Skeletal metastatic burden as a strong prognostic factor of survival in advanced NSCLC treated with chemo-immunotherapy: An exploratory imaging-based analysis

Abstract

Background

Metastatic burden is a key determinant of prognosis in advanced solid tumors, yet current RECIST 1.1-based assessment does not fully capture total or site-specific disease extent. Emerging evidence suggests that quantitative imaging-derived metrics may improve prognostic stratification beyond binary metastatic classification.

Methods

We retrospectively analyzed 52 patients with advanced non-oncogene-addicted lung cancer. Baseline clinical variables and imaging-derived metastatic lesions were assessed across multiple organ sites (lung, lymph nodes, bone, brain, liver and adrenal glands). Lesion area was calculated using an elliptical approximation and summed per organ and globally. Overall survival was evaluated using multivariable Cox regression models with restricted cubic splines to account for non-linear effects. Analyses included: binary metastatic status, continuous site-specific tumor burden, and relative contribution of each site to total metastatic load.

Results

Median age was 67 years, and bone (53%) and brain (27%) were frequent metastatic sites. Across all analytical levels, bone metastases consistently showed the strongest prognostic impact. Bone involvement was associated with worse overall survival both as a binary variable (HR 2.93, p = 0.020) and as continuous tumor burden (HR 2.82, p = 0.014). A higher relative contribution of bone disease to total tumor burden was also significantly associated with poorer survival (HR 3.01, p = 0.016). Other metastatic sites did not demonstrate consistent statistically significant associations with survival in this exploratory cohort, although these findings should not be interpreted as evidence of absence of prognostic relevance. Histological subtype and age remained independent prognostic factors across models.

Conclusion

Quantitative assessment of metastatic burden identified skeletal disease as a consistent prognostic signal in this exploratory cohort, supporting further investigation of imaging-derived measures of metastatic burden and spatial disease distribution.

Highlights

• •Quantitative imaging assessment of metastatic burden provides prognostic stratification beyond RECIST 1.1 in advanced NSCLC.
• •Skeletal metastases represent the most powerful site-specific prognostic determinant across binary, continuous, and relative burden models.
• •Bone metastatic burden is independently associated with reduced overall survival, demonstrating a dose-dependent effect.
• •Other metastatic sites show limited or inconsistent prognostic impact when modeled as continuous imaging-derived variables.
• •Spatial distribution of metastatic disease, particularly bone-dominant phenotypes, carries relevant prognostic information.

1. Introduction

The prognostic assessment of metastatic burden in patients with advanced non-small cell lung cancer (NSCLC) remains a critical challenge. In clinical practice, response assessment and disease characterization are still largely based on RECIST 1.1 criteria, which primarily rely on changes in linear dimensions of target lesions and dichotomous classification of metastatic involvement [1]. However, this framework has well-known limitations, as it relies on simple linear measurements and binary response categories that do not fully capture the complexity of tumor evolution.

The clinical landscape is significantly complicated by the assessment of metastatic sites that carry major prognostic weight, such as the brain, bone, and lung. Pulmonary metastatic burden, in particular, exhibits high intra-organ heterogeneity, where patterns such as diffuse bilateral (miliary) involvement, endobronchial extension, or cavitation serve as distinct markers of aggressive biology and systemic treatment resistance [2], [3], [4]. Similarly, bone and brain metastases are frequently underestimated by RECIST, despite their profound impact on morbidity and survival [5], [6].

This challenge is particularly important in the era of immuno-oncology (IO), where traditional RECIST metrics often fail to capture the full spectrum of patient response [7], [8]. Growing evidence suggests that total tumor burden (TTB) acts as a more robust, independent predictor of outcomes in IO-treated patients [9].

Based on these considerations, the aim of the present study was to evaluate the prognostic impact of metastatic burden quantified as lesion area across multiple organ sites in a cohort of patients with metastatic NSCLC. In particular, we investigated how site-specific tumor burden correlates with overall survival and whether quantitative imaging features provide incremental prognostic value beyond standard clinical variables. We also specifically addressed metastatic sites often underrepresented in classical RECIST evaluations, such as brain and bone, to provide a more comprehensive assessment of total disease burden and its prognostic implications [8], [10], [11].

