ABSTRACT
Background
Current staging systems for head and neck cutaneous squamous cell carcinoma (HNCSCC), such as AJCC8 and BWH, often fail to adequately stratify high‐risk patients, particularly due to the exclusion of important biological and host factors. This study aimed to identify independent prognostic factors for HNCSCC outcomes in the largest multi‐institutional retrospective series to date.
Methods
De‐identified data from 8,610 primary HNCSCC tumors, treated with curative intent surgery across multiple international centers, were analyzed. Outcomes included local recurrence (LR), locoregional metastasis (LM), distant metastasis (DM), and disease‐specific death (DSD). Multivariable Fine–Gray subdistribution hazard regression models with robust standard errors clustered at the patient level were fitted to generate exploratory hypotheses regarding independent hazard determinants.
Results
During a median follow‐up of 31.4 months (IQR 14.2–52.8), the cumulative incidence was 3.3% for LR, 2.7% for LM, 0.5% for DM, and 1.2% for DSD. Multivariable analysis confirmed that tumor diameter, poor differentiation, deep invasion, and large‐caliber perineural invasion (LCPNI) were the strongest predictors, being significantly associated with at least three of the four outcomes. Notably, factors currently excluded from or weighted differently in staging systems also proved critical: Immunosuppression was significantly associated with LR (SHR 2.03) and DSD (SHR 1.70); and lymphovascular invasion (LVI) was independently associated with LM (SHR 3.00). Poor differentiation demonstrated high prognostic impact, being a significant predictor for all four endpoints.
Conclusion
Tumor diameter, poor differentiation, deep invasion, LCPNI, immunosuppression status, and LVI are the most critical independent factors for HNCSCC prognosis. Our multi‐institutional data provide a robust framework for hypothesis generation, highlighting that incorporating host biological features and non‐anatomic variables could significantly refine future staging system discussions.
Keywords: cutaneous squamous cell carcinoma, head and neck cancer, metastasis, non‐melanoma skin cancer, oncology, outcomes, recurrence
1. Introduction
The American Joint Committee on Cancer (AJCC) staging system is the widely accepted standard for cancer stratification, playing a critical role in assigning prognosis and guiding treatment selection across oncology. A significant milestone was achieved in the AJCC 8th edition, which introduced a dedicated chapter for nonmelanoma skin cancers of the head and neck. This update incorporated key features such as tumor diameter, deep invasion (invasion beyond fat/depth of invasion [DOI]) and large‐caliber perineural invasion (LCPNI), thereby refining the pathological T‐stage classification [1, 2]. The changes introduced in AJCC8 represented a clear advancement over previous versions, a superiority that has been consistently demonstrated in the literature [3, 4].
Despite its overall improvement, the AJCC8 system still presents challenges, particularly when stratifying risk in advanced disease stages. Our consortium recently published a large‐scale validation study of the AJCC8 for HNCSCC, which demonstrated that, while the model has good distinctiveness for disease‐specific survival (DSS), its performance was limited for other critical outcomes, suggesting the need for further refinement [5].
The Brigham and Women's Hospital (BWH) staging system has demonstrated superior predictive accuracy (c‐index) compared to the AJCC8 in our cohort by utilizing a weighted factor‐grouping strategy [6, 7]. While the clinical impact of such cumulative risk factor burden was previously established within the AJCC8‐T3 [8] stage and more recently extended to BWH‐T2b [9] tumors, individual stratification remains limited. To address this, our group recently introduced the riSCC app, a personalized risk assessment tool designed to optimize traditional framework classifications [10].
The evidence suggests that current staging systems may be insufficient in capturing the full spectrum of risk associated with HNCSCC. Work from our international consortium and other groups has confirmed independent factors that could significantly enhance existing prognostic models, such as lymphovascular invasion (LVI) [11] or patient immunosuppression status [12, 13]. Integrating these high‐impact variables into a more comprehensive risk stratification tool may be the next logical step toward optimizing patient care.
Building upon the robust data derived from our international collaboration, we aimed to analyze a comprehensive panel of clinicopathological and host variables. Rather than attempting to propose a structural replacement or validating incremental classification metrics, the present study is designed as a large‐scale, exploratory, hypothesis‐generating framework. We aimed to evaluate the independent subdistribution hazard ratios (SHR) of these variables within a tightly selected primary cohort to provide robust epidemiological data for future staging adjustments.
2. Methods
2.1. Patient Cohort
Ethical approval was obtained from the Institutional Review Board (IRB) or ethics committees of each participating site. De‐identified data from each center were compiled into a master database. Only tumors that underwent curative‐intent surgical treatment for HNCSCC with either wide local excision (WLE) or peripheral and deep en‐face margin assessment (PDEMA)/Mohs micrographic surgery (MMS) were included. The selection period ranged from January 10, 1991, to December 31, 2023.
A total of 12,099 cases were initially reviewed. The following exclusions were applied: lip CSCC (lip skin or vermilion); cases treated following a previous HN cancer with associated regional disease; patients who received chemotherapy or radiotherapy for other HN neoplasms; cases metastatic at presentation; cases that received neoadjuvant therapy; cases with unknown follow‐up; cases with positive margins; cases receiving any kind of adjuvant therapy; and recurrent primary tumors. Crucially, to eliminate major therapeutic confounding factors from modern paradigms, all cases receiving immunotherapy were excluded. Only 15 patients in the original pool had been exposed to adjuvant immune checkpoint inhibitors, and they were fully omitted. Thus, the manuscript focuses precisely on prognostication in surgically treated, margin‐negative, non‐exposed to immunotherapy treated primary HNCSCC. After exclusions, the dataset included 8,610 cases.
