Abstract
Background
Survival rates in advanced non-small cell lung cancer (NSCLC) vary substantially, and changes in host body composition may affect prognosis. Serial computed tomography (CT) can repeatedly quantify skeletal muscle index (SMI) at the L3 level, but its dynamic predictive value and clinically usable thresholds have not been systematically validated. This study systematically evaluates the predictive value of longitudinal SMI changes based on serial CT for overall survival (OS), validate early treatment thresholds, and test its incremental effect over clinical models and Response Evaluation Criteria In Solid Tumors (RECIST).
Methods
A single-center retrospective cohort was established, including 451 patients with stage IIIB–IV NSCLC who received systemic therapy from 2018 to 2023. L3 SMI was extracted at baseline and follow-up CT; segmentation consistency intraclass correlation coefficient (ICC) ≥0.97. ΔSMI% was modeled as a time-dependent exposure (per 5% decrease) in multivariable Cox, and a 12-week landmark analysis was used to evaluate early ΔSMI% ≤−5%. Baseline missing data were handled with MICE (m=20). Combined with RECIST, we compared the C-index, Uno time-dependent area under the curve (AUC), and 12-month net reclassification improvement (NRI), and conducted decision curve analysis and 1,000 bootstrap calibration.
Results
Each 5% decrease in ΔSMI% was independently associated with increased death risk [hazard ratio (HR) =1.16, P=0.001], and higher baseline SMI was associated with reduced risk (per 10 cm2/m2 HR =0.87, P=0.006); the association between ΔSMI% and death risk was approximately linear (overall P=0.001). At the 12-week landmark analysis, early ΔSMI% ≤−5% was associated with worse survival (log-rank P=0.003), with an adjusted HR =1.68 (P=0.001). The correlation between early ΔSMI% and RECIST was weak (r=−0.18, P=0.001). On top of clinical covariates, adding baseline SMI and time-dependent ΔSMI% increased the corrected C-index from 0.67 to 0.71; 12-month NRI =0.19; AUC at 6/12 months was 0.71/0.70. Decision curves showed higher net benefit than treat all/none within threshold ranges 0.23–0.47 and 0.34–0.58; calibration was good.
Conclusions
SMI dynamic metrics derived from serial CT constitute an independent and actionable prognostic signal in advanced NSCLC, enable early identification of high-risk patients within routine follow-up windows, and provide incremental gains in discrimination and net benefit over clinical models.
Keywords: Non-small cell lung cancer (NSCLC), skeletal muscle index (SMI), multivariable prognostic model validation, time-dependent Cox, survival prediction
Highlight box.
Key findings
• Progressive decline of skeletal muscle index (SMI) measured from routine serial thoracic computed tomography (CT) independently predicts worse overall survival (OS) in patients with non-small cell lung cancer (NSCLC).
• Dynamic SMI loss over follow-up outperforms single baseline SMI as a robust prognostic marker for long-term survival stratification.
• SMI reduction is independently associated with worse OS, and may serve as an early indicator of impaired host reserve that warrants further nutritional and functional assessment.
What is known and what is new?
• Low baseline SMI is associated with unfavorable survival outcomes in lung cancer, yet most prior studies rely on one-time CT measurement without tracking longitudinal muscle changes during treatment.
• We verified that continuous SMI decline on standard follow-up CT provides extra prognostic value, enabling real-time risk assessment without additional radiation exposure.
What is the implication, and what should change now?
• Clinicians can quantify SMI from existing serial thoracic CT scans to dynamically stratify survival risk. Early SMI decline may identify patients who should be considered for additional nutritional, functional, or supportive-care assessment. Prospective interventional studies are required to determine whether SMI-guided interventions improve treatment tolerance or survival.
Introduction
Background
Lung cancer persists as a predominant cause of cancer mortality worldwide and in China, and non-small cell lung cancer (NSCLC) accounts for the vast majority. Many patients are already at stage IIIB–IV at diagnosis and require long-term systemic therapy and imaging follow-up (1,2). Immunotherapy and targeted therapy have improved outcomes, but overall survival (OS) varies substantially, and staging, Eastern Cooperative Oncology Group performance status (ECOG PS), and tumor burden alone cannot fully capture individual risk (3). Tumor-related cachexia features sustained skeletal muscle loss and is associated with fatigue, treatment intolerance, and poor survival (4). Chest and abdominal CT is routine for advanced NSCLC management and follow-up. Quantifying skeletal muscle area (SMA) at the L3 level and converting it to the SMI offers repeatability and comparability, providing a practical way to assess host status without extra tests (5,6). CT-based body composition assessment is gradually entering oncologic research and management.
Rationale and knowledge gap
Existing studies mostly treat sarcopenia as a baseline cross-sectional feature associated with survival or treatment toxicity, but have paid insufficient attention to changes in muscle mass during treatment (7). Muscle mass fluctuates continuously with disease progression and treatment stress; a single measurement cannot reflect the rate of depletion and easily conflates baseline reserves with progressive decline (8). Follow-up imaging time points are driven by clinical decision-making; if exposure is defined only after treatment or if a fixed timepoint difference is used to substitute for a continuous process, immortal time bias may be introduced and interpretability may be weakened (9). Studies differ considerably in threshold definitions and treatment populations, limiting the generalizability of conclusions. Whether dynamic SMI can yield an actionable early warning threshold within routine follow-up windows, whether it complements RECIST response, and whether it provides incremental value in discrimination, reclassification, and net benefit over clinical models still lack systematic evidence centered on time-dependent methods. Distinct from prior studies relying solely on single baseline SMI or static muscle loss measurements, the novelty of the present research lies in four integrated, unreported analytical components: (I) modeling serial ΔSMI% as a continuously updated time-dependent covariate to eliminate immortal time bias and capture dynamic muscle depletion over treatment; (II) validating a standardized 12-week early CT assessment window for rapid risk stratification in routine follow-up; (III) quantifying the weak independent correlation between longitudinal SMI decline and RECIST tumor response to confirm complementary prognostic information; (IV) comprehensively quantifying the incremental discrimination, reclassification and clinical net benefit of dynamic SMI beyond established clinical multivariable prognostic models. This work is a prognostic prediction validation study focusing on long-term OS, rather than diagnostic model development, and we aim to fill the gap in existing lung cancer prognostic models through this multi-dimensional integrated analysis of longitudinal skeletal muscle metrics.
Objective
We established a single-center retrospective cohort, extracted L3 SMI from serial CT before and after treatment, calculated ΔSMI% and used a time-dependent Cox model to evaluate its predictive effect on OS, and validated an early threshold with a 12-week landmark analysis. Combined with baseline SMI and RECIST, we evaluated discrimination, calibration, and net benefit to clarify the incremental value of dynamic SMI for early identification of high-risk patients. This study belongs to TRIPOD Type 4 validation research; we did not construct a new multivariate risk prediction model, and our core objective was to validate the incremental prognostic performance of dynamic SMI based on established clinical predictive covariates. We present this article in accordance with the TRIPOD reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1968/rc).
