Nomogram predicting overall and cancer specific prognosis for poorly differentiated lung adenocarcinoma after resection based on SEER cohort analysis

Abstract

The prognosis of poorly differentiated lung adenocarcinoma (PDLA) is determined by many clinicopathological factors. The aim of this study is identifying prognostic factors and developing reliable nomogram to predict the overall survival (OS) and cancer-specific survival (CSS) in patients with PDLA. Patient data from the Surveillance, Epidemiology and End Results (SEER) database was collected and analyzed. The SEER database was used to screen 1059 eligible patients as the study cohort. The whole cohort was randomly divided into a training cohort (n = 530) and a test cohort (n = 529). Cox proportional hazards analysis was used to identify variables and construct a nomogram based on the training cohort. C-index and calibration curves were performed to evaluate the performance of the model in the training cohort and test cohorts. For patients with PDLA, age at diagnosis, gender, tumor size were independent prognostic factors both for overall survival (OS) and cancer-specific survival (CSS), while race and number of nodes were specifically related to OS. The calibration curves presented excellent consistency between the actual and nomogram-predict survival probabilities in the training and test cohorts. The C-index values of the nomogram were 0.700 and 0.730 for OS and CSS, respectively. The novel nomogram provides new insights of the risk of each prognostic factor and can assist doctors in predicting the 1-year, 3-year and 5-year OS and CSS in patients with PDLA.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-024-73486-6.

Introduction

Lung cancer is the leading cause of cancer-related death worldwide, with adenocarcinoma being the most prevalent subtype¹. In the United States, the number of estimated new cases and deaths were 234,580 and 125,070, respectively, in 2024². In 2020, International Association for the Study of Lung Cancer (IASLC) has categorized invasive non-mucinous pulmonary adenocarcinoma into 3 grades, grade 1, grade 2 and grade 3, as defined by the combination of predominant subtype and percentage of high-grade patterns (solid, micropapillary, and/or complex glandular patterns) of growth³. However, for some rare histological types of lung adenocarcinoma, such as invasive mucinous adenocarcinoma, mixed invasive mucinous and non-mucinous adenocarcinoma, colloid adenocarcinoma, fetal adenocarcinoma, and enteric adenocarcinoma, this grading system does not provide detailed guidance. Patients with these histological subtypes often have a poorer prognosis^4–8. Due to the low incidence and the limited number of cases in single centers, there is currently a lack of effective prognostic evaluation systems in clinical practice. We collectively refer to these rare histological types with poorer prognosis, along with micropapillary and solid adenocarcinoma, as poorly differentiated lung adenocarcinoma. Patients with PDLA often present with advanced-stage disease at diagnosis, which contributes to the overall poor prognosis associated with this condition⁹.

Despite advancements in diagnostic techniques and therapeutic strategies, including targeted therapies and immunotherapy, the survival outcomes for patients with PDLA have not significantly improved compared to those with well-differentiated tumors^10,11. Several large validation studies have consistently demonstrated that poorly differentiated lung adenocarcinoma is linked to adverse prognosis, to higher risk of lymph node metastasis and local recurrence^12–14. This discrepancy underscores the need for a better understanding of the clinical and pathological factors that influence the prognosis of PDLA.

Prognostic models are essential tools in oncology, as they can aid clinicians in estimating patient outcomes and tailoring treatment approaches¹⁵. Nomograms, in particular, have emerged as user-friendly and simple visual instruments that integrate multiple prognostic factors into a single predictive model^16,17. By providing a visual representation of the risk of a specific outcome, nomograms can facilitate shared decision-making between clinicians and patients^18–20.

The Surveillance, Epidemiology, and End Results (SEER) program, a comprehensive cancer registry in the United States, offers a rich source of data for epidemiological research. It captures a wide range of information, including demographics, tumor characteristics, and survival data, which can be leveraged to develop and validate prognostic models for various cancer types, including PDLA. The SEER database has been widely used to construct prognostic models for various diseases. Previous studies have also utilized the SEER database to develop prognostic models for lung adenocarcinoma^21–23. However, there is still a lack of prognostic models specifically designed for poorly differentiated lung adenocarcinoma (PDLA). This absence results in a lack of specificity and accuracy in assessing the prognosis of PDLA patients.

