Abstract
Background
Pulmonary mucinous adenocarcinoma (PMA) represents a rare lung adenocarcinoma subtype, characterized by a lack of comprehensive pathological classification and prognostic factors. In this study, we propose a multimodal machine learning (ML) framework that integrates multiple data modalities to enhance PMA subtype classification and prognostic prediction.
Methods
This retrospective study enrolled 175 primary PMA cases with details of demographic, histopathological, computed tomography (CT) imaging, and epithelial growth factor receptor (EGFR) mutation data. Least absolute shrinkage and selection operator (LASSO) regularized logistic regression model was utilized for histological classification, and Cox proportional hazards model was employed for survival prediction with internal validation.
Results
Pure mucinous adenocarcinoma presented a higher prevalence of smaller radiographic or pathological tumor size, lower lobe localization, lower rates of lymph node metastasis and spread through air spaces (STAS), early pathological stage, carcinoembryonic antigen (CEA) ≤5 ng/mL, spiculated margin, Ki-67 low expression and lower frequency of EGFR exon 19 deletion (P=0.002, 0.003, <0.001, <0.001, 0.001, <0.001, 0.002, 0.03, 0.02, 0.001, respectively) compared to mix mucinous and mucin-secreting adenocarcinoma. The ML-derived nomogram achieved discriminative ability, with an area under the curve (AUC) of 0.792 in the training set and 0.687 in the validation set, and exhibited satisfactory calibration for the preoperative diagnosis of pure mucinous adenocarcinoma. Multivariate Cox modeling identified higher CEA levels, and EGFR exon 19 deletion as independent prognostic factors in PMA (P=0.04, 0.049, respectively). Moreover, Kaplan-Meier curves revealed significantly different outcomes between low- and high-risk groups stratified by the nomogram score (P<0.001).
Conclusions
Pulmonary pure mucinous adenocarcinoma exhibited lower malignancy compared to the mixed mucinous and mucin-secreting type. The nomogram model developed and validated in this study exhibited promising efficacy in predicting histological subtype preoperatively and survival in patients with PMA, offering guidance for clinicians in diagnosis and prognosis prediction.
Keywords: Histological characteristics, machine learning (ML), nomogram, prognosis, pulmonary mucinous adenocarcinoma (PMA)
Highlight box.
Key findings
• Pulmonary pure mucinous adenocarcinoma (PMA) exhibited lower malignancy compared with the mixed mucinous and mucin-secreting type. The developed nomogram model demonstrated exceptional efficacy in predicting the histological subtype of PMA. The prognostic nomogram model effectively predicted PMA survival.
What is known and what is new?
• PMA represents a rare subtype of lung adenocarcinoma and is a heterogeneous entity. However, its malignant potential and optimal pathological classification remain incompletely defined, and practical tools for predicting histological subtype and prognosis are lacking.
• This study provides evidence that PMA represents a distinct entity with reduced malignancy compared with the mixed mucinous and mucin-secreting type. It further introduces two practical predictive tools: a preoperative nomogram for predicting the histological subtype of PMA and a prognostic nomogram for predicting PMA survival. In addition, independent prognostic factors for PMA were identified.
What is the implication, and what should change now?
• These findings support the potential pathological classification of PMA, with the pure mucinous type recognized as a distinct entity with lower malignancy. The diagnostic and prognostic nomograms offer practical tools that can aid clinical decision-making and risk stratification for patients with PMA. Further validation in multicenter cohorts is warranted before these models are adopted into routine practice.
Introduction
Lung cancer poses a significant global health burden, being the most prevalent cancer and the primary cause of cancer-related mortality worldwide (1,2). Pulmonary mucinous adenocarcinoma (PMA), a rare subtype comprising less than 1% of non-small cell lung carcinomas (NSCLC), is characterized by the presence of malignant epithelial cells with extracellular mucin pools. Analysis of 2018 U.S. cancer registry data at the population level revealed that PMA constituted 0.15% of lung cancer diagnoses, underscoring its epidemiological rarity (3). PMA typically presents with nonspecific clinical symptoms that can overlap with those of inflammatory and metastatic conditions, leading to diagnostic complexities. While population-based studies have indicated similar survival outcomes between PMA and non-mucinous adenocarcinomas, emerging data suggest potential variations in treatment responses and recurrence patterns, highlighting the need for further molecular stratification (4,5).
Specifically, the three-dimensional distribution patterns of acinar/papillary components and mucin deposition density are associated with the risk of lymphovascular invasion and chemotherapy response rates in PMA. These findings highlight the importance of incorporating assessments of tumor heterogeneity into PMA prognostic frameworks and therapeutic decision-making algorithms (6). Furthermore, different pathological variants of PMA may exhibit distinct associations with various clinical characteristics. Therefore, a comprehensive analysis of factors across the pathological subgroups of PMA is crucial for optimizing both diagnosis and treatment strategies.
In 2015, the World Health Organization (WHO) established a classification system for mucinous adenocarcinoma of the lung, which includes mucinous adenocarcinoma in situ (AIS), mucinous minimally invasive adenocarcinoma (MIA), invasive mucinous adenocarcinoma (IMA), mixed invasive mucinous and non-mucinous adenocarcinoma, and colloid adenocarcinoma, albeit with a weak recommendation and low-quality evidence (7,8). A detailed pathological subtype classification for PMA is currently lacking. Our previous research has delineated the pathological features of PMA based on the WHO criteria, enabling a more refined classification into pure mucinous, mixed mucinous, and mucin-secreting subtypes (9). Moreover, the correlation between PMA subtypes and clinical characteristics remains uncertain, and there is a need to identify further diagnostic and prognostic factors for PMA patients. Therefore, identification of predictive factors for PMA holds important clinical implications for optimizing patient management.
Recently, machine learning (ML), a subset of artificial intelligence (AI), has been employed to extract valuable information and identify elusive risk factors to enhance prediction accuracy by improving the handling of missing data and nonlinear parameters in cancer research (10,11). Numerous studies have demonstrated the benefits of ML in diagnosis and predicting prognosis in various types of lung cancers: Gould et al. constructed an ML based model to enhance the accuracy of NSCLC diagnosis using clinical and laboratory parameters (12); Zeng et al. developed a nomogram model for predicting survival in surgically treated limited-stage small cell lung cancer patients by incorporating demographic and clinical variables (10); and Yang et al. identified risk factors derived from genomic features, clinical profiles, and demographics using ML models, offering valuable insights to aid clinicians in devising personalized treatment strategies (13).
While an increasing number of studies have integrated pathology, radiology and genomics to developed a predictive model in types of lung cancer, research on prediction model for PMA remains limited. This study employs ML techniques to analyze 175 cases of primary PMA, aiming to explore the relationship between PMA subtypes and clinical characteristics, develop predictive nomogram models for the diagnosis and prognosis of PMA patients, and validate the effectiveness of the predictive models using internal data. Thus, specific clinical features can serve as predictive markers for determining potential pathological subtypes and estimating survival in PMA patients in clinical practice. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1241/rc).
