Development and validation of a novel modified cancer cachexia index in patients with locally advanced gastric cancer undergoing neoadjuvant chemotherapy: a multicenter cohort study

Abstract

Background:

Although the cachexia index (CXI) is a well-established prognostic predictor in gastric cancer (GC) patients, its effectiveness in patients with locally advanced gastric cancer (LAGC) who have undergone neoadjuvant chemotherapy (NACT) remains unclear.

Methods:

This multicenter study included 600 LAGC patients treated with NACT from January 2010 to June 2022. A modified CXI was constructed based on Random Forest model, calculated as (post-NACT subcutaneous adipose tissue area at L3) × (post-NACT serum albumin)/(post-NACT platelet count). Patients were categorized into mCXI-low and mCXI-high.

Results:

In the training cohort, mCXI outperformed the traditional CXI in predicting of overall survival (OS) and tumor regression grades. The mCXI-high group had a significantly higher 3-year OS (73.0% vs. 58.9%, P = 0.002), recurrence-free survival (67.7% vs. 50.2%, P = 0.002), and disease-specific survival (74.4% vs. 62.5%, P = 0.012). Multivariate analysis confirmed that mCXI as an independent prognostic factor. The recurrence rate was significantly lower in the mCXI-high group (33.0% vs. 52.6%; P < 0.001). The mCXI-high group also had a lower recurrence rate (33.0% vs. 52.6%, P < 0.001) and a delayed recurrence peak (33.51 vs. 7.11 months). Similar results were obtained in the validation cohort. Further analysis showed that in mCXI-low patients with ypStage III disease, receiving more than 4 cycles of adjuvant chemotherapy (AC) significantly improved survival (3-year OS: 43.7% vs. 25.0%, P = 0.007). In mCXI-high patients, 4–6 AC cycles yielded optimal outcomes.

Conclusions:

mCXI was associated with the overall prognosis in patients with LAGC underwent NACT, is superior to traditional CXI, and may serve as a decision-making tool for guiding personalized postoperative AC.

Introduction

According to the most recent global cancer statistics for 2022, gastric cancer (GC), which has the fifth-highest incidence and mortality rates worldwide, poses a substantial global health burden^[1,2]. Radical surgical resection remains the only effective curative approach for patients with locally advanced gastric cancer (LAGC). However, even after radical resection, these patients face a considerable risk of postoperative recurrence. With the completion of the European MAGIC^[3] and FLOT4^[4] trials, it has been confirmed that perioperative chemotherapy combined with surgery improves overall survival and progression-free survival in patients with LAGC. In recent years, the RESOLVE study^[5] and the PRODIGY phase III trial^[6] conducted by Asian researchers have demonstrated that adding NACT before surgery enhances the prognostic benefits for patients with LAGC in Asia. Based on these results, perioperative chemotherapy has become the standard treatment for LAGC. This study follows the TITAN 2025 guidelines for transparent reporting of AI use in medical research^[7]

Recently, Jafri et al^[8] developed a cachexia index (CXI; CXI = SMI (cm²/m²) × ALB (g/L)/NLR) based on the clinical features of cachexia to evaluate the prognosis of patients with non-small cell lung cancer. Their research showed that the CXI is a useful predictor of cachexia risk in patients with small-cell lung cancer, aiding the prediction of survival and treatment response. Furthermore, numerous studies^[9-12] have demonstrated that the CXI has a significant prognostic value for patients with gastric malignancies who have not undergone NACT. The CXI comprises serum albumin (ALB) levels, neutrophil-to-lymphocyte ratio (NLR), and skeletal muscle index (SMI). Serum albumin is an indicator of disease severity and progression; low levels are associated with increased disease severity, higher progression risk, and reduced survival rates. The NLR is a simple and effective systemic inflammatory marker for assessing cancer prognosis. Thus, CXI integrates multiple assessments of nutrition, inflammation, and body composition, making it a promising predictor of cancer cachexia and prognosis.

Although patients with gastric cancer undergoing surgery alone often present varying degrees of cachexia at diagnosis, those receiving neoadjuvant therapy experience different physiological changes owing to advanced disease progression and the effects of adjuvant chemotherapy (AC), which distinguishes them from patients undergoing surgery without prior treatment. Therefore, traditional CXI may not be suitable for such patients. Gomes da Rocha et al^[13] indicate that chemotherapy can exacerbate cachexia, which in turn increases chemotherapy toxicity and affects patient tolerance to both chemotherapy and surgery. However, after NACT, cachexia may improve with changes in the tumor burden. Currently, there are few reports on the impact of changes in cachexia status following NACT on patient prognosis.

Therefore, this study aimed to explore the application value of the CXI in the prognostic assessment of patients undergoing NACT using multicenter data. Modifying the traditional CXI to better fit specific characteristics of patients with LAGC treated with NACT. Through this study, we hope the modified cachexia index to enhance the effectiveness of perioperative chemotherapy in the treatment of LAGC and ultimately improve overall patient outcomes.

Materials and methods

Study design and participants

Data were collected from 600 consecutive patients diagnosed with locally advanced gastric cancer (cT2-4NanyM0) between 1 January 2010, and 30 June 2022, identified at two large tertiary hospitals. The inclusion criteria for this study were as follows: (1) locally advanced gastric cancer (cT2-4NxM0) with clinical staging before neoadjuvant chemotherapy, (2) no history of other malignant tumors, (3) no evidence of distant metastasis or invasion of adjacent organs, and (4) patients who underwent radical gastrectomy following neoadjuvant chemotherapy. The exclusion criteria were as follows: (1) history of gastric surgery; (2) acute cardiovascular disease (such as cerebrovascular or coronary artery injury) within the past three months; (3) emergency surgery; (4) active infection or severe systemic immunodeficiency; and (5) incomplete serological, computed tomography, or anthropometric data. A total of 348 patients from Fujian Medical University Union Hospital (FJMUUH) were assigned to the internal training cohort, and 49 patients from Zhangzhou Affiliated Hospital of Fujian Medical University (ZZAFJMU) were designated as the external validation cohort. A treatment flowchart is shown in Figure S1 (http://links.lww.com/JS9/E389). This study has been reported in accordance with the STROCSS criteria^[14].