2.2. Data Retrieved
The information collected included patient demographics (age, sex, immune status), anatomical tumor location, histopathological findings (tumor diameter, DOI/Invasion beyond fat, histologic differentiation, LCPNI, and lymphovascular invasion [LVI]), details of surgical and adjuvant treatments, and patient status at the last follow‐up. Outcomes assessed were disease recurrence (local, regional, and distant) and disease‐specific death (DSD). All patients were followed at their respective institutions.
Clinical outcomes of interest included Local recurrence (LR), Locoregional metastasis (LM ‐ defined as in‐transit metastasis and/or nodal metastasis), distant metastasis (DM), and DSD. Although biologically distinct, in‐transit metastases were pooled into this single endpoint (LM) based on literature from independent retrospective cohorts demonstrating that they exhibit highly aggressive clinical behavior, carry a similarly reserved prognosis to advanced nodal disease, and dictate comparable multimodal therapeutic pathways—a clinical behavior also previously characterized by our group [14]. Key event dates—including those for LR, LM, DM, DSD, and the date of last appointment—were retrieved. For standard surgical excisions, margins were verified as negative, whereas for Mohs micrographic surgery (PDEMA), complete peripheral and deep margin evaluation was performed. Deep invasion was operationalized and defined as invasion beyond the subcutaneous fat or, when continuous histopathological measurement was applicable, DOI greater than 6 mm. When specific details were not stated on pathology reports, the following assumptions were made by each institution at the time of data collection, based on dermatopathology reporting conventions: 1. Tumor was considered well differentiated if the pathology did not state poorly or moderately differentiated. 2: Tumor depth was assumed to be confined to the dermis unless otherwise specified in the pathology or operative report. 3: The absence of LCPNI and LVI was assumed if not mentioned in the pathology report. Missing information was distributed as follows: immunosuppression (62 cases), tumor depth (6 cases), and tumor diameter (365 cases). Regarding tumor location, 429 cases were confirmed to have head and neck primary sites, although the specific subsite was not recorded. Additionally, data were missing for differentiation (1,316 cases) and for the caliber of the involved nerve in 216 perineural invasion (PNI) cases.
The unified variables, precise operational definitions, standardized collection protocols, and institution‐specific follow‐up strategies—including clinical and radiological surveillance protocols for recurrence and metastasis ascertainment—are fully detailed in our methodology publication, which serves as the definitive data dictionary for this cohort [15]. Any residual heterogeneity in surveillance frequencies across institutions and historical eras is acknowledged as an inherent trait of this multi‐institutional registry.
2.3. Statistical Analysis
All statistical analyses were performed using R (version 4.4.2; R Foundation for Statistical Computing, Vienna, Austria) and Stata (version 19; StataCorp LLC, College Station, TX). Descriptive characteristics of the cohort were summarized using means and standard deviations (SD) for continuous variables, and absolute counts with percentages for categorical variables.
Time‐to‐event analyses were conducted within a competing‐risks framework. Cumulative incidence functions (CIFs) were estimated for each event of interest: LR, LM, DM and DSD.
To identify prognostic factors, Fine–Gray subdistribution hazard regression models were fitted. First, univariable analyses were performed for each clinicopathological variable. Variables meeting a prespecified significance threshold (p < 0.10) and those considered clinically relevant were included in multivariable models. Adjusted subdistribution hazard ratios (SHR) with 95% confidence intervals (CI) were reported. For visualization purposes, adjusted CIF curves were generated for key predictors, holding the remaining covariates at their reference or mean values.
To satisfy the statistical requirement for non‐independent repeated observations within the same individual, all multivariable Fine–Gray regression models were fitted utilizing robust standard errors clustered at the patient level (using sandwich estimators). All‐cause non‐DSD was explicitly defined as the competing risk for recurrence/metastasis endpoints, and non‐HNCSCC death acted as the competing risk for DSD. Proportional subdistribution hazards assumptions were verified using Schoenfeld‐type residuals. To prevent model overfitting given low event counts for DM (47) and DSD (104), highly parsimonious variable entry was enforced following the 10‐events‐per‐variable convention.
Results are presented as 5‐year cumulative incidence estimates with their corresponding 95% CIs, together with univariable and multivariable SHR and p‐values. Statistical significance was defined as a two‐sided p‐value < 0.05.
3. Results
The clinicopathological characteristics of the cohort are summarized in Table 1. A total of 8,610 excised tumors from 6,398 patients were included. The cohort experienced a median follow‐up of 31.4 months (IQR 14.2–52.8). According to the AJCC8 system, the cohort was stratified as follows: 6905 cases (80.2%) were classified as T1, 654 (7.6%) as T2, 1029 (12.0%) as T3, and 22 (0.3%) as T4a. According to the BWH system, 6,324 cases (73.4%) were T1, 1428 (16.6%) were T2a, 826 (9.6%) were T2b, and 32 (0.4%) were T3. The MMS/PDEMA excision technique was used in 75.4% of tumors (n = 6489), while 24.6% of tumors were treated with standard excision (n = 2121). During follow‐up (31.4 months [IQR 14.2–52.8]), the cumulative incidence was 3.3% for LR, 2.7% for LM, 0.5% for DM, and 1.2% for DSD.
Table 1.
