Abstract
Background
Patients with coexisting emphysema and lung cancer present a complex clinical prognosis, yet current research evidence on this population remains limited. This study aimed to evaluate the prognostic value of CT-defined emphysema for extrapulmonary metastasis and develop a predictive machine-learning model.
Methods
A retrospective analysis was conducted on patients diagnosed with lung cancer between January 2015 and December 2018 and followed up until December 2024. CT-defined emphysema was quantified using the relative lung area with attenuation ≤ − 950 Hounsfield Units (LAA%), with severity stratified as follows: ≤6% (none), > 6% and ≤ 9% (mild), and > 9% (moderate-severe). Distant metastasis-free survival (DMFS) was analyzed using Kaplan-Meier and log-rank tests. Least absolute shrinkage and selection operator (LASSO) regression identified significant predictors for extrapulmonary metastasis. An eXtreme Gradient Boosting (XGBoost) model incorporating these features was developed and internally validated via stratified 10-fold cross-validation. Model performance was assessed using the area under the receiver operating characteristic curve (AUC) and decision curve analysis (DCA). SHAP analysis and a nomogram were performed based on the selected variables.
Results
A total of 542 patients were included in the study. Lung cancer patients with or without CT-defined emphysema showed significant differences in multiple clinical and pathological characteristics (sex, age, histological subtype, smoking status, BMI, etc.). Patients with moderate-severe CT-defined emphysema (LAA% >9%) exhibited worse DMFS than those with mild or no emphysema. XGBoost model showed better metastasis predictive efficiency than other machine learning models, especially when concurrent emphysema was considered (AUC = 0.820, NRI = 0.26, P < 0.001). SHAP analysis revealed the top three contributions for predicting distant metastasis in the model were tumor size, CT-defined emphysema, and smoking status.
Conclusion
CT-defined emphysema severity predicts worse DMFS and improves extrapulmonary metastasis risk in lung cancer. The developed interpretable XGBoost model and nomogram provide clinically valuable tools for personalized prognosis assessment in lung cancer patients with underlying emphysema.
Supplementary Information
The online version contains supplementary material available at 10.1186/s12880-026-02229-y.
Keywords: Lung cancer, Emphysema, Prognosis, Machine learning, Nomogram
Introduction
Globally, lung cancer is the primary cause of death related to malignancies, accounting for nearly 1.8 million fatalities in 2020 [1]. Metastatic dissemination occurs frequently in lung cancer, with nearly 40% of patients developing metastases in extrapulmonary sites over extended follow-up periods [2–4]. Our previous study further demonstrated that in lung cancer patients initially free of metastases, the cumulative probabilities of developing extrapulmonary metastases at 1, 3, and 5 years were 8.8%, 28.3%, and 42.7%, respectively [5]. Patients who develop extrapulmonary dissemination have a median survival of only 6 months, and their chance of surviving for 5 years is less than 1% [4, 6]. Imaging-based early identification of recurrent or metastatic disease could enhance survival outcomes by facilitating the prompt initiation of curative therapies [7]. However, research on risk factors predictive of extrapulmonary metastasis in lung cancer patients remains limited and underexplored, warranting further investigation.
Chronic obstructive pulmonary disease (COPD), which affects 10.3% of individuals aged 30 to 79 globally and is on the rise, is also frequently observed as a comorbidity in lung cancer patients [8–10]. Research indicates that individuals with COPD have a 2- to 6-fold increased risk of developing lung cancer compared to the general population, with COPD serving as an independent risk factor for lung cancer [11–13]. Furthermore, the severity of COPD is positively associated with increased incidence and mortality rates of lung cancer [14]. COPD can lead to systemic inflammatory changes due to localized chronic inflammation in the lungs, which potentially enhancing the likelihood of metastasis [15]. Additionally, lung cancer patients with COPD exhibit significantly lower 2-year and 3-year condition-specific survival rates compared to those without COPD, which correlates with an elevated risk of recurrence [16], indicating that early detection and treatment strategies for lung cancer patients with COPD urgently need improvement.
A quantitative evaluation of pulmonary emphysema using computed tomography (CT) imaging can precisely ascertain the extent of lung structural damage and classify subtypes of pulmonary emphysema, thereby eliminating the need for manual annotation [17]. Computer-assisted methodologies are capable of identifying all voxels with densities below a specified threshold, subsequently quantifying the volume of emphysema by multiplying the voxel count by the known voxel volume. While recent artificial intelligence approaches have demonstrated the potential of quantitative CT features in predicting aggressive tumor behaviors [18], more interpretable and clinically feasible imaging phenotypes have not been adequately evaluated for their association with extrapulmonary metastasis. Several studies have indicated that CT-defined emphysema serves as a predictor for the risk of lung cancer development [19, 20]. Nonetheless, the influence of coexisting CT-defined emphysema on the distant metastasis-free survival (DMFS) of lung cancer patients remains uncertain. The role of CT-defined emphysema in forecasting extrapulmonary metastases subsequent to lung cancer treatment is not well elucidated. Furthermore, while prognostic modeling tailored to lung cancer patients with concurrent emphysema is clinically significant, research in this domain is sparse.
Consequently, the objective of this longitudinal study was to assess the prognostic significance of CT-defined emphysema in predicting the development of extrapulmonary metastasis in lung cancer patients during follow-up. Additionally, the study examined the influence of patient and tumor characteristics on the incidence of extrapulmonary metastases in lung cancer. Furthermore, an optimal and interpretable machine-learning model was developed based on the significant characteristics to predict the risk of extrapulmonary metastases post-treatment in lung cancer patients, with consideration of emphysema.
Methods
Patients
This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. Ethical approval for this retrospective analysis was granted by the Medical Ethics Committee of Shanghai Chest Hospital (No. KS1956), and the requirement for informed consent was waived. This retrospective study enrolled patients with a pathological diagnosis of lung cancer who were admitted to the Shanghai Chest Hospital between January 2015 and December 2018. The inclusion criteria were as follows: (1) histopathological confirmation of primary lung cancer via biopsy or surgical specimen; (2) patients were treated for lung cancer at our hospital; (3) no distant metastases were found upon systemic examination at the initial visit; (4) underwent at least one follow-up visit at our institution ≥ 1 month after treatment to evaluate metastatic status. The exclusion criteria were as follows: (1) had a history of other malignancy; (2) incomplete medical records or survival data; (3) received prior anticancer treatment for lung cancer elsewhere; (4) suboptimal image quality (Fig. 1). The baseline characteristics of patients and the follow-up information were recorded by reviewing medical records. DMFS was defined as the time from treatment to the date of lung cancer-specific distant metastasis.
