Programmed Death Ligand-1 Combined Positive Score Concordance and Interrater Reliability in Primary Tumors and Synchronous Lymph Node Metastases in Resected Cases of p16⁺ Oropharyngeal Squamous Cell Carcinoma

Following institutional review board approval, a retrospective review of pathology reports from 2012 through 2020 at the Northshore University Health System (Evanston, Illinois) was performed. A total of 38 patients were identified who had resection of a p16⁺ OPSCC with concurrent cervical lymph node involvement. None of these patients had received preoperative neoadjuvant therapy.

Per the standard procedure of our department, specimens were fixed in formalin and processed via Sakura Tissue-Tek VIP. After embedding in paraffin, 4-μm-thick sections were cut and stained with hematoxylin and eosin using standard techniques.

PD-L1 IHC was performed using the SP263 anti–PD-L1 clone (Ventana Medical Systems, Tucson, Arizona) stained on the Ventana benchmark ULTRA platform optimized with the OptiView DAB IHC Detection kit (Ventana Medical Systems) according to the manufacturer's instructions. Sections of placenta were included as positive controls. In each case, single sections of the PT as well as a cervical lymph node metastatic deposit (LM) were stained. Sections were selected in order to maximize the amount of tumor present on the slide for review. The PT slides were consecutively reviewed as a group followed by the LM slides in order to minimize potential bias in reviewing both the PT and the LM from a single case consecutively. Explicit randomization of slides was not conducted.

All slides were reviewed by 4 pathologists (A.P., M.S., W.W., L.L.) with extensive experience in PD-L1 interpretation across multiple tumor types. p16 IHC had been performed in the course of routine care in all cases and deemed positive based on overexpression at the conventional 70% cut point.

CPS scoring was performed as described in the KEYNOTE-048 trial⁴  and elsewhere.⁹  Briefly, tumor cells were counted as positive if they displayed any degree of perceptible membranous staining. Tumor-associated immune cells were identified by their location within the tumor or tumor-associated stroma and deemed positive if they showed any degree of cytoplasmic or membranous staining. Inflammatory cells determined to be associated with necrosis or normal lymphoid tissue were excluded. The CPS score was then calculated by the equation ([number of positive tumor cells + number of immune cells]/[total number of tumor cells]) × 100.

Data were evaluated in regard to the clinically relevant CPS cut points of 1 and 20 and as a continuous variable. Cases with a CPS score of less than 1 were considered negative, those with a score between 1 and 20 were regarded as low positive, and those with a CPS score of 20 or greater were classified as high positive. For each specimen, an average CPS score was calculated based upon the ratings of the 4 observers. A consensus diagnostic category was also determined based upon the rating of the majority of observers. In cases where a majority did not exist, the average CPS score was used to determine the diagnostic category.

For purposes of comparison, we also evaluated scoring data from 66 head and neck squamous cell carcinoma cases (not limited to p16⁺ OPSCC) that had been consecutively scored in the course of routine clinical care at our institution from 2019 to 2021. These cases consisted of resection specimens, small biopsies, and cytology cell blocks and included PTs as well as metastatic lesions. As our institution uses consensus scoring for routine PD-L1 studies, all cases were prospectively scored at the time of initial review by between 2 and 4 (A.P., M.S., W.W., L.L.) raters who were blinded to the other observers' scores and all of whom participated in the evaluation of CPS scoring in the current series.

Results were summarized with frequencies and percentages. A scatterplot was used to display each pair of average CPS scores. Cohen κ and weighted κ coefficients with 95% CIs measured agreement as a categorical variable at the clinically relevant cut points of 1 and 20 between the PTs and LMs and also among observers for a given specimen. Additionally, reliability on a continuous scale was determined using the 2-way mixed-effects, single-measures, consistency intraclass correlation coefficient (ICC).

All statistical analysis was performed using SAS 9.4 (SAS Institute, Cary, North Carolina).

We identified 38 resected cases of treatment-naive p16⁺ OPSCC for which both the PT and a synchronous LM were available for review. Scoring details are summarized in Table 1. Low-positive (16 versus 19) and high-positive (22 versus 19) consensus CPS scores were seen in both PTs and LMs. No specimen had a negative consensus CPS score (Figure 1, A through D).

Overall, discordant consensus scoring between PT and LM at the clinically relevant cut points was noted in 9 of the 38 paired specimens (24%) (Table 2). There was fair to substantial agreement between paired specimens using negative, low-positive, and high-positive consensus groups (weighted κ = 0.53; 95% CI, 0.26–0.79). Examples of cases with concordant scoring are provided in Figures 2, A through D, and 3, A through D. As a continuous variable, concordance between LM and PT scoring was fair to good (ICC = 0.64; 95% CI, 0.56–0.72) (Figure 4).

In 6 of the 9 discordant cases, the consensus CPS score obtained from the PT was high positive and that obtained from the LM was low positive. In 3 cases, the LM score was high positive and the PT score was low positive (Figure 5, A through D).

