Oral Squamous Cell Carcinoma

ORIGINAL ARTICLE

Oral Squamous Cell Carcinoma Histologic Risk Assessment, but Not Margin Status, Is Strongly Predictive of Local Disease-free and Overall Survival Margaret Brandwein-Gensler, MD,*† Miriam S. Teixeira, MD,* Carol Ming Lewis, MD, MPH,§ Bryant Lee, MD,* Linda Rolnitzky, MS,‡ Johannes J. Hille, DDS,k Eric Genden, MD,* Mark L. Urken, MD,* and Beverly Yiyao Wang, MD†

Abstract: To analyze the impact of resection margin status and histologic prognosticators on local recurrence (LR) and overall survival (OS) for patients with oral squamous cell carcinoma (OSCC). This study was both retrospective and prospective in design. Cohort 1 refers to the entire group of 292 patients with OSCC. The slides from the earliest resection specimens from Cohort 1 were examined in an exploratory manner for multiple parameters. Cohort 2 refers to a subset of 203 patients, who did not receive any neoadjuvant therapy and had outcome data. Cohort 3 represents a subset of Cohort 2 (n = 168) wherein the histologic resection margin status could be reconfirmed. Cohort 4 refers a subset of 85 patients with tongue/floor of mouth tumors. Margin status was designated as follows: group 1, clearance of $5 mm with intraoperative analysis, no need for supplemental margins (n = 46); group 2, initial margins were measured as ,5 mm during intraoperative frozen section; supplemental resection margins were negative on final pathology (n = 73); group 3, the final pathology revealed resection margins ,5 mm (n = 30); group 4, the final pathology revealed frankly positive resection margins (n = 19). The endpoints of LR and OS were queried with respect to T stage, tumor site, margin status, and numerous histologic variables, by Cox regression and Kaplan-Meier survival analyses. Tumor stage (T) was significantly associated with LR (P = 0.028). Kaplan-Meier analysis for stage and for intraoral site was significantly associated with LR for T4 tumors. The increased likelihood of LR was higher for T4 OSCC of the buccal mucosa (75%), sinopalate (50%), and gingiva (100%) compared with mobile tongue (27%), and oropharynx (13%) (P = 0.013). Margin status was not associated with LR or OS (Cohort 3). This was so when all tumors were grouped together and when separate analyses were performed by tumor stage and oral subsite. No significance was demonstrated when margin status was examined for patients with similar treatment (surgery alone or surgery with adjuvant RT). However, the administration of adjuvant RT did significantly increase local

From the Departments of *Otolaryngology, †Pathology, and ‡Biomathematics, Mount Sinai Medical Center, New York, NY; §Department of Otolaryngology, Massachusetts General Hospital, Boston, MA; and kDepartment of Oral Pathology, University of Western Cape, Cape Town, South Africa. Reprints: Margaret Brandwein-Gensler, MD, Albert Einstein College of Medicine, Montefiore Medical Center, Moses Division, Department of Pathology, ENT, 111 East 210th Street, Bronx, NY 10467 (e-mail: mgensler@montefiore.org). Copyright Ó 2005 by Lippincott Williams & Wilkins

Am J Surg Pathol
Volume 29, Number 2, February 2005

disease-free survival (P = 0.0027 and P = 0.001 for T1 and T2 SCC, respectively). On exploratory analyses of histologic parameters, worst pattern of invasion was significantly associated with LR (P = 0.015) and OS (P , 0.001). Perineural invasion involving large nerves (.1 mm) was associated with LR (P = 0.005) and OS (P = 0.039). Limited lymphocytic response was also significantly associated with LR (P = 0.005) and OS (P = 0.001). When used as covariates in a multivariate Cox regression model, worst pattern of invasion, perineural invasion, and lymphocytic response were significant and independent predictors of both LR and OS, even when adjusting for margin status. Thus, these factors were used to generate our risk assessment. Our risk assessment classified patients into low-, intermediate-, or high-risk groups, with respect to LR (P = 0.0004) and OS (P , 0.0001). This classification retained significance when examining patients with uniform treatment. In separate analyses for each risk group, we found that administration of adjuvant radiation therapy is associated with increased local disease-free survival for high-risk patients only (P = 0.0296) but not low-risk or intermediate-risk patients. Resection margin status alone is not an independent predictor of LR and cannot be the sole variable in the decision-making process regarding adjuvant radiation therapy. We suggest that the recommendation for adjuvant radiation therapy be based on, not only traditional factors (inadequate margin, perineural invasion, bone invasion) but also histologic risk assessment. If clinicians want to avoid the debilitation of adjuvant radiation therapy, then a 5-mm margin standard may not be effective in the presence of high-risk score. Key Words: oral squamous carcinoma, risk, pathology, diagnosis, prognosis, radiotherapy, margins (Am J Surg Pathol 2005;29:167–178)

M

any pathologists have faced the query, ‘‘But I took the whole tumor out. Why did it recur?’’ One goal of this study was to examine the impact on resection margin status on outcomes for patients with oral squamous cell carcinoma (OSCC). We demonstrate that, even among tumors with extensive histologic documentation of adequate resection margins, local recurrence (LR) develops in a predictable subset of patients. Achieving histologically adequate margins does not constitute the entire picture. Correlating pathologic tumor grade alone with outcome has not been a uniformly satisfying venture. Tumor thickness

167

Am J Surg Pathol
Volume 29, Number 2, February 2005

Brandwein-Gensler et al

for T1 tongue squamous cell carcinoma (SCC) has strongly associated with propensity for lymph node metastasis and survival17 but is not a predictor for LR.13,25 Multiparameter prognostic histologic assessments have been developed and refined over the last two decades, based on variables that include pattern of invasion, degree of keratinization, nuclear pleomorphism, lymphocytic response, and mitotic rate.1–5,18,22 Pattern of tumor invasion (POI) refers to the manner in which cancer infiltrates tissue at the tumor/host interface. It is intuitive that neoplasia infiltrating in a widely dispersed manner is more aggressive than those growing in a bulky pushing fashion. Over the past two decades, POI alone, and as part of weighted scoring systems, has been demonstrated to predict LR and decreased overall survival (OS). In this study, we evaluated multiple histologic variables, analyzing them with respect to patient outcome. We present a novel, simple, highly predictive histologic risk scoring scheme that can serve as a rationale for decision-making viz-a-viz the necessity for adjuvant radiation therapy (RT) at the primary tumor site.

