Kaplan-Meier plot of probability of observed survival over time from surgery, stratified by combined margin grouping. Group margin definitions: group 1 includes all patients with negative frozen and specimen margins; group 2, all patients without a positive specimen margin who are not included in group 1; group 3, those with positive specimen margins but not positive frozen margins; and group 4, those with positive frozen and specimen margins.
eAppendix. Association of Main Specimen and Tumor Bed Margin Status With Local Recurrence and Survival in Oral Cancer Surgery
eTable 1. Frequency of Margin Results Under Various Definitions of Involved
eTable 2. Margin outcomes for various grouping categories, represented in 2x2 tables
eTable 3. Correlating Frozen Margins to Specimen Margins, including Local Recurrence Rate
eFigure 1. Observed Survival Based on Margin Groups
Customize your JAMA Network experience by selecting one or more topics from the list below.
Buchakjian MR, Tasche KK, Robinson RA, Pagedar NA, Sperry SM. Association of Main Specimen and Tumor Bed Margin Status With Local Recurrence and Survival in Oral Cancer Surgery. JAMA Otolaryngol Head Neck Surg. 2016;142(12):1191–1198. doi:10.1001/jamaoto.2016.2329
Copyright 2016 American Medical Association. All Rights Reserved.
What is the association between oral cavity cancer margins and local recurrence (LR), whether assessed on the main specimen or tumor bed?
In this retrospective cohort study of 406 patients with oral cavity cancer, prognostic stratification using both tumor bed frozen margins and main specimen margins was informative. Patients with involved intraoperative frozen specimens “cleared” by additional resection had 27% LR, which was not statistically different from microscopically positive margins that were not ultimately cleared.
The specimen margin is an important predictor of outcome, even if margins are taken from the tumor bed. Using intraoperative frozen margins to guide clearing margins does not improve prognosis associated with a microscopically positive margin.
There is controversy surrounding surgical margins in oral cavity squamous cell carcinoma (OCSCC), with debate regarding the assessment and prognostic value of margins.
To analyze a large cohort of OCSCC cases for correlation between tumor specimen margins and intraoperative tumor bed frozen margins and evaluate how margin status associates with local recurrence and survival.
Design, Setting, and Participants
Retrospective cohort study of 406 patients treated with OCSCC resection between 2005 and 2014 at the University of Iowa Hospitals and Clinics. Included cases underwent margin evaluation on the tumor specimen and intraoperative frozen margin assessment from the tumor bed.
Main Outcomes and Measures
Findings of intraoperative frozen margin analysis as a test of tumor specimen margins; local recurrence and survival based on margin findings; prognosis based on clearance of positive frozen margins. To evaluate whether additional resection to “clear” positive frozen margins affected prognosis, we compared local recurrence rates for patients in 3 groups: group A included those patients with negative margins on both intraoperative and permanent specimens; group B included those with positive intraoperative margins subsequently cleared by additional resection to negative margins; and group C included those with negative intraoperative but positive permanent specimen margins.
The median age of the 406 patients (234 men and 172 women) was 61 years (interquartile range, 53-72 years). When frozen margins were correlated with tumor specimen margins, frozen margin accuracy was 65%, with a 46% false-negative rate. We observed a local recurrence rate of 36% (95% CI, 24%-49%) when invasive carcinoma was present at an intraoperative frozen margin and 45% (95% CI, 34%-57%) when invasive carcinoma was found on the permanent specimen margin compared with 19% (95% CI, 14%-26%) and 13% (95% CI, 7%-22%) for completely negative frozen and permanent margin findings, respectively. There was a significant difference in local recurrence between group A (13%) and group B (27%) (absolute difference, 14%; 95% CI, 3%-26%) and between group A and group C (34%) (absolute difference, 21%; 95% CI, 8%-34%), but there was no difference between groups B and C (absolute difference, 7%; 95% CI, −8% to 22%), suggesting that additional resection to clear positive frozen margins does not improve prognosis.
Conclusions and Relevance
Intraoperative frozen margins from the tumor bed are not ideal predictors of positive margins on the main specimen. Both frozen and specimen margins are associated with local recurrence, but the specimen margin has the stronger association. Importantly, we demonstrate that clearing positive frozen margins from the tumor bed is not associated with improved outcomes.