Methods
Study Design
This is a registry-based single-center retrospective cohort study using consecutive clinical and serial CT imaging data extracted from our institutional electronic medical record and radiology archive database. The study was conducted at a single tertiary oncology-focused hospital: The Second Affiliated Hospital of Henan University of Chinese Medicine, located in central China (Henan Province). Only one medical center contributed all study participants, with no multi-center collaboration or external hospital data incorporated. No separate external validation dataset or random/non-random split sub-cohort was established; all eligible participants were included in a unified single dataset for all primary, secondary and sensitivity analyses. Internal validation of predictive model performance was conducted via 1,000 bootstrap resampling to correct for overfitting optimism, in accordance with standard prognostic model validation methodology.
We consecutively included advanced NSCLC patients who received systemic antitumor therapy at our cancer center from 01/01/2018 to 31/12/2023, with follow-up until 31/12/2024 or death, whichever occurred first. The final study size was determined by the total number of consecutive eligible patients within the prespecified enrollment period; no a priori formal sample size calculation was performed, and all patients meeting the eligibility criteria were included to maximize statistical power for prognostic model validation. A statistical analysis plan was prespecified and locked before data extraction. The data were accessed for the present analysis on 15/12/2025. All analyses were based on data generated by routine clinical workflows, with no additional intervention in diagnosis or treatment. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study received approval from the Ethics Committee of Henan Provincial Hospital of Traditional Chinese Medicine (The Second Affiliated Hospital of Henan University of Chinese Medicine) (No. HNSZYYYJS202302050). Written informed consent was waived for this retrospective analysis as de-identified data were utilized.
Study population and procedures
Inclusion criteria: age ≥18 years; pathologically confirmed NSCLC with stage IIIB–IV; completion of a contrast-enhanced chest–abdominal CT that included the L3 level within 28 days before initiation of systemic therapy at our institution (chemotherapy, immunotherapy, targeted therapy, or combination regimens) as the baseline scan; at least one follow-up CT that included L3 within 16 weeks after treatment initiation; availability of complete data on demographics, height, weight, ECOG PS, oncologic characteristics, and treatment information.
Exclusion criteria: small cell lung cancer or combined small-cell histology; concurrent presence of other primary malignancies requiring systemic therapy during the same period; receipt of best supportive care only without systemic therapy; CT images lacking the L3 slice or failure of muscle segmentation due to motion or metal artifacts; missing height precluding calculation of SMI.
Consecutive cases were identified through the electronic medical record and image archiving systems, and variables were de-identified and curated; OS outcomes were verified via the hospital medical record system, outpatient follow-up, and telephone follow-up.
Imaging acquisition and processing
CT scanning protocol and examination intervals
All images were derived from routine contrast-enhanced chest-abdominal CT. Multidetector CT scanners were used, with volumetric data acquired in the portal venous phase; tube voltage 100–120 kVp, automatic tube current modulation, slice thickness 2.5–3.0 mm, reconstruction matrix 512×512. Baseline CT was defined as the examination performed within 28 days prior to treatment initiation. Follow-up CT was performed approximately 6–12 weeks after treatment per routine practice, and thereafter at irregular intervals as deemed necessary by the physician. All follow-up CT time points that included L3 for each patient were recorded to construct a longitudinal muscle-mass series.
L3-level skeletal muscle segmentation workflow and quality control
At the level of the L3 vertebral body, a single axial slice at the midpoint of the L3 vertebral body was selected for segmentation; this anatomical criterion was applied consistently to all baseline and follow-up CT examinations for each patient. The selected slice was required to best depict the bilateral psoas, erector spinae, external oblique, internal oblique, and transversus abdominis muscles. A semi-automated thresholding method combined with region growing was used; CT attenuation thresholds for skeletal muscle were set at −29 to +150 Hounsfield units (10). After semi-automated segmentation, manual correction was generally minimal and was primarily limited to refining fascial plane boundaries and excluding adjacent non-muscle soft tissue. Segmentations were independently performed in CoveAI (Beijing Qixun Medical Technology Co., Ltd., Beijing, China) by two trained radiology researchers with 3 and 5 years of experience in CT-based body composition analysis, respectively; both investigators were blinded to patients’ survival outcomes and treatment response at the time of segmentation. The skeletal muscle cross-sectional area (SMA, cm2) of the slice was then automatically calculated. To assess consistency, 40 images were randomly selected for double-rater remeasurement and a two-way random-effects, absolute-agreement ICC was calculated; the same rater repeated measurements on the same subsample after a 2-week interval to obtain test–retest reliability. Image quality was reviewed by a senior radiologist; if the L3 level was poorly visualized or muscle boundaries were indiscernible, the segmentation was deemed unsuccessful and handled according to the exclusion criteria.
Calculation of SMI
Height was taken from the measurement on the day treatment began or the most recent physical examination. SMI was defined as the skeletal muscle cross-sectional area at the L3 level divided by height squared: SMI = SMA (cm2)/height (m)2 (11). Baseline SMI was taken from the baseline CT. To reduce the influence of extreme values, obviously unreasonable height and weight values were source-checked and corrected; if correction was not possible, they were treated as missing and entered into the imputation process.
Definition of longitudinal SMI change metrics
At each follow-up CT, the relative change from baseline (ΔSMI%) was calculated as follows: ΔSMI% = [(SMIt − SMI0)/SMI0] ×100% (12). ΔSMI% was incorporated as a time-dependent covariate in a piecewise-constant manner. Each ΔSMI% value derived from a given CT scan only entered the risk set and contributed to the survival model after the date of that CT examination; specifically, the ΔSMI% measured at a follow-up CT was assigned to the survival interval starting from the date of that CT scan and ending at the next follow-up CT, the date of death, or the date of last follow-up, whichever occurred first. ΔSMI% values were never applied retroactively to time intervals preceding the corresponding CT examination. As an illustrative example: if baseline CT was performed on day 0, and the first follow-up CT was conducted on day 84 (12 weeks), the ΔSMI% calculated at day 84 only affected the hazard estimation for the period from day 84 onward, and was not used for the day 0 to day 84 interval.
A 12-week (±14 days) early assessment window was prespecified, and the ΔSMI% within this window was calculated as the early indicator; an early ΔSMI% ≤−5% was prespecified as a “clinically significant decline” prior to data extraction and analysis (13-15). This threshold was selected based on prior body-composition research in oncology populations, where a 5% relative loss of skeletal muscle has been consistently associated with adverse clinical outcomes and is widely regarded as the minimal change of clinical relevance. This cut-off should be interpreted as a prespecified clinically relevant threshold evaluated via internal validation only, not an externally validated diagnostic cut-point. The individual-level SMI rate of change (slope) was also calculated by performing ordinary least squares linear regression of all available SMI values on time, with the regression coefficient taken as the rate of change per unit time (units: cm2/m2 per week).
Study variables and definitions
Primary exposure variables: dynamic SMI metrics
The primary exposure was time-updated ΔSMI% (continuous variable, with each 5% decrease as the increment unit); secondary exposures included the dichotomous early ΔSMI% (whether ≤−5%) and the SMI slope (continuous variable). To distinguish dynamic from static information, baseline SMI (continuous variable) was included in the model simultaneously.
Outcome variable: definition and ascertainment of OS
OS was defined as the duration from the commencement of systemic therapy to death due to any cause. Patients who remained alive were censored on the date of the last confirmed survival or at the conclusion of the follow-up. Death information was verified from multiple sources, including the inpatient medical record system, outpatient records, and telephone follow-up; in case of discrepancies, the earliest confirmed date of death was used.
