Systematic review and meta‐analysis of cervical metastases in oral maxillary squamous cell carcinoma

Abstract Background Management of the node‐negative neck in oral maxillary squamous cell carcinoma (SCC), encompassing the hard palate and upper alveolar subsites of the oral cavity, is controversial, with no clear international consensus or recommendation regarding elective neck dissection in the absence of cervical metastases. Aim To assess the occult metastatic rate in patients with clinically node negative oral maxillary SCC; both as an overall metastatic rate, and a comparison of patients managed with an elective neck dissection at index surgery, compared to excision of the primary with clinical observation of the neck. Methods and results A systematic review was performed by two independent investigators for studies relating to oral maxillary SCC and analysed according to PRISMA criteria. Data were extracted from Pubmed, Ovid MEDLINE, EMBASE, and SCOPUS via relevant MeSH terms. Grey literature was searched through Google Scholar and OpenGrey. Five hundred and fifty‐three articles were identified on the initial search, 483 unique articles underwent screening against eligibility criteria, and 29 studies were identified for final data extraction. Incidence of occult metastases in patients with clinically node negative oral maxillary SCC was identified either on primary elective neck dissection or on routine follow up. Meta‐analyses were performed. Of 553 relevant articles identified on initial search, 29 were included for analysis. The pooled overall rate of occult metastases in patients initially presenting with clinically node‐negative disease was 22.2%. There is a statistically significant effect of END on decreasing regional recurrence demonstrated in this study (RR 0.36, 95% CI 0.24, 0.59). Conclusion The results of this systematic review and meta‐analysis suggest elective neck dissection for patients presenting with hard palate or upper alveolar SCC, even in a clinically node negative neck.


| INTRODUCTION
Squamous cell carcinoma (SCC) is the most common malignant tumour within the oral cavity. The primary modality of therapy is surgical excision, with post-operative radiotherapy for advanced stage tumours or early-stage tumours with adverse features. 1 The presence of nodal metastases is thought to be the most significant prognostic factor in head and neck SCC, 2 and management of the clinically node negative (cN0) neck is guided by the risk of occult metastases or undetectable micro metastases. Based on modelling performed by Weiss et al., recommendations have suggested a 20% risk of cervical metastases as the threshold for treating the N0 neck with elective neck dissection. 3 Management of the cN0 neck with selective neck dissection of levels I-III is now well established for most oral cavity subsites, with two large randomised controlled trials in recent years proposing elective neck dissection even in early-stage oral cavity SCC. 4,5 The advantages of elective neck dissection include accurate prognostication, and assessment for the requirement for additional therapy such as radiotherapy and chemotherapy. 6,7 In addition to the prognostic information provided by nodal staging, the presence of extracapsular spread and multiple involved lymph nodes are both adverse prognostic features. 8 However, management of oral maxillary SCC, encompassing the hard palate and upper alveolar subsites of the oral cavity, presenting with the cN0 neck remains controversial due to a comparative lack in data, and a low metastatic rate in this cohort relative to other oral cavity subsites. 2 Traditionally, management of the cN0 neck has largely been conservative, with close surveillance. 2 The maxilla is thought to have limited lymphatic drainage compared to the rich lymphatics of the other subsites of the oral cavity, and oral cavity tumours of the maxilla are thought to be biologically similar to maxillary tumours arising in the sinonasal cavity, where elective neck management is classically not performed. 2 It has been suggested on recent studies that hard palate SCC behaves as aggressively as other oral cavity subsites, and that the cN0 neck should be managed surgically. [9][10][11] Given the persisting controversy, this study aims to assess the occult metastatic rate in patients with cN0 oral maxillary SCC; both as an overall metastatic rate, and a comparison of patients managed with an elective neck dissection at index surgery, compared to excision of the primary with clinical observation of the neck.

| Study identification
An initial search of published literature from database inception to 01 August 2020 was performed using Pubmed, Ovid Medline, Embase, and SCOPUS databases for all English-language literature. There were no restrictions to age, sex, or ethnicity applied in this search strategy.
Grey literature was searched through Google Scholar and OpenGrey.
The keywords 'maxilla', 'alveolar process', 'hard palate', 'squamous cell carcinoma of the head and neck', 'head and neck neoplasms', 'cervical metastases', and 'neck dissection' or 'lymph node excision' were mapped to MeSH terms and searched. A detailed search strategy for each database is available as a supplement (Supplement 1).
The PRISMA flowchart of the study is presented in Figure 1. Two independent reviewers (CC, SK) screened titles and abstracts of the retrieved articles in the initial screening phase. Full articles were obtained for relevant studies, and studies in which the title and abstract provided insufficient information. The individual reference lists of included articles and of similar literature reviews [12][13][14][15] were also screened for potential inclusion in the next phase.