Methods
Patients
The study cohort was derived from the prospectively maintained electronic medical record (EMR) system of Zhoushan Hospital, Zhejiang Province, China. This retrospective analysis enrolled 175 patients diagnosed with PMA who underwent surgical resection at the Department of Thoracic Surgery, Zhoushan Hospital, between January 2012 and December 2023. Demographic parameters consisted of age at diagnosis, sex, and smoking history. Patients underwent either lobectomy or segmentectomy with systematic of selective lymph node dissection, or wedge resection with lymph node sampling, depending on tumor location and invasiveness. Preoperative serum levels of carcinoembryonic antigen (CEA) were also collected using a Cobas e602 automated chemiluminescence analyzer (Roche, Mannheim, Germany). The pathological specimens were examined by two senior pathologists with expertise in lung cancer. The inclusion criteria consisted of: (I) histologically verified PMA; (II) prior surgical interventions; (III) comprehensive tumor-node-metastasis (TNM) staging data; (IV) follow-up information on mortality. Exclusion criteria included: (I) neoadjuvant treatment prior to surgery; (II) other histological subtypes of lung cancer; (III) multiple nodules. Ultimately, 175 cases were included, comprising 25 deaths (14.3%) and 150 survivors (85.7%). The patients with epidermal growth factor receptor (EGFR) mutation did not receive EGFR inhibitor therapy except one patient with brain metastasis. To ensure robust model evaluation and maximize data utility, the complete data set was randomly partitioned into a training set (70%) and an independent internal validation set (30%). The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Review Committee of Zhoushan Hospital (Approval No.: Zhoushan Hospital Ethical Review 2025-052-01) with informed consent waived due to its retrospective nature.
Pathological evaluation
Surgically resected tissue specimens were immediately fixed in 10% neutral formalin and then sectioned sequentially at 5 mm intervals to allow for thorough histological assessment. Full-thickness tumor sampling was performed to ensure complete histomorphological analysis. Two board-certified pathologists conducted independent blinded evaluation of hematoxylin and eosin-stained sections. Histopathological assessment focused on quantitatively evaluating predominant architectural patterns, including papillary, micropapillary, acinar, and solid subtypes. Each component was proportionally quantified as a percentage of the total tumor area, and the presence of spread through air spaces (STAS) was also assessed. The histopathological parameters encompassed PMA subclassification: the pure mucinous type, characterized by tumors composed entirely of mucinous cells; the mixed mucinous type, defined by tumors containing >10% mucinous cells along with other histological components (e.g., lepidic, papillary, acinar, or solid patterns); and the mucin-secreting type, referring to tumors with >10% extracellular mucin deposition (9), as shown in Figure 1.
Figure 1.
CT imaging and pathological categorization of PMA. (A) Pure mucinous type: CT imaging showed a relatively clear margin, lobulated, high-density nodule measuring 13 mm in maximum diameter in the posterior segment of the right upper lobe, adjacent to the pleura. Histopathological examination revealed adenocarcinoma composed exclusively of mucin-producing neoplastic cells with abundant extracellular mucin deposits. (B) Mixed mucinous type: CT imaging demonstrated a clear margin, spiculated appearance, pleural traction, and a mixed ground glass nodule measuring 20 mm in maximum diameter in the right upper lobe. Microscopic examination described the tumor tissue mainly characterized by acinar growth pattern, accompanied by 20% papillary growth pattern and 40% mucinous adenocarcinoma. (C) Mucin-secreting type: CT imaging indicated an indistinct margin, spiculated, lobulated, solid nodule with a maximum diameter of 14 mm in the left upper lobe. Microscopic examination demonstrated the tumor tissue was predominantly characterized by papillary/acinar growth pattern with lacking mucinous cells but rich in mucin. Pathological pictures were scanned by a digital pathology slide scanner and subsequently analyzed with K-viewer software (KF-PRO-005, KFBIO); hematoxylin and eosin-stained. CT, computed tomography; PMA, pulmonary mucinous adenocarcinoma.
Computed tomography (CT) examination and assessment
High-resolution CT (HRCT) scan was conducted preoperatively by using a 16-row multislice CT scanner with the lung window settings as follows: level =−500/width =1,500 HU, and two board-certified radiologists independently interpreted the images under blinded conditions. The CT characteristics of pulmonary lesions evaluated comprised tumor size, lobulation, spiculated edges, vacuole signs, pleural traction, alveolar collapse, margin clarity, mixed density, air bronchogram, and vascular convergence signs. The CT features specific to PMA subtypes were detailed in Figure 1.
Immunohistochemistry
Immunohistochemical staining was performed using 3,3'-diaminobenzidine chromogen. Two independent pathologists conducted blinded semi-quantitative assessments using microscopy (Olympus BX53, Japan), recording both the intensity and spatial distribution patterns of immunoreactivity. Staining intensity was scored as follows: 0= negative, 1= weak, 2= moderate, and 3= strong. The percentage of positive cells was scored as: 0 (≤10%), 1 (10–39%), 2 (40–69%), and 3 (≥70%). A total score was calculated as the product of the intensity and extent scores. Cases with a total score ≤2 (weak or negative staining) were categorized as low expression, whereas those with a score >2 (strong staining) were categorized as high expression. The antibodies included P63 (Maixin Biotech, Fuzhou, China), cytokeratin 7 (CK7) (Maixin Biotech, China), Ki-67 (ZSBio, Beijing, China), Napsin A (Maixin Biotech, China), and thyroid transcription factor-1 (TTF-1) (Maixin Biotech, China).
EGFR mutations
Paraffin-embedded tissues were employed for the detection of EGFR mutations in histologically confirmed patients. Nucleic acid was extracted from 2 to 5 sections by using QIAamp DNA FFPE tissue Kit (Qiagen, Hilden, Germany), followed by spectrophotometric quantification of DNA concentration and purity (A260/A280 ratio ≥1.8) using the Quawell Q3000 spectrophotometer (Quawell Technology, San Jose, CA, USA). EGFR mutations were detected on ABI 7500 Real-Time PCR system (ABI, Foster City, CA, USA) utilizing a commercial kit based on the amplification refractory mutation system (Yuanqi Diagnostics, Shanghai, China), following the manufacturer’s instructions.
Follow-up and death ascertainment
Disease progression data were collected through semi-annual structured telephone interviews, medical records, or the regional population death registry maintained by Zhoushan Municipal Center for Disease Control and Prevention (CDC). Follow-up was calculated from the date of curative resection to the time of death or the last follow-up on September 30, 2024. Overall survival (OS) served as the key outcome. The cohort achieved 100% follow-up completeness with no patients lost to follow-up. The adjudicators of the death outcome were blinded to the predictors.
Statistical analysis
Statistical analyses were performed using SPSS 29.0 (IBM Corp., Armonk, NY, USA) and R 4.4.1 (R Foundation, Vienna, Austria). The data set contained 8.3% missing data thus complete-case analysis was adopted. Prior to modeling, all predictor variables were standardized. Categorical variables were compared using Pearson’s Chi-squared test or Fisher’s exact test. To ensure robust model evaluation and maximize data utility, the complete data set was randomly partitioned into a training set (70%) and an internal validation set (30%). The randomization was performed using a computer-generated algorithm with a fixed seed (seed =123) to guarantee reproducibility. The training set was applied to the validation set to calculate the predicted risk for each patient. Additionally, stratified sampling approach was applied based on the outcome variable to preserve the distribution of event rates between the two sets.
