Objectives: The study aimed to determine the impact of computed tomography (CT)-based body composition and radiomics nomogram on the prediction of postoperative complications in gastric cancer. Methods: The clinical data of 457 individuals with surgically confirmed gastric cancer, 320 patients in the training cohort (TC) and 137 patients in the validation cohort (VC), were retrospectively analyzed. Body composition data were measured using CT. Postoperative complications were graded using the Clavien-Dindo system. Dedicated radiomics prototype software was used to segment lesions and extract characteristics from preoperative portal venous-phase CT images. Clinical, radiomics, and combined models were developed using logistic regression analysis. Model performance was evaluated using the area under the curve (AUC) of receiver operating characteristic curve, and the prediction ability of the optimal model was evaluated using calibration curves and decision curve analysis (DCA). Results: Nutritional Risk Screening 2002 (NRS2002) score, sarcopenia, and blood loss were independent predictors of postoperative complications in gastric cancer. A radiomics signature was created using 19 conserved radiomics features. The nomogram based on both the clinical model and radiomics signature showed the greatest predictive performance, with AUCs of 0.763 (95% confidence interval [CI], 0.708–0.817) and 0.748 (95% CI: 0.667–0.818) in the TC and VC, respectively. The calibration curve and DCA revealed that the nomogram was beneficial in clinical practice for the preoperative prediction of postoperative complications. Conclusions: The combined model consisting of NRS2002 score, sarcopenia, blood loss, and a radiomics signature holds potential application value for the individualized prediction of postoperative complications in gastric cancer patients.

Gastric cancer (GC) is the 5th leading cause of cancer in the world and the 4th leading cause of cancer-related deaths [1]. Gastrectomy with adequate lymph node dissection is the most effective therapy for possibly curable GC [2]. The rate of postoperative complications, however, remains unacceptable, ranging from 18.3% to 36% [3]. Postoperative complications can prolong hospital stay and adversely affect the recurrence-free and overall survival of patients. Therefore, reliable preoperative assessment tests are crucial for the prediction of postoperative complications in GC patients.

The literature remains mixed on the risk factors for the development of postoperative complications in GC patients. Tumor heterogeneity influences the growth and malignant progression of GC, and is considered an important factor affecting the failure of cancer treatment and inaccurate prognostication [4]. In addition, the progression of GC is correlated with a drastic decline in nutrition and function of the human body, which also leads to poor treatment response and worse prognosis [5]. Malnourishment is a widespread and serious issue among GC patients. Numerous studies have shown that malnutrition is a risk factor for postoperative complications in GC patients [6, 7]. Hence, tumor heterogeneity and nutritional status in GC patients warrant closer examination.

Nutrition Risk Screening 2002 (NRS2002), Patient-Generated Subjective Global Assessment, and Mini Nutritional Assessment, among others, are commonly used nutritional screening and assessment tools. However, all these tools are limited by their subjective assessment [8, 9]. Currently, body composition is becoming an increasingly popular topic of nutritional research. Fat tissue and skeletal muscle can reflect the nutritional status of patients [7, 10]. The current gold standard for assessing body composition is computed tomography (CT) [11]. Moreover, CT plays an important role in the diagnostic and therapeutic evaluation of GC patients. Accordingly, CT-quantified body composition measurements, which allow objective and dynamic monitoring of the patient’s nutritional status, are useful for optimizing treatment.

Radiomics is a rising field that enables noninvasive characterization of tumor heterogeneity through quantitative features derived from imaging data [12]. Radiomics is a promising technique for predicting histopathological characteristics, treatment response, and even clinical outcomes in GC [13-16]. Therefore, we hypothesized that CT-based radiomics might provide valuable additional information that may help predict postoperative complications in GC patients. To the best of our knowledge, a postoperative prediction model that integrates CT-based body composition and radiomic features in gastrectomy patients has never been published. Moreover, there is currently no consensus on which change in body composition can be used as an indicator for nutritional assessment.

The goal of present study was to develop a visual prognostic model that incorporates clinical risk factors (including body composition) and radiomics features to predict postoperative complications in patients with GC. Such a tool would aid in the identification of high-risk patients and in the formulation of individualized treatments, thereby improving the assessment of nutritional status and patient prognosis.

Patients

The local Institutional Review Board approved this retrospective study and waived the requirement for informed consent. In total, 457 GC patients who underwent radical gastrectomy between January 2017 and December 2019 were sequentially enrolled in the current research. The criteria for inclusion were as follows: (1) radical gastrectomy with R0 resection (no residual tumor) and D2 lymphadenectomy, (2) surgeries performed by clinicians in the same department and with the same seniority level, (3) pathologically confirmed GC, (4) contrast-enhanced abdominal CT scan conducted within 2 weeks before radical gastrectomy, and (5) complete clinical, pathological, and imaging data. The following served as exclusion criteria: (1) preoperative anti-tumor treatment, (2) palliative surgical resection, (3) other malignant tumors, (4) severe underlying disease, such as severe heart disease and malignant hypertension, (5) metabolic diseases, such as diabetes, hyperthyroidism, and adrenal cortical insufficiency, (6) tumor diameter <5 mm, as this was insufficient to delineate the region of interest (ROI), and (7) follow-up duration <30 days. Using a probability-based technique and simple random sampling, we assigned the patients to validation and training datasets in a ratio of 3:7. The Clavien-Dindo classification was used to classify postoperative complications occurring within 30 days after the surgery [17]. Overall complications were categorized according to Clavien-Dindo grades as grade IIor higher. Tumor grading was carried out in accordance with the 8th edition of the Union for International Cancer Control’s Tumor Node Metastasis classification system. The clinical characteristics of the 457 GC patients are outlined in Table 1.