Fig. 1.
Workflow of the study
Data collection
The factors measured at baseline and used in the analysis were as follows: sex, age, histological subtype (adenocarcinoma, squamous, large cell, small cell and others), smoking status, smoking index (pack-year), BMI, T stage of the 8th International Association for the Study of Lung Cancer (IASLC) TNM staging system, pulmonary lymph node metastasis, size of the largest tumor (≤ 3 cm, > 3 cm and ≤ 5 cm, and > 5 cm), EGFR mutation, ALK mutation, ROS1 mutation, carcinoembryonic antigen ([CEA], nanogram per milliliter), serum keratin 19 ([CYFRA21], nanogram per milliliter), squamous cell carcinoma antigen ([SCC], nanogram per milliliter), CT texture (pure solid, ground glass nodules, and mixed), initial CT-reported intrapulmonary metastases, miliary nodules, atelectasis, pleural effusion, pleural nodules, CT-defined emphysema subtype (centrilobular, paraseptal or panlobular), emphysema ratio, and initial treatment method (curative vs. palliative). In the initial treatment method, curative treatments included pulmonary resection (lobectomy, pneumonectomy or segmentectomy); palliative treatment included chemotherapy, radiation therapy, and systemic therapy. A subset of patients underwent spirometry at diagnosis, and forced expiratory volume in 1 s per forced vital capacity (FEV1/FVC) ≤ 70% was defined as airway obstruction [21].
Image acquisition
Chest CT was obtained with one of the following CT scanners: Discovery CT750, Revolution, Revolution-Apex (GE Healthcare, Milwaukee, USA); Ingenuity Core, Brilliance 256 iCT (Philips Healthcare, Amsterdam, Netherlands); SOMATOM Force (Siemens Healthcare, Erlangen, Germany); UCT-160 (United Imaging Healthcare, Shanghai, China). The scanning parameters were as follows: tube voltage, 120 kV; tube current, 220–300 mA; layer thickness, 5 mm; field of view, 350–400 mm. The scanning range was from the thoracic inlet to the costophrenic angles.
Image evaluation and outcome assessment
A radiologist (Z.Y.J., with 4 years of experience) independently reviewed the medical records, images of CT examinations, and standardized radiologic reports at the Picture Archiving and Communication System (PACS) workstation. CT-defined emphysema and the metastatic lesions were confirmed by two radiologists (Z.L. and Y.H., with 10 and 30 years of experience, respectively). The survival time was defined as the interval between the date of initial diagnosis and the date of events or the last follow-up visit before July 1, 2024. The severity of CT-defined emphysema was quantified on the baseline chest CT scans, and the Hounsfield Unit (HU) threshold for low attenuation area (LAA) in quantitative assessments was defined using − 950 HU [22]. The presence of CT-defined emphysema was defined using the 6% threshold of the relative area of the lung with density values below − 950 HU normalized to the lung volume at acquisition [23]. The patients were dichotomized into those with CT-defined emphysema (LAA% >6%) or without CT-defined emphysema (LAA% ≤6%) group. Further, the subset of patients with CT-defined emphysema was divided into those with mild (LAA% ≤9%) and those with moderate–severe (LAA% >9%) emphysema [24]. Extrapulmonary metastasis was confirmed based on pathology or other clinical evidence. Extrapulmonary metastasis was defined as metastasis occurring throughout the body. Image quantitative assessments were performed on uAI Discover software (United Imaging Intelligence Co., Ltd., Shanghai, China).
Statistical analysis
Comparison of continuous variables between groups was performed by an unpaired t test. Pearson’s χ2 test was used to compare categorical variables. Kaplan-Meier analysis was employed to estimate cumulative extrapulmonary metastasis rates, and group comparisons were conducted using Log-rank tests (between groups of categorical variables). Median follow-up duration was calculated by the reverse Kaplan-Meier estimation method.
The least absolute shrinkage and selection operator (LASSO) regression algorithm was used to select the significant features. Variance inflation factor (VIF) analysis was used to analyze the collinearity of variables, and all significant variables were included owing to no collinearity existing (VIF ≤ 10). Then, the variables screened by LASSO regression were used for model construction. For predicting extrapulmonary metastases, five types of machine learning algorithms were performed to model our data: eXtreme Gradient Boosting (XGBoost), k-Nearest Neighbor (KNN), Random Forest (RF), Support Vector Machine (SVM), and traditional Logistic Regression (LR). Hyperparameter tuning was performed using grid search combined with 10-fold cross-validation to ensure optimal performance, and the final optimized hyperparameters for each model are summarized in Supplementary Table 2. We conducted two sets of experiments. In the first model, all variables selected by the LASSO algorithm were pooled into the model to develop a preliminary diagnostic model. In the second model, the selected variables, excluding those related with CT-defined emphysema, were pooled into the model to develop a diagnostic model in order to investigate the value of CT-defined emphysema. To address the limitation of the small dataset and obtain an unbiased estimate of the model’s generalization performance, an internal stratified 10-fold cross-validation was performed [25]. The models were evaluated using the area under the curve (AUC) of the receiver operating characteristic (ROC) curves, accuracy, sensitivity, specificity, F1-score, precision and recall. Net reclassification index (NRI) were calculated to compare the performance between models. Decision curve analysis (DCA) and calibration curves to demonstrate true clinical utility. To interpret model decisions, the SHAP method generated a bar chart illustrating the contribution of each feature to the prediction outcomes, and SHAP values evaluated the impact of specific features on individual sample predictions. The selected features were then incorporated into a nomogram. A web-based calculator of the nomogram is available in the Supplementary File to help researchers and clinicians easily determine personalized risk predictions for extrapulmonary metastasis.
Machine learning models were conducted using uAI Discover software (United Imaging Intelligence Co., Ltd., Shanghai, China). SHAP analyses were performed using packages implemented in Python Software (version 3.6, Python Software Foundation, Wilmington, DE, USA). For statistical analysis, software SPSS 24.0 (IBM) was used, and R Studio (version 1.4.1717, RStudio Team) was used for generation of graphs. P < 0.05 was considered to indicate a statistically significant difference.
