Abstract
Background
The prognosis of elderly patients with distant metastasis (stage M1) of non-small cell lung cancer (NSCLC) is poor. This study aimed to develop and internally validate a prognostic model for estimating cancer-specific survival (CSS) in elderly patients with metastatic NSCLC.
Methods
Clinical data of stage M1 NSCLC patients (2010–2022) were extracted from the Surveillance, Epidemiology, and End Results database and randomly split into a training cohort and an internal validation cohort at a ratio of 7:3. The primary outcome was CSS, defined as death from NSCLC. Prespecified demographic, tumor, metastatic, and treatment-related variables were evaluated using univariate and multivariable Cox proportional hazards regression analyses, followed by bidirectional stepwise selection based on the Akaike information criterion. The cohort was randomly divided into training and internal validation cohorts at a ratio of 7:3. Model discrimination was assessed using the concordance indices (C-indexes) and time-dependent area under the receiver operating characteristic curve (AUC) at 1, 3, and 5 years, while calibration and decision curve analysis were used to evaluate model performance. A nomogram was developed based on the final model.
Results
A total of 69,516 patients were randomly allocated in the training (n=48,661) and validation (n=20,855) cohorts. In the training cohort, Cox regression analysis identified independent prognostic factors for CSS (all P<0.05), including age, gender, race, marital status, residential area, tumor size, primary tumor site, T stage, N stage, surgery, etc. The model showed moderate discrimination, with C-indices of 0.694 and 0.698 in the training and validation cohorts, respectively. The time-dependent AUCs at 1, 3, and 5 years were 0.740, 0.739, and 0.762 in the training cohort and 0.742, 0.739, and 0.766 in the validation cohort, respectively. The nomogram constructed based on the model could predict survival probability, and risk stratification showed significant differences in CSS among the three groups.
Conclusions
This study developed and internally validated a prognostic model and corresponding nomogram for estimating 1-, 3-, and 5-year CSS in elderly patients with stage M1 NSCLC. The model showed moderate and relatively consistent discriminative ability and may assist prognostic risk stratification.
Keywords: Non-small cell lung cancer (NSCLC), distant metastasis, elderly patients, cancer-specific survival (CSS), predictive model
Highlight box.
Key findings
• This study constructed and validated a cancer-specific survival (CSS) predictive model for elderly patients with stage M1 non-small cell lung cancer (NSCLC) that outperforms the traditional Tumor Node Metastasis (TNM) staging system in predicting 1-, 3-, and 5-year survival.
• It can provide clinicians with a scientific and reliable tool for assessing prognosis and formulating individualized treatment strategies.
What is known and what is new?
• The American Joint Committee on Cancer (AJCC) TNM staging system, the gold standard for NSCLC prognostic assessment, focuses solely on tumor pathological characteristics and fails to account for the individualized clinical factors of elderly patients. Existing predictive models are either constrained by small sample sizes or lack specificity for elderly patients with distant metastatic NSCLC, severely limiting their clinical utility.
• This study aims to construct and validate an accurate CSS predictive model for elderly NSCLC patients with distant metastasis, thereby providing clinicians with a scientific basis for formulating individualized treatment plans, precise prognosis assessment, and optimized follow-up management strategies.
What is the implication, and what should change now?
• We constructed a validated CSS model for elderly M1 NSCLC patients, superior to TNM staging, to support clinical prognostic evaluation and personalized treatment.
• The model exhibited excellent prognostic performance, with higher concordance indices (C-indexes) than the TNM staging system in both cohorts (training: 0.694 vs. 0.547; validation: 0.698 vs. 0.546). In addition, this model outperforms the traditional TNM staging system in predicting 1-, 3-, and 5-year survival.
Introduction
Background
Lung cancer remains a major global health burden and is the most frequently diagnosed malignancy and the leading cause of cancer-related mortality worldwide. In 2022, approximately 2.5 million new cases and 1.8 million deaths from lung cancer were estimated globally (1). Non-small cell lung cancer (NSCLC) accounts for the vast majority of lung cancer cases and remains a major contributor to cancer-related morbidity and mortality. Distant metastasis is a major contributor to poor outcomes in NSCLC, with the most common metastatic sites including bone, brain, and liver (2). Once the tumor breaks through local and regional barriers and develops distant organ metastasis, the 5-year overall survival rate of patients drops sharply to less than 5% (3), posing substantial challenges for treatment selection and survival management.
With the accelerating global aging, the proportion of elderly NSCLC patients (≥65 years old) has been rising annually, currently accounting for over half of all NSCLC patients and becoming the main affected population (4). Compared with younger patients, elderly NSCLC patients are often affected by comorbidities such as hypertension, diabetes, and cardiovascular diseases. Their physical functional reserve and tolerance to anti-tumor treatment are significantly reduced, which may narrow the range of options for treatment regimens such as surgery, chemotherapy, and radiotherapy, and ultimately results in worse overall prognosis (5). Particularly, elderly NSCLC patients with distant metastasis bear the dual risks of advanced tumors and age-related physiological frailty. Their cancer-specific mortality risk is significantly higher than that of younger metastatic patients or elderly non-metastatic patients, making them a challenging group in clinical prognosis assessment and treatment decision-making (6,7). However, the relative contribution of these factors and their ability to provide individualized survival estimates remain incompletely characterized in elderly patients with metastatic disease.