Baseline clinicopathological characteristics, treatment patterns, staging distribution, and event outcomes for the study cohort (n = 8610).
| Descriptive data (n = 8610) | |||||
|---|---|---|---|---|---|
| Clinicopathological characteristics | Treatment | ||||
| Median follow‐up (IQR) | 31.4 (14.2–52.8) | WLE | 2121 (24.6) | ||
| Mean age (SD) | 73.55 (11.49) | MMS/PDEMA | 6489 (75.4) | ||
| Age group | < 70 | 3026 (35.1) | Stratification systems | ||
| ≥ 70 | 5584 (64.9) | AJCC | T1 | 6905 (80.2) | |
| Sex | Male | 6741 (78.3) | T2 | 654 (7.6) | |
| Female | 1869 (21.7) | T3 | 1029 (12.0) | ||
| Immunosuppressed | No | 6795 (78.9) | T4a | 22 (0.3) | |
| Yes | 1815 (21.1) | BWH | T1 | 6324 (73.4) | |
| Invasion beyond fat | No | 7937 (92.2) | T2a | 1428 (16.6) | |
| Yes | 673 (7.8) | T2b | 826 (9.6) | ||
| Location | Other | 5763 (72.2) | T3 | 32 (0.4) | |
| Temple/Ear | 2216 (27.8) | Events | |||
| Mean diameter (SD) | 12.63 (10.54) | LR | No | 8324 (96.7) | |
| Diameter group | < 20 mm | 7011 (81.4) | Yes | 285 (3.3) | |
| ≥ 20 mm | 1599 (18.6) | Mean time to LR (SD) | 16.53 (16.97) | ||
| Mean DOI (SD) | 0.67 (2.33) | LM | No | 8379 (97.3) | |
| DOI group | ≤ 6 mm | 8299 (96.4) | Yes | 231 (2.7) | |
| > 6 mm | 311 (3.6) | Mean time to LM (SD) | 15.77 (16.55) | ||
| Differentiation | Well/Moderate* | 7914 (91.9) | DM | No | 8563 (99.5) |
| Poor | 696 (8.1) | Yes | 47 (0.5) | ||
| LCPNI | No | 8468 (98.4) | Mean time to DM (SD) | 21.86 (18.05) | |
| > 0.1 mm | 142 (1.6) | DSD | No | 8506 (98.8) | |
| LVI | No | 8548 (99.3) | Yes | 104 (1.2) | |
| Yes | 62 (0.7) | Mean time to DSD (SD) | 27.45 (19.61) | ||
Note: Continuous variables are presented as mean (standard deviation). Categorical variables are expressed as number (percentage).
To address reporting heterogeneity across calendar eras, missing/unstated pathological variables were systematically and deterministically reclassified into baseline categories per our pre‐specified protocol, ensuring complete cohort sums (100%) across all clinical and histopathological characteristics.
Abbreviations: AJCC, American Joint Committee on Cancer (8th edition); BWH, Brigham and Women's Hospital classification; DM, distant metastasis; DOI, Depth of Invasion; DSD, disease‐specific death; IQR, Interquartile Range; LCPNI, Large Caliber Perineural Invasion; LM, locoregional metastasis; LR, local recurrence; LVI, Lymphovascular Invasion; MMS, Mohs micrographic surgery; PDEMA, Peripheral and Deep En‐face Margin Assessment; SD, Standard Deviation; W/M, Well/Moderate differentiation; WLE, Wide Local Excision.
The univariable competing‐risks regression findings are presented in Table 2. In the multivariable competing‐risks regression (Table 3), DSD was independently associated with immunosuppression (SHR 1.70; 95% CI 1.11–2.60; p = 0.015), deep invasion (SHR 1.68; 95% CI 1.01–2.81; p = 0.048), large diameter (SHR 3.46; 95% CI 2.15–5.56; p < 0.001), poor differentiation (SHR 3.78; 95% CI 2.34–6.13; p < 0.001), and LCPNI (SHR 2.37; 95% CI 1.09–5.16; p = 0.029) (Figure 1). LM was associated with deep invasion (SHR 1.60; 95% CI 1.15–2.23; p = 0.006), large diameter (SHR 3.74; 95% CI 2.75–5.08; p < 0.001), poor differentiation (SHR 3.51; 95% CI 2.58–4.77; p < 0.001), LCPNI (SHR 2.77; 95% CI 1.66–4.62; p < 0.001), and LVI (SHR 3.00; 95% CI 1.49–6.07; p = 0.002). LR was independently associated with immunosuppression (SHR 2.03; 95% CI 1.58–2.61; p < 0.001), large diameter (SHR 2.09; 95% CI 1.57–2.79; p < 0.001), poor differentiation (SHR 3.11; 95% CI 2.31–4.20; p < 0.001), and LCPNI (SHR 2.16; 95% CI 1.29–3.61; p = 0.003). Invasion beyond fat showed a non‐significant trend toward for LR (SHR 1.41; 95% CI 0.99–2.00; p = 0.058), though it did not cross the formal significance threshold. DM was associated with large diameter (SHR 3.12; 95% CI 1.54–6.31; p = 0.002) and poor differentiation (SHR 4.21; 95% CI 2.20–8.06; p < 0.001).
Table 2.