In this study, we aimed to identify the key prognostic factors associated with PDLA and construct a nomogram that could predict overall survival (OS) and cancer-specific survival (CSS) for these patients. By leveraging the extensive dataset provided by the SEER program, we sought to develop a clinically relevant tool that could potentially guide treatment decisions and improve patient outcomes in the context of PDLA. Understanding the complex interplay of factors that contribute to the prognosis of PDLA is crucial for the development of more personalized and effective therapeutic strategies. Our nomogram, based on a comprehensive analysis of SEER data, represents a step forward in the quest to optimize the management of this aggressive form of lung cancer.

Materials and methods

Study Design

The present study is a retrospective cohort analysis aimed at identifying prognostic factors for lung adenocarcinoma and developing nomograms to predict OS and CSS based on the level of pathological differentiation. Data were extracted from the SEER database, encompassing the years 1975 to 2020 and including patients diagnosed with lung adenocarcinoma.

Data source and cohort selection

Eligible patients were identified from the SEER database, with the study cohort comprising individuals diagnosed with lung adenocarcinoma between 1975 and 2020 (http://seer.cancer.gov/). The inclusion criteria were as follows: patients who underwent surgical treatment, aged 18 years or older, with a diagnosis confirmed by histopathological report, and International Classification of Diseases for Oncology, Third Edition (ICD-O-3) codes indicating poorly differentiated lung adenocarcinoma as their first and primary cancer. ICD-O-3 were employed to identify the following subtypes: micropapillary adenocarcinoma (ICD-O-3 code 8265), solid adenocarcinoma (ICD-O-3 code 8230), invasive mucinous adenocarcinoma (ICD-O-3 code 8253), mixed invasive mucinous and non-mucinous adenocarcinoma (ICD-O-3 code 8254), colloid adenocarcinoma (ICD-O-3 code 8480), fetal adenocarcinoma (ICD-O-3 code 8333), and enteric adenocarcinoma (ICD-O-3 code 8144). Exclusion criteria included cases with missing values for key variables such as age, sex, race, marital status, laterality, grade, tumor size, number of positive regional nodes, stage, chemotherapy, and radiotherapy (Fig. 1). A total of 1059 patients ranged from 2004 to 2017 were retained after exclusion criteria. Staging system was translated to the 8th Edition of the American Joint Committee on Cancer (AJCC) Staging manual.

Fig. 1.

Poorly differentiated lung adenocarcinoma cohort selection.

Statistical analysis

Descriptive statistics were used to summarize the demographic and clinical characteristics of the study population. The One-Sample Kolmogorov-Smirnov Test (1 sample K-S Test) was utilized to assess the normality of continuous variables, which were presented as mean ± standard deviation for normally distributed data or median (interquartile range) for non-normally distributed data. Categorical data were expressed as frequencies and percentages (n (%)). Differences between groups were compared using independent sample t-tests for normally distributed continuous variables, Mann-Whitney U Tests for non-normally distributed continuous variables, and chi-square tests for categorical variables.

The study population was randomly divided into a training set (50% of cases) and a test set (remaining 50%) using R’s random sampling function. The baseline characteristics, including Age, Race, Gender, Marital status, Laterality, Stage, Tumor size, Number of positive regional nodes, Chemotherapy, Radiotherapy and Surgery, were compared between the two sets to ensure no significant differences existed.

Cox proportional hazards regression analysis was performed on the training set to identify factors associated with OS and CSS. Both univariate and multivariate models were used, with factors entered into the multivariate model using the Enter method. The impact of these factors on OS and CSS was assessed using hazard ratios (HR) and 95% confidence intervals (CI), with a p-value < 0.05 considered statistically significant. The Kaplan-Meier estimator was used to calculate the overall survival (OS) and cancer-specific survival (CSS) probabilities among different surgery options, along with their corresponding 95% confidence intervals (CIs). Differences between survival curves were evaluated using the Log-rank test, with a significance level set at P < 0.05.

A nomogram was constructed based on the significant factors identified in the multivariate analysis to predict 1-, 3-, and 5-year OS and CSS. The discrimination of the model was evaluated using the Harrell C-index, and the calibration was assessed using calibration curves. The diagnostic value of the nomogram was further evaluated using the area under the receiver operating characteristic curve (AUC) from the test set.

