CT-derived abdominal organ volumetrics for predicting recurrence-free and disease-free survival in resected non-small cell lung cancer: a multicenter retrospective cohort study

Introduction

Non-small cell lung cancer (NSCLC) is the most common histologic subtype of lung cancer and remains a leading cause of cancer mortality worldwide (1). For patients with resectable disease, curative-intent surgery represents the cornerstone of treatment; however, postoperative recurrence remains a major barrier to long-term survival (2). Recurrence risk is strongly associated with tumor-node-metastasis (TNM) stage. According to the most recent International Association for the Study of Lung Cancer (IASLC) data, the 5-year survival rate for stage IA disease can reach 82–94%, whereas it decreases to approximately 54–64% and 24–36% for stages II and III, respectively (3). Nevertheless, TNM staging primarily describes anatomic tumor burden and does not fully capture the biological heterogeneity of NSCLC. Established clinicopathologic and molecular factors, including histologic subtype, tumor grade, visceral pleural invasion, lymphovascular invasion, spread through air spaces (STAS), driver mutation status, resection extent, and receipt of adjuvant therapy, may further modify recurrence risk within the same TNM stage (4,5). For example, among patients with stage IA NSCLC, STAS-positive tumors have a significantly higher risk of recurrence than STAS-negative tumors, underscoring the limitations of anatomic staging alone (6).

In addition to tumor-intrinsic features, increasing evidence indicates that host-related factors contribute to postoperative outcomes in NSCLC. Cancer cachexia, sarcopenia, malnutrition, systemic inflammation, and impaired immune homeostasis have all been linked to poorer survival or recurrence-related outcomes in lung cancer and other malignancies (7-9). These factors may influence postoperative recovery, tolerance to adjuvant therapy, immune surveillance, and the dormancy or reactivation of micrometastatic disease (10,11). Preoperative computed tomography (CT) provides an opportunity to quantify such patient-level characteristics because CT is routinely obtained for diagnosis and staging and captures not only the primary tumor and nodal disease but also body composition and abdominal organ morphology. Prior studies have predominantly focused on skeletal muscle and adipose tissue, and CT-derived sarcopenia, myosteatosis, or adverse body-composition profiles have been associated with worse outcomes after surgical resection of NSCLC and across other malignancies (12-17). More recently, integrated models combining CT-based body composition with inflammatory or nutritional biomarkers have further highlighted the potential value of imaging-derived host phenotypes for postoperative risk stratification in NSCLC (18).

Beyond skeletal muscle and adipose tissue, abdominal solid organ volumes may capture additional dimensions of systemic host status because the liver, spleen, pancreas, kidneys, and adrenal glands participate in metabolic, immune-inflammatory, nutritional, renal, endocrine, and stress-related processes. Their morphology may therefore provide a CT-derived phenotype of host reserve that complements tumor-centered prognostic factors. Consistent with this concept, several preliminary studies support the feasibility and potential relevance of solid-organ volumetry in oncology. McGovern et al. evaluated liver volume (LVV) derived from preoperative CT in patients undergoing curative resection for non-metastatic colon cancer and examined its relationship with clinicopathologic characteristics and survival outcomes (19). In advanced NSCLC treated with immune checkpoint inhibitors, splenic volume and spleen-related indices have also been investigated as potential markers of immune-inflammatory status and treatment outcome (20,21). However, these studies were largely conducted in non-lung cancer populations or in patients with advanced disease receiving systemic therapy. Whether abdominal solid organ volumes provide independent prognostic information in surgically resected NSCLC remains insufficiently defined. Moreover, manual organ segmentation is time-consuming and operator-dependent, limiting the scalability and reproducibility of prior volumetric studies. Recent advances in automated deep-learning segmentation now allow high-throughput, reproducible quantification of multiple abdominal organs from routine CT imaging (22).

Against this background, we designed the present study to test whether abdominal organ volumetric metrics derived from routine preoperative CT could improve recurrence risk assessment in resected NSCLC. We included preoperative CT scans from 1,762 patients with surgically resected NSCLC across five centers and 3,263 healthy controls. Using an automated deep-learning segmentation pipeline, we quantified the volumes of seven abdominal organs, including the spleen, liver, pancreas, bilateral kidneys, and bilateral adrenal glands, and derived corresponding body surface area (BSA)-normalized organ indices. We compared these organ metrics between patients with NSCLC and healthy controls and examined their prognostic relevance for postoperative recurrence-related outcomes. Our goal was to determine whether abdominal organ volumetry may serve as an objective and scalable CT-based approach for recurrence risk stratification after curative-intent resection of NSCLC. We present this article in accordance with the STROBE reporting checklist (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0445/rc).

Methods

Ethics statement

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the ethics committees of all participating centers, including Hubei Provincial Hospital of Traditional Chinese Medicine (No. HBZY2026-C010-01), Wuhan Union Hospital (No. S0802), The Affiliated Traditional Chinese Medicine Hospital of Southwest Medical University (No. BY2025078), The First Affiliated Hospital of Bengbu Medical University (No. KY041), and Renmin Hospital of Wuhan University (No. WDRY2023-K004). The requirement for written informed consent was waived because of the retrospective nature of the study.

