Is central lung tumour location really predictive for occult mediastinal nodal disease in (suspected) non-small-cell lung cancer staged cN0 on ¹⁸F-fluorodeoxyglucose positron emission tomography–computed tomography?

Abstract

INTRODUCTION

Mediastinal lymph node involvement in patients with non-small-cell lung cancer (NSCLC) is an important predictive parameter of survival, and it allocates towards appropriate therapy [1]. Several studies found an association between central location of a lung tumour and unforeseen positive mediastinal nodes (pN2–3) after negative mediastinal imaging by computed tomography (CT) and/or positron emission tomography (PET) with a prevalence of 16–36% [2–6]. Therefore, current guidelines recommend invasive mediastinal staging by endosonography or videomediastinoscopy in patients with a centrally located tumour and negative mediastinum on PET and CT [1, 7–9].

Different definitions of central location of a lung tumour are being used in the literature. The National Comprehensive Cancer Network (NCCN) and European Society of Thoracic Surgeons (ESTS) guidelines define a lesion within the inner two-thirds of the lung as central [7, 8]. The guidelines of the American College of Chest Physicians (ACCP) consider lesions in the inner one-third as central [9]. Other definitions of central location are also used, such as endobronchial visualization by the standard bronchoscope [10, 11], contact with lobar or first segmental branches of vessels or bronchus [12] or distance of less than 2 cm from the bronchial tree [13]. An online survey showed huge variation in the definition of central lung tumour location among both pulmonologists and surgeons [14].

The objective of this study was to identify which definition of central location of the tumour would have the highest predictive value of unforeseen positive mediastinal lymph nodes in PET-CT-staged clinical N0 NSCLC.

PATIENTS AND METHODS

This is a retrospective study from a prospective single-centre database, managed by the respiratory oncology multidisciplinary team of the University Hospitals Leuven, Belgium (The Leuven Lung Cancer Group). The institutional review board’s approval was obtained (S59161). Data were collected and anonymized in accordance with the International Conference on Harmonization Guidelines of Good Clinical Practice. Patients with (suspected) NSCLC, clinically node negative (cN0) on imaging with an integrated ¹⁸F-fluorodeoxyglucose-PET and CT scan, resectable and referred for invasive mediastinal staging or anatomical resection were included. Our registry contains 5009 unique patient entries between 2005 and 2015. We excluded patients who did not undergo invasive mediastinal and/or surgical resection (n = 2444), patients with suspected positive lymph nodes on PET-CT (n = 1385) and patients with pretreatment diagnosis of neuroendocrine histology (carcinoid, large-cell neuroendocrine carcinoma or small-cell lung cancer) (n = 174), radio-occult lesions (n = 10), former therapy of lung cancer (n = 114) or without imaging for review (n = 69). Lymph nodes were considered radiologically suspicious if these had a short axis of more than 1 cm in CT or if these were ¹⁸F-fluorodeoxyglucose-PET positive by visual qualitative assessment (uptake higher than mediastinal blood pool). The International Association for the Study of Lung Cancer (IASLC) lymph node map was used to define the lymph node stations [15]. Clinical and pathological stages were converted to the 8th edition of the Union for International Cancer Control TNM classification [16]. Radiographic size, solid versus subsolid lesion, pretreatment histology, cT, cM, side and location in upper/middle versus lower lobe were investigated together with 5 different definitions of central location of the tumour:

1. Inner 1/3 (central) vs outer 2/3 (peripheral): patients were allocated to the ‘inner 1/3’ group if the centre of the tumour was located in the inner 1/3 of the lung parenchyma, i.e. adjacent to the mediastinum, 1/3 radially measured from the secondary carina on transverse CT imaging [9].

2. Inner 2/3 (central) vs outer 1/3 (peripheral): equally measured according to the first definition but at 2/3 [7, 8].

3. Contact with lobar or first segmental branches of pulmonary vessels or bronchus (central) vs no contact (peripheral): patients were allocated accordingly after review of transverse CT imaging [12].

4. Tumour ≤2 cm of the bronchial tree (central) or >2 cm from the bronchial tree (peripheral): patients were allocated to the central group if the tumour reached a zone of ≤2 cm around the distal 2 cm of trachea, carina, named major lobar bronchi, up to their first bifurcation [13].