Software and statistical significance

All statistical analyses were performed using R software version 4.2.2. A two-sided p-value of less than 0.05 was considered to indicate statistical significance.

Ethical considerations

This study was reviewed and deemed exempt by the Institutional Review Board due to the retrospective nature of the study and the use of de-identified data from the SEER database. No patient consent was required as per the board’s guidelines.

Results

Baseline characteristics

The study population was randomly divided into a training set (n = 530) and a test set (n = 529) using R’s random sampling function. As shown in Table 1, the baseline characteristics, including Age, Tumor size, Gender, Race, Marital status, Laterality, Stage, Number of positive regional nodes, Radiotherapy, Chemotherapy and Surgery, were compared between the two sets. No significant differences were observed between the training and test sets for any of the factors, indicating successful randomization and comparability of the groups.

Table 1.

Comparison of baseline information between training set and test set.

Factors	Training set (n = 530)	Test set (n = 529)	t/Z/χ2	P value
Age (year)	67.60 ± 10.74	66.87 ± 10.62	1.124	0.261
Tumor size (mm)	28.5 (17.0, 47.3)	28.0 (18.0, 45.0)	− 0.022	0.982
Tumor size, n (%)			0.145	0.930
1–20 mm	180 (34.0)	180 (34.0)
21–50 mm	236 (44.5)	240 (45.4)
>50 mm	114 (21.5)	109 (20.6)
Gender, n (%)			3.093	0.079
Male	222 (41.9)	250 (47.3)
Female	308 (58.1)	279 (52.7)
Race, n (%)			0.108	0.742
White	432 (81.5)	427 (80.7)
Others	98 (18.5)	102 (19.3)
Marital status			3.309	0.069
Unmarried	212 (40.0)	183 (34.6)
Married	318 (60.0)	346 (65.4)
Laterality, n (%)			0.599	0.439
Left	215(40.6)	227 (42.9)
Right	315 (59.4)	302 (57.1)
Stage, n (%)			0.531	0.912
I	369 (69.6)	377 (71.3)
II	77 (14.5)	69 (13.0)
III	66 (12.5)	65 (12.3)
IV	18 (3.4)	18 (3.4)
Number of positive regional nodes			0.873	0.646
0	448 (84.5)	447 (84.5)
1–3	61 (11.5)	66 (12.5)
>3	21 (4.0)	16 (3.0)
Radiotherapy, n (%)			1.385	0.239
No	503 (94.9)	493 (93.2)
Yes	27 (5.1)	36 (6.8)
Chemotherapy, n (%)			3.794	0.051
No/Unknown	414 (78.5)	386 (73.0)
Yes	116 (21.9)	143 (27.0)
Surgery, n (%)			1.545	0.462
Lobectomy	451 (85.1)	437 (82.6)
Limited resection	74 (14.0)	84 (15.9)
No surgery	5 (0.9)	8 (1.5)

Note: χ2, the statistical value of the Chi-square test; Z, the statistical value of the Mann-Whitney Test.

Survival analysis

We categorized all patients into three groups according to surgical approach: no surgery, limited resection, and lobectomy. The limited resection group includes both wedge resection and segmental resection. We plotted the survival curves for overall survival (OS) and cancer-specific survival (CSS) among different surgical approaches. For OS, the median survival times for the Lobectomy group, Limited Resection group, and No Surgery group were 93.0 (78.7, 107.3) months, 87.0 (63.4, 110.6) months, and 13.0 (2.3, 23.7) months, respectively. The differences among the three groups were significant (P = 0.02). However, there was no statistical difference between the Lobectomy and Limited Resection groups (HR: 1.143, 95% CI: 0.810–1.615, P = 0.447) (Fig. S1A). For CSS, the median survival times for the Lobectomy group, Limited Resection group, and No Surgery group were 185.0 (138.1, 231.9) months, 126.0 (83.9, 168.1) months, and 13.0 (2.3, 23.7) months, respectively. The differences among the three groups were significant (P = 0.0013). No statistical difference was found between the Lobectomy and Limited Resection groups (HR: 1.038, 95% CI: 0.661–1.630, P = 0.447) (Fig. S1B).