Predictor selection, hyperparameter tuning, and internal validation via bootstrapping, were conducted exclusively on the training set to develop the model. The internal validation set was held out entirely during this phase and was used only for the final, single assessment of the model’s discrimination, calibration, and overall clinical utility. For the diagnostic model, independent predictors of pathological type were identified using binary logistic regression via the base glm() function (binomial family), indicating odds ratios (ORs) with 95% confidence intervals (CIs). Least absolute shrinkage and selection operator (LASSO) regression (glmnet package) for feature selection of pathological subtypes, followed by predictive nomogram model construction. Model discrimination was assessed by the area under the receiver operating characteristic curve (AUC), while clinical utility was assessed via decision curve analysis (DCA) (dca.r package). For the prognostic model, survival outcomes were analyzed using Cox proportional hazards regression [coxph() in the survival package]. The proportional hazards assumption was verified using Schoenfeld residuals. Hazard ratios (HRs) with 95% CIs were demonstrated, and the Cox model was used to plot the nomogram. Model calibration was verified using bootstrap-corrected calibration curves (rms package). Survival probabilities were estimated by the Kaplan-Meier method, and intergroup differences were compared using the log-rank test. To stratify patients into low- and high-risk categories, the median of the nomogram-derived risk scores was employed as the primary cutoff, ensuring balanced group sizes and statistical power. Additionally, to evaluate the robustness of this cutoff, an exploratory analysis was performed to use X-tile software (version 3.6.1, Yale University, USA). All analyses adhered to two-tailed testing principles, with statistical significance defined as P<0.05.
Results
Characteristics of PMA and development and validation of a predictive model
Among the 175 patients diagnosed with PMA, 113 (64.6%) cases were classified as pure mucinous type, 36 (20.6%) as mixed mucinous type, and 26 (14.8%) cases with mucin-secreting type. Nevertheless, pure mucinous adenocarcinoma was notably associated with smaller radiographic or pathological tumor size, lower lobe localization, lower rates of lymph node metastasis and STAS, early pathological stage, CEA ≤5 ng/mL, spiculated margin, Ki-67 low expression and lower frequency of EGFR exon 19 deletion (P=0.002, 0.003, <0.001, <0.001, 0.001, <0.001, 0.002, 0.03, 0.02, 0.001, respectively, Table 1). In the CT imaging analysis, pure mucinous type demonstrated a significantly higher prevalence of pleural traction and spiculated margins compared to mixed mucinous type (P=0.02, 0.01, respectively). However, no statistically significant differences were observed between pure mucinous type and mucin-secreting type for these two imaging biomarkers (P=0.22, 0.57, respectively). Additionally, comparative analysis revealed no statistically significant difference between mixed mucinous and mucin-secreting subtypes in tumor size, tumor localization, lymph node metastasis, STAS, pathological stage, and CEA level (P all >0.05). Moreover, expression level of P63, CK7, Napsin A, and TTF-1 did not exhibit significant variations among different subtypes of PMA (P all >0.05, Table 1).
Table 1. Clinicopathological and CT characteristics between pulmonary mucinous adenocarcinoma with pure mucinous, mixed mucinous and mucin-secreting.
| Characteristics | Total | Pure mucinous | Mixed mucinous | Mucin-secreting | P value |
|---|---|---|---|---|---|
| Age | |||||
| ≤60 years | 61 (34.8) | 37 (32.7) | 11 (30.6) | 13 (50.0) | 0.23 |
| >60 years | 114 (65.2) | 76 (67.3) | 25 (69.4) | 13 (50.0) | |
| Sex | |||||
| Male | 70 (40.0) | 40 (35.4) | 25 (69.4) | 13 (50.0) | 0.24 |
| Female | 105 (60.0) | 73 (64.6) | 11 (30.6) | 13 (50.0) | |
| Smoking status | |||||
| Never | 141 (80.6) | 92 (81.4) | 31 (86.1) | 18 (69.2) | 0.22 |
| Current | 34 (19.4) | 21 (18.6) | 5 (13.9) | 8 (30.8) | |
| Surgical procedure | |||||
| Lobectomy | 130 (74.3) | 83 (73.5) | 27 (75.0) | 20 (76.9) | 0.93 |
| Sublobectomy | 45 (25.7) | 30 (26.5) | 9 (25.0) | 6 (23.1) | |
| Radiographic tumor size | |||||
| ≤2.0 cm | 124 (70.9) | 90 (79.6) | 21 (58.3) | 13 (50.0) | 0.002†‡** |
| >2.0 cm | 51 (29.1) | 23 (20.4) | 15 (41.7) | 13 (50.0) | |
| Pathological tumor size | |||||
| ≤2.0 cm | 125 (71.4) | 90 (79.6) | 22 (61.1) | 13 (50.0) | 0.003†‡ |
| >2.0 cm | 50 (28.6) | 23 (20.4) | 14 (38.9) | 13 (50.0) | |
| Tumor location | |||||
| Upper and middle lobe | 80 (45.7) | 39 (34.5) | 25 (69.4) | 16 (61.5) | <0.001†‡*** |
| Lower lobe | 95 (54.3) | 74 (65.5) | 11 (30.6) | 10 (38.5) | |
| Lymph node metastasis | |||||
| No | 154 (88.0) | 109 (96.5) | 28 (77.8) | 17 (65.4) | <0.001†‡*** |
| Yes | 21 (12.0) | 4 (3.5) | 8 (22.2) | 9 (34.6) | |
| STAS | |||||
| No | 160 (91.4) | 110 (97.3) | 29 (80.6) | 21 (80.8) | 0.001†‡** |
| Yes | 15 (8.6) | 3 (2.7) | 7 (19.4) | 5 (19.2) | |
| Pathological stage | |||||
| Ⅰ | 144 (82.3) | 103 (89.4) | 24 (66.7) | 17 (65.4) | <0.001†‡*** |
| Ⅱ and Ⅲ | 31 (17.7) | 10 (10.6) | 12 (33.3) | 9 (34.6) | |
| Serum CEA | |||||
| ≤5 ng/mL | 150 (71.4) | 105 (92.9) | 26 (72.2) | 19 (73.1) | 0.002†‡** |
| >5 ng/mL | 25 (28.6) | 8 (7.1) | 10 (27.8) | 7 (26.9) | |
| CT characteristics | |||||
| Margin | |||||
| Clear | 126 (72.0) | 81 (71.7) | 24 (66.7) | 21 (80.8) | 0.50 |
| Indistinct | 49 (28.0) | 32 (28.3) | 12 (33.3) | 5 (19.2) | |
| Mixed density | |||||
| No | 115 (65.7) | 68 (60.2) | 27 (75.0) | 20 (76.9) | 0.12 |
| Yes | 60 (34.3) | 45 (39.2) | 9 (25.0) | 6 (23.1) | |
| Vacuole sign | |||||
| No | 144 (82.3) | 90 (79.6) | 29 (80.6) | 25 (96.2) | 0.17 |
| Yes | 31 (17.7) | 23 (20.4) | 7 (19.4) | 1 (3.8) | |
| Pleural traction | |||||
| No | 147 (84.0) | 100 (88.5) | 26 (72.2) | 21 (80.8) | 0.056† |
| Yes | 28 (16.0) | 13 (11.5) | 10 (27.8) | 5 (19.2) | |
| Lobulation | |||||
| No | 109 (62.3) | 75 (66.4) | 18 (50.0) | 16 (61.5) | 0.21 |
| Yes | 66 (37.7) | 38 (33.6) | 18 (50.0) | 10 (38.5) | |
| Air bronchogram | |||||
| No | 172 (98.3) | 111 (98.2) | 35 (97.2) | 26 (100) | >0.99 |
| Yes | 3 (1.7) | 2 (1.8) | 1 (2.8) | 0 (0.0) | |
| Spiculated margin | |||||
| No | 111 (63.4) | 76 (67.3) | 16 (44.4) | 19 (73.1) | 0.03†§* |
| Yes | 64 (36.6) | 37 (32.7) | 20 (55.6) | 7 (26.9) | |
| Immunohistochemistry (n=125) | |||||
| p63 (n=125) | |||||
| Low expression | 99 (79.2) | 72 (80.9) | 14 (70.0) | 13 (81.3) | 0.29 |
| High expression | 26 (20.8) | 17 (19.1) | 6 (30.0) | 3 (18.7) | |
| CK7 (n=125) | |||||
| Low expression | 28 (22.4) | 18 (20.2) | 7 (35.0) | 3 (18.7) | 0.20 |
| High expression | 97 (77.6) | 71 (79.8) | 13 (65.0) | 13 (81.3) | |
| Ki-67 (n=125) | |||||
| Low expression | 113 (90.4) | 85 (95.5) | 16 (80.0) | 12 (75.0) | 0.02†‡* |
| High expression | 12 (9.6) | 4 (4.5) | 4 (20.0) | 4 (25.0) | |
| Napsin A (n=125) | |||||
| Low expression | 90 (72.0) | 65 (73.0) | 13 (65.0) | 12 (75.0) | 0.72 |
| High expression | 35 (28.0) | 24 (27.0) | 7 (35.0) | 4 (25.0) | |
| TTF-1 (n=125) | |||||
| Low expression | 40 (32.0) | 32 (36.0) | 5 (25.0) | 3 (18.7) | 0.17 |
| High expression | 85 (68.0) | 57 (64.0) | 15 (75.0) | 13 (81.3) | |
| EGFRexon 19 deletion (n=125) | |||||
| No | 119 (95.2) | 88 (98.9) | 19 (95.0) | 12 (75.0) | 0.001‡** |
| Yes | 6 (4.8) | 1 (1.1) | 1 (5.0) | 4 (25.0) | |
| EGFR exon 21 mutation (n=125) | |||||
| No | 122 (97.6) | 88 (98.9) | 19 (95.0) | 15 (93.7) | 0.21 |
| Yes | 3 (2.4) | 1 (1.1) | 1 (5.0) | 1 (6.3) |
Data are presented as n (%). *, P<0.05; **, P<0.01; ***, P<0.001. †, pure mucinous vs. mixed mucinous. ‡, pure mucinous vs. mucin-secreting. §, mixed mucinous vs. mucin-secreting showed significant differences. CEA, carcinoembryonic antigen; CK7, cytokeratin 7; CT, computed tomography; EGFR, epidermal growth factor receptor; STAS, spread through air spaces; TTF-1, thyroid transcription factor-1.