Table 1.

Comparison of clinical features in the TC and VC

Comparison of clinical features in the TC and VC
Comparison of clinical features in the TC and VC

CT Examination

CT examinations were performed using a 64-channel CT scanner (LightSpeed VCT, GE Healthcare), a 256-channel CT scanner (Revolution, GE Healthcare), and a dual-source CT scanner (SOMATOM Definition Flash; Siemens Healthcare). Detailed information on the CT protocols and variables acquired is presented in online supplementary Table A1 (for all online suppl. material, see www.karger.com/doi/10.1159/000526787). CT images in the venous phase were evaluated using a picture archiving and communication system.

Body Composition Assessment Using CT Images

All body composition data were processed on an AW4.6 workstation (GE HealthCare) and evaluated using the “X-Sect.” mode. The visceral fat area (VFA), subcutaneous fat area (SFA), and skeletal muscle area were calculated using a single slice taken at L3 in which both transverse processes were visible. The following tissue-specific criteria were used: −150 HU to −50 HU for visceral adipose tissue, −190 HU to −30 HU for subcutaneous adipose tissue, and −29 HU to 150 HU for skeletal muscle. The cross section of the skeletal muscle area was standardized to the square of the patient’s height (m2) to yield the skeletal muscle index. The visceral-to-subcutaneous ratio of adipose tissue area was used to assess the distribution of abdominal adipose tissue. Sarcopenia was defined as skeletal muscle index of ≤40.8 cm2/m2 in men and ≤34.9 cm2/m2 in women [18]. Visceral obesity was defined as a visceral-to-subcutaneous ratio of ≥1.33 in men and ≥0.93 in women [19]. Low SFA was defined as an SFA of <62.0 cm2 in women and <38.1 cm2 in men [5], and low VFA was defined as a VFA of <100 cm2 for both sexes [5].

Image Segmentation and Feature Extraction

The open-source program ITK-SNAP was used to segment the ROI. The greatest tumor cross section was chosen, and a two-dimensional ROI containing the tumor was segmented on portal venous-phase images with a slice thickness of 2.0–2.5 mm. All sections were assessed by two radiologists with more than 5 years of experience in diagnostic abdominal imaging. The radiologists highlighted the tumor locations on the relevant images. Radiologist A manually segmented the lesions for all subjects. The tumor was segmented on axial images, and intraluminal fluid and gas were carefully excluded. In addition, inter- and intra-class correlation coefficients (ICCs) were calculated for 50 subjects selected at random. The ROIs for each of these patients were determined by Radiologist A and Radiologist B. An ICC of more than 0.75 was deemed acceptable.

The A.K. software (Artificial Intelligence Kit, A.K.; GE Healthcare) was used to extract radiomics features. A total of 293 radiomics features were retrieved from the ROIs. Radiomics features were of 4 types: (1) 18 first-order statistics that depicted the voxel intensity distribution inside the tumor, (2) 14 shape features that reflected the size and shape of the ROI, (3) 103 second-order texture features, including gray-level co-occurrence matrix, gray-level size zone matrix, gray-level run length matrix, gray-level dependence matrix, and neighboring gray tone difference matrix, and (4) 158 filter features, including Laplacian of Gaussian filters. The original images were filtered using the Laplacian of Gaussian filters to enhance the edges in the images.

Construction of Prediction Models

Radiomics models were built with the open-source software FeAture Explorer Pro (FAEPro, version 0.3.7) on Python (3.7.6). Using a random series, all cases were allocated to the validation cohort (VC) and training cohort (TC) in a 3:7 ratio. The TC included 320 individuals (118/202 = positive/negative), and the VC included 137 patients (51/86 = positive/negative), with positive and negative signifying patients with and without postoperative complications, respectively. The synthetic minority oversampling technique was used to balance positive and negative samples in the training dataset. The feature matrix was normalized. Analysis of variance, recursive feature elimination, and relief were used for feature selection before model development. Among the top 20 features, the optimal combination of features was determined using a logistic regression classifier with 5-fold cross-validation.

Three prediction models were developed in this study: a clinical model, a radiomics model, and a combined model with both clinical and radiomics characteristics. Clinical characteristics and body composition data were included in the clinical model. Figure 1 depicts the method for creating these models.

Fig. 1.

Workflow of model construction.

Fig. 1.

Workflow of model construction.

Close modal

Nomogram Construction

We used receiver operating characteristic (ROC) curve analysis to examine the performance of the models. The area under the curve (AUC) of ROC curve was computed for quantification, and a cut-off value that maximized the Youden index, accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) was determined. Next, 95% confidence intervals (CIs) were estimated using the bootstrap method with 1,000 samples. Finally, the optimal model was used to establish a nomogram to visualize the model. The prediction ability of the nomogram was evaluated using a calibration curve and decision curve analysis (DCA).

