This retrospective cohort study was conducted at Union Hospital, Tongji Medical College, Huazhong University of Science and Technology, from March 2019 to June 2024. A total of 138 gastric cancer patients undergoing immunotherapy were enrolled. Inclusion criteria were: (I) histologically or cytologically confirmed gastric cancer; (II) age over 18 years; (III) adenocarcinoma pathology type; (IV) initial diagnosis and completion of treatment at this institution; (V) abdominal CT examination performed before initiation of ICIs therapy. Exclusion criteria included: (I) presence of other malignancies; (II) history of prior surgical intervention or immunotherapy; (III) coexisting chronic liver disease, systemic infectious diseases, or other conditions that may cause splenomegaly; (IV) missing medical records or relevant data; (V) poor-quality abdominal CT scans; (VI) splenic involvement.

Patient demographic and clinical baseline data were retrospectively collected, covering the entire period from admission to discharge or death. Follow-up assessments were conducted every six months via telephone, with a primary focus on disease progression and mortality, and continued until the patient’s death.

CT images were obtained using Siemens or Philips medical scanners, with tube voltage set at 120 kVp and automatic tube current modulation to optimize image quality. Advanced reconstruction techniques were employed using a 512×512 matrix and 64 mm×0.625 mm collimation. The standardized reconstruction slice thickness was 1.5 mm, and slice spacing ensured consistency and measurement accuracy. Images were stored and CT values measured using the Picture Archiving and Communication System (PACS).

Baseline SV was calculated using CT scans obtained within 4 weeks before starting ICI therapy. Changes in volume were assessed during follow-up imaging after ICIs treatment. Measurements were taken on both axial and coronal reformatted CT images, with SV calculated by a radiologist (L.L.L.) with 13 years of experience, blinded to clinical data. Inter-observer reproducibility was assessed in a randomly selected subset of 30 patients. Splenic volume measurements were independently performed by a second radiologist with equivalent experience, who was blinded to the initial measurements and all clinical outcomes, using the same standardized protocol. Inter-observer agreement was evaluated using a two-way random-effects intraclass correlation coefficient with absolute agreement [ICC(A,1)].

SV was determined using the formula derived by Prassopoulos et al. The formula is as follows: V (cm3) = 30 + 0.58 ×(W ×L ×T). Where W is the maximum width at the splenic hilum, T is the maximum thickness of the splenic hilum in a plane perpendicular to W, and L is the maximum caudo-cranial length (16) (Figure 1).

For each patient, the volume difference of the spleen was determined by comparing the two abdominal CT images with the largest time interval. The volumetric difference was divided by the time interval in months and then multiplied by twelve to calculate the annualized volumetric evolution, expressed as follows: ΔSV (%/year) = [(b − a) / a] × (12/c) Among them, a is the pre-treatment splenic volume (cm³), b is the splenic volume at the most recent follow-up (cm³), c is the interval between the two measurements (months), and ΔSV is the annualized splenic volume change (%). The ΔSV was subsequently analyzed as a binary variable, with the optimal cutoff point (14%) serving as the critical threshold (Supplementary Figure 1). To evaluate potential overfitting associated with the internally derived ΔSV cutoff, internal validation of the multivariable Cox model was performed using bootstrap resampling (1,000 repetitions) and repeated 5-fold cross-validation (50 repetitions). Model discrimination was assessed using Harrell’s concordance index (C-index).

To assess the robustness of ΔSV estimation under heterogeneous follow-up intervals, sensitivity analyses were conducted. These included stratification by CT follow-up duration, derivation of duration-adjusted ΔSV using linear mixed-effects modeling, and temporal standardization by interpolating splenic volume at a uniform 6-month post-treatment time point.

Radiological response was assessed using abdominal CT scans and evaluated according to the RECIST version 1.1 criteria. OS was defined as the time from study enrollment to death. PFS was defined as the time from first treatment administration to radiographic disease progression or death, whichever occurred first. Objective response rate (ORR) included patients achieving complete response (CR) or partial response (PR), while disease control rate (DCR) included patients achieving CR, PR, or stable disease (SD).