There are many points of controversy surrounding the assessment and prognostic use of margins in head and neck cancer surgery. The main limitation of previous retrospective series has been their heterogeneity, with wide variations in analyzed cancer subsites, margin sampling techniques, and definitions of positive and/or close margins. Two published surveys of head and neck surgeons and pathologists demonstrated differences of opinion1,2: 80% of surgeons believed that carcinoma in situ (CIS) should be considered positive; 46% considered that tumor more than 5 mm from the margin should be considered negative; and 69% classified tumor less than 5 mm from the margin as “close.” Ninety-seven percent of surgeons used frozen section margin assessment, with 76% sampling frozen specimens from the tumor resection bed and not the primary specimen, and 90% believing that initially positive margins resected to negative should ultimately be considered negative.
In reviewing the literature over the last 30 years, we found some controversial questions and practice deviations:
Whether a microscopically positive margin has prognostic significance;
Whether frozen section assessment accurately identifies a positive margin;
Whether the tumor bed frozen margin or resection specimen margin should be considered the true margin; and
Whether reexcision can “clear” a positive margin and whether this benefits the patient.
To address these questions, we undertook a retrospective assessment of our institutional experience surgically treating oral cavity squamous cell carcinoma (OCSCC), with specific attention to surgical margin assessment and outcomes of local recurrence (LR) and survival.
Approval for the study and a waiver of informed consent was obtained from the institutional review board of the University of Iowa Hospitals and Clinics. A retrospective review was conducted using the institutional tumor registry, clinic notes, and operative and pathology reports. Included were patients treated with surgical resection from 2005 to 2014 at the University of Iowa Hospitals and Clinics for primary OCSCC. Patients were excluded if their cancer was outside the oral cavity; histologic findings indicated something other than SCC or one of its variants; no invasive cancer was found in the specimen; cancer was recurrent; resection was not en bloc or not for curative intent; or gross disease remained after surgery. There were several patients with multiple OCSCCs over the study period, and these were included as distinct cancers provided they did not occur in adjoining sites.
Surgical practice involved tumor resection with attempted 3-dimensional 1-cm margins whenever possible. If tumor approached bone without clinical evidence of invasion, a conservative bone resection was performed; if bone invasion was seen, a radical bone resection was performed. Intraoperative margin assessment was performed from the edges of the resection (tumor bed) or directly from the tumor specimen. Almost all patients had intraoperative margins assessed from the tumor bed; therefore, we excluded the few patients with intraoperative margins from the specimen. If a margin was involved, further resection was performed in an attempt to obtain an uninvolved margin. We collected information about frozen section margins, including whether reexcision of tissue occurred, presence of tumor in reexcised tissue, and assessment of final margins based on frozen specimens and any additional resections and margins. For each frozen section, an intraoperative and final diagnosis was given; change from negative to dysplasia or CIS, or dysplasia or CIS to invasive carcinoma was defined as a false negative, while change from invasive carcinoma to dysplasia or CIS, or dysplasia or CIS to negative, was defined as a false positive.
A final margin assessment of the main tumor resection specimen after formalin fixation was also reported. For this study, margins were classified as positive if invasive cancer was present at the inked edge, very close if it was less than 1 mm from the edge, close if between 1 and 5 mm from the edge, and negative if 5 mm or more from the edge. If dysplasia or CIS was present at the margin, this was noted; mild and moderate dysplasia were classified together, as were severe dysplasia and CIS.
Outcome data including recurrence and survival were obtained from the tumor registry and verified via medical chart review. Local recurrence was defined as invasive cancer within 5 years of surgery at a contiguous site. Regional and distant recurrences were also noted. Patients without recurrence who did not survive 6 months following surgery were not included in calculating LR. All survival analyses were measured from the surgery date to the date of death or censor (last known follow-up).
Measures of test performance including sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV), were calculated according to standard definitions. Local recurrence was calculated as a proportion, and survival was estimated with the Kaplan-Meier method. Effect size (absolute difference in LR, relative risk, adjusted odds ratio [OR] for LR, and hazard ratio for survival) and 95% CIs were calculated where applicable. P < .05 was considered significant. All statistical analyses were performed with R: A Language and Environment for Statistical Computing (R Foundation for Statistical Computing, 2015; https://www.R-project.org).