Results
Patient characteristics
A total of 542 patients (mean age, 61 years ± 8 [SD]; 377 men) met inclusion criteria for this study. The baseline characteristics of patients and pulmonary tumors are summarized in Tables 1 and 2. The median follow-up period between initial diagnosis and extrapulmonary metastasis was 3.8 years (95% confidence interval [CI]: 3.6, 4.1).
Table 1.
Comparison of clinicopathologic characteristics of patients
| Characteristic | All patients (n = 542) | Patients without CT-defined emphysema (LC) (n = 318) |
Patients with CT-defined emphysema (n = 224) |
P Value § | |||
|---|---|---|---|---|---|---|---|
| All patients with CT-defined emphysema (n = 224) |
Without confirmed COPD (n = 155) | With confirmed COPD (n = 69) | LC vs. CT-defined emphysema | LC vs. COPD | |||
| Sex | < 0.001 | < 0.001 | |||||
| Male | 377(69.6) | 163(51.3) | 214(95.5) | 147(94.8) | 67(97.1) | ||
| Female | 165(30.4) | 155(48.7) | 10(4.5) | 8(5.2) | 2(2.9) | ||
| Age (years) | 61 ± 8(22–83) | 59 ± 9(22–78) | 63 ± 7(41–83) | 63 ± 7(41–83) | 64 ± 6(49–75) | < 0.001 | < 0.001 |
| <60 | 215(39.7) | 159(50.0) | 56(25.0) | 43(27.7) | 13(18.8) | ||
| ⩾60 | 327(60.3) | 159(50.0) | 168(75.0) | 112(72.3) | 56(81.2) | ||
| Histological subtype | < 0.001 | < 0.001 | |||||
| Adenocarcinoma | 381(70.3) | 256(80.5) | 125(55.8) | 90(58.1) | 35(50.7) | ||
| Squamous | 105(19.4) | 42(13.2) | 63(28.1) | 43(27.7) | 20(29.0) | ||
| Large cell | 11(2.0) | 4(1.3) | 7(3.1) | 5(3.2) | 2(2.9) | ||
| Small cell | 34(6.3) | 12(3.8) | 22(9.8) | 12(7.7) | 10(14.5) | ||
| Others | 11(2.0) | 4(1.3) | 7(3.1) | 5(3.2) | 2(2.9) | ||
| Smoking status | < 0.001 | < 0.001 | |||||
| Former or current smoker | 234(43.2) | 98(30.8) | 136(60.7) | 96(61.9) | 40(58.0) | ||
| Never smoker | 308(56.8) | 220(69.2) | 88(39.3) | 59(38.1) | 29(42.0) | ||
| Smoking Index (pack-year) * | 53 ± 76(4-730) | 62 ± 110(4-730) | 46 ± 35(5-350) | 41 ± 22(5-150) | 58 ± 55(10–350) | 0.09 | 0.013 |
| <30 | 340(62.7) | 240(75.5) | 100(44.6) | 67(43.2) | 33(47.8) | ||
| ⩾30 | 202(37.3) | 78(24.5) | 124(55.4) | 88(56.8) | 36(52.2) | ||
| BMI (kg/m2) | 23 ± 3(16–33) | 23.7 ± 3.2(17–33) | 22.6 ± 2.8(16–30) | 22.4 ± 2.7(16–28) | 23.0 ± 2.9(17–30) | < 0.001 | 0.13 |
| <25 | 416(76.8) | 221(69.5) | 195(87.1) | 138(89.0) | 57(82.6) | ||
| ⩾25 | 126(23.2) | 97(30.5) | 29(12.9) | 17(11.0) | 12(17.4) | ||
| T stage | 0.07 | 0.59 | |||||
| I-II | 339(62.5) | 209(65.7) | 130(58.0) | 87(56.1) | 43(62.3) | ||
| III-IV | 203(27.5) | 109(34.3) | 94(42.0) | 68(43.9) | 26(37.7) | ||
| Pulmonary Lymph Node Metastasis † | 377(69.6) | 216(67.9) | 161(71.9) | 114(73.5) | 47(68.1) | 0.33 | 0.98 |
| Initial treatment | < 0.001 | 0.19 | |||||
| Lobectomy or bilobectomy | 295(54.4) | 193(60.7) | 102(45.5) | 68(43.9) | 34(49.3) | ||
| Segmentectomy | 11(2.0) | 9(2.8) | 2(0.9) | 1(0.6) | 1(1.4) | ||
| Pneumonectomy | 9(1.7) | 6(1.9) | 3(1.3) | 0 | 3(4.3) | ||
| Ablation | 2(0.4) | 0 | 2(0.9) | 2(1.3) | 0 | ||
| Palliative treatment | 225(41.5) | 110(34.6) | 115(51.3) | 84(54.2) | 31(44.9) | ||
| Molecular pathology | |||||||
| EGFR mutation | 233(43.0) | 173(54.4) | 60(26.8) | 38(24.5) | 22(31.9) | < 0.001 | 0.001 |
| ALK mutation | 46(8.5) | 32(10.1) | 14(6.3) | 8(5.2) | 6(8.7) | 0.12 | 0.73 |
| ROS1 mutation | 44(8.1) | 30(9.4) | 14(6.3) | 9(5.8) | 5(7.2) | 0.18 | 0.57 |
| Blood tests | |||||||
| CEA (ng/mL) | 14.5 ± 36.8 | 14.8 ± 41.8 | 14.1 ± 27.9 | 15.4 ± 30.4 | 11.0 ± 20.7 | 0.003 | 0.08 |
| <5 | 304(56.1) | 192(60.4) | 112(50.0) | 79(51.0) | 33(47.8) | ||
| ⩾5 | 238(43.9) | 126(39.6) | 112(50.0) | 76(49.0 | 36(52.2) | ||
| CYFRA21 (ng/mL) | 5.6 ± 11.9 | 4.6 ± 11.4 | 7.1 ± 12.3 | 7.6 ± 13.6 | 5.8 ± 8.4 | < 0.001 | 0.008 |
| <5 | 393(72.5) | 256(80.5) | 137(61.2) | 89(57.4) | 48(69.6) | ||
| ⩾5 | 149(27.5) | 62(19.5) | 87(38.8) | 66(42.6) | 21(30.4) | ||
| SCC (ng/mL) | 1.6 ± 4.2 | 1.2 ± 2.2 | 2.1 ± 6.0 | 2.4 ± 7.0 | 1.5 ± 2.4 | < 0.001 | 0.002 |
| <1.5 | 434(80.1) | 271(85.2) | 163(72.8) | 115(74.2) | 48(69.6) | ||
| ⩾1.5 | 108(19.9) | 47(14.8) | 61(27.2) | 40(25.8) | 21(30.4) | ||
Note. — Data in parentheses are N (percentage) or Mean ± SD (range). *For patients who smoke or have a smoking history only. †The International Association for the Study of Lung Cancer (IASLC) pulmonary lymph node stations for lung cancer. §Between the lung cancer patients without CT-defined emphysema and the patients with CT-defined emphysema but without confirmed COPD or with confirmed COPD. LC = lung cancer patients without CT-defined emphysema, CEA = carcinoembryonic antigen, CYFRA21 = serum keratin 19, SCC = squamous cell carcinoma antigen
Table 2.
