 3 months for the first 2 years after surgery, every 6 months for the next 3 years, and annually thereafter. The follow-up duration was measured from the time of surgery to the last follow-up, and survival status at the last follow-up was documented. Disease-free survival was defined as the interval between surgery and recurrence at any site or all-cause death, whichever came first. Overall survival was defined as the interval between surgery and death from any cause.

The formalin-fixed, paraffin-embedded samples were prepared as 5-μm–thick sections for hematoxylin and eosin staining to determine the regions of interest, and the same regions on the other serial section were selected for multiphoton imaging. Two pathologists (Z.W. and L.L.) who were blinded to the clinical information independently reassessed the tumor region using a microscope. When the 2 pathologists had different opinions, a third pathologist (J.L.) was consulted and a consensus was reached through discussion. Finally, 5 regions of interest with a field of view of 500 × 500 μm per sample within the tumor region were randomly selected.

The images were acquired from a commercial laser scanning multiphoton microscope (LSM 880, Zeiss) as previously reported (eMethods in the Supplement).²⁷ The extraction of collagen features was performed via MATLAB 2015b (MathWorks).⁹ Four types of collagen features were extracted, including morphologic features and 3 types of textural features (histogram-based features, gray-level concurrence matrix–based features, and Gabor wavelet transform features). Detailed procedures are provided in the eMethods in the Supplement. Finally, a total of 146 features, including 12 morphologic features, 6 histogram-based features, 80 gray-level concurrence matrix–based features, and 48 Gabor wavelet transform features, were extracted (eTable 1 in the Supplement).

Cox proportional hazards regression with LASSO is an effective method to process high-dimensional predictors for survival analysis.^22,23 This method uses an L1 penalty to shrink the coefficients to 0. The penalty parameter λ controls the strength of the penalty. If we reduce λ and relax the penalty, more predictors can enter the model. In this study, Cox proportional hazards regression with LASSO with 10-fold cross-validation was used to select the most predictive features in the training cohort,^22,23 and a multiple feature–based TACS_GC was then constructed via a linear weighted combination of the selected features. The TACS_GC in the validation cohort was calculated by the selected features with their respective coefficients obtained from the training cohort.

An optimal cutoff value was identified via the maximally selected rank statistics to classify patients into high and low TACS_GC groups in the training cohort, and the same cutoff value was applied in the validation cohort.²⁸ Survival differences between the high and low TACS_GC groups were compared, and the associations of TACS_GC with DFS and OS were evaluated.

In the training cohort, the clinicopathological characteristics and TACS_GC were included in the univariate Cox proportional hazards regression analyses for DFS and OS. Variables with P < .05 were selected for the multivariable analyses. A backward stepwise regression was used to detect the independent predictors. Two integrated nomograms for the prediction of DFS and OS were developed based on the independent predictors.

To quantify the discrimination of the integrated nomograms, the C indexes were calculated, 5-year time-dependent receiver operator characteristic (ROC) curves were plotted, and the area under the ROC (AUROC) curves were computed.²⁹ The C index used the survival time as response, and the time-independent AUROC curve used the binary survival status at a given time point as response. The former quantified how well the prediction model could order the survival times, and the latter quantified how well the prediction model could order the survival status at a given time point.²⁹ To compare the agreement between the nomogram-predicted survival probabilities and the actual probabilities, calibration curves were generated.³⁰ To evaluate the clinical usefulness, decision curve analysis was used to assess the net benefits at different threshold probabilities.^31,32 The integrated nomograms were then applied in the validation cohort to validate the discrimination, calibration, and clinical usefulness.³³

To evaluate the incremental value of the TACS_GC to clinicopathological risk factors for prognosis prediction, 2 clinicopathological models, including independent clinicopathological risk factors for the prediction of DFS and OS, were built in the training cohort and then applied in the validation cohort. The incremental value of the TACS_GC to the clinicopathological models and TNM staging system was assessed with respect to discrimination, reclassification, and clinical usefulness.³³

