Lung cancer is a group of clinically and histologically heterogeneous diseases and is the leading cause of cancer-related mortality worldwide [
Visceral pleural invasion (VPI) has been identified as a poor prognostic factor in NSCLC and was first adopted as a
We must assess clinicopathological prognosis factors, including histological classification, tumor differentiation, tumor size, and so on, to estimate the accurate prognosis and deliver personalized treatment strategies for NSCLC patients. How to determine the subdivision level of VPI has some limitations in clinical application, which is usually due to being diagnosed by imaging features and clinical judgment rather than pathological diagnosis. VPI in NSCLC patients is usually determined by the tumor site, radiographic findings, and other clinical characteristics; therefore, patients have variable clinical outcomes. Moreover, although VPI has been incorporated into the existing TNM staging system of AJCC, the relationship between the subdivision of VPI and prognosis in the NSCLC patients continues to have a great deal of ambiguity. Based on these reasons, we are proposing the use of additional prognostic models to evaluate the survival outcomes of these patients.
Nomography, as a prognostic tool, can integrate predictive factors to establish a statistical model and has been extensively applied to predict the survival outcomes of various cancer patients [
The SEER database, which has primary sources of population-based cancer statistics in the USA, capturing approximately 97% of cancer incidence and covering about 28% of the USA population in 17 SEER registries, is maintained by the National Cancer Institute (NCI) [
The study variables were recorded as follows: gender (female and male), race (black, white, and others),
We obtained permission to retrieve the SEER research data files with the reference number 16828-Nov2017. The research data used in this study did not involve the subjects or individual identification information. Therefore, this study did not require informed consent and ethical approval.
We used R software version 3.5.1 to perform statistical analysis and generate graphics (R Foundation for Statistical Computing, Vienna, Austria). The continuous variables were converted to categorical variables and calculated using the Chi-squared test. The survival curve for OS was displayed using the Kaplan–Meier method, and differences were tested using a log-rank test stratified based on the prognostic factors. The Cox proportional hazards multivariate regression was used to analyze further the variables with
The graphical nomogram was obtained using the logistic regression model from the training group using the R package rms. The maximum points for each variable were set to 100. The discrimination and predictive powers of the nomogram model were evaluated using the concordance index (C-index) [
This study initially included a total of 87,045 lung cancer patients from the SEER research database from 2010 to 2015 that subsequently was randomly divided into either the training group (
The flowchart of patients included and grouped.
Characteristics of NSCLC patients with the different subdivision levels of VPI.
Variables | All patients | Training cohort | Validation cohort |
---|---|---|---|
Female | 42,199 (48.5) | 29,550 (48.5) | 12,649 (48.4) |
Male | 44,846 (51.5) | 31,383 (51.5) | 13,463 (51.6) |
<60 | 15,510 (17.8) | 10,813 (17.7) | 4,697 (18.0) |
≥60 | 71,535 (82.2) | 50,120 (82.3) | 21,415 (82.0) |
Black | 9,583 (11.0) | 6,758 (11.1) | 2,825 (10.8) |
White | 71,259 (81.9) | 49,896 (81.9) | 21,363 (81.8) |
Others | 6,203 (7.1) | 4,279 (7.0) | 1,924 (7.4) |
Married | 46,595 (53.5) | 32,584 (53.5) | 14,011 (53.7) |
Divorced | 10,766 (12.4) | 7,584 (12.4) | 3,182 (12.1) |
Single | 11,420 (13.1) | 7,946 (13.0) | 3,474 (13.3) |
Widowed | 14,727 (16.9) | 10,330 (17.0) | 4,397 (16.8) |
Unknown | 3,537 (4.1) | 2,489 (4.1) | 1,048 (4.0) |
AC | 49,530 (56.9) | 34,766 (57.1) | 14,764 (56.5) |
BAC | 5,366 (6.1) | 3,736 (6.1) | 1,630 (6.2) |
LCC | 1,710 (2.0) | 1,202 (2.0) | 508 (2.0) |
SC | 30,439 (35.0) | 21,229 (34.8) | 9,210 (35.3) |
I | 11,539 (13.3) | 8,065 (13.2) | 3,474 (13.3) |
II | 35,660 (41.0) | 24,918 (40.9) | 10,742 (41.1) |
III | 38,508 (44.2) | 26,979 (44.3) | 11,529 (44.2) |
IV | 1,338 (1.5) | 971 (1.6) | 367 (1.4) |
| 29,605 (34.0) | 20,790 (34.1) | 8,815 (33.76) |
| 25,870 (29.7) | 18,047 (29.6) | 7,823 (29.96) |
| 16,523 (19.0) | 11,607 (19.1) | 4,916 (18.83) |
| 14,972 (17.2) | 10,432 (17.1) | 4,540 (17.39) |
| 50,120 (57.6) | 35,040 (57.5) | 15,080 (57.8) |
| 8,122 (9.4) | 5,736 (9.4) | 2,476 (9.5) |
| 21,968 (25.2) | 15,446 (25.4) | 6,522 (25.0) |
| 6,745 (7.8) | 4,711 (7.7) | 2,034 (7.7) |
| 63,034 (72.4) | 44,062 (72.3) | 18,972 (72.7) |
| 24,011 (27.6) | 16,871 (27.7) | 7,140 (27.3) |
PLX | 45,747 (52.6) | 31,976 (52.5) | 13,771 (52.7) |
| 34,690 (39.9) | 24,389 (40.0) | 10,301 (39.4) |
| 2,977 (3.4) | 2,064 (3.4) | 913 (3.5) |
| 2,529 (2.9) | 1,763 (2.9) | 766 (3.0) |
| 1,102 (1.3) | 741 (1.2) | 361 (1.4) |
Note: AC: adenocarcinoma; BAC: bronchioalveolar carcinoma; LCC: large-cell carcinoma; SC: squamous cell carcinoma.
The variables of the training group including gender, age, race, marital status, histology classification, tumor grade, TNM stage, and the subdivision of the VPI, were incorporated into the univariate analysis. All selected variables with a
Univariate and multivariate Cox regression analysis of the training group.
Variable | Univariate analysis | Multivariate analysis | |
---|---|---|---|
HR (95% CI) | |||
<0.001 | |||
Female | Reference | ||
Male | 1.370 (1.332–1.410) | <0.001 | |
<0.001 | |||
<60 | Reference | ||
≥60 | 1.374 (1.324–1.427) | <0.001 | |
<0.001 | |||
Black | Reference | ||
White | 1.039 (0.997–1.084) | 0.0708 | |
Other | 0.820 (0.767–0.877) | <0.001 | |
<0.001 | |||
Married | Reference | ||
Divorced | 1.203 (1.153–1.255) | <0.001 | |
Single | 1.172 (1.124–1.223) | <0.001 | |
Widowed | 1.371 (1.320–1.424) | <0.001 | |
Unknown | 1.084 (1.008–1.165) | 0.0287 | |
<0.001 | |||
AC | Reference | ||
BAC | 0.702 (0.649–0.759) | <0.001 | |
LCC | 1.388 (1.262–1.528) | <0.001 | |
SC | 1.193 (1.158–1.228) | <0.001 | |
<0.001 | |||
I | Reference | ||
II | 1.204 (1.141–1.271) | <0.001 | |
III | 1.467 (1.390–1.547) | <0.001 | |
IV | 1.453 (1.293–1.634) | <0.001 | |
<0.001 | |||
| Reference | ||
| 1.395 (1.339–1.454) | <0.001 | |
| 1.642 (1.571–1.717) | <0.001 | |
| 1.745 (1.667–1.826) | <0.001 | |
<0.001 | |||
Reference | |||
1.278 (1.228–1.330) | <0.001 | ||
1.515 (1.472–1.559) | <0.001 | ||
1.453 (1.395–1.514) | <0.001 | ||
<0.001 | |||
| Reference | ||
| 2.427 (2.361–2.494) | <0.001 | |
<0.001 | |||
PLX | Reference | ||
| 0.499 (0.484–0.515) | <0.001 | |
| 0.465 (0.430–0.503) | <0.001 | |
| 0.478 (0.441–0.519) | <0.001 | |
| 0.605 (0.546–0.670) | <0.001 |
Based on these results, these factors were determined to be independent prognostic factors and then incorporated into the construction of the nomogram predicting 3- and 5-year overall survival (OS) in the training group (Figure
Nomogram for survival predicted probability. Each variable was individually evaluated for every patient and given a score subsequently. The higher scores obtained by summing the points of each variable from different patients indicate the poorer survival probability. Abbreviations: BAC, bronchioalveolar carcinoma; AC, adenocarcinoma; LCC, large-cell carcinoma; and SC, squamous cell carcinoma.
Calibration plots of nomogram in both groups. (a and b). The nomogram was calibrated in the training group by predicting the 3- and 5-year survival. (c and d). The nomogram was also calibrated in the validation group by predicting the 3- and 5-year survival. All results showed better fitting effects.
Kaplan–Meier curve analysis in both groups. (a) Risk classification of patient survival with the different degrees of VPI in the training group (log-rank
The incidence rate of VPI accounts for approximately 11.5% of NSCLC patients and varies between different histological types [
Nomograms are prognostic tools that make complex statistical models simpler with terse diagrams, which supply more exact and understandable prognosis predicting results and are widely applied in clinical practice [
Many clinical studies have shown that NSCLC patients with VPI have poorer survival as compared with those without. The possible factors for poor prognosis are associated with higher involvement in mediastinal lymph nodes [
There are several limitations in this population-based study. All of the data from the SEER database, which has a retrospective nature, contain bias that should be taken into consideration. First, the different medical agencies and professionals, including pathologists, surgeons, and so on, influence the detection rates and the quality of pleural invasion. Moreover, the SEER database fails to provide other information regarding NSCLC patients with VPI including the method of detection, complications, pulmonary function, and treatment method, which could generate underlying bias as well. Thus, it is useful to improve the accuracy of predictive models by incorporating novel predictors and introducing competing risk models. Moreover, the inclusion of the degree of VPI of early-stage NSCLC patients might also be a good candidate to control the confounding factors. Second, it is a common issue that improved accuracy of the predictive model is usually accompanied by a compromise between the increasing complexity of predictive factors and the decreasing understandability of the model during the modeling process of the nomogram. Considering the aforementioned, variables of clinical importance and high repeatable practicability would be preferred. Moreover, the nomogram itself needs to be confirmed using calibration plots and the C-index due to its uncertainty. Nonetheless, the nomogram is a powerful supplement for clinician judgment and clinical decision-making. Third, occult pleural metastases, which cannot be assessed by routine pathological examination, can only be detected during thoracotomy. Considering this study’s retrospective nature, the selection bias could not be avoided. In this study, whether patients diagnosed as PL0 actually have occult pleural metastases is still unclear. Besides this, patients who displayed no signs of pleural invasion based on clinical and/or radiographic judgment cannot exclude the possibility of occult micrometastases. Risk factors of these patients were relatively higher than those diagnosed with VPI. Fourth, based on the SEER database, we randomly divided the data into the training and the validation groups at the ratio of 7:3. This method of nomogram construction and validation is common, whereas further external verification was not available. Hence, we will focus on the data from multiple medical centers in further research to perform the validation of the external cohort.
We developed and validated a population-based nomogram model to predict survival differences among the different subdivision levels of VPI in NSCLC patients. This study provides a novel perspective that helps clinicians determine survival prognosis and establish personalized treatment strategies for NSCLC patients with varying degrees of VPI, which is an effective supplement to traditional TNM staging.
The data used to support the findings of this study are available from the corresponding author upon request.
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. SEER database is the leading cancer statistics database in the United States, and it is globally open and shared. The authors obtained the data by applying to the authorities.
The authors declare that they have no conflicts of interest.
Conceptualization was performed by Fan Wang. Methodology was developed by Fan Wang and Pei Li. Software was provided by Fan Wang. Validation was performed by Fan Wang and Pei Li. Formal Analysis was performed by Fan Wang. Investigation was done by Fan Wang and Pei Li. Original draft was written by Fan Wang. Reviewing and editing were performed by Fengsen Li. Supervision was done by Fengsen Li.
This research received a specific grant from the Natural Science Foundation of Xinjiang Uygur Autonomous Region (no. 2019D01A06) and Special Project of Xinjiang Public Health Key Technology Research and Development and Epidemic Prevention System Construction (no. 2020A03004).