A total of 513 pathology records were reviewed. Of these, 76 (15%) were excluded for not meeting study criteria. There were 8 patients with no reported margin evaluation from the main specimen and 20 with no intraoperative frozen margins; these were excluded from further analysis. There were 7 patients with intraoperative frozen sections from the main specimen only, and these were excluded to decrease heterogeneity. The results of this study are intended to be applicable only to a method of intraoperative margin assessment from the tumor bed. The 406 remaining patients were included in the analysis.
The characteristics of this study population are summarized in Table 1 for the overall cohort and in relationship to the frozen and specimen margins (involved margin including positive, CIS, and very close for the tumor specimen). The study population was 58% male with a median age of 61 years. Tumor subsites included 45% oral tongue, 20% alveolus, 18% floor of the mouth, and 17% other (eg, palate, retromolar trigone, cheek). The distribution of T stage was 45% T1, 21% T2, 4% T3, and 30% T4. The small number of T3 patients were combined into a category of T3/T4 patients. The distribution of N stage was 71% N0, 10% N1, 1% N2a, 14% N2b, 4% N2c, and less than 1% N3. The small number of patients in various node-positive groups were collapsed into N0, N1/N2a, and N2b/N2c/N3. Factors significantly associated with an involved specimen margin included T stage, N stage, extracapsular extension, perineural invasion, lymphovascular invasion, bone invasion, and initial intraoperative frozen margin. The only factors significantly correlated with an involved initial frozen margin were patient sex and tumor specimen margin.
There were 3308 intraoperative frozen sections sampled from the tumor bed in 406 patients (see eTable 1 in the Supplement for numbers of involved frozen, final operative, and tumor bed margins). In comparing the intraoperative diagnosis to final frozen section diagnosis (as reevaluated by the pathologist postoperatively), we found that frozen section accuracy was 99%, with 97% sensitivity (95% CI, 94%-98%) and nearly 100% specificity (95% CI, 99%-100%). There were 14 false-negative diagnoses (<1%) and 8 false-positive diagnoses (<1%), with a less than 4% false-negative rate for intraoperative frozen margins. The prevalence of involved margins found on frozen samples was 12%; the NPV was nearly 100% (95% CI, 99%-100%).
We next evaluated the performance of intraoperative frozen margins as a test of final specimen margins. For intraoperative frozen margins, there were 149 patients with initial involved margins (including invasive cancer and CIS); for margins assessed on the corresponding tumor specimen, there were 115 involved margins, providing a sensitivity of 55% (95% CI, 45%-64), specificity of 70% (95% CI, 65%-76%), and accuracy of 66% for frozen section as a test of final specimen margins (false-negative rate, 45%). When we included very close margins (<1 mm) on the tumor specimen, the prevalence of an involved margin was 43%, and the performance of frozen sections in identifying an involved margin was as follows: sensitivity, 48% (95% CI, 41%-56%); specificity, 72% (95% CI, 66%-78%); NPV, 65% (95% CI, 58%-70%); and PPV, 57% (95% CI, 49%-65%). Alternatively, when we applied a more strict definition of involved as only invasive cancer at the frozen and specimen margin, the prevalence was 23%; sensitivity, 35% (95% CI, 26%-46%); specificity, 87% (95% CI, 83%-90%); NPV, 82% (95% CI, 77%-86%), and PPV, 45% (95% CI, 33%-57%) (see eTable 2 in the Supplement for detailed tables of these margin categories).
To determine an accurate LR rate, we excluded patients who had recurrence other than LR or who died before 6 months of follow-up without LR, leaving 344 patients for the LR analysis (Table 2). The overall LR rate was 23%. For the specimen margin, the LR rates for strictly negative margins (n = 79) vs positive margins (n = 77) were 13% vs 45%, and 18% for patients with other margin statuses (CIS, dysplasia, close, or very close; n = 188). For the initial intraoperative frozen margins, the LR rates for strictly negative margins (n = 191) vs positive margins (n = 59) were 19% vs 36%, and 22% for the other margin statuses (n = 94). For the final margin status based on all frozen specimens and additional resections, the LR rates for strictly negative margins (n = 268) vs positive margins (n = 8) were 22% vs 62%, and 22% for other margin statuses (n = 68). Overall, the tumor specimen margins were significantly associated with LR, while the initial intraoperative frozen margins and final operative margins were not significantly associated with LR.