Continuous variables were compared by the independent sample, unpaired t test if normally distributed or the Mann-Whitney U test if nonnormally distributed. When a categorical variable was compared, if the table was a 2 × 2 contingency table with at least 1 expected cell count less than 5, the Fisher exact test was calculated or the χ² test was used; if the table was a 2 × C (C > 2) contingency table, the χ² test was performed. C indexes and time-independent AUROC curves were compared by the z score test and DeLong test, respectively. The Kaplan-Meier method and log-rank test were used to estimate DFS and OS, and Cox proportional hazards regression was conducted to compute the hazard ratios (HRs) with 95% CIs. Interactions between the TACS_GC level (high vs low) and adjuvant chemotherapy (yes vs no) were tested by Cox proportional hazards regression. All statistical analyses were performed using R software, version 3.6.2 (R Foundation for Statistical Computing) and SPSS software, version 19.0 (IBM Inc). A 2-sided P < .05 was considered statistically significant.

Of the 519 patients included in the study (median age, 57 years [IQR, 49-65 years]; 360 [69.4%] male), 294 consecutive patients (median age, 57 years [IQR, 49-64 years]; 208 [70.7%] male) were included in the training cohort and 225 consecutive patients (median age, 58 years [IQR, 51-65 years]; 152 [67.6%] male) were included in the validation cohort. The clinicopathological characteristics of all patients are summarized in Table 1. No significant difference in the clinicopathological characteristics was found between the 2 cohorts. The clinicopathological characteristics between patients with and without complete data were similar (eTable 2 in the Supplement).

The construction framework of the TACS_GC is presented in Figure 1. A TACS_GC that included 9 collagen features was constructed (eFigure 2 in the Supplement). On the basis of the training cohort, an optimal cutoff value of 2.81 was determined, and all included patients were divided into high and low TACS_GC groups (eFigure 3 in the Supplement). The distribution of TACS_GC with the corresponding survival status is shown in eFigure 4 in the Supplement. Patients with higher TACS_GC values were more likely to experience disease recurrence or death. No significant difference was found in the TACS_GC values between the training (median, 2.47 [IQR 2.08-2.91]) and validation (median, 2.55 [IQR 2.08-2.96]) cohorts (median difference, −0.05; 95% CI, −0.15 to 0.05; P = .23). The TACS_GC level was significantly associated with tumor differentiation, depth of invasion, lymph node metastasis, and distant metastasis (eTables 3-5 in the Supplement).

In the training cohort, the 5-year DFS and OS were 71.7% (95% CI, 65.7%-78.2%) and 74.7% (95% CI, 68.9%-81.0%), respectively, in the low TACS_GC group. In the high TACS_GC group, the 5-year DFS and OS were significantly decreased to 25.9% (95% CI, 18.4%-36.3%) and 29.2% (95% CI, 21.4%-39.8%), respectively (Figure 2A and B). These results were further verified in the validation cohort (Figure 2C and D). The TACS_GC remained a significant prognostic indicator after stratification by clinicopathological variables, demonstrating the independent association of the TACS_GC with prognosis (eFigures 5 and 6 in the Supplement).

Univariate Cox proportional hazards regression analyses showed that TACS_GC was significantly associated with worse DFS and OS in both the training (DFS: HR, 3.57 [95% CI, 2.45-5.20]; OS: HR, 3.54 [95% CI, 2.41-5.20]; P < .001 for both comparisons) and validation (DFS: HR, 3.10 [95% CI, 2.26-4.27]; OS: HR, 3.24 [95% CI, 2.33-4.50]; P < .001 for both comparisons) cohorts (eTables 6 and 7 in the Supplement). Backward stepwise multivariable Cox proportional hazards regression analyses found that the TACS_GC was still independently associated with DFS and OS after adjustment for CA 19-9, T stage, N stage, and M stage in the training (DFS: HR, 2.03 [95% CI, 1.40-2.95]; OS: HR, 1.95 [95% CI, 1.33-2.88]; P < .001 for both comparisons) and validation (DFS: HR, 2.61 [95% CI, 1.82-3.76]; OS: HR, 2.39 [95% CI, 1.69-3.38]; P < .001 for both comparisons) cohorts (Table 2).

