Introduction

Brandwein et al. proposed the worst pattern of invasion (WPOI) as a histologic variable in 2005 [1]. It has since been validated as a prognosticator in oral squamous carcinoma (OSC) [1,2,3,4,5,6,7,8,9,10,11,12,13]. The histologic risk model incorporated a new class of WPOI known as WPOI-5, which portends significantly poorer outcomes in OSC patients as compared to WPOI-4 [1].

Currently, WPOI-5 is an American Joint Committee on Cancer (AJCC) 8th edition registry data collection variable and a reporting element in the College of American Pathologists (CAP) synoptic for oral cancers [14, 15]. Previous studies have established that WPOI-5 is significantly predictive of locoregional recurrence (LRR) and disease specific survival (DSS) [3]. Thus, the ability to accurately identify WPOI-5 intraoperatively has implications for real-time surgical decision-making. Kohler et al. demonstrate that WPOI affects the extent of tumor resection since more aggressive tumors (WPOI-4 and WPOI-5) require more extensive margins to minimize risk of LRR [16]. WPOI-5 has significant implications for the management of regional lymph nodes as well. The objectives of this study are to determine the following: (1) The association of WPOI-5 with occult cervical metastases (OCM), and (2) The accuracy of WPOI-5 identification at frozen section analysis. Additionally, a link to a teaching module on the recognition of WPOI-5 is included in this manuscript.

Materials and Methods

Retrospective Study

The Institutional Review Board approved the study of OSC patients who were clinically/radiologically cN0 and who underwent primary resection and elective neck dissection (END). The data collected included 8th edition AJCC T and N stage, depth of invasion (DOI), WPOI, size of positive lymph nodes, perineural invasion (PNI), and extranodal extension (ENE). All cases were originally diagnosed by MBW and then re-reviewed for this study. DOI was measured either using a digital pathology platform or, for pre-digital cases, by overlaying the glass slides with an acetate-printed millimeter ruler. The DOI was measured from the estimated position of basal reserve cells to the furthest invading tumor islands. Data were stored in a secure database. Sensitivity, specificity, and predictive values tests were performed online using the MedCalc© diagnostic test evaluation calculator (MedCalc Software Ltd. 2022).

Prospective Study

Following Institutional Review Board approval, 47 patients undergoing 51 surgeries were enrolled over 30 months from a single institution (Mount Sinai West Hospital, New York, NY). Informed consent was obtained from each patient. All patients underwent standard of care surgery for biopsy-proven OSC—either primary resection (n = 47) or salvage resection (n = 4). Concordance between intraoperative and final pathology WPOI classification was examined. WPOI was determined as part of routine resection margin assessment as follows: T1 and T2 carcinomas are examined in entirety and T3/T4 tumors are examined generously at the time of frozen section during margin assessment. Clinicopathologic information (surgical procedure, frozen WPOI, final WPOI, and pathological T-stage) were recorded and stored in a secure database on the internal Mount Sinai Hospital network. The frozen section WPOI data were collected either from (1) an intraoperative frozen section WPOI determination recorded in real-time by a single pathologist with expertise in head and neck pathology or based on (2) hematoxylin and eosin frozen section slides reviewed ex post facto by the same pathologist, blinded to permanent section results. WPOI was classified as either non-aggressive (WPOI 1–3), WPOI-4, or WPOI-5. Final WPOI from permanent section was retrieved from Dr. Brandwein-Weber’s pathology reports. Each frozen WPOI was compared with the corresponding final WPOI. Since the variable of interest in the current study is WPOI-5, for purposes of assessing accuracy, WPOI-4 was lumped together with non-aggressive patterns of invasion. Sensitivity, specificity, and predictive value tests were performed online using the MedCalc© diagnostic test evaluation calculator (MedCalc Software Ltd. 2022).

Results

The retrospective study was comprised of 228 patients with pT1/pT2 cN0 OCM (staged by AJCC 8th edition) who underwent primary resection and END. Thirteen pT1 patients had occult cervical metastases; 10 with WPOI-5, 2 with WPOI-4, and 1 with WPOI-3 (Table 1). The majority of pT1 WPOI-5 carcinomas (24/26) had DOI ≥ 4 mm, with most occurring between 4 and 5 mm (Table 2). Only two pT1 WPOI-5 carcinomas had DOI < 4 mm (3.0 and 3.5 mm), and both patients had OCM. PNI data is known for 20/26 pT1 patients with OCM. This includes only one of the two pT1 cases with both WPOI-5 and DOI < 4 mm. PNI was present in two of 20 cases; the aforementioned case with both WPOI-5 and DOI < 4 mm was negative for PNI.