MATERIALS AND METHODS Patient Cohorts Clinicopathologic data were analyzed on 292 patients (193 men, 99 women; age range, 18–95 years, mean age, 63 years) with OSCC treated at this institution over the last two decades (defined as Cohort 1, Table 1). The tumor sites of origin were determined after examination of the resection specimens and correlating these findings with the clinical impressions. Cohort 1 included patients receiving primary therapy at other institutions and patients receiving neoadjuvant therapy. The adjuvant treatment status was not confirmed for 65 patients. Clinical outcomes were available for 259 patients. TABLE 1. Frequency Distributions of Cohort 1 Variable Site

T

N

Stage

168

Mobile tongue, floor of mouth Tongue base Posterior oral cavity, pharynx Gingiva Palate and maxilla Buccal 1 2 3 4 X 0 1 2 3 X 1 2 3 4 X

N

%

138 29 67 7 25 18 82 92 35 73 7 148 47 73 12 10 51 40 47 138 11

47 10 23 2 9 6 28 32 12 25 2 51 16 25 4 3 17 14 16 47 4

Cohort 2 (n = 203) represents a group culled from Cohort 1, with no neoadjuvant therapy, and confirmed treatment status and clinical follow-up. Seventy-nine patients were uniformly treated by primary surgery alone. A total of 124 patients received primary surgery plus adjuvant radiotherapy. Cohort 3 (n = 168) represents a subset of Cohort 2, satisfying the following requirements: 1) they received their primary surgery at this institution between 1989 and 2002; 2) no neoadjuvant therapy was administered; 3) the specimen was adequately sampled, and all or most of the slides were available for rereview to confirm margin status; 4) postoperative treatment and disease status was confirmed; and 5) all disease-free patients had a minimum of 18 months follow-up. Of these 168 patients, 105 received adjuvant radiotherapy and 63 did not. Cohort 4 refers to a subset of Cohort 3 with tumors of the mobile tongue and floor of mouth.

Histologic Variables: Pattern of Invasion All glass slides from the earliest available resection specimens of Cohort 1 were examined. Table 2 summarizes the variables analyzed. Tumor POI was examined at the host/tumor interface (Fig. 1). POI types 1 through 4 have been previously defined by Bryne et al.3–5 POI type 1 represents tumor invasion in a broad pushing manner. POI type 2 represents tumor invasion with broad pushing ‘‘fingers,’’ or separate large tumor islands, with a stellate appearance. POI type 3 represents invasive islands of tumor greater than 15 cells per island. POI type 4 represents invasive tumor islands smaller than 15 cells per island. This includes single cell invasion. POI type 4 also includes strands of tumor cells in a single-cell filing pattern, regardless of island size. Herein, we introduce a pattern 5 to recognize a widely dispersed pattern of tumor infiltrate. We define this as tumor satellites of any size with 1 mm or greater distance of intervening normal tissue (not fibrosis) at the tumor/host interface. Adherence to these well-defined criteria resulted in good interobserver concordance. We are presently undertaking a study of interobserver concordance of these variables in a larger group of pathologists. All prior publications concerning POI were either limited to biopsy specimens or were based on the most aggressive POI present. To validate this practice, we created two new variables: predominant POI (PPOI) and worst POI (WPOI). PPOI was determined by measuring POI at the tumor interface of each slide of the resection specimen. PPOI was tallied as the most common POI found; in case of a tie (eg, four slides of POI 3 and four slides POI 4), the higher score was assigned. The WPOI was taken as the highest score present, no matter how focal. Lymphoid infiltrate at the tumor/host interface was quantified as a three-tiered variable (Fig. 2). Pattern 1 was assigned when a continuous and dense rim of lymphoid tissue was present at the interface. Pattern 2 was assigned when patches of dense lymphoid infiltrate were present at the interface but the inflammation was discontinuous along the interface. The presence of any lymphoid patch qualifies a tumor as lymphocyte pattern 2. Pattern 3 was assigned for limited response, that did not form lymphoid patches, or for no lymphoid response. (We are presently undertaking a study comparing the reproducibility of both the three-tiered and q 2005 Lippincott Williams & Wilkins

Am J Surg Pathol
Volume 29, Number 2, February 2005

Oral Squamous Cell Carcinoma

TABLE 2. Frequency Distribution of Histologic Variables for Cohort 1 Variable Worst pattern of invasion (WPOI)

Predominant pattern of invasion (PPOI)

Grade

Multifocal carcinoma in situ

Keratin

Foreign body reaction

Eosinophilia

Lymphocytic infiltrate

Vascular/lymphatic invasion

Perineural invasion

Grade

N

%

1 2 3 4 5 1 2 3 4 5 X 1 2 3 None 1 slide .1 slide Not evaluated Most = 1 Intermediate = 2 Least = 3 None Any Not evaluated None Any Not evaluated Dense continuous band = 1 Large patches = 2 Little or none = 3 None 1 or 2 vessels More than 3 vessels Not evaluated None Small nerves Nerves larger than 1 mm Not evaluated

2 7 51 159 70 7 54 144 70 13 4 33 109 150 236 25 27 4 132 106 54 226 62 4 202 85 5 67 128 97 223 34 31 4 217 42 30 1

0.7 2.4 17.5 54.5 24 2 18 49 24 4 1 11 37 52 81 9 9 1 45 36 19 77 21 1 69 29 2 23 44 33 76 12 11 1 76 14 10 0.3

a four-tiered grading.5) Perineural invasion (PNI) was defined as carcinoma specifically tracking along or within a nerve. PNI was further classified as involving large nerves (diameter $1 mm) or small nerves (diameter ,1 mm) (Fig. 3). Other variables examined were as follows: keratinization (quantified as .50%, ,50%, and ,10%), nuclear grade (well, intermediate, poor), foreign body reaction (1 or more high power field (HPF) vs. none), eosinophilia (1 or more HPF vs. none), lymphatic or vascular tumor emboli, and carcinoma in situ in mucosal adjacent to tumor (no slides, 1 slide, more than 1 slide).

