CT Body Composition Enhances Survival Risk Stratification

Background and study aims:

Gastric carcinoma remains the fifth most common malignancy and the second leading cause of cancer-related mortality worldwide. For patients with locally advanced disease, standard treatment includes radical gastrectomy with D2 lymphadenectomy followed by (neo)adjuvant chemotherapy and, more recently, immune checkpoint inhibitors (ICIs). Despite these therapeutic advances, outcomes for stage III gastric cancer remain poor, with 5-year overall survival rates below 40%. Considerable variability in prognosis persists even within the same the American Joint Committee on Cancer (AJCC) substage, highlighting the importance of host-related factors such as nutritional status, systemic inflammation, and immune competence in shaping survival.

Computed tomography-based body composition (CTBC) analysis offers an objective means to quantify skeletal muscle, subcutaneous adipose tissue (SAT), and visceral adipose tissue (VAT), capturing key dimensions of patient physiology that are not accounted for in traditional staging systems. However, its prognostic value in gastric cancer has been inconsistently reported.

Manual CTBC segmentation is time-consuming and prone to inter-observer variation, limiting its routinely clinical applicability. Advances in deep learning, particularly U-Net-based architectures, enable rapid, automated body composition analysis with high concordance to expert annotations. This innovation greatly supports scalable investigations into the prognostic relevance of adipose distribution in cancer.

Here, the investigators prepare to apply automated CTBC analysis to a homogeneous cohort of 300 patients with AJCC8 stage III gastric cancer to determine whether visceral adiposity-related metrics improve survival risk stratification beyond TNM staging. To clarify the underlying mechanisms, the investigators further integrated plasma metabolomic profiling and tumor immune-metabolic phenotyping. This comprehensive approach aims to delineate the systemic and tumor-intrinsic consequences of fat distribution and identify potential metabolic or immunologic vulnerabilities relevant to patient stratification and therapy.

Patients and Methods:

This study was conducted under the approval of the institutional review board of Chang Gung Memorial Hospital at Linkou, Taiwan (IRB 202301368B0). The investigators retrospectively reviewed patients with gastric carcinoma who had undergone curative-intent gastrectomy from 2007 to 2022 at Department of Surgery of CGMH. Inclusion criteria included 1) patients with pathological stages III gastric cancer according to the 8th edition of the AJCC Staging System, and 2) patients had received high-quality baseline CT as work-up for clinical staging within one month before the definite surgery. A total of 227 patients eligible enrolled into this study. Clinicopathological data were collected based on our prospectively-constructed database, including body mass index (BMI, kg/m^2). Metastatic lymph nodes retrieved ratio (LN ratio) was defined as the ratio between metastatic lymph nodes and the total number of lymph nodes examined. Disease-free survival (DFS) was measured from the date of radical gastrectomy until disease recurrence, while overall survival (OS) was measured from the date of radical gastrectomy until death. The last outcome review was set on Dec 31, 2024. The median follow-up was 33.1 months (range, 5.1 to 189.8 months).

CTBC analysis through manual and automated segmentation approach Body composition parameters were acquired form non-contrast CT scans of the lower thorax to the pelvis with 5mm slice thickness. Three consecutive slices at lumbar 3 vertebral body level were included and measurements averaged over these three images. For manual segmentation and following analyses, the detailed procedures had been described elsewhere.12 For automated body composition segmentation, the investigators used UNet++ as the architecture of the muscle and adipose tissue segmentation models.15 The investigators trained five segmentation models for abdominal muscles, paraspinal muscles, psoas muscle, SAT, and VAT. Abdominal muscle included psoas, erector, quadratus lumborum, rectus abdominis, external and internal oblique, and transverse abdominis muscles. Abdominal fat tissue was divided into SAT and VAT for separate analysis. To mitigate the variation of body size, normalizing measures of abdominal muscle and fat area to patient stature was performed by dividing the cross-sectional area by the patient's height (meter)2. Skeletal muscle index (SMI), SAT index and VAT index were calculated as such. VAT-to-SAT ratio was acquired by VAT index divided by SAT index.