Table 1 Occult metastases for pT1/pT2 cN0 by WPOI (N = 228) (%)
Full size table
Table 2 Distribution of occult metastases for pT1 by depth of invasion (DOI) (%)
Full size table

WPOI-5 was significantly predictive of OCM in 26 of 79 pT1 patients as compared to WPOI-4/WPOI-3 (p < 0.0001) (Table 1). The sensitivity, specificity, positive predictive value, negative predictive value, and accuracy of predicting OCM in pT1 patients were 50%, 94.3%, 81.3%, 79.4% and 81.1%, respectively.

If all pT1 pN0 patients are treated with END, the expected overtreatment rate is 80%. This rate can be reduced to approximately 50% if only patients with pT1 AND WPOI-5 disease received END. However, in doing so, OCM would be missed in 4% of patients (pT1 pN + WPOI-4 / WPOI-3). If END were performed in our cohort for all pT1 ≥ 4 mm (regardless of WPOI) or all WPOI-5 (regardless of DOI), then no OCM would have been missed (p = 0.017, 100% sensitivity, 29% specificity, 77% positive predictive value, 23% negative predictive value).

In the prospective study of the 51 tumor samples (Table 3), 15 were WPOI-5, 15 were WPOI-4, and 21 were non-aggressive (WPOI 1-3) on final pathology (Table 4). Intraoperative consultation correctly assigned WPOI in 42 of 51 cases (82.4%). Eleven pairs were classified as “true positive,” meaning the diagnosis of WPOI-5 was rendered on both frozen and permanent sections. Thirty-six pairs were considered “true negative,” meaning the tumor was not classified as WPOI-5 on frozen or permanent section.

Table 3 Prospective cohort: 47 patients with 51 resections
Full size table
Table 4 WPOI concordance between intraoperative/permanent section (N = 51, p = 0.017)
Full size table

Four pairs were classified as “false negative,” meaning the tumor was reclassified to WPOI-5 on permanent section after being classified as non-aggressive or WPOI-4 on frozen section. There were no false positives. We tested the hypothesis that this discrepancy, which is related to sampling error, correlates with increased tumor size (pT1/pT2 vs pT3/pT4 or recurrent tumor ≥ 4 cm). We found a trend (p = 0.059) which supports this hypothesis (Fisher' exact test, one-tailed).

With respect to identifying WPOI-5, the accuracy, sensitivity, and specificity were 92.16%, 73.33%, and 100.0%, respectively (Table 5). The most common discrepancy was between non-aggressive on frozen and WPOI-4 on permanent (n = 4). This was followed by WPOI-4 on frozen to WPOI-5 on permanent (n = 3). All these discrepancies were due to sampling issues. There was one case of non-aggressive on frozen reclassified to WPOI-5 on permanent (sampling error) and one case of WPOI-4 on frozen reclassified to non-aggressive on permanent (interpretive error).

Table 5 WPOI-5 identification during intraoperative consultation compared to final pathology results
Full size table

Discussion

WPOI was proposed as a prognosticator in OSC by Brandwein et al. in 2005 [1], validated internally [2, 3] and by numerous external international groups [4,5,6,7,8,9,10,11,12,13]. WPOI is categorized as WPOI-1 through WPOI-5. Tumors identified as WPOI-1, WPOI-2, or WPOI-3 are all considered “non-aggressive patterns of invasion,” whereas WPOI-4 and WPOI-5 are considered “aggressive patterns of invasion.” Generally, tumors with non-aggressive pattern of invasion do not extend beyond their perimeter. WPOI-1 is defined as a pushing border, WPOI-2 is defined as finger-like growth, and WPOI-3 is defined as large separate islands, attached or detached but confined within the tumor perimeter, with more than 15 cells per island. Tumors with aggressive pattern of invasion demonstrate convincingly discontiguous cancer satellites. WPOI-4 is defined as small tumor islands, separated from the main tumor mass, with 15 or fewer cells per island. Carcinomas are categorized as WPOI-5 if their satellites are dispersed (≥ 1 mm away from the main mass or neighboring satellites). The 8th edition of the AJCC Staging Manual: Oral Cavity includes WPOI-5 as a recommended reported feature, as this variable is significantly predictive of poorer outcomes [14]. While the most common tumor dispersion phenotype is spread through soft tissue, dispersion may also be the result of extratumoral PNI or lymphovascular tumor emboli in more rare instances [3, 17]. WPOI-5 demonstrates a positive predictive value of 42% for LRR and is significantly predictive of DSS on multivariate analysis when adjusted for confounders [3]. Given the predictive nature of WPOI-5 for disease progression and survival outcomes, this variable can be a useful tool to inform clinical decision-making.