Covariates: demographic, oncologic, and treatment information
Prespecified covariates included age, sex, smoking history, body mass index (BMI), ECOG PS, histology (adenocarcinoma or squamous cell carcinoma), stage (IIIB or IV), line of therapy (initial first-line or later-line), treatment category (chemotherapy, immunotherapy, targeted therapy, or combination regimens), baseline tumor burden (sum of the longest diameters of RECIST 1.1 target lesions, mm), and presence of liver metastasis. All covariates were measured at treatment initiation and fixed at baseline.
Radiologic response assessment: RECIST and tumor burden metrics
Response was assessed using RECIST 1.1 criteria at the first post-treatment imaging follow-up. The median interval from baseline CT to the first follow-up CT (i.e., the first RECIST assessment) was 8.8 weeks [interquartile range (IQR), 6.2–11.9 weeks]; baseline CT was obtained within 28 days before treatment initiation. For model incorporation, the percentage change in the sum of the longest diameters of target lesions from baseline was entered as a continuous variable, rather than as a categorical RECIST response [complete response (CR)/partial response (PR)/stable disease (SD)/progressive disease (PD)]. The baseline sum of the longest diameters of target lesions was also recorded as a quantitative tumor burden metric and used in combined models to test the complementarity between host muscle metrics and tumor burden information.
Statistical analysis
Statistical analyses were conducted using R 4.3.2. Continuous variables were described as mean with standard deviation or median with IQR, and categorical variables as frequency and percentage. The primary analysis used a time-dependent Cox model, with ΔSMI% included as a time-updated continuous covariate; effects were presented as hazard ratios (HRs) per 5% decrease in SMI, with adjustment for baseline SMI and prespecified covariates. If the Schoenfeld residuals test indicated violation of the proportional hazards assumption, interaction terms between the violating variables and time were specified or a stratified Cox model was used. To evaluate nonlinearity, a four-knot restricted cubic spline was employed to model the relationship between ΔSMI% and HR. In the presence of a turning point, both the continuous effect and two-piecewise linear effects based on the turning point were presented.
The early-window analysis used a 12-week landmark analysis design. Patients alive at 12 weeks and having completed imaging assessment were included. Early ΔSMI% was regarded as the exposure. Cox analysis of OS beyond the landmark was carried out, and Kaplan-Meier curves and log-rank test results were plotted.
All model comparisons involving RECIST (Model D) were restricted to patients with an available first post-treatment RECIST assessment. For the landmark-based model comparisons, all four nested models (A–D) were fitted in the identical patient subset—patients alive at the 12-week landmark analysis with complete RECIST and covariate data—and shared the same prediction starting time at the landmark, to ensure fair comparison and avoid sample-derived bias.
To conduct a comparison between “dynamic versus static”, four nested models were established: Model A (comprising only clinical covariates), Model B (Model A augmented with baseline SMI), Model C (Model B supplemented with ΔSMI%), and Model D (Model C incorporated with the percentage change in target-lesion longest diameter at the first RECIST assessment, entered as a continuous variable). Discrimination was compared through the application of the Harrell C-index and the Uno time-dependent AUC, and improvements in model fit were evaluated using the likelihood ratio test and Akaike information criterion (AIC). At the 12-month prediction time point, the net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were calculated, with 95% confidence intervals reported. To investigate the complementarity between host muscle metrics and RECIST, the gains in the C-index and NRI of Model C relative to Model D were compared, and the variance inflation factor (VIF) was assessed to ensure the absence of significant multicollinearity.
Clinical utility was appraised through decision curve analysis to evaluate the net benefit across risk-threshold ranges at the 6-month and 12-month time points. Model calibration employed 1,000 bootstrap resamples to acquire optimism-corrected calibration curves and Brier scores.
Sample size adequacy was evaluated based on the events-per-variable (EPV) ratio for multivariable Cox model validation. With 271 observed all-cause death events and up to 12 candidate covariates in the fully adjusted model, the EPV ratio exceeded 20, which is well above the widely accepted minimum threshold of 10 for stable survival model estimation. This sample size was deemed sufficient to support all planned analyses, including discrimination assessment, bootstrap calibration, reclassification improvement, and decision curve analysis.
Missing data were assumed “conditionally missing at random”. Multiple imputation by chained equations (20 imputed datasets) was performed for baseline covariates. Predictive mean matching was used for continuous variables and logistic regression for binary variables. The imputation model included the outcome and all covariates. Imaging exposures (ΔSMI% and SMI slope) were not imputed and were used only within observable intervals.
Three robustness analyses were prespecified: (I) replacing ΔSMI% with the SMI slope to repeat the primary model; (II) repeating the early-window analysis in a subsample restricting the first imaging window to 8–12 weeks; (III) using a joint model (linear mixed-effects + Cox shared random effects) to address irregular measurement times and informative dropout.
Subgroup and interaction analyses were conducted within treatment category (chemotherapy, immunotherapy, targeted therapy, or combination regimens), sex, and BMI (<25 vs. ≥25 kg/m2), with interaction terms specified in the model to test consistency. All tests were two-sided, with a significance threshold of α=0.05. If the events-to-variables ratio was insufficient to support the full set of covariates, ridge-penalized Cox regression was used and the basis for selecting the penalty parameter was reported to reduce overfitting risk.
Data management
Segmentation workflows, parameters, and key quality-control points were compiled into a standard operating procedure and implemented after investigator training; a target threshold of between-rater ICC ≥0.90 was set for segmentation consistency. All data were stored on an institution-controlled server with role-based access management; imaging and clinical data were linked by a unique study identifier.
Results
Study population and imaging quality control
From 2018 to 2023, we identified 1,307 lung cancer patients who received systemic therapy; after applying clinical criteria including histologic type, stage, age, and concomitant other primary malignancies, 743 met the clinical inclusion criteria; a further 292 were excluded based on baseline and follow-up CT requirements and imaging quality control, and 451 patients were finally included in the analysis set (Figure 1). Among the 451 included advanced NSCLC patients, the median age was 63 years; male 63.64%, smoking history 61.86%, ECOG PS 0–1 72.73%. Baseline median BMI was 22.9 kg/m2 and SMI was 46.8 cm2/m2; median tumor burden was 89 mm, and liver metastasis was present in 30.82%. Median follow-up was 18.4 months, with 271 deaths (Table 1). Each patient had at least 2 CT time points that included L3 (median 3), the median interval from baseline to the first follow-up was 8.8 weeks, and the median interval between successive follow-up scans was 9.3 weeks; a total of 1,441 scans were included for analysis. Consistency of L3 skeletal muscle segmentation was very high, with both between-rater and within-rater ICC ≥0.97 (Table 2). The missing rates for age, BMI, and baseline tumor burden were 0.44%, 6.87%, and 8.20%, respectively [corresponding fraction of missing information (FMI) 0.009, 0.13, and 0.19]. After multiple imputation by chained equations (m=20), imputation diagnostics showed no obvious abnormalities, and the resulting data were used for subsequent survival analyses (Table 3).
Figure 1.
Flowchart of patient selection. CT, computed tomography; NSCLC, non-small cell lung cancer; SCLC, small cell lung cancer; SMI, skeletal muscle index.