Statistical Analysis

SPSS software (version 26.0), R software package (version 4.1.1), and MedCalc software (version 19.0.4) were used for statistical analysis. Continuous data were expressed as medians and interquartile ranges, and the Mann-Whitney U test was used to assess between-group differences in these data. Categorical variables were presented using numbers and percentages, and the χ2 or Fisher exact test was used to assess between-group differences in these variables. Logistic regression was used to perform univariate and multivariate analyses of postoperative complications, and logit transformation was used to standardize the distribution of continuous variables. The MedCalc software was used to plot the ROC curves of multiple models. The ROCs of the two cohorts were compared using the DeLong test. Differences in the aforementioned statistical analysis were judged to be statistically significant when the p value was less than 0.05. The nomogram and calibration curve was plotted using the rms package, and DCA was performed using the rmda package. The Hosmer-Lemeshow test was used to determine the goodness-of-fit of the nomogram in both cohorts, and it was satisfying when the p value exceeded 0.05.

Patients and Clinical Data

Analysis of the short-term surgical outcomes showed that a total of 169 patients (36.9%) had postoperative complications. Inflammatory complications were present in 99 patients (21.6%) and included pulmonary infections (n = 62), intra-abdominal infections (n = 11), anastomotic leakages (n = 15), urinary tract infections (n = 5), bloodstream infections (n = 3), and wound infections (n = 3). The overall complication rate was greater in patients with sarcopenia than in patients without sarcopenia (55.6% vs. 33.0%, p = 0.004). The incidence of inflammatory complications was higher in patients with visceral obesity than in patients without visceral obesity (35.8% vs. 19.8%, p = 0.009; Table 2). There was no statistically significant variation in the distribution of clinical factors between the TC and VC (p > 0.05; Table 1).

Table 2.

Postoperative outcomes

Postoperative outcomes
Postoperative outcomes

Construction of Clinical Model

Multivariate analysis revealed that the following clinical factors were independent predictors of postoperative complications: NRS2002 score, sarcopenia, and blood loss. The clinical model was constructed using these independent factors (Table 3).

Table 3.

Multivariate risk factor analysis for the clinical model

Multivariate risk factor analysis for the clinical model
Multivariate risk factor analysis for the clinical model

Construction of Radiomics Model and Combined Model

A total of 293 radiomics features were retrieved, and 271 of these features had ICC values exceeding 0.75. Using the FAEPro software, we performed logistic regression and selected 19 optimal radiomics characteristics with nonzero coefficients (online suppl. Table A2). These included 4 first-order statistics, 3 shape features, 3 second-order texture features (2 gray-level co-occurrence matrix and 1 neighboring gray tone difference matrix), and 9 filter features. The optimal features were used to construct a radiomics model. The clinical and radiomics models were integrated to create the combined model.

Comparison of the Prediction Performance of the Three Models

Among the three models, the combined model had the highest AUC (0.763 and 0.748 for TC and VC, respectively) and the highest accuracy (0.741 and 0.693 for TC and VC, respectively; Tables 4, 5). There were no significant differences in AUCs between the clinical and radiomics models in the two cohorts (DeLong test, p = 0.447 for TC, p = 0.960 for VC); the AUCs of the combined model significantly differed from those of the clinical and radiomics models (DeLong test, p = 0.007 and 0.002, respectively, for TC; p = 0.032 and 0.042, respectively, for VC; Fig. 2a, b). Therefore, the combined model was judged to be the best predictive model for postoperative complications in GC.

Table 4.

Model comparison in TC

Model comparison in TC
Model comparison in TC
Table 5.

Model comparison in VC

Model comparison in VC
Model comparison in VC
Fig. 2.

Receiver operating characteristic (ROC) curves of the three models in the training (a) and validation (b) cohorts. The area under the curve (AUC) of the ROC curve is a comprehensive embodiment of a model’s accuracy, sensitivity, and specificity. The combined model had significantly higher AUCs in both the training and validation cohorts than the clinical or radiomics model alone. The combined model was judged to be the best predictive model for postoperative complications in gastric cancer.

Fig. 2.

Receiver operating characteristic (ROC) curves of the three models in the training (a) and validation (b) cohorts. The area under the curve (AUC) of the ROC curve is a comprehensive embodiment of a model’s accuracy, sensitivity, and specificity. The combined model had significantly higher AUCs in both the training and validation cohorts than the clinical or radiomics model alone. The combined model was judged to be the best predictive model for postoperative complications in gastric cancer.

Close modal

Construction and Clinical Utility Analysis of the Nomogram

A nomogram was constructed based on the optimal model (Fig. 3). The higher the value calculated from the nomogram, the higher the likelihood of postoperative complications in GC. For example, the total score of a GC patient with a preoperative NRS2002 score ≥3 points, sarcopenia, intraoperative blood loss of 400 mL, and a radiomics score of 0.5 would be 33 + 16 + 32 + 50 = 131, and the risk of postoperative complications would be 69%. Physicians and patients can use the nomogram to predict the risk of postoperative complications of GC surgery and more accurately assess individual patients, which would help them select a more appropriate treatment plan.