Five‐year cumulative incidence functions (5y‐CIF) and subdistribution hazard ratios (SHR) estimated using univariate Fine–Gray competing‐risk models for local recurrence (LR), locoregional metastasis (LM), distant metastasis (DM), and disease‐specific death (DSD).
| LR | LM | ||||||
|---|---|---|---|---|---|---|---|
| Univariate model | 5y‐CIF (%) | SHR (95% CI) | p | 5y‐CIF (%) | SHR (95% CI) | p | |
| Age | — | 1.00 (0.99–1.01) | 0.550 | — | 1.01 (1.00–1.02) | 0.130 | |
| Age group | < 70 | 4.3 (3.63–5.03) | — | 3.0 (2.50–3.65) | — | — | |
| ≥ 70 | 4.1 (3.45–4.90) | 0.97 (0.76–1.23) | 0.790 | 3.5 (2.90–4.17) | 1.18 (0.91–1.53) | 0.210 | |
| Sex | Male | 4.3 (3.78–4.93) | — | 3.3 (2.87–3.85) | — | — | |
| Female | 3.8 (2.83–4.94) | 0.84 (0.62–1.14) | 0.280 | 3.0 (2.17–3.91) | 0.90 (0.65–1.25) | 0.540 | |
| Immunosuppressed | No | 3.6 (3.05–4.14) | — | 3.1 (2.68–3.63) | — | — | |
| Yes | 6.4 (5.27–7.71) | 2.03 (1.59–2.59) | < 0.001*** | 3.7 (2.81–4.69) | 1.18 (0.87–1.58) | 0.290 | |
| Invasion beyond fat | No | 3.8 (3.31–4.32) | — | 2.6 (2.24–3.05) | — | — | |
| Yes | 8.6 (6.46–11.04) | 2.48 (1.83–3.37) | < 0.001*** | 10.1 (7.82–12.67) | 4.11 (3.06–5.53) | < 0.001*** | |
| Location | Other | 4.2 (3.57–4.81) | — | 3.4 (2.89–3.96) | — | — | |
| Temple + ear | 4.6 (3.60–5.70) | 1.12 (0.85–1.47) | 0.420 | 3.4 (2.62–4.35) | 1.01 (0.75–1.36) | 0.960 | |
| Diameter | — | 1.03 (1.02–1.04) | < 0.001*** | — | 1.04 (1.03–1.05) | < 0.001*** | |
| Diameter group | < 20 mm | 3.2 (2.71–3.71) | — | 1.8 (1.46–2.18) | — | — | |
| ≥ 20 mm | 8.6 (7.12–10.27) | 2.92 (2.29–3.72) | < 0.001*** | 9.5 (7.99–11.20) | 5.72 (4.40–7.44) | < 0.001*** | |
| DOI | — | 1.08 (1.05–1.11) | < 0.001*** | — | 1.11 (1.08–1.14) | < 0.001*** | |
| DOI group | ≤ 6 mm | 4.0 (3.56–4.57) | — | 2.9 (2.51–3.33) | — | — | |
| > 6 mm | 8.3 (5.35–12.05) | 2.19 (1.41–3.39) | < 0.001*** | 12.3 (8.71–16.43) | 4.71 (3.26–6.81) | < 0.001*** | |
| Differentiation | Well/Moderate | 3.4 (2.97–3.94) | — | 2.4 (2.00–2.76) | — | — | |
| Poor | 12.4 (9.86–15.25) | 4.08 (3.12–5.33) | < 0.001*** | 12.7 (10.13–15.53) | 5.72 (4.34–7.54) | < 0.001*** | |
| PNI | No | 4.0 (3.53–4.52) | — | 3.1 (2.66–3.50) | — | — | |
| > 0.1 mm | 15.1 (9.20–22.25) | 3.99 (2.47–6.45) | < 0.001*** | 14.1 (8.67–20.82) | 5.20 (3.17–8.54) | < 0.001*** | |
| LVI | No | 4.2 (3.68–4.69) | — | 3.1 (2.74–3.58) | — | — | |
| Yes | 10.1 (3.67–20.47) | 2.55 (1.06–6.14) | 0.037* | 18.1 (9.26–29.26) | 6.95 (3.64–13.30) | < 0.001*** | |
| DM | DSD | ||||||
|---|---|---|---|---|---|---|---|
| Univariate model | 5y‐CIF (%) | SHR (95% CI) | p | 5y‐CIF (%) | SHR (95% CI) | P | |
| Age | — | 1.01 (0.98–1.04) | 0.440 | — | 1.01 (1.00–1.03) | 0.090 | |
| Age group | < 70 | 0.7 (0.45–1.09) | — | — | 1.5 (1.12–2.01) | — | — |
| ≥ 70 | 0.7 (0.46–1.07) | 1.11 (0.61–2.00) | 0.730 | 1.9 (1.41–2.48) | 1.18 (0.79–1.75) | 0.420 | |
| Sex | Male | 0.7 (0.51–1.00) | — | — | 1.7 (1.36–2.14) | — | — |
| Female | 0.7 (0.34–1.33) | 0.96 (0.46–1.99) | 0.910 | 1.6 (0.98–2.48) | 0.90 (0.55–1.49) | 0.690 | |
| Immunosuppressed | No | 0.7 (0.47–0.97) | — | — | 1.5 (1.13–1.87) | — | — |
| Yes | 0.8 (0.44–1.40) | 1.24 (0.64–2.40) | 0.530 | 2.5 (1.70–3.40) | 1.70 (1.11–2.58) | 0.014* | |
| Invasion beyond fat | No | 0.6 (0.38–0.78) | — | — | 1.4 (1.07–1.74) | — | — |
| Yes | 2.4 (1.31–3.88) | 4.40 (2.30–8.42) | < 0.001*** | 4.9 (3.30–6.93) | 3.97 (2.54–6.22) | < 0.001*** | |
| Location | Other | 0.8 (0.52–1.08) | — | — | 1.8 (1.42–2.30) | — | — |
| Temple + ear | 0.7 (0.33–1.19) | 0.86 (0.42–1.75) | 0.670 | 1.6 (1.06–2.42) | 0.94 (0.59–1.49) | 0.780 | |
| Diameter | — | 1.04 (1.03–1.05) | < 0.001*** | — | 1.05 (1.04–1.05) | < 0.001*** | |
| Diameter group | < 20 mm | 0.4 (0.26–0.64) | — | — | 1.0 (0.72–1.33) | — | — |