Study design and patient selection

We conducted a multicenter retrospective cohort study at five medical centers. We retrospectively screened all consecutive patients with clinically and pathologically confirmed NSCLC who underwent curative-intent surgical resection between April 2020 and July 2025 using the institutional electronic medical record, radiology, surgical, and pathology databases. Patients who met the predefined inclusion and exclusion criteria were included in the final analysis. The patient selection flow diagram is shown in Figure 1. The inclusion criteria were as follows: (I) clinically and pathologically confirmed NSCLC treated with curative intent; (II) CT performed within 1 month before surgery; and (III) complete clinicopathologic and follow-up data. The exclusion criteria were as follows: (I) poor CT image quality or incomplete scan coverage; (II) incomplete clinical or prognostic data; and (III) concomitant other primary malignancy. We also enrolled a healthy control cohort from the physical examination center of Wuhan Union Hospital as a reference population for baseline organ-volumetric comparison. All controls were confirmed to be in good health based on clinical assessment and underwent CT scanning with complete abdominal coverage. Controls with poor CT image quality, incomplete scan coverage, incomplete key clinical data, or a history of malignancy were excluded.

Figure 1 The overview of the study design and workflow. BYFYY, The First Affiliated Hospital of Bengbu Medical University; CT, computed tomography; HPH-TCM, Hubei Provincial Hospital of Traditional Chinese Medicine; NSCLC, non-small cell lung cancer; RHWU, Renmin Hospital of Wuhan University; SMU-TCM, The Affiliated Traditional Chinese Medicine Hospital, Southwest Medical University; WHUH, Wuhan Union Hospital.

Figure 1 illustrates the overall study design and workflow. We processed CT images from eligible participants using a deep learning-based automated segmentation algorithm to segment seven abdominal organs, including the bilateral adrenal glands, bilateral kidneys, pancreas, liver, and spleen, and to extract organ volumes. We calculated organ indices by normalizing organ volumes to BSA. We then compared organ volumes and indices between patients with NSCLC and healthy controls. Correlation analyses were used to evaluate associations among organ features and between organ features and clinical variables. Survival analyses were performed in the full NSCLC cohort to assess associations between organ features and clinical outcomes and to examine prognostic stratification across clinical subgroups.

Quantification of abdominal organ volumes and indices

All patients underwent CT scanning within 1 month before surgical resection. CT acquisition parameters for each institution are summarized in Table S1. All CT images were processed using a pretrained neural network based on the nnU-Net framework for automated organ segmentation. For each organ, the algorithm generated segmentation masks. Absolute organ volumes were calculated by counting voxels within each mask and multiplying by the voxel spacing. Organ indices were then derived by normalizing organ volumes to BSA. Organ volumes were reported in cm³, and BSA-normalized organ indices were reported in cm³/m². For each patient, we obtained spleen volume (SPV), right kidney volume (RKV), left kidney volume (LKV), LVV, pancreas volume (PNV), right adrenal gland volume (RAGV), LAGV, spleen index (SPI), right kidney index (RKI), left kidney index (LKI), liver index (LVI), pancreas index (PNI), right adrenal gland index (RAGI), and left adrenal gland index (LAGI). In total, 14 anatomic quantitative parameters (seven volumes and seven organ indices) were obtained; detailed definitions and calculation procedures are provided in Table S2. All segmentation and feature extraction were performed on an Ubuntu 22.04 Linux server equipped with an NVIDIA RTX 4090 graphics processing unit (GPU).

To validate the reliability of the automated segmentation algorithm, we randomly selected 50 patients, and a radiologist with 15 years of experience in thoracic diagnostic imaging manually delineated the seven abdominal organs. Agreement between manual and automated measurements was high (Figure S1A-S1G), and the automated approach markedly reduced the time required for quantification (Figure S1H). Therefore, all subsequent analyses used organ metrics derived from automated segmentation, which helped reduce observer-related measurement variability compared with manual delineation.

Clinical data and outcomes

Clinical data were obtained from the electronic medical record systems of the participating institutions. The collected variables included sex, age, height, weight, diabetes, hypertension, smoking, drinking, albumin, neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), pathological types, and T stage. The date of surgical resection was defined as the index date. Clinical variables were obtained from preoperative medical records. Laboratory variables were derived from the most recent routine preoperative tests before surgery, which were generally performed within 7 days; if multiple results were available, the result closest to surgery was used. Body mass index (BMI) was calculated from preoperative height and weight. Pathological types and T stage were extracted from the final surgical pathology report.