Fig. 3.

Nomogram created using typical clinical risk variables and radiomics signature.

Fig. 3.

Nomogram created using typical clinical risk variables and radiomics signature.

Close modal

The calibration curves of the nomogram model demonstrated good agreement in both the TC and VC (Fig. 4a, b). The DCA showed that for a given patient, if the threshold probability is between 0.15 and 0.58, then using any of the three models in the current study to predict postoperative complications adds more benefit than the treat-all-patients scheme or the treat-none scheme. Within this range, the net benefit was comparable among the three models, with the best net benefit in the combined model (Fig. 5). The p value for the Hosmer-Lemeshow statistic was 0.884 for the TC and 0.703 for the VC, indicating that the Hosmer-Lemeshow test was not statistically significant and that the model was a good fit.

Fig. 4.

Calibration curves of the nomogram in the training (a) and validation (b) cohorts. Calibration curves depict the calibration of the nomogram in terms of the agreement between the predicted risks of postoperative complications and observed outcomes of postoperative complications. The x-axis represents the nomogram-predicted probability. The y-axis represents the actual postoperative complication rate. The diagonal dotted line represents a perfect prediction by an ideal model. The black dotted line represents the performance of the nomogram, and the black solid line is bias-corrected by bootstrapping (B= 1,000 repetitions). The closer the black dotted and solid lines fit to the diagonal dotted line, the better the prediction.

Fig. 4.

Calibration curves of the nomogram in the training (a) and validation (b) cohorts. Calibration curves depict the calibration of the nomogram in terms of the agreement between the predicted risks of postoperative complications and observed outcomes of postoperative complications. The x-axis represents the nomogram-predicted probability. The y-axis represents the actual postoperative complication rate. The diagonal dotted line represents a perfect prediction by an ideal model. The black dotted line represents the performance of the nomogram, and the black solid line is bias-corrected by bootstrapping (B= 1,000 repetitions). The closer the black dotted and solid lines fit to the diagonal dotted line, the better the prediction.

Close modal
Fig. 5.

Decision curve analysis for the three models. The light gray line represents the assumption that all patients have postoperative complications. The dark black horizontal line represents the assumption that no patients have postoperative complications. The x-axis represents the threshold probability. The y-axis measures the net benefit. The green line represents the radiomics model. The blue line represents the clinical model. The red line represents the combined model. The net benefit was calculated by subtracting the proportion of all patients who tested false positive from the proportion who tested true positive, and weighting by the relative harm of forgoing treatment compared with the negative consequences of an unnecessary treatment. The decision curve showed that for a given patient, if the threshold probability is between 0.15 and 0.58, then using any of the three models in the current study to predict postoperative complications adds more benefit than the treat-all-patients scheme or the treat-none scheme. The net benefit was best in the combined model.

Fig. 5.

Decision curve analysis for the three models. The light gray line represents the assumption that all patients have postoperative complications. The dark black horizontal line represents the assumption that no patients have postoperative complications. The x-axis represents the threshold probability. The y-axis measures the net benefit. The green line represents the radiomics model. The blue line represents the clinical model. The red line represents the combined model. The net benefit was calculated by subtracting the proportion of all patients who tested false positive from the proportion who tested true positive, and weighting by the relative harm of forgoing treatment compared with the negative consequences of an unnecessary treatment. The decision curve showed that for a given patient, if the threshold probability is between 0.15 and 0.58, then using any of the three models in the current study to predict postoperative complications adds more benefit than the treat-all-patients scheme or the treat-none scheme. The net benefit was best in the combined model.

Close modal

In this research, we created and verified a novel, CT-based combined prediction model to improve the assessment of nutritional status and the prediction of postoperative complications in GC patients. The model is a nomogram incorporating the NRS2002 score, sarcopenia, blood loss, and a radiomics signature. Compared to the radiomics signature alone and the clinical model based on clinical features and body composition data, the nomogram provided better predictive accuracy, which shows the incremental value of the combined model for the prediction of postoperative complications in GC. Moreover, by quantifying the prediction model, the current study proved that the nomogram, which enables visualization, was an intuitive and reliable personalized decision-making tool.