Model Development and discrimination

The nomogram models for predicting 1-, 3-, and 5-year overall survival (OS) and cancer-specific survival (CSS) were developed using the training set (Fig. 2A and B). The models incorporated significant prognostic factors identified through univariate and multivariate Cox regression analysis. These factors included age, gender, tumor size, race, and the number of positive regional nodes for OS (Table 2). While age, tumor size, stage, and gender were significant for CSS (Table 3).

Fig. 2.

A Nomogram Model’s 1/3/5-Year OS Prediction Results. B Nomogram Model’s 1/3/5-Year CSS Prediction Results.

Table 2.

Univariate and multivariate Cox regression analysis of OS influencing factors.

Characteristics	HR (95% CI)	P value	HR (95% CI)	P value
Age (year)	1.030 (1.018, 1.043)	< 0.001	1.027 (1.014, 1.040)	< 0.001
Tumor size, n (%)		< 0.001		< 0.001
1–20 mm	Reference		Reference
21–50 mm	1.670 (1.226, 2.274)	0.001	1.511 (1.103, 2.070)	0.010
> 50 mm	3.213 (2.288, 4.512)	< 0.001	2.836 (1.980, 4.062)	< 0.001
Gender, n (%)
Male	Reference		Reference
Female	0.683 (0.535, 0.872)	0.002	0.742 (0.575, 0.956)	0.021
Race, n (%)
Others	Reference		Reference
White	1.821 (1.257, 2.638)	0.002	1.932 (1.314, 2.841)	0.001
Marital status
Unmarried	Reference
Married	1.182 (0.915, 1.526)	0.201
Laterality, n (%)
Left	Reference
Right	0.895 (0.699, 1.147)	0.382
Stage, n (%)		< 0.001		0.064
I	Reference		Reference
II	1.883 (1.313, 2.701)	0.001	1.087 (0.689, 1.714)	0.721
III	2.116 (1.509, 2.966)	< 0.001	1.349 (0.875, 2.082)	0.176
IV	2.901 (1.639, 5.132)	< 0.001	2.418 (1.228, 4.761)	0.011
Number of positive regional nodes		< 0.001		0.001
0	Reference		Reference
1–3	2.664 (1.902, 3.732)	< 0.001	2.319 (1.447, 3.716)	< 0.001
> 3	2.767 (1.570, 4.877)	< 0.001	2.720 (1.367, 5.410)	0.004
Radiotherapy, n (%)
No	Reference		Reference
Yes	2.164 (1.301, 3.601)	0.003	1.033 (0.565, 1.889)	0.916
Chemotherapy, n (%)
No/Unknown	Reference		Reference
Yes	1.389 (1.042, 1.852)	0.025	0.834 (0.559, 1.242)	0.371
Surgery, n (%)		0.033		0.063
Lobectomy	Reference		Reference
Limited resection	1.143 (0.810, 1.615)	0.447	1.360 (0.950, 1.947)	0.093
No surgery	3.622 (1.343, 9.766)	0.011	2.556 (0.871, 7.503)	0.088

Table 3.

Univariate and multivariate Cox regression analysis of CSS influencing factors.

Characteristics	HR (95% CI)	P value	HR (95% CI)	P value
Age (year)	1.018 (1.003, 1.033)	0.021	1.020 (1.004, 1.036)	0.014
Tumor size, n (%)		< 0.001		< 0.001
1–20 mm	Reference		Reference
21–50 mm	1.829 (1.202, 2.783)	0.005	1.715 (1.126, 2.613)	0.012
> 50 mm	4.351 (2.813, 6.732)	< 0.001	3.812 (2.415, 6.017)	< 0.001
Gender, n (%)
Male	Reference
Female	0.668 (0.489, 0.912)	0.011	0.706 (0.510, 0.977)	0.036
Race, n (%)
Others	Reference
White	1.522 (0.978, 2.368)	0.063
Marital status
Unmarried	Reference
Married	1.245 (0.899, 1.724)	0.188
Laterality, n (%)
Left	Reference
Right	0.933 (0.680, 1.279)	0.665
Stage, n (%)		< 0.001		0.002
I	Reference		Reference
II	2.275 (1.465, 3.533)	< 0.001	1.118 (0.655, 1.910)	0.682
III	3.278 (2.215, 4.853)	< 0.001	1.974 (1.199, 3.248)	0.007
IV	5.107 (2.761, 9.446)	< 0.001	3.646 (1.684, 7.896)	0.001
Number of positive regional nodes		< 0.001		0.101
0	Reference		Reference
1–3	3.202 (2.155, 4.757)	< 0.001	1.735 (1.023, 2.943)	0.041
> 3	3.210 (1.674, 6.155)	< 0.001	1.743 (0.801, 3.794)	0.161
Radiotherapy, n (%)
No	Reference		Reference
Yes	2.931 (1.687, 5.090)	< 0.001	0.922 (0.474, 1.797)	0.812
Chemotherapy, n (%)
No/Unknown	Reference		Reference
Yes	2.248 (1.618, 3.123)	< 0.001	1.185 (0.750, 1.872)	0.466
Surgery, n (%)		0.005		0.105
Lobectomy	Reference		Reference
Limited resection	1.038 (0.661, 1.630)	0.872	1.457 (0.911, 2.331)	0.116
No surgery	5.220 (1.925, 14.154)	0.001	2.343 (0.780, 7.042)	0.129

