We reviewed a total of 89 patients with head and neck cancer who were treated with ICIs between November 2015 and December 2020 in our institution. Among them, 52 patients met the following inclusion criteria: (1) had squamous cell carcinoma; (2) had FDG PET/CT before initiation of immunotherapy; (3) did not undergo surgery or other conventional chemotherapy between the PET/CT and ICI treatment; (4) were without double primary malignancy. The following exclusion criteria were applied to these 52 patients: (1) patients with a time interval of more than 60 days between the date of PET/CT and the date of immunotherapy; (2) those without measurable lesions to be evaluated on PET/CT; (3) those with short-term follow-up loss within 1 month (not due to death). Thirty-nine patients were finally enrolled. This retrospective study was approved by the institutional review board (IRB File No. SMC 2021-02-013) and the need for informed consent was waived.

Clinical variables included age, sex, stage, and type of immunotherapy (single regimen or combined regimen). Stage was divided into recurrence group with initial stages 1, 2 or 3, and advanced group with initial stage 4. Furthermore, we obtained serum absolute neutrophil count (ANC) and leukocyte count within 2 weeks from the date of PET/CT. The derived neutrophil-to-lymphocyte ratio (dNLR) was calculated as follows: dNLR = neutrophils/(leukocytes minus neutrophils) (12). The dNLR values were dichotomized by the optimal cut-off obtained from receiver operating characteristic (ROC) curves for progression and death.

Patients underwent neck CT (or MRI) and chest CT scans every 3–6 months to evaluate the therapy response. In some patients, abdominal CT or brain MRI or ¹⁸F-FDG PET/CT was added depending on the tumor location or new symptoms. Progression was determined by an increment of the lesion size or an appearance of the new lesion, which was confirmed by serial follow-up imaging studies. Progression-free survival (PFS) was defined as the period from the start of immunotherapy to the date of the imaging in which the clue finding was first identified. In the absence of progression, PFS was defined as the period to the date of the last clinical visit. Overall survival (OS) was defined as the period from the start of immunotherapy to the date of death or to the date of the last clinical visit without death. All clinical information was obtained through medical record reviews and imaging reviews.

Positron emission tomography/computed tomography images were mostly obtained (27/39 patients, 69%) from a GE STE PET/CT scanner (Milwaukee, WI, USA). Nine cases were obtained from a GE Discovery MIDR PET/CT scanner and three cases from a GE Discovery LS PET/CT scanner. Patients were instructed to fast for at least 6 h and were injected with 5 MBq/kg of ¹⁸F-FDG. Blood sugar levels of all patients were below 200 mg/dl. After 60 min, a CT scan was performed (STE, 16-slice, 140 keV, 30–170 mA; MIDR, 128-slice, 120 keV, 30–100 mA; LS, 8-slice, 140 keV, 40–120 mA), and a PET scan from the skull base to the thigh was subsequently obtained (STE, 2.5 mm/frame, 3D mode; MIDR, 2 mm/frame, 3D mode; LS, 4 mm/frame, 2D mode). Image reconstruction was performed using an ordered-subsets expectation maximization (OSEM) algorithm for the STE scanner and the LS scanner (2 iterations, 20 subset, matrix size 128 × 128, voxel size 3.9 × 3.9 × 3.3 mm; 2 iterations, 28 subset, matrix size 128 × 128, voxel size 4.3 × 4.3 × 3.9 mm, respectively). OSEM with time-of-flight and point-spread-function was used for reconstruction in the MIDR scanner (4 iterations, 18 subset, matrix size 192 × 192, voxel size 2.6 × 2.6 × 3.3 mm).