Intraoperative Margin Assessment T1 tumors were mapped and entirely blocked in during intraoperative examination. T2 to T4 tumors were mapped out q 2005 Lippincott Williams & Wilkins

and sampled generously. This involves entirely sampling all of the closest margins, plus representative sections from the remaining margins. The distances between carcinoma (either infiltrating or mucosal carcinoma in situ) and the inked resection margins were microscopically measured using a 43 ocular and a ruler. If the distance is #5 mm at the time of frozen section, then that margin was designated as inadequate. A resection margin map is thus generated. The surgeon is informed of the results and the specific problematic sites and examines either a drawing or the actual remaining specimen. The sites of inadequate margins are then revisited in situ, and additional tissue cuffs are submitted as ‘‘supplemental margins’’ (guided supplemental margins). The supplemental margins are entirely examined; if cancer-free, then tumor extirpation is complete. Our treatment policy is to continue harvesting supplemental margins until resection adequacy is achieved. For this study, all the permanent and frozen section histologic slides on the primary resection specimens were rereviewed, and microscopic measurements were retaken using a 43 ocular piece and a ruler. Patient margin status was retrospectively classified into four groups based on examination of the formalin-fixed paraffin embedded histology slides, plus the frozen section slides. The rule of thumb is ‘‘worst news wins.’’ For example, if a margin measures 3 mm on frozen section and the corresponding permanent section measures 5 mm, the shorter distance was taken as the margin distance. Also, while 5 mm was considered close on intraoperative analysis, a measurement of 5 mm on formalin-fixed, paraffin-embedded tissue was considered adequate retrospectively. This strategy was used to avoid the scenario of negative margins at frozen section converting to close margins after further sectioning on permanent sections. Interestingly, the impact of tissue shrinkage had limited impact when frozen section margin distances were compared with permanent sections.

Margin Analysis The four margin groups were classified as follows: group 1 was categorized $5 mm clearance at initial resection; group 2 was categorized as inadequate margins during intraoperative consultation, final margins $5 mm after harvesting supplemental margins; group 3 was categorized as final permanent pathology revealing close (,5 mm) margins; and group 4 was categorized as final permanent pathology revealing frankly positive margins. Cohort 3 was subjected to analyses regarding margin status.

Analysis Endpoints Follow-up data were collected through patient visits, hospitalizations, or telephone interviews. ‘‘First failure’’ was defined as the first manifestation of disease failure per site. Locoregional disease recurrence was subclassified as either local (at the primary tumor site) or regional (at the site of cervical metastases) or both (local plus regional). This was accomplished by reexamining slides from the resection specimen of the first recurrence and correlating these findings with the margins of the previous resection and the clinical impression. The endpoint for Cox regression model and Kaplan-Meier analyses was 1) time to first LR versus 2)

169

Brandwein-Gensler et al

Am J Surg Pathol
Volume 29, Number 2, February 2005

FIGURE 1. Pattern of invasion (POI). A, POI type 1: broad pushing front. B, POI type 2: ‘‘finger-like’’ pushing pattern. C, POI type 2: note the high nuclear grade despite the low-grade POI. D, POI type 3: islands at tumor periphery greater larger than 15 cells/island. E, POI type 4: individual tumor cells infiltrating at interface. F, POI type 4: strands of infiltrating tumor cells. G, POI type 4: tumor islands composed of 15 cells or less. H, POI type 5: 1 mm or greater of normal tissue between tumor satellite islands at interface.

duration of failure-free follow-up, and 3) time to death versus duration of OS. Distant metastases, or the development of a new primary carcinoma, were less common manifestations of ‘‘first failures.’’ The date of death was confirmed with the Social Security Death Index (www.ancestry.com/search/rectype/vital/ssdi/main.htm) rather than death certificates. Our endpoint for OS does not distinguish cause of death; this approach is valid as it assumes that deaths unrelated to cancer are equally distributed between both groups. Significance level was set at P , 0.05. Spearman rank correlation coefficients were used to examine the interrelationships between the histologic variables of Table 2 with each other and with patient age, gender, tumor site, tumor stage (T), nodal stage (N), and stage grouping. The Cox regression survival model and Kaplan-Meier survival analyses were used to analyze data on the variables margin status,

170

WPOI, PPOI, tumor grade, multifocal carcinoma in situ, keratin production, foreign body reaction, eosinophilia, lymphocytic infiltrate, vascular/lymphatic invasion, perineural invasion, tumor site, and T stage with respect to LR, regional metastasis, and OS.

RESULTS Patient Outcome With Respect to ‘‘First Failures’’ For Cohort 2, 97 patients remained disease-free after median follow-up of 30 months (range, 6 months to 9.8 years). LR developed in 38 patients after a median of 9 months (range, 50 days to 2.7 years). Regional cervical metastases developed in 30 patients after a median of 8 months (range, 55 days to 2 years). Eleven patients developed synchronous LR and q 2005 Lippincott Williams & Wilkins

Am J Surg Pathol
Volume 29, Number 2, February 2005

Oral Squamous Cell Carcinoma

FIGURE 3. Perineural invasion of a large nerve.

manifestation of disease failure for 11 patients after a median of 23 months (range, 3 months to 7 years). SPTs were defined as any malignancy developing at an upper aerodigestive tract site not adjacent to the index tumor. Pulmonary adenocarcinomas, or SCC histologically distinct from the index tumor, were also considered as SPTs. With respect to margin status, there were 46 patients classified as group 1, 73 as group 2, 30 as group 3, and 19 as group 4 (total 168, Cohort 3). Table 3 demonstrates the frequency distributions of all ‘‘first failure events’’ for Cohort 3, per tumor site. Table 4 demonstrates the LR rates only, for all sites, across all margin groups for Cohort 3. Table 5 breaks down for 85 patients with tongue/floor of mouth tumors (Cohort 4) with respect to LR rate, margin status, T stage, and adjuvant RT status. Across all tumor sites, T stage did significantly correlate with LR (Pearson x2, P = 0.028). When stratified for both stage and site (but not margin group), this significance was seen for T4 tumors. The LR rate was significantly higher for T4 tumors of the buccal mucosa (75%), sinopalatal (50%), and gingival (100%) as compared with mobile tongue (27%) and oropharynx (13%) (Pearson x2, P = 0.013). Margin status was significantly associated with perineural spread (Pearson x2, P = 0.003,). Patients with close or positive margins were more likely to have perineural spread than patients with negative margins (Table 6).