2. Methods

We retrospectively analyzed a cohort of 52 treatment-naïve patients with advanced non-oncogene addicted lung cancer who received first-line chemotherapy plus immune checkpoint inhibitor (ICI) regimens at the Medical Oncology Department of the Policlinico Universitario Campus Bio-Medico (Rome, Italy) between February 2020 and April 2025. Patients were managed according to routine clinical practice and followed at the discretion of the treating physicians. Data were retrospectively collected, with a cutoff date of May 2025. Baseline clinical data included age, sex, ECOG performance status (PS), PD-L1 expression, smoking status, histology, and subsequent systemic therapy. Overall survival (OS) was defined as the time from metastatic diagnosis to death from any cause. Patients who were alive at the last available follow-up were censored at the date of last follow-up.

All available baseline contrast-enhanced CT scans of the chest, brain, abdomen, and neck acquired before treatment initiation were independently reviewed by two board-certified radiologists, who were blinded to survival outcomes. Discrepancies were resolved by consensus.

Metastatic lesions were identified across predefined anatomical sites, including the adrenal glands, lungs, lymph nodes, brain, bone, and liver. For each lesion, the maximum long-axis diameter and its perpendicular short-axis diameter were measured on axial CT images. Lesion area was estimated assuming an elliptical shape using the formula π × (long axis × short axis) / 4. Organ-specific metastatic burden was calculated as the sum of lesion areas within each anatomical site (mm²), whereas total metastatic burden was defined as the sum of lesion areas across all sites.A Associations between metastatic burden and OS were evaluated using Cox proportional hazards models. For each site, burden was modeled as a continuous variable, including patients with and without metastatic involvement of the specific anatomical site being analyzed. To account for potential non-linear effects of metastatic burden on survival, restricted cubic spline functions were applied to continuous lesion area variables within Cox models. For sites with a wide range of values, we used a log1p transformation to stabilize the data.

All models were adjusted for age, ECOG-PS, histological subtype, and PD-L1 expression, which were selected a priori based on their established clinical relevance as key prognostic covariates. This predefined selection was adopted to avoid overadjustment and multivariable model overfitting. The study strategy is summarized in Fig. 1. All statistical analyses were performed using R software (R Foundation for Statistical Computing).

Fig. 1.

Schematic representation of the retrospective multicenter analysis aimed at evaluating the prognostic impact of site-specific metastatic burden in advanced non-oncogene-addicted lung cancer.

3. Results

A total of 52 patients with metastatic non-oncogene-addicted lung cancer were included in the analysis. Baseline clinical and pathological characteristics are detailed in Table 1. The cohort had a median age of 67 years (IQR: 58–72), with a male predominance of 76%. Smoking history was highly prevalent, with 59% of patients being former smokers and 36% current smokers, while only 4.5% were never-smokers. Regarding performance status (ECOG PS), 93% of patients presented with a score of 0 or 1, and 6.8% with a score of 2.Adenocarcinoma was the predominant histological subtype (66%), followed by squamous cell carcinoma (14%) and other histologies (20%), including adenosquamous and poorly differentiated carcinomas.

Table 1.

Baseline clinical and pathological characteristics of the study cohort (n = 52). Abbreviations: ECOG PS, Eastern Cooperative Oncology Group performance status; IQR, interquartile range; PD-L1, programmed death-ligand 1.

Characteristic	N = 52
Age	67 (58, 72)
Sex	38 (76%)
Lung metastases	46 (90%)
Bone metastases	27 (53%)
Lymph nodes	43 (84%)
Adrenal metastases	13 (25%)
Brain metastases	14 (27%)
Liver metastases	11 (22%)
Smoking status
Never	2 (4.5%)
Ex	26 (59%)
Current	16 (36%)
ECOG performance status
0	25 (57%)
1	16 (36%)
2	3 (6.8%)
Histology
Adenocarcinoma	29 (66%)
Squamous	6 (14%)
Other	9 (20%)
PD-L1 status
0%	14 (33%)
1–49%;	28 (65%)
>50%	1 (2.3%)

¹Median (Q1, Q3); n (%)

PD-L1 expression was negative (0%) in approximately one-third of patients (33%), while the majority (65%) showed low-to-intermediate expression (1–49%), and only a small fraction (2.3%) presented high expression (≥50%).

