Multicentric evaluation of the impact of central tumour location when comparing rates of N1 upstaging in patients undergoing video-assisted and open surgery for clinical Stage I non-small-cell lung cancer†

Abstract

INTRODUCTION

In 10–20% of patients with clinical Stage I (cStage-I) non-small-cell lung cancer (NSCLC) unforeseen positive lymph nodes are found during resection, resulting in pathological nodal upstaging [1, 2]. In case of comparable patient selection, preoperative staging and pathological examination, nodal upstaging can be used as a quality indicator for lymph node assessment during surgery. Although survival remains the ultimate outcome parameter, studying nodal upstaging is useful as it requires less patients to detect a quality difference [3].

Several retrospective studies have demonstrated lower rates of N1 upstaging (i.e. cN0 to pN1) after video-assisted thoracic surgery (VATS) compared with open resections, suggesting inferior lymph node assessment by VATS [2–6]. This comes in contrast to the similar or superior survival rates found in the VATS group [5].

We postulated that lower N1 upstaging might be the result of selection bias and relates to variables not captured in these retrospective analyses, even if propensity matching was used. One such unaccounted variable might be the location of the tumour (central versus peripheral).

The aim of this study is not to claim equivalence or superiority of VATS to open surgery in terms of nodal staging. This would only be feasible after conducting a well-powered randomized controlled trial (RCT) with an appropriate definition of a negligible difference in N1 upstaging. The objective of this study is to validate in a multicentric setup whether the reputedly lower probability of N1 upstaging after VATS resection disappears if central tumour location is part of the multivariable analysis [7]. We hypothesized that in a cohort of consecutive patients operated for cStage-I NSCLC, relatively more central tumours would be found in the open (versus VATS) group due to differences in patient selection. Additionally, we hypothesized that central tumours are associated with a higher incidence of unforeseen positive N1 nodes compared with peripheral tumour location during resection of cStage-I NSCLC.

METHODS

Institutional Review Board approval was obtained (S58903). The data were obtained and anonymized in accordance with the International Conference on Harmonization Guidelines of Good Clinical Practice. The study was registered at ClinicalTrials.gov (NCT02730897).

Data of patients with cStage-I NSCLC, consecutively operated in 2014, were collected in 11 European centres. Exclusion criteria included previous therapy for a thoracic cancer, metastatic disease, induction chemo- or radiotherapy, non-anatomical resections (wedge), previous lymph node disease and unavailability of preoperative fluorodeoxyglucose–positron emission tomography (PET). Lymph nodes were considered radiologically suspicious if they had a short axis of more than 1 cm at computer tomography (CT) scan or were fluorodeoxyglucose-PET positive by visual qualitative assessment (i.e. uptake higher than mediastinal blood pool).

Collected data included age, gender, clinical TNM, preoperative invasive mediastinal staging, surgical approach (VATS or open surgery), conversion, central tumour location, forced expiratory volume in 1 s, diffusion capacity of the lungs for carbon monoxide, side, histology and pathological TNM. Staging was performed according to the 7th edition of the TNM for lung tumours criteria of the American Joint Committee on Cancer [8].

Two definitions of central tumour location were used, and statistical analysis was separately performed with both definitions:

1. CT confirmed contact between tumour and the lobar pulmonary artery, vein or bronchus or their first segmental branches.

2. Tumour visualization on standard bronchoscopy (typically a 5–6-mm bronchoscope, i.e. not by paediatric scope or endobronchial ultrasound).

All CT scans were reviewed by the local principal investigators at each centre (cTNM staging and location). CT images of patients with lesions that were located in an ambiguous position (borderline central versus peripheral) were sent to all investigators to obtain a more uniform interpretation of the CT imaging.

Following an intent-to-treat principle, data of patients requiring a conversion from VATS to an open technique were analysed within the VATS group. A separate analysis was performed with conversions included in the open group. All contributing surgeons routinely follow the European Society of Thoracic Surgeons (ESTS) guidelines on systematic lymph node dissection [9].

Statistics

The Fisher’s exact test (Fisher–Freeman–Halton) and the Mann–Whitney U-test were used to compare categorical and continuous variables between groups. Odds ratios (ORs) and 95% confidence intervals (CIs) were reported from univariable binary logistic regression models for N1 upstaging (versus no upstaging). A random centre effect was added in each model to handle the clustering of patients within a centre. Results are also given for the analysis stratified by tumour location. A multivariable binary logistic regression model with a random effect of centre (0-mean normal distributed) was used to evaluate the relationship between surgical technique and N1 upstaging (versus no upstaging) after correction for a prespecified list of predictors, i.e. cT1 vs cT2a, gender, histology and preoperative mediastinal staging. Two versions of the model were considered, without (Model A) and with central tumour location on CT (Model B) included as a predictor. The multivariable models were performed on patients with complete information on all predictors. The standard deviation of the random effects was interpreted using the ‘divide by 4’ rule of thumb [10]. A likelihood ratio test (with a 50:50 mixture of χ²s with 1 and 0 degrees of freedom, respectively) was used to test for between-centre variability in N1 upstaging.