On the basis of the multivariable Cox proportional hazards regression analyses in the training cohort, 2 integrated nomograms for the prediction of DFS and OS, including TACS_GC, CA 19-9, T stage, N stage, and M stage, were developed (eFigure 7 in the Supplement). The integrated nomograms displayed C indexes of 0.80 (95% CI, 0.73-0.88) for DFS and 0.81 (95% CI, 0.75-0.88) for OS in the training cohort. In the validation cohort, the C indexes were 0.78 (95% CI, 0.70-0.87) for DFS and 0.78 (95% CI, 0.69-0.86) for OS. The calibration curves of the 2 integrated nomograms showed satisfactory agreement between the nomogram-predicted survival and actual survival in both the training and validation cohorts (eFigure 8 in the Supplement).

To evaluate the incremental value of the TACS_GC for prognosis prediction, 2 clinicopathological models, including T stage, N stage, M stage, and CA 19-9, for the prediction of DFS and OS were built (eTable 8 in the Supplement). Compared with the integrated nomograms, the clinicopathological models showed significantly decreased C indexes of 0.78 (95% CI, 0.71-0.85; P = .03) for DFS and 0.80 (95% CI, 0.73-0.86; P = .03) for OS in the training cohort. Similarly, significantly reduced C indexes of 0.76 (95% CI, 0.67-0.84; P = .006) for DFS and 0.75 (95% CI, 0.67-0.84; P = .002) for OS were detected in the validation cohort. In addition, the C indexes of the TNM staging system alone for DFS and OS were also reduced (eTable 9 in the Supplement). Moreover, significantly improved 5-year AUROCs were also found in the integrated nomograms for DFS and OS compared with the clinicopathological models (eFigure 9 in the Supplement). In addition, the integrated nomograms yielded net reclassification improvement values of 0.21 (95% CI, 0.05-0.33; P = .01) for DFS and 0.21 (95% CI, 0.02-0.32; P = .03) for OS to the clinicopathological models in the training cohort. Similar results were observed in the validation cohort (eTable 10 in the Supplement). Finally, the decision curve analysis showed that the integrated nomograms had a higher net benefit than the clinicopathological models and TNM staging system across most of the range of threshold probabilities (eFigure 10 in the Supplement).

The clinicopathological characteristics of patients with stage II and III GC according to adjuvant chemotherapy are listed in eTable 11 in the Supplement. An interaction test between the TACS_GC level and adjuvant chemotherapy indicated that patients with low TACS_GC levels had a superior benefit from adjuvant chemotherapy compared with patients with high TACS_GC levels (P < .05 for all comparisons) (eTable 12 in the Supplement). Patients with stage II and III GC and low TACS_GC levels rather than high TACS_GC levels had a favorable response to adjuvant chemotherapy (DFS: HR, 0.65 [95% CI, 0.43-0.96]; P = .03; OS: HR, 0.55 [95% CI, 0.36-0.82]; P = .004; dichotomized variable, P < .001 for interaction for both comparisons) (Figure 3). Subgroup analyses according to the TNM stage demonstrated that patients with low TACS_GC levels could experience significant survival benefits in the subgroups with stage II and III GC, but for patients with high TACS_GC levels, the survival benefits were less clear in the subgroup with stage II GC and were clinically small in the subgroup with stage III GC (eFigure 11 and 12 in the Supplement).

An accurate prediction of prognosis and survival benefits from adjuvant chemotherapy in patients with GC is integral to treatment decision-making. In this study, we constructed a TACS_GC in the tumor microenvironment of GC based on multiphoton imaging and demonstrated that the TACS_GC was significantly associated with prognosis of GC. Moreover, by incorporating TACS_GC into clinicopathological models, we found that the TACS_GC could provide additional prognostic information. Furthermore, we showed that the TACS_GC was a potential indicator of survival benefits associated with adjuvant chemotherapy in patients with stage II and III GC.