Considering the distinct clinicopathological characteristics observed in pure mucinous type compared to mixed mucinous type and mucin-secreting type, our analysis focused on establishing differential criteria to distinguish pure mucinous lesions from these two variant forms preoperatively and the results presented significant difference in radiographic tumor size, tumor location, serum CEA level, mixed density, and pleural traction (P=0.001, <0.001, 0.001, 0.04, 0.03, respectively, Table 2). Multivariate logistic analysis revealed radiographic tumor size (OR =2.473, 95% CI: 1.132–5.399, P=0.02), tumor location (OR =0.327, 95% CI: 0.162–0.663, P=0.002), serum CEA level (OR =3.200, 95% CI: 1.169–8.756, P=0.02) were the independent factors for pure mucinous adenocarcinoma (Table 2).
Table 2. Univariate and multivariate logistic regression analysis for pure mucinous adenocarcinoma.
| Characteristic | Group | Univariate analysis | Multivariate analysis | ||||||
|---|---|---|---|---|---|---|---|---|---|
| β | OR | 95% CI | P | β | OR | 95% CI | P | ||
| Age | ≤60 vs. >60 years | –0.260 | 0.771 | 0.405–1.469 | 0.43 | ||||
| Sex | Female vs. male | 0.537 | 1.711 | 0.911–3.212 | 0.10 | ||||
| Smoking status | Never vs. current | 0.150 | 1.162 | 0.536–2.520 | 0.70 | ||||
| Surgical procedure | Sublobectomy vs. lobectomy | 0.124 | 1.113 | 0.554–2.315 | 0.73 | ||||
| Radiographic tumor size | ≤2.0 vs. >2.0 cm | 1.170 | 3.323 | 1.635–6.348 | <0.001*** | 0.905 | 2.473 | 1.132–5.399 | 0.02* |
| Tumor location | Upper and middle vs. lower lobe | –1.310 | 0.270 | 0.140–0.511 | <0.001*** | –1.117 | 0.327 | 0.162–0.663 | 0.002** |
| Serum CEA level | ≤5 vs. >5 ng/mL | 1.601 | 4.958 | 1.996–12.318 | <0.001*** | 1.163 | 3.200 | 1.169–8.756 | 0.03* |
| Margin | Clear vs. indistinct | –0.045 | 0.956 | 0.479–1.910 | 0.90 | ||||
| Mixed density | No vs. yes | –0.729 | 0.482 | 0.241–0.964 | 0.04* | –0.751 | 0.472 | 0.213–1.046 | 0.07 |
| Vacuole sign | No vs. yes | –0.545 | 0.580 | 0.242–1.387 | 0.22 | ||||
| Pleural traction | No vs. yes | 0.898 | 2.455 | 1.082–5.572 | 0.03* | 0.760 | 2.139 | 0.880–5.202 | 0.09 |
| Lobulation | No vs. yes | 0.486 | 1.625 | 0.862–3.065 | 0.13 | ||||
| Air bronchogram | No vs. yes | –0.094 | 0.910 | 0.081–10.239 | 0.94 | ||||
| Spiculated margin | No vs. yes | 0.460 | 1.585 | 0.838–2.998 | 0.16 | ||||
*, P<0.05; **, P<0.01; ***, P<0.001. CEA, carcinoembryonic antigen; CI, confidence interval; OR, odds ratio.
The baseline characteristics and differences between the training and validation sets were shown in Table S1, with no significant differences observed between the two sets (all P>0.05). LASSO regression was then applied to identify parameters associated with different subtypes of PMA (Figure 2A,2B), followed by the development of a nomogram based on the four selected parameters (Figure 2C). The nomogram is primarily intended for preoperative stratification, with non-lower lobe tumor location, elevated serum CEA, radiographic tumor size, and mixed density retained as key predictive covariates for identifying the non-pure subtype. According to the nomogram scoring, tumor location contributed most strongly to the total score. The model achieved an AUC of 0.792 in the training set (Figure 2D) and 0.687 in the validation set (Figure 2E), indicating satisfactory discriminative performance. The calibration plot demonstrated good agreement [mean absolute error (MAE) =0.041], with some deviations observed within the intermediate probability ranges (Figure 3A). The DCA showed that the nomogram model provided greater net benefit across a wide range of high-risk thresholds compared to dichotomous risk classification strategies. A sharp decline in the nomogram curve was observed around a threshold of 0.2 (Figure 3B). This decline implied that the model’s false-negative or false-positive rate increased within this range, which reduced its net benefit.
Figure 2.
Classification of PMA. (A) Coefficient shrinkage profile demonstrating regularization paths of 35 candidate predictors. The upper x-axis indicates cumulative non-zero coefficients, while the lower x-axis represents log-transformed penalty parameter (λ). (B) 10-fold cross-validation with minimum criteria (dashed vertical line) determining optimal λ selection. The y-axis quantifies cross-validated partial likelihood deviance (mean ± SE). Optimal feature selection occurred at log (λ) =−3.6. The optimal number of knots for curve interception was determined to be between 4 and 10. Following variable selection, the ML algorithm identified and retained 4 predictors with non-zero coefficients for inclusion in the subsequent model. (C) Nomogram for predicting the pathological classification of PMA by using 5 predictors. (D) ROC curve of the training set. (E) ROC curve of the internal validation set. AUC, area under the curve; CEA, carcinoembryonic antigen; ML, machine learning; PMA, pulmonary mucinous adenocarcinoma; ROC, receiver operating characteristic; SE, standard error.