Several prognostic models have previously been developed for patients with metastatic or elderly NSCLC. For example, one study based on the Surveillance, Epidemiology, and End Results (SEER) database developed and validated nomograms for elderly patients (8). Other SEER-based models have examined metastatic NSCLC in broader populations or focused on specific elderly subgroups, including patients with squamous cell carcinoma or locally advanced disease (9,10). These studies indicate that clinicopathological and treatment-related variables can improve prognostic discrimination beyond conventional Tumor Node Metastasis (TNM) staging. Nevertheless, important differences remain among existing models in terms of the target population, outcome definition, predictor selection, modeling strategy, and validation approach. In particular, relatively few models have specifically focused on patients aged ≥65 years with M1 NSCLC and cancer-specific survival (CSS) using contemporary population-based data.
The SEER database provides a large, population-based cohort with standardized information on demographic, tumor, treatment, and survival characteristics, making it suitable for developing population-level prognostic models. Nevertheless, SEER lacks several clinically important variables, including performance status, smoking history, frailty, detailed comorbidity burden, driver mutations, PD-L1 expression, and detailed information on targeted therapy, immunotherapy, and treatment sequence. These limitations may constrain the predictive accuracy and generalizability of SEER-based models and should therefore be considered when interpreting their clinical utility.
The current mainstream tool for NSCLC prognosis assessment is the American Joint Committee on Cancer (AJCC) TNM staging system (11). However, TNM staging does not incorporate many patient-level and treatment-related factors that may influence prognosis, particularly in older patients with metastatic disease. Therefore, a model integrating readily available demographic, tumor, metastatic, and treatment-related characteristics may provide complementary prognostic risk stratification beyond TNM staging.
Objective
Based on the SEER database, this study aimed to develop and internally validate a prognostic model for CSS in patients aged ≥65 years with M1 NSCLC. Given the inherent limitations of SEER, the model was intended as a population-based prognostic tool rather than a comprehensive individualized prediction model. We present this article in accordance with the TRIPOD reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1475/rc).
Methods
Study cohort
This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was based on a SEER-derived dataset of patients with NSCLC who had documented distant metastases at diagnosis. Distant metastatic status was determined primarily using SEER stage- and metastasis-related variables, including AJCC stage, Extent of Disease stage, and SEER Combined Mets at DX variables. Because these variables capture disease stage or metastatic status at diagnosis, the study population comprised patients with distant metastatic disease at initial diagnosis rather than patients who developed distant metastases during subsequent follow-up.
Patients diagnosed with NSCLC between 2010 and 2022 were selected, with the following inclusion criteria: (I) aged 65–85 years; (II) pathologically confirmed NSCLC, with tumor location corresponding to the International Classification of Diseases for Oncology, 3rd Edition (ICD-O-3) code C34.X (lung and bronchus), and tumor stage classified as distant metastasis (M1 stage); (III) histological types defined as adenocarcinoma, squamous cell carcinoma, and other NSCLC subtypes based on specific ICD-O-3 codes (detailed code categories are presented in Table S1).
Patients were excluded if they had unknown or invalid survival data, unknown or inapplicable surgery status, unknown or missing urban-rural classification, unavailable or invalid T or N stage, or unknown status of bone, brain, liver, or lung metastases. After application of the eligibility and exclusion criteria, 69,516 patients were included in the final study cohort. The detailed patient selection process is presented in Figure 1.
Figure 1.
Flowchart of patient selection. ICD-O-3, the International Classification of Diseases for Oncology, 3rd Edition; NSCLC, non-small cell lung cancer; SEER, Surveillance, Epidemiology, and End Results.
Variables
The extracted clinical variables were categorized into three groups, with demographic variables including age, gender, race, marital status, and residential location; tumor characteristics covering T and N stages, tumor size, primary site, histological type, and metastatic sites; and treatment-related variables involving surgery, radiation, and chemotherapy. The primary outcome of this study was CSS, defined as the time interval from the date of NSCLC diagnosis to death due to NSCLC. For the CSS analysis, death from NSCLC was defined as the event (coded as 1), whereas patients who remained alive at the end of follow-up or died from causes other than NSCLC were coded as 0 and treated as censored observations.
Statistical analysis
All statistical analyses were performed using R software (Version 4.4.1), and a two-tailed P<0.05 was considered statistically significant.
Missing data were handled using multiple imputation. Multiple imputation was performed using the mice package in R, with a fixed random seed to ensure reproducibility. Five imputed datasets were generated with 20 iterations per dataset. Predictive mean matching was used for continuous variables, logistic regression for binary variables, and classification and regression tree (CART) methods for multicategorical variables. The imputation model included the candidate predictors and outcome-related variables. Multiple imputation was used to reduce potential bias associated with missing data and to preserve statistical power.
The study population was randomly divided into training and internal validation cohorts at a ratio of 7:3. The model was developed in the training cohort and internally validated in the validation cohort. Continuous variables were presented as mean ± standard deviation (SD) or median, and inter-group comparisons were conducted using the independent samples t-test or Wilcoxon rank-sum test, respectively. Categorical variables were expressed as frequency and percentage, with inter-group differences compared using the chi-square test or Fisher’s exact test.