Margin Status: No Association With LR or OS FIGURE 2. Lymphocytic infiltrate. A, Pattern 1: dense rim of lymphoid tissue at the interface. B, Pattern 2: patches of lymphoid infiltrate present at the interface (L), but the inflammation was discontinuous (*). C, Pattern 3: limited lymphoid response without formation of lymphoid patches.

regional metastases after a median of 9 months (range, 4.4 months to 5 years). Analyses concerning LR included the sum of LR alone, plus LR with synchronous regional metastases. Distant metastases in all but 1 patient involved the lungs; the remaining patient developed brain metastases. Sixteen patients developed distant metastasis as the first manifestation of disease failure after a median of 7 months (range, 37 days to 2 years). A second primary tumor (SPT) was the first q 2005 Lippincott Williams & Wilkins

Kaplan-Meier analyses for the endpoints of local disease-free survival and OS demonstrated no significant association between margin status and either of these endpoints for Cohort 3 (Figs. 4, 5) or Cohort 4. There was no significant relationship between margin status and LR, or OS, for all sites and T stages, combined and separate. As an example, there were 12 patients with T1 carcinomas who were resected with adequate margins (group 1). None of them received adjuvant RT. Yet 25% of this group developed local and/or local plus regional recurrence. This is explained by the histologic risk factors reported below (Table 5). To control for the confounding effect of adjuvant RT, we also analyzed within treatment groups of Cohort 3 for the patients treated only with surgery (n = 63), and for those also receiving adjuvant RT (n = 105), across margin groups. No

171

Am J Surg Pathol
Volume 29, Number 2, February 2005

Brandwein-Gensler et al

TABLE 3. First Failures Stratified by Oral Subsite for Cohort 3 Site

N*

Mobile tongue and/or floor of mouth Tonsil, pharynx, retromolar trigone Palatomaxillary Base of tongue Buccal Gingiva Total

85 34 19 16 10 4 168

Local Recurrence (LR) 12 3 3 2 5 4 29

Regional Metastasis (RM)

(14%) (9%) (14%) (12%) (50%) (100%) (17%)

12 4 2 6 0 0 24

LR + RM

(14%) (12%) (10%) (37%)

5 1 2 0 0 0 8

(14%)

(6%) (3%) (10%)

(5%)

Distant Metastasis

New Primary

8 (9%) 4 (12%) 0 0 0 0 12 (7%)

3 (4%) 3 (9%) 0 0 0 0 6 (4%)

Note: Percentages go across rows per site. *Patients with follow-up more than 18 months.

also associated with N . N0 at diagnosis (two-tailed x2, P , 0.001). With regard to LR, statistical significance was achieved when WPOI pattern 5 was compared with pattern 4 (Table 8), (Pearson x2, P , 0.001), but no significance was achieved with PPOI. Both PPOI (x2, P = 0.003; 95% confidence interval, 1.68, 19.98) and WPOI (x2, P = 0.000; 95% confidence interval, 1.49, 28.00) patterns 4 and 5 were significantly associated with OS (Table 7). Both PNI and lymphocytic response were strongly associated with OS and LR (Table 8). No associations were found between tumor grade, multifocal carcinoma in situ, keratinization, foreign body reaction, eosinophilia, vascular/lymphatic invasion, as well as age and gender, and LR or OS.

significant associations were demonstrated between margin status and LR and OS for uniformly treated groups either.

Associations With Histologic Features Spearman rank correlation coefficients performed on the data from Cohort 1 confirmed that all of the histopathologic variables, POI (ungrouped), tumor grade, multifocal carcinoma in situ, keratin, foreign body reaction, eosinophilia, lymphocytic infiltrate, vascular/lymphatic invasion, and perineural invasion, were independent of each other, and also independent of age, gender, tumor, node, and stage. The coefficient of correlation (r) was less than 0.2, with the exception of keratin and grade (r = 0.41). However, exploratory analyses revealed that WPOI patterns 4 and 5 versus grouped patterns 1, 2, and 3 were associated with margin status (Pearson x2, P = 0.000) and also PNI (Pearson x2, P = 0.003). The frequency distributions of these variables are shown in Table 7. WPOI patterns 4 and 5 versus grouped patterns 1, 2, and 3 were

Introducing Multifactorial Risk Assessment Given the strong associations of WPOI, PNI, and lymphocytic response with the outcomes of LR and OS, we

TABLE 4. Local Recurrences (LR) for Cohort 3, Stratified for Tumor Site Margin Status Mobile tongue +/or floor of mouth Tonsil, pharynx, retromolar trigone Palatomaxillary Base of tongue Buccal Gingiva Total

N total N LR N total N LR N total N LR N total N LR N total N LR N total N LR N total N LR

N

Group 1

Group 2

Group 3

Group 4

85 17 34 4 19 5 16 2 10 5 4 4 168 37

22 4 8 1 7 3 5 0 3 0 1 1 46 9

39 9 15 3 7 1 5 1 4 2 3 3 73 19

15 2 7 0 3 0 4 1 1 1 0 0 30 4

9 2 4 0 2 1 2 0 2 2 0 0 19 5

N per site per margin Corresponding LR N per site per margin Corresponding LR N per site per margin Corresponding LR N per site per margin Corresponding LR N per site per margin Corresponding LR N per site per margin Corresponding LR N per site per margin Corresponding LR

group group group group group group group

N, number of patients; N LR, number of patients with local recurrence. Note: 1. Disease-free patients had a minimum of 18 months follow-up. 2. Patients were excluded if margin status could not be histologically re-reviewed. 3. Local recurrence rates refers to the first recurrence only and is calculated as number of patients who experienced only local recurrence plus patients developing concomitant local and regional recurrence.