We then examined the correlation of LR between intraoperative frozen margins and permanent specimen margins (eTable 3 in the Supplement). There was a very large difference in LR between negative frozen–negative specimen margins (7%, n = 57) and positive frozen–positive specimen margins (65%, n = 26). For patients with a positive specimen margin and any finding other than positive on the frozen margin (including negative), the LR rate was 35% (n = 51). For patients with a negative specimen margin but frozen margin other than negative (n = 22), the LR rate was 29%.
We also looked at LR with various intermediate margins. For patients with CIS at the specimen margin, LR was 24%. For patients with a specimen margin less than 1 mm, LR was 25% (n = 51). For patients with a close specimen margin (1-5 mm), recurrence was 15% (n = 117). There were only 3 patients with mild or moderate dysplasia at the specimen margin, and there were no recurrences in these 3 cases.
Recognizing that frozen margin assessment from the tumor bed adds information in predicting LR, we stratified margin results by placing patients into 4 groups: group 1 included patients with negative frozen margins and negative specimen margins (n = 57); group 2, any patient without a positive specimen margin who was not included in group 1 (ie, close, very close, dysplasia at either margin, or positive at the frozen margin) (n = 210); group 3, all patients with positive specimen margins, but not a positive frozen margin (n = 51); and group 4, those with positive frozen and specimen margins (n = 26). The LR rates for these groups are as follows: group 1, 7% (95% CI, 2%-18%); group 2, 19% (95% CI, 14%-25%); group 3, 35% (95% CI, 23%-50%), and group 4, 65% (95% CI, 44%-82%), and these differences were significant for the group overall (P <.001). Compared with group 1, the relative risk for LR in groups 2, 3, and 4 were 2.7 (95% CI, 1.0-7.3), 5.0 (95% CI, 1.8-13.9), and 9.3 (95% CI, 3.5-25.0), respectively (Table 3).
There was a significant difference in observed survival for each of the specimen, frozen, final, and combined margin groups (eFigure 1 in the Supplement). Interestingly, the difference was greatest for stratification based on specimen margin, and even greater when specimen and frozen margin information was combined to stratify the groups (hazard ratios for margin groups relative to group 1: group 2, 1.50 (95% CI, 0.92-2.45); group 3, 2.27 (95% CI, 1.27-4.04); and group 4, 4.51 (95% CI, 2.46-8.27). The Kaplan-Meier estimates of 5-year survival based on combined margin grouping of negative-negative findings, intermediate results, positive specimen but not positive frozen, and positive-positive specimens (groups 1-4) were 72%, 61%, 43%, and 19%, respectively (Figure).
Of patients with intraoperative frozen sections, 37% (n = 149) had involved frozen margins including invasive disease and CIS. All of these except 33 (22%) had additional tissue removed from the involved margin. Demonstration of a final negative margin based on intraoperative frozen sampling and additional resection was achieved for all except 40 (27%).
To examine the association between prognosis and additional resection to clear a positive frozen margin, 3 groups were created: group A included patients with uninvolved intraoperative and tumor specimen margins (no invasive cancer <1 mm or CIS); group B, those with involved intraoperative margins (invasive cancer or CIS) who underwent additional resection to clear the margins with negative final intraoperative margins; and group C, those with uninvolved intraoperative margins and no additional resection but involved margins on main tumor specimen (invasive cancer <1 mm or CIS). The LR rates for groups A, B, C were 13% (95% CI, 8%-20%), 27% (95% CI, 18%-37%), and 34% (95% CI, 24%-46%), respectively (Table 4). A pairwise test of significance for each group showed a significant difference in LR between group A and group B (absolute difference, 14%; 95% CI, 3%-26%) and between group A and group C (absolute difference, 21%; 95% CI, 8%-34%) but no difference in LR between groups B and C (absolute difference, 7%; 95% CI, 8%-22%). In addition, for purposes of analysis, a fourth classification was created to include patients with a final intraoperative margin that, despite attempts, could not be cleared and remained involved (invasive or CIS); the LR for this group (group D) was 29% (95% CI, 16%-48%), which was not different from group B or C. This indicates that achieving a negative margin from additional tissue resection after an initial involved frozen margin does not change the prognostic information derived from a microscopically positive resection margin.
