Plaveiselcelcarcinomen van de mondholte metastaseren primair lymfogeen naar de lymfeklieren van de hals. Aangetoonde metastasen worden doorgaans behandeld met een halsklierdissectie. Voor kleine mondholtetumoren (T1 en T2) die klinisch en radiologisch een negatieve hals laten zien (cN0) en waarbij de hals niet geopend hoeft te worden voor resectie van de primaire tumor of reconstructie van het defect zijn verschillende strategieën mogelijk aangaande behandeling van de hals. Er is bekend dat 30% van deze patiënten occulte metastasen heeft. Er kan gekozen worden voor een electieve halsklierdissectie dan wel voor “watch-and-wait” strategie waarbij de hals wordt opgevolgd bijvoorbeeld met echografische controle en pas behandeld wordt bij een manifeste metastase. Elke strategie heeft zijn eigen voor- en nadelen. Bij een electieve halsklierdissectie worden occulte metastasen direct verwijderd. In geval van grote chirurgische ingrepen zoals commando-procedures met of zonder reconstructies, wordt ongeacht de stadiering van de hals meestal een halsklierdissectie uitgevoerd vanwege de noodzakelijke chirurgische benadering. Bij een “watch-and-wait” strategie ondergaan minder patiënten een behandeling van de hals, maar bestaat het risico dat door de verlate diagnose van eventuele lymfekliermetastasen de behandeling uitgebreider moet zijn dan bij een electieve halsklierdissectie (bijvoorbeeld een gemodificeerd radicale halsklierdissectie met opofferen van structuren in de hals (bijvoorbeeld vena jugularis interna, nervus accessorius en/of musculus sternocleidomastoideus). Ook kan een niet-juiste stadiering van de hals leiden tot het niet uitvoeren van een (selectieve) halsklierdissectie of adjuvante radiotherapie of zelfs dat de metastasen door de delay van het ontdekken inoperabele ziekte zou kunnen ontstaan. Verder moet een afweging moet gemaakt worden op basis van onder andere morbiditeit van een halsklierdissectie, verwijderen van een vangnet voor metastasen van een recidief of tweede primaire tumoren, gebruik van (schaarse) middelen en kosten.

Description of studies

Ibrahim (2020) performed a systematic review and included 24 articles, of which 4 were RCTs (Fakih, 1989; Kligerman, 1994; Yuen, 2009; D’Cruz, 2015). An exact search date was not provided and it was unclear how keywords were combined in the search. Excluded studies in the full-text selection phase were not described or referenced and reasons for excluding literature were not provided. Data was extracted by two authors independently and was cross-checked. The authors did not perform a risk of bias appraisal of the included studies. Potential publication bias was assessed and the authors reported that there was no evidence of publication bias.

Massey (2019) included the same four RCTs. An exact search date was not provided, however it was reported that databases were searched from inception to June 2018. It was unclear how the key words were combined in the search. Excluded studies in the full-text selection phase were not described, however reasons for exclusion were provided in the flow-diagram describing the study selection. Some studies containing overlapping datasets were referenced and excluded in some analyses. Three authors extracted data independently which were checked for consistency. Two authors independently assessed the risk of bias of RCTs by using the Cochrane risk of bias tool. Discrepancies in data-extraction and risk of bias assessments were resolved by consensus. Massey (2019) also included one RCT that recruited patients with T3 disease, which was excluded for the current data-analysis. The authors assessed the potential for publication bias and reported that no publication bias was detected.

Study and/or sample characteristics reported by Ibrahim (2020) and Massey (2019) are summarized in Table 1. Other data about potentially important prognostic factors in the sample (such as depth of invasion, perineural invasion, extracapsular spread) was not reported in the systematic reviews. Data about the watch and wait procedures were extracted from the original RCT papers, separately from the systematic reviews.

Table 1 Summary of study and sample characteristics as reported in Ibrahim (2020) and/or Massey (2019). Data from the watch and wait strategy was extracted from the original papers (Fakih, 1989; Kligerman, 1994; Yuen, 2009; D’Cruz, 2015)

Results

Neck recurrence

Ibrahim (2020) pooled data on neck recurrence from the four included RCTs which compared an elective neck dissection (45 events from n=343) to a watch and wait management (158 events from n=361). A pooled relative risk of 0.36 was found (95%CI: 0.20-0.66, random effects, I² = 69.24%), favoring an elective neck dissection. For the pooled risk difference, a fixed methods meta-analyses by Ibrahim (2020) found a difference of -0.302 (95%CI: -0.364 to -0.240, I² = 0%), favoring an elective neck dissection. A fixed-effects meta-analysis of risk differences by Ibrahim (2020) found a pooled risk difference of -0.302 (95%CI: -0.364 to -0.240, I² = 0.0%), favoring elective neck dissection.