HIGHLIGHTS

• The cachexia index (CXI), its effectiveness in patients with locally advanced gastric cancer (LAGC) who have undergone neoadjuvant chemotherapy (NACT) remains unclear.

• Analysis of 397 LAGC patients showed that mCXI outperformed traditional CXI in predicting overall survival (OS) and TRG grade. The mCXI-high group had significantly better 3-year OS (73.0% vs. 58.9%), RFS (67.7% vs. 50.2%), and DSS (74.4% vs. 62.5%). mCXI was identified as an independent prognostic factor. The different mCXI groups can selectively determine how many cycles of adjuvant chemotherapy a patient needs postoperatively.

• mCXI is a more accurate prognostic tool than traditional CXI and can guide postoperative chemotherapy decisions, improving survival outcomes.

All procedures were in accordance with the ethical standards of the responsible committee on human experimentation (institutional and national) and with the Helsinki Declaration of 1964 and later versions. Informed consent or substitute for it was obtained from all patients for being included in the study. This study was approved by the Institutional Review Board of Fujian Medical University Union Hospital and Zhangzhou Affiliated Hospital of Fujian Medical University. This study is registered at ClinicalTrials.gov (Identifier: NCT06689462).

Perioperative treatment and surgical procedures

All participants in this investigation were subjected to a minimum of two cycles of NACT. Given the diversity of perioperative chemotherapy regimens among patients, we categorized the specific chemotherapy protocols into the following groups: 1. Platinum-based regimens (platinum + capecitabine/platinum + S-1/platinum + 5-FU); 2. paclitaxel regimens (paclitaxel + capecitabine/paclitaxel + S-1/paclitaxel + 5-FU); and 3. Other regimens (paclitaxel + platinum/single agent S-1)^[15]. Postoperative adjuvant chemotherapy is recommended for patients with LAGC who have a stable general condition and preserved organ function. The doses of all administered medications were determined based on instructions provided by the drug manufacturers, guidelines, and the patient’s body surface area. Experienced oncologists at each center adjusted the chemotherapy regimen based on the tumor response and treatment toxicity.

Data collection and definition

All patients who underwent hematological tests and CT images before the initiation of neoadjuvant chemotherapy and the last test before surgery were included.

Tumor regression following neoadjuvant treatment was evaluated by professional radiologists at each center using the RECIST criteria 1.1^[16]. Concurrently, senior pathologists at each center assessed the pathological response of the primary tumor based on the Becker TRG standard, categorized as follows: TRG 0 (complete response, no residual tumor cells), TRG 1 (near-complete response, only single cells or small clusters of cancer cells), TRG 2 (partial response, residual tumor cells remaining), and TRG 3 (poor response, minimal or no tumor cell death). The T and N stages were defined according to the 8th edition of the American Joint Committee on Cancer (AJCC) TNM staging system^[17].

Measurement of the nutrition, body composition and inflammatory indicators

Body composition was measured using axial CT images at the L3 vertebral level obtained from the Picture Archiving and Communication System in Digital Imaging and Communication in Medicine format for all patients. A single CT image of the third lumbar vertebra (L3) was selected to quantify muscle and adipose tissue characteristics because this anatomical location is strongly associated with whole-body volume^[18,19]. A single trained investigator analyzed all CT images using Slice-O-Matic software (version 5.0; Fast Vision, Montreal, Canada)^[20]. The software precisely identified and quantified the cross-sectional areas of skeletal muscle (29–150 HU), visceral adipose tissue (−150 to −50 HU), and subcutaneous adipose tissue (−190 to −30 HU) using standardized Hounsfield unit (HU) ranges. Additionally, the skeletal muscle area (SMA, cm²), subcutaneous adipose tissue area (SAT, cm²), and visceral adipose tissue area (VAT, cm²) were measured (Figure S2, http://links.lww.com/JS9/E389). SMI = SMA (cm²)/height²(m²).

The NLR(neutrophil to lymphocyte ratio), LMR(lymphocyte to monocyte ratio), PLR(platelet to lymphocyte ratio), SII(systemic immune inflammation index), and SIRI(system inflammation response index), were calculated using the following formulas: NLR = neutrophil count/lymphocyte count; LMR = lymphocyte count/monocyte count; PLR = platelet count/lymphocyte count; SII = (neutrophil count × platelet count/lymphocyte count; and SIRI = (neutrophil count × monocyte count/lymphocyte count^[21-23] (Table S6, http://links.lww.com/JS9/E388).

The calculation formulas for all corresponding dynamic change indicators (Δ) are as follows: (post-NACT indicator − pre-NACT indicator)/pre-NACT indicator^[15].

Method for constructing the modified cachexia index formula

The selected indicators in each formula were defined based on their positive or negative predictive impact on the endpoint (survival). For example, higher levels of albumin (ALB), subcutaneous adipose tissue (SAT), visceral adipose tissue (VAT), the lymphocyte-to-monocyte ratio (LMR), and lymphocytes (LC) are associated with better patient prognosis. To amplify their predictive power, these indicators were placed in the numerator of the formula. Conversely, fibrinogen (FIB) and platelets (PLT), which are associated with poorer prognosis when elevated, were placed in the denominator to enhance the overall predictive efficiency. As the values of the positive predictors (numerator) increase and the negative predictors (denominator) decrease, the resulting mCXI value increases, indicating a better nutritional status and stronger predictive power for the endpoint. Conversely, when the positive predictors decrease and the negative predictors increase, the mCXI value decreases, indicating poorer nutritional status and weaker predictive power. The formulation used in this study aligns with the conceptual framework of traditional CXI indicators.

Modified cachexia index and traditional cachexia index

The random forest model (Supplementary Statistical Material 1, http://links.lww.com/JS9/E387) was used to rank the importance of inflammatory, nutritional, and body composition indicators at different time points. Modified cachexia index were developed based on the dynamic changes in inflammation, nutritional status, and body composition before, during and after neoadjuvant chemotherapy (Fig. 1A–C). The formulae are as follows:

Figure 1.

Selection of inflammation indicators, nutrition indicators, and body composition using the random forest model.