The P-values <0.05 were considered to be significant. All analyses were performed using SAS software, version 9.4 of the SAS System for Windows.

RESULTS

Data of 956 patients with cStage-I NSCLC, operated in 2014, were gathered from 11 centres. We excluded the following: histology other than NSCLC (n = 38); no anatomical resection (n = 3) and non-cStage I (n = 20). Data from 895 patients were used for the analysis. The number of patients operated on in each centre is shown in Fig. 1. All patients underwent fluorodeoxyglucose-PET and CT scan. The patient characteristics are summarized in Table 1. Preoperative invasive mediastinal staging was performed in 22% of patients (EBUS 15%, mediastinoscopy 5% and combination 2%). Most patients underwent a VATS procedure (n = 699). Conversion rate was 9% (n = 62). Variation per centre can be seen in Fig. 1. Resections consisted of 12 pneumonectomies, 77 anatomical segmentectomies and 806 lobectomies. In 6% (n = 53) of patients, the procedures were robot assisted. One centre had little experience with VATS in 2014 and provided 43% (n = 85/196) of the open procedures. Survival at 90 days and 2 years was 97.8% and 86.1%, respectively. Median follow-up period was 21 months (interquartile range 16–25 months).

Figure 1

Patients included per centre. VATS: video-assisted thoracic surgery.

Open in new tabDownload slide

Central tumour location

Central tumour location, defined as contact with lobar or first segmental bronchovascular structures on CT, was seen in 16% (n = 140) of all patients. In the open group, 28% of patients had central tumours and 12% in the VATS group (P < 0.001). Tumour was visible by standard bronchoscopy in 4% (n = 27) of patients in the VATS group vs 17% (n = 34) in the open group (P < 0.001).

When patients treated in the single centre with little VATS experience were excluded, the incidence of central tumours in the open group would rise from 28% to 41% (n = 45/111) according to CT definition and from 17% to 28% (n = 31/111) according to bronchoscopic definition.

Nodal upstaging

In 79% (n = 708) of patients, the pathological stage remained Stage I after resection. Overall upstaging was seen in 21% [VATS 19% (n = 136) vs open surgery 26% (n = 51), Table 2]. Upstaging by a higher T-factor (pT>2a pN0) was found in 5% (n = 47). Overall incidence of N1 and N2 upstaging was 9% and 6%, respectively. N1 upstaging was 8% after VATS vs 15% after open surgery. N2 upstaging was 7% after VATS vs 6% after open surgery. Univariable analysis showed higher chance for N1 upstaging in case of open surgery, squamous tumours, cT2a versus cT1 and central tumour location (by CT and/or bronchoscopy) (Table 3).

Multivariable analysis

Multivariable analysis with gender, VATS vs open surgery, histology, T-factor, invasive mediastinal staging and centre (as random effect) showed a significant lower incidence of N1 upstaging after VATS when central location was not taken into account (OR 0.43, 95% CI 0.23–0.79); P = 0.007) (Model A, Table 4). However, after correction for tumour location, the difference between VATS and the open technique diminished and the evidence for a relation disappeared (OR 0.76, 95% CI 0.41–1.42; P = 0.4) (Model B, Table 4). Only tumour location (CT definition) was a significant factor in the latter multivariable model, with a higher probability for N1 upstaging for central tumours compared with peripheral tumours (OR 6.2, 95% CI 3.56–10.77); P < 0.001) (Fig. 2). A similar result was found when ‘visualized on bronchoscopy’ was used as definition of central tumour location (OR 4.67, 95% CI 2.23–9.8; P < 0.001) (based on 690 subjects with complete data set).

Figure 2

Multivariable analysis of upstaging from cN0 to pN1. Odds ratio and 95% CI are shown for the following variables: central versus peripheral location of the tumour, VATS versus open surgery, cT2a versus cT1, squamous versus non-squamous, male versus female and invasive staging versus no invasive staging. CI: confidence interval; VATS: video-assisted thoracic surgery.