Table 1. Baseline characteristics of patients.
| Variable | Overall (N=451) |
|---|---|
| Age (years) | 63 [56–71] |
| Sex | |
| Male | 287 (63.64) |
| Female | 164 (36.36) |
| Smoking history | |
| Never | 172 (38.14) |
| Former or current | 279 (61.86) |
| BMI (kg/m2) | 22.9 [20.8–25.7] |
| ECOG PS | |
| 0–1 | 328 (72.73) |
| ≥2 | 123 (27.27) |
| Histology | |
| Adenocarcinoma | 296 (65.63) |
| Squamous cell carcinoma | 155 (34.37) |
| Stage | |
| IIIB | 103 (22.84) |
| IV | 348 (77.16) |
| Line of therapy | |
| First-line | 379 (84.04) |
| Later-line | 72 (15.96) |
| Treatment category | |
| Chemotherapy | 142 (31.49) |
| Immunotherapy | 151 (33.48) |
| Targeted therapy | 96 (21.29) |
| Combination regimens | 62 (13.75) |
| Baseline tumor burden (sum of the longest diameters of RECIST 1.1 target lesions, mm) | 89 [63–136] |
| Liver metastasis | |
| Present | 139 (30.82) |
| Absent | 312 (69.18) |
| Baseline SMI (cm2/m2) | 46.8 [40.9–53.1] |
| Follow-up duration (months) | 18.4 [9.7–28.9] |
| Events (deaths) | 271 (60.09) |
Categorical variables are presented as n (%), and continuous variables are presented as median [interquartile range]. BMI, body mass index; ECOG PS, Eastern Cooperative Oncology Group performance status; RECIST, Response Evaluation Criteria In Solid Tumors; SMI, skeletal muscle index.
Table 2. Imaging acquisition and segmentation quality metrics.
| Metric | Value | 95% CI |
|---|---|---|
| Number of available CT time points per patient | ||
| 2 times | 138 (30.60) | – |
| 3 times | 166 (36.81) | – |
| 4 times | 94 (20.84) | – |
| ≥5 times | 53 (11.75) | – |
| Interval from baseline to first follow-up (weeks) | 8.8 [6.2–11.9] | – |
| All intervals between successive follow-ups (weeks) | 9.3 [6.7–13.4] | – |
| Total number of usable scans (n) | 1,441 | – |
| Between-rater ICC | ||
| SMA | 0.98 | 0.97–0.99 |
| SMI | 0.97 | 0.96–0.98 |
| Within-rater ICC | ||
| SMA | 0.99 | 0.98–0.99 |
| SMI | 0.99 | 0.98–0.99 |
Categorical variables are presented as n (%), and continuous variables are presented as median [interquartile range]. SMA denotes the L3-level skeletal muscle cross-sectional area; time points refer to available CT examinations that included L3. CI, confidence interval; CT, computed tomography; ICC, intraclass correlation coefficient; SMA, skeletal muscle area; SMI, skeletal muscle index.
Table 3. Overview of missing data and multiple imputation diagnostics.
| Variable | Missing number | Missing rate (%) | Imputation model type (MICE) | FMI |
|---|---|---|---|---|
| Age | 2 | 0.44 | PMM | 0.009 |
| BMI | 31 | 6.87 | PMM | 0.13 |
| Baseline tumor burden | 37 | 8.2 | PMM | 0.19 |
Apart from the variables listed in the table, no other baseline covariates had missing data; imaging exposures ΔSMI% and SMI slope were not imputed and were used only within observable intervals. BMI, body mass index; FMI, the fraction of missing information under Rubin’s rules; MICE, multiple imputation by chained equations; PMM, predictive mean matching; SMI, skeletal muscle index.
Independent effect of dynamic SMI on OS
After adjustment for covariates, the time-dependent Cox model showed that each 5% decrease in ΔSMI% was independently associated with an increased risk of death (HR =1.16, 95% CI: 1.09–1.24, P=0.001); higher baseline SMI was associated with a lower risk of death (per 10 cm2/m2: HR =0.87, 95% CI: 0.79–0.96, P=0.006). ECOG PS ≥2, stage IV, later-line therapy, higher tumor burden, and liver metastasis were all independent adverse prognostic factors (all P<0.05) (Table 4). Further testing with a four-knot restricted cubic spline showed that ΔSMI% was approximately linearly associated with mortality risk (overall P=0.001, nonlinearity P=0.07) (Figure 2).
Table 4. Multivariable time-dependent Cox model.
| Variable | HR | 95% CI | P value |
|---|---|---|---|
| Time-dependent ΔSMI% (per 5% decrease) | 1.16 | 1.09–1.24 | 0.001 |
| Baseline SMI (per 10 cm2/m2) | 0.87 | 0.79–0.96 | 0.006 |
| Age (per 10 years) | 1.07 | 0.98–1.16 | 0.11 |
| Sex (female vs. male) | 0.92 | 0.78–1.09 | 0.32 |
| Smoking history (former or current vs. never) | 1.11 | 0.93–1.33 | 0.24 |
| BMI (per 5 kg/m2) | 0.93 | 0.84–1.03 | 0.15 |
| ECOG PS (≥2 vs. 0–1) | 1.46 | 1.18–1.79 | 0.001 |
| Histology (squamous cell carcinoma vs. adenocarcinoma) | 1.17 | 0.97–1.41 | 0.10 |
| Stage (IV vs. IIIB) | 1.28 | 1.02–1.62 | 0.03 |
| Line of therapy (later-line vs. first-line) | 1.31 | 1.06–1.64 | 0.01 |
| Treatment category | |||
| Immunotherapy vs. chemotherapy | 0.88 | 0.72–1.07 | 0.19 |
| Targeted therapy vs. chemotherapy | 0.82 | 0.66–1.02 | 0.07 |
| Combination regimens vs. chemotherapy | 0.96 | 0.75–1.23 | 0.74 |
| Baseline tumor burden (per 50 mm) | 1.12 | 1.03–1.22 | 0.007 |
| Liver metastasis (present vs. absent) | 1.29 | 1.09–1.53 | 0.006 |
ΔSMI% represents relative change from baseline; negative values indicate a decrease. BMI, body mass index; ECOG PS, Eastern Cooperative Oncology Group Performance Status; SMI, skeletal muscle index.
Figure 2.
Restricted cubic spline curve of ΔSMI% and mortality risk. CI, confidence interval; HR, hazard ratio; SMI, skeletal muscle index.
Early treatment window analysis
Kaplan-Meier analysis (follow-up after the 12-week landmark analysis) showed that the survival curve of the early ΔSMI% ≤−5% group was clearly lower than that of the early ΔSMI% >−5% group (log-rank P=0.003) (Figure 3). Using a 12-week landmark Cox model, after adjusting for covariates, early ΔSMI% ≤−5% was independently associated with a higher risk of death (HR =1.68, 95% CI: 1.26–2.23, P=0.001). In the same model, ECOG PS ≥2, stage IV, later-line therapy, higher baseline tumor burden, and liver metastasis were also adverse prognostic factors, whereas higher baseline SMI was associated with a lower risk of death (all P<0.05) (Table 5).
Figure 3.