Univariate Cox proportional hazards regression analyses were initially performed for the prespecified candidate predictors to examine their individual associations with CSS. Subsequently, all prespecified candidate predictors were entered into an initial multivariable Cox proportional hazards model in the training cohort. A bidirectional stepwise selection procedure based on the Akaike information criterion was then applied to identify the final set of predictors. The variables retained in the final model were used to construct the CSS prognostic model and nomogram. Age and tumor size were included as continuous variables in the Cox regression model without any manual categorization, to avoid information loss resulting from arbitrary truncation. For categorical variables, reference categories were specified during the modeling process, and these variables were incorporated into the Cox proportional hazards model as dummy variables.
Model discrimination was evaluated using the concordance index (C-index) and time-dependent receiver operating characteristic (ROC) curves. Time-dependent area under the ROC curve (AUC) values were calculated at 12, 36, and 60 months to assess the discrimination of the model for 1-, 3-, and 5-year CSS, respectively. The corresponding 95% confidence intervals (CIs) were reported. Sensitivity and specificity based on a single fixed cutoff were not used as primary performance measures because no prespecified clinical threshold was available for this time-to-event prediction model. Calibration was assessed by comparing the predicted and observed CSS probabilities at 12, 36, and 60 months using calibration curves in both the training and internal validation cohorts. Decision curve analysis (DCA) was performed to evaluate the potential net benefit of the model across a range of threshold probabilities. The performance of the prognostic model was also compared with that of the TNM staging system using the C-index and time-dependent AUC.
Based on the screened independent prognostic factors, a nomogram was developed. For this nomogram, each independent prognostic factor was assigned a corresponding score, and the total score of each patient was calculated by summing these individual scores. The CSS probability of each patient was then predicted based on their total score. According to the risk scores of patients in the training cohort, patients were stratified into three risk groups: low-risk (score <25th percentile), intermediate-risk (25th percentile ≤ score ≤ 75th percentile), and high-risk (score 75th percentile). Finally, Kaplan-Meier (K-M) survival curves and the log-rank test were used to analyze CSS outcomes across different risk groups, compare inter-group survival differences, and verify the validity of the risk stratification.
Results
Baseline characteristics of patients
A total of 69,516 patients with stage M1 NSCLC were included, consisting of 15,604 patients in the other group and 53,912 patients in the death group. Before imputation, missing data were identified for tumor size, marital status, and race. The detailed missing-data patterns are presented in Table S2. The mean CSS time in the death group was significantly shorter (9.85±13.36 vs. 21.76±26.48 months, P<0.001). The death group had a slightly higher mean age (73.82±5.72 vs. 73.38±5.61 years, P<0.001), a higher proportion of males (55.2% vs. 49.8%, P<0.001), and a higher proportion of White individuals (80.4% vs. 75.8%, P<0.001) (Table 1). Comparison of tumor characteristics revealed that the death group had a larger mean tumor diameter (48.72 vs. 43.92 mm), higher proportions of stage T4 (36.6% vs. 35.1%) and stage N2 (43.8% vs. 36.6%), and more frequent bone metastasis (40.4% vs. 33.7%), brain metastasis (24.9% vs. 22.1%), and liver metastasis (17.6% vs. 11.6%).
Table 1. Baseline characteristics of NSCLC patients.
| Characteristics | Level | Other (N=15,604) | Cancer-specific death (N=53,912) | P |
|---|---|---|---|---|
| Survival (months) | 21.76 (26.48) | 9.85 (13.36) | <0.001 | |
| Chemotherapy | No | 6,031 (38.7) | 25,115 (46.6) | <0.001 |
| Yes | 9,573 (61.3) | 28,797 (53.4) | ||
| Radiation | No | 9,407 (60.3) | 30,578 (56.7) | <0.001 |
| Yes | 6,197 (39.7) | 23,334 (43.3) | ||
| Surgery | No | 14,628 (93.7) | 52,278 (97.0) | <0.001 |
| Yes | 976 (6.3) | 1,634 (3.0) | ||
| Age (years) | 73.38 (5.61) | 73.82 (5.72) | <0.001 | |
| Gender | Female | 7,836 (50.2) | 24,171 (44.8) | <0.001 |
| Male | 7,768 (49.8) | 29,741 (55.2) | ||
| Race | AP/AI/AN | 2,040 (13.1) | 4,754 (8.8) | <0.001 |
| Black | 1,732 (11.1) | 5,791 (10.7) | ||
| White | 11,832 (75.8) | 43,367 (80.4) | ||
| Marital | Married | 8,766 (56.2) | 30,017 (55.7) | 0.27 |
| Single | 6,838 (43.8) | 23,895 (44.3) | ||