| ≥ 20 mm | 2.0 (1.28–2.87) | 5.16 (2.85–9.35) | < 0.001*** | 4.6 (3.52–5.99) | 5.42 (3.63–8.10) | < 0.001*** | |
| DOI | — | 1.13 (1.09–1.17) | < 0.001*** | — | 1.13 (1.09–1.17) | < 0.001*** | |
| DOI group | ≤ 6 mm | 0.6 (0.40–0.81) | — | — | 1.5 (1.15–1.81) | — | — |
| > 6 mm | 4.1 (2.17–6.99) | 8.21 (4.12–16.35) | < 0.001*** | 7.2 (4.39–10.81) | 5.51 (3.29–9.23) | < 0.001*** | |
| Differentiation | Well/Moderate | 0.5 (0.31–0.68) | — | — | 1.2 (0.91–1.51) | — | — |
| Poor | 3.2 (1.89–4.93) | 6.59 (3.60–12.08) | < 0.001*** | 6.8 (4.79–9.23) | 5.92 (3.91–8.95) | < 0.001*** | |
| LCPNI | No | 0.7 (0.49–0.93) | — | — | 1.6 (1.28–1.95) | — | — |
| > 0.1 mm | 2.4 (0.66–6.36) | 3.91 (1.22–12.60) | 0.020* | 7.0 (3.22–12.69) | 4.79 (2.32–9.89) | < 0.001*** | |
| LVI | No | 0.7 (0.50–0.92) | — | — | 1.7 (1.34–2.03) | — | — |
| Yes | 4.7 (0.79–14.50) | 6.49 (1.57–26.76) | 0.010** | 5.8 (1.51–14.43) | 4.44 (1.38–14.30) | 0.012* | |
Note: The 5y‐CIF is presented with its 95% confidence interval. SHR values correspond to univariate subdistribution hazard models. A dash (–) indicates the reference category or continuous variables without a meaningful CIF estimate. Statistical significance is denoted as follows: p < 0.05 (*), p < 0.01 (**), p < 0.001 (***). Statistically significant values (p < 0.05) are highlighted in bold.
Abbreviations: DOI, Depth of invasion; LCPNI, Large Caliber Perineural Invasion; LVI, Lymphovascular Invasion; W/M, Well/Moderate differentiation.
Table 3.
Multivariable subdistribution hazard ratios (SHR) for local recurrence (LR), locoregional metastasis (LM), distant metastasis (DM), and disease‐specific death (DSD), estimated using Fine–Gray competing‐risk regression.
| Multivariate model | SHR (95% CI) | p | ||
|---|---|---|---|---|
| LR | Immunosuppressed | No | — | |
| Yes | 2.03 (1.58–2.61) | < 0.001*** | ||
| Invasion beyond fat | No | — | ||
| Yes | 1.41 (0.99–2.00) | 0.058° | ||
| Diameter | < 20 mm | — | ||
| ≥ 20 mm | 2.09 (1.57–2.79) | < 0.001*** | ||
| Differentiation | Well/Moderate | — | ||
| Poor | 3.11 (2.31–4.20) | < 0.001*** | ||
| LCPNI | No | — | ||
| Yes | 2.16 (1.29–3.61) | 0.003** | ||
| LVI | No | — | ||
| Yes | 1.49 (0.60–3.67) | 0.390 | ||
| LM | Invasion beyond fat | No | — | — |
| Yes | 1.60 (1.15–2.23) | 0.006** | ||
| Diameter | < 20 mm | — | — | |
| ≥ 20 mm | 3.74 (2.75–5.08) | < 0.001*** | ||
| Differentiation | Well/Moderate | — | — | |
| Poor | 3.51 (2.58–4.77) | < 0.001*** | ||
| LCPNI | No | — | — | |
| Yes | 2.77 (1.66–4.62) | < 0.001*** | ||
| LVI | No | — | — | |
| Yes | 3.00 (1.49–6.07) | 0.002** | ||
| DM | Invasion beyond fat | No | — | — |
| Yes | 1.81 (0.87–3.79) | 0.114 | ||
| Diameter | < 20 mm | — | — | |
| ≥ 20 mm | 3.12 (1.54–6.31) | 0.002** | ||
| Differentiation | Well/Moderate | — | — | |
| Poor | 4.21 (2.20–8.06) | < 0.001*** | ||
| LCPNI | No | — | — | |
| Yes | 2.08 (0.60–7.26) | 0.248 | ||
| LVI | No | — | — | |
| Yes | 2.67 (0.64–11.12) | 0.177 | ||
| DSD | Immunosuppressed | No | — | — |
| Yes | 1.70 (1.11–2.60) | 0.015* | ||
| Invasion beyond fat | No | — | — | |
| Yes | 1.68 (1.01–2.81) | 0.048* | ||
| Diameter | < 20 mm | — | — | |
| ≥ 20 mm | 3.46 (2.15–5.56) | < 0.001*** | ||
| Differentiation | Well/Moderate | — | — | |
| Poor | 3.78 (2.34–6.13) | < 0.001*** | ||
| LCPNI | No | — | — | |
| Yes | 2.37 (1.09–5.16) | 0.029* | ||
| LVI | No | — | — | |
| Yes | 1.99 (0.58–6.89) | 0.277 | ||
Note: A dash (–) indicates the reference category. SHR values are presented with their 95% confidence intervals. Statistical significance is denoted as follows: p < 0.05 (*), p < 0.01 (**), p < 0.001 (***). The symbol (°) indicates a marginal statistical trend (0.05 ≤ p < 0.10). Statistically significant values (p < 0.05) are highlighted in bold.