Baseline metastatic distribution revealed a high prevalence of thoracic involvement, with lung and lymph node metastases observed in 90% and 84% of patients, respectively. Extrathoracic disease was also common, with bone metastases identified in 53% of cases, followed by brain (27%), adrenal (25%), and liver (22%) metastases. Overall, the study population exhibited a diverse metastatic profile, reflecting a clinically representative distribution of systemic disease across multiple organ systems.

In the first level of multivariable analysis, we evaluated the association between baseline clinical characteristics, metastatic site involvement (as binary variables), and overall survival. Age was independently associated with worse overall survival (HR 1.06, 95% CI 1.01–1.11, p = 0.027), indicating a progressive increase in risk with increasing age. Among metastatic sites, bone involvement was significantly associated with poorer survival (HR 2.93, 95% CI 1.19–7.24, p = 0.020). No statistically significant associations were observed for the other metastatic sites in this model. Histological squamous subtype was also independently associated with survival (HR 2.42, 95% CI 1.34–4.37, p = 0.003), confirming its prognostic relevance within this cohort. COG performance status and PD-L1 expression were not statistically significantly associated with overall survival in this model. (Table 2).

Table 2.

Multivariable Cox regression analysis of clinical characteristics and binary metastatic site involvement. Multivariable analysis assessing the prognostic impact of baseline clinical variables and the presence (binary) of metastases in specific organ sites on overall survival. Hazard ratios (HR) and 95% confidence intervals (CI) are reported.

Characteristic	HR OS (Univariate)	95% CI	p-value	HR OS (Multivariate)	95% CI	p-value
Age	1.04	1.00, 1.08	0.077	1.06	1.01, 1.11	0.027
Lung metastases	5.37	0.72, 40.23	0.102	1.69	0.19, 15.26	0.639
Bone metastases	1.84	0.93, 3.66	0.082	2.93	1.19, 7.24	0.020
Adrenal metastases	1.59	0.79, 3.18	0.193	0.89	0.38, 2.10	0.796
Brain metastases	2.12	1.04, 4.32	0.038	1.79	0.57, 5.62	0.321
Liver metastases	0.94	0.41, 2.15	0.887	0.44	0.14, 1.33	0.145
Lymph nodes metastases	1.42	0.50, 4.04	0.509	1.06	0.31, 3.66	0.926
ECOG-PS	1.50	0.89, 2.52	0.129	1.23	0.65, 2.33	0.520
Histology	1.42	0.95, 2.12	0.086	2.42	1.34, 4.37	0.003
PD-L1	0.62	0.33, 1.15	0.130	1.01	0.45, 2.25	0.986

In the second level of analysis, we evaluated the association between continuous metastatic burden (lesion area in mm²) at each organ site and overall survival using multivariable Cox regression models. Increasing bone metastatic burden was significantly associated with worse overall survival (HR 2.82, 95% CI 1.23–6.44, p = 0.014), confirming a strong dose-dependent prognostic impact of skeletal involvement. Histological subtype remained an independent prognostic factor in this model (HR 2.76, 95% CI 1.45–5.27, p = 0.002), consistently showing a strong association with survival across analytical levels. Age also retained a significant association with outcome (HR 1.06, 95% CI 1.01–1.11, p = 0.015). No significant associations were observed between overall survival and metastatic burden in lung, lymph node, adrenal, or brain sites when modeled as continuous variables. ECOG performance status and PD-L1 expression were also not significantly associated with survival in this model. (Table 3).

Table 3.

Multivariable Cox regression analysis of continuous metastatic burden (lesion area). Multivariable assessment of the prognostic impact of metastatic burden, quantified as total lesion area per organ site.