| Income | $50,000–$70,000 | 3,342 (21.4) | 14,077 (26.1) | <0.001 |
| <$50,000 | 897 (5.7) | 4,227 (7.8) | ||
| >$70,000 | 11,365 (72.8) | 35,608 (66.0) | ||
| Residential area | Metropolitan | 13,593 (87.1) | 45,779 (84.9) | <0.001 |
| Non-metropolitan | 2,011 (12.9) | 8,133 (15.1) | ||
| T stage | T0 | 166 (1.1) | 515 (1.0) | <0.001 |
| T1 | 2,673 (17.1) | 6,458 (12.0) | ||
| T2 | 3,769 (24.2) | 13,921 (25.8) | ||
| T3 | 3,512 (22.5) | 13,260 (24.6) | ||
| T4 | 5,484 (35.1) | 19,758 (36.6) | ||
| N stage | N0 | 5,372 (34.4) | 14,396 (26.7) | <0.001 |
| N1 | 1,378 (8.8) | 4,727 (8.8) | ||
| N2 | 5,707 (36.6) | 23,602 (43.8) | ||
| N3 | 3,147 (20.2) | 11,187 (20.8) | ||
| Tumor size (mm) | 43.92 (23.61) | 48.72 (24.61) | <0.001 | |
| Primary site | Lower lobe lung | 4,458 (28.6) | 15,165 (28.1) | <0.001 |
| Lung NOS | 1,551 (9.9) | 6,252 (11.6) | ||
| Main bronchus | 494 (3.2) | 2,315 (4.3) | ||
| Middle lobe lung | 670 (4.3) | 2,157 (4.0) | ||
| Overlapping lesion of lung | 139 (0.9) | 562 (1.0) | ||
| Upper lobe lung | 8,292 (53.1) | 27,461 (50.9) | ||
| Bone metastasis | No | 10,344 (66.3) | 32,129 (59.6) | <0.001 |
| Yes | 5,260 (33.7) | 21,783 (40.4) | ||
| Brain metastasis | No | 12,160 (77.9) | 40,508 (75.1) | <0.001 |
| Yes | 3,444 (22.1) | 13,404 (24.9) | ||
| Liver metastasis | No | 13,792 (88.4) | 44,436 (82.4) | <0.001 |
| Yes | 1,812 (11.6) | 9,476 (17.6) | ||
| Lung metastasis | No | 10,631 (68.1) | 3,6580 (67.9) | 0.52 |
| Yes | 4,973 (31.9) | 17,332 (32.1) | ||
| Histologic type | Adenocarcinoma | 10,555 (67.6) | 32,074 (59.5) | <0.001 |
| Other | 1,881 (12.1) | 9,304 (17.3) | ||
| Squamous cell carcinoma | 3,168 (20.3) | 12,534 (23.2) |
Data are presented as n (%) or mean (standard deviation). AI, American Indian; AN, Alaska native; AP, Asian/Pacific Islander; N, node; NOS, not otherwise specified; NSCLC, non-small cell lung cancer; T, tumor.
The 69,516 patients were randomly divided into two cohorts at a ratio of 7:3, including a training cohort (n=48,661) and an internal validation cohort (n=20,855). Baseline characteristics of the two cohorts are shown in Table S3. Results indicated that the training cohort and validation cohort were highly consistent in key indicators such as the distribution of distant metastatic sites, composition of histological types, and proportion of treatment modalities (all P>0.05), indicating that the data splitting was reasonable and suitable for subsequent model construction and validation.
Screening of independent prognostic factors in the training cohort
To identify independent factors associated with CSS, we performed univariate and multivariate Cox regression analyses. Univariate Cox regression analysis revealed that age, gender, race, marital status, income, residential area, tumor size, primary site, T stage, N stage, surgery, chemotherapy, distant metastatic sites, and histological types with CSS (P<0.05, Table 2). All prespecified candidate predictors were subsequently entered into an initial multivariable Cox proportional hazards model. This analysis confirmed that all 14 factors remained independent prognostic factors for CSS (P<0.05, Table 3).
Table 2. Univariate Cox regression for the risk factors of CSS.
| Variable | Level | HR (95% CI) | P |
|---|---|---|---|
| Age | 1.02 (1.02–1.02) | <0.001 | |
| Tumor size | 1.01 (1.01–1.01) | <0.001 | |
| Marital | Single | 1.10 (1.08–1.13) | <0.001 |
| Residential area | Non-metropolitan | 1.16 (1.13–1.19) | <0.001 |
| Gender | Male | 1.22 (1.20–1.25) | <0.001 |
| Bone metastasis | Yes | 1.33 (1.30–1.35) | <0.001 |
| Brain metastasis | Yes | 1.24 (1.21–1.27) | <0.001 |
| Liver metastasis | Yes | 1.51 (1.47–1.55) | <0.001 |
| Lung metastasis | Yes | 0.96 (0.94–0.98) | <0.001 |
| Surgery | Yes | 0.47 (0.44–0.50) | <0.001 |
| Radiation | Yes | 1.01 (0.99–1.03) | 0.19 |
| Chemotherapy | Yes | 0.51 (0.50–0.52) | <0.001 |
| Histologic type | Other | 1.44 (1.40–1.48) | <0.001 |
| Squamous cell carcinoma | 1.28 (1.25–1.31) | <0.001 | |
| Income | <$50,000 | 1.06 (1.02–1.11) | 0.003 |
| >$70,000 | 0.85 (0.83–0.87) | <0.001 | |
| N stage | N1 | 1.17 (1.12–1.21) | <0.001 |
| N2 | 1.34 (1.31–1.37) | <0.001 | |
| N3 | 1.28 (1.24–1.32) | <0.001 | |
| Primary site | Lung NOS | 1.17 (1.13–1.21) | <0.001 |
| Main bronchus | 1.32 (1.25–1.39) | <0.001 | |
| Middle lobe lung | 0.94 (0.89–1.00) | 0.03 | |
| Overlapping lesion of lung | 1.08 (0.97–1.19) | 0.15 | |
| Upper lobe lung | 0.99 (0.97–1.02) | 0.52 | |
| Race | Black | 1.41 (1.34–1.47) | <0.001 |
| White | 1.36 (1.31–1.41) | <0.001 | |
| T stage | T1 | 0.78 (0.70–0.87) | <0.001 |
| T2 | 1.00 (0.90–1.11) | 0.96 | |
| T3 | 1.06 (0.96–1.18) | 0.26 | |
| T4 | 1.08 (0.98–1.20) | 0.13 |
CI, confidence interval; CSS, cancer-specific survival; HR, hazard ratio; N, node; NOS, not otherwise specified; T, tumor.