Malnutrition frequently occurs in GC patients (incidence, 37–63%), especially in hospitalized patients, as both the disease and its treatments adversely affect the patient’s nutritional status; however, this type of malnutrition is often overlooked or under-treated [20, 21]. In our combined model, the NRS2002 score, sarcopenia, and blood loss were independent clinical predictors of overall postoperative complications in GC, which is consistent with the results of other studies [22-24]. The European Society of Parenteral and Enteral Nutrition recommends the NRS2002 as the instrument of choice for NRS in hospitalized patients [25]. However, NRS2002 has certain limitations since it cannot determine the degree of malnutrition and is not sufficient to assess changes in body composition. We focused on skeletal muscle and adipose tissue in the body composition data and found that sarcopenia was associated with overall postoperative complications, while visceral obesity was associated with postoperative inflammatory complications in GC. Sarcopenia is a condition characterized by a gradual and universal loss of skeletal muscle mass and strength, with or without adipose tissue loss [26]. The impact of body composition on postoperative complications may be explained as follows: first, sarcopenia serves as a marker of malnutrition, which would increase the rate of complications [27]. Second, weight loss in GC patients is primarily associated with sarcopenia; these patients are frail and lack amino acids for tissue repair, and their physiological and repair functions are compromised, leading to an increased incidence of postoperative complications [28, 29]. Third, visceral fat is an important metabolic tissue that secretes pro-inflammatory substances, which are associated with the risk of inflammatory complications. Visceral fat also secretes angiogenic stimulants such as vascular endothelial growth factor and leptin, the continued expression of which can lead to the uncontrolled formation of fragile, disorganized, easy-to-bleed vessels, leading to an increased risk of intraoperative blood loss [30]. Furthermore, excessive intraoperative blood loss can lead to unfavorable postoperative conditions. In patients with sarcopenia, blood loss can further reduce the body’s immunity and ability to fight off cancer cells, leading to a worse postoperative prognosis [31]. For this reason, we should focus on the above clinical risk factors and enhance nutritional support to improve treatment outcomes.

Radiomics has been a research hotspot in recent years, as this technology enables the noninvasive discovery of features that are difficult to identify by visual inspection or conventional imaging tools [32]. On reviewing the reports of radiomics in the prognosis of GC, we found that previous studies primarily focused on long-term prognosis [14, 33], and few reports investigated postoperative complications. Our research combines a radiomics signature and clinical risk factors, which differs from and complements previous studies. Tumor heterogeneity is one of the critical features of GC. Tumor heterogeneity refers to the temporal and spatial heterogeneity of solid cancers at the genetic, cellular, protein, microenvironment, tissue, and organ levels [34]. CT texture analysis is a method for quantifying tumor heterogeneity [35]. The impact of tumor heterogeneity on postoperative complications may be attributed to the following factors: first, GC cells can secrete cytokines such as pro-inflammatory cytokines and tumor necrosis factor to different degrees, leading to accelerated muscle atrophy and lipolysis [20], which increase the incidence of postoperative complications by affecting the nutritional status of the patients. Second, macrophages are significantly correlated with tumor heterogeneity [36], and they may play a direct role in tumor growth through their many functions, such as inflammation, intravascular injection, angiogenesis, and matrix remodeling [36]; this would lead to compromised immune function and an increased incidence of complications. Therefore, CT-based radiomics provides valuable information for the prediction of postoperative complications in GC, which validates our hypothesis.

Overall, the combined model in this study showed optimal predictive performance in both the TC and VC, with AUCs of 0.763 (95% CI: 0.708–0.817) and 0.748 (95% CI: 0.667–0.818), respectively, implying that the radiomics signature has additive value to the clinical model. Similar combined models have been studied for other conditions. For example, Kim et al. [37] found that a combined model based on MRI and clinicopathologic variables had optimal discrimination and calibration ability in predicting pathologic complete response after neoadjuvant chemotherapy for breast cancer. Liang et al. [38] showed that their proposed combined nomogram model may be beneficial in distinguishing between grade 1 and grade 2/3 tumors in individuals with pancreatic neuroendocrine tumors. Combined models have also been investigated in lung cancer, liver cancer, and metastatic tumors, and shown similar results [39-41]. This is further evidence of the good diagnostic efficiency of combined models. Our combined model incorporates more characteristic informative indicators, such as NRS2002 and muscle index, to screen for patients at nutritional risk and further identify the degree of malnutrition in patients, while the radiomics features in our model provide information on the tumor microenvironment. Thus, our combined model has potential clinical applications and higher predictive stability than single-factor models. Finally, to facilitate clinical application, we constructed a nomogram based on the combined model. This nomogram showed good performance and high compatibility. This scoring system was used to generate postoperative complication probabilities, which can help doctors distinguish high- and low-risk patients, and formulate more appropriate treatment plans. For high-risk patients, in particular, a multidisciplinary consultation involving the nutrition, gastrointestinal surgery, medical oncology, and radiology departments can be conducted to formulate an individualized treatment plan. This is conducive to improving the prognosis of patients and is in line with the present trend of individualized precision medicine.

The current research has some shortcomings. First, this research is a single-center retrospective evaluation that might be subject to selection bias and requires multi-center validation to extend the generalizability of the findings. Second, the use of various scanners from different manufacturers as well as varying acquisition parameters may have had an effect on the radiomics variables. Third, the tumor biology might not be fully characterized by segmenting the ROI on the largest cross section rather than on the whole tumor volume.

In conclusion, this study developed a new combined prediction model for postoperative complications in GC, mainly from a nutritional and imaging perspective. The model takes into account the NRS2002 score, sarcopenia, blood loss, and a radiomics signature. The constructed nomogram visualizes risk factors and is an objective individualized predictor with potential applications in assisting clinical decisions and enabling precision medicine.

We thank Shunzu Lu for analyzing the cross-sectional CT images at the third lumbar vertebra (L3) to obtain the body composition data. We thank Medjaden Inc. for scientific editing of this manuscript. And we would like to thank all of researchers and study participants for their contributions.

The Ethics Committee of the First Affiliated Hospital of Guangxi Medical University approved the current retrospective investigation (approval number: 2022-KY-E-183), and the requirement for written informed consent was waived.