Postoperative follow-up information was obtained from outpatient and inpatient records, institutional imaging systems, and telephone follow-up. The administrative cutoff date for follow-up was December 31, 2025. Across the five centers, patients were followed using broadly similar surveillance protocols, generally every 3–6 months during years 0–2 after surgery, every 6–12 months during years 3–5, and annually thereafter. Chest CT was the core imaging modality for routine surveillance, while brain magnetic resonance imaging (MRI), positron emission tomography/CT (PET/CT), bone scintigraphy, abdominal CT, or ultrasonography was performed when clinically indicated. Recurrence was defined as locoregional recurrence or distant metastasis after curative-intent resection and was determined from follow-up examination reports, with pathological confirmation used when available. The date of recurrence was defined as the date of the first imaging examination showing findings consistent with recurrence. Recurrence-free survival (RFS) and disease-free survival (DFS) were evaluated as recurrence-related outcomes. RFS was defined as the time from surgical resection to the recurrence date defined above. DFS was defined as the time from surgical resection to the first documented recurrence or death from any cause, whichever occurred first. Patients without endpoint events were censored at the last available disease assessment or follow-up before the administrative cutoff date, and those lost to follow-up were censored at the date on which they were last known to be event-free. Overall survival (OS) was not evaluated as a study endpoint because only 22 deaths had occurred by the administrative cutoff date and complete, uniformly ascertained vital status data were not available across all centers. Therefore, this study focused on recurrence-based endpoints.

Statistical analysis

Continuous variables are presented as medians and interquartile ranges (IQRs) and were compared using Student’s t-test or the Mann-Whitney U test, as appropriate. Categorical variables are presented as counts and percentages and were compared using the chi-square test or Fisher’s exact test. Normality was assessed using the Shapiro-Wilk test. Agreement between manual measurements and measurements derived from automated segmentation was evaluated using the intraclass correlation coefficient (ICC) and Bland-Altman plots. Propensity score matching (PSM) was used for baseline organ-volumetric comparisons between patients with NSCLC and healthy controls. Propensity scores were estimated using age, sex, height, and weight. Nearest-neighbor matching was performed with a caliper of 0.1, and standardized mean differences (SMDs) were used to assess covariate balance before and after matching. Pearson’s correlation coefficient (r) was used to evaluate associations between continuous variables, and the point-biserial correlation coefficient (r_pb) was used for associations between continuous and binary variables. One-way analysis of variance (ANOVA) was applied to test differences in continuous variables across multi-category groups, and effect size was reported as η². Optimal cutoff values for organ volumes and indices were determined using X-tile (Yale University School of Medicine, New Haven, CT, USA). Kaplan-Meier curves and the log-rank test were used to compare DFS and RFS between organ-defined groups. Multivariable Cox proportional hazards regression was performed to assess associations between organ volumes/indices and DFS/RFS. Covariates were selected a priori based on clinical relevance and availability, including age, sex, T stage, pathological types, diabetes, hypertension, smoking, drinking, BMI, albumin, NLR, and PLR. Missing covariate data were handled using multiple imputation by chained equations with 20 imputed datasets and 20 iterations. Cox models were fitted separately within each imputed dataset, and estimates were pooled according to Rubin’s rules. Collinearity among covariates was assessed using generalized variance inflation factors (GVIFs), with degree-of-freedom adjustment for categorical variables. The proportional hazards assumption was tested and satisfied for all Cox models. Subgroup analyses were conducted across clinically relevant strata, and hazard ratios (HRs) with 95% confidence intervals (CIs) were reported. To evaluate the robustness of the primary findings, two sensitivity analyses were performed. First, the participating center was added to the multivariable Cox models. Second, to assess the potential influence of shorter follow-up among recently treated patients, the multivariable Cox analyses were repeated after excluding patients treated in 2025. All tests were two-sided, and P<0.05 was considered statistically significant. Statistical analyses were performed using R (R Foundation for Statistical Computing, version 4.3.0) and Python (Python Software Foundation, version 3.7.12).

Results

Patient characteristics

A total of 1,762 patients with NSCLC from five medical centers were included in the final analysis, and all participants met the predefined inclusion and exclusion criteria. Baseline demographic and clinical characteristics by center are summarized in Table 1. The median age of the full NSCLC cohort was 60 years, and 934 patients (53.0%) were women. The administrative cutoff date for follow-up was December 31, 2025. The median follow-up duration was 39.7 months for RFS and 39.4 months for DFS. During follow-up, 198 RFS events and 219 DFS events occurred.

We also included 3,263 healthy controls from the physical examination center of Wuhan Union Hospital. Compared with the NSCLC cohort, the healthy cohort had a higher proportion of males (62.7% vs. 47.0%, P<0.001) and a younger median age (43 vs. 60 years, P<0.001). Differences were also observed in height (167 vs. 162 cm, P<0.001) and weight (68 vs. 61 kg, P<0.001). To reduce these baseline imbalances for patient-controlled organ-volumetric comparisons, PSM was performed. After matching, 1,202 patients with NSCLC and 1,202 healthy controls remained. Baseline characteristics before and after PSM are shown in Table S3, and the PSM results and subsequent organ-volume comparisons are summarized in Figure 2; specifically, Figure 2A shows that the absolute SMDs for the matched covariates were substantially reduced after matching. The PSM-matched cohort was used only for baseline organ-volumetric comparisons; all prognostic analyses were performed in the full NSCLC cohort of 1,762 patients.

Figure 2 Differences in organ volume between NSCLC patients and healthy controls after PSM. (A) The love plot shows the changes in SMD of each covariate before and after PSM. (B-H) The bean plots illustrate the difference in organ volume between NSCLC patients and healthy controls after PSM. **, P<0.01; ***, P<0.001. LAGV, left adrenal gland volume; LKV, left kidney volume; LVV, liver volume; NSCLC, non-small cell lung cancer; PNV, pancreas volume; PSM, propensity score matching; RAGV, right adrenal gland volume; RKV, right kidney volume; SMD, standardized mean difference; SPV, spleen volume.