Statistical analyses were performed using IBM SPSS Statistics (version 28) and R software (version 4.4.2). Descriptive statistics were used to assess variable distributions. Continuous variables were expressed as mean ± standard deviation or median (interquartile range [IQR]), while categorical variables were presented as counts and percentages. The optimal cutoff values for continuous variables were determined using the survminer package in R, and these thresholds were then used to dichotomize the variables. Optimal cutoff values for neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), systemic immune-inflammation index (SII), lymphocyte-to-monocyte ratio (LMR), prognostic nutritional index (PNI), ΔSV, and baseline splenic volume are summarized in Supplementary Table 1.

Based on the selected splenic volume threshold, patients were divided into two groups: “increased group” and “non-increased group.” Kaplan-Meier survival analysis was used to assess PFS and OS, and the log-rank test was applied to compare the two groups. Univariate and multivariate Cox regression models were used to examine the effect of ΔSV on survival. Variables with p-values less than 0.10 were included in the multivariate analysis model. Hazard ratios (HR) and corresponding 95% confidence intervals (CIs) were calculated.

Bland–Altman analysis showed that most measurements were within the 95% limits of agreement, without evident systematic bias (Supplementary Figure 2). Intraclass correlation coefficient analysis using a two-way random-effects model with absolute agreement yielded an ICC(A,1) of 0.938 (95% CI: 0.889–0.967; P< 0.001).

A total of 138 gastric cancer patients were enrolled in this study (87 in the splenic volume increased group and 51 in the non-increased group), and all patients received chemotherapy combined with immune checkpoint inhibitor treatment. Table 1 presents the baseline demographic and clinical characteristics. Among all enrolled patients, 68 (49.3%) were over 60 years old, and 70 (50.7%) were under 60. Additionally, 95 patients (68.8%) were male, and 43 (31.2%) were female. Overall, no significant differences were observed between the two groups in terms of gender, age, smoking history, disease stages, inflammatory markers, or basic biochemical indicators. Furthermore, when patients were categorized based on the median baseline splenic volume, the baseline characteristics were summarized in Supplementary Table 2. In the baseline volume cohort, no statistically significant association was found between splenic volume size and progression-free survival (HR, 0.74 [95% CI, 0.45–1.20]; P = 0.220) or overall survival (HR, 1.13 [95% CI, 0.65–1.98]; P = 0.664) (Supplementary Figure 3). Given the lack of prognostic significance of baseline splenic volume, we next focused on the prognostic value of ΔSV.

The median baseline splenic volume for all enrolled patients was 166.1 cm³ (IQR: 133.80–212.94 cm³). The median final splenic volume after chemotherapy combined with immune checkpoint inhibitor treatment was 203.9 cm³ (IQR: 150.20–259.67 cm³). To assess whether splenic volume increased, a correlation analysis was performed between the two splenic volume measurements for the same subject. Paired sample boxplots showed a significant difference in splenic volume before and after immunotherapy in gastric cancer patients, with a notable increase (P < 0.001) (Supplementary Figure 4).

The tumor response of the splenic volume increased group and the non-increased group is summarized in Supplementary Table 3. Overall, no significant differences were observed between the two groups in terms of short-term treatment outcomes. The ORR and DCR were 37.9% and 94.3% in the increased group, and 47.1% and 98.0% in the non-increased group, respectively, with no statistically significant differences (ORR, P = 0.293; DCR, P = 0.535). Notably, the proportion of patients with progressive disease (PD) was higher in the increased group than in the non-increased group (5.7% vs. 2.0%).

Kaplan-Meier survival curves were plotted for PFS and OS in patients with ΔSV ≤ 14% and ΔSV > 14%. Log-rank tests showed that the “increased group” had significantly shorter PFS (14.8 months vs. 26.6 months, P < 0.001) and OS (18.9 months vs. 30.9 months, P < 0.001) compared to the “non-increased group” (Figures 2A, B). The HR for PFS in the non-increased group compared to the increased group was 0.36 (95% CI, 0.21–0.64), and the HR for OS was 0.22 (95% CI, 0.10–0.45). Additionally, time-dependent receiver operating characteristic (ROC) curves were used to explore the prognostic predictive value of splenic volume changes in advanced gastric cancer, showing that splenic volume change had good predictive value for both PFS and OS (Supplementary Figure 5). The area under the curve (AUC) for ΔSV was 0.711, 0.693, and 0.809 for 1-year, 1.5-year, and 2-year PFS, respectively, and 0.814, 0.729, and 0.808 for 1-year, 1.5-year, and 2-year OS, respectively.