Figure 3.
Evaluation of the logistic regression model for PMA subtyping. (A) Calibration plot of PMA pathological subtyping predicted by the logistic regression model. (B) DCA employed for a model forecasting the pathological classifications of PMA. C.L., confidence level; DCA, decision curve analysis; PMA, pulmonary mucinous adenocarcinoma.
Analysis of prognostic factors and predictive modeling for PMA
Univariate Cox proportional hazards analysis revealed larger tumor size, lymph node metastasis, mixed mucinous and mucin-secreting type, advanced pathological stage, higher CEA levels, indistinct margin, Ki-67 high expression and EGFR exon 19 deletion were distinctly correlated with poor prognosis (P=0.002, <0.001, 0.008, <0.001, <0.001, 0.003, 0.01, 0.02, respectively, Table 3). Subsequently, variables showing univariate significance were included in the multivariate Cox proportional hazards regression. Higher CEA levels (HR =4.759, 95% CI: 1.094–20.695, P=0.04), and EGFR exon 19 deletion (HR =4.728, 95% CI: 1.005–22.246, P=0.049), were identified as independent prognostic factors in PMA (Table 3).
Table 3. Univariate and multivariate Cox regression analysis of postoperative survival in PMA.
| Characteristic | Group | Univariate analysis | Multivariate analysis | ||||||
|---|---|---|---|---|---|---|---|---|---|
| β | HR | 95% CI | P | β | HR | 95% CI | P | ||
| Age | ≤60 vs. >60 years | –0.509 | 0.601 | 0.274–1.319 | 0.20 | ||||
| Sex | Female vs. male | 0.030 | 1.301 | 0.462–2.302 | 0.94 | ||||
| Smoking status | Never vs. current | 0.130 | 1.149 | 0.425–3.052 | 0.80 | ||||
| Surgical procedure | Non-lobectomy vs. lobectomy | –0.521 | 0.594 | 0.254–1.391 | 0.23 | ||||
| Pathological tumor size | ≤2.0 vs. >2.0 cm | 1.253 | 3.501 | 1.588–7.721 | 0.002** | –0.249 | 0.780 | 0.142–4.277 | 0.78 |
| Lymph node metastasis | No vs. yes | 1.719 | 5.579 | 2.530–12.30 | <0.001*** | 0.121 | 1.129 | 0.176–7.236 | 0.90 |
| STAS | No vs. yes | 0.972 | 2.642 | 0.784–8.913 | 0.12 | ||||
| Tumor location | Upper and middle vs. lower lobe | –0.612 | 0.542 | 0.243–1.211 | 0.14 | ||||
| Histological type | Pure mucinous vs. mixed mucinous and mucin-secreting | 1.116 | 3.053 | 1.346–6.926 | 0.008** | –1.130 | 0.323 | 0.081–1.292 | 0.11 |
| Pathological stage | Ⅰ vs. Ⅱ and Ⅲ | 1.749 | 5.748 | 2.607–12.670 | <0.001*** | 1.330 | 3.782 | 0.435–32.871 | 0.23 |
| Serum CEA | ≤5 vs. >5 ng/mL | 1.892 | 6.633 | 2.918–15.078 | <0.001*** | 1.560 | 4.759 | 1.094–20.695 | 0.04* |
| Margin | Clear vs. indistinct | 1.188 | 3.282 | 1.480–7.276 | 0.003** | –0.324 | 0.723 | 0.198–2.642 | 0.62 |
| Mixed density | No vs. yes | 0.016 | 1.016 | 0.448–2.301 | 0.97 | ||||
| Vacuole sign | No vs. yes | –1.498 | 0.224 | 0.030–1.655 | 0.14 | ||||
| Pleural traction | No vs. yes | 0.322 | 1.380 | 0.514–3.700 | 0.52 | ||||
| Lobulation | No vs. yes | 0.103 | 1.108 | 0.502–2.443 | 0.80 | ||||
| Air bronchogram | No vs. yes | 0.665 | 1.944 | 0.258–14.617 | 0.52 | ||||
| Spiculated margin | No vs. yes | 0.520 | 1.683 | 0.761–3.718 | 0.20 | ||||
| p63 (n=125) | Low vs. high expression | 0.009 | 1.009 | 0.283–3.593 | 0.99 | ||||
| CK7 (n=125) | Low vs. high expression | 0.190 | 1.209 | 0.324–4.507 | 0.78 | ||||
| Ki-67 (n=125) | Low vs. high expression | 1.385 | 3.994 | 1.382–11.548 | 0.01* | 0.003 | 1.003 | 0.269–3.733 | >0.99 |
| Napsin A (n=125) | Low vs. high expression | –1.197 | 0.302 | 0.039–2.340 | 0.25 | ||||
| TTF-1 (n=125) | Low vs. high expression | –0.275 | 0.759 | 0.275–2.097 | 0.60 | ||||
| EGFRexon 19 deletion (n=125) | No vs. yes | 1.482 | 4.403 | 1.225–15.822 | 0.02* | 1.554 | 4.728 | 1.005–22.246 | 0.049* |
| EGFRexon 21 mutation (n=125) | No vs. yes | 0.635 | 1.887 | 0.246–14.455 | 0.54 | ||||
*, P<0.05; **, P<0.01; ***, P<0.001. CEA, carcinoembryonic Antigen; CI, confidence interval; CK7, cytokeratin 7; CT, computed tomography; EGFR, epidermal growth factor receptor; HR, hazard ratio; PMA, pulmonary mucinous adenocarcinoma; STAS, spread through air spaces; TTF-1, thyroid transcription factor-1.
LASSO regression analysis identified CEA levels, Ki-67 expression, EGFR exon 19 deletion, tumor size, lymph node metastasis, histological type, indistinct margin, and stage as factors associated with postoperative survival in patients with PMA (Figure 4A,4B). These parameters emerged as significant indicators of adverse survival, highlighting their clinical relevance in risk stratification. The prognostic model integrated critical determinants including pathological tumor size >2.0 cm, lymph node metastasis, advanced pathological stage, non-pure histology, CEA levels >5 ng/mL, indistinct margins, Ki-67 high expression and EGFR exon 19 deletion. Based on these predictive factors, a nomogram was developed to predict 3- and 5-year OS, and among which CEA level contributed most strongly to the nomogram score (Figure 4C). In the integrated prognostic model, the Harrell’s concordance index (C-index) was 0.888 (95% CI: 0.808–0.896) in the training set and 0.770 (95% CI: 0.606–0.934) in the validation set. Calibration curves (Figure 5) for 3- and 5-year survival yielded MAE of 0.038 and 0.047 in the training set (Figure 5A,5C), and 0.041 and 0.056 in the validation set (Figure 5B,5D), respectively. By contrast, a benchmark model constructed using pathological stage alone exhibited notably inferior discrimination, with C-indices of 0.738 (95% CI: 0.603–0.873) and 0.583 (95% CI: 0.432–0.734) across the two cohorts, and corresponding calibration MAEs of 0.023/0.021 and 0.039/0.045 for 3- and 5-year predictions (Figure S1A-S1D). These data confirm that our model provides incremental prognostic accuracy beyond conventional staging. Furthermore, we performed an ablation analysis to evaluate the models independence on established clinicopathological variables. Upon systematic exclusion of canonical predictors, including pathological stage, tumor size, lymph node status, STAS, and Ki-67 expression, the model retained robust discriminative capacity, achieving a C-index of 0.840 (95% CI: 0.772–0.958) in the training set and 0.668 (95% CI: 0.493–0.843) in the validation set, with calibration MAEs of 0.027/0.029 and 0.038/0.049 for 3- and 5-year endpoints, respectively (Figure S1E-S1H). This demonstrates that the model maintains substantial predictive value independent of well-characterized prognostic covariates. In addition, we constructed a series of stepwise models to evaluate the incremental value of multimodal integration: a clinical characters model, a model combining clinical features with CT findings, a model combining clinical features with CT findings and serum CEA, and the full multimodal model. The corresponding C-indices were 0.838 (95% CI: 0.747–0.929), 0.841 (95% CI: 0.751–0.931), 0.873 (95% CI: 0.784–0.962), and 0.875 (95% CI: 0.781–0.991), respectively. The stepwise addition of CT features and serum CEA progressively improved model discrimination, confirming the incremental benefit of multimodal integration.
Figure 4.
LASSO-based feature selection and prognostic nomogram for PMA. (A) LASSO coefficient curve for 35 features (y-axis). The numbers on the upper x-axis represent the average number of predictor variables. The x-axis below represents log (λ). (B) Select the tuning parameter (λ) for the LASSO model through 10-fold cross-validation based on the minimum criterion. The y-axis represents the partial likelihood deviance. Following variable selection, the ML algorithm identified and retained 8 predictors with non-zero coefficients for inclusion in the subsequent model. (C) A nomogram was constructed to predict 3- and 5-year OS. CEA, carcinoembryonic antigen; LASSO, least absolute shrinkage and selection operator; ML, machine learning; OS, overall survival; PMA, pulmonary mucinous adenocarcinoma.
Figure 5.
Calibration plots of the full multimodal model for predicting 3- and 5-year OS in patients with PMA. Calibration curve for 3-year OS (A) in the training set and (B) validation set. Calibration curve for 5-year OS (C) in the training set and (D) validation set. (E) Kaplan-Meier survival curves for low- and high-risk groups of PMA. (F) Kaplan-Meier survival curves for groups of PMA with pure mucinous type, mixed mucinous type and mucin-secreting type. OS, overall survival; PMA, pulmonary mucinous adenocarcinoma.
By calculating the nomogram-derived risk scores (range: 0–375 points) for each patient, we determined the statistically optimal cutoff value through survival-based cutoff point analysis. A threshold of 34 points was established to stratify patients into different prognostic groups: individuals with scores ≤34 points formed the low-risk group (n=119), while those with scores >34 points constituted the high-risk group (n=56). Kaplan-Meier survival analysis revealed significantly distinct clinical outcomes between the two risk groups, as evidenced by a remarkable separation in the OS curves (P<0.001, Figure 5E). Moreover, although histological subtype of PMA was not identified as an independent factor for the outcome of patients with PMA, Kaplan-Meier curve displayed that cases with pure mucinous type had longer OS than those with mixed mucinous type and mucin-secreting type (P=0.03, 0.01, respectively, Figure 5F).
Discussion
PMA represents a rare subtype of lung adenocarcinoma, characterized by its low incidence rate yet sharing a similar epidemiological profile with other lung adenocarcinoma subtypes (14). Noteworthy distinctions in clinical, pathological, genetic, and CT characteristics set PMA apart from non-mucinous lung adenocarcinomas (15-17). Notably, our previous research identified heterogeneous PMA subtypes based on variations in mucinous cellularity and extracellular mucinous matrix composition (9). Despite these advancements, inconsistent survival persistently exists, likely due to limited sample sizes and diverse analytical approaches. A key challenge lies in integrating multifaceted data including clincopathological, genetic and CT imaging features, to accurately predict prognosis. ML-based nomogram models offer a promising tool for distinguishing PMA from non-mucinous adenocarcinomas by leveraging high-dimensional data to capture nonlinear relationships and enhance predictive accuracy (18). However, the systematic application of nomogram models to assess histological subtype and survival outcomes in PMA, remains scarce, highlighting an understudied area. Addressing this gap could refine risk stratification and guide personalized therapeutic interventions for this rare yet clinically complex malignancy.
Our study reaffirmed previous observations that PMA is predominantly associated with the lower lobe of the lung, T1 stage, N0 stage, and pathological stage I (19). Notably, we found that the pure mucinous type of PMA exhibited less aggressive behavior compared with the mixed mucinous and mucin-secreting types, as characterized by smaller tumor size, earlier pathological stage, CEA level ≤5 ng/mL, lower rates of lymph node metastasis and STAS, low Ki-67 expression, and a lower frequency of EGFR exon 19 deletion. Hwang et al. classified the malignant degree of PMA based on the extent of lepidic and non-lepidic invasive patterns, noting that invasive patterns were least common in small tumors and progressively increased in larger tumors, suggesting a correlation between histological subtype of PMA and tumor progression (20). Moreover, consistent with prior reports, EGFR mutation rates in PMA vary significantly across populations (0–33.3%) due to racial and technical disparities (21-23). A subsequent investigation revealed EGFR mutations in 3 out of 41 metastatic PMA cases, which showed favorable clinical responses to EGFR-targeted therapy (24). Our results indicated EGFRexon 19 and 21 mutations accounted for 4.8% and 2.4%, respectively. Crucially, mucin-secreting type exhibited a higher frequency of EGFR mutation compared to pure mucinous type, consistent with another Chinese cohort (23). Additionally, in CT imaging features, the pure mucinous type showed a notably higher prevalence of pleural traction and spiculated margins compared to mixed mucinous type, while being similar to mucin-secreting type. Compared to non-mucinous adenocarcinoma, mucinous adenocarcinoma were more likely to exhibit indistinct margins and pleural retraction (18,25,26), and the spiculated margins being more common in pure mucinous adenocarcinoma than in mixed mucinous adenocarcinoma (17).
Significantly, we have pioneered the development and validation of a predictive model for distinguishing pure PMA utilizing logistic and LASSO regression analyses within an ML framework. According to the nomogram scoring, tumor location in the lower lobe contributed most strongly to the identification of pure PMA, which reflects the distinctive anatomical predilection of PMA. The underlying mechanisms may involve gravity-dependent mucus clearance and region-specific inflammatory microenvironments. The preoperative model demonstrated robust performance with relatively high AUC values in both the training and validation sets, along with well-calibrated curves. This predictive model offers a promising approach for accurately identifying pure PMA, providing a tool for facilitating more precise therapeutic interventions for patients.
To enhance the accuracy of prognostic modeling for PMA, we initially employed Cox regression analysis to identify prognostic factors of PMA, identifying higher CEA level and EGFR exon 19 deletion as independent prognostic factors for PMA patients. CEA level also made the strongest contribution to the nomogram score. CEA has been established as a clinically significant tumor biomarker known to facilitate malignant progression in lung cancer (27). Our previous cohort study revealed a correlation between elevated serum CEA level and adverse clinical outcomes (28). Notably, downstream genes activation by EGFR mutants has been linked to anti-apoptosis, pyroptosis, metastasis, and immune evasion (29-31). While specific subtypes of EGFR mutations are rarely used as distinguishing factors in prognostic assessments for PMA (23), our previous findings indicated that EGFRexon 19 deletion was an unfavourable prognostic marker in lung adenocarcinoma patients, with worse outcomes compared to EGFRexon 21 mutation cases (28). Moreover, Deng et al. demonstrated EGFR mutation was a poor prognostic factor in PMA and advanced stage lung adenocarcinoma (32), which was correlated with our study. One reason is that, except for one patient who received EGFR inhibitor therapy when brain metastasis occurred three years later, none of the EGFR mutation cases in our cohort received EGFR inhibitor therapy. Therefore, our results differ from those of some studies showing that EGFR mutation is a favorable prognostic predictor for NSCLC patients who received EGFR inhibitor therapy (33-36).
LASSO regression technique from ML was utilized to incorporate variables in the development of a nomogram with the calibration curves demonstrated strong consistency. Our stepwise model comparison demonstrated the incremental value of multimodal integration, with the full model achieving a significantly higher C-index than the clinical-only model. The ablation model also retained a relatively high C-index. Despite not achieving high C-index, the survival curve displayed significant differentiation between the two stratified groups, indicating the effectively discrimination ability of the nomogram and its potential utility in clinical prediction. Consistent with prior studies that had also developed satisfactory nomogram model despite moderate C-index values (10), our findings underscored the importance of histopathological classification in PMA by demonstrating that pure mucinous adenocarcinoma exhibited significantly superior clinical outcome compared to other PMA subtype, validating the prognostic relevance of histopathological classification in PMA. Additionally, our findings indicated that lobectomy for PMA provided comparable OS to sublobectomy, and that pure mucinous adenocarcinoma exhibited lower aggressiveness than the mixed mucinous and mucin-secreting type. Based on these observations, we suggest that segmentectomy or wedge resection may be sufficient for patients with pure mucinous adenocarcinoma, and that extending the resection to lobectomy is not routinely necessary. Similarly, Yang and colleagues reported that sublobectomy yielded OS comparable to lobectomy in small-sized PMA (37). Therefore, preoperative prediction of histologic subtype is clinically important to guide the surgical decision-making.
This study exhibits notable strengths and limitations. Initially, the data collection was confined to a single center, potentially introducing bias, and further multicenter cohort studies are warranted to enhance the dataset’s breadth and credibility. Secondly, the restricted patient population with PMA and the frequency of fatalities might result in data bias when partitioning patients into training and validation sets. The limited sample size and reduced event rate could lead to increased variability in calibration assessments. Thirdly, although the proportion of missing data in this study was relatively low, multiple imputation was not performed, which may have introduced some degree of bias into the results. Additionally, due to constraints in laboratory infrastructure in the archipelago setting, KRAS mutations were not identified in the patient cohort. Consequently, expanding the sample size, refining grouping strategies, or employing advanced corrective methods is recommended. Furthermore, it is crucial to validate the model’s performance across multiple datasets to ensure its generalizability.
Furthermore, our findings suggested that the models captureed both established and novel signals. The diagnostic model identified novel multimodal patterns, particularly the combination of CT-derived features and serum markers, which together improved subtype prediction beyond CT characters alone preoperatively. The prognostic model confirmed expected associations, such as smaller tumors, negative lymph node status, early stage, lower CEA, and less aggressive histology, as predictors of better outcomes. Additionally, we highlighted that EGFR exon 19 deletion were identified as a novel predictor of postoperative survival, adding new insights beyond established risk factors. These findings suggest that the models are not merely recapitulating known clinical knowledge but are extracting complementary information from multimodal inputs, supporting the incremental value of our approach.
Conclusions
In conclusion, we have established a refined histopathological classification approach for PMA with our validation study revealing distinctive clinicopathological characteristics indicating lower aggressiveness in pure mucinous adenocarcinoma compared to mixed mucinous and mucin-secreting types. Moreover, our developed nomogram model exhibited performance in predicting histological subtype of PMA preoperatively. Importantly, higher CEA level, and EGFR exon 19 deletion were confirmed as independent prognostic factors for PMA. These findings offer valuable insights for histopathological diagnosis and prognostic evaluation in the context of PMA.
Supplementary
The article’s supplementary files as
Acknowledgments
We thank the anonymous reviewers for reviewing the manuscript and providing constuctive comments.
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 retrospective cohort study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Institutional Review Board of Zhoushan Hospital (Approval No. Zhoushan Hospital Ethical Review 2025-052-01). The requirement for informed consent was formally waived by the ethics committee due to its retrospective nature. Individual patient contact was impractical given the study’s historical design.
Footnotes
Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1241/rc
Funding: This study was supported in part by grants from the Health Commission of Zhejiang Province, and National Traditional Chinese Medicine Comprehensive Reform Demonstration Zone (grant Nos. 2022RC292 and GZY-KJS-ZJ-2025-091 to W.Z.), and the Scientific Research Program of Traditional Chinese Medicine of Zhejiang Province (grant No. 2021ZB349 to L.L.).
Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1241/coif). The authors have no conflicts of interest to declare.
Data Sharing Statement
Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1241/dss
References
- 1. Siegel RL, Kratzer TB, Giaquinto AN, et al. Cancer statistics, 2025. CA Cancer J Clin 2025;75:10-45.. 10.3322/caac.21871 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Li C, Lei S, Ding L, et al. Global burden and trends of lung cancer incidence and mortality. Chin Med J (Engl) 2023;136:1583-90.. 10.1097/CM9.0000000000002529 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Moon SW, Choi SY, Moon MH. Effect of invasive mucinous adenocarcinoma on lung cancer-specific survival after surgical resection: a population-based study. J Thorac Dis 2018;10:3595-608.. 10.21037/jtd.2018.06.09 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Zhu D, Zhang Q, Rui Z, et al. Pulmonary invasive mucinous adenocarcinoma mimicking pulmonary actinomycosis. BMC Pulm Med 2022;22:181.. 10.1186/s12890-022-01971-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Yin YH, Qi YG, Wang B. Differential diagnosis of pulmonary nodular mucinous adenocarcinoma and tuberculoma with dynamic CT: a retrospective study. J Thorac Dis 2022;14:1225-31.. 10.21037/jtd-22-372 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Li W, Yang Y, Yang M, et al. Clinicopathologic Features and Survival Outcomes of Primary Lung Mucinous Adenocarcinoma Based on Different Radiologic Subtypes. Ann Surg Oncol 2024;31:167-77.. 10.1245/s10434-023-14193-w [DOI] [PubMed] [Google Scholar]
- 7. Travis WD, Brambilla E, Noguchi M, et al. International association for the study of lung cancer/american thoracic society/european respiratory society international multidisciplinary classification of lung adenocarcinoma. J Thorac Oncol 2011;6:244-85.. 10.1097/JTO.0b013e318206a221 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Travis WD, Brambilla E, Noguchi M, et al. International Association for the Study of Lung Cancer/American Thoracic Society/European Respiratory Society: international multidisciplinary classification of lung adenocarcinoma: executive summary. Proc Am Thorac Soc 2011;8:381-5.. 10.1513/pats.201107-042ST [DOI] [PubMed] [Google Scholar]