This figure presents the importance ranking of inflammatory indicators, nutritional indicators, and body composition at different time points using random forests. Panel A shows the selection of pre-neoadjuvant indicators, panel B shows the selection of post-neoadjuvant indicators, and panel C show the selection of dynamic changes in indicators during neoadjuvant chemotherapy. The formulas are as follows: pre-NACT mCXI = (pre-SAT × pre-LMR)/pre-FIB; post-NACT mCXI = (post-SAT × post-ALB)/post-PLT; Δ-NACT mCXI = ΔVAT × ΔALB × ΔLC and traditional CXI = (SMI × ALB)/ NLR.

Pre-NACT CXI = pre-SAT (cm²) × pre-LMR/pre-FIB (g/L)

Post-NACT CXI = post-SAT (cm²) × post-ALB (g/L)/post-PLT

Dynamic change during NACT: ΔCXI = ΔVAT × ΔALB × ΔLC

The traditional cachexia index was calculated as follows:

Traditional CXI = SMI (cm²/m²) × ALB (g/L)/NLR^[9,24]

Primary and secondary outcome

The primary outcome is overall survival (OS), and the secondary outcomes were recurrence-free survival (RFS) and disease-specific survival (DSS). OS was defined as the time from the end of surgery to death. RFS was defined as the time from the end of surgery to the first recorded instance of cancer recurrence, metastasis, or occurrence of a new cancer. DSS was defined as the time from the end of surgery to death, specifically due to tumor recurrence or metastasis.

Local-regional recurrence refers to tumors clinically confirmed to have recurred in the residual stomach, anastomosis, or surgical site. Multisite recurrence involves simultaneous recurrence at two or more metastatic sites such as the peritoneum, liver, lungs, bones, brain, distant lymph nodes, or other hematogenous metastatic sites. Other site recurrences include those occurring in locations other than the liver, such as the lungs, bones, brain, pancreas, ovaries, adrenal glands, and rectum. Patients with lymph node recurrence or recurrence at unspecified locations were also included.

Postoperative follow-up strategy

The outpatient review included physical examination, laboratory evaluation, chest radiography, abdominal computed tomography, and ultrasound. Additionally, an annual endoscopy is advised to evaluate the anastomosis site and residual stomach, with a biopsy performed as needed to exclude local recurrence. Patients were followed up once every 3–6 months within the first 2 years, every 6–12 months within the next 3–5 years, and once every year after 5 years. The time of follow-up evaluation could be advanced, and the frequency could be increased depending on the patients’ specific situation.

Statistical analyses

Continuous variables were presented as mean ± standard deviation (SD) if they followed a normal distribution; otherwise, they were reported as median (interquartile range, IQR). Chi-square tests or Fisher’s exact tests were used to analyze categorical variables. The randomForest package in R (version 4.3.2) was employed to evaluate the predictive importance of variables for overall survival (OS). The model was configured with parameters ntree = 1000 and mtry = √p (where p denotes the total number of variables), ensuring stable estimation of feature importance across multiple iterations. A 5-fold cross-validation approach was applied to assess model stability, and the average error rate was reported. Variable importance was ranked based on two metrics: “Mean Decrease in Accuracy” and “Mean Decrease Gini.” A random forest regression algorithm was employed to rank and select the importance of continuous prognostic variables. Associations between various factors were evaluated using a heatmap with color encoding to illustrate the degree of correlation. The “ggcoxzph” function was used to plot residuals, and the “ggcoxfunctional” function was employed to test for linearity between continuous variables and the log hazard in the Cox model (Supplementary Statistical Material 2, http://links.lww.com/JS9/E387). Cox proportional hazard models with robust standard errors were used for univariate and multivariate analyses to identify independent predictors of overall survival. The area under the curve (AUC) was computed with a 95% confidence interval based on 2000 stratified bootstrap replicates. Restricted cubic splines (RCS) were used to evaluate the relationship between postoperative adjuvant chemotherapy cycles and OS in patients with LAGC (Supplementary Statistical Material 3, http://links.lww.com/JS9/E387). Data analysis was performed using SPSS (version 25; IBM, Armonk, NY, USA) and R 4.3.2 (R Foundation for Statistical Computing, Vienna, Austria). Statistical significance was set at P <0.05.

Results

Selection of the modified cachexia index

In the training cohort, ROC curves were used to compare the predictive performance of the traditional CXI for OS and TRG2/3. The results indicated that the AUC values for CXI before, after, and during dynamic changes throughout NACT were as follows: OS (0.504 vs. 0.524 vs. 0.543, respectively) and TRG2/3 (0.519 vs. 0.530 vs. 0.501, respectively) (Figure S3, http://links.lww.com/JS9/E389).

While, mCXI demonstrated superior predictive power for OS compared to traditional CXI across different time points, as assessed by ROC curves (AUC_{pre-nact mCXI} vs. AUC_{pre-nact traditional CXI} = 0.607 vs. 0.504; AUC_{post-nact mCXI} vs. AUC_{post-nact traditional CXI} = 0.638 vs. 0.524; AUC_{Δnact mCXI} vs. AUC_{Δnact traditional CXI} = 0.544 vs. 0.543) (Fig. 2A–C). Similarly, for predicting TRG 2/3, the mCXI outperformed the traditional CXI across all time points (AUC_{pre-nact mCXI} vs. AUC_{pre-nact traditional CXI} = 0.561 vs. 0.519; AUC_{post-nact mCXI} vs. AUC_{post-nact traditional CXI} = 0.605 vs. 0.530; AUC_{Δnact mCXI} vs. AUC_{Δnact traditional CXI} = 0.502 vs. 0.501) (Fig. 2D–F).

Figure 2.

Comparison of receiver operating characteristic (ROC) curves between modified cancer cachexia index and traditional cancer cachexia index in predicting postoperative TRG grade and overall survival.

Panels A-C compare the predictive abilities of pre-neoadjuvant, post-neoadjuvant, and dynamically changing mCXI and traditional CXI at different time points for postoperative overall survival. Panels D-F compare the predictive abilities of pre-neoadjuvant, post-neoadjuvant, and dynamically changing mCXI and traditional CXI at different time points for postoperative TRG grade.