Characteristic	HR OS (Univariate)	95% CI	p-value	HR OS (Multivariate)	95% CI	p-value
Age	1.04	1.00, 1.08	0.077	1.06	1.01, 1.11	0.015
Lung metastases	3.50	0.95, 12.86	0.060	0.87	0.17, 4.39	0.862
Bone metastases	1.80	0.96, 3.38	0.066	2.82	1.23, 6.44	0.014
Adrenal metastases	1.13	0.51, 2.50	0.760	0.76	0.31, 1.89	0.559
Brain metastases	0.93	0.44, 1.96	0.842	0.41	0.15, 1.09	0.073
Liver metastases	1.50	0.79, 2.83	0.213	0.87	0.40, 1.92	0.731
Lymph nodes metastases	2.15	1.07, 4.29	0.030	1.70	0.60, 4.83	0.317
ECOG-PS	1.50	0.89, 2.52	0.129	1.32	0.71, 2.48	0.380
Histology	1.42	0.95, 2.12	0.086	2.76	1.45, 5.27	0.002
PD-L1	0.62	0.33, 1.15	0.130	0.96	0.43, 2.13	0.911

In the third analytical level, we evaluated the prognostic impact of the relative contribution of each metastatic site to the total tumor burden, expressed as the ratio between site-specific lesion area and global metastatic area. To avoid multicollinearity between competing organ-specific contributions, each metastatic site was modeled in separate multivariable Cox regression models. Across models, histological subtype consistently emerged as an independent prognostic factor, being significantly associated with overall survival in all analyses. A higher relative contribution of bone metastatic burden was significantly associated with worse overall survival (HR 3.01, 95% CI 1.23–7.36, p = 0.016), confirming skeletal disease as the most prognostically impactful metastatic compartment when accounting for global disease distribution. No significant associations were observed for adrenal, brain, or lymph node metastatic ratios in relation to overall survival. ECOG performance status, PD-L1 expression, and age showed no consistent significant associations across models, with the exception of a borderline effect of age in the bone ratio model (HR 1.04, 95% CI 1.00–1.09, p = 0.048). (Table 4).

Table 4.

Multivariable Cox regression analysis of the relative contribution of metastatic sites to total tumor burden. Prognostic impact of the relative contribution of individual metastatic sites to the global tumor load, expressed as a ratio (site-specific lesion area / total metastatic area). Each site is analyzed in a separate multivariable model to avoid multicollinearity.

Characteristic	HR	95% CI	p-value
Adrenal ratio
Adrenal ratio	1.42	0.60, 3.32	0.424
Age	1.03	0.99, 1.07	0.167
PD-L1	0.63	0.35, 1.16	0.140
ECOG-PS	1.44	0.84, 2.45	0.181
Histology	1.64	1.07, 2.54	0.025
Bone ratio
Bone ratio	3.01	1.23, 7.36	0.016
Age	1.04	1.00, 1.09	0.048
PD-L1	0.70	0.38, 1.28	0.246
ECOG-PS	1.47	0.88, 2.46	0.141
Histology	2.04	1.25, 3.32	0.004
Brain ratio
Brain ratio	1.50	0.44, 5.07	0.514
Age	1.03	0.99, 1.07	0.195
PD-L1	0.71	0.34, 1.46	0.348
ECOG-PS	1.33	0.72, 2.46	0.371
Histology	1.59	1.03, 2.45	0.035
Liver ratio
Liver ratio	0.40	0.13, 1.26	0.120
Age	1.04	1.00, 1.08	0.072
PD-L1	0.66	0.36, 1.24	0.197
ECOG-PS	1.48	0.86, 2.56	0.157
Histology	2.06	1.22, 3.46	0.007
Lymph node ratio
Lymph node ratio	0.78	0.27, 2.21	0.636
Age	1.03	0.99, 1.07	0.218
PD-L1	0.61	0.33, 1.12	0.109
ECOG-PS	1.52	0.87, 2.66	0.143
Histology	1.67	1.07, 2.62	0.024