Abdominal organ metrics in patients with NSCLC and healthy controls

Using an automated approach based on deep learning, we performed segmentation of seven abdominal organs from preoperative CT images and quantified organ volumes (Figure 1). Pairwise correlations among organ volumes varied in magnitude (Figure 3A). Strong positive correlations were observed between RKV and LKV (r=0.859, P<0.001) and between RAGV and LAGV (r=0.834, P<0.001). Organ volumes were also associated with age and BSA (Figure 3B). All organ volumes were positively correlated with BSA (range of r: 0.246–0.556; all P<0.001). RAGV (r=0.107, P<0.001) and LAGV (r=0.128, P<0.001) showed modest positive correlations with age, whereas SPV, LVV, and PNV were negatively correlated with age (range of r: −0.144 to −0.062; all P<0.05). To account for differences in body size between individuals, organ volumes were normalized by BSA to derive organ indices. Associations between organ volumes/indices and clinical variables are summarized in Tables S4,S5. Ridgeline plots indicated similar distributions of organ volumes and indices across the participating centers. Principal component analysis (PCA) demonstrated substantial overlap in the distributions of organ volumes and indices among centers, suggesting good consistency (Figure S2). Despite differences in sex distribution across centers, organ volumes and indices were generally comparable between male and female patients (Figure S3).

Figure 3 Visualization of organ volume correlations and their associations with clinical parameters. (A) Correlation matrix of organ volumes. (B) Scatter plots showing correlations of organ volumes with age and BSA. ***, P<0.001. BSA, body surface area; LAGV, left adrenal gland volume; LKV, left kidney volume; LVV, liver volume; PNV, pancreas volume; RAGV, right adrenal gland volume; RKV, right kidney volume; SPV, spleen volume.

Compared with matched healthy controls, patients with NSCLC showed higher SPV (184.5 vs. 165.3, P<0.001), RKV (137.4 vs. 132.2, P=0.009), LKV (141.7 vs. 136.9, P=0.002), LVV (1,318.8 vs. 1,261.1, P<0.001), RAGV (2.8 vs. 2.5, P<0.001), and LAGV (3.3 vs. 3.0, P<0.001), whereas PNV was significantly lower (72.5 vs. 77.3, P<0.001) (Figure 2B-2H). After normalization to organ indices, the overall pattern remained largely consistent. Specifically, SPI (106.3 vs. 97.4, P<0.001), RKI (80.6 vs. 77.8, P=0.002), LKI (83.1 vs. 80.6, P<0.001), LVI (775.1 vs. 742.7, P<0.001), RAGI (1.6 vs. 1.5, P<0.001), and LAGI (1.9 vs. 1.8, P<0.001) were increased, whereas PNI was significantly decreased (42.5 vs. 45.5, P<0.001) (Figure S4).

Associations between abdominal organ features and clinical outcomes

We investigated the associations between organ volumes/indices and clinical outcomes. In univariable Cox regression analyses, PNV and PNI were associated with RFS and DFS, and LVI was associated with RFS (Figure 4A). To account for potential confounding, we further evaluated these associations using multivariable Cox proportional hazards models adjusted for the predefined covariates described in the “Statistical analysis” section. Collinearity diagnostics showed no substantial collinearity in the primary multivariable models, with maximum adjusted GVIFs of 1.425 for RFS and 1.426 for DFS. In the multivariable analyses, lower PNV (RFS: HR =0.735, P<0.001; DFS: HR =0.791, P=0.002), lower PNI (RFS: HR =0.729, P<0.001; DFS: HR =0.787, P=0.001), and lower LAGI (RFS: HR =0.845, P=0.030; DFS: HR =0.833, P=0.013) were independently associated with shorter RFS and DFS (Figure 4A; Table 2). In addition, lower LAGV was independently associated with shorter DFS (HR =0.843, 95% CI: 0.725–0.981, P=0.03), and lower LVI was independently associated with shorter RFS (HR =0.834, 95% CI: 0.713–0.976, P=0.02). To further assess prognostic relevance, optimal cutoff values for organ volumes and indices were determined using X-tile (Figure S5). When organ features were analyzed as categorical variables, multivariable Cox regression yielded consistent results. Specifically, PNV (RFS: HR =0.578, P<0.001; DFS: HR =0.639, P=0.002), PNI (RFS: HR =0.570, P<0.001; DFS: HR =0.644, P=0.005), and LAGI (RFS: HR =0.656, P=0.009; DFS: HR =0.629, P=0.002) remained independent predictors of both RFS and DFS. Moreover, LAGV was an independent predictor of DFS (HR =0.694, P=0.002), whereas LVI was an independent predictor of RFS (HR =0.728, P=0.03; Table S6).