12-week landmark Kaplan-Meier survival curves stratified by early ΔSMI%. Analysis included patients alive at 12 weeks with completed imaging assessment; censoring was based on the last follow-up date. SMI, skeletal muscle index.
Table 5. Twelve-week landmark Cox analysis of early ΔSMI% and overall survival.
| Variable | HR | 95% CI | P value |
|---|---|---|---|
| Early ΔSMI% ≤−5% (vs. >−5%) | 1.68 | 1.26–2.23 | 0.001 |
| Age (per 10 years) | 1.08 | 0.97–1.20 | 0.16 |
| Sex (female vs. male) | 0.91 | 0.76–1.10 | 0.32 |
| Smoking history (former or current vs. never) | 1.12 | 0.92–1.35 | 0.25 |
| BMI (per 5 kg/m2) | 0.93 | 0.83–1.05 | 0.21 |
| ECOG PS (≥2 vs. 0–1) | 1.44 | 1.14–1.81 | 0.002 |
| Histology (squamous cell carcinoma vs. adenocarcinoma) | 1.15 | 0.94–1.41 | 0.17 |
| Stage (IV vs. IIIB) | 1.26 | 1.003–1.60 | 0.047 |
| Line of therapy (later-line vs. first-line) | 1.29 | 1.05–1.60 | 0.01 |
| Treatment category | |||
| Immunotherapy vs. chemotherapy | 0.86 | 0.70–1.06 | 0.14 |
| Targeted therapy vs. chemotherapy | 0.81 | 0.64–1.02 | 0.07 |
| Combination regimens vs. chemotherapy | 0.95 | 0.73–1.24 | 0.69 |
| Baseline tumor burden (per 50 mm) | 1.11 | 1.01–1.22 | 0.02 |
| Liver metastasis (present vs. absent) | 1.25 | 1.04–1.51 | 0.01 |
| Baseline SMI (per 10 cm2/m2) | 0.88 | 0.79–0.99 | 0.03 |
At the 12-week landmark, the at-risk population was n=339 (early ΔSMI% ≤−5% n=119; >−5% n=220); there were 227 events after the landmark (≤−5% group 93; >−5% group 134). BMI, body mass index; CI, confidence interval; ECOG PS, Eastern Cooperative Oncology Group Performance Status; HR, hazard ratio; SMI, skeletal muscle index.
Complementarity with RECIST and incremental value of models
Pearson correlation analysis in 339 patients with available early ΔSMI% and first RECIST assessment data showed a weak negative correlation between early ΔSMI% and the change in the longest diameter of target lesions at the first RECIST assessment (r=−0.18, P=0.001), indicating low correlation between the magnitude of muscle loss and tumor volume change (Figure 4A). Model comparisons showed that, on the basis of clinical covariates, adding baseline SMI slightly increased the C-index (0.67→0.69), and further inclusion of time-dependent ΔSMI% continued to improve predictive performance (C-index increased to 0.71, NRI =0.19). After adding RECIST information, the C-index increased slightly to 0.73 (Table 6). Using the Uno method to estimate the time-dependent AUC(t) and applying bias correction with 1,000 bootstrap resamples, Model C achieved AUC(t) values of 0.71 and 0.70 at the key time points of 6 and 12 months, respectively, and was overall higher than Models A and B during the 0–24-month follow-up period, suggesting that inclusion of time-dependent ΔSMI% improves discrimination (Figure 4B). Decision curve analysis showed that Model C yielded higher net benefit than Models A/B at 6 and 12 months, and outperformed the Treat all/none strategy within threshold ranges of 0.23–0.47 (6 months) and 0.34–0.58 (12 months); after adding RECIST, the net benefit of Model D increased slightly further (Figure 4C). Calibration analysis using 1,000 bootstrap samples showed that, after the 12-week landmark, predictions of 6- and 12-month mortality risk by Model C were generally consistent with observed risks, with calibration slope close to 1 and intercept close to 0 (6 months: 0.982/0.011; 12 months: 0.957/0.018), indicating good calibration (Figure 4D).
Figure 4.
Complementary prognostic performance of dynamic ΔSMI% relative to clinical covariates and RECIST for predicting overall survival after the 12-week landmark analysis. (A) Correlation between early ΔSMI% and tumor response. Assessed by RECIST 1.1; negative values indicate tumor shrinkage. (B) Time-dependent AUC of prediction models. Model A: clinical covariates; Model B: Model A + baseline SMI; Model C: Model B + time-dependent ΔSMI%. AUC(t) estimated via Uno’s method with 1,000 bootstrap resamples. (C) Decision curve analysis. Model A: clinical covariates; Model B: Model A + baseline SMI; Model C: Model B + time-dependent ΔSMI%; Model D: Model C + RECIST. (D) Calibration curves. AUC, area under the curve; KM, Kaplan-Meier; RECIST, Response Evaluation Criteria In Solid Tumors; SMI, skeletal muscle index.
Table 6. Comparison of model performance and incremental value.
| Metric | Model A | Model B | Model C | Model D |
|---|---|---|---|---|
| C-index (corrected) | 0.67 | 0.69 | 0.71 | 0.73 |
| Likelihood ratio χ2 | – | 11.72 | 16.83 | 6.27 |
| P value | – | 0.001 | 0.001 | 0.01 |
| AIC | 3,081.42 | 3,074.33 | 3,059.87 | 3,051.74 |
| BIC | 3,148.63 | 3,138.27 | 3,126.94 | 3,119.38 |
| 12-month NRI (95% CI) | – | 0.12 (0.04–0.21) | 0.19 (0.09–0.30) | 0.07 (0.003–0.14) |
| 12-month IDI (95% CI) | – | 0.02 (0.002–0.03) | 0.03 (0.01–0.04) | 0.01 (0.001–0.02) |
| 12-month Brier score | 0.19 | 0.18 | 0.17 | 0.16 |
Likelihood ratio χ2 and P values are for nested comparisons with the immediately preceding model (B vs. A, C vs. B, D vs. C). NRI and IDI are category-free at 12 months, calculated relative to the immediately preceding model, with 95% CIs reported. Model A includes prespecified clinical baseline covariates; Model B adds baseline SMI; Model C adds time-dependent ΔSMI%; Model D adds the first RECIST response or change in target lesions. AIC, Akaike information criterion; BIC, Bayesian information criterion; CI, confidence interval; IDI, integrated discrimination improvement; NRI, net reclassification improvement; RECIST, Response Evaluation Criteria In Solid Tumors; SMI, skeletal muscle index.
Robustness, sensitivity, and model diagnostics
Using the same modeling framework as the primary analysis, when replacing ΔSMI% with the SMI slope, each −0.10 cm2/m2/week decrease in SMI was associated with an increased risk of death (HR =1.19); when narrowing the early window to 8–12 weeks, early ΔSMI% ≤−5% still indicated worse prognosis (HR =1.62); the joint model also showed that higher current SMI was associated with a lower risk of death (per 10 cm2/m2: HR =0.90). All of the above results reached statistical significance (P<0.05) (Table 7). Multivariable Cox interaction analyses showed that each 5% decrease in ΔSMI% was associated with higher mortality risk across the main subgroups; the 95% CIs crossed 1 in the targeted therapy and BMI ≥25 subgroups. Interactions were not significant (all P>0.05), and the overall effect direction was consistent (Figure 5). Multicollinearity diagnostics showed that VIFs for all covariates in the final model were low (all VIF <5), suggesting no significant multicollinearity in the model (Table 8).