Interrater agreement based upon the clinically relevant cut points in regard to the 4 individual observers was fair to substantial and was similar for both PT and LM scoring (κ = 0.54 [95% CI, 0.36–0.71] and 0.51 [95% CI, 0.33–0.70], respectively) (Table 3). As a continuous variable, agreement between observers was good to excellent for both PT and LM (ICC = 0.72 [95% CI, 0.65–0.80] and ICC = 0.67 [95% CI, 0.60–0.75], respectively).

Interspecimen reliability and concordance for paired PT and LM was calculated for each of the 4 observers (Supplemental Table 1; see supplemental digital content, containing 2 tables, at https://meridian.allenpress.com/aplm in the April 2023 table of contents). Discordant scoring based on the clinically relevant cut points ranged from 18% to 32% for scoring performed by the individual observers, and agreement between PT and LM scoring ranged from fair to moderate (κ = 0.36–58).

Review of scoring data from our series of 66 prospectively evaluated head and neck squamous cell carcinoma cases demonstrated substantial to almost perfect agreement (κ = 0.84; 95% CI, 0.74–0.95) (Table 4; Supplemental Table 2) between observers that was superior to that seen in our current series of p16⁺ OPSSCs when considering either PT or LM. Levels of agreement are similar when considering only the 26 resected cases in our prospective series (κ = 0.85; 95% CI, 0.69–1.00). None of the 26 resected cases received a negative consensus score from the majority of raters, whereas 7 of 40 small biopsy/cell block specimens (18%) received a negative score from the majority of raters.

Pembrolizumab as a single agent was approved by the FDA for use as first-line therapy in head and neck squamous cell carcinoma in 2019 for patients with a CPS score of 1 or greater based on the findings of the KEYNOTE-048 trial.⁴  The eligibility criteria for this trial allowed the use of material from either PTs or metastatic lesions and permitted the use of small biopsies as well as resected tumors. At the time that this trial was conducted, it was unclear if substantial differences would exist between synchronous PT and LM specimens from the same patient.

In the context of non–small cell lung carcinoma, it has been well established that significant differences can exist in PD-L1 IHC scoring using the tumor proportion score system of evaluation in the setting of both metachronous and synchronous metastatic lesions. Depending on the clinically relevant cut point in question, discrepancies between synchronous resected PTs and mediastinal LMs have been found to occur in 33% to 45% of cases.^10,11  Similar observations highlighting significant rates of discordance between resected PTs and synchronous resected LMs have also been made in triple-negative breast cancer using immune cell area scoring,^12,13  as well as in invasive urothelial carcinoma.¹⁴ 

As such, the finding that CPS scoring at the clinically relevant cut point of 20 differs between resected p16⁺ OPSCCs and synchronous LMs in 24% of cases is not entirely surprising. Our results are largely in agreement with prior work that has also demonstrated variability between PTs and LMs in head and neck squamous cell carcinoma. Brcic et al¹⁵  found only moderate correlation between LMs and resected PTs. However, in contrast to our work, the tumor proportion scoring system was used, and scoring was evaluated as a continuous variable. Similar to our work, Schneider et al¹⁶  also examined PD-L1 expression in paired lymph nodes and PTs and found discordance in staining in 21 of 69 cases based on a nonstandard scoring system. Of note, their analysis used a PD-L1 antibody clone, 5H1, not commonly used in clinical practice. De Meulenaere et al¹⁷  performed PD-L1 IHC on 99 resected paired lymph nodes and PTs and found only fair agreement in tumor cell labeling, varying somewhat based on whether the 22C3 (κ = 0.258) or SP142 (κ = 0.338) antibody clone was used. CPS scoring was not included in this study.

Six of the discordant paired cases in our series displayed a greater degree of staining in the PT, whereas in 3, a greater degree of staining was seen in the LM. Recent work has documented both increased¹⁸  and decreased¹⁶  staining for PD-L1 IHC, albeit using nonstandard scoring techniques, in LMs when compared with the PT in head and neck squamous cell carcinoma. Although it is difficult to generalize based on the number of discordant cases in our series, there does not appear to be a clear bias toward increased or decreased staining in LMs when compared with synchronous PTs.

Another interesting observation is that negative scores (CPS <1) were uncommon, and no specimen in our series of resected p16⁺ OPSCC cases received a negative score from the majority of raters. This is a significant departure from the rate of negative cases (15%) seen in the KEYNOTE-048 study,⁴  which validated the use of PD-L1 inhibitors as first-line therapy in head and neck squamous cell carcinoma. In contrast to the patients in our study, the KEYNOTE-048 trial⁴  included patients with nonoropharyngeal head and neck squamous cell carcinoma as well as p16⁻ OPSCCs. An important factor in our study design that likely contributed to the lower rate of negative scoring seen in the current series is our exclusive use of resected cases, as opposed to the KEYNOTE-048 study,⁴  which allowed for testing of small biopsies. As we have demonstrated in prior work, the rate of negativity is significantly lower when resected specimens are evaluated as opposed to small biopsies and aspirate cell blocks.¹⁹  Furthermore, a recent reference laboratory series published by Huang et al²⁰  documented a rate of negative CPS scores in head and neck squamous cell carcinoma of 5% across 312 cases. This series was not limited to resected cases and also included small biopsy and cytology material, suggesting that our proportion of negative cases is reflective of that likely to be seen in clinical practice. Of note, 17% of small biopsy and cell block specimens from our prospective series of head and neck squamous cell carcinoma cases received a negative score from the majority of raters.