Figure 4 Association of organ volumes and indices with clinical outcomes. (A) Bubble charts illustrate the association between continuous variables of organ features and clinical outcomes. HRs were derived from multivariable Cox proportional hazards models, with each continuous organ metric scaled per 1-standard-deviation increase, and adjusted for age, sex, T stage, pathological types, diabetes, hypertension, smoking, drinking, BMI, albumin, NLR, and PLR. (B-E) Kaplan-Meier survival curves of DFS in patients stratified by different organ volumes and indices. (F-I) Kaplan-Meier survival curves of RFS in patients stratified by different organ volumes and indices. The optimal cutoff values for organ volume and index were obtained using the X-tile software. BMI, body mass index; DFS, disease-free survival; HR, hazard ratio; LAGI, left adrenal gland index; LAGV, left adrenal gland volume; LKI, left kidney index; LKV, left kidney volume; LVI, liver index; LVV, liver volume; NLR, neutrophil-to-lymphocyte ratio; PLR, platelet-to-lymphocyte ratio; PNI, pancreas index; PNV, pancreas volume; RAGI, right adrenal gland index; RAGV, right adrenal gland volume; RFS, recurrence-free survival; RKI, right kidney index; RKV, right kidney volume; SPI, spleen index; SPV, spleen volume.

Kaplan-Meier analyses showed that PNV, PNI, LAGI, and LAGV significantly stratified DFS (all log-rank P<0.05; Figure 4B-4E), whereas LAGI, LVI, PNI, and PNV significantly stratified RFS (all log-rank P<0.05; Figure 4F-4I). The log-rank P values were 0.04 for LAGV and 0.04 for LVI. By contrast, the remaining organ volumes and indices shown in Figure S6 did not demonstrate significant survival discrimination. Subgroup analyses further indicated that the prognostic performance of PNV, PNI, LVI, and LAGI was generally consistent across clinically relevant subgroups (Figures S7,S8).

Sensitivity analyses supported the robustness of the primary findings. After additional adjustment for the participating center, the associations of PNV, PNI, and LAGI with RFS and DFS remained stable, while other findings were generally directionally consistent (Table S7). Collinearity remained low in the center-adjusted models, with maximum adjusted GVIFs of 1.462 for RFS and 1.461 for DFS. After excluding 21 patients treated in 2025, none of whom had RFS or DFS events, the cohort included 1,741 patients with 198 RFS events and 219 DFS events, and the main associations remained nearly unchanged (Table S8).

Discussion

Based on preoperative CT from a multicenter cohort of patients with resectable NSCLC, we used an nnU-Net-based deep learning model to automatically segment seven abdominal organs, quantify organ volumes, and derive BSA-normalized organ indices. We observed baseline differences in organ volumes and indices between patients with NSCLC and healthy controls, as well as correlation patterns among organ volumes and their associations with age and BSA. In the full NSCLC cohort, lower PNV, PNI, and LAGI were independently associated with shorter RFS and DFS after adjustment for clinical covariates. Lower LAGV was associated with DFS only, whereas lower LVI was associated with RFS only.

The present findings are best interpreted in the context of host-related risk rather than as evidence that abdominal organ volumes directly determine postoperative recurrence. Abdominal solid organ volumes are more likely to represent integrated imaging phenotypes of systemic conditions, including metabolic reserve, immune-inflammatory status, nutritional status, comorbidity burden, and cancer-associated catabolism. Although the multivariable models adjusted for available tumor- and host-related factors, residual confounding by unmeasured inflammatory markers, nutritional parameters, detailed body-composition measures, perioperative condition, and treatment selection cannot be excluded. Therefore, abdominal organ volumetry should be viewed as a complementary CT-derived imaging biomarker that may help characterize patient-level vulnerability and refine recurrence risk stratification.

We did not observe a stable independent association between SPV/SPI and postoperative recurrence risk in our resectable NSCLC cohort. Although a trend was noted in univariable analyses, the association was not statistically significant after adjustment for age, stage, histologic subtype, and inflammatory markers. This finding differs from reports in advanced populations receiving immunotherapy. Galland et al. (20) reported that in advanced NSCLC treated with immune checkpoint inhibitors, either larger baseline SPV or an increase in SPV during treatment was associated with worse survival (P=0.001) and may serve as an imaging surrogate of an immunosuppressive state. Fu et al. (23) observed a negative association between baseline SPV and survival in a hepatocellular carcinoma (HCC) cohort receiving immunotherapy and reported a nonlinear relationship using restricted cubic spline analysis, suggesting a potential threshold effect. Aslan et al. (24) further showed in metastatic renal cell carcinoma (mRCC) that an increase in SPV during treatment was associated with a higher risk of immunotherapy failure. These discrepancies may reflect differences in patient populations and treatment contexts. In advanced disease, changes in SPV during immunotherapy may better capture the interplay among tumor burden, immune responses, and immune modulation, potentially involving myeloid-derived suppressor cell (MDSC) expansion and broader immune dysregulation (25). In contrast, our cohort primarily comprised early-stage resectable patients with less pronounced systemic immune and inflammatory perturbations, which may limit the prognostic signal captured by a single preoperative SPV measurement. Moreover, prior positive studies mainly evaluated treatment-associated SPV dynamics, whereas our analysis relied on a single preoperative assessment. Finally, SPV may overlap with peripheral inflammatory biomarkers such as NLR and PLR; this overlap may attenuate its independent prognostic contribution in multivariable models.