Table 7. Summary of robustness and sensitivity analyses.
| Analysis scenario | Primary effect measure | HR (95% CI) | P value | Sample size, n | Events, n |
|---|---|---|---|---|---|
| Time-dependent Cox replacing ΔSMI% with the SMI slope | Per −0.10 cm2/m2/week | 1.19 (1.08–1.31) | 0.001 | 451 | 271 |
| Landmark Cox restricting the early window to 8–12 weeks | Early ΔSMI% ≤−5% vs. >−5% | 1.62 (1.19–2.20) | 0.002 | 311 | 201 |
| Joint model (linear mixed-effects + Cox shared random effects) | Current SMI value (per 10 cm2/m2) | 0.90 (0.82–0.99) | 0.03 | 451 | 271 |
The unit of the SMI slope was cm2/m2 per week; the shared random effects of the joint model were used to handle irregular measurement times and informative dropout. Modeling in each scenario was consistent with the primary analysis. CI, confidence interval; HR, hazard ratio; SMI, skeletal muscle index.
Figure 5.
Subgroup analysis of ΔSMI% effect on overall survival. Note: HR represents the risk ratio per 5% decrease in ΔSMI%; adjusted for covariates consistent with the main multivariate Cox model. BMI, body mass index; HR, hazard ratio; SMI, skeletal muscle index.
Table 8. Multicollinearity diagnostics.
| Variable | VIF |
|---|---|
| Time-dependent ΔSMI% | 1.74 |
| Baseline SMI | 1.96 |
| Age | 1.12 |
| Sex | 1.08 |
| Smoking history | 1.14 |
| BMI | 1.23 |
| ECOG PS | 1.19 |
| Histology | 1.09 |
| Stage | 1.17 |
| Line of therapy | 1.29 |
| Treatment category | |
| Immunotherapy vs. chemotherapy | 1.47 |
| Targeted therapy vs. chemotherapy | 1.52 |
| Combination regimens vs. chemotherapy | 1.33 |
| Baseline tumor burden | 1.26 |
| Liver metastasis | 1.18 |
VIF <5 was considered indicative of no significant multicollinearity. BMI, body mass index; ECOG PS, Eastern Cooperative Oncology Group Performance Status; SMI, skeletal muscle index; VIF, variance inflation factor.
Discussion
Given reliable segmentation consistency and a follow-up cadence consistent with real-world clinical practice, longitudinal declines in SMI quantified on serial CT remained independently associated with an increased risk of death and showed an approximately linear dose–response relationship, suggesting that reliance on a single baseline muscle measurement cannot reflect the accumulation of risk over the disease course. Skeletal muscle is the body’s largest protein reservoir and a primary target organ of inflammation and energy-metabolic stress (16). In advanced NSCLC, tumor burden and treatment stress drive systemic inflammation and neuroendocrine responses, activating protein degradation pathways and exacerbating anabolic resistance. This restricts muscle protein synthesis and enhances breakdown, leading to muscle mass loss (17). Muscle loss reduces reserves, weakens immunometabolic support, increases risks, and affects treatment, resulting in a survival disadvantage (18). The protective effect of baseline SMI shows initial reserves matter, and dynamic changes reflect depletion rate, enabling risk trajectory differentiation at the same baseline. Previous lung cancer body composition studies mainly reported prognostic differences around baseline sarcopenia thresholds, affected by threshold selection and single -measurement variability (19). In recent years, more studies emphasize that muscle mass loss during treatment is related to survival, toxicity, and functional outcomes (20). The approximate linear relationship and metric consistency make dynamic SMI more suitable for viewing cachexia as a continuous spectrum and provide evidence for incorporating muscle mass monitoring into routine imaging follow-up.
In addition, we incorporated recent published evidence regarding CT-assessed sarcopenia in metastatic and elderly NSCLC populations reported in Current Oncology (21). This cross-sectional work confirmed baseline sarcopenia as an unfavorable prognostic factor for advanced lung cancer, yet it lacked time-dependent longitudinal modeling of on-treatment muscle changes, did not define a standardized 12-week early predictive window, and failed to systematically compare body composition metrics with RECIST tumor response or quantify incremental predictive gains over conventional clinical models. Our study substantially extends this prior evidence by addressing these critical methodological gaps and provides an actionable risk stratification tool for routine clinical follow-up.
After evaluating early SMI changes with a 12-week landmark analysis design, patients with early ΔSMI% ≤−5% had persistently worse survival after the landmark. This association remained independent after adjusting for key covariates like baseline tumor burden, ECOG PS, stage, and line of therapy. In the sensitivity analysis with an imaging window of 8–12 weeks, the conclusions were consistent, suggesting the signal is not time-dependent and has reproducible risk-stratification capability. A rapid decline in muscle mass at treatment initiation often means the body enters a strong catabolic state quickly. Inflammation-driven protein degradation, insufficient intake, reduced activity, and treatment-related adverse events lead to rapid consumption of skeletal muscle as a metabolic buffer (22). This process is often recognized earlier than changes in body weight and may directly diminish locomotor and respiratory muscle reserves, reduce immunometabolic support and drug tolerance, and increase the likelihood of dose adjustments, delays, or interruptions, thereby affecting subsequent disease control and survival (23). The 12-week window aligns closely with the routine imaging follow-up cadence in clinical practice, making CT-based body composition assessment feasible to implement within real-world care pathways. Early SMI decline may serve as a risk signal that prompts consideration of additional nutritional, functional, or supportive-care assessment, rather than directly triggering specific interventions. Prospective interventional studies are still required to determine whether SMI-guided nutritional or exercise interventions improve treatment tolerance or survival (24). Previous cachexia and supportive care research in lung cancer has generally regarded early declines in body weight or muscle mass as an adverse signal, but body weight is substantially influenced by fluid retention, changes in adiposity, and measurement error, and a single sarcopenia threshold also fails to capture dynamic deterioration (25). Defining early SMI decline using a landmark approach helps reduce immortal time bias, brings early CT metrics closer to a translatable clinical early-warning tool, and provides a basis for prospective validation of the benefits of interventions. It should be emphasized that the continuous time-dependent ΔSMI% analysis represents the primary result of this study, while the dichotomous −5% threshold is presented as a potentially practical bedside risk-stratification tool that warrants further prospective confirmation in independent external cohorts.