The authors have no conflicts of interest to declare.

This article was supported by a grant from National Natural Science Foundation of China (No. 82060310) and the Research Project of Guangxi Health Committee (Z20180916).

The authors’ responsibilities were as follows: Qiaoqing Lan and Xuechun Guan contributed equally to this article, designed the research, and drafted the article; Qiaoqing Lan, Shunzu Lu, and Zijian Jiang collected the data; Qiaoqing Lan and Huashan Lin analyzed the data; Liling Long and Wenzhao Yuan revised the paper; and all authors read and approved the final manuscript. The authors report that they have no conflicts of interest.

All the data (pooled hazard ratios with 95% CIs of odds ratio, body composition data, or radiomic features) used to support the findings of this study are included within the article and its online supplementary material. Please contact author Qiaoqing Lan for data requests.

1.
Sung
H
,
Ferlay
J
,
Siegel
RL
,
Laversanne
M
,
Soerjomataram
I
,
Jemal
A
,
Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries
.
CA Cancer J Clin
.
2021 May
;
71
(
3
):
209
49
. .
2.
Smyth
EC
,
Nilsson
M
,
Grabsch
HI
,
van Grieken
NC
,
Lordick
F
.
Gastric cancer
.
Lancet
.
2020 Aug
;
396
(
10251
):
635
48
. .
3.
Yuan
P
,
Wu
Z
,
Li
Z
,
Bu
Z
,
Wu
A
,
Wu
X
,
Impact of postoperative major complications on long-term survival after radical resection of gastric cancer
.
BMC Cancer
.
2019 Aug
;
19
(
1
):
833
. .
4.
Dercle
L
,
Ammari
S
,
Bateson
M
,
Durand
PB
,
Haspinger
E
,
Massard
C
,
Limits of radiomic-based entropy as a surrogate of tumor heterogeneity: ROI-area, acquisition protocol and tissue site exert substantial influence
.
Sci Rep
.
2017 Aug
;
7
(
1
):
7952
. .
5.
Dong
QT
,
Cai
HY
,
Zhang
Z
,
Zou
HB
,
Dong
WX
,
Wang
WB
,
Influence of body composition, muscle strength, and physical performance on the postoperative complications and survival after radical gastrectomy for gastric cancer: a comprehensive analysis from a large-scale prospective study
.
Clin Nutr
.
2021 May
;
40
(
5
):
3360
9
. .
6.
Arends
J
,
Bachmann
P
,
Baracos
V
,
Barthelemy
N
,
Bertz
H
,
Bozzetti
F
,
ESPEN guidelines on nutrition in cancer patients
.
Clin Nutr
.
2017 Feb
;
36
(
1
):
11
48
. .
7.
Lim
HS
,
Lee
B
,
Cho
I
,
Cho
GS
.
Nutritional and clinical factors affecting weight and fat-free mass loss after gastrectomy in patients with gastric cancer
.
Nutrients
.
2020 Jun
;
12
(
7
):
1905
. .
8.
Xu
J
,
Cao
J
,
Wang
Y
,
Yao
X
,
Wang
Y
,
He
Z
,
Novel preoperative nutritional assessment tool and prognostic model for ESCC patients
.
J Cancer
.
2019 Jun
;
10
(
17
):
3883
92
. .
9.
Chen
XY
,
Zhang
XZ
,
Ma
BW
,
Li
B
,
Zhou
DL
,
Liu
ZC
,
A comparison of four common malnutrition risk screening tools for detecting cachexia in patients with curable gastric cancer
.
Nutrition
.
2020 Feb
;
70
:
110498
. .
10.
White
JV
,
Guenter
P
,
Jensen
G
,
Malone
A
,
Schofield
M
;
A
cadem
y Malnutrition Work Group
,
Consensus statement: Academy of Nutrition and Dietetics and American Society for Parenteral and Enteral Nutrition – characteristics recommended for the identification and documentation of adult malnutrition (undernutrition)
.
J Parenter Enteral Nutr
.
2012 May
;
36
(
3
):
275
83
. .
11.
Magudia
K
,
Bridge
CP
,
Bay
CP
,
Babic
A
,
Fintelmann
FJ
,
Troschel
FM
,
Population-scale CT-based body composition analysis of a large outpatient population using deep learning to derive age-sex-and race-specific reference curves
.
Radiology
.
2021 Feb
;
298
(
2
):
319
29
. .
12.
Wang
Y
,
Liu
W
,
Yu
Y
,
Liu
JJ
,
Xue
HD
,
Qi
YF
,
CT radiomics nomogram for the preoperative prediction of lymph node metastasis in gastric cancer
.
Eur Radiol
.
2020 Feb
;
30
(
2
):
976
86
. .
13.
Huang
J
,
Chen
Y
,
Zhang
Y
,
Xie
J
,
Liang
Y
,
Yuan
W
,
Comparison of clinical-computed tomography model with 2D and 3D radiomics models to predict occult peritoneal metastases in advanced gastric cancer
.
Abdom Radiol
.
2022 Jan
;
47
(
1
):
66
75
. .
14.
Li
W
,
Zhang
L
,
Tian
C
,
Song
H
,
Fang
M
,
Hu
C
,