In contrast to the spleen, pancreatic volume and the corresponding index showed a more stable and consistent independent association with both RFS and DFS in our cohort. Fukumoto et al. reported that pancreatic volume normalized by BSA was significantly reduced and associated with surgical prognosis, suggesting that pancreatic atrophy may reflect a chronic burden of inflammation and fibrosis as well as impaired nutritional and metabolic reserve (26). Although this evidence was derived primarily from primary pancreatic malignancies, it supports a broader concept that pancreatic volume may serve as an imaging biomarker integrating digestive and absorptive capacity, exocrine function, glucose and lipid homeostasis, and overall metabolic reserve (27). Reduced pancreatic volume may be related to acinar cell atrophy, degranulation, and fibrotic remodeling. Given the high protein synthesis and secretory demands of acinar cells, they are sensitive to amino acid availability and metabolic stress. In murine models, stress conditions suppress pathways involved in protein synthesis, such as mechanistic target of rapamycin complex 1 (mTORC1), which may impair digestive enzyme synthesis and zymogen granule maintenance, contributing to reductions in acinar structure and pancreatic volume (28). In addition, impaired islet cell function may reduce local insulin signaling and disrupt the insulo-acinar axis, further exacerbating exocrine dysfunction and organ atrophy (29). Clinically, reduced pancreatic volume may indicate compromised nutritional and metabolic reserve, which could increase recurrence risk by limiting postoperative recovery, attenuating immune surveillance against residual tumor cells or micrometastatic disease, and reducing tolerance to adjuvant therapies (30). Similarly, LAGI and LAGV were independently associated with clinical outcomes, suggesting that morphologic features of organs related to the hypothalamic-pituitary-adrenal (HPA) axis may provide additional information for recurrence risk stratification. The adrenal gland is a key effector organ for glucocorticoid secretion and stress responses, playing an important role in inflammatory regulation, immune homeostasis, and metabolic stress adaptation (31). We observed that a lower LAGI was associated with shorter recurrence-related outcomes, which may be related to the cumulative effects of chronic stress and systemic inflammatory burden. Sustained dysregulation of the HPA axis may alter long-term patterns of glucocorticoid exposure, including basal levels, stress responsiveness, and circadian rhythmicity (32,33). Prolonged abnormalities in glucocorticoid signaling may contribute to immunosuppression and reduced immune effector function and may be associated with immune inflammatory imbalance, thereby weakening postoperative immune surveillance and clearance of residual tumor cells or micrometastatic lesions and increasing the risk of early recurrence (34,35). In parallel, chronic comorbidities and long-term medication exposure may also contribute to adrenal morphologic changes and their association with recurrence risk. LVI showed an association with RFS in the primary and sensitivity analyses, but not with DFS. This finding may reflect liver-related metabolic and synthetic reserve, consistent with prior evidence linking LVV or functional reserve to outcomes in oncology (36,37), although the effect was less consistent than that of pancreatic and left adrenal metrics.

To our knowledge, this is the first study in resectable NSCLC to systematically leverage preoperative CT and automated segmentation using deep learning to quantify seven abdominal organs and to evaluate their associations with postoperative recurrence risk. We achieved high-throughput extraction of volumes and normalized indices for seven organs, including the pancreas, adrenal glands, and liver. Using the healthy control cohort as a reference, we characterized baseline organ-volumetric differences between patients with NSCLC and healthy individuals. In the full NSCLC cohort, pancreatic volume, pancreatic index, and LAGI were identified as independent predictors of both RFS and DFS. These findings support the potential value of CT-derived abdominal organ quantitative metrics as readily obtainable imaging biomarkers of patient-related host characteristics. All quantitative metrics in this study were derived from routine preoperative CT. The use of automated segmentation using deep learning mitigated the efficiency and reproducibility limitations of manual delineation and enabled standardized, high-throughput analysis across multicenter data. This technical framework extends opportunistic CT quantification beyond traditional body composition measures to include abdominal organs, offering a complementary imaging approach for assessing patient-level factors.

Several limitations should be acknowledged. First, the retrospective design may introduce selection bias and residual confounding, despite the use of multivariable Cox regression, multiple imputation, and sensitivity analyses. Second, the healthy control cohort was derived from a single physical examination center, whereas the NSCLC cohort was collected from five medical centers; therefore, the patient-control comparison may be affected by unmeasured population and center-level differences. Third, although all centers used broadly similar CT-based surveillance protocols and center-adjusted sensitivity analyses yielded generally consistent results, differences in follow-up intensity or additional imaging use may have influenced the timing of recurrence detection. Fourth, the lack of mature and uniformly ascertained OS data limited our ability to evaluate the association between organ volumetrics and OS. Finally, our analyses were based on a single preoperative CT time point and could not characterize perioperative or post-treatment changes in organ morphology. Future multicenter prospective studies with longer follow-up, multicenter control cohorts, and serial CT imaging are needed to validate these findings.