Notably, post-NACT mCXI exhibited higher AUC values for predicting OS and TRG grade than both pre-NACT mCXI and ΔmCXI as well as any individual inflammatory, nutritional, or body composition parameters at any time point (Tables S2–S4, http://links.lww.com/JS9/E388). Consequently, this study used post-NACT mCXI as mCXI for further analyses. Patients were stratified into mCXI-low (mCXI < 17) and mCXI-high (mCXI ≥ 17) groups using the “Maxtast” function in R software (Figure S4, http://links.lww.com/JS9/E389).

Patient characteristics

Comparisons of the patient characteristics are shown in Table 1. In the training cohort, patients were classified into mCXI-low (245/348) and mCXI-high (103/348) groups. Age distribution was comparable between the groups (median 63 years, IQR [56–69] vs. median 65 years, IQR [58–69]; P = 0.424). The mCXI-low group had a significantly higher proportion of male patients (196 [80%] vs. 64 [62.14%]; P < 0.001). Additionally, patients in the mCXI-low group had a lower BMI (median 21.3, IQR [19.4–23.6] vs. median 24.2, IQR [22.2–25.9]; P < 0.001), more advanced ypT staging (ypT3_low vs. ypT3_high = 130 (53.06) vs 49 (47.57); ypT4_low vs. ypT4_high = 59 (24.08) vs. 13 (12.62); P < 0.001), higher TRG grades (TRG2/3_low vs. TRG2/3_high = 188 (76.73%) vs. 68 (66.02); P = 0.039), and poorer RECIST response (SD/PD_low vs. SD/PD_high = 149 (60.82) vs. 42 (40.78); P < 0.001). Further comparisons revealed significant differences in certain inflammatory indicators and nutritional indicators between the groups (Figures S5–S7, http://links.lww.com/JS9/E389). However, no statistically significant differences were observed in other clinicopathological variables. The baseline characteristics of the external validation cohort were similar to those of the internal training cohort, which also showed significantly lower BMI, higher TRG grading, and poorer RECIST response in mCXI-low group (Table 1).

Table 1.

Clinicopathological characteristics at staging by modified cachexia index

Characteristic	FJMUUH (training cohort)				ZZAFJMU (validation cohort)
	Overall, N = 348a	mCXI-low N = 245a	mCXI-high N = 103a	P value	Overall, N = 49a	mCXI-low N = 28a	mCXI-high N = 21a	P value
Age	64 (57, 69)	63 (56, 69)	65 (58, 69)	0.424b	66 (59, 71)	65 (59, 71)	67 (61, 70)	0.627b
Sex				<0.001c				0.201c
Male	260 (74.71%)	196 (80.00%)	64 (62.14%)		35 (71.43%)	22 (78.57%)	13 (61.90%)
Female	88 (25.29%)	49 (20.00%)	39 (37.86%)		14 (28.57%)	6 (21.43%)	8 (38.10%)
BMI (kg/m2)	22.2 (20.2, 24.3)	21.3 (19.4, 23.6)	24.2 (22.2, 25.9)	<0.001b	21.5 (19.4, 22.9)	20.6 (18.5, 22.2)	22.5 (21.5, 25.4)	0.004b
ASA				0.776c				0.559d
1	27 (7.76%)	20 (8.16%)	7 (6.80%)		13 (26.53%)	9 (32.14%)	4 (19.05%)
2	265 (76.15%)	184 (75.10%)	81 (78.64%)		29 (59.18%)	15 (53.57%)	14 (66.67%)
3	56 (16.09%)	41 (16.74%)	15 (14.56%)		7 (14.29%)	4 (14.29%)	3 (14.28%)
cT (AJCC 8th)				0.619d				0.398c
T2	13 (3.74%)	8 (3.26%)	5 (4.86%)		12 (24.49%)	5 (17.86%)	7 (33.33%)
T3	52 (14.94%)	35 (14.29%)	17 (16.50%)		19 (38.78%)	11 (39.28%)	8 (38.10%)
T4	283 (81.32%)	202 (82.45%)	81 (78.64%)		18 (36.73%)	12 (42.86%)	6 (28.57%)
cN (AJCC 8th)				0.848c				0.055c
N0	32 (9.20%)	23 (9.39%)	9 (8.74%)		12 (24.49%)	4 (14.29%)	8 (38.10%)
N +	316 (90.80%)	222 (90.61%)	94 (91.26%)		37 (75.51%)	24 (85.71%)	13 (61.90%)
ypT (AJCC 8th)				0.006c				0.777d
T0	27 (7.76%)	14 (5.72%)	13 (12.62%)		4 (8.16%)	3 (10.71%)	1 (4.76%)
T1	33 (9.48%)	22 (8.98%)	11 (10.68%)		6 (12.24%)	2 (7.14%)	4 (19.05%)
T2	37 (10.63%)	20 (8.16%)	17 (16.50%)		2 (4.09%)	1 (3.57%)	1 (4.76%)
T3	179 (51.44%)	130 (53.06%)	49 (47.57%)		9 (18.37%)	5 (17.86%)	4 (19.05%)
T4	72 (20.69%)	59 (24.08%)	13 (12.63%)		28 (57.14%)	17 (60.72%)	11 (52.38%)
ypN (AJCC 8th)				0.178c				0.507d
N0	141 (40.52%)	92 (37.55%)	49 (47.57%)		18 (36.73%)	8 (28.57%)	10 (47.62%)
N1	69 (19.83%)	47 (19.18%)	22 (21.36%)		9 (18.37%)	6 (21.43%)	3 (14.29%)
N2	55 (15.80%)	41 (16.74%)	14 (13.59%)		14 (28.57%)	8 (28.57%)	6 (28.57%)
N3	83 (23.85%)	65 (26.53%)	18 (17.48%)		8 (16.33%)	6 (21.43%)	2 (9.52%)
NACT cycles	4.00 (3.00, 4.00)	4.00 (2.00, 4.00)	4.00 (3.00, 4.00)	0.729b	3.00 (3.00, 3.00)	3.00 (3.00, 3.00)	3.00 (3.00, 3.00)	0.646b
AC cycles	4.00 (2.00, 6.00)	4.00 (2.00, 6.00)	4.00 (2.00, 4.50)	0.074b	2.00 (1.00, 4.00)	3.00 (1.75, 4.00)	1.00 (1.00, 4.00)	0.113b
AC				0.756c				0.301d
No	26 (7.47%)	19 (7.76%)	7 (6.80%)		4 (8.16%)	1 (3.57%)	3 (14.29%)
Yes	322 (92.53%)	226 (92.24%)	96 (93.20%)		45 (91.84%)	27 (96.43%)	18 (85.71%)
Tumor size (cm)	4.00 (3.00, 6.00)	4.50 (3.00, 6.00)	4.00 (3.00, 5.00)	0.074b	4.50 (3.50, 5.60)	4.50 (3.50, 5.70)	4.50 (3.50, 5.00)	>0.999b
Tumor location				0.250c				>0.999d
Upper 1/3	139 (39.94%)	90 (36.73%)	49 (47.57%)		20 (40.82%)	11 (39.28%)	9 (42.85%)
Middle 1/3	63 (18.10%)	46 (18.78%)	17 (16.50%)		10 (20.41%)	6 (21.43%)	4 (19.05%)
Lower 1/3	113 (32.47%)	86 (35.10%)	27 (26.21%)		9 (18.37%)	5 (17.86%)	4 (19.05%)
Mixed	33 (9.48%)	23 (9.39%)	10 (9.71%)		10 (20.41%)	6 (21.43%)	4 (19.05%)
Differentiation				0.190c				0.600c
Well/moderate	126 (36.21%)	95 (38.78%)	31 (30.10%)		16 (32.65%)	8 (28.57%)	8 (38.10%)
Poor/undifferentiated	192 (55.17%)	132 (53.88%)	60 (58.25%)		31 (63.27%)	18 (64.29%)	13 (61.90%)
Unknown	30 (8.62%)	18 (7.34%)	12 (11.65%)		2 (4.08%)	2 (7.14%)	0 (0.00%)
Lymphovascular invasion				0.245c				0.100c
No	210 (60.34%)	143 (58.37%)	67 (65.05%)		17 (34.69%)	7 (25.00%)	10 (47.62%)
Yes	138 (39.66%)	102 (41.63%)	36 (34.95%)		32 (65.31%)	21 (75.00%)	11 (52.38%)
Perineural invasion				0.666c				0.325c
No	183 (52.59%)	127 (51.84%)	56 (54.37%)		15 (30.61%)	7 (25.00%)	8 (38.10%)
Yes	165 (47.41%)	118 (48.16%)	47 (45.63%)		34 (69.39%)	21 (75.00%)	13 (61.90%)
TRG grade				0.039c				0.055c
0/1	92 (26.44%)	57 (23.27%)	35 (33.98%)		12 (24.49%)	4 (14.29%)	8 (38.10%)
2/3	256 (73.56%)	188 (76.73%)	68 (66.02%)		37 (75.51%)	24 (85.71%)	13 (61.90%)
RECIST criteria				<0.001b				0.005c
CR/PR	157 (45.11%)	96 (39.18%)	61 (59.22%)		37 (75.51%)	17 (60.71%)	20 (95.24%)
SD/PD	191 (54.89%)	149 (60.82%)	42 (40.78%)		12 (24.49%)	11 (39.29%)	1 (4.76%)
LOS (days)	8.0 (7.0, 11.0)	8.0 (7.0, 11.0)	8.0 (7.0, 10.0)	0.113b	7.00 (6.00, 8.00)	7.00 (6.00, 8.00)	7.00 (7.00, 8.00)	0.834b
Operation duration (min)	180 (160, 221)	180 (157, 220)	180 (169, 225)	0.648b	225 (202, 250)	220 (200, 251)	229 (206, 245)	0.486b
Intraoperative blood loss (ml)	40 (30, 50)	50 (30, 60)	35 (30, 50)	0.359b	50 (30, 50)	50 (45, 50)	50 (30, 50)	>0.999b

Bold values indicate that the P value is statistically significant.

AC, adjuvant chemotherapy; ASA, American Society of Anesthesiologists Physical Status Classification System; BMI, body mass index; CR, complete response; LOS, length of stay; NACT, neoadjuvant chemotherapy; PD, progressive disease; PR, partial response; SD, stable disease; TRG, tumor regression grade. ^aMedian (IQR); ^bWilcoxon rank sum test; ^cPearson’s chi-squared test; ^dFisher’s exact test.

Association of modified cachexia index with clinicalpathological characteristic and long-term survival outcomes

In the training cohort, the heatmap revealed no significant interactions between mCXI and other clinicopathological variables (Figure S8A, http://links.lww.com/JS9/E389). Moreover, Kaplan-Meier survival analysis was performed to assess survival differences across mCXI subgroups, indicating that: the mCXI-low group showed a significantly lower 3-year OS rate (58.9% vs. 73.0%; P = 0.002), 3-year RFS rate (50.2% vs. 67.7%; P = 0.002), and 3-year DSS rate (62.5% vs. 74.4%; P = 0.012) (Fig. 3A–C). Additionally, adjusted Cox proportional hazards model analysis showed that mCXI was an independent prognostic factor for OS (HR: 0.62, 95% CI [0.35–0.92]; P = 0.044) (Fig. 4), but not for RFS (HR: 0.73, 95% CI [0.48–1.11]) or DSS (HR: 0.82, 95% CI [0.51–1.31]) (Figure S9A and B, http://links.lww.com/JS9/E389).

Figure 3.

Kaplan-Meier survival curves comparing overall survival (OS), recurrence-free survival (RFS), and disease-specific survival (DSS) between mCXI-low and mCXI-high subgroups.

Panels A–C represent the internal training cohort, and panels D–F represent the external validation cohort. Survival differences were calculated using the Kaplan-Meier method and compared using the log-rank test. The mCXI-high subgroup demonstrated significantly better 3-year OS, RFS, and DSS in the internal cohort, and consistent trends were observed in the validation cohort. These findings highlight the clinical relevance of mCXI in stratifying patient prognosis after neoadjuvant chemotherapy and guiding individualized postoperative management for locally advanced gastric cancer.

Figure 4.

Univariate and multivariate Cox regression analyses associated with overall survival.

This figure shows the correlation between patients’ clinicopathological characteristics and overall survival. The univariate analysis identifies whether the variables are prognostic factors for postoperative survival, and all variables are then included in the Cox regression model for adjustment to select independent prognostic factors.