Abbreviations: LCPNI, Large Caliber Perineural Invasion; LVI, Lymphovascular Invasion.
Figure 1.
Adjusted cumulative incidence functions (CIF) for disease‐specific death according to key prognostic factors, estimated using multivariable Fine–Gray competing‐risk models. Curves represent adjusted predictions for: (a) immunosuppression status; (b) invasion beyond fat; (c) tumor diameter group; (d) histologic differentiation; and (e) perineural invasion (PNI, operating specifically as large‐caliber perineural invasion [LCPNI] in the multivariable model). All curves are adjusted for the remaining covariates included in the model. The y‐axis shows the cumulative incidence (%) and the x‐axis represents time in years.
4. Discussion
The current study represents the largest series to date providing a comprehensive analysis of prognostic factors in head and neck cutaneous squamous cell carcinoma (HNCSCC), thereby offering valuable evidence for clinical practice. Our findings confirm the prognostic significance of established factors. Specifically, tumor diameter and poor differentiation were the most important predictors of poor outcomes, demonstrating the highest subdistribution hazard ratios (SHR) and being significantly associated with all four analyzed endpoints (LR, LM, DM, and DSD). Furthermore, deep invasion and LCPNI were confirmed as major risk factors, significantly affecting three of the four outcomes analyzed. These results strongly support the inclusion of these parameters in current staging models, underscoring their critical role in risk stratification.
Our data also highlights factors currently not included in the standard AJCC and BWH staging systems that could provide necessary refinement, particularly in high‐risk subsets. Patient immunosuppression status proved to be a highly significant risk factor, associated with two major outcomes (LR and DSD), as already observed in other works [12, 13]. Similarly, while LVI was associated only with LM in our series, it has also been associated with DSD in the literature [11], making it a relevant prognostic factor in CSCC.
Differentiation is considered by BWH, but was excluded from AJCC8 based on the potential subjectivity associated with its definition [1]. Nevertheless, several works have identified its impact on CSCC prognosis. Indeed, considering poor differentiation could lead to an upgrade in the current AJCC8 staging [16]. For example, T3‐AJCC8 tumors with poor differentiation exhibit almost a one‐third risk of evolving nodal metastasis [4], and T2a‐BWH cases where poor differentiation is the sole risk factor belong to a higher risk group within T2a‐BWH [16]. However, previous publications from South America suggested that its prognostic value may differ in the Brazilian population compared to North American and European centers [17]. This discrepancy is most likely due to the lack of a universally standardized grading system for CSCC differentiation, a deficiency which indeed prompted the AJCC to exclude it from the AJCC8 staging system [1]. Reliability studies have repeatedly shown suboptimal inter‐ and intra‐rater reliability in grading CSCC differentiation among dermatopathologists and Mohs surgeons [18]. While efforts are underway to improve this standardization, our findings underscore the need for models that can compensate for this lack of uniformity, perhaps by relying on more objective measures like deep invasion and PNI.
The need for refined stratification is particularly important in heterogeneous groups, especially AJCC8‐T3 and BWH‐T2b disease. Previous works have demonstrated that the cumulative burden of risk factors can help refine patient stratification, both for T3‐AJCC8 [8] and, more recently, for BWH‐T2b [9]. This evidence strongly supports a shift toward a weighted factor‐based model. Indeed, patients with three or more risk factors combined exhibit a significantly higher risk of DSD in T3‐AJCC8 [8], and those with three accumulated risk factors show worse outcomes compared to those with two risk factors in T2b‐BWH [9, 19]. Consequently, applying a cumulative risk factor approach would help to better stratify patient staging.
We acknowledge that identifying independent predictors via hazard ratios is methodologically distinct from structural staging updates, which demand formal discrimination and calibration assessments. Our study does not attempt to redefine current algorithms but underscores those biological variables omitted from the AJCC8 framework—specifically histological differentiation, host immunosuppression, and LVI—exhibit strong independent associations with poor outcomes. This multi‐institutional standardized evidence highlights clear paths where future classification consensus committees could focus refinement efforts.
The main limitation of this study is its retrospective design. While our large, multi‐institutional, international cohort provided vastly improved statistical power and generalizability compared to smaller‐scale studies, there is an inherent potential for data heterogeneity between centers. Specifically, the definitions and recording practices for pathological variables such as immunosuppression and LR are not universally standardized and have been shown to be inconsistently recorded [15]. Future prospective studies and the development of consensus guidelines for the pathological reporting of HNCSCC are necessary to mitigate these issues. Certain limitations regarding data completeness must be acknowledged. Information was missing for variables such as immunosuppression status, tumor diameter, differentiation, and specific subsite classification for a subset of head and neck primary tumors. Additionally, the lack of reported nerve caliber in several PNI cases may limit our ability to further stratify risk. While these conservative, non‐directional recategorizations could theoretically skew estimates toward a lower risk profile and represent a limitation inherent to long‐term retrospective series, this conservative approach prevents overestimating risk and ensures a robust sample size for multivariable competing‐risks modeling. Furthermore, by identifying robust, independent factors, this work provides the necessary foundation for the development of future prospective, refined staging systems that move beyond simple anatomic measures to integrate biological and host factors, ultimately improving the accuracy of HNCSCC prognosis.