Table 3. Multivariate Cox regression for the risk factors of CSS.
| Variable | Level | HR (95% CI) | P |
|---|---|---|---|
| Age | 1.01 (1.01–1.01) | <0.001 | |
| Tumor size | 1.00 (1.00–1.01) | <0.001 | |
| Marital | Single | 1.08 (1.06–1.10) | <0.001 |
| Residential area | Non-metropolitan | 1.04 (1.00–1.07) | 0.047 |
| Gender | Male | 1.21 (1.18–1.23) | <0.001 |
| Bone metastasis | Yes | 1.38 (1.35–1.41) | <0.001 |
| Brain metastasis | Yes | 1.34 (1.31–1.38) | <0.001 |
| Liver metastasis | Yes | 1.39 (1.35–1.43) | <0.001 |
| Lung metastasis | Yes | 0.96 (0.94–0.98) | <0.001 |
| Surgery | Yes | 0.55 (0.52–0.58) | <0.001 |
| Chemotherapy | Yes | 0.48 (0.47–0.49) | <0.001 |
| Histologic type | Other | 1.27 (1.23–1.30) | <0.001 |
| Squamous cell carcinoma | 1.16 (1.13–1.19) | <0.001 | |
| Income | <$50000 | 1.03 (0.99–1.08) | 0.13 |
| >$70000 | 0.89 (0.87–0.91) | <0.001 | |
| N stage | N1 | 1.17 (1.12–1.22) | <0.001 |
| N2 | 1.32 (1.28–1.35) | <0.001 | |
| N3 | 1.30 (1.26–1.34) | <0.001 | |
| Primary site | Lung NOS | 1.20 (1.16–1.24) | <0.001 |
| Main bronchus | 1.20 (1.13–1.26) | <0.001 | |
| Middle lobe lung | 0.96 (0.91–1.02) | 0.17 | |
| Overlapping lesion of lung | 1.03 (0.93–1.14) | 0.53 | |
| Upper lobe lung | 0.96 (0.94–0.99) | 0.001 | |
| Race | Black | 1.28 (1.22–1.34) | <0.001 |
| White | 1.30 (1.25–1.34) | <0.001 | |
| T stage | T1 | 0.99 (0.89–1.10) | 0.82 |
| T2 | 1.18 (1.07–1.32) | 0.002 | |
| T3 | 1.19 (1.07–1.33) | 0.001 | |
| T4 | 1.20 (1.08–1.34) | <0.001 |
CI, confidence interval; CSS, cancer-specific survival; HR, hazard ratio; N, node; NOS, not otherwise specified; T, tumor.
Evaluation of the predictive model
The discriminative ability of the model was evaluated using the C-index (Table 4) and AUC (Figure 2). In the training cohort, the Cox predictive model constructed based on independent prognostic factors had a C-index of 0.694, and the AUC values for 1-year, 3-year, and 5-year CSS prediction were 0.740 (95% CI: 0.736–0.744), 0.739 (95% CI: 0.732–0.745), and 0.762 (95% CI: 0.752–0.772), respectively (Figure 2A). In the validation cohort, the model had a C-index of 0.698, and the AUC values for 1-, 3-, and 5-year CSS prediction were 0.742 (95% CI: 0.735–0.749), 0.739 (95% CI: 0.727–0.751), and 0.766 (95% CI: 0.747–0.785), respectively (Figure 2B). These findings indicated that the model demonstrated moderate and relatively consistent discriminative ability in both the training and internal validation cohorts.
Table 4. Comparison of C-indexes between the Cox model and TNM staging system.
| Model | Training cohort | Validation cohort |
|---|---|---|
| Cox model | 0.694 (0.691–0.697) | 0.698 (0.694–0.702) |
| TNM staging system | 0.547 (0.544–0.551) | 0.546 (0.541–0.551) |
Data are presented as C-indexes (95% CI). CI, confidence interval; C-indexes, concordance indices; TNM, tumor-node-metastasis.
