Should a neck dissection be performed on patients with cN0 adenoid cystic carcinoma? A REFCOR propensity score matching study

Sarah Atallah , Antoine Moya-Plana , Olivier Malard , Gilles Poissonnet , Nicolas Fakhry , Jérémie Bettoni , Patrice Gallet , Pierre Ransy , Sebastien Vergez , Ludovic De Gabory , Diane Evrard , Maria Lesnik , Pierre Philouze , Chloé Bertolus , Philippe Schultz , Sylvain Morinière , Vianney Bastit , Christian A. Righini , Haitham Mirghani , Franck Jegoux , Olivier Mauvais , Julian Biau , Xavier Dufour , Nicolas Bon-Mardion , Esteban Brenet , Benjamin Verillaud , Bertrand Baujat *


Background
Adenoid Cystic Carcinoma (ACC) accounts for 1% of head and neck cancers and 10% of salivary gland cancers [1].
It is characterised by slow evolution, rare lymph node invasion and high rate of local and distant recurrences, which make it a deadly tumour [2].
The therapeutic strategy remains unconsensual, with the exception of surgical management at the localised and resectable stage [3]. Guidelines are based on mostly retrospective series [4,5].
ACC is often described as having relatively infrequent lymphophilic extension. Lymph node metastases (LNM) rates vary from 4% to 33% depending on the tumour sites [6e8]. LNM is a factor of negative prognosis [9e12].
The need for prophylactic neck dissection (ND) on cN0 patients is therefore questionable: the few studies carried out so far have not shown any benefit in terms of overall survival (OS) when ND was performed [14].
The main objective of our study is to evaluate the impact on EFS of ND for cN0 ACC patients. The secondary objectives are to study this impact on local recurrence-free survival (RFS), metastasis-free survival (MFS) and OS.

Population and database
A multicentre prospective inclusion of incident cases was conducted from January 2009 to February 2018. Data collection was carried out using a standardised questionnaire. Data were integrated into the REFCOR (French National Network on rare head and neck cancers) database. The comparative study protocol was elaborated in 2018. Analyses were performed after updating and quality control of the database.
Missing and inconsistent data were corrected after reviewing the files. A second data quality control was carried out by a checking of the histological reports. Records of patients that were found to be pNþ after dissection were reviewed.
Any patient managed in a REFCOR centre with a confirmed ACC was included in the series.
Only patients without clinical LNM at the diagnostic stage (cN0) who were surgically treated at the primary tumour site were included.
Exclusion criteria were metastatic patients at the diagnostic stage, unknown treatment status and underage patients <18 years (Fig. 1).
In order to take into account the type of postoperative radiotherapy (PORT) and to verify the absence of influence of this treatment on survival in the compared groups, a survey was conducted among the network's radiotherapists.

Propensity score matching
A propensity score analysis was conducted to minimise the selection bias and potential confounding factors between groups.
This score was calculated for each patient as the expected probability of receiving ND, based on a multivariate logistic regression that included the main confounding factors associated with survival: age, body mass index (BMI), tumour site and T stage. These criteria were chosen following a meeting between expert clinicians because of their clinical relevance. In the series, BMI and age were identified as significantly influencing patient prognosis and included de facto. Only variables known at the time of the surgery were eligible [15]. Each patient was weighted by the inverse probability of being part of the group of patients with ND (NDþ) versus the group without ND (NDÀ), in order to balance observable characteristics. The concordance tolerance (caliper) was set at 0.02. Given the presence of missing data, multiple imputation was used using the MICE method (R software). The results obtained on the series before and after imputation were compared in order to validate them. Sensitivity analyses had been scheduled a priori: exclusion of sinus tumours (where ND is only rarely recommended) and exclusion of BMI (12% of missing data).

Evaluation criteria
EFS was defined as the time between diagnosis and local or distant recurrence, death or the date of last follow-up for censored patients.
OS was defined as the time between diagnosis and death, or the date of last follow-up for censored patients.
MFS was defined as the time between diagnosis and distant recurrence, death or the date of last follow-up for censored patients.
RFS was defined as the time between diagnosis and local recurrence, death or the date of last follow-up for censored patients.

Statistical analysis
The categorical variables were characterised by their proportion and compared using Pearson's chi-squared test or Fisher's exact test.
Continuous variables were described by their mean or median and their range. They were then compared using Student's t-test or ManneWhitney test.
All statistical tests were two-sided. A p value less than 0.05 was considered statistically significant.
The KaplaneMeier method and a Cox model were used to analyse EFS, OS, MFS and RFS.