Comparison of CT imaging characteristics of patients
| Characteristic | All patients (n = 542) | Patients without CT-defined emphysema (LC) (n = 318) |
Patients with CT-defined emphysema (n = 224) |
P Value § | |||
|---|---|---|---|---|---|---|---|
| All patients with CT-defined emphysema (n = 224) |
Without confirmed COPD (n = 155) | With confirmed COPD (n = 69) | LC vs. CT-defined emphysema | LC vs. COPD | |||
| CT texture classification | 0.086 | 0.14 | |||||
| Pure solid | 475(87.6) | 272(85.5) | 203(90.6) | 138(89.0) | 65(94.2) | ||
| Ground glass nodules (GGO) | 17(3.1) | 14(4.4) | 3(1.3) | 3(1.9) | 0 | ||
| Mixed | 50(9.2) | 32(10.1) | 18(8.0) | 14(9.0) | 4(5.8) | ||
| Size of the largest tumor | < 0.001 | 0.012 | |||||
| ≤3 cm | 294(54.2) | 184(57.9) | 110(49.1) | 67(43.2) | 43(62.3) | ||
| >3 cm to ≤ 5 cm | 158(29.2) | 106(33.3) | 52(23.2) | 42(27.1) | 10(14.5) | ||
| >5 cm | 90(16.6) | 28(8.8) | 62(27.7) | 46(29.7) | 16(23.2) | ||
| CT-reported intrapulmonary metastases | 131(24.2) | 80(25.2) | 51(22.8) | 36(23.2) | 15(21.7) | 0.51 | 0.61 |
| Miliary nodules | 63(11.6) | 24(7.5) | 39(17.4) | 26(16.8) | 13(18.8) | < 0.001 | 0.046 |
| Atelectasis | 40(7.4) | 20(6.3) | 20(8.9) | 15(9.7) | 5(7.2) | 0.25 | 0.96 |
| Pleural effusion | 111(20.5) | 69(21.7) | 42(18.8) | 32(20.6) | 10(14.5) | 0.39 | 0.19 |
| Pleural Nodules | 42(7.7) | 25(7.9) | 17(7.6) | 14(9.0) | 3(4.3) | 0.9 | 0.26 |
| CT-defined emphysema | 224(41.3) | - | 224(100.0) | 155(100.0) | 69(100.0) | < 0.001 | < 0.001 |
| CT-defined emphysema subtype* | - | - | |||||
| Centrilobular | 103(46.0) | - | 103(46.0) | 73(47.1) | 30(43.5) | ||
| Paraseptal | 46(20.5) | - | 46(20.5) | 38(24.5) | 8(11.6) | ||
| Panlobular | 75(33.5) | - | 75(33.5) | 44(28.4) | 31(44.9) | ||
| LAA%* | - | - | 8.6 ± 3.9(6.1–28.1) | 8.6 ± 4.1(6.1–28.1) | 8.4 ± 3.4(6.1–20.2) | - | - |
| Emphysema Ratio | - | - | |||||
| No (≤ 6%) | 318(58.7) | 318 | - | - | - | ||
| Mild (> 6% to ≤ 9%) | 167(30.8) | - | 167(74.6) | 116(74.8) | 51(73.9) | ||
| Moderate/severe (> 9%) | 57(10.5) | - | 57(25.4) | 39(25.2) | 18(26.1) | ||
Note. — Data in parentheses are N (percentage) or Mean ± SD (range). *For patients who were diagnosed with emphysema on CT imaging only. CT-defined emphysema was the lung relative area below 950 Hounsfield units > 6%. §Between the lung cancer patients without CT-defined emphysema and the patients with CT-defined emphysema but without confirmed COPD or with confirmed COPD. LC = lung cancer patients without CT-defined emphysema, LAA = low attenuation area
During the follow-up period, the 1-, 3-, and 5-year cumulative incidences of extrapulmonary metastasis were 10.3% (95% CI: 7.8, 12.9), 26.8% (95%CI: 23.0, 30.5), and 37.6% (95% CI: 33.5, 41.7), respectively (Fig. 2). Extrapulmonary metastases were identified in 220 patients during follow-up. The most common sites included bone (n = 88), liver (n = 75), and brain (n = 52), followed by adrenal glands (n = 41), peritoneal lymph node (n = 14), kidney (n = 10), pancreas (n = 5), peritoneum (n = 4), spleen (n = 4), esophagus (n = 1), breast (n = 1), stomach (n = 1).
Fig. 2.
Kaplan-Meier curves show the cumulative extrapulmonary metastasis rate
In the current study, 224 of 542 (41.3%) patients had CT-defined emphysema at diagnosis (mean LAA%, 8.6 ± 3.9 [SD]), of which, 167 (30.8%) were mild (LAA% ≤9%) and 57 (10.5%) were moderate-severe (LAA% >9%). Centrilobular emphysema was the predominant type in the cohort (46.0%, 103 out of 224). In the group with CT-defined emphysema, 30.8% (69 out of 224) of patients were confirmed with COPD by spirometry. Comparative analysis revealed significant differences in sex, age, histological subtype, smoking status, BMI, initial treatment, EGFR mutation status, CEA, CYFRA21, and SCC between lung cancer patients with and without CT-defined emphysema. Additionally, smoking index was significantly different between lung cancer patients with or without COPD, while CEA showed no statistically significant difference (Table 2).