Figure 2.
ROC curves for the Cox predictive model or TNM staging system of CSS on the training cohort and validation cohort. (A) Cox predictive model on training cohort; (B) Cox predictive model on validation cohort; (C) TNM staging system on training cohort; (D) TNM staging system on validation cohort. CI, confidence interval; CSS, cancer-specific survival; ROC, receiver operating characteristic; T, time; TNM, Tumor Node Metastasis.
In contrast, the traditional TNM staging system showed poor predictive performance: in the training cohort, its C-index was only 0.547, and the AUC values for 1-, 3-, and 5-year CSS prediction were 0.566 (95% CI: 0.562–0.571), 0.579 (95% CI: 0.571–0.587), and 0.591 (95% CI: 0.578–0.603), respectively (Figure 2C); in the validation cohort, its C-index was 0.546, and the AUC values for 1-, 3-, and 5-year CSS prediction were 0.559 (95% CI: 0.550–0.568), 0.598 (95% CI: 0.582–0.611), and 0.620 (95% CI: 0.597–0.643), respectively (Figure 2D).
Results of DCA showed that, in both the training and validation cohorts, the Cox model constructed in this study (Figure 3) had significantly higher net benefits than the traditional TNM staging system (Figure 4) within the range of most threshold probabilities for 1-, 3-, and 5-year CSS prediction. These findings suggested that the model may provide greater potential clinical utility than the TNM staging system within the evaluated population.
Figure 3.
DCA for the Cox predictive model of 1-, 3-, 5-year CSS on the training cohort (A-C) as well as in the validation cohort (D-F). AIC, Akaike Information Criterion; CSS, cancer-specific survival; DCA, decision curve analysis.
Figure 4.
DCA for the TNM staging system of 1-, 3-, 5-year CSS on the training cohort (A-C) as well as in the validation cohort (D-F). CSS, cancer-specific survival; DCA, decision curve analysis; TNM, Tumor Node Metastasis.
Calibration curves showed relatively good agreement between the predicted and observed probabilities of 1-, 3-, and 5-year CSS in both the training and internal validation cohorts (Figure S1). The calibration performance was generally consistent across the evaluated time points, although some deviations between predicted and observed probabilities were observed.
Construction of the nomogram and risk stratification
A visual nomogram (Figure 5) was created based on the independent prognostic factors identified by multivariate Cox regression to intuitively estimate the 1-, 3-, and 5-year CSS probabilities of patients. In the nomogram, each independent prognostic factor was assigned a corresponding score according to its regression coefficient.
Figure 5.
Prognostic nomograms of 1-, 3-, and 5-year CSS. AI, American Indian; AN, Alaska native; AP, Asian/Pacific Islander; CSS, cancer-specific survival; N, node; NOS, not otherwise specified; T, tumor.
Patients were stratified into three risk groups based on the risk scores in the training cohort: low-risk group (score <0.102), intermediate-risk group (0.102≤ score ≤0.933), and high-risk group (score >0.933). Results of K-M survival analysis showed that there were significant differences in CSS curves among the three groups in both the training and validation cohorts (log-rank P<0.0001, Figure 6). This indicated that the nomogram-derived risk stratification could distinguish patients with different CSS outcomes within the study population.
Figure 6.
Kaplan-Meier survival curves for CSS after risk stratification in the training cohort (A) and validation cohort (B). CSS, cancer-specific survival.
Discussion
Based on a large-sample cohort from the SEER database, this study targeted elderly patients with stage M1 NSCLC. Through univariate and multivariate Cox proportional hazards regression models, we identified multi-dimensional independent prognostic factors covering demographic characteristics, tumor pathological features, and treatment modalities, constructed and validated a predictive model, and simultaneously developed a visual nomogram to quantify patients' survival risk. This may assist prognostic risk assessment decision-making in elderly patients with stage M1 NSCLC.
Univariate and multivariate Cox analyses revealed that age, gender, race, marital status, income, residential area, tumor size, primary site, T stage, N stage, surgery, chemotherapy, distant metastatic sites, and histological types were independent factors affecting CSS, which highly consistent with the results of similar studies (12,13). With increasing age, the structure and function of the human immune system undergo changes. Elderly patients are prone to immunosenescence, which enables tumors to evade immune surveillance (14); meanwhile, the decline in physiological function increases the risk of treatment-related toxicity, indirectly affecting survival. Male gender and White race were associated with poorer prognosis. Racial differences in NSCLC survival have also been documented, although the direction and magnitude of these disparities vary across racial groups and clinical settings (15). Barquín et al. found that female NSCLC patients had significantly longer survival than male patients (16). This gender difference may be related to variations in hormone and receptor expression levels (17).
Patients with higher household income or residing in metropolitan areas had better CSS, which may be attributed to differences in access to medical resources. Metropolitan areas have dense distribution of oncology centers, allowing patients to more easily receive standardized multi-disciplinary treatment; at the same time, high-income groups show better treatment adherence, which can reduce prognosis deterioration caused by treatment delay or interruption (18,19). Similarly, married patients usually have more stable care and emotional support, which helps improve treatment tolerance and follow-up completeness (20).