Survival

Four randomized controlled trials reported several survival outcomes at different time-points (Fakih, 1989; Kligerman, 1994; Yuen, 2009; D’Cruz, 2015). Data is summarized in Table 2.

Table 2 Overview of survival outcomes

Mortality

Klingerman (1994) reported 7 deaths (7/34, 20.6%) in the elective neck dissection group (n=4 death related to the disease, n=2 death due to new primary tumor, n=1 unknown cause) and 15 deaths (15/33, 45.5%) in the observation group (n=14 disease related deaths (of which 2 brain and pleural metastases), n=1 heart failure) over the course of 3 years. From this data we calculated a relative risk of 0.45 (95%CI: 0.21-0.97), favoring an elective neck dissection.

D’Cruz (2015) reported 50 deaths in the elective neck dissection group (50/243, 20.6%) and 79 deaths in the observation group (79/253, 31.2%) over the course of the study (median follow-up: 39 months, IQR: 16-76). From the reported data, a relative risk of 0.66 (95%CI: 0.48-0.90) was calculated, favoring elective neck dissection.

Level of evidence of the literature

The level of evidence regarding the outcome measure neck recurrence was downgraded by 2 levels because of study limitations (2 level for risk of bias: Massey (2019) indicated that two RCTs had a moderate risk of bias and two RCTs had a high risk of bias. However, the authors did not provide reasons or appraisals on the individual domains of the risk of bias tool); publication bias was not assessed (although publication bias was assessed by the authors of both systematic reviews by using both randomized and non-randomized studies together (Ibrahim, 2020; Massey, 2019), we included 4 RCTs which is not enough to reliably assess publication bias).

The level of evidence regarding the outcome measure overall survival was downgraded by 1 level because of study limitations (1 level for risk of bias: Massey (2019) indicated that the RCT had a high risk of bias. However, the authors did not provide reasons or appraisals on the individual domains of the risk of bias tool, however blinding is one of the domains in the tool which probably had a high risk of bias considering the 4 RCTs did not report any procedures regarding blinding. Blinding does not affect a hard outcome such as overall survival, and therefore we only subtracted 1 level); number of included patients (1 level for imprecision: event calculator revealed that for HR= 0.64 about 158 events were needed (alpha=0.05, beta=0.2, q1=0.49, q0=0.51, relative hazard = 0.64), over the course of the whole study 129 events were reported; furthermore the confidence interval crossed the pre-defined border of 0.7); publication bias was not assessed (although publication bias was assessed by the authors of both systematic reviews by using both randomized and non-randomized studies together (Ibrahim, 2020; Massey, 2019), we included 1 RCTs which is not enough to reliably assess publication bias).

The level of evidence regarding the outcome measure disease-free survival was downgraded by 2 levels because of study limitations (2 level for risk of bias: Massey (2019) indicated that one RCT had a moderate risk of bias and 2 RCTs had a high risk of bias. However, the authors did not provide reasons or appraisals on the individual domains of the risk of bias tool); conflicting results (inconsistency); applicability (bias due to indirectness); we did not downgrade for number of included patients (imprecision, because: event calculator revealed that for HR= 0.45 about 49 events were needed (alpha=0.05, beta=0.2, q1=0.49, q0=0.51, relative hazard = 0.45), furthermore the confidence interval of the largest RCT did not cross the pre-defined border of 0.7); publication bias was not assessed (although publication bias was assessed by the authors of both systematic reviews by using both randomized and non-randomized studies together (Ibrahim, 2020; Massey, 2019), we included 3 RCTs which is not enough to reliably assess publication bias).

The level of evidence regarding the outcome measure disease-specific survival was downgraded by 2 levels because of study limitations (2 level for risk of bias: Massey (2019) indicated that the RCT had a moderate risk of bias. However, the authors did not provide reasons or appraisals on the individual domains of the risk of bias tool); number of included patients (2 levels for imprecision: the study only contained a sample of n=71); publication bias was not assessed (although publication bias was assessed by the authors of both systematic reviews by using both randomized and non-randomized studies together (Ibrahim, 2020; Massey, 2019), we included 1 RCT which is not enough to reliably assess publication bias).

The level of evidence regarding the outcome measure mortality was downgraded by 2 levels because of study limitations (1 level for risk of bias: Massey (2019) indicated that the two RCTs had a high risk of bias. The authors did not provide reasons or appraisals on the individual domains of the risk of bias tool, however blinding is one of the domains in the tool which probably had a high risk of bias considering the 4 RCTs did not report any procedures regarding blinding. Blinding does not affect a hard outcome such as mortality, and therefore we only subtracted 1 level); number of included patients (1 level for imprecision: Confidence intervals cross the pre-defined border of 0.75); publication bias was not assessed (although publication bias was assessed by the authors of both systematic reviews by using both randomized and non-randomized studies together (Ibrahim, 2020; Massey, 2019), we included 2 RCTs which is not enough to reliably assess publication bias).