Association between CT-defined emphysema and DMFS
The stratified survival analysis by severity of CT-defined emphysema showed that patients with moderate-severe CT-defined emphysema (LAA% >9%) had significantly worse DMFS compared to those without it (LAA% ≤6%) or the mild group (6% < LAA% ≤9%) (log-rank test, P < 0.001; Fig. 3A). There was no significant difference between the patients without CT-defined emphysema (LAA% ≤6%) and the mild group (6% < LAA% ≤9%) (log-rank test, P = 0.167).
Fig. 3.
Kaplan–Meier curves of (A) overall distant metastasis-free survival (DMFS) and (B-H) stratified DMFS among the age, subtype, treatment, CEA, CYFRA21, SCC, and size of tumors groups with different CT-defined emphysema status differed significantly (Log-rank P < 0.001)
Furthermore, we assessed DMFS differences among lung cancer patients after adjustment for the aforementioned key factors (Fig. 3B-H and Supplementary Fig. 1). Patients with CT-defined emphysema exhibited a significantly higher risk of extrapulmonary metastasis compared to those without CT-defined emphysema across all histological subtypes (HR for adenocarcinoma with CTE vs. adenocarcinoma without CTE, 1.5; [95% CI: 1.1–2.1; P = 0.019]; HR for other subtypes with CTE vs. other subtypes without CTE, 2.0; [95% CI: 1.1–3.4; P = 0.017]). When stratified by age, the elevated metastasis risk associated with CT-defined emphysema persisted in patients with age < 60 years (HR 1.7; 95% CI: 1.0–2.7; P = 0.045). Conversely, no significant difference in metastasis risk was observed between CT-defined emphysema-positive and -negative subgroups among patients with age ≥ 60 years (Supplementary Table 1). Additionally, we found that patients without CT-defined emphysema exhibited a significantly higher risk of extrapulmonary metastasis when ≥ 60 years old, non-Adenocarcinoma, initially palliative treatment, CEA ≥ 5 ng/mL, CYFRA21 ≥ 5 ng/mL, SCC ≥ 1.5 ng/mL or largest tumor≥3 cm, while only largest tumor size was related with DMFS in patients with CT-defined emphysema (HR = 2.4, P < 0.001). The detailed values are shown in Supplementary Table 1.
Screening of characteristic variables and construction of models
In order to screen variables with high contribution ratios, all characteristics were included in LASSO regression analysis to screen extrapulmonary metastases relevant features and the ten-fold cross-validation method was adopted for iterative analysis. The variables finally included were smoking status(current or former smoker vs. never smoke), smoking index (<30 pack-year vs. ≥30 pack-year), initial treatment (curative vs. palliative), CEA (<5 ng/mL vs. ≥5 ng/mL), size of tumor (<3 cm vs. ≥3 cm), CT-defined emphysema (LAA% ≤6% vs. LAA% >6%)and moderate-severe CT-defined emphysema (LAA% ≤9% vs. LAA% >9%). All of the above variables have no collinearity (VIF ≤ 10).
To identify the optimal algorithm for predicting extrapulmonary metastasis in lung cancer patients, five machine learning models were developed and evaluated. The XGBoost model demonstrated superior discriminative performance and was consequently selected for subsequent analyses (Supplementary Table 3 and Supplementary Fig. 2). Based on the AUC, the XGBoost model with CT-defined emphysema outperformed the model without CT-defined emphysema. A detailed description of the two models’ performance can be found in Table 3 and Fig. 4. The XGBoost model with CT-defined emphysema reached an AUC of 0.820 during the training phase and 0.748 during the testing phase. The NRI was 0.26 between models with or without CT-defined emphysema indicator (P < 0.001), revealing that the integration of CT-defined emphysema into the XGBoost model performed satisfactorily, indicating CT-defined emphysema may improve classification accuracy for extrapulmonary metastases prediction. The DCA curves confirmed the superior clinical utility of the XGBoost model with CT-defined emphysema (Fig. 4B). The calibration curves of the two models were shown in Fig. 4C, demonstrating good agreement between the predicted probabilities of models and observed outcomes.
Table 3.
Performance of machine learning models in predicting the extrapulmonary metastasis of patients
| Model | AUC (95% CI) | Sensitivity | Specificity | Accuracy | Precision | F1 Score | |
|---|---|---|---|---|---|---|---|
| XGBoost with CT-defined Emphysema | Train | 0.820(0.784–0.858) | 0.682 | 0.803 | 0.754 | 0.703 | 0.692 |
| Test | 0.748(0.624–0.892) | 0.773 | 0.781 | 0.778 | 0.708 | 0.739 | |
| XGBoost without CT-defined Emphysema | Train | 0.734(0.694–0.782) | 0.641 | 0.714 | 0.684 | 0.605 | 0.623 |
| Test | 0.678(0.558–0.84) | 0.773 | 0.531 | 0.63 | 0.531 | 0.63 |
Fig. 4.
The ROC curves of the XGBoost machine learning model with CT-defined emphysema and without CT-defined emphysema in (A) training set and (B) validation set. (C) DCA curves was performed to evaluate the clinical utility of the models. (D) Calibration curves of the models for prediction of the likelihood of extrapulmonary metastasis
SHAP
The comprehensive population picture of SHAP analysis in patients developing extrapulmonary metastases during follow-up illustrates the variables in the XGBoost model (Fig. 5A, B). The results showed that the contribution of each variable, in descending order, was: size of tumor, CT-defined emphysema, smoke status, CEA, initial treatment, moderate/severe CT-defined emphysema, and smoke index. A detailed case study showing the model’s predictive course for a specific patient (Fig. 5C). To further demonstrate how variables shift predictions in specific patients, SHAP waterfall plots are presented for a low-risk and a high-risk case (Fig. 5D, E). The red indicator indicates a positive contribution to the prediction, while the blue indicator indicates a negative impact. The f(x) value represents the actual SHAP value for each factor.
Fig. 5.