Among tumor pathological features, tumor size, primary tumor site, T stage, N stage, histological type, and metastatic sites were core prognostic factors. Larger tumors and more advanced T/N stages were associated with higher mortality risk, which aligns with the pathological principle that “higher tumor burden increases the probability of vascular/lymphatic invasion and accelerates the progression of distant metastasis” (21,22). The AJCC TNM staging system further subdivides T1 into three subgroups (T1a, T1b, T1c) based on every 1 cm increment in tumor size, which also confirms that tumor size is an extremely important prognostic factor for NSCLC (11). In addition, multiple studies have reported that poorer histological differentiation, advanced T stage, and advanced N stage are significantly associated with increased NSCLC mortality (23,24). Among histological types, adenocarcinoma was associated with better prognosis than squamous cell carcinoma and other subtypes, which is highly relevant to current advances in clinical treatment. Adenocarcinoma patients have higher mutation rates of driver genes such as epidermal growth factor receptor, anaplastic lymphoma kinase and ROS1, and their survival can be significantly prolonged through targeted therapy (25,26). Patients who received surgery or chemotherapy had higher survival rates, indicating that despite the decline in physical function in elderly patients, active treatment can still yield survival benefits when their physical condition permits (9).
Finally, we validated the performance of the model. Results showed that the C-index and ROC analysis of the predictive model performed well in both the training and validation cohorts and were superior to those of the traditional TNM staging system. DCA further demonstrated that the newly developed model had higher net benefits and greater clinical applicability than the TNM staging system. The development of a visual nomogram also lowered the threshold for clinical application.
However, this study has several limitations. First, it is subject to inherent limitations of the SEER database: details on treatment (e.g., chemotherapy regimens, use of targeted drugs, and exposure to immunotherapy) are lacking. Therefore, although the large sample size and substantial number of CSS events provided a favorable basis for model development, the limited availability of clinically relevant predictors may constrain the predictive performance of the model, and a large sample size alone does not guarantee high predictive accuracy. Second, the study excluded patients with incomplete clinical information, leading to a certain degree of selection bias. Third, although CSS was selected to specifically assess NSCLC-related mortality in this elderly population, deaths from other causes were treated as censored observations in the primary analysis. Non-cancer mortality is an important competing event in older patients and may therefore influence CSS estimates. In addition, cause of death in the SEER database may be subject to misclassification. These factors should be considered when interpreting the CSS predictions. Finally, although the model demonstrated relatively consistent performance between the training and internal validation cohorts, validation was performed using a cohort derived from the same SEER population. Therefore, internal validation alone cannot establish the external transportability or generalizability of the model to other populations or clinical settings. In the future, multi-center external validation and prospective validation can be conducted, which is expected to further promote the development of precision medicine for elderly NSCLC patients.
Conclusions
This study developed and internally validated a prognostic model and corresponding nomogram for estimating 1-, 3-, and 5-year CSS in elderly patients with M1 NSCLC. The model demonstrated moderate discrimination and could stratify patients into groups with different CSS outcomes within the SEER cohort. However, its clinical applicability and generalizability remain uncertain and require validation in independent external cohorts.
Supplementary
The article’s supplementary files as
Acknowledgments
We would like to thank the SEER database for providing publicly available data for this research.
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.
Footnotes
Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1475/rc
Funding: This work was supported by the Basic Public Welfare Research Project of Pingyang County (No. Y2025A14, to W.L.).
Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1475/coif). W.L. reports that this work was supported by the Basic Public Welfare Research Project of Pingyang County (No. Y2025A14). The other authors have no conflicts of interest to declare.
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;74:229-263.. 10.3322/caac.21834 [DOI] [PubMed] [Google Scholar]
- 2. Wang S, Tang W, Jin F, et al. Comprehensive Analysis of Lung Cancer Metastasis: Sites, Rates, Survival, and Risk Factors-A Systematic Review and Meta-Analysis. Clin Respir J 2025;19:e70107.. 10.1111/crj.70107 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Arbour KC, Riely GJ. Systemic Therapy for Locally Advanced and Metastatic Non-Small Cell Lung Cancer: A Review. JAMA 2019;322:764-774.. 10.1001/jama.2019.11058 [DOI] [PubMed] [Google Scholar]
- 4. Chen Q, Ying S, Qin J, et al. Optimization of treatment strategies for elderly patients with advanced non-small cell lung cancer. Front Oncol 2024;14:1384906.. 10.3389/fonc.2024.1384906 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Driessen EJ, Bootsma GP, Hendriks LE, et al. Stage III Non-Small Cell Lung Cancer in the elderly: Patient characteristics predictive for tolerance and survival of chemoradiation in daily clinical practice. Radiother Oncol 2016;121:26-31.. 10.1016/j.radonc.2016.07.025 [DOI] [PubMed] [Google Scholar]