Intraoperative consultation during head and neck resections is the standard of care with respect to margin assessment. Several studies specifically analyzed the consistency between frozen section and permanent section margin analysis. Layfield et al. analyzed 1796 corresponding pairs of frozen and permanent sections to determine the accuracy of intraoperative margins for primary head and neck squamous carcinomas [18]. Concordance was 97%, and discrepancies were identified in only 55 pairs. These discrepancies were predominantly false negatives, in that the frozen section was negative and the permanent section was positive or close [18]. Similar yet smaller studies concur that frozen section is a highly accurate method for clearing tumor margins intraoperatively [19,20,21,22,23]. These studies also shed light on the limitations associated with frozen sections [20,21,22,23,24,25]. One limitation is the unusual event of positive margins in the final report which were not detected during frozen tissue analysis [23]. Possible sources of sampling errors may come from either undetected tumor in deeper sections within the frozen tissue block or from sampling additional tissues for permanent sections. Serinelli et al. analyzed whether sampling additional deeper levels from the frozen section blocks improves concordance [26]. They compared 654 tissue blocks: 532 had two slides cut during frozen section and 122 blocks had ≥ 3 slides cut. They found no significant difference in concordance, suggesting that examination of deeper frozen sections might not reduce discordance [26]. Sampling the entire specimen in pT1 and pT2 OSC obviates the possibility of tissue sampling discordance.

This is the first study to demonstrate that WPOI-5 can be accurately identified intraoperatively (92.2%). The specificity of identifying WPOI-5 on frozen section was 100%. In other words, no cases were misassigned as WPOI-5 on frozen section. This is meaningful as it prevents unnecessarily aggressive surgical management. Four cases were identified as WPOI-5 only on permanent section (false negative) due to sampling errors, which is an expected limitation of intraoperative assessment.

If the patient’s tumor is identified as WPOI-5 only at permanent section, the patient can return to the operating room for more surgery, if required. On the other hand, more aggressive surgery (e.g., wider resection of the tumor, END) should be considered if WPOI-5 is identified on frozen section [16] since WPOI-5 is associated with inadequate resection margins [1]. Importantly, Kohler and colleagues have recently confirmed that optimal margin distance is influenced by WPOI in a large retrospective study of 772 intraoral cancer patients. They demonstrated that the optimal resection margin distance for tumors with non-aggressive WPOI was 1.7 mm, whereas the optimal distance for grouped WPOI-4/WPOI-5 tumors was 7.8 mm [16]. While our current standards of care will not be changed solely based on conclusions drawn from retrospective studies, their study should serve as the basis for prospective study designs that will arrive at new treatment recommendations.

A potential study limitation is that three frozen section cases were reviewed ex post facto. However, given the time lapse (“wash out” period) between signing out the permanent section results and reviewing the frozen slides while blinded to the final report, this did not influence study results.

Beyond resection margin distances, depth of invasion (DOI) is another important feature that can be determined during surgery, as it predicts occult cervical metastases (OCM) and influences decisions regarding END. The accuracy of intraoperative DOI is 96.8%, confirming its reliability as a frozen section parameter [27]. Van Lanschot and colleagues investigated the optimal cut-off value for indicating END in early-stage oral cavity SCC (n = 300) and report that DOI ≥ 4 mm was appropriate for END [28]. Some of the debate regarding the optimal DOI cut point may be caused by inaccurately considering tumor thickness (TT) and DOI as synonymous [29, 30]. When accurate and distinct definitions are utilized for these two prognostic features, only DOI accounts for exophytic and ulcerative tumors, which strengthens its value as a prognostic factor [30]. Multiple reports have supported a cut-off value of 4 ± 0.5 mm [29,30,31].

Other groups have also looked at WPOI-5 in the context of OCM. Verma et al. demonstrated that WPOI-5 was predictive of OCM (p = 0.0213) for a combined group of 189 pT1/pT2 patients but did not stratify for DOI [32]. Larson and colleagues were unable to demonstrate an association between WPOI-5 and occult cervical metastasis in a smaller group of pT1-T2/cN0 patients undergoing END [33]. This small cohort size of 35 pT1/cN0 patients included only one patient with OCM (personal communication). Shan and colleagues identified only one WPOI-5 tumor in a cohort of 145 OCM patients, casting doubt on their assessment [34]. Shimizu and colleagues studied 91 patients with clinically low-stage oral cancer; 75% of cancers were < 4 mm. Only 12 patients (13%) underwent END and only three patients (3.2%) had occult cervical metastasis [8]. Therefore, this study was not optimally designed to address predictors of OCM.

