Oral Cavity Cancer: Neck Management

Hi Team,

Hope everyone had a good weekend. We're into the back half of August, which means we're into oral cavity for the next three weeks. For foreshadowing purposes, this week is management of the neck, next week (8/26) is margins, and we'll close out on 9/2 with management of the mandible and adjuvant therapy. As always we're meeting Wednesday mornings @ 7AM sharp and donuts for those who can join in person (I'll make sure I bring breakfast treats this time around!). If Wednesday doesn't work for you, email me or grab any of our head and neck surgeons while you're in the OR and we'll talk it through there.

This week's topic is management of the neck in oral cavity cancer. There is a lot of historical precedent as to why dissect certain levels of the neck (in clinical N0 and in clinically N+ disease). There is also a bit of leeway in which levels you need to dissect depending on how you interpret the literature (ie level 4) and whether you perform a sentinel lymph node biopsy on a cT1N0 oral tongue SCCa.

Why levels I to III? 30 year old papers using 50 year old data.

Shah Cancer 1990 and Medina Head Neck 1989 are the papers that built the selective neck dissection we do every week. Shah mapped metastatic patterns from radical neck specimens and showed oral cavity cancers drain to levels I, II, and III while oropharynx and larynx drain to II, III, and IV. Medina made the operative case for the supraomohyoid neck dissection off the back of it. These are curated retrospective cohorts from two institutions (MSK and MD Anderson) covering patients treated from the 1950s through the 1980s, and they have driven standard of care for three decades. Worth asking yourself when you read them: how long did it take for this to actually become standard, and what would it take today to displace a practice built on this kind of evidence?

In cN0 and low nodal burden cN+ disease, level IV has a bit of controversy. Byers Head Neck 1997 described "skip metastases" in oral tongue where you only have nodal disease in the lower neck. Across 277 previously untreated oral tongue patients at MD Anderson the frequency was 15.8%, but read that definition carefully: it combines isolated level IV disease with cases where level III was the only positive node and levels I and II were clear (how do pathologists define levels?). Of 91 patients whose elective dissection stopped above level IV (retrospective analysis of a third of the patients where there are likely other selection biases), nine recurred in level IV with the primary still controlled, and six of those nine died of neck disease. That is why Byers argued for taking levels I to IV in oral tongue. Modern series are far less alarming. Warshavsky JAMA Otolaryngol 2019 pooled 13 studies and 1,359 cN0 patients, finding overall level IV involvement of 2.53% and true isolated skip metastasis to level IV of 0.50%. TL;DR: level IV is real, it is rare, and it is not a reason to routinely extend a cN0 dissection to level IV. Know the numbers for boards, and know that reasonable surgeons still take level IV on an oral tongue primary with a deep tumor (a lot of time describing this analysis but I think it's demonstrative of how you can take a retrospective review of a few hundred patients and either try to turn it into dogma or try to use it as a nuanced way of taking care of patients).

Elective neck dissection versus watchful waiting

The occult metastasis rate in early oral cavity cancer sits around 30%, which is exactly the range where nobody's intuition is reliable. Huang Cancer 2009 is the meta-analysis that popularized a 4mm tumor thickness cutoff for electively dissecting the neck when someone is cN0. Read it, but read it knowing that tumor thickness is not depth of invasion. Thickness measures from the surface (so an exophytic tumor looks thick without being deep) while DOI measures from the adjacent normal basement membrane. AJCC 8th edition moved us to DOI to better define pT stage (if you use tumor thickness and depth of invasion interchangeably on an oral boards exam you will get burned).