Univariate Cox proportional hazards models were used to analyze demographic, clinical, and cancer-related factors (Tables 2, 3). Specifically, in univariate regression analysis for PFS, Eastern Cooperative Oncology Group Performance Status (ECOG PS), PNI, splenic volume changes, and stages were associated with prognosis; for OS, ECOG PS, PNI, splenic volume changes, and stages were identified as potential predictors. These covariates were then included in multivariate regression analysis. In the multivariate analysis, higher ECOG PS (HR, 3.558 [95% CI, 1.801–7.029]; P < 0.001) and splenic volume increased (HR, 0.419 [95% CI, 0.233–0.751]; P = 0.004) were associated with a higher risk of disease progression. High PLR (HR, 2.143 [95% CI, 1.098–4.181]; P = 0.025), high ECOG scores (HR, 3.172 [95% CI, 1.509–6.669]; P = 0.002), stage IV disease (HR, 2.700 [95% CI, 1.067–6.829]; P = 0.025), and splenic volume increased (HR, 0.233 [95% CI, 0.109–0.501]; P< 0.001) were positively associated with shorter overall survival.

Internal validation of the multivariable Cox model incorporating ΔSV was performed. The apparent C-index of the model was 0.775. After bootstrap resampling, the optimism-corrected C-index was 0.756. Repeated 5-fold cross-validation yielded a mean C-index of 0.752 with a standard deviation of 0.020 (95% CI: 0.712–0.788). The internal validation results are summarized in Supplementary Table 4.

Sensitivity analyses were conducted to evaluate the effect of heterogeneous follow-up intervals on ΔSV estimation. Stratified analyses according to follow-up duration (<6 months, 6–12 months, and >12 months) yielded hazard ratios of comparable magnitude for both progression-free survival and overall survival. After correction using linear mixed-effects modeling, ΔSV remained significantly associated with survival outcomes. Similar results were observed after temporal standardization of splenic volume at a 6-month post-treatment time point. Detailed results are presented in Supplementary Tables 5–8.

Subgroup analyses based on baseline characteristics revealed consistent results for both PFS and OS. In both PFS and OS analyses (Supplementary Figures 6, 7), the risk of progression in the group with increased ΔSV was higher than that in the group with non-increased ΔSV, though some subgroups did not show statistical significance.

After adjusting for covariates, we used restricted cubic spline (RCS) curves to explore the relationship between splenic volume change and PFS and OS. A significant nonlinear association was found between splenic volume change and PFS (Figure 3A) (overall P < 0.001, nonlinear P = 0.032). When the splenic volume change was small, the hazard ratio for disease progression remained low; as splenic volume increased, the risk of disease progression gradually rose. A similar nonlinear association was observed between splenic volume change and OS (Figure 3B) (overall P < 0.001, nonlinear P = 0.014). When splenic volume change was minimal, the hazard ratio for death was close to 1; as splenic volume increased, the hazard ratio rose rapidly, although the increase slowed over time, and the hazard ratio remained relatively high.

Based on the independent risk factors identified in the multivariate Cox regression, we combined ΔSV with key clinical features to create a nomogram for predicting 1-year and 2-year overall survival in patients with advanced gastric cancer (Figure 4). The predictive accuracy of the model was assessed using time-dependent AUC and calibration curves. The calibration curves confirmed a high consistency between predicted and actual outcomes. Compared with the time-dependent AUC curves for the single-factor stages prognosis, the multifactor prognostic model’s time-dependent AUC values for 1-year, 1.5-year, and 2-year survival were 0.802, 0.762, and 0.883, respectively, while the time-dependent AUC values for the single-factor stages model were only 0.564, 0.575, and 0.686 (Supplementary Figure 8).

Decision curve analysis was conducted to evaluate the net benefit of the nomogram for predicting 1-year and 2-year overall survival. As shown in Supplementary Figure 9, for 1-year overall survival, the nomogram yielded a higher net benefit than both the “treat-all” and “treat-none” strategies when the threshold probability was below approximately 0.65. For 2-year overall survival, the nomogram similarly outperformed the default strategies across threshold probabilities up to approximately 0.60.