The current study is the first demonstration that WPOI-5 significantly predicts OCM for pT1 OSC. Thirteen pT1 patients had occult cervical metastases; 10 with WPOI-5, 2 with WPOI-4, and 1 with WPOI-3 (Table 1). Most pT1 WPOI-5 carcinomas are clustered at a DOI between 4 and 5 mm, thus these patients would have received END based on DOI. The three WPOI-3/WPOI-4 patients with OCM would also have received END based on DOI (Table 1). However, two patients with WPOI-5 had DOI < 4 mm, and both demonstrated OCM. As significance was lost for pT2 patients, we conclude that DOI > 5 mm is the dominant OCM predictor. If END were performed in our cohort for all pT1 ≥ 4 mm (regardless of WPOI) or all WPOI-5 (regardless of DOI), then no OCM would have been missed (p = 0.017, 100% sensitivity, 29% specificity, 77% positive predictive value, 23% negative predictive value). DOI ≥ 4 mm is the dominant determinant of OCM in this cohort. We would like to stress that WPOI-5 tumors with DOI < 4 mm are extremely rare. However, it is reasonable to suggest, based on these data, that END be performed for Stage I OSC with DOI < 4 mm and WPOI-5.

Our study assures that classifying WPOI both on frozen and permanent sections use the same approach as proposed by Brandwein-Gensler et al. [1]. For example, in both instances, if there are no satellites convincingly separate from the main tumor, then this is a non-aggressive pattern of invasion. If one sees separate tumor satellites at the advancing edge that are large (> 15 cells), this too represents non-aggressive pattern of invasion. If the separate tumor satellites at the advancing edge are small (≤ 15 cells), then the tumor is classified as WPOI-4. Tumor dispersion, characteristic of WPOI-5, can usually be appreciated at low-power as satellites are interspersed with sizable regions of normal soft tissue. The size of the satellites become immaterial. Dispersion is measured either between a satellite and the closest point within the main tumor, or between waves of satellites. A dispersion distance of at least 1 mm represents WPOI-5. We emphasize that, irrespective of frozen or permanent setting, having knowledge and experience with WPOI classification is essential for accurate risk stratification and prediction of OCM, as we demonstrate here.

The art of pathology, just like the art of surgery, has always been passed from teacher to student by experiential one-on-one teaching, akin to apprenticeship, be it at the multi-headed microscope, a computer screen, or at the operating room table [35, 36]. Samulski et al. emphasized the challenges associated with teaching and learning a predominantly visual-based field, such as pathology [35]. Assessing risk score and recognizing WPOI-5 are perfect examples within pathology that require experiential learning. We demonstrated substantial interrater agreement [κ = 0.64, 95% CI (0.46–0.79)] as well as substantial agreement between raters and the standard [κ = 0.87, 95% CI (0.69–1.00)] for risk classification [2]. Even with such high concordance, agreement was substantially improved after lecture participation, review of printed materials, and a multi-headed microscope teaching session, as compared with risk scoring after only reading published criteria (data not shown). Chang and colleagues expressed doubt regarding the reproducibility of WPOI-5 [37]. Heerema and colleagues investigated the reproducibility of pattern of invasion scoring and found no more than moderate interobserver agreement [38]. In this study, a significant limitation is the lack of comparison between their reads and the gold standard. Such suboptimal performance reflects inherent learning limitations when the experiential component is lacking. We have previously cautioned, “Surgical pathologists at other institutions who would like to use the Risk Model should first seek out practical training sessions” [3]. Within the histologic risk model, the rationale for migrating from high-risk categorization (evaluating three variables with a total of nine possible categories) to WPOI-5 as a single binary variable (yes or no) was to promote the inclusion of WPOI as a recommended feature in 8th AJCC Staging. To further increase adaptation and reproducibility of this prognosticator, we created a teaching module (YouTube video) to provide an active visual demonstration rather than solely relying on written text.

YouTube Link for WPOI Teaching Module by Dr. Margaret Brandwein-Weber: https://www.youtube.com/watch?v=k2dMAgml1H8

Conclusions

WPOI-5 predicts OCM in pT1 patients, which can impact decisions regarding END. The expected rate of overtreatment is 80% if all pT1 pN0 patients are treated with END. This rate can be reduced to approximately 50% if only patients with pT1 AND WPOI-5 disease receive END. However, in doing so, OCM would be missed in 4% of patients (pT1 pN + WPOI-4 / WPOI-3). Typically, DOI ≥ 4 is an indication for the surgeon to proceed with sentinel node or neck dissection. We show that for rare cases of WPOI-5 presenting with DOI < 4 mm, WPOI-5 should override the DOI. In such cases, it is reasonable for the surgeon to proceed with END. Since WPOI-5 is known to predict inadequate resection margins and is associated with a high risk of LRR [1, 17], intraoperative identification of WPOI-5 can have significant implications for the extent of tumor resection.