(A, B) Presents a comprehensive population map of extrapulmonary metastasis in lung cancer patients, illustrating the variables in the XGBoost model. (C) A detailed case study showing the model’s predictive course for a specific patient. (D, E) SHAP waterfall plots illustrating the stepwise contribution of each variable to the predicted probability of extrapulmonary metastasis in two representative patients. Red features increase risk; blue features decrease risk
Nomogram development
Based on the LASSO regression analysis and XGBoost algorithm, the 7 variables were included in the prediction model. We established an individualized prediction nomogram model for extrapulmonary metastasis after treatment in lung cancer patients. Based on the nomogram, the points obtained corresponding to each prediction indicator, the sum of the points was recorded as the total score, and the “diagnostic possibility” corresponding to the total score was the extrapulmonary metastasis probability of the patient (Fig. 6).
Fig. 6.
Nomogram to predict the extrapulmonary metastasis probability of lung cancer patients
Discussion
To the best of our knowledge, this study represents the first retrospective analysis to explore the potential impact of CT-defined emphysema on the prognosis of extrapulmonary metastasis in lung cancer patients during post-treatment follow-up. Our findings indicate that lung cancer patients with CT-defined emphysema exhibit poorer DMFS and present differences in age, subtype, treatment, CEA, CYFRA21, SCC, and tumor size. Furthermore, this study developed an interpretable model based on XGBoost to predict the risk of extrapulmonary metastasis in lung cancer patients.
Previous studies had demonstrated that emphysema was associated with reduced overall survival and progression-free survival in lung cancer [26–28], and CT-defined emphysema serves as an independent prognostic factor for non-small cell lung cancer [19]. However, the role of CT-defined emphysema in the prognosis of metastases in lung cancer remains unclear. The results of our study revealed a significantly lower DMFS in lung cancer patients with moderate-severe CT-defined emphysema, suggesting that emphysema increase the risk of distant metastasis in both adenocarcinoma and non-adenocarcinoma subtypes. This finding aligns with previous reports by Wu, et al. [29], which indicated that comorbid COPD is associated with poorer prognosis and reduced survival in non-small cell lung cancer patients. Earlier studies have reported associations between coexisting emphysema or COPD and clinical variables in lung cancer patients, including age, histological subtype, treatment methods, and CEA levels [26–28, 30, 31], which is consistent with our findings of significant difference in these factors between lung cancer patients with emphysema and those without. In stratified analyses, we observed that among lung cancer patients without emphysema, age ≥ 60 years, non-adenocarcinoma histology, initial palliative treatment, CEA ≥ 5 ng/mL, CYFRA 21 ≥ 5 ng/mL, SCC ≥ 1.5 ng/mL, and size of the largest tumor ≥ 3 cm were significantly associated with an increased risk of extrapulmonary metastasis. However, among patients with CT-defined emphysema, only size of the largest tumor remained significantly correlated with extrapulmonary metastasis. This observation aligns with Jeong, et al. [32], who reported CT-measured tumor size as a predictor of reduced disease-free survival in lung cancer patients. Notably, in lung cancer patients with CT-defined emphysema, only tumor size influenced distant metastasis. We hypothesize that the presence of emphysema may attenuate the effects of other risk factors, possibly through complex interactions, warranting further in-depth investigation.
According to the National Comprehensive Cancer Network (NCCN) guidelines for Non-Small Cell Lung Cancer, early detection of metastatic disease enables timely clinical intervention, thereby improving survival outcomes [33]. In this study, we investigated the potential for curative-intent therapy in carefully selected patients with limited local recurrences, who may benefit from curative-intent therapy, and achieve long-term survival. We utilized LASSO regression to identify predictors of extrapulmonary metastasis in patients with post-treatment lung cancer. The selected features included variables such as smoking status, smoking index, initial treatment, CEA, tumor size, and the presence of CT-defined emphysema, including moderate to severe cases. Our findings align with previous research, indicating that moderate or severe CT-defined emphysema is associated with a poor prognosis in lung cancer [19, 34]. Furthermore, the NRI results demonstrated that the XGBoost model, which incorporated CT-defined emphysema characteristics, outperformed other models, suggesting that CT-defined emphysema is beneficial for predicting extrapulmonary metastasis in lung cancer. While recent large-scale studies have established promising clinicopathological nomograms for lung cancer prognosis [35, 36], machine learning approaches offer enhanced flexibility in integrating heterogeneous data types. The XGBoost algorithm, a tree-based machine learning model, is renowned for its superior performance in various recent machine learning challenges [37]. Mascalchi, et al. [24] incorporated baseline CT-derived measures of emphysema severity, coronary artery calcification scores, age, sex, and smoking history into an XGBoost machine learning model, achieving moderate predictive efficacy for all-cause mortality among smokers with emphysema, as indicated by an AUC of 0.71. However, the model demonstrated limited accuracy in predicting lung cancer-specific mortality, with an AUC of 0.61. Ding, et al. [38] developed machine learning models using XGBoost and RF algorithms to predict the risk of distant metastasis in patients with testicular germ cell tumors, achieving AUC values ranging from 0.814 to 0.816 in the training cohort. In comparison, our model integrates CT-derived tumor size and tumor-specific markers, resulting in superior discriminative performance, with an AUC of 0.82 for predicting lung cancer extrapulmonary metastasis.
This study is subject to several limitations. Firstly, as it was a retrospective study conducted at a single institution, there is a potential for selection bias due to non-random patient enrollment and institutional practice patterns. Secondly, the size of the COPD subgroup was relatively small due to some patients being unable to complete the spirometry test, which limited our ability to fully assess the impact of COPD severity on metastatic risk. Thirdly, only internal validation was conducted, which may restrict the generalizability of the model results. External validation in independent, multicenter cohorts is needed to confirm generalizability. Furthermore, while XGBoost performed well, incorporating alternative or ensemble algorithms may improve predictive accuracy further.
Conclusion
In conclusion, CT-defined emphysema is independently associated with worse DMFS in lung cancer patients. The interpretable machine learning model developed using baseline clinical and CT characteristics showed robust performance in predicting post-treatment extrapulmonary metastasis in lung cancer patients. These findings highlight the prognostic relevance of coexisting emphysema and suggest that its integration into risk stratification may help identify high-risk patients earlier, potentially supporting more personalized management strategies.