The construction of the TACS_GC was determined mainly by 3 key factors. First, the use of multiphoton imaging could specifically visualize the morphologic features of collagen because of its physical origins.¹⁶ Compared with the pathological staining methods of collagen, such as Masson trichrome staining, multiphoton imaging is acquired based on the intrinsic signals, thus avoiding additional reagents and batch-to-batch variations, which yields more sensitive information about collagen and ensures reproducible results.³⁴ Second, an objectively quantitative method for the extraction of collagen features is essential for describing collagen alterations in the tumor microenvironment. For this purpose, we established a standardized quantification framework for the quantitative measurement of high-dimensional collagen features from multiphoton imaging.^11,35 The third factor was a statistical model that could integrate high-dimensional features into a single signature. Cox proportional hazards regression with LASSO, a state-of-the-art statistical model that has satisfactory performance for dealing with high-dimensional data, was therefore used.^22,23,36 After considering the 3 key factors, the TACS_GC was constructed.

Collagen is usually organized as an isotropic meshwork in normal extracellular matrix but is realigned during tumor invasion.³⁷ Provenzano et al¹³ revealed that increased collagen density accelerated tumorigenesis, local invasion, and metastasis, causally linking increased stromal collagen to tumor formation and progression. A recent investigation¹⁰ also reported that a collagen-rich tumor microenvironment facilitated the early metastatic onset of tumor cells via interaction with integrin-α₂, thus promoting cell migration and anoikis resistance. In this study, a higher TACS_GC level was an indicator of poorer prognosis. According to the calculation formula, TACS_GC was positively correlated with collagen area, collagen straightness, and collagen cross-link density, which was in accordance with previous findings.⁷ However, the underlying mechanisms remain unclear. Therefore, additional work should focus on the underlying mechanisms of the interaction of tumor cells and collagen.

In addition to the morphologic features, textural features have also been measured for a more comprehensive description of collagen organization.³⁸ Currently, a common consensus about what types of textural features should be included to represent collagen alterations has not yet been reached. However, the textural features mainly included first-order statistics (eg, histogram-based features), second-order statistics (eg, gray-level concurrence matrix–based features), and wavelet transformation (eg, Gabor wavelet transform features),^38-40 which have been proven to have powerful potential for disease diagnosis. Stromal collagen information could be robustly measured and quantified via the combination of morphologic and textural features.⁴¹ Although most published studies^38-41 only assessed the correlation of textural features and patient outcomes, we assumed that there is an underlying molecular assignment for different textural features.

According to the so-called seed and soil hypothesis, cancer metastasis originated from the intricate interaction between tumor cells and their microenvironment.⁴² The tumor cells play the principal role and the tumor microenvironment plays the secondary role during the process of cancer metastasis.³⁷ In the integrated nomograms, CA 19-9, depth of invasion, lymph node metastasis, and distant metastasis represented the tumor cells, and the TACS_GC represented the structural tumor microenvironment. Therefore, despite prognosis prediction reaching statistical significance, the incremental value of TACS_GC was small. From the perspective of clinicians, the TACS_GC provided additional prognostic information and helped researchers to understand the interactions between tumor cells and their structural microenvironment; therefore, the TACS_GC was clinically relevant and worth investigation.

Adjuvant chemotherapy is the standard treatment for patients with stage II and III GC to improve survival outcomes.^2-6 However, many patients with stage II and III GC do not experience survival benefits from adjuvant chemotherapy, indicating that some patients might be overtreated because their tumors are not sensitive to the given type of chemotherapy.⁴³ Xiong et al⁴⁴ proposed that increased intratumoral collagen deposition was correlated with resistance to chemotherapy and with reduced survival in patients with breast cancer. In the present study, improved oncologic outcomes were detected in patients with low TACS_GC levels who received adjuvant chemotherapy but not in patients with high TACS_GC levels. Therefore, the TACS_GC level might be helpful for identifying patients with stage II and III GC who might benefit from adjuvant chemotherapy.

Multiphoton imaging has been recognized as a novel method for optical biopsy of samples.¹⁷ Because of the comparable results between multiphoton imaging and hematoxylin and eosin staining, pathologists or clinicians could differentiate the cancerous and normal tissues of the stomach after minimal training.^45,46 Routine formalin fixation and paraffin embedding were reported to have negligible influence on multiphoton image acquisition,⁴⁷ and approximately 10 minutes is needed to perform multiphoton imaging; thus, multiphoton imaging and TACS_GC are promising for clinical application. We believe that clinicians could obtain TACS_GC in the near future using multiphoton imaging, and the integrated nomograms could be used in clinical practice.