5. Visualized on the standard bronchoscopy (central) vs not visualized (peripheral): patients were allocated to the central group if the tumour was visualized by the standard bronchoscopy (typically 5–6 mm bronchoscope, i.e. not by paediatric scope or endosonography) [10, 11].

All CT scans were primarily reviewed by 1 author (H.D.C.). CT images of patients with lesions that were located in an ambiguous position (borderline central versus peripheral) were discussed in group (H.D.C., C.D. and P.D.L.). The ESTS guidelines on perioperative systematic nodal dissection were routinely followed by all involved surgeons [17].

Aims

The primary aim was to evaluate the association between each of the 5 different definitions of central tumour location and the prevalence of positive mediastinal nodal disease (pN2–3). The secondary aim was the evaluation of the association between central tumour location and any unforeseen positive nodes, i.e. pN1 and pN2–3 combined (pN+).

Statistics

Odds ratios and 95% confidence intervals (CIs) were reported from binary logistic regression models for pN2–3 upstaging (versus pN0–1) and pN+ upstaging (versus pN0, i.e. no upstaging). For each of the 5 definitions of central tumour location separately, a multivariable logistic regression model was fitted for pN+ upstaging with the following selected parameters: central tumour location, cT, pretreatment histology, lobar location (upper/lower), solid versus subsolid, cM and radiographic size. Given the low prevalence, no multivariable analyses were performed for pN2–3 upstaging. The discriminative ability has been quantified using the area under the curve (0.5 = random prediction and 1 = perfect discrimination) and compared between models using a test proposed by DeLong et al. [18]. The area under the curves of multivariable models were obtained after leave-one-out cross-validation. P-values <0.05 were considered significant. All analyses were performed using the SAS software, version 9.4 of the SAS System for Windows.

RESULTS

Patient characteristics

We identified 813 patients with (suspected) NSCLC, cN0 after PET-CT and referred for mediastinal staging and/or surgical resection after multidisciplinary team meeting between January 2005 and December 2015. Pretreatment demographics are listed in Table 1. In half of the patients, pathological tumour confirmation was absent prior to invasive staging or resection, and 4% were benign at final pathology. Subsolid lesions (semisolid or pure ground glass) were observed in 7%.

Central location

The different definitions of central tumour location resulted in different percentages of lesions being central. The most conservative definition was ‘visualized by bronchoscopy’ with 23% of lesions designated as being central. Location in the inner 1/3 of the lung, tumour contact with bronchovascular structures or distance ≤2 cm of the proximal bronchial tree resulted in 32–36% of lesions being central. The least selective was location in the inner 2/3 of the lung with half of the patients being central (Table 1).

Management of patients

Thirty-one percent of patients underwent invasive mediastinal nodal staging, with 93% of them diagnosed by videomediastinoscopy (Table 2). Primary resection was performed in 751 patients and in 189 patients after invasive staging. Three patients underwent explorative surgery only. Thirty-nine patients underwent induction treatment followed by surgery (Table 2). Twenty patients did not undergo resection after invasive mediastinal staging. Eleven of them had positive mediastinal nodes at invasive staging. Resections were performed using video-assisted thoracic surgery (VATS) in 42% of patients (n = 335/790). Central lesions (inner 1/3) were operated using VATS in 10% of the patients (n = 24/243). Overall, the median number of examined lymph node stations was 5 [interquartile range (IQR) of 3–6], and median number of mediastinal lymph node stations was 3 (IQR 2–4) (Table 2). We found no difference in the number of lymph node stations assessed between VATS and open surgery [5 (IQR 3–5) vs 5 (IQR 3–6), P = 0.33]. More stations were evaluated in case of central tumour location versus peripheral [for the inner 1/3 definition: 5 (IQR 4–6) vs 4 IQR (3–5), P < 0.001]. There were 324 observed deaths. The mean follow-up was 84 months. The 5-year survival was 61% (95% CI 58–64) with 258 patients at risk at 5 years.