Supplementary Information
Below is the link to the electronic supplementary material.
Acknowledgements
Not applicable.
Author contributions
Y.Z analyzed and interpreted the patient data and drafted the work. Q.C contributed to the analysis of data. H.Z contributed to the acquisition of data. K.N contributed to the acquisition of data. L.Z contributed to the acquisition of data. L.Y interpreted the data. G.T contributed to the software used in the work. J.X contributed to the acquisition of data. S.L contributed to the acquisition of data. Y.S contributed to the acquisition of data. Q.N contributed to the acquisition of data. W.K contributed to the acquisition of data. H.Y substantively revised the work. L.Z contributed to the conception of the work and substantively revised the work. All authors read and approved the final manuscript.
Funding
This work was supported in part by grants from the National Natural Science Foundation of China (82071873), Young Scientists Fund of the National Natural Science Foundation of China (82302188), National Key R&D Program of China (2021YFC2500700), Shanghai Health Research Foundation for Talents (2022YQ060), Shanghai Science and Technology Innovation Action Plan (22Y11911100), and Shanghai Innovative Medical Product Application Demonstration Project (24SF1904000).
Data availability
The datasets generated and/or analysed during the current study are not publicly available due the privacy of patients but are available from the corresponding author on reasonable request.
Declarations
Ethics approval and consent to participate
This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The retrospective study was approved by the Shanghai Chest Hospital review board (No. KS1956) and the requirement for informed consent was waived.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Yunjing Zhu, Qunhui Chen and Huiyuan Zhu contributed equally as first authors.
Hong Yu and Lin Zhu contributed equally to this work as senior authors.
Contributor Information
Hong Yu, Email: yuhongphd@163.com.
Lin Zhu, Email: monica_zhul@163.com.
References
- 1.Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, 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(3):209–49. [DOI] [PubMed] [Google Scholar]
- 2.Yang J, Peng AM, Wang B, Gusdon AM, Sun XT, Jiang GN, et al. The prognostic impact of lymph node metastasis in patients with non-small cell lung cancer and distant organ metastasis. Clin Exp Metastas. 2019;36(5):457–66. [DOI] [PubMed] [Google Scholar]
- 3.Hirsch FR, Scagliotti GV, Mulshine JL, Kwon R, Curran WJ, Wu YL, et al. Lung cancer: current therapies and new targeted treatments. Lancet. 2017;389(10066):299–311. [DOI] [PubMed] [Google Scholar]
- 4.Schiller JH, Harrington D, Belani CP, Langer C, Sandler A, Krook J, et al. Comparison of four chemotherapy regimens for advanced non-small-cell lung cancer. New Engl J Med. 2002;346(2):92–8. [DOI] [PubMed] [Google Scholar]
- 5.Zhu Y, Liu J, Nie K, Chen Q, Yu L, Zhang Y, et al. Added value of abdominal CT after treatment of lung cancer. Quant Imaging Med Surg. 2025;15(8):6787–800. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Eberhardt WEE, Mitchell A, Crowley J, Kondo H, Kim YT, Turrisi A, et al. The IASLC lung cancer staging project. J Thorac Oncol. 2015;10(11):1515–22. [DOI] [PubMed] [Google Scholar]
- 7.Westeel V, Foucher P, Scherpereel A, Domas J, Girard P, Tredaniel J, et al. Chest CT scan plus x-ray versus chest x-ray for the follow-up of completely resected non-small-cell lung cancer (IFCT-0302) a multicentre, open-label, randomised, phase 3 trial. Lancet Oncol. 2022;23(9):1180–8. [DOI] [PubMed] [Google Scholar]
- 8.Agusti A, Vogelmeier CF, Halpin DMG. Tackling the global burden of lung disease through prevention and early diagnosis. Lancet Respiratory Med. 2022;10(11):1013–5. [DOI] [PubMed] [Google Scholar]
- 9.Venkatesan P. GOLD COPD report: 2023 update. Lancet Respir Med. 2023;11(1):18. [DOI] [PubMed] [Google Scholar]
- 10.Adeloye D, Song P, Zhu Y, Campbell H, Sheikh A, Rudan I, et al. Global, regional, and National prevalence of, and risk factors for, chronic obstructive pulmonary disease (COPD) in 2019: a systematic review and modelling analysis. Lancet Respir Med. 2022;10(5):447–58. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Carr LL, Jacobson S, Lynch DA, Foreman MG, Flenaugh EL, Hersh CP, et al. Features of COPD as predictors of lung cancer. Chest. 2018;153(6):1326–35. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Young RP, Hopkins RJ, Christmas T, Black PN, Metcalf P, Gamble GD. COPD prevalence is increased in lung cancer, independent of age, sex and smoking history. Eur Respir J. 2009;34(2):380–6. [DOI] [PubMed] [Google Scholar]
- 13.Mouronte-Roibas C, Leiro-Fernandez V, Fernandez-Villar A, Botana-Rial M, Ramos-Hernandez C, Ruano-Ravina A. COPD, emphysema and the onset of lung cancer. A systematic review. Cancer Lett. 2016;382(2):240–4. [DOI] [PubMed] [Google Scholar]
- 14.Young RP, Ward RC, Scott RJ, Gamble GD, Silvestri G. Airflow limitation and mortality during cancer screening in the National lung screening trial: why quantifying airflow limitation matters. Thorax. 2023;78(7):690–7. [DOI] [PubMed] [Google Scholar]
- 15.Mantovani A, Cancer. Inflaming Metastasis Nat. 2009;457(7225):36–7. [DOI] [PubMed] [Google Scholar]
- 16.Lopez-Encuentra A, Astudillo J, Cerezal J, Gonzalez-Aragoneses F, Novoa N, Sanchez-Palencia A, et al. Prognostic value of chronic obstructive pulmonary disease in 2994 cases of lung cancer. Eur J Cardiothorac Surg. 2005;27(1):8–13. [DOI] [PubMed] [Google Scholar]