172

q 2005 Lippincott Williams & Wilkins

Am J Surg Pathol
Volume 29, Number 2, February 2005

Oral Squamous Cell Carcinoma

TABLE 5. Local Recurrences of 85 Mobile Tongue/Floor of Mouth SCC (Cohort 4) Per Margin Status and T Stage, Plus Adjuvant RT Status Group 1 T1

T2

T3

T4

Tx

Total

N N N N N N N N N N N N N N N N N N

all RT LR all RT LR (X) all RT LR (X) all RT LR (X) all RT LR all RT LR

12 0 3 6 5 0 1 0 3 2 1 (0) 0 22 7 4

Group 2

Group 3

Group 4

Total

15 5 0 15 9 5 (3) 7 5 3 (2) 1 1 0 1 1 1 39 21 9

6 3 0 5 4 1 (0)

2 2 0 3 1 1

0

0

4 2 1 (1)

3 3 1 1 1 0 9 7 2

35 10 3 29 19 17 8 5 3 11 8 3 2 2 1 85 44 17

0 15 9 2

N, number of patients; N LR, number of patients with local recurrence; N RT, number of patients receiving adjuvant radiotherapy; (X), number of patients with local recurrence who did not receive adjuvant radiotherapy; Tx, unknown T stage.

assigned a score value to these variables (Table 9). We now propose a novel risk classification that is strongly predictive of LR and OS. We were able to develop risk classification algorithms that generated the best fit, and classified patients into low, intermediate, and high risk. The rationale behind the point assignments (0, 1, 3) rather than (0, 1, 2) is that we postulated that certain features (POI type 5 vs. type 4, PNI of large nerve vs. small nerves) represented significant steps ‘‘downward’’ viz-a-viz prognosis. Thus, we reasoned that this point assignment would better discriminate intermediate- from high-grade risk. We scored Cohort 1 according to this threetiered risk assessment. Kaplan-Meier survival analysis demonstrated that our risk assessment was predictive of both OS (Fig. 6) and LR (not shown) (log rank, P , 0.0001 and P = 0.0004, respectively). The curve representing intermediate risk is relatively equidistant from the low- and high-risk curves graph (Fig. 6), consistent with the reasoning stated above.

FIGURE 4. Kaplan-Meier analysis for LR (Cohort 3) per margin group (P = 0.2117).

Risk Assignment, Treatment, and Overall Survival: Cohort 2 To investigate the impact of treatment on this group, we studied Cohort 2 patients treated with primary resection (n = 79), and those also receiving postoperative, adjuvant RT (n = 124). The addition of adjuvant RT had no impact on OS, whereas the risk assignment significantly predicted OS. Cox regression analysis revealed that the risk status was strongly associated with OS (P , 0.001), whereas the addition of adjuvant RT had no impact (P = 0.830). Separate analyses were then performed for Cohort 2 patients sorted by each risk group. Adjuvant RT had no impact on OS for low-risk patients (log rank, P = 0.87) or for intermediate-risk patients (log rank, P = 0.96). However, a trend was seen for increased OS with adjuvant RT administered to high-risk patients (log rank, P = 0.06) (graphs not shown).

TABLE 6. Frequency Distribution of Perineural Invasion With Margin Status (Chi-square P = 0.003) Perineural Invasion Margin Status Group Group Group Group

1 2 3 4

None

,1 mm

.1 mm

90% 69% 75% 61%

7% 18% 19% 16%

3% 13% 6% 23%

Note: x2 (P = 0.003). Group 1, initially .5 mm intraoperatively; group 2, .5 mm after harvesting supplemental margins intraoperatively; group 3, final pathology reveals margins ,5 mm; group 4, final pathology reveals frankly positive margins.

q 2005 Lippincott Williams & Wilkins

FIGURE 5. Kaplan-Meier analysis for OS (Cohort 3) per margin group (P = 0.8351).

173

Am J Surg Pathol
Volume 29, Number 2, February 2005

Brandwein-Gensler et al

TABLE 7. Distribution Frequencies of Worst Pattern of Invasion, With Respect to Margin Status (P = 0.0001) and Perineural Invasion (P = 0.003)

Group 1 Group 2 Group 3 Group 4 No PNI PNI ,1 mm PNI .1 mm

WPOI 1, 2, 3 Grouped (%)

WPOI 4 (%)

WPOI 5 (%)

32 15 17 19 26 5 0

65 55 55 42 56 64 38

3 30 28 39 17 31 62

Note: Percentages are across rows. Group 1, initially .5 mm intraoperatively; group 2, .5 mm after harvesting supplemental margins intraoperatively; group 3, final pathology reveals margins ,5 mm; group 4, final pathology reveals frankly positive margins; PNI, perineural spread; WPOI, worst pattern of infiltration.

Risk Assignment, Treatment, and Local Disease-free Survival: Cohort 2 Treatment was analyzed for local disease-free survival using Cox survival analysis. When the 203 patients were grouped together, our risk assignment (P = 0.017) and administration of adjuvant RT (P = 0.027) both impacted LR by Cox regression analysis (graphs not shown). For this reason, we performed separate Kaplan-Meier analyses on each risk group. The results demonstrate that for low-risk patients (n = 20), the addition of adjuvant RT has no impact on local disease-free survival (log rank, P = 0.56). For intermediaterisk patients (n = 86), the addition of adjuvant RT has no impact on local disease-free survival (log rank, P = 0.53). However for high-risk patients (n = 98), the addition of adjuvant RT does improve local disease-free survival (log rank, P = 0.0296) (Fig. 7).

DISCUSSION Meaning of Margins The report by Byers et al, in 1978, on 216 patients with OSCC treated by surgery alone, serves as an important historical control regarding the significant improvement in local disease-free and OS for patients with negative resection margins.6 An extensive review of the impact of resection margins is beyond the scope of this manuscript. But to