Conclusions

This multicenter study identified a distinct pattern of baseline abdominal organ volumetric differences between patients with surgically resected NSCLC and healthy controls. Pancreatic and adrenal gland volumetric metrics were independently associated with RFS and DFS after surgery. CT-derived abdominal organ quantitative features may provide complementary information for postoperative recurrence risk stratification beyond conventional clinicopathologic factors.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the STROBE reporting checklist. Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0445/rc

Data Sharing Statement: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0445/dss

Peer Review File: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0445/prf

Funding: This work was supported by the National Natural Science Foundation of China (Nos. 825B2054, 82472058, 82172034, and 82502329); the Major Special Project for Technology Innovation of Hubei Province (No. 2023BCB014); the Noncommunicable Chronic Diseases-National Science and Technology Major Project (No. 2024ZD0522802); and the Research Fund of Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology (No. AI2025B06).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0445/coif). All authors report that the work was supported by competitive research grants from National Natural Science Foundation of China, the Major Special Project for Technology Innovation of Hubei Province, the Noncommunicable Chronic Diseases-National Science and Technology Major Project, and the Research Fund of Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology; such support was provided to the participating institutions. The authors have no other conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the ethics committees of all participating centers, including Hubei Provincial Hospital of Traditional Chinese Medicine (No. HBZY2026-C010-01), Wuhan Union Hospital (No. S0802), The Affiliated Traditional Chinese Medicine Hospital of Southwest Medical University (No. BY2025078), The First Affiliated Hospital of Bengbu Medical University (No. KY041), and Renmin Hospital of Wuhan University (No. WDRY2023-K004). The requirement for written informed consent was waived because of the retrospective nature of the study.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Bray F, Laversanne M, Sung H, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2024;74:229-63. [Crossref] [PubMed]
2. Rajaram R, Huang Q, Li RZ, et al. Recurrence-Free Survival in Patients With Surgically Resected Non-Small Cell Lung Cancer: A Systematic Literature Review and Meta-Analysis. Chest 2024;165:1260-70. [Crossref] [PubMed]
3. Rami-Porta R, Nishimura KK, Giroux DJ, et al. The International Association for the Study of Lung Cancer Lung Cancer Staging Project: Proposals for Revision of the TNM Stage Groups in the Forthcoming (Ninth) Edition of the TNM Classification for Lung Cancer. J Thorac Oncol 2024;19:1007-27. [Crossref] [PubMed]
4. Zer A, Ahn MJ, Barlesi F, et al. Early and locally advanced non-small-cell lung cancer: ESMO Clinical Practice Guideline for diagnosis, treatment and follow-up. Ann Oncol 2025;36:1245-62. [Crossref] [PubMed]
5. Jung HA, Lim J, Choi YL, et al. Clinical, Pathologic, and Molecular Prognostic Factors in Patients with Early-Stage EGFR-Mutant NSCLC. Clin Cancer Res 2022;28:4312-21. [Crossref] [PubMed]
6. Shimomura M, Miyagawa-Hayashino A, Omatsu I, et al. Spread through air spaces is a powerful prognostic predictor in patients with completely resected pathological stage I lung adenocarcinoma. Lung Cancer 2022;174:165-71. [Crossref] [PubMed]
7. Fearon K, Strasser F, Anker SD, et al. Definition and classification of cancer cachexia: an international consensus. Lancet Oncol 2011;12:489-95. [Crossref] [PubMed]
8. Al-Sawaf O, Weiss J, Skrzypski M, et al. Body composition and lung cancer-associated cachexia in TRACERx. Nat Med 2023;29:846-58. [Crossref] [PubMed]
9. Mizuguchi S, Izumi N, Tsukioka T, et al. Neutrophil-lymphocyte ratio predicts recurrence in patients with resected stage 1 non-small cell lung cancer. J Cardiothorac Surg 2018;13:78. [Crossref] [PubMed]
10. Hanahan D. Hallmarks of Cancer: New Dimensions. Cancer Discov 2022;12:31-46. [Crossref] [PubMed]
11. Abbosh C, Frankell AM, Harrison T, et al. Tracking early lung cancer metastatic dissemination in TRACERx using ctDNA. Nature 2023;616:553-62. [Crossref] [PubMed]
12. Bates DDB, Pickhardt PJ. CT-Derived Body Composition Assessment as a Prognostic Tool in Oncologic Patients: From Opportunistic Research to Artificial Intelligence-Based Clinical Implementation. AJR Am J Roentgenol 2022;219:671-80. [Crossref] [PubMed]
13. Mantz L, Mercaldo ND, Simon J, et al. Preoperative Chest CT Myosteatosis Indicates Worse Postoperative Survival in Stage 0-IIB Non-Small Cell Lung Cancer. Radiology 2025;314:e240282. [Crossref] [PubMed]