A systematic review of the literature was performed to answer the following question:

What are the (un)beneficial effects of elective neck dissection on neck recurrence, survival, and mortality compared to an observation strategy in patients with cT1-2N0 oral cavity carcinomas?

P: patients with a resected primary cT1-2N0 oral cavity carcinoma;

I: Elective neck dissection;

C: watchful waiting;

O: neck recurrence, overall survival, disease-free survival, disease-specific survival, mortality.

Relevant outcome measures

The guideline development group considered neck recurrence and overall survival as a critical outcome measure for decision making; and disease-free survival, disease-specific survival, and mortality as an important outcome measure for decision making.

A priori, the working group did not define the outcome measures listed above but used the definitions used in the studies.

The working group defined a minimal clinically relevant difference as:

• 0.8 or 1.25 as borders for clinical decision-making for risk or odds ratios of neck recurrence.
• 5% difference or more (absolute) and HR< 0.7 in disease-specific survival.
• 5% difference or more (absolute) and HR< 0.7 in overall survival.
• A difference of 10 points on the quality of life instrument EORTC QLQ-C30 or a difference of a similar magnitude on other quality of life instruments.
• Statistically significant less complications/adverse events.
• Statistically significant better functional outcomes, work participation.

Search and select (Methods)

The working group identified two published systematic reviews that included studies comparing elective neck dissection to watchful waiting (Ibrahim, 2020; Massey, 2019).

The combined reporting of both systematic reviews was used and only data from randomized studies were considered, even though both systematic reviews included observational studies.

Ibrahim (2020) searched Medline, Google Scholar and Scopus for relevant studies published from 1989 up to 2018. Keywords used in the search were reported in the published manuscript (Ibrahim, 2020). Studies were selected when published in English, patients had pathologically proven T1-2N0M0 oral squamous cell carcinoma (oral tongue, buccal mucosa, hard palate, alveolar ridge, floor of mouth, retromolar trigone), patients had a clinically and radiologically negative neck, patients only received surgical treatment without previous neck radiotherapy, and when elective neck dissection was compared to watchful waiting. Exclusion criteria were not reported. Over 1200 studies were identified by Ibrahim (2020) and, finally, 24 studies were selected (including 4 randomized studies).

Massey (2019) searched Medline, Google Scholar, and Scopus up to June 2018. Keywords used in the search were reported in the published manuscript (Massey, 2019). The reference lists of systematic reviews and meta-analyses deemed relevant were hand-searched. Two authors independently screened articles. Potentially relevant articles were assessed in detail by the two reviewers independently. Studies were selected when patients had T1-2N0M0 (clinical negative nodes) squamous cell carcinoma of the oral cavity with a verification of diagnosis, when the sample size was 30 or more, the oral cavity was defined in accordance with the AJCC or UICC, the TNM staging (AJCC or UICC) was used, patients did not have prior head and neck treatment (surgery, radiotherapy, chemotherapy), occult lymph node metastasis was clearly defined (as: the presence of a metastasis in the sampled lymph node of a clinically disease-free neck at elective neck dissection), and the techniques of neck dissection were well defined (as: a neck dissection in a patient without clinically detectable disease). Studies were excluded when cancers of the oral cavity were non-squamous cell, the primary tumor site was outside the oral cavity, patients had signs of clinical nodal disease, or when the study focused on laboratory application. Massey (2019) found 1840 search hits, eventually leading to the inclusion of 39 studies (including 5 randomized studies).

Results

Two systematic reviews (Ibrahim, 2020; Massey, 2019) were used to identify relevant randomized trials to be included in the analysis of the literature. Both systematic reviews included the same four randomized studies (Fakih, 1989; Kligerman, 1994; Yuen, 2009; D’Cruz, 2015). Massey (2019) included one additional randomized study, published in 1980, which included patients with T3 carcinomas. Ibrahim (2020) did not include that study because of their search period (from 1989 up to 2018). Due to the inclusion of T3 disease in the study’s sample (21% in the elective neck dissection group; 6% in the therapeutic neck dissection group), it was decided not to consider the study in the current analysis of literature (Vandenbrouck, 1980). Important study characteristics and results are summarized in the evidence tables from the combined reporting of the two systematic reviews. The assessment of the risk of bias of the systematic reviews is summarized in the risk of bias tables. The risk of bias assessment for the individual randomized studies as reported by Massey (2019) was used for the GRADEing of outcomes.