5. Conclusions
These findings highlight a consistent clinical trend suggesting that non‐anatomic host and biological variables are heavily tied to adverse endpoints. While not entirely novel individually, our standardized results across an unprecedented head‐and‐neck exclusive sample serve as a robust hypothesis‐generating foundation to guide future staging system refinements.
Disclosures of Relationships and Activities
JC: Grants to institution: Sanofi/Regeneron, Castle‐Biosciences; Lectures honoraria: Sanofi, Almirall, Abbvie, Regeneron, Sun Pharma; Payment for expert testimony: Sanofi, Regeneron, Almirall; Support for attending meetings: Pfizer, Almirall, Lilly, Castle‐Biosciences; Advisory board: Almirall, Sanofi, Regeneron, Kyowa, Roche, InflaRx; Funding: support by the Gerencia Regional de Salud de Castilla y León (GRS2549/A/22) and by the Instituto de Salud Carlos III (PI21/01207), co‐financed by European Union. ESR: Serves as a consultant Regeneron Inc., Checkpoint Therapeutic, Feldan Pharmaceuticals, Merck Pharmaceuticals. She serves as a Principal Investigator/Co‐Investigator for the following companies: Regeneron (PI/Co‐I), Merck (Co‐I), Castle Biosciences (PI/Co‐I). She is on the Board of Directors of Skin Cancer Outcomes Consortium (SCOUT); DRC: Investigator for Regeneron, no direct funds; ARM: He has consulted for Phlecs BV, Kyowa, Eli Lilly, Momenta, UCB, and Regeneron in the past, greater than 24 months ago. He has consulted for Incyte, Soligenix, Clarivate, Argenyx, and Bristol Myers Squibb in the past, less than 12 months ago. He consults for Leo Pharma, PPD, Mallinckrodt, Nuvig, Tourmaline Bio, Janssen, Boehringer Ingelheim, and Biocryst currently. He consults for Regeneron and Pfizer currently with payments to the institution. He has grant support from Kyowa, Miragen, Regeneron, Corbus, Pfizer, Incyte, Eli Lilly, Argenx, Palvella, Abbvie, Priovant, Bristol Myers Squibb, Merck in the last 24 months. Beyond 24 months, grant support has come from Sun Pharma, Elorac, Novartis, and Janssen. He has received royalties from Adelphi Values and Clarivate. His current patents include Methods and Materials for assessing and treating cutaneous squamous cell carcinoma (PCT/US2023/078902), Use of Oral Jaki in Lichen Planus‐ PCT/US2024/020149; and Topical Ruxolitinib in Lichen Planus‐ PCT/US2021/053149, 2023‐520085, & 21805700.8, respectively. Methods and Materials for Treating Lichen Planopilaris (Registration Number: 53,103), Machine‐Learning Models for Tumor Grading Using Rank‐Aware Contextual Reasoning on Whole Slide Images (63/616,287). JAC: Receives funding for investigator initiated basic science research from Regeneron and is a PI for a clinical trial sponsored by Regeneron. SAK: Advisory board (paid): Merck, BMS, Regeneron, Galera therapeutics; Advisory board (unpaid): Castle Biosciences; Research support: Castle biosciences, Merck, BMS, Regeneron; Honoraria: UpToDate. ATV: Research Support – Castle; Advisory Board ‐ Inhibitor Therapeutics; Honoraria ‐ Up to Date. AW: Institutional research grant from Castle Biosciences. The remaining have no conflicts to disclose.
6. Synopsis
This multi‐institutional study of 8,610 primary HNCSCC tumors identifies tumor diameter, poor differentiation, deep invasion, large‐caliber perineural invasion, immunosuppression, and lymphovascular invasion as critical independent prognostic factors.
Acknowledgements
This study was funded by a research grant from the American College of Mohs Surgery Foundation. This organization was not involved in the design and conduct of the study, approval of the manuscript, or decision to submit the manuscript for publication.
Data Availability Statement
The authors have nothing to report.