The discrimination of the models was assessed using the Harrell’s C-index, which yielded values of 0.700 (95% CI: 0.667–0.733) for OS and 0.730 (95% CI: 0.689–0.771) for CSS, indicating good predictive accuracy.

Calibration and validation

Calibration curves were used to assess the agreement between the nomogram-predicted probabilities and the actual observed outcomes for OS and CSS at 1-, 3-, and 5-year intervals. (Figures 3 and 4) The curves demonstrated close alignment between predicted and actual probabilities, suggesting that the models were well-calibrated.

Fig. 3.

Calibration curves for predictions for the 1-year (a), 3-year (b), 5-year (c) OS in the training cohort (A) and in the testing cohort (B). The nomogram-predicted probability of OS is plotted on the X-axis, and the actual OS is plotted on the Y- axis. OS, overall survival.

Fig. 4.

Calibration curves for predictions for the 1-year (a), 3-year (b), 5-year (c) CSS in the training cohort (A) and in the testing cohort (B). The nomogram-predicted probability of CSS is plotted on the X-axis, and the actual OS is plotted on the Y- axis. CSS, cancer-specific survival.

The predictive models were further validated using the test set. The calibration curves for the test set were consistent with those of the training set, confirming the generalizability of the models.

ROC Analysis

The diagnostic value of the nomogram models was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC). For OS, the AUC values were 0.755, 0.759, and 0.742 for the training set, and 0.763, 0.724, and 0.696 for the test set at 1-, 3-, and 5-year intervals, respectively (Fig. 5). For CSS, the AUC values were 0.758, 0.778, and 0.753 for the training set, and 0.781, 0.725, and 0.706 for the test set at 1-, 3-, and 5-year intervals, respectively (Fig. 6). These AUC values indicate a high diagnostic performance of the models.

Fig. 5.

The AUC of the ROC curve for the 1-year (a), 3-year (b), 5-year (c) OS in the training cohort (A) and in the testing cohort (B).

Fig. 6.

The AUC of the ROC curve for the 1-year (a), 3-year (b), 5-year (c) CSS in the training cohort (A) and in the testing cohort (B).

In conclusion, the nomogram models developed in this study provide a reliable and accurate tool for predicting the survival outcomes of patients with poorly differentiated lung adenocarcinoma. The models demonstrated good discrimination, calibration, and diagnostic value, making them a valuable asset for clinical decision-making and patient prognosis communication.

Discussion

Poorly differentiated lung adenocarcinoma is a subtype of non-small cell lung cancer (NSCLC) characterized by aggressive behavior and poor clinical outcomes⁵. The prognosis of PDLA is influenced by a multitude of factors^2,24–27 Despite the International Association for the Study of Lung Cancer (IASLC) pathology committee proposing a new grading system for lung adenocarcinoma in 2020, it does not provide clear guidelines for some rare and poorly prognostic subtypes of lung adenocarcinoma. Currently, there is also a lack of effective prognostic models for patients with these poorly differentiated lung adenocarcinomas³. In this study, we aimed to develop a nomogram to predict the overall prognosis for patients diagnosed with poorly differentiated lung adenocarcinoma using a large-scale SEER cohort.