4. Discussion

In this study, we investigated the prognostic role of metastatic burden across different organ sites in a cohort of 52 treatment-naïve patients with advanced non-oncogene-addicted NSCLC using a hierarchical analytical framework integrating binary metastatic status, continuous lesion area, and relative burden distribution. Across all analytical levels, a consistent and biologically coherent signal emerged, with skeletal metastatic involvement showing the most robust and reproducible association with overall survival. Bone metastases were independently associated with worse prognosis both when modeled as a binary variable and, more importantly, when assessed as continuous tumor burden and as relative contribution to total metastatic load, suggesting a dose-dependent and context-dependent prognostic effect. These findings are in line with previous evidence highlighting the adverse prognostic role of bone involvement in advanced lung cancer, but extend current knowledge by suggesting that not only the presence but also the quantitative dominance of skeletal disease within the metastatic landscape may carry additional prognostic information [10], [12], [13].In contrast, other metastatic sites, including lung, lymph nodes, adrenal glands, and brain, did not show consistent associations with overall survival across modeling strategies, either as binary variables or when assessed using continuous burden metrics. These findings should not be interpreted as evidence that these metastatic sites, or established clinical prognostic factors such as ECOG performance status and PD-L1 expression, lack prognostic relevance. The relatively small sample size, limited number of events, and heterogeneity in metastatic patterns may have reduced the statistical power to detect associations of clinically meaningful magnitude. In particular, the absence of a significant association between brain metastatic burden and overall survival should be interpreted with caution. Only 27% of patients in our cohort had brain metastases, which may have limited the statistical power to detect an association within this subgroup. Moreover, brain metastatic disease is clinically and biologically heterogeneous, with relevant differences in number, size, symptomatic status, local treatment, that were not comprehensively captured by our quantitative burden assessment. Therefore, the finding that skeletal burden showed a stronger statistical association with survival than other sites should not be interpreted as evidence that skeletal disease is intrinsically more clinically or biologically important than others. Rather, it should be considered an exploratory observation requiring validation in larger and independently characterized cohorts.

The ratio-based analysis further suggested that the relative distribution of metastatic burden across organs may provide complementary prognostic information beyond absolute tumor load, with bone-dominant disease phenotypes appearing to be associated with worse outcomes, supporting the concept that spatial patterns of metastatic dissemination may be clinically relevant. However, these findings should not be interpreted as establishing a definitive hierarchy of prognostic importance among metastatic sites.

From a methodological point of view, our results support the application of flexible survival models incorporating continuous imaging-derived metrics and non-linear effects, in line with recent advances in quantitative imaging biomarker research. Moreover, using advanced statistical models—such as spline-based Cox regression—allows us to better analyze tumor burden. These methods are more effective than traditional approaches because they can capture the complex, non-linear relationships between the tumor size and a patient's survival, allowing a more refined characterization of the relationship between tumor burden and survival compared with conventional dichotomous approaches [14], [15], [16], [17], [18], [19]. Importantly, our framework addresses baseline prognostic assessment rather than treatment response. RECIST 1.1 is primarily designed for treatment response assessment and remains the predominant framework used in clinical trials; however, bone metastases are generally considered non-measurable lesions unless an associated measurable soft-tissue component is present [1]. The MD Anderson (MDA) criteria provide a validated approach for assessing treatment response in bone metastases, although they are not routinely used as the primary response criteria in most clinical trials [20].

Furthermore, skeletal metastases represent a unique, clinically aggressive biological niche. Growing evidence suggests that the bone marrow microenvironment (TME) in patients with skeletal metastases is characterized by a profoundly immunosuppressive landscape. Bone-derived factors, such as TGF-β and osteoclast-activating cytokines, create a sanctuary site that not only promotes tumor progression but actively reduces T-cell infiltration and effector function [21]. Our findings suggest that quantitative assessment of skeletal burden may provide additional prognostic information beyond the simple presence of bone metastases; however, the present study cannot determine whether skeletal burden directly contributes to therapeutic resistance or poorer outcomes.

From a clinical perspective, these findings suggest that incorporating quantitative assessment of bone tumor burden into staging protocols could enhance prognostic stratification. Moreover, the prognostic relevance of skeletal involvement further underscores the importance of appropriate management of bone metastases, including the integration of antiresorptive agents according to established clinical indications [22], [23]. While some studies have explored the potential synergistic role of these agents in NSCLC patients undergoing immunotherapy, hypothesizing that by mitigating the skeletal “vicious cycle,” they might modulate the bone microenvironment, clinical evidence remains exploratory, and these agents are currently primarily utilized for skeletal-related event (SRE) prevention rather than as primary anti-tumor therapies [22], [23].

Furthermore, the development of automated or semi-automated tools for longitudinal tracking of organ-specific burden could facilitate real-time monitoring of disease evolution.