References
- 1. Amin M. B., Greene F. L., Edge S. B., et al., “AJCC Cancer Staging Manual.” International Publishing: American Joint Commission on Cancer (New York, NY: Springer, 2017). 8th ed.. [Google Scholar]
- 2. Zanoni D. K., Patel S. G., and Shah J. P., “Changes in the 8th Edition of the American Joint Committee on Cancer (AJCC) Staging of Head and Neck Cancer: Rationale and Implications,” Current Oncology Reports 21, no. 6 (2019): 52, 10.1007/s11912-019-0799-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Karia P. S., Morgan F. C., Califano J. A., and Schmults C. D., “Comparison of Tumor Classifications for Cutaneous Squamous Cell Carcinoma of the Head and Neck in the 7th Vs 8th Edition of the AJCC Cancer Staging Manual ,” JAMA Dermatology 154, no. 2 (2018): 175–181, 10.1001/jamadermatol.2017.3960. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Cañueto J., Burguillo J., Moyano‐Bueno D., et al., “Comparing the Eighth and the Seventh Editions of the American Joint Committee on Cancer Staging System and the Brigham and Women's Hospital Alternative Staging System for Cutaneous Squamous Cell Carcinoma: Implications for Clinical Practice,” Journal of the American Academy of Dermatology 80, no. 1 (2019): 106–113.e2, 10.1016/j.jaad.2018.06.060. [DOI] [PubMed] [Google Scholar]
- 5. Girardi F. M., Wagner V. P., Machado C. D. C., et al., “Validation of Current Staging Systems in HNCSCC: A Multinational Cohort Study,” Head & Neck 47, no. 12 (2025): 3385–3393, 10.1002/hed.28259. [DOI] [PubMed] [Google Scholar]
- 6. Jambusaria‐Pahlajani A., Kanetsky P. A., Karia P. S., et al., “Evaluation of AJCC Tumor Staging for Cutaneous Squamous Cell Carcinoma and a Proposed Alternative Tumor Staging System,” JAMA Dermatology 149, no. 4 (2013): 402–410, 10.1001/jamadermatol.2013.2456. [DOI] [PubMed] [Google Scholar]
- 7. Ruiz E. S., Karia P. S., Besaw R., and Schmults C. D., “Performance of the American Joint Committee on Cancer Staging Manual, 8th Edition vs the Brigham and Women's Hospital Tumor Classification System for Cutaneous Squamous Cell Carcinoma,” JAMA Dermatology 155, no. 7 (2019): 819–825, 10.1001/jamadermatol.2019.0032. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Conde‐Ferreirós A., Corchete L. A., Puebla‐Tornero L., et al., “Definition of Prognostic Subgroups in the T3 Stage of the Eighth Edition of the American Joint Committee on Cancer Staging System for Cutaneous Squamous Cell Carcinoma: Tentative T3 Stage Subclassification,” Journal of the American Academy of Dermatology 85, no. 5 (2021): 1168–1177, 10.1016/j.jaad.2020.03.088. [DOI] [PubMed] [Google Scholar]
- 9. Ran N. A., Granger E. E., Brodland D. G., et al., “Risk Factor Number and Recurrence, Metastasis, and Disease‐Related Death in Cutaneous Squamous Cell Carcinoma,” JAMA Dermatology 161, no. 6 (2025): 597–604, 10.1001/jamadermatol.2025.0128. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Jambusaria‐Pahlajani A., Jeanselme V., Wang D. M., et al., “riSCC: A Personalized Risk Model for the Development of Poor Outcomes in Cutaneous Squamous Cell Carcinoma,” Journal of the American Academy of Dermatology 93, no. 1 (2025): 73–81, 10.1016/j.jaad.2025.02.076. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Hirotsu K. E., Aasi S. Z., Samson K. K., et al., “Lymphovascular Invasion Is An Independent Predictor of Metastasis and Disease‐Specific Death in Cutaneous Squamous Cell Carcinoma: A Multicenter Retrospective Study,” Journal of the American Academy of Dermatology 93, no. 2 (2025): 368–377, 10.1016/j.jaad.2025.04.029. [DOI] [PubMed] [Google Scholar]
- 12. Klein J. C., Shahwan K. T., Petric U. B., et al., “Impact of Immunosuppression on Cutaneous Squamous Cell Carcinoma Outcomes,” Journal of the American Academy of Dermatology 94, no. 1 (2026): 151–160, 10.1016/j.jaad.2025.09.042. [DOI] [PubMed] [Google Scholar]
- 13. Cardona‐Machado C., Martín‐Vallejo J., Becerril‐Andrés S., et al., “Immunosuppression Is Associated With An Increased Risk of Distant Metastases in High‐Risk Cutaneous Squamous Cell Carcinoma: A Retrospective Cohort Study,” JAAD International 15 (2024): 74–77, 10.1016/j.jdin.2023.11.006. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Pahalyants V., Jairath N. K., Maas D. E., et al., “Satellitosis or In‐Transit Metastasis in Cutaneous Squamous Cell Carcinoma: Risk Factors and the Prognostic Significance,” Journal of the American Academy of Dermatology 93, no. 6 (2025): 1432–1440, 10.1016/j.jaad.2025.07.029. [DOI] [PubMed] [Google Scholar]
- 15. Hallak D., Ruiz E., Ran N., et al., “Protocol for Establishing a Large, International, Multicenter Cutaneous Squamous Cell Carcinoma Database,” Archives of Dermatological Research 317 (2025): 829, 10.1007/s00403-025-04330-8. [DOI] [Google Scholar]
- 16. Gupta N., Mulvaney P. M., Murad F., et al., “Improved Homogeneity and Monotonicity of American Joint Committee on Cancer Staging Manual, 8th Edition on Cutaneous Squamous Cell Carcinoma With Addition of Poor Differentiation,” Journal of the American Academy of Dermatology 90, no. 2 (2024): 423–425, 10.1016/j.jaad.2023.10.009. [DOI] [PubMed] [Google Scholar]
- 17. Girardi F. M., Wagner V. P., Martins M. D., Abentroth A. L., and Hauth L. A., “Impact of Ajcc 8 Pt Staging in Cutaneous Head and Neck Squamous Cell Carcinoma in a Nonselected Real‐World Patient Sample,” Head & Neck 45, no. 2 (2023): 337–346, 10.1002/hed.27233. [DOI] [PubMed] [Google Scholar]
- 18. Diede C., Walker T., Carr D. R., and Shahwan K. T., “Grading Differentiation in Cutaneous Squamous Cell Carcinoma: A Review of the Literature,” Archives of Dermatological Research 316, no. 7 (2024): 434, 10.1007/s00403-024-03184-w. [DOI] [PubMed] [Google Scholar]
- 19. Cardona‐Machado C., Puebla‐Tornero L., Martín‐Vallejo J., Ruiz‐Villanueva A., Cardeñoso‐Álvarez E., and Cañueto J., “Refined Prognostic Stratification of Cutaneous Squamous Cell Carcinoma Using Top Eight Risk Factors: A Retrospective Cohort Study,” Journal of the European Academy of Dermatology and Venereology 39, no. 7 (2025): e572–e576, 10.1111/jdv.20398. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The authors have nothing to report.