- 6. Dong Y, Zhou S, Li J, et al. Distant metastatic patterns in young and old non-small cell lung cancer patients: A dose‒response analysis based on SEER population. Heliyon 2024;10:e36657.. 10.1016/j.heliyon.2024.e36657 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Gu Y, Zhang J, Zhou Z, et al. Metastasis Patterns and Prognosis of Octogenarians with NSCLC: A Population-based Study. Aging Dis 2020;11:82-92.. 10.14336/ad.2019.0414 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Sun H, Liu M, Yang X, et al. Construction and validation of prognostic nomograms for elderly patients with metastatic non-small cell lung cancer. Clin Respir J 2022;16:380-393.. 10.1111/crj.13491 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Chen S, Gao C, Du Q, et al. A prognostic model for elderly patients with squamous non-small cell lung cancer: a population-based study. J Transl Med 2020;18:436.. 10.1186/s12967-020-02606-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Guo M, Li B, Yu Y, et al. Delineating the pattern of treatment for elderly locally advanced NSCLC and predicting outcomes by a validated model: A SEER based analysis. Cancer Med 2019;8:2587-2598.. 10.1002/cam4.2127 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Rami-Porta R, Asamura H, Travis WD, et al. Lung cancer - major changes in the American Joint Committee on Cancer eighth edition cancer staging manual. CA Cancer J Clin 2017;67:138-155.. 10.3322/caac.21390 [DOI] [PubMed] [Google Scholar]
- 12. Liang W, Zhang L, Jiang G, et al. Development and validation of a nomogram for predicting survival in patients with resected non-small-cell lung cancer. J Clin Oncol 2015;33:861-869.. 10.1200/jco.2014.56.6661 [DOI] [PubMed] [Google Scholar]
- 13. Zhou L, Zhang Y, Chen W, et al. Development and validation of a prognostic nomogram for early stage non-small cell lung cancer: a study based on the SEER database and a Chinese cohort. BMC Cancer 2022;22:980.. 10.1186/s12885-022-10067-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Berben L, Floris G, Kenis C, et al. Age-related remodelling of the blood immunological portrait and the local tumor immune response in patients with luminal breast cancer. Clin Transl Immunology 2020;9:e1184.. 10.1002/cti2.1184 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Wannaphut C, Ponvilawan B, Kanitthamniyom C, et al. Racial disparities in non-small cell lung cancer survival outcomes: a systematic review and meta-analysis. Proc (Bayl Univ Med Cent) 2025;38:722-731.. 10.1080/08998280.2025.2524792 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Barquín M, Calvo V, García-García F, et al. Sex is a strong prognostic factor in stage IV non-small-cell lung cancer patients and should be considered in survival rate estimation. Cancer Epidemiol 2020;67:101737.. 10.1016/j.canep.2020.101737 [DOI] [PubMed] [Google Scholar]
- 17. Thaker R, Lee UJ, Strange TJ, et al. Estrogen receptor subtypes and survival outcomes in non-small cell lung cancer. Pathol Res Pract 2025;268:155843.. 10.1016/j.prp.2025.155843 [DOI] [PubMed] [Google Scholar]
- 18. Forrest LF, Adams J, Rubin G, et al. The role of receipt and timeliness of treatment in socioeconomic inequalities in lung cancer survival: population-based, data-linkage study. Thorax 2015;70:138-145.. 10.1136/thoraxjnl-2014-205517 [DOI] [PubMed] [Google Scholar]
- 19. Yang X, Deng L, Li M, et al. Impact of socioeconomic status on cancer staging, survival in non-small cell lung cancer. Front Public Health 2022;10:992944.. 10.3389/fpubh.2022.992944 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Aizer AA, Chen MH, McCarthy EP, et al. Marital status and survival in patients with cancer. J Clin Oncol 2013;31:3869-3876.. 10.1200/jco.2013.49.6489 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Ma Q, Dieterich LC, Ikenberg K, et al. Unexpected contribution of lymphatic vessels to promotion of distant metastatic tumor spread. Sci Adv 2018;4:eaat4758.. 10.1126/sciadv.aat4758 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Zhang J, Gold KA, Lin HY, et al. Relationship between tumor size and survival in non-small-cell lung cancer (NSCLC): an analysis of the surveillance, epidemiology, and end results (SEER) registry. J Thorac Oncol 2015;10:682-690.. 10.1097/jto.0000000000000456 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Wang Q, Wang Y, Wang X, et al. Development and validation of a dynamic survival nomogram for metastatic non-small cell lung cancer based on the SEER database and an external validation cohort. Transl Lung Cancer Res 2022;11:1678-1691.. 10.21037/tlcr-22-544 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Yang F, Gao L, Wang Q, et al. Development and validation of nomograms to predict early death in non-small cell lung cancer patients with brain metastasis: a retrospective study in the SEER database. Transl Cancer Res 2023;12:473-489.. 10.21037/tcr-22-2323 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Guo Z, Yi F, Yin W, et al. Clinical value of the new International Association for the Study of Lung Cancer/American Thoracic Society/European Respiratory Society classification of lung adenocarcinoma. Thorac Cancer 2017;8:159-169.. 10.1111/1759-7714.12419 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Huang CY, Chen BH, Chou WC, et al. Factors associated with the prognosis and long-term survival of patients with metastatic lung adenocarcinoma: a retrospective analysis. J Thorac Dis 2018;10:2070-2078.. 10.21037/jtd.2018.03.143 [DOI] [PMC free article] [PubMed] [Google Scholar]