It does not appear that additional tissue resection is significantly related to observed survival: patients whose frozen specimen was involved and then cleared by further excision (group B) or those whose specimen margin was involved, but no additional resection occurred (group C) had no survival difference compared with each other. Both of these groups, and the added group D, had significantly worse survival outcomes than patients with uninvolved margins in both intraoperative and tumor specimens (group A) (hazard ratios relative to group A: group B, 1.60; 95% CI, 1.08-2.36; group C, 1.85; 95% CI, 1.24-2.74; and group D, 2.14; 95% CI, 1.29-3.55). Performing intraoperative frozen assessment with the intention to clear an involved tumor bed margin was not associated with improved overall prognosis.
In this large, single-institution review of OCSCC surgery, we demonstrate several important results, including the prognostic significance of the tumor specimen margin, additional value of tumor bed frozen margin, relatively low prognostic utility of final margins, and lack of improved outcomes with additional resection to clear a positive frozen margin. This analysis benefits from a relatively homogenous surgical technique as well as a broad sample of patients with oral cavity cancer, making it both technique- and oral cavity site–specific and most likely applicable to patients treated with wide local excision and intraoperative tumor bed margin assessment. One limitation of this study is a lack of comprehensive, reliable comorbidity data for the patient data set, which might potentially influence observed survival results. Other limitations include surgical practice by a relatively small number of surgeons, and an inability to compare our results with frozen sections taken from the main specimen owing to the paucity of cases evaluated by this method.
It is a tenet of surgical oncology to attempt complete tumor extirpation as assessed by a cuff of normal tissue surrounding the neoplasm. Even if all detectable tumor is removed, recurrences and cancer-related deaths still occur. This raises the question of whether microscopic examination of resection edges provides prognostic information about the likelihood of recurrence and survival.
For head and neck carcinomas, these questions have been extensively studied. Landmark articles3,4 have found that microscopic margins are associated with recurrence and survival, recommending that intraoperative frozen sections be obtained and analyzed to increase the likelihood of a negative margin. Many studies support these conclusions,5-7 while others provide contradictory evidence,1,2,8-10 suggesting that a positive margin is one of several factors signifying aggressive tumor biology, but it is not independently sufficient to determine prognosis. These studies were similar in they all assessed margins based on the main specimen, but they used different definitions of involved.
The present study differs from prior analyses in that it bases the intraoperative margin evaluation strictly on tumor bed sampling. We also considered a variety of margin assessments (initial frozen, final margin, tumor specimen) and findings (invasive cancer, CIS, dysplasia, closeness). We clearly demonstrate that the tumor specimen margin and initial frozen margin are associated with recurrence and survival under a variety of definitions of an involved margin, but the final operative margin is associated with recurrence only when invasive cancer remains.
Ninety-nine percent of head and neck surgeons use intraoperative frozen sections to analyze margins and assume that these specimens will provide an accurate assessment.2-4 In this study, we confirm that the reliability of intraoperative frozen specimen diagnosis is quite good relative to the final diagnosis of the same tissue specimen, with an accuracy of 99%, and a false-negative rate of 3.5%. However, as in other smaller studies,5-7,11-15 we also confirm that concordance between tumor bed and tumor specimen margins is not good. Defining involved margins as those with invasive carcinoma or CIS, we found that the false-negative rate for assessment by frozen specimen from the tumor bed is 45%, with an NPV of 79%. The intraoperative margin sampling from the tumor bed did not detect more than half the cases of an involved margin on the tumor specimen.
There is a lack of reliability in margin sampling from the tumor bed that would allow independent reviewers of the pathology report to know whether a main specimen margin had been replaced by a subsequent resection. In addition, there is a recognized issue with site relocalization, as it may be difficult to accurately identify a particular location within the tumor resection bed.16
The current study demonstrates a lack of perfect concordance between tumor bed margins and main specimen margins. Cases of main specimen margin involvement with no corresponding findings in tumor bed frozen sections are common; conversely, frozen sections may show involvement, while the main specimen margins are negative. We demonstrate that prognostic information is contained within findings from both the frozen and specimen margins. We therefore propose that neither should be considered the “true” margin, and each provides information to help develop prognoses. The results of both the frozen and main specimen margin analyses provide an approximation of whether tumor will recur locally, and each has predictive value.