Measurement of plasma metabolites EDTA-plasma samples stored at -800 C were subjected to metabolomic analysis using the Biocrates AbsoluteIDQ® p180 kit (Biocreates Life Sciences AG, Innsbruck, Austria). This kit is designed to determine the levels of amino acids (n=21), biogenic amines (n=21), monosaccharides (n=1), acylcarnitines (n=40), glycerophospholipids (n=90) and sphingomyelins (n=15) of body fluid.

Immunohistochemical (IHC) staining Formalin fixed and paraffin embedded tissues of surgical specimen were cut into 4-µm sections and mounted on the glue-coated slides. A modified avidin-biotin- peroxidase complex IHC method was performed. Primary antibodies against palmitoyltransferase (CPT)1 (Abcam, ab128568, UK), CPT2 (Abcam, ab181114), carnitine-acylcarnitine translocase (CACT) (Thermo Fisher, 19363-1-AP, US), indoleamine 2,3-dioxygenase 1 (IDO1) (OriGene, TA506402, US) and aryl hydrocarbon receptor (AHR) (Cell Signaling, 83200, US) were applied. Semi-quantification of targets was performed using a scale of - (negative), + (weakly positive), ++ (moderately positive), and +++ (strongly positive), taking account the percentage of stained cells and the intensity of staining. For statistical clarity, those with - and + were designated as immuno-negative, while those with ++ and +++ were designated as immuno-positive.

Immunoblot Analysis Proteins were transferred onto a PVDF Transfer Membrane (Millipore). The membranes were blocked with 3% skim milk in TBST buffer (100 mM Tris-HCl (pH 7.6), 150 mM NaCl, and 0.05% Tween 20) at RT for 1 h. Protein blots were incubated with primary antibodies including CPT1 (1/1000, Abcam), CPT2 (1/1000, Abcam), CACT (1/1000, Thermo Fisher), IDO1 (1/1000, Origene), AHR (1/1000, Cell signaling), and GAPDH (1/3000, Abcam).

Multi-parameter flow cytometer A 35-color multi-parameter flow cytometer (BD Bioscience) panel was employed to analyze the expression of immune markers in the collected frozen-stored tumor samples retrieved during the surgeries. The detailed immune subsets and executing protocol were described elsewhere.16 Flow cytometric analysis was performed using a FACSymphony A5.2 instrument (BD Bioscience) to acquire the data. The collected data were subsequently analyzed using FlowJo software (BD Bioscience) and the CATALYST package in R.

Statistical Analysis Numerical data were expressed as mean ± standard deviation (SD), median [interquartile range (IQR), Q3-Q1], or frequency (%), as appropriate. Group differences were assessed using Fisher's exact test or Pearson's chi-square test. Optimal cutoff values for the LNR were determined using survival tree analysis in R software (version 4.0). Cutoff points for CTBC metrics were identified using the coxph function from the survival package and the cutp function from the survMisc package. Associations between clinicopathological variables or CTBC parameters and DFS or OS were analyzed using logistic regression. To evaluate the added prognostic value of CTBC metrics, models with and without CTBC parameters were compared using the likelihood ratio test (LRT) and Akaike's information criterion (AIC), with a ΔAIC >4 considered strong evidence of model improvement. Kaplan-Meier survival curves with log-rank tests were used to compare DFS and OS. For flow cytometry analysis, cohort data were sorted in descending order by each CTBC metric. Thresholds were defined to classify patients into high and low groups, with the high group threshold set at twice that of the low group. A two-tailed p value < 0.05 was considered statistically significant. All statistical analyses were performed using SPSS version 20.0 (SPSS Inc., Chicago, IL, USA) and R (R Core Team, 2024).