Open in new tabDownload slide

The evidence of an effect of surgical approach was also absent in an analysis stratified by tumour location. In patients with peripheral tumours on CT, the incidence of N1 upstaging was 6% (n = 33/590) vs 5% (n = 6/134) after VATS and the open technique, respectively (OR 1.15, 95% CI 0.42–3.19; P = 0.8). In the smaller subset of patients with central tumours on CT, N1 upstaging was 22% (n = 19/86) after VATS and 35% (n = 19/54) after open surgery (OR 0.53, 95% CI 0.24–1.19; P = 0.12). Although a difference in the effect of VATS as a function of tumour location was observed, this interaction was not significant (P = 0.22).

In a multivariable model for N2 upstaging (versus pN0–pN1), there was no evidence for a relationship with VATS, with (OR 1.29, 95% CI 0.62–2.69; P = 0.5) or without (OR 1.11, 95% CI 0.54–2.29; P = 0.8) correction for tumour location on CT.

In the multivariable Model B, there was no evidence for between-centre variability. The standard deviation of the random effects distribution equalled 0.088 on the logit scale. This implies that centres only differed by approximately ±2.2% on the probability scale (over and above the factors included in the multivariable model). This between-centre variability was not only low but also insignificant (P = 0.32).

Conversions

Conversion rate was 9% (n = 62/699). Central tumours were seen in 11% (n = 7/62) of patients who underwent a conversion from VATS to open surgery, and N1 upstaging was found in 6%. Multivariable analysis with conversions included in the open group showed no relation between type of surgery and N1 upstaging (OR 0.82, 95% CI 0.46–1.45; P = 0.5) and a strong relation between central tumour location on CT and N1 upstaging (OR 6.33, 95% CI 3.67–10.92; P < 0.001).

Tumour location and N1 upstaging in video-assisted thoracic surgery-predominant centres

In 7 centres, more than 80% of all cases were performed by VATS (VATS-predominant centres). These centres submitted 670 patients, of whom 91% (n = 613) of patients were operated on by VATS. Central tumours (on CT) were seen in 11% (n = 70) of the VATS group and in 51% (n = 29) of the open group. N1 upstaging was seen in 7% of VATS patients (n = 45) and in 30% (n = 17) of open cases.

DISCUSSION

Only a well-conducted and large RCT can demonstrate the superiority of N1 nodal dissection with VATS over open surgery and evaluate the clinical significance of a potential difference. In this multicentre retrospective analysis of operated patients with cStage I NSCLC, we found significantly higher N1 upstaging (cN0–pN1) in case of central tumour location compared with peripheral tumours. Secondly, we found more central tumours in patients operated by thoracotomy, probably due to selection bias. Thirdly, in a multivariable analysis, the initially presumed lower incidence of N1 upstaging after VATS compared with open surgery disappeared. Only central tumour location remained a significant parameter. This confirms the results of our previous single-centre analysis: patients with central tumours are negatively selected for VATS and have relatively more unforeseen positive N1 nodes [7]. Without disproving that a lower incidence of N1 upstaging might be related to the VATS technique, this study clearly shows that the difference in the probability of N1 upstaging when compared with open surgery is overestimated. This finding is further corroborated by recent data showing a survival advantage in VATS patients in spite of a lower incidence of upstaging [5].

Several large retrospective studies investigated and reported lower incidence of N1 upstaging after VATS compared with thoracotomy [2–6] (Table 5). This finding is important as it suggests a poorer lymph node dissection by VATS. Although the ultimate oncological outcome is survival, assessing N1 upstaging as a quality indicator is interesting as less patients would be needed to find significant differences in performance [3]. The quality of the surgical resection is thought to have a much higher impact on the long-term survival compared with the short-term outcomes [11, 12]. Furthermore, N1 upstaging is less related to the quality of the preoperative mediastinal lymph node evaluation by PET-CT or invasive mediastinal staging in comparison with the study of N2 upstaging [7]. In our opinion, it is therefore a better parameter to evaluate the quality of the surgical resection.

The only RCT comparing N1 upstaging after VATS (n = 2/48) with open surgery (n = 3/52) was too small to detect a difference and recruitment ended in 1994 [13]. A more recent RCT on the number of lymph nodes removed was even smaller (n = 66) and found an overall N1 upstaging rate of 11%. However, it did not report upstaging after VATS versus open surgery separately [14].

The first large study was conducted by Boffa et al. [2] with more than 11 000 patients from the voluntary database of the Society of Thoracic Surgeons (STS). It showed a similar rate of N2 upstaging but significantly less N1 upstaging in the VATS group, even after propensity-matched analysis (Table 5). One of the strongest arguments in favour of a selection bias was the finding of 17% cN0 to N1 upstaging in patients operated through thoracotomy by VATS-predominant surgeons (≥80% of submitted cases by VATS). This suggests that cases with a higher chance of upstaging were intentionally operated using an open technique [2, 7]. A subanalysis of the 7 VATS-predominant participating centres in our study revealed a high 30% rate of N1 upstaging in case of open surgery. More than half of those patients had central tumours. This suggests an important selection bias, in which directing cases with a higher potential for N1 upstaging towards favoured for open surgery.