To further investigate the prognostic value of mCXI across different chemotherapy regimens, patients within both the mCXI-low and mCXI-high subgroups were stratified into four chemotherapy groups: Paclitaxel-based, platinum-based, Paclitaxel plus platinum based, and other regimens. Survival analyses showed no significant differences in OS or RFS among the different chemotherapy regimens within either subgroup (all P > 0.05, Figure S13A–D, http://links.lww.com/JS9/E389).

Recurrence pattern and peak recurrence survival time

By the time of the last follow-up, in the internal training cohort, recurrence was observed in 129 patients (52.6%) and 34 patients (33.0%) in the mCXI-low and mCXI-high groups, respectively (Table 2). The difference in the recurrence rates between the two groups was statistically significant (P = 0.001). Peritoneal recurrence is the most common recurrence. However, no significant statistical differences were observed between the groups in recurrence rates for any recurrence type, including locoregional [14 (10.8%) vs. 4 (11.7%); P = 0.880], liver [17 (13.2%) vs. 3 (9%); P = 0.491], multiple-sites [24 (18.6%) vs. 6 (17.6%); P = 0.898], and other-sites (bone, brain, lung, distant lymph nodes, pancreas, adrenal gland, etc.) [37 (28.7%) vs. 14 (41.2%); P = 0.162].

Table 2.

Recurrence pattern

	mCXI-low (n = 245,%)	mCXI-high (n = 103,%)	P valuea
Recurrenceb	129(52.6%)	34(33%)	0.001
Recurrence location
Locoregional	14(10.8%)	4(11.7%)	0.880
Peritoneal Metastasis	37(28.7%)	7(20.5%)	0.344
Liver	17(13.2%)	3(9%)	0.491
Multiple sitesc	24(18.6%)	6(17.6%)	0.898
Others	37(28.7%)	14(41.2%)	0.162
Bone	8	3	-
Brain	5	2	-
Distant LN	6	4	-
Lung	2	2	-
Other sitesd	16	3	-

Bold values indicate that the P value is statistically significant.

^aP for Pearson’s chi-squared test.

^bRefers only to first-time recurrence, even though patients can have recurrences at multiple times.

^cIncludes patients who have recurrence simultaneously in 2 or more metastatic sites, including peritoneum, liver, lung, bone, brain, distant lymph node, or other hematogenous metastatic sites.

^dIncluding pancreas, ovaries, adrenal glands, rectum, etc.

Further analysis demonstrated a significant later in the time to the first recurrence peak in the mCXI-high group compared to the mCXI-low group (33.51 months vs. 7.11 months) (Fig. 5B). Additionally, comparing the time to the first recurrence peak across different sites revealed that locoregional recurrences had the earliest peak at 5.87 months. In contrast, multiple-site recurrences exhibited a later peak at 11.11 months. The peak times for peritoneal, liver, distant lymph node, and lung recurrence ranged from 6.28 to 11.08 months (Fig. 5A).

Figure 5.

Peak time of recurrence.

Panel A shows the peak time of first recurrence at different recurrence sites in the training cohort. Panel B shows the peak time of first recurrence in the mCXI-low and mCXI-high groups in the training cohort. Panel C shows the peak time of first recurrence in the mCXI-low and mCXI-high groups in the validation cohort.

External validation

In the external validation cohort, the correlation between the mCXI and other clinical variables was consistent with findings from the internal training cohort, demonstrating relative independence (Figure S8B, http://links.lww.com/JS9/E389). mCXI was significantly associated with postoperative OS and RFS, with mCXI-high patients showing better OS and RFS than mCXI-low patients (P for OS = 0.041; P for RFS = 0.013) (Fig. 3D and E). However, the association with DSS was only marginally significant (P = 0.074) (Fig. 3F). Additionally, in the validation cohort, the peak time of recurrence was consistent with that in the internal cohort, with mCXI-high patients experiencing delayed peak recurrence compared with mCXI-low patients (19.69 months vs. 16 months) (Fig. 5C).

Prediction of postoperative adjuvant chemotherapy effects by modified cachexia index

In the entire cohort, RCS analysis revealed a nonlinear relationship between the number of postoperative AC cycles and OS (P = 0.0245) (Figure S10, http://links.lww.com/JS9/E389). Based on this, AC cycles were categorized into three groups: fewer than four cycles, four to six cycles, and more than six cycles (Table S5, http://links.lww.com/JS9/E388).

Subgroups analysis according to the different AC cycles in the mCXI-low group revealed that there was no significant survival difference in OS of different postoperative chemotherapy cycles (3-year OS: AC cycle < 4 vs. 4 ≤ AC cycle ≤6 vs. AC>6 = 50.7% vs. 65.7% vs. 51.8%; P = 0.165) (Figure S11A, http://links.lww.com/JS9/E389). While, for the ypStage III in mCXI-low patients who received four or more cycles of postoperative chemotherapy had significantly better 3-year OS than those who received fewer (3-year OS: AC cycle ≥ 4 vs. AC cycle <4 = 43.7% vs. 24.2%; P = 0.005) (Figure S12C and D, http://links.lww.com/JS9/E389). Similar results were not observed in the ypCR/ypI and ypII groups (all P > 0.05) (Figure S12A and B, http://links.lww.com/JS9/E389).

However, in the mCXI-high group of the entire cohort, patients who received 4-6 cycles of AC had a significantly better 3-year OS compared to those who received fewer than 4 cycles or more than 6 cycles (3-year OS: < 4 cycles vs. 4-6 cycles vs. > 6 cycles = 65.6% vs. 87.4% vs. 50%; P = 0.005) (Figure S11B, http://links.lww.com/JS9/E389).

Discussion

This study proposes a novel mCXI as a potential marker of nutritional status. It evaluates its prognostic value and utility in guiding postoperative adjuvant chemotherapy decisions in patients undergoing NACT for LAGC. The mCXI, a composite index, incorporates post-NACT inflammation, nutrition, and body composition indicators. We observed that patients in the mCXI-low group had significantly worse comprehensive survival outcomes than those in the mCXI-high group. This negative effect remained evident even after adjusting for known confounders. Despite comparable recurrence rates at various sites, patients with a low mCXI experienced an earlier peak in the time to first recurrence following NACT. Further analysis revealed that in the mCXI-low group, patients with ypIII stage disease who received more than four cycles of adjuvant chemotherapy had improved survival outcomes compared to those receiving fewer cycles. However, a significant difference in overall prognosis between patients receiving four to six cycles and those receiving fewer or more cycles was observed in the mCXI-high group. These findings suggest that the mCXI could serve as a simple, objective, and effective approach for assessing prognosis and guiding individualized postoperative AC recommendations in patients with LAGC.