Prevalence of nodal upstaging

Any nodal upstaging (cN0 to pN+) was found in 171 (21%) patients. This was pN1 in 106 (13%) patients and pN2–3 in 65 (8%) patients (Table 3). The negative predictive value [true negatives (n = 748)/all negatives (n = 813)] of PET-CT in the detection of positive mediastinal nodes was, therefore, 92%. Positive mediastinal nodes were detected by invasive mediastinal staging in 17 (26%) patients, by lymphadenectomy at resection after false-negative invasive staging in 9 (14%) patients and by surgery without preresection invasive mediastinal staging in 39 (60%) patients. The negative predictive value of mediastinal invasive staging was 97% [true negatives (n = 223)/all negatives (n= 231)] at a prevalence of 11% [n = (9 + 17)/248].

Mediastinal nodal upstaging (pN2–3)

Univariable analysis did not demonstrate a relationship between central tumour location and unforeseen positive mediastinal nodes (pN2–3) for any of the 5 definitions of central tumour location (Table 4). The prevalence of pN2–3 was 8% overall for the complete cohort and never higher than 10% in patients with central tumour location whatever definition used (Table 4). The univariable analysis demonstrated that pretreatment non-squamous histology was predictive of mediastinal nodal disease (Table 5). Also, in none of the multivariable analyses, a significant relationship between central tumour location by any of the 5 definitions and pN2–3 upstaging was obtained. Non-squamous pretreatment histology was the only independent predictor in all 5 multivariable models.

Any nodal upstaging (pN+)

Central tumour location was related to overall nodal upstaging (from cN0 to pN+) in a univariable analysis with odds ratios between 3.36 (95% CI 2.32–4.88) and 4.21 (95% CI 2.96–5.99) depending on the definition of central tumour location (Table 4). Irrespective of the definition of centrality, pretreatment histology, centrality and tumour location in the upper lobe were significant in the multivariable model (with the exception of histology in the model with visualization by bronchoscopy as definition of central location). Table 6 summarizes the result for the model with the inner 1/3 as definition for centrality. The odds ratios for the effect of centrality equalled 3.90 (2.24–6.77), 2.99 (1.66–5.36), 4.46 (2.47–8.06), 4.21 (2.36–7.51) and 3.27 (1.79–5.97) in multivariable models with centrality definitions 1–5, respectively. The area under the curve of multivariable models ranged between 0.736 and 0.754. AUCs did not differ significantly between the 5 definitions, neither in the univariable model nor in the multivariable model (P > 0.10 for all pairwise comparisons).

DISCUSSION

Our main finding is that in a cohort of (suspected) NSCLC without nodal involvement after PET-CT (cN0), the prevalence of unforeseen positive mediastinal nodes was only 8%, varying between 6 and 10% according to the 5 verified definitions of central tumour location. The negative predictive value of imaging with PET-CT was, therefore, 92%. A review of the historical series with CT imaging without PET demonstrated a median negative predictive value of 82% (range 55–91%) [9]. With integrated PET-CT, the mean negative predictive value increased to 90% (range 83–100%), which is in line with this study [9].

This is the first study comparing 5 definitions of central tumour location and their relationship with unforeseen mediastinal disease in patients with cN0 (suspected) NSCLC and referred to invasive mediastinal and/or surgery. We were unable to identify a definition of central tumour location with a stronger predictive value for unforeseen mediastinal disease over another. In fact, none of the 5 definitions of central tumour location were related to pN2–3 in the univariable or the multivariable analysis. We could argue that the best definition is the one that is the most reproducible and easy to obtain. Further studies would be necessary to answer this question, although the clinical impact might be small if the prevalence of unforeseen mediastinal nodes is confirmed to be under 10% in studies with modern integrated PET-CT.