However, several limitations must be acknowledged, including the retrospective single-center design, the relatively small sample size, and the exploratory nature of the analyses, which collectively limit the generalizability of the findings. In addition, despite multivariable adjustment for key clinical covariates, residual confounding cannot be excluded. Moreover, the use of two-dimensional lesion area as a surrogate for tumor burden, while practical and reproducible, does not fully capture the true volumetric disease burden. A significant concern in our modeling strategy was the potential for overfitting, given the limited sample size relative to the complexity of metastatic patterns. To mitigate this risk, we adopted a highly conservative approach: we restricted our multivariable models to a pre-defined key clinical covariate known to influence survival. The limited sample size and number of events may have resulted in insufficient statistical power to detect established prognostic associations, particularly for less frequent metastatic sites and clinical covariates. Finally, the findings should be interpreted as hypothesis-generating and require external validation in larger, independent cohorts before any clinical translation can be considered.

In conclusion, in treatment-naïve patients with advanced non-oncogene addicted lung cancer, skeletal metastatic burden emerged as the strongest prognostic signal within this exploratory cohort. However, this finding should be interpreted as an association rather than evidence of a causal relationship or a definitive hierarchy of prognostic importance among metastatic sites.

Three complementary levels of analysis were performed: (i) assessment of the association between clinical variables and OS, as well as the presence or absence of metastases across individual organ sites; (ii) evaluation of the relationship between site-specific total lesion area and survival outcomes; (iii) calculation of the ratio between site-specific tumor burden and total metastatic burden (site area divided by total metastatic area).

Availability of data and material

The datasets generated and analyzed during the current study are available from the corresponding author on reasonable request.

AI-Assisted writing statement

During the preparation of this work, the authors used ChatGPT (OpenAI, San Francisco, CA, USA) to improve the clarity, flow, grammar, and readability of the manuscript. After using this tool, the authors carefully reviewed and edited the content as needed and take full responsibility for the final version of the publication.

CRediT authorship contribution statement

Mariam Grazia Polito: Writing – original draft, Formal analysis, Data curation, Conceptualization. Consolo Andrea: Software, Resources, Methodology. Cariglia Marcello: Software, Resources, Methodology. Simone Foderaro: Conceptualization. Marco Russano: Writing – review & editing. Alessia Vendittelli: Data curation. Giulia La Cava: Data curation. Emanuele Claudio Mingo: Data curation. Matteo Fiorenti: Data curation. Ludovica Esposito: Data curation. Cecilia Teresa Saracino: Data curation. Federica Giaccio: Data curation. Federica Bernieri: Data curation. Camilla Landone: Data curation. Beatrice De Santis: Data curation. Vittoria Colella: Project administration, Methodology. Luisana Sisca: Investigation, Data curation. Francesco Pantano: Writing – review & editing. Bruno Vincenzi: Writing – review & editing. Giuseppe Tonini: Writing – review & editing. Bruno Beomonte Zobel: Writing – review & editing. Alessio Cortellini: Writing – review & editing, Supervision, Investigation. Carlo Augusto Mallio: Writing – review & editing, Supervision.

Consent for publication

Not applicable.

Ethics approval and consent to participate

Ethical approval and informed consent were waived for this study, as it involved a retrospective analysis of previously collected clinical data.

Funding

None.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: FP reports participation in advisory boards for Lilly, Gilead, AstraZeneca, Novartis, and Daiichi Sankyo. AC reports consulting fees from Bristol-Myers Squibb, AstraZeneca, MSD Oncology, Regeneron, Amgen, Daiichi Sankyo/AstraZeneca, Roche, Access Infinity, Ardelis Health, IQVIA, Alpha Sight, Capvision, techspert.io, Atheneum, Alira Health, Tegus, and Johnson & Johnson/Janssen; honoraria from AstraZeneca, MSD Oncology, Sanofi/Regeneron, Roche, and Johnson & Johnson/Janssen; research funding (to institution) from the International Association for the Study of Lung Cancer; and travel support from Roche and MSD Oncology. BV reports consulting/advisory roles with Eisai, Eli Lilly, Novartis, PharmaMar, Abbott, GSK and Dechiphera; honoraria from Novartis, PharmaMar, Abbott, GSK, Accord and Dechiphera; institutional research funding from Eli Lilly, Novartis and PharmaMar; and testimony activities for Abbott, GSK and Accord. He declares no employment, stock ownership or other conflicts of interest. All remaining authors have declared no conflicts of interest.