Fat loss is a hallmark of cancer cachexia^[27], and studies have shown that reductions in adipose tissue exacerbate energy metabolism disorders, impair quality of life, and shorten patient survival^[28,29]. Recently, numerous studies have confirmed the prognostic relevance of subcutaneous adipose tissue (SAT) across multiple cancer types, including breast cancer^[30], esophageal cancer^[31], pancreatic cancer^[32], and multiple myeloma^[33]. A recent study focusing on gastric cancer patients also found significantly lower survival rates among those with low SAT compared to those with higher SAT, suggesting the potential value of SAT in cancer cachexia risk assessment^[34]. Moreover, ALB is the sole biochemical parameter consistently included in various nutritional scoring systems such as the Nutritional Risk Index (NRI), Prognostic Nutritional Index (PNI), and CONUT score, underscoring its stability and applicability. In our study, ALB demonstrated superior prognostic value compared to other nutritional markers. Recent studies have highlighted that platelets play critical roles beyond hemostasis, including tumor progression, angiogenesis, immune evasion, and metastasis^[26,35]. Under cancer cachexia conditions, tumor cells can stimulate the release of pro-inflammatory cytokines such as interleukin-6 (IL-6), which in turn activate platelets and immune cells, triggering a cascade of inflammatory responses that lead to increased lipolysis and muscle wasting^[36,37]. Furthermore, platelets can actively secrete inflammatory mediators, exacerbating cachexia progression; clinically, thrombocytosis is often associated with advanced tumors, high tumor burden, and poor prognosis^[38,39]. Based on the above mechanisms and clinical evidence, our study incorporated these key indicators – subcutaneous adipose tissue (SAT), albumin (ALB), and platelet count (PLT) – into the mCXI model to more comprehensively and accurately assess the prognostic risk of cancer cachexia patients following neoadjuvant chemotherapy.

However, in the training cohort, the overall AUC values of mCXI for predicting OS and TRG2/3 were relatively low, which may raise concerns about its discriminative performance. This could be closely related to its intended function. mCXI was designed as a prognostic stratification tool to aid in multifactorial risk assessment, rather than as an independent diagnostic marker. In patients with locally advanced gastric cancer receiving neoadjuvant chemotherapy, prognosis is influenced by multiple factors, making it difficult for a single indicator to achieve high discriminative accuracy. Therefore, moderately low AUC values are common in similar studies. For instance, Wang et al^[40] recently proposed the Systemic Nutrition-Inflammation Index (SNII), which, despite outperforming several traditional indices, had an AUC range of only 0.534–0.612. This reflects the inherent limitations of using nutritional and inflammatory indicators to predict outcomes in complex disease settings. Moreover, mCXI consistently showed predictive advantages over traditional CXI across multiple time points and endpoints, and this trend was validated in an external cohort, indicating its strong stability and generalizability.

Jafri et al^[8] introduced the Cancer Cachexia Index (CXI) as a composite measure of various clinical characteristics of cachexia. Their research highlighted that patients with metastatic non-small cell lung cancer exhibiting a low CXI had reduced overall and progression-free survival rates. Similar negative survival outcomes have been documented for other cancer types. Gong et al^[10] and Sakurai et al^[12] assessed the CXI in patients undergoing surgical resection for gastric cancer and found that a low CXI was associated with poorer survival. However, unlike these studies, our mCXI incorporates neoadjuvant chemotherapy-specific factors, providing a more tailored approach for LAGC patients.

Historically, the primary focus in cancer cachexia theory has been muscle loss, which is considered a key complication of this condition. To date, research on cachexia has predominantly focused on the muscle tissue^[24,41,42]. However, Argilés et al^[43] highlighted the significant role of non-muscle tissue and intertissue communication in cancer cachexia, leading to the exploration of new therapeutic strategies. Furthermore, Kays et al^[28] demonstrated that in patients with advanced pancreatic ductal adenocarcinoma (PDAC) undergoing neoadjuvant therapy, fat loss alone had a prognostic value equivalent to that of muscle loss. Their research underscores the critical role of adipose tissue in cachexia and PDAC mortality. In our study, random forest analysis revealed that SAT had greater prognostic relevance than SMI in patients with neoadjuvant gastric cancer. This pattern may reflect the rapid metabolic demands placed on adipose reserves during chemotherapy, which precedes muscle catabolism, especially in patients with advanced disease. Consequently, the SAT may be a more significant prognostic indicator than the SMI. This pattern may reflect the rapid metabolic demands placed on adipose reserves during chemotherapy, which precedes muscle catabolism, especially in patients with advanced disease.

Tumor-associated coagulopathies are commonly observed in patients with malignancies, and both fibrinogen (FIB) and platelets (PLT) are critical components of the coagulation system^[44]. Fibrinogen is an acute-phase protein synthesized by the liver, and its levels increase during systemic inflammation. Sustained chronic inflammation may accelerate multi-organ dysfunction and is closely associated with tumor progression^[45]. Previous studies have demonstrated that elevated fibrinogen levels are not only associated with tumor stage, angiogenesis, and metastasis, but also promote tumor cell adhesion, immune escape, and treatment resistance in gastrointestinal and lung cancers^[25,46,47]. Cancer cachexia is often accompanied by systemic inflammation and coagulation abnormalities. Platelets, as key mediators of both inflammation and coagulation, may be influenced by nutritional status, which in turn affects tumor behavior and progression^[26]. In gastrointestinal tumors, platelet activation is frequently observed, driven by the prothrombotic tumor microenvironment. Moreover, platelets can facilitate tumor angiogenesis via the release of vascular endothelial growth factor (VEGF), and promote metastasis through direct interaction with tumor cells^[35].