| Study selection
Studies were included according to the following criteria: (1) English language literature, (2) oral cavity tumours of the hard palate, upper alveolar ridge, and maxillary gingiva (3) primary treatment included surgical resection, (4) sample size was greater than five patients.
Studies were excluded based on the following criteria: (1) case reports or small case series with a sample size of five or less, (2) maxillary tumours not originating from the oral cavity subsite of interest, and (3) data was inseparable due to the heterogeneity of patient population. Insufficiently discriminatory data is defined as study data wherein the outcomes of cN0 patients who would otherwise fit inclusion criteria could not be discriminated from clinically node positive (cN+) tumours, tumours invading other sites in the oral cavity, and tumours which could not be discriminated from non-SCC tumours. All patients who had surgical treatment of their primary were included, regardless of radiotherapy status.
Both reviewers (CC, SK) reviewed the full text of each article, and selected studies for data extraction based on inclusion and exclusion criteria. Both reviewers discussed any differences or discrepancies between final selection, and any further discrepancies were managed in consultation with a third reviewer (ZH).
All studies which fit the inclusion criteria were included into the systematic review. Studies that provided individual patient-level regional recurrence data stratified into upfront neck dissection and observation subgroups were included into the final meta-analysis.

| Outcomes
The occult cervical metastatic rate of clinically node negative (cN0) patients, determined based on clinical examination and radiological findings, was extracted in three separate subgroups: 1. Occult metastatic rate on pathologically examined nodes following upfront neck dissection (cN0pN+) and; 2. Occult metastatic rate on cN0 patients with upfront neck dissection on long term follow-up (cN0 and END) and; 3. Occult metastatic rate on cN0 patients without neck dissection on long term follow-up (cN0 and no END).
Occult metastases were identified in two ways: (1) a neck dissection performed electively at initial surgery for cN0 disease, revealing pN+ disease; and (2) patients with cN0 disease on initial presentation, then presenting with cN+ disease on follow up, after primary surgery. cN0 patients presenting with delayed locoregional recurrence were not included in calculations, as this was thought to be a recurrence of the initial primary, rather than occult nodal metastasis manifesting in delayed cervical lymphadenopathy. The occult metastatic rate at two time points, on initial presentation, and on follow-up, was calculated.

| Secondary outcomes
Predictive and prognostic factors were identified and recorded as secondary outcomes for subgroup analysis: site of primary tumour, early (T1/2) and late (T3/4) stage disease, pathological grade, and survival outcomes.

| Data extraction
The following data points were extracted: basic demographics, TNM stage and grade, nodal stage, and follow-up data. In patients treated with END, pathological nodal status at initial upfront neck dissection was recorded (cN0pN+). As individual studies defined and reported the rates of occult metastasis differently, published data were examined and classified according to this study's definition of occult metastases. For all follow-up data on our three groups, survival outcomes and nodal recurrence including time to regional recurrence were analysed. Wherein clinical and pathological TNM data was available, clinical tumour staging was used for consistency between studies.
Data was pooled to give an overall follow-up occult metastasis rate across the entire study.
To prevent patient duplication, cN0pN+ patients who proceeded to have regional recurrence were counted as a single event. Data which could not be discriminated between included and excluded patients were excluded from data analysis, such as those where regional recurrence could not be discriminated between patients initially presenting as cN0 and cN+, or wherein hard palate data was mixed with other oral cavity cancer subsites.
Large database studies 16,17 were excluded due to the heterogeneity of data and inability to extract accurate and sufficient follow up data. Case reports [18][19][20][21][22][23] were excluded due to insufficient sample size.
The relative risk, odds ratio (OR), risk ratio (RR), or pooled incidence were used as summary statistics and reported with 95% confidence intervals (CI). Meta-analyses were performed using random-effected models to take into account the anticipated clinical and methodological diversity between studies. The I 2 statistic was used to estimate the percentage of total variation across studies due F I G U R E 1 PRISMA flow diagram to heterogeneity rather than chance, with values exceeding 50% indicative of considerable heterogeneity.
Publication bias was visually assessed using funnel plots comparing logit of event rates with precision. Egger's linear regression method was used to quantitatively assess for funnel plot asymmetry, 24 and the Trim-and-Fill method was used to explore the impact of studies potentially missing due to publication bias. 25 Statistical analysis was conducted with Review Manger Version 5.3 (Cochrane Collaboration, Oxford, UK) ad Comprehensive Metaanalysis v3.0 (Biostat Inc., Englewood, USA). All P-values were twosided, and values <0.05 were considered statistically significant.

| Quality appraisal
Risk of bias was assessed by two independent reviewers with the ROBIN-I tool for the 16 retrospective cohort studies included in meta-analysis. 26 Both reviewers discussed any differences or discrepancies, and a graphical representation of the final results was created with the ROBVIS tool. 27 As all studies included in the meta-analysis are retrospective observational studies of patient records, general flaws of retrospective cohort studies such as the assumption of accurate recordkeeping apply to studies which did not undergo formal risk of bias assessment. The outcomes of interest, the incidence of nodal disease, and the method of determining regional metastasis on followup, are applicable across most tertiary centres.