This study has limitations. First, it was retrospective in nature, and all specimens were acquired from 2 medical centers in China; thus, potential bias was inevitable. A prospective, multicenter trial is needed to validate the performance of TACS_GC. Second, the underlying mechanism of the TACS_GC for the prediction of prognosis and adjuvant chemotherapy benefits remains unclear; therefore, further investigations are needed to better understand the role of TACS_GC in tumor progression and chemotherapy response.

The findings suggest that TACS_GC could provide additional prognostic information to the TNM staging system for prognosis prediction of GC. Moreover, the TACS_GC value might be useful to distinguish patients with stage II and III GC who might benefit from adjuvant chemotherapy. Thus, the TACS_GC might be helpful for patient counseling, decision-making regarding adjuvant chemotherapy, and follow-up scheduling.

Accepted for Publication: October 3, 2021.

Published: November 30, 2021. doi:10.1001/jamanetworkopen.2021.36388

Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2021 Chen D et al. JAMA Network Open.

Corresponding Authors: Jun Yan, MD, PhD, Department of General Surgery, Guangdong Provincial Key Laboratory of Precision Medicine for Gastrointestinal Tumor, Nanfang Hospital, The First School of Clinical Medicine, Southern Medical University, 1838 N Guangzhou Ave, Guangzhou 510515, People’s Republic of China (yanjunfudan@163.com); Shuangmu Zhuo, PhD, School of Science, Jimei University, 185 Yinjiang Road, Xiamen 361021, People’s Republic of China (shuangmuzhuo@gmail.com).

Author Contributions: Drs D. Chen, H. Chen, Chi, and Fu contributed equally to this study and should be considered co–first authors. Dr Yan had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: D. Chen, X. Xu, Li, Zhuo, Yan.

Acquisition, analysis, or interpretation of data: D. Chen, H. Chen, Chi, Fu, Wang, Wu, S. Xu, Sun, Lin, Cheng, Jiang, Dong, Lu, Zheng, G. Chen, Zhuo, Yan.

Drafting of the manuscript: D. Chen, Chi, Fu, Wu, Sun, X. Xu, Lin, Cheng, Zhuo.

Critical revision of the manuscript for important intellectual content: D. Chen, H. Chen, Chi, Fu, Wang, S. Xu, Jiang, Dong, Lu, Zheng, G. Chen, Li, Zhuo, Yan.

Statistical analysis: D. Chen, H. Chen, Chi, Fu, S. Xu, X. Xu, Jiang, Dong, Zheng.

Obtained funding: D. Chen, G. Chen, Fu, Li, Yan.

Administrative, technical, or material support: Wang, S. Xu, G. Chen, Li, Zhuo, Yan.

Supervision: D. Chen, Li, Zhuo, Yan.

Conflict of Interest Disclosures: None reported.

Funding/Support: This study was supported by grants 82103041 (Dr D. Chen), 82102693 (Dr Fu), 81773117 (Dr Yan), and 81771881 (Dr Zhuo) from the National Natural Science Foundation of China; grant 2020B121201004 (Dr Li) from the Guangdong Provincial Key Laboratory of Precision Medicine for Gastrointestinal Cancer; grant 2019JC05Y361 (Dr Li) from the Guangdong Provincial Major Talents Project; grant 2018J07004 (Dr Zhuo) from the Natural Science Foundation of Fujian Province; grant 2019-WJ-21 (Dr Zhuo) from the Joint Funds of Fujian Provincial Health and Education Research; grants 2018Y2003 (Dr G. Chen), 2019L3018 (Dr G. Chen), and 2019YZ016006 (Dr G. Chen) from the Science and Technology Program of Fujian Province; and grants 2018CR034 (Dr Yan) and 2020CR001 (Dr D. Chen) from the Clinical Research Project of Nanfang Hospital.

Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: American Journal Experts (Durham, North Carolina) edited the manuscript for English language and grammar. They were compensated for their work.