Prognostic value of computed tomography radiomics features in patients with gastric cancer following curative resection
.
Eur Radiol
.
2019 Jun
;
29
(
6
):
3079
89
. .
15.
Liu
S
,
Liu
S
,
Ji
C
,
Zheng
H
,
Pan
X
,
Zhang
Y
,
Application of CT texture analysis in predicting histopathological characteristics of gastric cancers
.
Eur Radiol
.
2017 Dec
;
27
(
12
):
4951
9
. .
16.
Wang
Y
,
Liu
W
,
Yu
Y
,
Han
W
,
Liu
JJ
,
Xue
HD
,
Potential value of CT radiomics in the distinction of intestinal-type gastric adenocarcinomas
.
Eur Radiol
.
2020 May
;
30
(
5
):
2934
44
. .
17.
Dindo
D
,
Demartines
N
,
Clavien
PA
.
Classification of surgical complications: a new proposal with evaluation in a cohort of 6336 patients and results of a survey
.
Ann Surg
.
2004 Aug
;
240
(
2
):
205
13
. .
18.
Zhuang
CL
,
Huang
DD
,
Pang
WY
,
Zhou
CJ
,
Wang
SL
,
Lou
N
,
Sarcopenia is an independent predictor of severe postoperative complications and long-term survival after radical gastrectomy for gastric cancer: analysis from a large-scale cohort
.
Medicine
.
2016 Mar
;
95
(
13
):
e3164
. .
19.
Fujiwara
N
,
Nakagawa
H
,
Kudo
Y
,
Tateishi
R
,
Taguri
M
,
Watadani
T
,
Sarcopenia, intramuscular fat deposition, and visceral adiposity independently predict the outcomes of hepatocellular carcinoma
.
J Hepatol
.
2015 Jul
;
63
(
1
):
131
40
. .
20.
Arends
J
,
Baracos
V
,
Bertz
H
,
Bozzetti
F
,
Calder
PC
,
Deutz
NEP
,
ESPEN expert group recommendations for action against cancer-related malnutrition
.
Clin Nutr
.
2017 Oct
;
36
(
5
):
1187
96
. .
21.
Grace
EM
,
Shaw
C
,
Lalji
A
,
Mohammed
K
,
Andreyev
HJN
,
Whelan
K
.
Nutritional status, the development and persistence of malnutrition and dietary intake in oesophago-gastric cancer: a longitudinal cohort study
.
J Hum Nutr Diet
.
2018 Dec
;
31
(
6
):
785
92
. .
22.
Ito
Y
,
Kanda
M
,
Ito
S
,
Mochizuki
Y
,
Teramoto
H
,
Ishigure
K
,
Intraoperative blood loss is associated with shortened postoperative survival of patients with stage II/III gastric cancer: analysis of a multi-institutional dataset
.
World J Surg
.
2019 Mar
;
43
(
3
):
870
7
. .
23.
Shi
B
,
Liu
S
,
Chen
J
,
Liu
J
,
Luo
Y
,
Long
L
,
Sarcopenia is associated with perioperative outcomes in gastric cancer patients undergoing gastrectomy
.
Ann Nutr Metab
.
2019 Dec
;
75
(
4
):
213
22
. .
24.
Hersberger
L
,
Bargetzi
L
,
Bargetzi
A
,
Tribolet
P
,
Fehr
R
,
Baechli
V
,
Nutritional risk screening (NRS 2002) is a strong and modifiable predictor risk score for short-term and long-term clinical outcomes: secondary analysis of a prospective randomised trial
.
Clin Nutr
.
2020 Sep
;
39
(
9
):
2720
9
. .
25.
Kondrup
J
,
Allison
SP
,
Elia
M
,
Vellas
B
,
Plauth
M
;
Educational and Clinical Practice Committe
,
European Society of Parenteral and Enteral Nutrition ESPEN
.
ESPEN guidelines for nutrition screening 2002
.
Clin Nutr
.
2003 Aug
;
22
(
4
):
415
21
. .
26.
Cruz-Jentoft
AJ
,
Baeyens
JP
,
Bauer
JM
,
Boirie
Y
,
Cederholm
T
,
Landi
F
,
Sarcopenia: European consensus on definition and diagnosis: report of the European Working Group on Sarcopenia in Older People
.
Age Ageing
.
2010 Jul
;
39
(
4
):
412
23
. .
27.
Kuwada
K
,
Kuroda
S
,
Kikuchi
S
,
Yoshida
R
,
Nishizaki
M
,
Kagawa
S
,
Clinical impact of sarcopenia on gastric cancer
.
Anticancer Res
.
2019 May
;
39
(
5
):
2241
9
. .
28.
Prado
CMM
,
Heymsfield
SB
.
Lean tissue imaging: a new era for nutritional assessment and intervention
.
J Parenter Enteral Nutr
.
2014 Nov
;
38
(
8
):
940
53
. .
29.
Tamura
T
,
Sakurai
K
,
Nambara
M
,
Miki
Y
,
Toyokawa
T
,
Kubo
N
,
Adverse effects of preoperative sarcopenia on postoperative complications of patients with gastric cancer
.
Anticancer Res
.
2019 Feb
;
39
(
2
):
987
92
. .
30.
Macshut
M
,
Kaido
T
,
Yao
S
,
Miyachi
Y
,
Sharshar
M
,
Iwamura
S
,
Visceral adiposity is an independent risk factor for high intra-operative blood loss during living-donor liver transplantation; could preoperative rehabilitation and nutritional therapy mitigate that risk?
Clin Nutr
.
2021 Mar
;
40
(
3
):
956
65
. .
31.
Nakanishi
