A predictive model of lymph node metastasis for thymic epithelial tumours

Abstract

Open in new tabDownload slide

OBJECTIVES

Thymic epithelial tumours (TETs) are relatively rare indolent malignancies in the mediastinum. Lymph node metastasis (LNM) is an important prognostic indicator for TETs; however, the pattern of LNM involved in TETs has yet to be elucidated.

METHODS

Patients diagnosed with histologically confirmed thymoma (A–B3), thymic carcinomas and thymic neuroendocrine tumours, between 1988 and 2016 were identified from the Surveillance, Epidemiology, and End Results database. Univariable and multivariable logistic regression analyses were applied to identify the predictors for LNM. The predictive nomogram was built from the independent risk factors and measured using the concordance statistic.

RESULTS

The overall proportion of TETs with LNM was 18.5% (200/1048). The rate of LNM in thymoma, thymic carcinomas and thymic neuroendocrine tumours was 6.8% (42/622), 30.2% (100/331) and 61.1% (58/95), respectively. According to the logistic regression analysis, histology type and T stage were independent factors correlated with LNM. A predictive nomogram model was developed with a concordance statistic of 0.807 (95% confidence interval: 0.773–0.841), which was significantly better than the T stage (P < 0.001) while had limited benefit to the histology type (P = 0.047). The calibration curve for the nomogram comparing the predicted and actual probabilities after bias correction showed good agreement.

CONCLUSIONS

Nodal involvement was not uncommon in TETs. Main factors related to LNM in TETs were histology type and T stage. The probability of LNM could be well calculated using the predictive model.

INTRODUCTION

The thymus is an important lymphoid organ that gradually deteriorates with the development of the adaptive immune system in childhood and is ultimately replaced by fat in adulthood [1]. Under certain circumstances, the residual epithelial tissue can develop into a tumour. Thymic malignancies are a relatively rare type of thoracic solid tumour and include tumours derived from epithelial cells, germ cells, lymphocytes and soft tissues. According to the pathological classification of the 4th edition of the World Health Organization, most of thymic epithelial tumours (TETs) are regarded as malignant tumours. TETs are further classified as thymomas, thymic carcinomas (TCs) or thymic neuroendocrine tumours (TNETs) [2].

Lymph node metastasis (LNM) is an important prognostic indicator of thymic malignancy [3]. In the past, it was generally believed that LNM was rare in TETs; thus, lymph node dissection (LND) or sampling was rarely done during surgical resection. However, in recent years, LNM in TETs has garnered greater attention [4]. Previously, the commonly used staging system for thymic malignancies was Masaoka–Koga staging and LNM was generally classified as stage IVb, which is the same staging as distant metastases [5]. In the 8th edition of the Union for International Cancer Control/American Joint Committee on Cancer stage program, tumour node metastasis (TNM) staging was used to distinguish thymic tumours from distant metastases and to identify the corresponding lymph node division in thymic tumours. Presently, the National Comprehensive Cancer Network and European Society for Medical Oncology guidelines emphasize the importance of LND in thymectomy, though the importance of lymphadenectomy in TETs has still not received widespread attention in current clinical practice [6, 7].

In the current study, a comprehensive, in-depth examination of the cancer registration database was done to determine and analyse the occurrence of LNM in thymic malignancies and to analyse the factors associated with a high risk of LNM. A predictive nomogram model was built based on the independent risk variables and the nomogram performance was measured.

PATIENTS AND METHODS

This research was a retrospective observational study conducted according to the Declaration of Helsinki and its amendments. Patient data were extracted from the Surveillance, Epidemiology, and End Results (SEER) database program. This database is a large-scale cancer registry database with various comprehensive data of anonymized cancer patients covering >30% of the US population [8]. The use of public available data is consistent with the requirements of the institutional review and the study protocol has been approved by the ethics committees of Klinikum Ernst von Bergmann (EA1/213/17). Consent was waived considering the anonymous, observational and registry-based nature of the study. The authors present the following article in accordance with the TRIPOD and STROBE reporting checklist.

Researchers consecutively screened and downloaded the data of Incidence-SEER 18 Regs Custom Data (with additional treatment fields), November 2018 Sub (1975–2016 varying), National Cancer Institute, Surveillance Research Program, released April 2019, based on the November 2018 submission. The inclusion criteria for the study exploring the factors related to LNM in TETs were as follows: (i) patients with histologically confirmed (positive histology) primary malignancies (code number: /3) located in the thymus (primary site-labelled: C37.9); (ii) patients who underwent surgical treatment and had 1 or more lymph nodes surgically resected and pathologically examined; (iii) since detailed information regarding dissected and positive lymph nodes was documented from 1988, the period studied was 1988–2016; and (iv) patients classified as having the following pathologies were selected: thymoma, TCs and TNETs. Case-complete analysis methods were applied to fully analysed all enrolled patients with the complete available data.

The diagnosis time, age, gender, race, marital status and other information at the time of diagnosis were obtained from the SEER database. All the variables were unified and coded. Information related to tumour pathology mainly included: tumour size, pathological type, number of LND, number of positive lymph nodes and T stage. T stage was divided into T1/2 and T3/4 groups depending on whether the peripheral organs were infiltrated or not, according to the 8th edition TNM staging [9]. Treatment-related information including surgery type and surgical treatment method were classified as limited resection, extended thymectomy or debulking, according to the SEER variable coding.

Logistic regression was used to explore variables associated with LNM in TETs using the rms R package. All research variables were included in the univariable logistic regression analysis and the odds ratio (OR) and 95% confidence interval (CI) of different parameters were compared. Factors significantly associated with LNM (P < 0.05) in the univariable analysis were included in the multivariable analysis to obtain the independent risk factors associated with LNM using the stepwise methods.

A predictive nomogram model was formulated using foreign, regplot and rms R packages with selected variables that achieved significance at P < 0.05, based on the results of the multivariable logistic regression analysis. In this study, the verification of the nomogram model was completed in 2 steps: discrimination evaluation and calibration evaluation. The discrimination of the nomogram model refers to its ability to correctly distinguish whether an outcome event occurred or not and was calculated using the concordance statistic (C-statistic). Model discrimination is generally considered to be good when the C-statistic exceeds 0.7 and superior when it exceeds 0.8 [10]. The area under the receiver operating characteristic (ROC) curve (AUC) was used to compare its performance with those independently related variables. The significant difference between AUCs of different models was tested with the DeLong method using the pROC software R package. Internal calibration plots with 1000 bootstrap resamples from the total cohort were used to evaluate the nomogram performance, which compared the predicted and observed probabilities of LNM. The model was considered to have good prediction consistency when the predicted value and the actual value fit on the 45° line. Decision curve analysis (DCA) is a tool used to evaluate the clinical value and application prospects of predictive models [11].

The statistical analyses in this research were performed using R software 4.0.5 (https://www.r-project.org) and SPSS Statistics 24.0 (IBM Corp., Armonk, NY, USA). The number of observations and proportions of categorical variables were calculated, and the significant differences between groups were compared via chi square test or Fisher's exact test. Continuous variables that did not meet the normal distribution, median and interquartile range (IQR) were analysed for continuous variables by the Mann–Whitney test (U) test or Wilcox signed rank test. Two-sided P-values <0.05 was considered statistically significant. Blinding analysis and the adjusted association between each candidate predictor/outcome were not involved in this registry population-based research.

RESULTS

Characteristics of enrolled patients

A total of 1048 patients met the research criteria and were included in this retrospective study. Of these, 59.4% (622/1048) had thymoma, 31.6% (331/1048) had TCs and 9.1% (95/1048) had TNETs. The majority of patients were male (53.6%, 562/1048) and Caucasian (71.1%, 745/1048), with a median age of 60 years (IQR: 49–70). The median tumour size of all patients was 6.5 cm (IQR: 4.8–9.0). The median number of lymph nodes removed was 3 (IQR: 1–6).

The overall proportion of LNM in thymic tumours was 18.5% (200/1048). The rate of LNM in thymomas, TCs and TNETs was 6.8% (42/622), 30.2% (100/331) and 61.1% (58/95), respectively (P < 0.001) (Table 1). The percentage of LNM in the T3/4 group (24.5%, 120/490) was significantly higher than that in the T1/2 group (12.6%, 58/460). Details of the LNM percentages in different pathological subtypes of TETs are shown in Fig. 1.

Univariable and multivariable analyses of lymph node metastasis

Univariable logistic analysis was performed on all variables of patients who underwent LND to explore the factors that correlated with LNM. Race (P = 0.011), histology type (P < 0.001), T stage (P < 0.001), surgery type (P < 0.001) and number of LNDs (P = 0.001) were significantly correlated with LNM (Table 2). These factors were included in the multivariable logistic regression analysis. The results showed that B3 thymoma (OR = 3.616, 95% CI : 1.729–7.563, P < 0.001), TCs (OR = 8.536, 95% CI: 4.569–15.950, P < 0.001), TNETs (OR = 39.360, 95% CI: 18.803–84.430, P < 0.001) and T3/4 stage (OR = 1.141, 95% CI: 1.042–1.250, P = 0.004) were independent risk factors for LNM. The later T stage or the more advanced histological type, the more likely the occurrence of LNM (Table 3).

The development of the predictive model

A predictive nomogram model was developed from the variables that independently correlated with LNM as determined by the multivariable logistic regression analysis (Fig. 2). According to the prediction model, the probability of LNM in patients with T1/2 TCs and TNETs was 19.9% (95% CI: 14.6–26.5%) and 51.5% (95% CI: 39.9–63.0%), respectively. The probability of LNM was 36.0% (95% CI: 29.8–42.7%) and 70.7% (95% CI: 59.5–79.9%) in patients with TCs and TNETs at T3/4 stage, respectively. In thymoma, the probability of LNM was 17.8% (95% CI: 12.0–25.5%) in T3/4 type B3 thymoma and was <10% for other subtypes.

The evaluation of the predictive model

The model demonstrated a good accuracy for predicting LNM with a C-statistic of 0.807 (95% CI: 0.773–0.841), indicating that the nomogram had good discrimination ability for estimating the status of LNM. The prediction model performed better, according to the Delong test for the area under the ROC curve (AUC = 0.807), when compared with predictors of T stage (AUC = 0.597, P < 0.001) or histology type (AUC = 0.790, P = 0.047) alone (Fig. 3A). The calibration curve of the nomogram comparing the predicted and actual probabilities after bias correction showed good agreement (Fig. 3B). DCA was used to further evaluate the clinical utility of this graphical model. The curves indicated that, if the threshold probability was in the range of 0.10–0.85, this graphical model provided a net benefit when used to predict LNM (Fig. 3C). The results from the DCA showed that the nomogram model was of more benefit than the simple T stage model or histology type model alone.

DISCUSSION

This study systematically analysed the data from TET patients in the cancer registration database who had received surgical treatment and LND and analysed factors that were potentially related to or affected LNM via univariable and multivariable logistic regression analysis. The pathological type and T stage of TETs were found to be independent factors related to LNM. A prediction model for LNM in TETs was developed based on the results of the multivariable analysis.

TETs is a relatively rare type of indolent thoracic cancer [12]. It is precisely because of the rarity of TETs and the low-grade malignancy of some pathological subtypes that research on LNM of this kind of tumour is sparse, and LND or sampling is rarely mentioned during surgical resection. In recent years, the issue of LNM in TETs has garnered more attention. The staging systems widely used in clinical practice are the Masaoka–Koga and TNM staging systems. Masaoka–Koga staging was commonly used in the past and generally classified LNM and distant metastases as stage IVb. The 8th edition of Union for International Cancer Control/American Joint Committee on Cancer staging was adopted in 2017, in which comprehensive staging is carried out according to the dimension of TNM [13]. According to the mediastinal lymph node map, the lymph node status is classified as N0, N1, N2 and Nx [14].

In recent years, some scholars have studied LNM in TETs and the impact of LNM on prognosis. Kondo et al. conducted a retrospective study of data from 1320 TETs patients at 115 medical centres collected in 2003. It was found that the rate of LNM was 1.8% in thymoma, 26.8% in TCs and as high as 27.5% in TNETs [15]. In 2015, Weksler et al. [16, 17] evaluated data from the SEER database between 1988 and 2011 and found that the proportions of LNM in cases of thymoma, TCs and TNETs were 13.3%, 33.5% and 62.3%, respectively. The prognoses of patients who were LNM positive were significantly worse than in patients who were not LNM positive. The proportions of TCs and TNETs positive for LNM were similar to what was reported in previous studies. The current study analysed the occurrence of LNM in different thymoma pathological subtypes; thus, thymomas of unclear pathological types (8580/3: not otherwise specified) were not included; however, these were included in the Weksler et al. study. The percentage of thymomas with LNM found in the current study was 6.8%, which was quite different from the percentage found in the previous study (13.3%). Like our study, Hwang et al. [18] found an LNM rate of 5.1% in thymomas in a Korean cohort study. A prospective multicentre clinical study from the Chinese Alliance for Research in Thymomas (ChART) database found that 275 patients with TETs underwent intentional lymph node sampling or dissection [19]. The proportion of thymoma, TCs and TNETs patients with LNM in this study was 2.1%, 25% and 50%, respectively, which is similar to the percentage of patients with LNM found in our study.

In addition to the pathological subtypes of TETs, the T stage is also an important factor that is associated with LNM. The rate of LNM becomes higher significantly as the tumour T stage increases. This study found that T3/4 is an independent risk factor for LNM in TETs. In addition, the relationship between tumour size and LNM was examined by univariable analysis, and there was no significant correlation found. Therefore, although the relationship between LNM and tumour size is still controversial, the relationship between LNM and tumour local invasion is relatively definitive [20]. Similar findings appeared in previous studies, suggesting that LNM in TETs is related to the invasiveness of tumours to a certain extent [19, 20].

In the univariable analysis, race, number of LND and surgery type were associated with LNM, though these variables were not significantly related to LNM in the multivariable analysis. Previous studies found that Asian people have a higher incidence of TETs than Caucasians [21]. However, there are no detailed studies on differences in LNM related to race. The current study found that a greater number of LND and more extended thymectomies were associated with LNM. To some extent, this reflects that a more thorough resection can more effectively detect potential suspicious lymph nodes and a more complete pathological examination can be conducted. However, no significant correlation of LND or extent of thymectomy with LNM was found in multivariable analysis. A multicentre study from ChART found that deep mediastinal (N2) dissection was an independent factor for the detection of LNM. The lymph node stations resected may be more important than the number of LND [22]. At present, both National Comprehensive Cancer Network and European Society for Medical Oncology guidelines emphasize the importance of LND in the radical surgical treatment of TETs [23].

According to the guidelines’ description of LND, all suspicious lymph nodes need to be removed [14]. LND in TCs and TNETs can clarify the status of LNM and provide more accurately evaluation of patients’ long-term prognosis, as previously published by our research team [4]. The present study extends the study subjects to all the TETs pathology types (thymoma, TCs and TNETs) to further explore the high-risk factors related to LNM. Dissection of lymph nodes around the tumour and anterior mediastinum is recommended for patients with clinical stage I and II TETs. For patients with clinical stage III, systematic dissection of anterior mediastinal lymph nodes and sampling of some intrathoracic lymph nodes are recommended. For patients with suspected or confirmed TCs or TNETs, it is recommended that the systematic sampling of at least the anterior mediastinal, intrathoracic, supraclavicular and inferior cervical lymph node stations should be performed [3]. Complete R0 thymectomy usually includes the resection of N1 region anterior mediastinal lymph nodes. For other mediastinal lymph node stations, the location of the resected lymph nodes should be marked referring to ITMIG/IASLC standard lymph node map and submitted for examination, according to the characteristics of TETs [24].

Currently, the preoperative assessment of lymph node status in patients with TETs has not received sufficient attention. Computed tomography (CT) and magnetic resonance imaging (MRI) studies are the main preoperative radiology evaluation methods for TETs, having both still some limitations in pathological type determination [18]. Previous studies have found that 18F-fluorodeoxyglucose positron emission-computed tomography (PET-CT) has good efficacy in differentiating and assessing the pathological type and tumour aggressiveness of TETs [25, 26]. Moreover, PET–CT is also of great clinical value in the evaluation of lymph node status and distant metastasis. To determine its significance in LNM assessment in TETs, more studies are needed. For TET patients with much advanced pathological types or aggressive malignancies, a more complete mediastinal regional LND (N1 + N2) should be performed as recommended [24]. Surgeons should perform a more detailed and in-depth preoperative evaluation of the lymph node status and clarify the potential clinical N stage in TETs patents with reference to the ITMIG lymph node map [13]. The management of the lymph nodes (the specific location, sampling or dissection of the lymph nodes) should be recorded in the operative notes.

In this study, a predictive model for lymph node involvement in TETs was constructed based on the results of logistic regression multivariable analysis. The analysis results of the ROC curve indicated that the predictive model had a good prediction efficiency and better accuracy compared with individual predictors alone. Good consistency was obtained based on the internal verification results of the calibration curve. The DCA curve showed that the predictive model would be beneficial for clinical application. At present, only 1 study in the ChART database was found that examined predictive variables for LNM in TETs [19]. This prospective landmark study found 3 independent LNM risk factors, including pathological type (B3/TCs/TNETs), T staging (T3/T4) and N2 dissection (Yes). The study included data from 275 TETs patients and constructed predictive models, ushering in a new era for LND in TETs. The current study confirmed relevant results of previous studies and has the following advantages: our study included a much larger number of patients compared to the prospective study. In addition, good results were obtained using several multidimensional evaluation methods that were applied to assess the discrimination, calibration and clinical practicality of the nomogram model.

Limitations

The limitations of this study were as follows: first, this was a retrospective study that contained some inevitable natural deviations, and the variability in lymph node sampling/dissection according to the institutions and to the histology and T stage may well influence their findings on the probability of lymph node involvement. The data were extracted from the cancer registry database, and some important clinicopathological variables could not be fully included in the study (e.g. neoadjuvant treatment, specific N1 or N2 stations of LND and metastasis). Due to the relative rarity of TETs, LND is still not routinely performed along with thymectomy. Therefore, there was a lack of sufficient data to carry out further external verification.

In general, nodal involvement in TETs is not uncommon, especially in tumours of more invasive pathological types (TCs and TNETs). This study found that in tumours with a higher T stage, more severe local invasion is associated with a higher probability of LNM. Therefore, when TETs are surgically treated, the resection of lymph nodes should also be considered. Removal of any suspected metastatic lymph nodes during thymectomy is recommended. Systematic mediastinal LND is recommended for patients with high T stage (T3 and more) and advanced histological malignancies (especially TNETs or TCs). In this study, the nomogram prediction model of LNM in TETs was constructed based on the results of multivariable logistic regression, and the probability of LNM in TETs of different pathological types and T stages was assessed. The prediction efficiency of the model was evaluated using a variety of methods and the model was found to have a superior predictive effect, compared to the T stage while also having limited advantage than the histology types. In clinical practice, more thorough LND for high-risk patients are of great clinical significance.

Presented at the 35th Annual Meeting of the European Association for Cardio-Thoracic Surgery, Barcelona, Spain, 13–16 October 2021.

ACKNOWLEDGEMENTS

The authors acknowledge the statistical assistance from Dr. Jiani Wang (Institute for Social Medicine, Epidemiology and Health Economics, Charité-Universitätsmedizin zu Berlin).

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Conflict of interest: none declared.

Data Availability Statement

The data in this study were obtained from the SEER project, which was a public database. The data underlying this article will be shared at reasonable request to the corresponding author.

Author contributions

Zi-Ming Wang: Conceptualization; Data curation; Formal analysis; Methodology; Software; Visualization; Writing—original draft; Writing—review & editing. Feng Li: Data curation; Methodology; Software; Visualization; Writing—original draft; Writing—review & editing. Lara Sarigül: Methodology; Writing—original draft; Writing—review & editing. Dania Nachira: Methodology; Writing—original draft; Writing—review & editing. Diego Gonzalez-Rivas: Methodology; Resources; Writing—review & editing. Harun Badakhshi: Project administration; Supervision; Writing—review & editing. Jens-C. Rückert: Project administration; Resources; Supervision. Calvin S.H. Ng: Resources; Supervision. Mahmoud Ismail: Conceptualization; Resources; Supervision; Validation; Writing—original draft; Writing—review & editing.

Reviewer information

European Journal of Cardio-Thoracic Surgery thanks Clemens Aigner, Yukinori Sakao and the other, anonymous reviewer(s) for their contribution to the peer review process of this article.

REFERENCES

ABBREVIATIONS

ABBREVIATIONS
 
• AUC

  Area under the receiver operating characteristic curve

 
• ChART

  Chinese Alliance for Research in Thymomas

 
• CI

  Confidence interval

 
• DCA

  Decision curve analyses

 
• LND

  Lymph node dissection

 
• LNM

  Lymph node metastasis

 
• OR

  Odds ratio

 
• ROC

  Receiver operating characteristic

 
• SEER

  Surveillance, Epidemiology, and End Results

 
• TCs

  Thymic carcinomas

 
• TETs

  Thymic epithelial tumours

 
• TNETs

  Thymic neuroendocrine tumours

 
• TNM

  Tumour node metastasis