summarize, many groups have demonstrated the local diseasefree and OS benefits of achieving negative resection margins.6,7,11,12,15,16,19–22 We could not demonstrate the same survival benefits with negative resection margins (Figs. 4, 5). The impact of resection margins was not obfuscated by the confounding factor of adjuvant RT, as we had corrected for this. There are at least two explanations for our findings. First, the definition of ‘‘positive margins’’ varies, so that what may be deemed adequate by one group is considered inadequate by another. A perusal of Table 10 reveals that the local recurrence rate (LRR) for patients with positive margins varies widely from 22% to 80%. A trend can also be seen, as the LRR for patients with inadequate resection margins decreases between the 1980s and 1990s. When comparing only the reported studies in which the definition of inadequate margins is similar to our definition, the LRR ranges from 22% to 71%.7,15,16,20,21 The weighted average LRR for inadequate margins for these five reports is 31%. Our LRR for inadequate margins is 18% (9 of 49 patients, Table 4, all sites, all stages), then compares favorably with corresponding reports. Resection margins from the 1970s were not the same as resection margins nowadays. Advances in free-flap reconstructive techniques have allowed surgeons to remove tumors previously deemed unresectable. So our inability to demonstrate significant differences between in LRR and OS in patients with adequate versus inadequate resection margins may be due to the overall improved outcome of this later group. Second, the definition of LR may also vary with respect to SPT. Recently, van Es et al reported on a series of 82 patients with adequately resected T1 and T2 oral SCC treated by surgery alone.23 Negative resection margins were defined as at least 5 mm, or greater than the intertumor satellite distance. Their reported LRR was significantly lower than ours (4% vs. 16%, respectively, for T1 and T2 SCC of tongue) (Table 5). The series reported by van Es et al23 and Slootweg et al21 defined SPT adjacent to the index tumor as any tumor developing 2 years or later after resection (personal communication). Molecular profiling studies can conclusively distinguish SPT adjacent to the index tumor from LR. Since these tests were not part of this study, we took a very conservative approach in our diagnoses of SPT; any SCC that subsequently developed adjacent to the first SCC was considered LR. Our rate of SPT for tongue primary SCC is 4%. If, for instance, we applied the same definition to distinguish LR from SPT (2 years), then our SPT rate would

TABLE 8. Comparison of POI, Lymphocytic Response, and Perineural Invasion Variable

Local Recurrence

Regional Metastasis

Overall Survival

PPOI 4 PPOI 5 WPOI 4 WPOI 5 Lymphocytic response, weak or none Perineural invasion ,1 mm Perineural invasion .1 mm

NS NS WPOI 4 vs. 5 95% CI 0.86, 5.01 P = 0.015

NS NS NS NS NS NS NS

95% CI 1.07, 3.60 P = 0.024 95% CI 1.78, 8.38 P = 0.001 HR 2.0 95% CI 1.62, 8.77 P = 0.004 HR 6.4 95% CI 2.43, 13.97 P = 0.001 HR 6.2 95% CI 2.88, 14.18 P = 0.001 HR 2.3 95% CI 1.36, 3.95 P = 0.002 HR 1.9 95% CI 1.42, 4.81 P = 0.039

95% CI 1.47, 9.21 P = 0.005 NS 95% CI 1.43, 7.89 P = 0.005

NS, not significant; HR, hazard ratio.

174

q 2005 Lippincott Williams & Wilkins

Am J Surg Pathol
Volume 29, Number 2, February 2005

Oral Squamous Cell Carcinoma

TABLE 9. Proposed Risk Assessment for Oral Squamous Cell Carcinoma Point Assignment for Risk Scoring Histologic Variable

0

Perineural invasion Lymphocytic infiltrate at interface WPOI at interface

None Continuous band 1 or 2 or 3

Small nerves Large patches 4

1

3

Risk Score (sum of all point assignments)

Risk for local Recurrence

Overall Survival Probability

Adjuvant Treatment Recommendations

Score = 0 1 or 2 3 to 9

Low Intermediate High

Good Intermediate Poor

No local disease-free benefit seen for adjuvant RT No local disease-free benefit seen for adjuvant RT RT regardless of 5 mm margins

Large nerves Little or none 5

rise to 8%, while our LRR would fall to 12% for 65 patients with T1 and T2 tongue SCC. We analyzed Cohort 4 (mobile tongue/floor of mouth cancers) patients, as this anatomic region allows for relatively simple harvesting of supplemental margins. We would logically predict that, when stratified for T stage, the LRR for group 1 margins would be lowest, followed by group 2, then group 3, and lastly group 4. The closer the LRR for group 2 margins is to group 1 margins, the more effective our supplemental margins, in response to the intraoperative feedback. We do see this effect when grouping T1/T2 tongue carcinomas. The LRR for group 1 margins is 16% (3 of 18) and for group 2 margins is also 16% (5 of 30). However, this association breaks down when we stratify for T stages. For T1 tongue carcinomas (all margin groups), the likelihood of remaining disease free at the primary site is seen as a function of adjuvant RT (Table 5) (0% LRR for 10 RT+ patients, 13% LRR for 22 RT2 patients) (P = 0.0027). The only recurrences in this group developed in patients within the group 1 margins

who were not treated with adjuvant RT. For T2 tumors, we also see the impact of RT on LRR: 16% for RT+ versus 40% for RT2 patients (P = 0.001) but no direct association between margin status and LRR.

FIGURE 6. Kaplan-Meier OS curves classified by risk assessment (Cohort 1).

FIGURE 7. Kaplan-Meier local disease-free survival curves for Cohort 2 high-risk patients only (n = 98).

q 2005 Lippincott Williams & Wilkins

Significance of Pattern of Invasion Our intention was to build upon previous studies1–5 regarding the predictive value of POI at the tumor interface. We expanded on the traditional Bryne POI, adding a pattern 5, which is defined as tumor satellites (regardless of size) dispersed 1 mm or farther from the closest intervening tumor island. We also validated the process of considering only the worst pattern of invasion by comparing PPOI with worst pattern of invasion (WPOI) at the tumor/host interface. Using the Cox model, we found that both WPOI 4 (P = 0.004) and WPOI 5 (P = 0.001) were significantly related to OS when compared with WPOI 1-3. As WPOI and PPOI were both predictive of OS, it is valid to use WPOI as a variable in place of PPOI. We found, indeed, that WPOI was associated with LR, whereas PPOI was not.

175

Am J Surg Pathol
Volume 29, Number 2, February 2005

Brandwein-Gensler et al

TABLE 10. Historical Review of Local Recurrence Rates and Definitions for Resection Adequacy Year

Author

N

Site

T

62

Intraoral plus lip

All

1976

Looser

1978

Byers

216

All sites

All

1986

Scholl

268

Tongue

All

1987

Chen

270

All sites

All

1990

Loree

303

Oral cavity

T1, T2

1993

Jacobs

687

All

1996

Van Es

82

All sites Tongue

1996

Jones

352

All sites

All

1998

Yuen

50

Tongue

All

1999

Woolgar

200

Intraoral

1999

Spiro

150

2001

Sieczka

2002

2004

176

T1, T2

Definition of Margins Negative .5 mm; close ,5 mm; positive = cut through or CIS Positive = cut-through or CIS

Supplemental Margins?