Current guidelines recommend invasive mediastinal staging in case of central tumour location [1, 7–9]. Previous studies demonstrating a relationship between central tumour location and unforeseen mediastinal nodes after imaging were different from this study as they combined cN0 and cN1 patients [3] and were based on CT imaging only [4, 5] or on non-integrated CT and PET [2]. Apart from our study, 3 other studies with cN0 NSCLC patients after PET-CT failed to demonstrate a significant difference in mediastinal upstaging between central or peripheral lesions [19–21]. We thus question the prediction of unforeseen mediastinal disease in case of cN0 NSCLC solely based on central location of the tumour. To our knowledge, the only study demonstrating that central location was a predictor of pN2 disease in patients with cN0 NSCLC after integrated PET-CT was recently published by Gao et al. [6]. Of note, they reported a high prevalence of subsolid lesions (56% vs 7% in this study), 19% (n = 53/284) of patients were not referred for resection (after invasive staging) and the number of assessed lymph node stations was not recorded. It is unclear whether relative more resections with lymphadenectomy were performed in case of central lesions, and non-surgical treatment was more frequent in peripheral lesions. With this conflicting literature, we argue that guidelines should not catalogue patients with central tumours together with cN1 disease on imaging. This was the case in the ACCP guidelines due to the difficulty to assess the N1 nodes in case of central tumour location [9]. Unlike studies concerning central tumours, studies with focus on cN1 lesions have consistently demonstrated 20–30% prevalence of occult N2 disease [22].

To make a prediction on the chance of unforeseen mediastinal disease, and therefore, the usefulness of invasive mediastinal staging, a model more complex than currently used by guidelines seems to be necessary. Farjah et al. [23] developed a prediction model for pathological N2 disease in patients with a negative mediastinum by PET. In this model with 6 risk factors for N2 disease including central tumour location, only cN1 disease by PET was associated with pN2 disease after the multivariable analysis. These patients were not included in our study as it is clear that cN1 disease warrants invasive mediastinal staging. The rate of unforeseen pN2 is also variable according to histological grade [23]. In our practice, this factor would be only partially helpful as only half of the cN0 patients referred to surgery had known histology. Multidisciplinary discussion of patient profile and imaging resulted in the resection of benign lesions in only 8% of patients without pretreatment histology or 4% overall.

Looking at any nodal upstaging, i.e. pN1 and pN2-N3 together (pN+), all 5 definitions of central tumour location were similarly predictive. Visualization by bronchoscopy was most selective with the highest chance of unforeseen lymph node disease (42% in the central group). Area under the curve was similar for 5 definitions. The significant relationship between central tumour location and any nodal upstaging (cN0 to pN1 or pN2) is driven by cN0 to pN1 upstaging. We previously demonstrated in a multicentre retrospective study that more than a quarter of patients with a central cStage I NSCLC had unforeseen positive N1 nodes at resection [24]. As surgeons often choose an open approach for central lesions (i.e. with higher intrinsic chance of N1 upstaging), failure to include central tumour location in several retrospective studies comparing N1 upstaging between VATS and open surgery leads to a selection bias with seemingly lower N1 upstaging in the VATS group [24].

Limitations

The following are limitations of this study. Except from the characteristic problems with retrospective analyses, we should be especially careful while drawing conclusions from the prevalence of mediastinal disease in subgroups, as these are based on low numbers. We did not include the primary tumour maximal standardized uptake value (SUV_max) as a parameter—even if 1 study suggested a relationship with unforeseen mediastinal nodal disease [25]—for several reasons, in our prospective experience, analysis with an SUV threshold was not superior to the use of visual interpretation scale, an ideal cut-off value has not been determined and the values are not standardized from centre to centre or even between scanners in the same centre [7, 26, 27]. Furthermore, a prospective study found a prevalence of only 5.6% N2 disease in cT1–2cN0 tumours with an SUV_max of greater than 10, i.e. presumed high-risk patients for unforeseen mediastinal disease [28]. We did not investigate potential differences in evaluation of centrality of tumours among different investigators. ‘Visualized by bronchoscopy’ might be an interesting parameter for large databases and studies as this is readily available in patient files, potentially less investigator dependent and causes less workload than retrospective CT analysis [11].

CONCLUSION

We found no correlation between 5 different definitions of central lung tumour location and unforeseen mediastinal nodal disease in patients with cN0 (suspected) NSCLC after imaging with contemporary PET-CT. The prevalence of N2 disease was 8%, and prevalence of any nodal upstaging (pN+) was 21%. All definitions of central tumour location were predictive for pN+. A more complex model than central location alone seems necessary to accurately predict mediastinal positive nodes when modern PET-CT demonstrates cN0 NSCLC.

Conflict of interest: none declared.

REFERENCES