The development of our nomogram for predicting the overall prognosis in poorly differentiated lung adenocarcinoma is grounded in a methodical analysis of a SEER cohort. By identifying key prognostic factors such as age, tumor size, gender, and race, we have created a tool that not only predicts survival probabilities but also elucidates the relative impact of each factor on patient outcomes¹⁶. Compared to previous studies that have developed prognostic prediction models for lung adenocarcinoma using the SEER database, we have developed a unified prognostic prediction model specifically targeting poorly differentiated adenocarcinomas. This includes micropapillary adenocarcinoma, solid adenocarcinoma, invasive mucinous adenocarcinoma, mixed invasive mucinous and non-mucinous adenocarcinoma, colloid adenocarcinoma, fetal adenocarcinoma, and enteric adenocarcinoma. Given the small number of cases for each subtype, developing separate prognostic models for each would likely reduce the accuracy of the models and complicate their clinical application^28–30, Conversely, constructing a unified prognostic model for all lung adenocarcinoma patients, regardless of subtype, could lead to an inappropriate overestimation of the prognosis for patients with PDLA^31–33. Our model facilitates more accurate and streamlined assessment of these patients in clinical practice. This approach avoids the bias that can arise from small sample sizes, enhances the ease of use of the model, and improves the accuracy of prognostic predictions by considering the distinct pathological features of this subgroup. Additionally, we have incorporated staging into the construction of the prediction model, which enhances its applicability compared to previous studies that focused solely on a single stage, thereby making the model relevant for a broader range of patients.

Our analysis identified several independent prognostic factors that significantly influence the overall survival (OS) of patients with poorly differentiated lung adenocarcinoma. These factors include age、gender、tumor size、race、number of positive regional nodes. The nomogram integrates these factors to provide a personalized estimate of OS at 1-, 3-, and 5-year intervals.

The C-index values for both OS and CSS models indicate that the nomograms have great discrimination, with scores above 0.7, which is considered acceptable for clinical use. This suggests that the models can effectively separate patients with different risks of mortality due to poorly differentiated lung adenocarcinoma. The calibration curves showed a high level of agreement between the predicted probabilities and actual outcomes, both in the training and test sets. This close correspondence indicates that the models are reliable and can be used with confidence to predict survival probabilities in individual patients. The ROC analysis further confirmed the diagnostic value of the models, with AUC values generally above 0.7 for both OS and CSS, indicating that the models have a good ability to discriminate between patients who will survive and those who will not.

The nomogram’s predictive accuracy, as evidenced by the C-index and calibration curves. However, the slightly higher actual survival rates observed in the test cohort compared to the nomogram predictions for 1-year CSS suggest that there may be other factors at play that were not captured in our model. These could include advancements in treatment modalities, emerging targeted therapies, or the influence of patient behaviors such as smoking cessation.

The significance of age as a prognostic factor aligns with the biological understanding that older patients may experience more aggressive disease progression and have a higher risk of comorbidities, which can complicate treatment and affect survival^34,35. The positive correlation between age and the risk of death in this study suggests that elderly patients may face poorer prognoses due to declines in physiological function and reduced tolerance to treatment³⁶. This finding underscores the need for more nuanced assessments in treatment decisions for elderly patients to ensure the safety and efficacy of treatment plans³⁷ .

Tumor size, a classic indicator of cancer prognosis, was once again confirmed as important in this study. An increase in tumor volume was closely associated with a reduction in survival time, which may be related to the biological behavior, aggressiveness, and treatment response of the tumor. The influence of tumor size on prognosis underscores the importance of early detection and intervention, as larger tumors are often associated with a more advanced stage at diagnosis and a poorer prognosis³⁸. Therefore, early detection and treatment are crucial for improving patient outcomes.

The role of gender differences in cancer prognosis has long been an area of interest. The gender disparity observed, with female patients showing a better prognosis, may be attributed to biological differences, hormonal factors, or variations in tumor biology. This finding warrants further investigation to understand the underlying mechanisms and to explore potential gender-specific therapeutic strategies^26,39–41.

While racial differences were observed for OS, these effects did not remain significant for CSS. This result indicates that racial factors may be confounded by other socio-economic and treatment-related factors. The impact of race on survival outcomes is a complex issue that may reflect socioeconomic disparities, access to healthcare, and genetic factors². The higher hazard ratio observed for non-white patients suggests a need for targeted interventions to address these disparities and improve outcomes across diverse populations.