Therapy

LR for Positive and Close Margins

LR for Negative Margins

Survival, Positive Margins

Survival, Negative Margins

No

17: preop RT; all surgery

71%

32%

30.6% 5-yr OS

35.9% 5-yr OS

Yes

Surgery only

80%

12–18% depending on site

10 0% 2-yr OS

Positive = cut-through

Yes

Mixed regimens

15%

Negative .5 mm; close ,5 mm; positive = cut through or CIS Negative .5 mm; close ,5 mm; positive = cut through

No

Mixed regimens

Small group with poor survival 7% 5-yr DF

59–77% 2-yr OS (46–86% 2-yr OS for negative supplemental group) 49% 5-yr OS

No

Surgery +/2 RT

No

Mixed regimens Surgery only

Close ,5 mm or intertumor distance . distance to margin; positive = cut through Positive = cut through, or up to margin; close not defined

No

55% for positive, 40% for close 30%

17% for T1, T2 27% T3, T4 38/215 (18%)

25%

11%

NA

4% (3/76)

No

Neoadjuvant radiation and surgical salvage

66% at 5 yr

47% at 5 yr

Close ,10 mm positive = cut through

No

16% (8/49)

NA

All

Close ,5 mm positive = cut through

No

All with surgery, 5 with preop RT; some with postop RT Surgery, adjuvant RT for margins ,5 mm

Tongue

All

Close = 1 HPF; Positive = cut through

No

Not explicitly delineated

36%

12%

27

Buccal

All

No

Most: surgery alone

66%

50%

Slootweg

394

All sites

All

No

Not specified

22% (41/187)

4% (8/207)

Brandwein

168

All sites

All

Close ,5 mm; positive = cut through Included any dysplasia, plus margins ,5 mm or intertumor distance .distance to margin Close ,5 mm positive = cut through

13% group 3; 26% group 4

19% group 1; 26% group 2

Yes

Not addressed

52% 5-yr OS (all T’s)

39% 5-yr DF

60% 5-yr OS (all T’s)

Disease-specific survival after recurrence and surgical salvage lower for patients with positive margins as compared to negative margins, however not statistically significant Not addressed

Margin status (clear vs. close vs. involved) strongly correlated with disease-free survival (P = .0001), however, no rates supplied 5-yr OS was similar in both groups (60%) 75% 66% 5-year 5-year survival survival

q 2005 Lippincott Williams & Wilkins

Am J Surg Pathol
Volume 29, Number 2, February 2005

By what mechanism does POI impact upon survival? It is well established that the N status at presentation is the single-most predictive value for OSCC. WPOI has been previously demonstrated as significantly associated with positive lymph nodes.13,24 We also demonstrate the association between WPOI and initial N status. WPOI (collapsed 1, 2, 3 vs. 4, 5) did associate with initially positive lymph node status as a two-tailed x2 analysis (P , 0.001).

Significance of Lymphocytic Response and Perineural Invasion We demonstrate that weak or limited lymphocyte response at the tumor/host interface is strongly associated with LR (P = 0.005) and death (P = 0.001). We have found that shifting from the traditional four-tiered scoring of lymphocyte response to this three-tiered scoring enhances interobserver reproducibility. Other studies have also demonstrated an inverse relationship between lymphocytic infiltrate and potential for lymph node metastasis9 as well as survival.14 Hosal et al have demonstrated that poor lymphocyte response was associated with locoregional recurrence.10 To our knowledge, our study is the first to demonstrate a strong, inverse correlation between lymphocytic infiltrate and LR upon multivariate analysis. PNI is a known prognosticator of survival and LR.8 We found that PNI of small and large nerves was associated with OS (P = 0.002 and 0.039, respectively). PNI of large nerves was associated with LR (P = 0.005).

Our Risk Classification Based on these findings, we propose a novel risk assessment classification based on three histologic features. This scoring, in our opinion, can serve as the rationale for whether adjuvant RT to the primary site is indicated. To correct for the confounding effect of adjuvant RT, we studied Cohort 2 with respect to risk classification, studying patients who were uniformly treated with primary resection versus those who also received adjuvant RT. Our findings do confirm that there is no local disease-free benefit in administering adjuvant RT for low-risk and intermediate-risk patients. However, a definite local disease-free benefit was observed for high-risk patients treated with adjuvant RT. For these analyses, we chose to perform separate Kaplan-Meier analyses instead of Cox regression analysis. The Kaplan-Meier analysis is not modeled on any underlying assumptions, and it provides more realistic survival curves. We previously referred to 12 patients with T1 SCC of the mobile tongue, primarily treated by surgery alone, and group 1 margins, who had developed LR. When we revisited these cases, it was no surprise that all of these patients had at least one of the three histologic findings associated with elevated risk. Eleven of these patients had high-risk scores, and 1 patient had an intermediate-risk score.

Future Directions From a statistical viewpoint, the burden of validation lies in the application of our risk assessment on a new data set. From a clinicopathologic viewpoint, this is not presently q 2005 Lippincott Williams & Wilkins