Two other large studies, both based on the National Cancer Database (NCDB, USA), were published recently. Medbery et al. [6] confirmed the previous finding of a lower incidence of N1 upstaging after VATS, while Yang et al. did not [15]. Both studies used an intent-to-treat analysis and propensity score matching, although they included different variables to construct the propensity score [16]. The study by Yang et al. was larger with 9390 matched pairs, covered a more recent study period and included surgery after induction therapy (only 1.5% of the patients) and robotic procedures (analysis of VATS versus robotic did not show differences). Interestingly, Medbery et al. found no difference in nodal upstaging between open or VATS lobectomy in centres with academic/research programmes. The quality of surgery is dependent on both the technique and the surgeon. Most importantly, upstaging is directly related to the extent of lymph node assessment by surgeon and pathologist [17]. The majority of patients in the NCDB (upstaged or not upstaged) did not have adequate lymph node assessment as recommended by current guidelines [16].

Still, in the majority of the retrospective studies, the incidence of N1 upstaging is lower after VATS compared with open surgery (Table 5). We believe that not including the central location of a tumour among the adjusting factors conceals an important selection bias in these retrospective studies. Unavailability of the variable ‘central location of the tumour’ in the databases is an important issue. Reassessing the CT scans is time-consuming. Not only is information on the ‘central location on CT’ often unavailable and a burden to search retrospectively from large data sets, the definition of central location is not uniform. We therefore used ‘visibility on standard bronchoscopy’ as a definition in a previous study and evaluated its use in this study [7]. Although less patients are included in the central tumour group with this definition, the incidence of pN1 was the highest of all subgroups (36%, Table 3) and resulted in similar findings, i.e. central location is the solitary significant parameter N1 upstaging in multivariable analysis. As ‘visibility on standard bronchoscopy’ is often readily available in patient records, we think it is a promising parameter and should be added in large databases.

Limitations

This is a retrospective multicentre study investigating a potential selection bias. However, other biases might be present. The results of the selected centres, willing to participate in this study, might be not representative of other centres. Only patients with NSCLC on final histology were included: an analysis including patients without preoperative diagnosis can therefore potentially show different results. We included only cStage-I disease, based on a PET-CT in all patients. We did not provide details on the number of lymph nodes or the number of lymph node stations examined. The number of lymph nodes is highly variable and is not reliable in case of fragmentation [7, 20]. The number of stations is often used to investigate the quality of mediastinal dissection but is less useful for N1 nodes. Nodal levels 10 and 11 are a collection of different nodal locations. For example, the removal of a number 11 node does not equal extensive dissection of all possible inter-lobar nodes. These are further reasons why N1 upstaging is a more attractive quality parameter compared with the number of nodes or stations. Another limitation is that we excluded the pN2 patients from the regression model for N1 upstaging. In the pN2 group, some patients also had positive hilar nodes (pN2 N1+) and some had only skip lesions (pN2 N1−). We previously published a monocentric study with both ‘pN1’ and ‘positive N1 nodes’ (i.e. pN1 and pN2 N1+ combined) as outcomes. A similar effect of central tumour location was demonstrated in both multivariable models [7].

CONCLUSION

More than a quarter of patients with a central cStage-I NSCLC had unforeseen positive N1 nodes at resection. In this retrospective multicentre cohort analysis, relatively more patients with central tumours were operated on by open surgery than by VATS. The apparent difference in the incidence of N1 upstaging (cN0–pN1), which was lower after VATS compared with open surgery, disappeared in a multivariable analysis when including central tumour location. This suggests that the lower incidence of N1 upstaging in large previous studies should be interpreted with caution as central tumour location was not used as an adjusting factor in the analysis. Selection bias with relatively more patients with central tumour location in the open surgery group with a higher chance of N1 upstaging might play an important role.

ACKNOWLEDGEMENTS

We are grateful to all the thoracic surgeons from the different centres, who have contributed with patient data. We thank the members of MITIG for their suggestions to improve this study and article.

Funding

HD is supported by grant from Clinical Research Fund - University Hospitals Leuven and Funds for Research–Flanders (FWO 1701018N).

Conflict of interest: Herbert Decaluwé is speaker for Medtronic; René Horsleben Petersen is speaker for Medtronic, Ethicon and Medela; Jean-Marc Baste is speaker for Medtronic, Ethicon and Intuitive and Henrik Jessen Hansen is speaker for Medtronic, Medela and Bard.

REFERENCES

Author notes