Moreover, the tumor immune-inflammatory microenvironment contributes significantly to cancer progression^[48-50]. Hanahan et al^[51] demonstrated that immune and inflammatory cells affect tumor growth through cytokines and chemokines produced via autocrine and paracrine signaling. The high metabolic rate and abnormal proliferation of cancer cells make patients more susceptible to malnutrition, leading to loss of muscle mass, fat, and overall body weight. Shi et al^[11] revealed that malnutrition can adversely affect the immune-inflammatory microenvironment, leading to an imbalance between immune suppression and tumor proliferation. This imbalance ultimately leads to a vicious cycle of “malnutrition-immune suppression-tumor progression” in patients. Previous research on cachexia indicators for non-small cell lung cancer^[8] and diffuse large B-cell lymphoma^[24] employed the NLR as a key inflammatory marker in traditional CXI models to predict patient outcomes, demonstrating significant predictive accuracy. In our study of patients with LAGC undergoing NACT, we found that the PLT count after neoadjuvant treatment offered superior prognostic value compared to the NLR and had greater clinical applicability. This observation is consistent with the findings of Shi et al, who showed that platelet count (PLT) had a more independent association with overall prognosis in patients with gastrointestinal cancer than the NLR and other markers^[52].

Prospective studies on perioperative chemotherapy have consistently demonstrated significant improvements in the 3- and 5-year survival rates of patients with locally advanced gastric cancer compared with those undergoing surgery alone^[53-57]. However, in patients with advanced gastric cancer and poor nutritional status, rapid tumor progression coupled with disease-related wasting and chemotherapy toxicity exacerbates the decline in physical function and nutritional status. Without timely nutritional interventions and targeted adjuvant therapies, these patients are at a heightened risk of poor clinical outcomes. Although the therapeutic and toxic effects of perioperative chemotherapy are pivotal, conflicting factors influence patient prognostic outcomes. Therefore, tailoring the postoperative adjuvant treatment for individuals with neoadjuvant-treated gastric cancer has emerged as a critical challenge in current clinical practice. In our study, we found that for mCXI-low patients with postoperative pathological stage ypIII, a minimum of four cycles of adjuvant chemotherapy was required to achieve optimal overall survival outcomes. For patients in the mCXI-high group with better prognoses, receiving 4-6 cycles of adjuvant chemotherapy significantly improved overall survival. However, since the number of patients in this group receiving more than 6 cycles of chemotherapy was relatively small, further investigation is needed to determine whether extended chemotherapy cycles provide additional benefits. Consequently, clinicians should consider the nutritional status of these patients when planning postoperative adjuvant chemotherapy to avoid overtreatment. The application of the mCXI in identifying individual differences among patients with neoadjuvant-treated gastric cancer and guiding targeted therapeutic strategies may be crucial for optimizing comprehensive care. Notably, this finding has not been reported previously.

Several limitations should be acknowledged. First, as a retrospective study, it was inherently susceptible to recall and selection biases. However, the prospectively collected data and an external validation cohort may have reduced this type of bias. Additionally, owing to data limitations, the toxicities experienced during chemotherapy and dose adjustments were not collected, resulting in a lack of analysis of adverse reactions during these periods. Finally, although this study conducted a comprehensive assessment of body composition using a substantial and complete dataset of pre- and post-NACT CT images, the relatively small sample size remains a limitation. Future studies should focus on validating these findings in larger, multi-center cohorts and incorporate real-time toxicity data to refine mCXI’s utility in guiding chemotherapy regimens.

Conclusion

This multicenter cohort study demonstrated that the modified Cancer Cachexia Index (mCXI), which integrates post-NACT nutritional, inflammatory, and body composition parameters, is a strong prognostic tool for patients with LAGC. The mCXI showed better performance than the traditional CXI in predicting long-term outcomes and provided useful insights for guiding adjuvant chemotherapy decisions. Its clinical application may improve individualized treatment and optimize prognosis in this population.

What this study adds?

• Proposes a modified Cancer Cachexia Index (mCXI) adapted to the peri-neoadjuvant chemotherapy (NACT) setting in locally advanced gastric cancer (LAGC).

• Demonstrates that nutritional-inflammatory predictors differ before and after NACT, supporting the need for stage-specific modeling.

• Shows that mCXI consistently outperforms traditional CXI in predicting both overall survival and treatment response across internal and external cohorts.

• Highlights the biological relevance of SAT, ALB, and PLT in cancer cachexia and prognosis.

Ethical approval

The study was approved by the Institutional Review Board at the Fujian Medical University Union Hospital and Zhangzhou Affiliated Hospital of Fujian Medical University (2024KY039).

Consent

Participants were informed of the study's purpose, procedures, potential risks and benefits, and their right to withdraw from the study at any time without penalty. Data collected were kept confidential and used solely for research purposes. Patients signed informed consent regarding publishing their data.

Sources of funding

This study was supported by the Fujian Province Medical “Creating high-level hospitals, high-level medical centers and key specialty projects” (MWYZ [2021] No. 76), Fujian provincial health technology project (2022QNA026) and Fujian Provincial Natural Science Foundation of China (2024J01586). The funders 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.

Author contributions

L.K.Z., H.L.Z., C.M.H., and J.W.X.: concept and design; L.K.Z., H.L.Z., B.B.X., G.T.L., H.H.Z., Y.H.T., and P.L.: acquisition, analysis, or interpretation of data; L.K.Z., H.L.Z., Q.Y.C., and J.W.X.: drafting of the manuscript; L.K.Z., H.L.Z., X.Y.Z., and J.W.X.: statistical analysis; X.Y.Z., Z.W.Z., C.M.H., P.L., J.W.X., and C.H.Z.: administrative, technical, or material support; H.L.Z., C.M.H., and J.W.X.: supervision.

Conflicts of interest disclosure

There are no conflicts of interest or financial ties to disclose from any of author.

Research registration unique identifying number (UIN)

Name of the registry: Clinical Trials.gov. Unique Identifying number or registration ID:NCT06689462. Hyperlink to your specific registration (must be publicly accessible and will be checked): https://clinicaltrials.gov/ct2/results?cond=NCT06689462&term=&cntry=&state=&city=&dist=.

Guarantor

Jian-Wei Xie.

Provenance and peer review

Not commissioned, externally peer-reviewed.

Data availability statement

The dataset analyzed for this study is available from the corresponding author on reasonable request.