- 17.Yang J, Angelini ED, Balte PP, Hoffman EA, Austin JHM, Smith BM, et al. Novel subtypes of pulmonary emphysema based on Spatially-Informed lung texture learning: the Multi-Ethnic study of atherosclerosis (MESA) COPD study. Ieee T Med Imaging. 2021;40(12):3652–62. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Zuo ZC, Fan XH, Tang Y, Zhang Y, Peng X, Zeng WH, et al. Deep learning-powered 3D segmentation derives factors associated with lymphovascular invasion and prognosis in clinical T1 stage non-small cell lung cancer. Heliyon. 2023;9(4). [DOI] [PMC free article] [PubMed]
- 19.Zhang Y, Yi J, Sun D, Su Y, Zuo Y, Zhu M, et al. The association of chest computed tomography-defined visual emphysema and prognosis in patients with nonsmall cell lung cancer. Erj Open Res. 2023;9(6). [DOI] [PMC free article] [PubMed]
- 20.Yang XF, Wisselink HJ, Vliegenthart R, Heuvelmans MA, Groen HJM, Vonder M, et al. Association between chest CT-defined emphysema and lung cancer: A systematic review and Meta-Analysis. Radiology. 2022;304(2):322–30. [DOI] [PubMed] [Google Scholar]
- 21.Agusti A, Celli BR, Criner GJ, Halpin D, Anzueto A, Barnes P, et al. Global initiative for chronic obstructive lung disease 2023 report: GOLD executive summary. Eur Respir J. 2023;61(4). [DOI] [PMC free article] [PubMed]
- 22.Labaki WW, Xia M, Murray S, Hatt CR, Al-Abcha A, Ferrera MC, et al. Quantitative emphysema on Low-Dose CT imaging of the chest and risk of lung cancer and airflow obstruction: an analysis of the National lung screening trial. Chest. 2021;159(5):1812–20. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Lynch DA, Austin JH, Hogg JC, Grenier PA, Kauczor HU, Bankier AA, et al. CT-Definable subtypes of chronic obstructive pulmonary disease: A statement of the Fleischner society. Radiology. 2015;277(1):192–205. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Mascalchi M, Romei C, Marzi C, Diciotti S, Picozzi G, Pistelli F, et al. Pulmonary emphysema and coronary artery calcifications at baseline LDCT and long-term mortality in smokers and former smokers of the ITALUNG screening trial. Eur Radiol. 2023;33(5):3115–23. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Varma S, Simon R. Bias in error Estimation when using cross-validation for model selection. BMC Bioinformatics. 2006;7. [DOI] [PMC free article] [PubMed]
- 26.Zhai RH, Yu XJ, Shafer A, Wain JC, Christiani DC. The impact of coexisting COPD on survival of patients with Early-Stage Non-small cell lung cancer undergoing surgical resection. Chest. 2014;145(2):346–53. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Wang W, Dou S, Dong WY, Xie MS, Cui LW, Zheng CY, et al. Impact of COPD on prognosis of lung cancer: from a perspective on disease heterogeneity. Int J Chronic Obstr. 2018;13:3767–76. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Ishida M, Mimae T, Kamigaichi A, Kawamoto N, Tsubokawa N, Miyata Y, et al. The role of emphysema on postoperative prognosis in Early-Stage nonsmall cell lung cancer. Ann Surg Oncol. 2024;31(8):5055–63. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Wu K, Wang J, Zhao LM, Wang P, Duan Q. The prognosis of non-small cell lung cancer combined with chronic obstructive pulmonary disease: A systematic review and meta-analysis. Clin Respiratory J. 2020;14(4):389–96. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Lee SJ, Lee J, Park YS, Lee CH, Lee SM, Yim JJ, et al. Impact of chronic obstructive pulmonary disease on the mortality of patients with Non-Small-Cell lung cancer. J Thorac Oncol. 2014;9(6):812–7. [DOI] [PubMed] [Google Scholar]
- 31.Jin LL, Zhou SL, Huang L, He YT, Zhang JY, Chen L, et al. Neutrophil-to-lymphocyte ratio as a prognostic marker for lung cancer in combined pulmonary fibrosis and emphysema patients. Eur J Med Res. 2025;30(1). [DOI] [PMC free article] [PubMed]
- 32.Jeong B, Lee SM, Park S, Choe J, Choi S, Do KH, et al. Prognostic performance in lung cancer according to tumor size: comparison of axial, multiplanar, and 3-dimensional CT measurement to pathological size. Eur J Radiol. 2021;144. [DOI] [PubMed]
- 33.Riely GJ, Wood DE, Ettinger DS, Aisner DL, Akerley W, Bauman JR, et al. Non-Small cell lung Cancer, version 4.2024, NCCN clinical practice guidelines in oncology. J Natl Compr Canc Netw. 2024;22(4):249–74. [DOI] [PubMed] [Google Scholar]
- 34.Shiraishi Y, Tanabe N, Shimizu K, Oguma A, Shima H, Sakamoto R, et al. Stronger associations of centrilobular than paraseptal emphysema with longitudinal changes in diffusing capacity and mortality in COPD. Chest. 2023;164(2):327–38. [DOI] [PubMed] [Google Scholar]
- 35.Zuo ZC, Zhang GC, Song P, Yang J, Li ST, Zhong Z, et al. Survival nomogram for stage IB Non-Small-Cell lung cancer Patients, based on the SEER database and an external validation cohort. Ann Surg Oncol. 2021;28(7):3941–50. [DOI] [PubMed] [Google Scholar]
- 36.Zuo ZC, Zhang GC, Lin SY, Xue Q, Qi WY, Zhang W, et al. Radiomics nomogram based on optimal volume of interest derived from High-Resolution CT for preoperative prediction of IASLC grading in clinical IA lung adenocarcinomas: A Multi-Center, Large-Population study. Technol Cancer Res T. 2024;23. [DOI] [PMC free article] [PubMed]
- 37.Ester M, Kriegel HP, Xu X, XGBoost. A scalable tree boosting system. In Proceedings of the 22Nd ACM SIGKDD International Conference on Knowledge Discovery and Data Miningvol, pg 785, (2016). Geogr Anal. 2022.
- 38.Ding L, Wang K, Zhang C, Zhang Y, Wang KLR, Li W, et al. A machine learning algorithm for predicting the risk of developing to M1b stage of patients with germ cell testicular cancer. Front Public Health. 2022;10. [DOI] [PMC free article] [PubMed]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The datasets generated and/or analysed during the current study are not publicly available due the privacy of patients but are available from the corresponding author on reasonable request.