K
,
Kanda
M
,
Kodera
Y
.
Long-lasting discussion: adverse effects of intraoperative blood loss and allogeneic transfusion on prognosis of patients with gastric cancer
.
World J Gastroenterol
.
2019 Jun
;
25
(
22
):
2743
51
. .
32.
Aerts
HJWL
,
Velazquez
ER
,
Leijenaar
RTH
,
Parmar
C
,
Grossmann
P
,
Carvalho
S
,
Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach
.
Nat Commun
.
2014 Jun
;
5
:
4006
. .
33.
Gao
X
,
Ma
T
,
Bai
S
,
Liu
Y
,
Zhang
Y
,
Wu
Y
,
A CT-based radiomics signature for evaluating tumor infiltrating Treg cells and outcome prediction of gastric cancer
.
Ann Transl Med
.
2020 Apr
;
8
(
7
):
469
. .
34.
Lambin
P
,
Rios-Velazquez
E
,
Leijenaar
R
,
Carvalho
S
,
van Stiphout
RGPM
,
Granton
P
,
Radiomics: extracting more information from medical images using advanced feature analysis
.
Eur J Cancer
.
2012 Mar
;
48
(
4
):
441
6
. .
35.
Ganeshan
B
,
Goh
V
,
Mandeville
HC
,
Ng
QS
,
Hoskin
PJ
,
Miles
KA
.
Non-small cell lung cancer: histopathologic correlates for texture parameters at CT
.
Radiology
.
2013 Jan
;
266
(
1
):
326
36
. .
36.
Goto
K
,
Honda
Y
,
Ikeda
K
,
Takemoto
K
,
Higaki
T
,
Hayashi
T
,
Tumor heterogeneity evaluated by computed tomography detects muscle-invasive upper tract urothelial carcinoma that is associated with inflammatory tumor microenvironment
.
Sci Rep
.
2021 Jul
;
11
(
1
):
14251
. .
37.
Kim
SY
,
Cho
N
,
Choi
Y
,
Lee
SH
,
Ha
SM
,
Kim
ES
,
Factors affecting pathologic complete response following neoadjuvant chemotherapy in breast cancer: development and validation of a predictive nomogram
.
Radiology
.
2021 May
;
299
(
2
):
290
300
. .
38.
Liang
W
,
Yang
P
,
Huang
R
,
Xu
L
,
Wang
J
,
Liu
W
,
A combined nomogram model to preoperatively predict histologic grade in pancreatic neuroendocrine tumors
.
Clin Cancer Res
.
2019 Jan
;
25
(
2
):
584
94
. .
39.
Sun
Y
,
Li
C
,
Jin
L
,
Gao
P
,
Zhao
W
,
Ma
W
,
Radiomics for lung adenocarcinoma manifesting as pure ground-glass nodules: invasive prediction
.
Eur Radiol
.
2020 Jul
;
30
(
7
):
3650
9
. .
40.
Serenari
M
,
Han
KH
,
Ravaioli
F
,
Kim
SU
,
Cucchetti
A
,
Han
DH
,
A nomogram based on liver stiffness predicts postoperative complications in patients with hepatocellular carcinoma
.
J Hepatol
.
2020 Oct
;
73
(
4
):
855
62
. .
41.
Zhang
W
,
Mao
N
,
Wang
Y
,
Xie
H
,
Duan
S
,
Zhang
X
,
A Radiomics nomogram for predicting bone metastasis in newly diagnosed prostate cancer patients
.
Eur J Radiol
.
2020 Jul
;
128
:
109020
. .

Additional information

Qiaoqing Lan and Xuechun Guan contributed equally to this work.

Open Access License / Drug Dosage / Disclaimer
This article is licensed under the Creative Commons Attribution 4.0 International License (CC BY). Usage, derivative works and distribution are permitted provided that proper credit is given to the author and the original publisher.Drug Dosage: The authors and the publisher have exerted every effort to ensure that drug selection and dosage set forth in this text are in accord with current recommendations and practice at the time of publication. However, in view of ongoing research, changes in government regulations, and the constant flow of information relating to drug therapy and drug reactions, the reader is urged to check the package insert for each drug for any changes in indications and dosage and for added warnings and precautions. This is particularly important when the recommended agent is a new and/or infrequently employed drug. Disclaimer: The statements, opinions and data contained in this publication are solely those of the individual authors and contributors and not of the publishers and the editor(s). The appearance of advertisements or/and product references in the publication is not a warranty, endorsement, or approval of the products or services advertised or of their effectiveness, quality or safety. The publisher and the editor(s) disclaim responsibility for any injury to persons or property resulting from any ideas, methods, instructions or products referred to in the content or advertisements.