Early ΔSMI% was only weakly correlated with RECIST target lesion change, suggesting that changes in muscle mass and tumor shrinkage reflect different dimensions. After incorporating baseline SMI into the model, predictive performance improved. Adding time-dependent ΔSMI% further enhanced discrimination and reclassification, and maintained higher time-dependent AUC(t) during follow-up. Decision curve analysis showed that models with dynamic SMI achieved higher net benefit across multiple threshold ranges, and adding RECIST information led to only a slight additional increase. After the 12-week landmark, calibration for short- and mid- term risk prediction was stable, and model outputs were consistent with observed risk levels. Multicollinearity diagnostics and subgroup interaction analyses indicated that predictors were independent with consistent effect directions. Clinically, RECIST mainly reflects tumor burden changes, while dynamic SMI more reflects host metabolic and functional reserve changes (26). Even when tumors shrink in advanced NSCLC patients, systemic inflammation and anabolic resistance may not be alleviated, and continued muscle mass loss can affect treatment tolerance and infection risk. So, dynamic SMI can provide prognostic information independent of tumor response (27). The net benefit of dynamic SMI indicates it can reduce ineffective interventions or missed identification of high-risk patients, better meeting clinical decision-making needs (28). The incremental predictive gain from RECIST was limited, in line with their partially complementary nature. Prognostic research in lung cancer has shifted from single tumor indicators to combined tumor and host indicators, and CT-based body composition has translational potential due to its repeatability and ease of integration into follow-up workflows (29). Existing studies likewise suggest limited correlation between changes in muscle mass and RECIST, supporting a framework in which tumor control and host wasting are assessed in parallel (30), highlighting the clinical value of dynamic SMI.
Beyond skeletal muscle, the broader concept that routine oncologic CT contains prognostically relevant host-related information beyond tumor morphology is increasingly recognized. In particular, cardiac biomarkers and CT-derived coronary artery calcium have emerged as extra-tumoral imaging metrics with prognostic significance in NSCLC patients. Nardone et al. demonstrated that CT-assessed coronary artery calcium score [coronary artery disease (CAD) score] and cardiac ejection fraction were independent predictors of OS in NSCLC patients, supporting the integration of cardiovascular imaging biomarkers into risk stratification (31). Furthermore, the CARE-RT trial protocol highlights the value of integrated cardio-imaging evaluation—including cardiac CT and magnetic resonance imaging (MRI)—in patients undergoing radical radiotherapy for lung cancer, aiming to detect and quantify treatment-related cardiac damage and guide individualized management (32). Together with our findings on dynamic SMI, these studies support a paradigm shift from tumor-only response assessment toward a combined tumor-and-host imaging framework, in which routine CT can simultaneously capture both tumor burden changes and host reserve status, including musculoskeletal and cardiovascular dimensions, without requiring additional examinations.
Limitations
As a single-center retrospective study, case sources depended on prior visits and image retention, and the included population tended to be those who completed baseline and early follow-up CT, which may underestimate the risk of patients with rapid progression; selection bias cannot be fully avoided. CT re-examinations were driven by clinical decisions and the intervals were not investigator-defined; some patients had fewer scans due to disease worsening, transfer of care, or treatment interruption; even with the time-dependent Cox model, the 8–12-week sensitivity analysis, and the joint model, residual informative dropout may persist. Covariates covered only routine clinical information; inflammatory–nutritional markers, glucocorticoid use, and the intensity of nutritional support and exercise interventions were not systematically collected, and unmeasured confounding may affect interpretation of effect sizes. Imaging metrics focused on L3 cross-sectional SMI; excluding muscle density, fat infiltration, and whole-body muscle distribution evaluation, which limits a complete body composition change characterization. Model performance was optimism-corrected by bootstrap, calibrated, and analyzed with decision curves, but it’s internal validation lacking external data for generalizability. Future studies should conduct multicenter prospective registries, harmonize CT acquisition windows and segmentation workflows, record key inflammatory-nutritional markers and dose intensity without extra exams, validate the stability of the 12-week threshold, and embed nutrition and exercise interventions triggered by early muscle mass decline to evaluate their impact on treatment tolerance and survival. Looking ahead, longitudinal SMI measurements may eventually be integrated with tumor imaging, CT radiomics, molecular characteristics, and circulating biomarkers to build more comprehensive prognostic models; recent work combining CT radiomics with liquid biopsy in NSCLC illustrates the potential of such multimodal patient characterization (33). In parallel, artificial intelligence–assisted longitudinal quantitative CT evaluation is demonstrating growing feasibility for automated extraction and consistent longitudinal monitoring of quantitative imaging biomarkers (34,35); although these initial investigations focused on pulmonary parenchymal abnormalities, the same automated approaches may be extended to skeletal muscle quantification in future work. These multimodal and automated developments represent promising future directions rather than currently established clinical tools.
Conclusions
Longitudinal declines in SMI on serial CT are an independent adverse prognostic factor for OS in advanced NSCLC, with an approximately linear dose–response. Dynamic metrics are more predictive than a single baseline measurement and complement baseline info. Early decline within 12 weeks can identify high-risk patients at routine follow-up nodes and is actionable. Dynamic SMI is weakly correlated with RECIST, provides host-status info beyond tumor burden, and improves discrimination, reclassification, and decision net benefit in clinical models while maintaining good calibration. Dynamic SMI monitoring has potential utility as a routine risk-stratification tool during follow-up, and early decline may identify patients who warrant additional nutritional, functional, or supportive-care assessment. Future prospective interventional studies are needed to determine whether SMI-guided interventions can optimize treatment tolerance and survival outcomes.
Supplementary
The article’s supplementary files as
Acknowledgments
None.
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. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Henan Provincial Hospital of Traditional Chinese Medicine (The Second Affiliated Hospital of Henan University of Chinese Medicine) (No. HNSZYYYJS202302050), and individual consent for this retrospective analysis was waived.
Footnotes
Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1968/rc
Funding: None.
Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1968/coif). The authors have no conflicts of interest to declare.