D'Cruz NEJM 2015 is a randomized control trial out of Tata Memorial Hospital in India that has had a big impact on management of the neck in early T-stage oral cavity SCCa. 596 patients with early-stage cN0 oral cancer randomized to elective neck dissection versus therapeutic dissection at relapse. At three years, overall survival was 80.0% with END versus 67.5% with observation, and disease-free survival was 69.5% versus 45.9%. The effect size is large and the trial is well designed. The caveat we always raise: this was a single-center Indian cohort, heavily tobacco and betel quid exposed, with tumors and follow-up patterns that do not map cleanly onto the patients we see in West Philadelphia (or many other regions of the USA). It is still the best randomized data we have, and it is why our default at Penn is to dissect the neck with some additional weight given to depth of invasion (4mm for oral cavity and a lower threshold for subsites like the floor of mouth based on Penn data and MSK data).

Sentinel Lymph Node Biopsy - Non-standard in the US, Standard in Europe

Civantos JCO 2010 (the ACOSOG multi-institutional trial) and Schilling Eur J Cancer 2015 (SENT) are the prospective work establishing that sentinel node biopsy can stage the cN0 neck. Civantos: 140 T1-T2 cN0 patients across 25 institutions, negative predictive value 94% on routine H&E rising to 96% with step sectioning and immunohistochemistry, false-negative rate 9.8% overall (10% tongue, 25% floor of mouth), and NPV 100% for T1 lesions. SENT is the larger dataset: 415 patients at 14 European centres, sentinel node found in 99.5% and positive in 23%, sensitivity 86% and NPV 95%, false-negative rate 14%, disease-specific survival 94%, and only 3% minor complications. Many of the SENT investigators in Denmark and the Netherlands consider SLNB standard of care and would find our routine END mildly barbaric. We continue to perform elective neck dissection though! For a taste of why the argument persists, Naegeli-Pullankavumkal Cancers 2025 is a recent single-center retrospective comparing SLNB (n=55) to END (n=27) in T1-T2 tongue and floor of mouth. Recurrence-free survival was not statistically different overall, but the SLNB arm had 23.6% recurrence versus 7.4% in the END arm and, in the pN+ subgroup, the difference approached significance. This is small, retrospective, and underpowered, so do not over-read it. Do notice that every one of these single-institution series lands somewhere different, which is precisely the problem a randomized trial exists to solve (see HN006 below).

Nodal Burden and Nodal Yield

Ho JCO 2017 looked at metastatic nodal burden and found that the raw number of positive nodes tracks with survival in a continuous, dose-dependent way rather than in the categorical buckets our staging system uses. The finding itself is unsurprising. The interesting question is the one underneath it: how do we decide which variables earn a place in a staging system, and how much prognostic resolution are we willing to trade for something a surgeon can remember?

On the quality side, Ebrahimi Cancer 2011 and Divi JCO 2016 (using RTOG 9501 and 0234 data) both support nodal yield as a proxy for adequacy of dissection, with roughly 18 nodes as the commonly cited minimum number that should be dissected in a neck. Finally, Koyfman JCO 2019 is the ASCO guideline synthesizing all of this (including oropharyngeal cancer), and is a useful read to dig a bit deeper, especially if you look through the 179 references in that paper.

Future directions

NRG-HN006 is the trial that should settle SLNB versus END. The phase II portion completed accrual in December 2024 with 261 patients, met its criteria, and the trial re-opened to phase III in November 2025, seeking 425 additional patients for a total of 686. It carries co-primary endpoints: superior patient-reported neck and shoulder function at six months with SLNB, and non-inferior disease-free survival. Dr. Stephen Lai, one of the driving investigators, is a Penn residency graduate.

Looking further out, Lin and Myers Curr Oncol Rep 2026 review where prediction of occult nodal disease is heading: multimodal models fusing radiomics, digital pathology, and methylation signatures into what they call a "panoramic virtual biopsy." Performance beats imaging alone in early work. Whether it survives external validation is the open question, and for the informatics-minded among you, the barriers they list (data heterogeneity, class imbalance, algorithmic opacity) are the ones every clinical prediction model hits.

Thanks for taking the time to read this and looking forward to discussing further!

Have a great week!

Rob