Patient characteristics
Our sample included 322 patients from 33 centres with cN0 ACC, a majority of which were female (58%; sex ratio 1:4).
Smoking and alcohol consumption were associated with 31% and 14% of patients, respectively. Patients had immunosuppression (diabetes or immunosuppressive therapy) in 6% of cases.
The proportion of patients treated for sinus tumours varied among centres, depending on their degree of specialisation in sinus surgery. Half of the patients had advanced T-stages at diagnosis: 20% of T3 and 32% of T4.
All the 322 patients underwent surgery of the tumour site: 46% of them NDþ, 149 patients, versus 173 NDÀ.
Tumour resection margins were positive in 40% of patients and close in 22% of patients.
The presence of a solid component representing >30% of the tumour (histological grade 3), perineural invasion and vascular embols was found in 20%, 68% and 28% of patients, respectively.
Among the 149 patients who had a ND, histological reports of 85 of them could be reviewed. Among those, the results of 5 bilateral ND and 77 unilateral ND could be analysed. The average number of nodes was 22, median was 18 and range was [3e76].
PORT was performed on 75% of patients. Information regarding the type of radiotherapy and the irradiated site (on the tumour bed and/or lymph node bed) was missing in 70% of cases.
For NDÀ cN0 ACC patients, 85% of our surveyed radiotherapists performed prophylactic irradiation (50e60 Gy). In the case of NDþ, 57% of them performed prophylactic irradiation even if pN0.
In addition, 13% of patients received chemotherapy along with PORT.
The characteristics of the two patient groups are presented in Table 1. They were comparable for the majority of characteristics, with the exception of BMI and tumour site.
Two groups matched by propensity score were generated (96 pairs, n Z 194 patients). Matching erased clinical differences between groups (Table 1).

Characteristics of pNþ patients
Seven cN0 patients were found to be pNþ, which implies an occult LNM rate of 5%. Four of them had capsular effraction Table 2.
The pNþ involvement in the patient whose tumour site was parotid was a lymph node invasion by tumour contiguity.
For two patients, tumour invasion was localised on a boundary zone between two anatomical entities. The tumour of the first patient was located between the maxillary sinus and the oral cavity. It infiltrated the palatal mucosa. The second patient's tumour was located between the submandibular gland and the oral cavity. It infiltrated the mouth floor mucosa.
All in all, 6 of the 7 pNþ patients had an invasion of their oral cavity's mucosa.

Survival
The median follow-up was 25 months.

Main judgement criteria
Before the matching analysis, the median EFS of the NDþ group was 72 months (95% Confidence Interval (CI) [55e112]) versus 76 months (95% CI [59e114]) for the ND-group. The 5-year EFS rate for NDþ patients Table 1 Characteristics of patients according to "Neck Dissection"/"No Neck Dissection" group before and after propensity score matching (*p 0.05) (NA Z data not available).   Fig. 2).

Secondary judgment criteria (results expressed after propensity score)
The rates of 5-year survival in the NDþ group compared to the ND-group were:  Fig. 3).