A potential limitation of our study is that we used the SP263 antibody clone as opposed to the 22C3 clone that was used in the KEYNOTE-048 trial⁴  and approved by the FDA for use as a companion diagnostic test for the selection of patients for treatment with pembrolizumab. When performing a direct comparison of CPS scoring in a series of 43 head and neck squamous cell carcinoma resections and small biopsies, Cerbelli et al²¹  documented excellent agreement between categorical scores obtained using the SP263 and 22C3 antibody clones (κ = 0.878) as well as a similar proportion of negative cases. Although this study and most other observers have found that these 2 assays produce comparable CPS scores in head and neck squamous cell carcinoma,^22,23  some have observed significantly higher rates of positive cases when using the SP263 clone.⁹  As the rate of negative cases in our prospective series of head and neck squamous cell carcinoma cases is comparable to that seen in the KEYNOTE-048 trial⁴  and by Huang et al²⁰  when using the 22C3 antibody clone, it is likely that our distribution of negative, low-positive, and high-positive cases will be similar to that seen in most practices irrespective of the antibody clone being used. Furthermore, as the 22C3 antibody clone and Dako IHC platform are not used by a significant proportion of laboratories performing PD-L1 IHC testing in North America,²⁴  it is likely that a large number of laboratories are using the widely available SP263 clone, which uses the Ventana platform.

A strength of our study is the use of 4 experienced raters for PD-L1 scoring. Based upon the clinically relevant cut points, we documented only fair to substantial (κ = 0.54 and 0.51, respectively) interrater agreement for both the PT and LM in our series of resected p16⁺ OPSCCs. This is significantly lower than the rates of agreement documented by Cerbelli et al²¹  (κ = 0.878) in their series consisting of biopsies, cytology cell blocks, and resections from head and neck squamous cell carcinoma cases. The level of interrater agreement documented in our own practice in the course of routine care in head and neck squamous cell carcinoma cases is similarly significantly higher (κ = 0.84).

Based upon the design of our study, it is not possible to definitively determine the cause of the relatively low levels of interrater agreement seen in our series of resected p16⁺ OPSCCs. In the course of this study and routine clinical practice, we have subjectively observed that CPS scoring in this subset of cases is particularly challenging owing to the dense lymphoid stroma present in both the PTs and LMs. As these tumors frequently elicit minimal desmoplasia and have a large number of tumor-associated lymphocytes, performing an accurate assessment of PD-L1 expression, particularly in the peritumoral lymphocytes, is often difficult.

CPS scoring of multiple blocks from the PT and LM was beyond the scope of our study. We chose to examine the block that contained the greatest amount of tumor from each specimen in order to best mimic clinical practice. It is possible that examination of multiple blocks from each specimen would have resulted in different rates of concordance between specimens than we observed.

We elected to focus on p16⁺ OPSCCs in order to construct a relatively uniform set of cases for analysis. Additionally, as these are tumors that frequently present with LMs, the selection of tumor site for PD-L1 testing is likely to be a common scenario. It is currently unclear what the relationship between p16 immunoreactivity or high-risk human papillomavirus (hrHPV) positivity and PD-L1 immunoreactivity is. Although some studies have demonstrated similar levels of PD-L1 mRNA and protein expression in hrHPV-positive and -negative tumors,^25,26  others have documented higher rates of PD-L1 immunoreactivity in hrHPV-positive tumors.²⁷ 

In the current clinical paradigm, accurate PD-L1 IHC scoring is central in determining therapy in patients with advanced head and neck squamous cell carcinoma.²⁸  In this context, our observations have several implications in regard to the selection of material for PD-L1 IHC testing. First, because 24% of patients with p16⁺ OPSCC will exhibit discordant staining between the PT and metastatic nodal deposits, it may be of value to test multiple sites of disease, if available, although which site of disease correlates best to response remains an open question. Furthermore, in cases where the CPS scores from different sites are discordant, it is unclear if response correlates best with the higher or lower value.

Additionally, in agreement with the recent work of Huang et al,²⁰  we found that negative scores were uncommon in our series of resected p16⁺ OPSCCs, and no specimen received a negative consensus score. For patients with multiple sites of disease, it may be of value to obtain additional material if initial testing is negative and the patient is otherwise a good candidate for pembrolizumab monotherapy. This is particularly the case in patients with only small biopsies available for evaluation, as we have previously shown that the majority of negative cases in this setting are associated with a low-positive result in the paired resection specimens.¹⁹  Finally, we observe that the level of interrater agreement in this series of resected p16⁺ OPSCCs is significantly lower than that documented by our group and others in head and neck squamous cell carcinomas in general. As such, review by multiple observers may be of value in these cases.