One to five tumor lesions were evaluated in each patient according to PERCIST 1.0 criteria (up to two lesions per organ, Figure 1) (13). On a dedicated GE workstation, we set the volume of interest (VOI) of each lesion, which was carefully drawn not to include brain parenchyma. Maximum standardized uptake value (SUV_max), peak standardized uptake value (SUV_peak), metabolic tumor volume (MTV), and total lesion glycolysis (TLG) were measured for each lesion. SUV is defined as follows: radioactivity ingested per gram of tissue/radioactivity injected per kilogram of body weight. SUV_max represents the highest single-pixel SUV value within the VOI, while SUV_peak is an average SUV within a small area containing the hottest uptake and the around. The highest SUV_max and SUV_peak among all lesions of each patient were selected as representative values for the patient. MTV represents the metabolically active volume by summing the voxels above a threshold. Four thresholds consisting of SUV2.5, 30% of SUV_max, 40% of SUV_max, and 50% of SUV_max were used for automatically contouring the VOI. TLG is calculated by multiplying MTV and SUV_mean. Total MTV (TMTV) and total TLG (TTLG) were calculated as the sums of the MTV and TLG of all lesions of each patient. Liver SUV_max and spleen SUV_max were additionally obtained by setting 3-cm VOIs on the right hepatic lobe and spleen (14). Metastatic lesions were not included within the VOI. Spleen-to-liver ratio (SLR) was calculated by dividing spleen SUV_max by liver SUV_max (14). The aforementioned PET parameters were dichotomized by optimal cut-off values obtained from ROC curves for progression and death. Combined parameters (MTV + dNLR) were classified into three groups as follows: Group 1, low TMTV and low dNLR; Group 2, low TMTV and high dNLR OR high TMTV and low dNLR; Group 3, high TMTV and high dNLR. Imaging parameters were assessed by two experienced nuclear medicine physicians (SH Moon and HR Kwon).

Subgroup analysis was performed as a preliminary study on 18 patients with pre-existing PD-L1 (22C3, Dako, Santa Clara, CA, USA) staining results at the primary sites or metastatic lesions (15). Details of the immunohistochemical stains are provided in Supplementary Table 1. The proportion of cells stained with PD-L1 (%) was expressed as a tumor proportion score (TPS) and a combined positive score (CPS). TPS represents the ratio of PD-L1-positive tumor cells among all viable tumor cells, while CPS represents the ratio of PD-L1-positive any cells (including tumor cells, lymphocytes, and macrophages) among all viable tumor cells (15).

Another preliminary analysis was performed on 12 patients with tissues capable of CD8 (SP57, Ventana, Oro Valley, AZ, USA) and granzyme B (11F1, Novocastra, Buffalo Grove, IL, USA) staining. Details of the immunohistochemical stains are provided in Supplementary Table 1. After CD8 and granzyme B staining was performed on formalin-fixed and paraffin-embedded tissue specimens, the ratio (%) of the number of stain-positive cells to the total number of cells was investigated using QuPath software.

The aforementioned procedures were performed by an experienced pathology physician (J Cho). SUV_max, SUV_peak, MTV, and TLG of the target lesions were measured to evaluate the association with IHC biomarkers. PET/CT images for this analysis were taken within 1 month from the date of tissue collection.

Survival curves were evaluated for PFS and OS by Kaplan-Meier method and log-rank test. Univariable and multivariable analyses for prognosis were performed using Cox proportional hazard method. Binary PET parameters, combined parameters, and clinical variables were included. Multivariable analysis with enter mode was performed using significant variables of univariable analysis. Multivariable analysis was performed via different models for TMTV, TTLG, and combined parameters to avoid multicollinearity. Spearman's correlation was used to evaluate association between immunostaining indicators and PET parameters. IBM SPSS Statistics software (version 27.0) was used, and a p-value < 0.05 was considered statistically significant.

The characteristics of the subjects are presented in Table 1. Thirty-nine patients with a mean age of 60.7 years and with male predominance (84.6%) were enrolled. Primary tumor sites included nasopharynx (n = 6), oropharynx (n = 1), hypopharynx (n = 11), maxillary sinus (n = 2), nasal cavity (n = 3), tonsil (n = 6), tongue (n = 8), and oral mucosa (n = 2). Sixteen patients received a single regimen of immunotherapy and 23 patients received a combined regimen. Pembrolizumab (PD-1 inhibitor, 33%, 13/39), nivolumab (PD-1 inhibitor, 36%, 14/39), durvalumab (PD-L1 inhibitor, 28%, 11/39), and avelumab (PD-L1 inhibitor, 3%, 1/39) were used for ICIs. Combined therapies used with ICI included conventional chemotherapy (such as gemcitabine, 5-fluorouracil, and cisplatin), radiotherapy, and proton therapy. Twenty-seven patients (69%) had recurrence and the mean PFS was 238.9 days. These patients were determined to be progressive disease by imaging studies including neck CT (13 cases), neck CT+chest CT (two cases), neck CT+chest CT+abdominal CT (one case), chest CT (six cases), abdominal CT (one case), neck MRI (one case), brain MRI (two cases), and ¹⁸F-FDG PET/CT (one case). The recurrence sites included neck (13 cases), lung (six cases), neck + lung (two cases), liver (one case), brain (two cases), and bone (three cases). Eleven patients (28%) died and the mean OS was 384.1 days.