14. Tao J, Fang J, Chen L, et al. Increased adipose tissue is associated with improved overall survival, independent of skeletal muscle mass in non-small cell lung cancer. J Cachexia Sarcopenia Muscle 2023;14:2591-601. [Crossref] [PubMed]
15. Xie H, Wei L, Liu M, et al. Preoperative computed tomography-assessed sarcopenia as a predictor of complications and long-term prognosis in patients with colorectal cancer: a systematic review and meta-analysis. Langenbecks Arch Surg 2021;406:1775-88. [Crossref] [PubMed]
16. Lading T, Media AS, Tellefsen B, et al. Sarcopenia and survival in surgically treated non-small cell lung cancer: A systematic review and meta-analysis. Eur J Surg Oncol 2025;51:110538. [Crossref] [PubMed]
17. Gezer NS, Bandos AI, Beeche CA, et al. CT-derived body composition associated with lung cancer recurrence after surgery. Lung Cancer 2023;179:107189. [Crossref] [PubMed]
18. Zou F, Zhao J, Zhong L, et al. CT-based body composition and inflammatory nutritional biomarker nomogram for predicting early postoperative recurrence of non-small cell lung cancer: a multicenter study. J Thorac Dis 2025;17:8046-62. [Crossref] [PubMed]
19. McGovern J, Mackay C, Freireich R, et al. The Relationship between Liver Volume, Clinicopathological Characteristics and Survival in Patients Undergoing Resection with Curative Intent for Non-Metastatic Colonic Cancer. Tomography 2024;10:349-59. [Crossref] [PubMed]
20. Galland L, Lecuelle J, Favier L, et al. Splenic Volume as a Surrogate Marker of Immune Checkpoint Inhibitor Efficacy in Metastatic Non Small Cell Lung Cancer. Cancers (Basel) 2021;13:3020. [Crossref] [PubMed]
21. Aslan V, Karabörk Kılıç AC, Rustamova Cennet N, et al. Splenic index score as a predictor of outcomes in metastatic non small cell lung cancer patients treated with immune checkpoint inhibitors. Sci Rep 2025;15:15781. [Crossref] [PubMed]
22. Wasserthal J, Breit HC, Meyer MT, et al. TotalSegmentator: Robust Segmentation of 104 Anatomic Structures in CT Images. Radiol Artif Intell 2023;5:e230024. [Crossref] [PubMed]
23. Fu X, Jiang S, Li Y, et al. Spleen Volume Dynamics and Survival Outcomes in HCC Patients Undergoing Immune Checkpoint Inhibitors: A Retrospective Analysis. J Hepatocell Carcinoma 2025;12:1069-82. [Crossref] [PubMed]
24. Aslan V, Karabörk Kılıç AC, Özet A, et al. The role of spleen volume change in predicting immunotherapy response in metastatic renal cell carcinoma. BMC Cancer 2023;23:1045. [Crossref] [PubMed]
25. Wu C, Hua Q, Zheng L. Generation of Myeloid Cells in Cancer: The Spleen Matters. Front Immunol 2020;11:1126. [Crossref] [PubMed]
26. Fukumoto T, Watanabe T, Hirai I, et al. Pancreatic volume is one of the independent prognostic factors for resectable pancreatic ductal adenocarcinomas. J Hepatobiliary Pancreat Sci 2016;23:472-9. [Crossref] [PubMed]
27. Ichida H, Imamura H, Takahashi A, et al. Evaluation of pancreatic morphometric parameters, exocrine function, and nutritional status and their causal relationships in long-term survivors following pancreatectomy. Surgery 2024;176:1189-97. [Crossref] [PubMed]
28. Sans MD, Crozier SJ, Vogel NL, et al. Dietary Protein and Amino Acid Deficiency Inhibit Pancreatic Digestive Enzyme mRNA Translation by Multiple Mechanisms. Cell Mol Gastroenterol Hepatol 2021;11:99-115. [Crossref] [PubMed]
29. Hu C, Chen Y, Yin X, et al. Pancreatic endocrine and exocrine signaling and crosstalk in physiological and pathological status. Signal Transduct Target Ther 2025;10:39. [Crossref] [PubMed]
30. Whitcomb DC, Buchner AM, Forsmark CE. AGA Clinical Practice Update on the Epidemiology, Evaluation, and Management of Exocrine Pancreatic Insufficiency: Expert Review. Gastroenterology 2023;165:1292-301. [Crossref] [PubMed]
31. Quatrini L, Ugolini S. New insights into the cell- and tissue-specificity of glucocorticoid actions. Cell Mol Immunol 2021;18:269-78. [Crossref] [PubMed]
32. Ring M. An Integrative Approach to HPA Axis Dysfunction: From Recognition to Recovery. Am J Med 2025;138:1451-63. [Crossref] [PubMed]
33. Olejniczak I, Oster H, Ray DW. Glucocorticoid circadian rhythms in immune function. Semin Immunopathol 2022;44:153-63. [Crossref] [PubMed]
34. Acharya N, Madi A, Zhang H, et al. Endogenous Glucocorticoid Signaling Regulates CD8(+) T Cell Differentiation and Development of Dysfunction in the Tumor Microenvironment. Immunity 2020;53:658-671.e6. [Crossref] [PubMed]
35. Khadka S, Druffner SR, Duncan BC, et al. Glucocorticoid regulation of cancer development and progression. Front Endocrinol (Lausanne) 2023;14:1161768. [Crossref] [PubMed]
36. Zuo D, Li Y, Liu H, et al. Value of Non-tumoral Liver Volume in the Prognosis of Large Hepatocellular Carcinoma Patients After R0 Resection. J Clin Transl Hepatol 2023;11:560-71. [Crossref] [PubMed]
37. D'Avola D, Granito A, Torre-Aláez M, et al. The importance of liver functional reserve in the non-surgical treatment of hepatocellular carcinoma. J Hepatol 2022;76:1185-98. [Crossref] [PubMed]