Data Sharing Statement
Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1968/dss
References
- 1. Sung H, Ferlay J, Siegel RL, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 2021;71:209-49.. 10.3322/caac.21660 [DOI] [PubMed] [Google Scholar]
- 2. Xie W, Liu S, Li G, et al. The evolving treatment paradigm of lung cancer in China. Acta Pharm Sin B 2022;12:1536-7.. 10.1016/j.apsb.2022.01.010 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Wang Y, Lin X, Sun D. A narrative review of prognosis prediction models for non-small cell lung cancer: what kind of predictors should be selected and how to improve models?. Ann Transl Med 2021;9:1597.. 10.21037/atm-21-4733 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Arends J, Strasser F, Gonella S, et al. Cancer cachexia in adult patients: ESMO Clinical Practice Guidelines☆. ESMO Open 2021;6:100092. Erratum in: ESMO Open. 2022 Dec;7(6):100614.. 10.1016/j.esmoop.2021.100092 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Wang M, Soth V, Liu X, et al. Impact of Response Assessment Intervals on Survival and Economic Burden in Long-Term Responders to Immunotherapy for Advanced Non-Small-Cell Lung Cancer. Cancers (Basel) 2025;17:3312.. 10.3390/cancers17203312 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Klassen PN, Mazurak VC, Thorlakson J, et al. Call for standardization in assessment and reporting of muscle and adipose change using computed tomography analysis in oncology: A scoping review. J Cachexia Sarcopenia Muscle 2023;14:1918-31.. 10.1002/jcsm.13318 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Lin TY, Chen YF, Wu WT, et al. Impact of sarcopenia on the prognosis and treatment of lung cancer: an umbrella review. Discov Oncol 2022;13:115.. 10.1007/s12672-022-00576-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Lee J, Kim EY, Kim E, et al. Longitudinal changes in skeletal muscle mass in patients with advanced squamous cell lung cancer. Thorac Cancer 2021;12:1662-7.. 10.1111/1759-7714.13958 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Weymann D, Krebs E, Regier DA. Addressing immortal time bias in precision medicine: Practical guidance and methods development. Health Serv Res 2025;60:e14376.. 10.1111/1475-6773.14376 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Rollins KE, Gopinath A, Awwad A, et al. Computed tomography-based psoas skeletal muscle area and radiodensity are poor sentinels for whole L3 skeletal muscle values. Clin Nutr 2020;39:2227-32.. 10.1016/j.clnu.2019.10.003 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Derstine BA, Holcombe SA, Ross BE, et al. Optimal body size adjustment of L3 CT skeletal muscle area for sarcopenia assessment. Sci Rep 2021;11:279.. 10.1038/s41598-020-79471-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Matsumoto H, Tsuchiya K, Nakanishi H, et al. Clinical Usefulness of Monitoring Muscle Volume during Atezolizumab Plus Bevacizumab Therapy in Patients with Unresectable Hepatocellular Carcinoma. Cancers (Basel) 2022;14:3551.. 10.3390/cancers14143551 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Hanna L, Nguo K, Porter J, et al. Malnutrition risk as a predictor of quality of life and skeletal muscle depletion following upper gastrointestinal cancer diagnosis: A longitudinal analysis. J Nutr Health Aging 2025;29:100623.. 10.1016/j.jnha.2025.100623 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Best TD, Roeland EJ, Horick NK, et al. Muscle Loss Is Associated with Overall Survival in Patients with Metastatic Colorectal Cancer Independent of Tumor Mutational Status and Weight Loss. Oncologist 2021;26:e963-70.. 10.1002/onco.13774 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Huang CY, Yang YC, Chen TC, et al. Muscle loss during primary debulking surgery and chemotherapy predicts poor survival in advanced-stage ovarian cancer. J Cachexia Sarcopenia Muscle 2020;11:534-46.. 10.1002/jcsm.12524 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Webster JM, Kempen LJAP, Hardy RS, et al. Inflammation and Skeletal Muscle Wasting During Cachexia. Front Physiol 2020;11:597675.. 10.3389/fphys.2020.597675 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Setiawan T, Sari IN, Wijaya YT, et al. Cancer cachexia: molecular mechanisms and treatment strategies. J Hematol Oncol 2023;16:54.. 10.1186/s13045-023-01454-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Davis MP, Panikkar R. Sarcopenia associated with chemotherapy and targeted agents for cancer therapy. Ann Palliat Med 2019;8:86-101.. 10.21037/apm.2018.08.02 [DOI] [PubMed] [Google Scholar]
- 19. Leão Mendes J, Ferreira RQ, Mata I, et al. Body Composition Analysis in Metastatic Non-Small-Cell Lung Cancer: Depicting Sarcopenia in Portuguese Tertiary Care. Cancers (Basel) 2025;17:539.. 10.3390/cancers17030539 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Jensen S, Bloch Z, Quist M, et al. Sarcopenia and loss of muscle mass in patients with lung cancer undergoing chemotherapy treatment: a systematic review and meta-analysis. Acta Oncol 2023;62:318-28.. 10.1080/0284186X.2023.2180660 [DOI] [PubMed] [Google Scholar]
- 21. Nardone V, Reginelli A, Patanè V, et al. Prognostic Value of Sarcopenia in Elderly Patients with Metastatic Non-Small-Cell Lung Cancer Undergoing Radiotherapy. Curr Oncol 2024;31:6673-85.. 10.3390/curroncol31110492 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Fairman CM, Lønbro S, Cardaci TD, et al. Muscle wasting in cancer: opportunities and challenges for exercise in clinical cancer trials. JCSM Rapid Commun 2022;5:52-67.. 10.1002/rco2.56 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Murphy BT, Mackrill JJ, O'Halloran KD. Impact of cancer cachexia on respiratory muscle function and the therapeutic potential of exercise. J Physiol 2022;600:4979-5004.. 10.1113/JP283569 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Degens JHRJ, Sanders KJC, de Jong EEC, et al. The prognostic value of early onset, CT derived loss of muscle and adipose tissue during chemotherapy in metastatic non-small cell lung cancer. Lung Cancer 2019;133:130-5.. 10.1016/j.lungcan.2019.05.021 [DOI] [PubMed] [Google Scholar]
- 25. Ni J, Zhang L. Cancer Cachexia: Definition, Staging, and Emerging Treatments. Cancer Manag Res 2020;12:5597-605.. 10.2147/CMAR.S261585 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Trestini I, Belluomini L, Dodi A, et al. Body composition derangements in lung cancer patients treated with first-line pembrolizumab: A multicentre observational study. J Cachexia Sarcopenia Muscle 2024;15:2349-60.. 10.1002/jcsm.13568 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Roch B, Coffy A, Jean-Baptiste S, et al. Cachexia - sarcopenia as a determinant of disease control rate and survival in non-small lung cancer patients receiving immune-checkpoint inhibitors. Lung Cancer 2020;143:19-26.. 10.1016/j.lungcan.2020.03.003 [DOI] [PubMed] [Google Scholar]
- 28. Vickers AJ, van Calster B, Steyerberg EW. A simple, step-by-step guide to interpreting decision curve analysis. Diagn Progn Res 2019;3:18.. 10.1186/s41512-019-0064-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Verkoulen KCHA, Geenen L, Franssen AJPM, et al. Beyond the tumor: towards a cachexia-based host phenotype through body composition analysis in patients with resectable lung cancer. Transl Lung Cancer Res 2025;14:3183-95.. 10.21037/tlcr-2025-511 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Rounis K, Makrakis D, Tsigkas AP, et al. Cancer cachexia syndrome and clinical outcome in patients with metastatic non-small cell lung cancer treated with PD-1/PD-L1 inhibitors: results from a prospective, observational study. Transl Lung Cancer Res 2021;10:3538-49. Erratum in: Transl Lung Cancer Res. 2023 Oct 31;12(10):2146-2147.. 10.21037/tlcr-21-460 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Nardone V, Reginelli A, De Marco G, et al. Role of Cardiac Biomarkers in Non-Small Cell Lung Cancer Patients. Diagnostics (Basel) 2023;13:400.. 10.3390/diagnostics13030400 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Nardone V, Belfiore MP, De Chiara M, et al. CARdioimaging in Lung Cancer PatiEnts Undergoing Radical RadioTherapy: CARE-RT Trial. Diagnostics (Basel) 2023;13:1717.. 10.3390/diagnostics13101717 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Belfiore MP, Sansone M, Ciani G, et al. Radiomic Analysis and Liquid Biopsy in Preoperative CT of NSCLC: An Explorative Experience. Thorac Cancer 2025;16:e70115.. 10.1111/1759-7714.70115 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Russo A, Patanè V, Oliva A, et al. AI-Based HRCT Quantification in Connective Tissue Disease-Associated Interstitial Lung Disease. Diagnostics (Basel) 2025;15:2179.. 10.3390/diagnostics15172179 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Russo A, Patanè V, Ruotolo F, et al. Artificial Intelligence-Assisted Quantification of Longitudinal HRCT Changes During Treatment of Pulmonary Tuberculosis: An Exploratory Proof-of-Concept Study. Diagnostics (Basel) 2026;16:1822.. 10.3390/diagnostics16121822 [DOI] [PMC free article] [PubMed] [Google Scholar]