Oral Squamous Cell Carcinoma

feasible, as our study represents the accrual of data collected over one decade. Other hospital centers can retrospectively apply this risk scoring system to validate its significance. While our data support the idea that intermediate- and low-risk patients do not benefit from adjuvant RT, it should not be implied that inadequate resection margins are acceptable. The relative percentages of frankly positive resection margins in our low-risk and intermediate-risk patients were small (2 of 20 and 6 of 86, respectively). Of the intermediate-risk patients, 5 of 6 with positive margins received adjuvant RT. Of the low-risk patients, 1 of 2 with positive margins received adjuvant RT. Do high-risk patients have disease that obligates them to receive adjuvant RT? Or is there some margin distance (eg, 2 cm) that would obviate this need in high-risk patients. It may be possible that high-risk patients might not have margin-dependent disease. It remains to be seen as to which modality (greater margin distance vs. adjuvant therapy) could be of benefit to these patients. Future prospective studies are necessary. One of the immediate practical applications of our data is that we can logically recommend adjuvant RT to a ‘‘new’’ group: patients with T1/T2 N0/N1 SCC who have negative margins but are classified as high-risk. Perineural spread is a ‘‘traditional’’ finding that usually merits adjuvant RT. But in our risk classification, one may be classified as high risk without having perineural tumor spread. For instance, limited lymphocytic response, plus any WPOI, would yield a score of 3 or more, which is automatically classified as high risk. On the other hand, there are implications of our data that may, at face value, be more difficult to accept. For instance, our data support the idea that patients with low or intermediate risk do not benefit from adjuvant RT. Therefore, low- or intermediate-risk patients with inadequate margins might be better treated by re-resection rather than adjuvant RT. This type of issue is better addressed by future prospective studies. In summary, we advocate a very simple scoring system that can group patients into three categories (low-risk, intermediate-risk, and high-risk) and are predictive of LR and also OS. We demonstrate that high-risk patients can benefit from adjuvant RT to the primary site, with respect to local disease-free survival, but no benefit is observed for low- and intermediate-risk patients. The optimum tumor clearance distance is not established, but we can conclude that if the patient has a high-risk score, adjuvant RT is indicated despite resection margins of 5-mm clearance. ACKNOWLEDGMENT The authors thank Dr. Dave Marshall, MD, Department of Radiation Oncology, MSSM, for helpful discussions. REFERENCES 1. Anneroth G, Batsakis J, Luna M. Review of the literature and a recommended system of malignancy grading in oral squamous cell carcinoma. Scan J Dent Res. 1987;95:229–249. 2. Bundegaard T, Rossen K, Henriksen SD, et al. Histologic parameters in the evaluation of T1 squamous cell carcinomas of the oral cavity. Head Neck. 2002;24:656–660.

177

Brandwein-Gensler et al

3. Bryne M, Koppang HS, Lilleng R, et al. New malignancy grading is a better prognostic indicator than Broders’ grading in oral squamous cell carcinomas. J Oral Pathol Med. 1989;18:432–437. 4. Bryne M, Koppang HS, Lilleng R, et al. Malignancy grading of the deep invasive margins of oral squamous cell carcinomas has high prognostic value. J Pathol. 1992;166:375–381. 5. Bryne M, Jenssen N, Boysen M. Histologic grading in the deep invasive front of T1 and T2 glottic squamous cell carcinomas has high prognostic value. Virchows Arch. 1995;427:277–281. 6. Byers RM, Bland KI, Borlase B, et al. The prognostic and therapeutic value of frozen section determination in the surgical treatment of squamous cell carcinoma of the head and neck. Am J Surg. 1978;136:525–528. 7. Chen TY, Edmrich LJ, Driscoll DL. The clinical significance of pathological findings in surgically resected margins of the primary tumor in head and neck carcinoma. Int J Radiat Oncol Biol Phys. 1987;13:833–837. 8. Fagan JJ, Collins B, Barnes L, et al. Perineural invasion in squamous cell carcinoma of the head and neck. Arch Otolaryngol Head Neck Surg. 1998;124:637–640. 9. Hiratsuka H, Miyakawa A, Nakamori K, et al. Multivariate analysis of occult lymph node metastasis as a prognostic indicator for patients with squamous cell carcinoma of the oral cavity. Cancer. 1997;80:351–356. 10. Hosal AS, Unal OF, Ayhan A. Possible prognostic value of histopathologic parameters in patients with carcinoma of the oral tongue. Eur Arch Otolaryngol. 1998;225:216–219. 11. Jacobs JR, Ahmad K, Casiano R, et al. Implications of positive surgical margins. Laryngoscope. 1993;103:64–68. 12. Jones AS. Prognosis in mouth cancer: tumor factors. Oral Oncol Eur J Cancer. 1994;30B:8–15. 13. Kim HC, Kusukawa J, Kameyama T. Clinicopathologic parameters in predicting cervical nodal metastasis in early squamous cell carcinoma of the oral cavity. Kurume Med J. 1993;40:183–192. 14. Kurokawa H, Yamashita Y, Murata T, et al. Histologic grading of malignancy correlates with regional lymph node metastasis and survival of patients with oral squamous cell carcinoma. Fukuoka Acta Med. 1998;89:225–231.

178

Am J Surg Pathol
Volume 29, Number 2, February 2005 15. Looser KG, Shah JP, Strong EW. The significance of ÔpositiveÕ margins in surgically resected epidermoid carcinomas. Head Neck Surg. 1978;1:107– 111. 16. Loree TR, Strong EW. Significance of positive margins in oral cavity squamous carcinoma. Am J Surg. 1990;160:410–414. 17. Martinez-Gimeno C, Rodriguez EM, Vila CN, et al. Squamous cell carcinoma of the oral cavity: a clinicopathologic scoring system for evaluating risk of cervical lymph node metastasis. Laryngoscope. 1995;105:728–733. 18. Odell EW, Jani P, Sherriff M, et al. The prognostic value of individual histology grading parameters in small lingual squamous cell carcinomas. Cancer. 1994;74:789–794. 19. Scholl P, Byers RM, Batsakis JG, et al. Microscopic cut-through of cancer in the surgical treatment of squamous carcinoma of the tongue: prognostic and therapeutic implications. Am J Surg. 1986;152:354–360. 20. Sieczka E, Datta R, Singh A, et al. Cancer of the buccal mucosa: are margins and T-stage accurate predictors of local control? Am J Otolaryngol. 2001;22:395–399. 21. Slootweg PJ, Hordijk GJ, Schade Y, et al. Treatment failure and margin status in head and neck cancer: a critical view on the potential value of molecular pathology. Oral Oncol. 2002;38:500–503. 22. Spiro RH, Guillamondegui O, Paulino AF, et al. Pattern of invasion and margin assessment in patients with oral tongue cancer. Head Neck. 1999;21:408–413. 23. van Es RJ, van Nieuw Amerongen N, Slootweg PJ, et al. Resection margin as a predictor of recurrence at the primary site for T1 and T2 oral cancers: evaluation of histopathologic variables. Arch Otolaryngol Head Neck Surg. 1996;122:521–525. 24. Yamamoto E, Miyakawa A, Kohama GI. Mode of invasion and lymph node metastasis in squamous cell carcinoma of the oral cavity. Head Neck Surg. 1984;6:938–947. 25. Yuen APW, Lam KY, Lam LK, et al. Prognostic factors of clinically Stage I and Stage II oral tongue carcinoma: a comparative study of stage, thickness, shape, growth pattern, grading, Martinez-Gimeno score, and pathologic features. Head Neck. 2002;24:513–520.

q 2005 Lippincott Williams & Wilkins