The cut-off values of PET parameters by each threshold and dNLR were obtained and the results are presented in Supplementary Table 2. Parameters with 40% of SUV_max showed the best performance in the analysis, so we decided to describe the findings focusing on this threshold (data with SUV2.5, 30 and 50% thresholds are not shown).

In univariable analysis for PFS using binary PET parameters and clinical variables (Table 2), TMTV (<10.01), TTLG (<138.00), age, dNLR (<2.02), and Group 1 of combined parameters (low TMTV + low dNLR) were significant favorable predictors (p = 0.018, p = 0.027, p = 0.005, p = 0.022, and p = 0.021, respectively). In multivariable analysis with these variables (Table 3), only the combined parameter and age showed statistical significance. Group 3 had significantly worse PFS than Group 1 [hazard ratio (HR) = 8.91 and p = 0.036]. However, Group 1 and Group 2 had no significant difference in progression risk (p = 0.116). The older age had a lower risk of progression than the younger age (HR = 0.98 and p = 0.048). In Kaplan-Meier curves for PFS (Figure 2), TMTV, TTLG, dNLR, and combined parameters were statistically significant (p = 0.011, p = 0.022, p = 0.016, and p = 0.008, respectively).

Spleen-to-liver ratio was the only significant prognostic factor in univariable analysis for OS (Table 4). Patients with high SLRs had a higher risk of death (HR = 4.39 and p = 0.019). Similarly, only SLR was statistically significant in survival curve analysis for OS among all PET and clinical parameters (p = 0.011, Figure 2).

To increase the homogeneity of the cohort, the same analysis as above was performed using a group consisting only of the STE PET scanner, which accounts for the majority of study subjects (69.2%, 27/39; Supplementary Table 3). In univariable analysis, TTLG, dNLR, and the combined parameter were significant predictors for PFS (TTLG, HR = 2.80, p = 0.037; dNLR, HR = 5.61, p = 0.023; TMTV+dNLR, HR = 8.17, p = 0.044). Younger age or higher TMTV was more likely to have a progression, but they were not statistically significant (p = 0.058 and p = 0.062). In multivariable analysis for PFS, Group 3 of the combined parameters was tend to have a higher progression risk than Group 1, but was not statistically significant (p = 0.064). There were no statistically significant factors in univariable analysis for OS.

We additionally performed individual analyses for the single regimen group and the combined regimen group with respect to immunotherapy since the treatment type could affect the disease course (Supplementary Tables 4, 5). For single regimen group (n = 16), age and the combined parameter were statistically significant in univariable analysis for PFS (HR = 0.94, p = 0.026; HR = 10.72, p = 0.041, respectively). No significant predictors were found in multivariable analysis for PFS and univariable analysis for OS. For combined regimen group (n = 23), TMTV and TTLG were prognostic factors for PFS in univariable analysis (HR = 4.16, p = 0.028; HR = 3.24, p = 0.044, respectively). There were no independent predictive factors in multivariable analysis for PFS and univariable analysis for OS. SLR showed a marginal significance for OS (p = 0.051).

Spearman's correlation analysis between PD-L1 and PET parameters is presented in Table 5. TPS (%) and MTV showed a moderate negative correlation at all thresholds (MTV2.5, γ = −0.494, p = 0.037; MTV30%, γ = −0.619, p = 0.006; MTV40%, γ = −0.554, p = 0.017; MTV50%, γ = −0.627, p = 0.005). CPS (%) and MTV also showed a moderate negative correlation at most of the thresholds (MTV30%, γ = −0.558 p = 0.016; MTV40%, γ = −0.487, p = 0.040; MTV50%, γ = −0.570, p = 0.013). The correlation between TPS and TLG showed a statistical significance only in the threshold of 50% (p = 0.045). There was no significant correlation between CPS and TLG. SUVs showed no significant relationships with TPS or CPS.

Regarding the number and activity of cytotoxic T cells, both CD8 and granzyme B biomarkers were not related to metabolic parameters. The p-values of SUV_max, SUV_peak, MTV40%, and TLG40% for association with CD8 positivity (%) were p = 0.557, p = 0.681, p = 0.542, and p = 0.471, respectively. For association with granzyme B positivity (%), the p-values were p = 0.395, p = 0.415, p = 0.913, and p = 0.595, respectively.