Our analysis also revealed that patients with a greater number of positive regional nodes at diagnosis have a significantly worse prognosis. This underscores the importance of accurate staging and the potential role of regional lymph node assessment in treatment planning.

Different surgical approaches result in varying prognoses for lung cancer patients. Research by Dai et al. demonstrated that for patients with NSCLC ≤ 1 cm and > 1 to 2 cm, lobectomy provides better outcomes compared to sublobar resection⁴². Conversely, a randomized controlled trial showed that for peripheral small lung cancers with a consolidation-to-tumor ratio (CTR) > 0.5 and ≤ 2 cm, segmentectomy offers similar recurrence-free survival (RFS) but superior overall survival (OS) compared to lobectomy⁴³. In our study, we found that surgical treatment is more effective than non-surgical treatment for poorly differentiated adenocarcinoma. However, there is no significant difference in OS and cancer-specific survival (CSS) between lobectomy and limited resection. This finding is consistent with previous research on the prognostic impact of different surgical methods in invasive lung adenocarcinomas with predominant micropapillary and solid components⁴⁴. The high likelihood of early distant metastasis in poorly differentiated adenocarcinoma may explain why extending the surgical resection does not significantly improve patient outcomes. Further studies with larger sample sizes and longer follow-up periods are needed to confirm these findings.

The International Association for the Study of Lung Cancer (IASLC) recently published the ninth edition of the TNM classification for lung cancer⁴⁵. The 9th edition TNM staging system recognizes the prognostic significance of tumor size and the number of involved lymph nodes, which aligns with our nomogram’s emphasis on these factors. Pathological and molecular subtyping were not included in the 9th edition TNM staging system. However, a large number of studies have shown that pathological subtypes are closely related to the prognosis and the choice of surgical methods for patients with lung adenocarcinoma^44,46. Our analysis, which identified age, gender, tumor size, race, and the number of positive regional nodes as significant prognostic factors, complements the 9th edition TNM staging system by providing a personalized risk assessment tool. Our study noted some key differences, although the TNM staging system emphasizes the importance of tumor size and lymph node status, our analysis indicates that a patient’s race and gender may also have a significant impact on prognosis¹⁶.

Additionally, the system now accounts for the presence of pleural or pericardial metastasis, which was not explicitly included in our nomogram but is crucial for accurate staging and prognosis.

It is important to note that the models developed in this study are based on a large, population-based dataset from the SEER program, which enhances the generalizability of the findings. However, the retrospective nature of the SEER database, which may be subject to selection bias and incomplete data. Additionally, the lack of molecular data limits our understanding of the biological heterogeneity of lung adenocarcinoma, which could influence prognosis and response to therapy. And the lack of detailed information on certain clinical variables such as the specifics of systemic therapy and the exact histological subtypes of adenocarcinoma, which also could potentially influence prognosis. The proportion of different pathological subtypes also has an impact on the prognosis of patients with lung adenocarcinoma⁴⁷. Detailed information is not provided in the SEER database. Due to the low incidence and limited number of cases in single centers, we did not use external data to validate the model’s generalizability. Future studies involving multicenter and large sample sizes are needed to further verify the reliability and broad applicability of this prognostic model.

In summary, this study has successfully developed and validated nomograms for predicting OS and CSS in patients with poorly differentiated lung adenocarcinoma, which provides a valuable predictive tool for those patients. These nomograms, with their demonstrated accuracy and reliability, have the potential to aid clinicians in prognostic estimation and decision-making regarding treatment strategies. The identification of key prognostic factors highlights areas for further research and potential targets for intervention. Future studies should incorporate molecular data and explore the role of emerging therapies to refine prognostic models and improve patient outcomes. The integration of molecular profiling and biomarker information into the nomogram could enhance its predictive power, as these factors are increasingly recognized as important determinants of patient outcomes in the era of precision medicine.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

WS and JS contributed equally to this article. All authors contributed to the study conception and design. Data collection: WS, JS; Data analysis: BZ, XM, ML; Manuscript writing: WS, JS; Study design: YG, WS. All authors read and approved the final manuscript.

Funding

This work was supported by National Key R&D Program of China (Grant no. 2020YFE02022200).

Data availability

The datasets generated and analyzed during the current study are available in the SEER database. (http://seer.cancer.gov/).

Declarations

Competing interests

The authors have no relevant financial or non-financial interests to disclose.