- 9. Pan X, Fang R, Zhang B, et al. Pathological and imaging features of pulmonary invasive mucinous adenocarcinoma-a retrospective cohort study. Transl Lung Cancer Res 2024;13:1376-82.. 10.21037/tlcr-24-526 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Zeng Q, Li J, Tan F, et al. Development and Validation of a Nomogram Prognostic Model for Resected Limited-Stage Small Cell Lung Cancer Patients. Ann Surg Oncol 2021;28:4893-904.. 10.1245/s10434-020-09552-w [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Li Y, Wu X, Yang P, et al. Machine Learning for Lung Cancer Diagnosis, Treatment, and Prognosis. Genomics Proteomics Bioinformatics 2022;20:850-66.. 10.1016/j.gpb.2022.11.003 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Gould MK, Huang BZ, Tammemagi MC, et al. Machine Learning for Early Lung Cancer Identification Using Routine Clinical and Laboratory Data. Am J Respir Crit Care Med 2021;204:445-53.. 10.1164/rccm.202007-2791OC [DOI] [PubMed] [Google Scholar]
- 13. Yang Y, Xu L, Sun L, et al. Machine learning application in personalised lung cancer recurrence and survivability prediction. Comput Struct Biotechnol J 2022;20:1811-20.. 10.1016/j.csbj.2022.03.035 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Xu L, Li C, Lu H. Invasive mucinous adenocarcinoma of the lung. Transl Cancer Res 2019;8:2924-32.. 10.21037/tcr.2019.11.02 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Gow CH, Hsieh MS, Liu YN, et al. Clinicopathological Features and Survival Outcomes of Primary Pulmonary Invasive Mucinous Adenocarcinoma. Cancers (Basel) 2021;13:4103.. 10.3390/cancers13164103 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Xiao Z, Chen J, Feng X, et al. Use of CT-derived radiomic features to preoperatively identify invasive mucinous adenocarcinoma in solitary pulmonary nodules ≤3 cm. Heliyon 2024;10:e30209.. 10.1016/j.heliyon.2024.e30209 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Zhang J, Hao L, Xu Q, et al. Radiomics and Clinical Characters Based Gaussian Naive Bayes (GNB) Model for Preoperative Differentiation of Pulmonary Pure Invasive Mucinous Adenocarcinoma From Mixed Mucinous Adenocarcinoma. Technol Cancer Res Treat 2024;23:15330338241258415.. 10.1177/15330338241258415 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Sun C, Sun J, Wei F, et al. Hybrid Deep Learning-Machine Learning Fusion of Clinical, Radiomic and Deep Learning Features for Preoperative Differentiation of Solitary Pulmonary Mucinous Adenocarcinoma. Diagnostics (Basel) 2026;16:1651.. 10.3390/diagnostics16111651 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Cui D, Xie S, Liu Q. Postoperative survival of pulmonary invasive mucinous adenocarcinoma versus non-mucinous invasive adenocarcinoma. BMC Pulm Med 2023;23:9.. 10.1186/s12890-023-02305-x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Hwang S, Han J, Choi M, et al. Size of Non-lepidic Invasive Pattern Predicts Recurrence in Pulmonary Mucinous Adenocarcinoma: Morphologic Analysis of 188 Resected Cases with Reappraisal of Invasion Criteria. J Pathol Transl Med 2017;51:56-68.. 10.4132/jptm.2016.09.17 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Kim M, Hwang J, Kim KA, et al. Genomic characteristics of invasive mucinous adenocarcinoma of the lung with multiple pulmonary sites of involvement. Mod Pathol 2022;35:202-9.. 10.1038/s41379-021-00872-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Xu X, Li N, Wang D, et al. Clinical Relevance of PD-L1 Expression and CD8+ T Cells' Infiltration in Patients With Lung Invasive Mucinous Adenocarcinoma. Front Oncol 2021;11:683432.. 10.3389/fonc.2021.683432 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Cai L, Wang J, Yan J, et al. Genomic Profiling and Prognostic Value Analysis of Genetic Alterations in Chinese Resected Lung Cancer With Invasive Mucinous Adenocarcinoma. Front Oncol 2020;10:603671.. 10.3389/fonc.2020.603671 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Demir T, Araz M, Moloney C, et al. Efficacy of Systemic Treatments in Patients With Metastatic Lung Invasive Mucinous Adenocarcinoma. Clin Lung Cancer 2024;25:e316-22.. 10.1016/j.cllc.2024.05.004 [DOI] [PubMed] [Google Scholar]
- 25. Hong R, Ping X, Liu Y, et al. Combined CT-Based Radiomics and Clinic-Radiological Characteristics for Preoperative Differentiation of Solitary-Type Invasive Mucinous and Non-Mucinous Lung Adenocarcinoma. Int J Gen Med 2024;17:4267-79.. 10.2147/IJGM.S479978 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Zhang J, Hao L, Li M, et al. CT Radiomics Combined With Clinicopathological Features to Predict Invasive Mucinous Adenocarcinoma in Patients With Lung Adenocarcinoma. Technol Cancer Res Treat 2023;22:15330338231174306.. 10.1177/15330338231174306 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Yuan J, Sun Y, Wang K, et al. Development and validation of reassigned CEA, CYFRA21-1 and NSE-based models for lung cancer diagnosis and prognosis prediction. BMC Cancer 2022;22:686.. 10.1186/s12885-022-09728-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Zhu WY, Li HF, Fang KX, et al. Epidermal Growth Factor Receptor Mutations and Their Prognostic Value with Carcinoembryonic Antigen in Pathological T1 Lung Adenocarcinoma. Dis Markers 2018;2018:2942618.. 10.1155/2018/2942618 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Hu LY, Zhuang WT, Chen MJ, et al. EGFR Oncogenic Mutations in NSCLC Impair Macrophage Phagocytosis and Mediate Innate Immune Evasion Through Up-Regulation of CD47. J Thorac Oncol 2024;19:1186-200.. 10.1016/j.jtho.2024.03.019 [DOI] [PubMed] [Google Scholar]
- 30. Salama MF, Liu M, Clarke CJ, et al. PKCα is required for Akt-mTORC1 activation in non-small cell lung carcinoma (NSCLC) with EGFR mutation. Oncogene 2019;38:7311-28.. 10.1038/s41388-019-0950-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Kamle S, Ma B, Schor G, et al. Chitinase 3-like-1 (CHI3L1) in the pathogenesis of epidermal growth factor receptor mutant non-small cell lung cancer. Transl Oncol 2024;49:102108.. 10.1016/j.tranon.2024.102108 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Deng C, Zhang Y, Ma Z, et al. Prognostic value of epidermal growth factor receptor gene mutation in resected lung adenocarcinoma. J Thorac Cardiovasc Surg 2021;162:664-674.e7.. 10.1016/j.jtcvs.2020.05.099 [DOI] [PubMed] [Google Scholar]
- 33. Yang G, Tian L, He D, et al. Aumolertinib in non-small cell lung cancer with uncommon EGFR exon 19 deletions: a real-world dual-center study. BMC Cancer 2026;26:767.. 10.1186/s12885-026-16119-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Zhang R, Yan H, Tian F, et al. Prognostic implications of uncommon EGFR exon 19 deletion-insertion mutations in non-small cell lung cancer treated with third-generation EGFR-TKIs. Lung Cancer 2025;208:108755.. 10.1016/j.lungcan.2025.108755 [DOI] [PubMed] [Google Scholar]
- 35. Guan M, Shi Q, Ye W, et al. Patients with Advanced Non-Small Cell Lung Cancer and the EGFR Exon 19 Deletion p.L747 Benefit from Chemotherapy and First-Generation Tyrosine Kinase Inhibitors Compared with Patients with p.E746. Onco Targets Ther 2025;18:1227-41.. 10.2147/OTT.S535531 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Chan SK, Choi HC, Lee VH. Overall Survival Benefits of First-Line Treatments for Asian Patients with Advanced Epidermal Growth Factor Receptor-Mutated NSCLC Harboring Exon 19 Deletion: A Systematic Review and Network Meta-Analysis. Cancers (Basel) 2022;14:3362.. 10.3390/cancers14143362 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37. Yang Z, Yang F, Li Z, et al. Sublobectomy versus lobectomy for peripheral small-sized pulmonary mucinous adenocarcinoma. J Thorac Dis 2025;17:390-9.. 10.21037/jtd-2024-2096 [DOI] [PMC free article] [PubMed] [Google Scholar]