Discussion
Our study found a low rate of occult LNM and no difference in EFS after adjustment on biases between the two groups of patients treated with ND or not. To our knowledge, our study is the first to assess the influence of ND on EFS. It differs from previous studies by analysing survival with a propensity score that allows for a better comparison, despite the lack of randomisation [16]. The relevance of the question is supported by the finding in our series that patient characteristics are balanced between the NDþ and NDÀ groups, suggesting that the decision to perform ND is not based on a standardised attitude.
In our series, the cN þ rate at the time of diagnosis is low (11%, 48 patients). In comparison, Ning's meta-analysis finds a high variability of results (4%e33%) depending on the series, with an average of 16% of cNþ [13].
For the main salivary glands, a literature review suggests a 18.6% rate of LNM. The prevalence of LNM for ACC is 14.5% for the parotid gland, 22.5% for the submandibular gland and 24.7% for the sublingual gland. In this study, cervical LNM occur more frequently in the T3eT4 stage and are mainly found in zones II and III [17].
Few studies present the occurrence rate of occult LNM among the cN0 after ND. It varies from 8% to 23%, with an average of 14% in Ning's meta-analysis [7,13,17,18].
The variability of this rate can be explained by the difference in the proportion of cN0 patients who received ND, the number of lymph nodes collected and the small size of the studied populations [19].
In our study, the occurrence of cN0pNþ among the NDþ patients is low (5%). This rate could be artificially lowered by the fact that our definition of ND does not include the number of lymph nodes sampled due to 45% of missing data. It could also be due to a possible reclassification a posteriori as cNþ by technicians of a few patients initially considered as cN0.
After reviewing the seven records of cN0pNþ patients, it is observed that six of the tumours shared a common characteristic: they invaded the mucosa of the oral cavity. The seventh case corresponded more to an invasion by contiguity. These results are in line with other series [8,20,21]. More unexpected is the high proportion of grade 1 (5/7) among these patients, as histological grade is supposed to predict the risk of node invasion. It is although difficult to make hypotheses out of such a small number of patients.
Current data on ACC were insufficient to determine whether prophylactic ND should be recommended for cN0 patients: studies are mainly retrospective. Several of them show the limited benefit in terms of survival of treating patients with ND [8,14,22]. Others use only the rate of LNM as a basis for recommendations: for the main salivary glands, the International Head/Neck Scientific Group recommends that ND be performed in N0 patients only at the T3eT4 stage [17], whereas for sinus, lacrimal gland and external auricular duct tumours with a low metastasis rate (5.3%), it recommends not performing a ND [23].
For ACC in the oral cavity and oropharynx, an international review recommends that, in the face of a higher occult metastases rate within these sites, ND should be performed when patients have negative prognostic factors and postoperative radiotherapy is not being considered [24]. Our study is in line with this conclusion: involvement of the oral mucosa is probably a risk factor of occult nodal involvement and should be considered in the decision process. For ACC in the larynx, an international review found only 12% LNM and recommends not to perform ND in cN0 patients [25].
Our results are in line with these studies, showing that ND does not bring any benefit in terms of survival for cN0 patients. The trend towards better survival in the ND-group is a strong argument to suggest that this negative result is not only due to a lack of power. The confidence interval of the matched HR (HR Z 1.33; 95% CI [0.82e2.16]) leaves little space for a HR below 1, which would favour the ND.
PORT is often applied to patients with ACC, because of the frequency of negative histological prognosis criteria. It has a positive impact on the survival rate of patients with advanced T-stage and positive margins [26].
In our series, the majority of patients received PORT. In addition, both groups of NDþ versus NDÀ patients had PORT in the same proportion.
A possible explanation for the lack of observed benefit in favour of ND could therefore be linked to PORT, which seems e according to our survey of the network's radiotherapists e to be little influenced by the use of ND: radiotherapists tend to include at least the proximal nodes in their irradiation field, whether or not the cN0 patient has had a ND. This finding supports the absence of a potential bias induced by PORT.
Systematic ND can still be used to stage patients with the pN classification, which can theoretically guide the choice of post-operative treatment. However, it is an excessively invasive procedure for 95% of the cN0 patients in our series: Whether being functional or radical, ND remains a source of post-operative complications and morbidity in 35% of cases [27]. Functional disorders still occur frequently. They are often associated with damage to the spinal nerve, but shoulder morbidity can occur even if the nerve has been preserved [28,29].
However, the presence of LNM is a negative prognostic factor [9,10,12,30]. ND is therefore recommended if LNM are found during the initial assessment [4].
In order not to underestimate cNþ patients, we recommend carrying out a qualitative pre-operative assessment by imaging. Positron emission tomography with 2-18-fluoro-2deoxy-D-glucose coupled with a CT scan (PET-CT-FDG) plays a major role in the detection and staging of patients with head and neck cancer, but most studies focus on squamous cell carcinoma (SCC) [31]. However, ACC absorbs less FDG than SCC [32].
PET-CT-FDG could be an added value in the assessment of disseminated disease for patients with ACC, and therefore in determining treatment planning, but still needs to be evaluated [33].
Sentinel lymph node (SLN) technique was validated for the treatment of tumours of the oral cavity and oropharynx in 2005 by a consensus conference [34]. It enables to target the lymph node procedure and morbidity to be reduced. Our study suggests that it may be advisable to consider proposing the SLN technique for ACC involving the mucosa of the oral cavity, especially because they are the most suitable for this technique [29].
One of the limitations of our study is the number of missing data: it involved seven of the variables with an average of 6%. BMI variable had 12% missing data. Sensitivity analyses were performed by excluding this variable from our model. The results were not altered. Moreover, the analysis of missing data by individual and by variable showed a random distribution of these missing data suggesting the feasibility of multiple imputation [35].

Conclusion
The incidence of occult metastases in patients with cN0 ACC is low.
Patients not treated with ND do not show a difference in EFS compared to those treated with it, as long as post-operative radiotherapy is still considered.
Our results suggest that systematic ND should not be performed on cN0 patients with minor or major salivary gland ACC, except for infiltrating tumours of the oral cavity.
However, it is necessary to carry out a close preoperative assessment of the intervention and quality imaging as well as early post-operative lymph node monitoring.
Evaluation of the SLN technique for patients with oral cavity ACC could be an alternative approach to ND.