| RESULTS
A total of 553 articles were identified from an initial search: 440 from database searching, five systematic reviews of oral cavity squamous cell cancers with subgroup data specific to upper alveolar subsites; and 38 from hand searching reference lists of these five systematic reviews. Four hundred and eighty-three unique articles were selected for screening against inclusion and exclusion criteria, and 133 articles met the criteria for full text review. Twenty nine studies underwent data extraction; all were retrospective cohort studies (Oxford Centre for Evidence-Based Medicine level 3), 28 and no randomised controlled trials were identified ( Figure 1). As all articles selected for full-text screening and data extraction were available, no authors were contacted. There was moderate risk of bias of 16 retrospective cohort studies included in meta-analysis (Figure 2), based on a detailed breakdown of individual studies (Supplement 2).
Data pertaining to cN0 patients (n = 1644) from 2001 to 2020 were extracted. These were further analysed in subgroups to examine desired outcomes ( Table 1). The sample size and event rate for each data set was recorded ( Table 2). Of 1905 patients presenting with maxillary SCC, 86.30% were determined to be clinically node negative.
Of all cN0 patients undergoing elective neck dissection, the occult metastatic rate at initial presentation (cN0pN+) is 18.62% (n = 92/ 494). When these cN0pN+ patients are added to the 261 patients presenting with cN+ disease, the true incidence of nodal disease on initial management is 18.53% (n = 353/1905, 29 studies).
For patients with upfront elective neck dissection who were found to be pathologically node negative (cN0pN0), the follow-up recurrence rate was 3.44% (n = 17/494). Hence, the total incidence of nodal disease in patients with upfront END was 22.06% (n = 109/ 494). For cN0 patients who did not undergo END, the incidence of There is a statistically significant effect of END on decreasing regional recurrence (RR 0.36, 95% CI 0.24, 0.59) ( Figure 5(A)).
The reported follow-up period of cN0 patients was pooled to give an average of 50.46 months (SD ±13.94 months, n = 457). 10  Clinically-node negative (cN0) patients who received upfront elective neck dissection (END), who were found to be pathologically node negative (cN0pN0), and then developed regional recurrence.  (Figure 5(B)).
Five studies reported pathological grading for 306 cN0 patients (Table 5). 38 therefore be postulated that no patient was excluded from END due to being medically unfit for surgery. Additionally, there was a paucity of data relating to specific node levels involved in occult cervical metastases; this was also not able to be reliably extrapolated or analysed from the included sample size.
There was overall moderate risk of bias for the 16 studies included into the final meta-analysis ( Figure 2). Selection of patients for elective neck dissection appeared inconsistent across the studies analysed, reflecting the lack of a general consensus regarding management of this cohort. Eight studies did not report their criteria in selecting patients for END. 11,[31][32][33][34]36,37,44 Reported indications for neck dissection ranged from surgeon preference, 9,10,29,45,50 prophylactic clearance as part of a flap reconstruction, 10,30 and intraoperative findings. 30 There was a statistically significant difference in operative technique for one of the studies where free-flap reconstruction was more likely in the END group. 10 One study used T4 staging as a factor in choosing to perform END, however only one of six T4 patients had END. 40 Additionally, due to the role of MDT and individual clinician decisions regarding management following primary resection, patients in the non-operative arm were either treated with a watch and wait approach, or via radiotherapy in five studies. 10,33,34,36,37 Additionally, pathological findings of perineural involvement 40   which reported an occult metastatic rate of the cN0 neck of 21% at index surgery, with an overall metastatic rate of 32%, including both cN0pN+ and cN+ patients. 13 Differences in results between the two systematic reviews can be attributed to differences in study design.
Utilising a different study design and tighter exclusion criteria allowed Two cross-sectional studies derived from large reporting databases were found on initial literature search pertaining to occult metastatic rates in cN0 patients at initial diagnosis; however, no follow-up data could be extrapolated from these studies and as such were excluded from meta-analysis to avoid patient duplication. Lin et al.
reported an incidence of cN0 disease at 86.3% from 725 patients, similar to the 86.5% incidence reported in this study. Ninety-nine patients (13.6%) in this study with either cN0 or cN+ oral maxillary cancer had regional recurrence, however due to the non-discrimatory nature of the reported data between regional recurrence in node positive and node negative patients, these patients were excluded from meta-analysis. 17 Obayemi et al. reported a rate of 14% (n = 59/422) occult metastasis upon neck dissection at primary resection. However, statistical analysis of the study concluded that after controlling for tumour, patient and treatment factors, patients who underwent END had statistically significant improved OS outcomes over an 11-year period (HR 0.74, P = 0.002). 16 There are several limitations to this review. Many of the studies analysed had were small, retrospective studies with varying levels of heterogeneity in different data points across different studies. Furthermore, data points such as age and gender could not be collected.
Although it is hoped this study will influence clinical management of these patients, higher level evidence may come from prospective multi-institutional data which are currently lacking.

| CONCLUSION
The data in this systematic review and meta-analysis of 29 articles validate the role of elective neck dissection in hard palate and upper alveolar SCC even in the absence of cervical metastases.

CONFLICT OF INTEREST
None of the authors have any conflicts of interest or funding sources to disclose.

AUTHORS' CONTRIBUTIONS
All authors had full access to the data in the study and take responsi-

ETHICAL STATEMENT
Not applicable.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.