There are a number of factors to consider when deciding how or whether to sample intraoperative margins. Samples taken from the tumor bed may be of varying width, length, and depth and are difficult to correlate to the main specimen, yet this is the most prevalent sampling technique used by head and neck surgeons.2 Prior studies comparing the 2 frozen margin techniques suggest that false negatives occur equally with either approach,15 although margin sampling from the tumor specimen had the best correlation with local control and survival.14,17,18
In the current study, we demonstrate that main specimen margin assessment has a stronger correlation with recurrence and survival than tumor bed margin sampling. We do not evaluate intraoperative main specimen margin sampling, and we have not directly compared the 2 methods.
The ability to respond to a positive margin by additional surgical resection has been described as the rationale to assess intraoperative margins. However, it is not clear that this practice benefits patients. Byers et al3 report similar LR and 2-year survival whether the margin was negative or initially positive on frozen section and subsequently cleared. However, outcomes were worse for patients with positive margins at the conclusion of surgery (LR, 80%; survival, 5%).3 The study concludes that intraoperative frozen analysis identifying a positive margin may benefit the patient if the margin is further excised to negative.
Frozen section analysis is performed on many more patients than could ultimately benefit. Byers et al3 calculated that 30% of patients derived benefit from frozen assessment and suggested that it was reasonable to obtain frozen sections from everyone for this level of potential gain. However, it is important to realize that the study excluded nearly 60% of patients from analysis for a variety of reasons. Other studies have calculated that between 2% and 15% of patients could potentially benefit from intraoperative frozen sections.11,19 A number of retrospective studies have reexamined the findings of Byers et al,3 and some have agreed,19 while others suggest that no benefit is realized if a positive margin is cleared, and yet others describe a prognostic significance for microscopic tumor cut-through.9,13,14,18,20
Our study does not find evidence to support intraoperative frozen sampling from the tumor bed as a therapeutic measure. This reinforces the hypothesis that a positive margin despite intentions to gain a cuff of normal tissue is a sign of aggressive tumor biology and/or anatomic constraints of the resection. We should be careful to clarify that we do not believe that this provides evidence that a known positive margin should not be pursued with additional excision. We emphasize that this study highlights the prognostic importance of margin status and that an involved margin, either on the initial tumor bed frozen sections or the tumor specimen, is associated with worse outcomes regardless of the final margin result.
In summary, this study has made several important observations that should influence the use and interpretation of margin data in OCSCC surgery: (1) findings on microscopic assessment of margins from the tumor specimen or the tumor bed yield prognostic information for recurrence and survival; (2) the permanent specimen margin has a stronger association with outcomes than the tumor bed frozen sections; and (3) there is no evidence that additional removal of tissue guided by clearing intraoperative margins has therapeutic or prognostic value.
Corresponding Author: Steven M. Sperry, MD, Department of Otolaryngology–Head and Neck Surgery, University of Iowa Hospitals and Clinics, 200 Hawkins Dr, Iowa City, IA 52242 (email@example.com).
Accepted for Publication: June 28, 2016.
Published Online: July 17, 2016. doi:10.1001/jamaoto.2016.2329
Author Contributions: Dr Sperry had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Sperry.
Acquisition, analysis, or interpretation of data: Buchakjian, Tasche, Robinson, Pagedar, Sperry.
Drafting of the manuscript: Buchakjian, Sperry.
Critical revision of the manuscript for important intellectual content: Buchakjian, Tasche, Robinson, Pagedar, Sperry.
Statistical analysis: Tasche, Sperry.
Administrative, technical, or material support: Robinson, Sperry.
Study supervision: Robinson, Sperry.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none were reported.
Previous Presentation: This study was presented at the American Head & Neck Society Ninth International Conference on Head and Neck Cancer; July 17, 2016; Seattle, Washington.
Create a personal account or sign in to: