We retrospectively included 259 patients with HNSCC diagnosed and treated in the Department of Otorhinolaryngology–Head and Neck Surgery (ORL-HNS), the First Affiliated Hospital of Xi’an Jiaotong University from April 2008 to December 2017. The inclusion criteria were as follows: (1) having undergone curative resection surgery, radiotherapy, or chemoradiotherapy; (2) pathological diagnosis of primary HNSCC; and (3) availability of complete baseline clinicopathological data and follow-up information. Exclusion criteria were as follows: (1) previous diagnosis of other malignancies; (2) history of underlying health conditions about inflammatory, autoimmune, or hematological; and (3) prognostic survival of less than one month; and (4) with insufficient clinical data. The enrolled patients were arbitrarily split into a training set of 181 individuals and a validation set of 78 individuals, with a 9:4 ratio.

The study adhered to the principles outlined in the Declaration of Helsinki and was approved by the Ethics Committee of the First Affiliated Hospital of Xi’an Jiaotong University (Approval No. 2022-321). Informed consent was secured from every participant prior to their inclusion in the research. During the proofreading phase, the manuscript was reviewed and refined using GPT-4.0, a generative AI model by OpenAI, to enhance the clarity and fluency of the text.

The variables collected for this study included both demographic and tumor-related factors: age, sex, smoking index, lymph node status, and tumor stage based on the 8th edition of the American Joint Committee on Cancer (AJCC) Tumor Node Metastasis (TNM) classification system. The smoking index was determined by multiplying the average daily cigarette consumption by the total duration of smoking years of the individual. Tumor staging followed the AJCC’s 8th edition TNM guidelines. Blood samples were collected within seven days before the initiation of surgery, radiotherapy (RT), or chemoradiotherapy (CRT).

Pre-treatment hematological parameters were assessed, including counts of lymphocytes, neutrophils, monocytes, and platelets, as well as levels of albumin (ALB), fibrinogen (FIB), total bilirubin (TBIL), creatinine (Cr), and alanine aminotransferase (ALT). Thirteen biomarkers related to immune function, inflammation, and nutrition were examined: lymphocyte, neutrophil, monocyte, platelet, FIB, ALB, TBIL, Cr, ALT, NLR, PLR, PNI, SII, ALBI score. The biomarkers were determined as follows: PLR = platelets/lymphocytes, NLR = neutrophils/lymphocytes, PNI = albumin (g/L) + 5 × lymphocytes, SII = platelets × neutrophils/lymphocytes, and ALBI = log₁₀ total bilirubin (µmol/L) × 0.66 - albumin (g/L) × 0.085.

The primary outcome, overall survival (OS), was measured as the interval (in months) from the initial confirmed diagnosis to either death or the most recent follow-up, which occurred on December 31, 2022. Recurrence-free survival (RFS), the secondary endpoint, was measured as the period from either curative surgery or initiation of RT/CRT to first relapse last follow-up, whichever occurred first.

All data, including demographic and clinical variables, were collected from our hospital’s electronic medical records. OS and RFS outcomes were determined using multiple standardized protocols to ensure accuracy and consistency.

All statistical analyses were conducted using IBM SPSS (version 26.0) and R software (version 4.4.1), with support from R Studio (version 2024.04.2 + 764). Constant variables were reported as means with standard deviations (SD), while categorical variables were described as frequencies with percentages. To compare clinicopathological features between the training and validation sets, various methods were used. Categorical variables were analyzed using Pearson’s Chi-square test, the Mann-Whitney U test, and Fisher’s exact test, while constant variables were assessed by Student’s t-test and the Mann-Whitney U test. The results of these comparisons are summarized in Table 1 , with statistical significance set at p < 0.05.

To identify independent prognostic biomarkers associated with immune function, nutritional status, and inflammatory response in head and neck squamous cell carcinoma patients, we conducted a comprehensive analysis using the least absolute shrinkage and selection operator (LASSO) Cox regression. This method was applied to 13 preselected variables in the training set, enabling the identification of the most statistically significant and clinically relevant predictors. At an optimal λ value of 0.08, six variables with non-zero coefficients emerged as significant: prognostic nutritional index (PNI), platelet-to-lymphocyte ratio (PLR), systemic immune-inflammation index (SII), albumin-bilirubin index (ALBI), fibrinogen (FIB), and monocyte count.

These six variables were subsequently utilized to construct a nomogram designed to estimate 1-year, 3-year, and 5-year OS for HNSCC patients. The nomogram offers a visual representation of how these biomarkers collectively influence patient outcomes, facilitating individualized prognostic assessments.

Building on this foundation, we developed the systemic immune-inflammation-nutrition (SIIN) score. This score integrates the coefficients derived from the nomogram into a single formula, allowing clinicians to calculate a comprehensive score reflecting the interplay between immune function, inflammatory response, and nutritional status.

The validation of the new indicator was developed. We preliminarily explored the connections between SIIN score and other survival indices using the Mann-Whitney U test and Kruskal-Wallis test. Total points derived from the SIIN score were then incorporated into receiver operating characteristic (ROC) analysis to determine the area under the ROC curve (AUC), which was used to obtain optimal cutoff values and assess the predictive accuracy of SIIN score in both the training and validation sets. The Kaplan-Meier (K-M) curves, along with the log-rank test, and Cox proportional hazards regression models performing univariate and multivariate analyses by stepwise backward selection, were used to test the ability of the SIIN score to stratify the risk of mortality of SIIN score for OS and RFS, where statistical significance was defined as a two-sided p-value of less than 0.05. Calibration curves and decision curve analysis (DCA) were utilized to evaluate the accuracy and clinical utility of the SIIN nomogram.

A total of 279 HNSCC patients who participated in regular examinations and follow-up visits between April 2008 and December 2017 were excluded based on exclusion criteria, and 259 patients were included eventually, whose baseline demographic and clinical characteristics were arbitrarily assigned to either the training set (n = 181) or the validation set (n = 78), as detailed in Table 1 . 238 (91.9%) patients of them were male, and the age of 144 (55.6%) were over 60-year-old, with no significant difference found in baseline clinicopathological parameters between training and validation sets (p > 0.05).

To identify independent prognostic biomarkers, all immune, inflammatory, and nutritional variables were incorporated in a LASSO Cox regression analysis. Using the optimal λ value corresponding to the smallest cross-validated error, this analysis identified six biomarkers - PNI, PLR, SII, ALBI, FIB, and monocyte count - with non-zero coefficients. These findings highlighted their significant association with the prognosis of HNSCC patients undergoing surgery with or without RT/CRT ( Figure 1 ). The markers filtered out were put into nomogram model to construct the formula: Risk score = - PNI × 0.115 + PLR × 0.026 + SII × 0.003 + ALBI × 0.563 + FIB × 1.233 + Monocyte × 10.034 ( Figure 2 ).

We conducted multiple analyses to rigorously validate the prognostic capability of the SIIN score in cancer patients. Initially, we explored its correlation with key clinicopathological variables by constructing boxplots. These visualizations indicated no significant difference in SIIN scores across age and sex groups ( Figures 3A, B ). However, a clear association was observed between more advanced disease stages and higher SIIN scores. Specifically, patients with higher TNM stages and those with lymph node metastasis (N stage), consistently exhibited elevated SIIN scores ( Figures 3C, D ). This suggests that the SIIN score may reflect the degree of systemic immune, inflammatory, and nutritional alterations associated with disease progression. Consequently, the SIIN score could serve as a useful prognostic indicator in evaluating the severity of the disease and guiding treatment strategies for patients with HNSCC.

To further assess the predictive strength of the SIIN score across diverse patient populations, we developed time-dependent ROC curves based on OS and RFS at different intervals. The SIIN score showed robust predictive value in the training set, with AUCs for forecasting 3- and 5-year OS of 0.736 (95% confidence interval [CI]=0.657-0.815) and 0.700 (95% CI = 0.621-0.778) ( Figure 4A ), 3- and 5-year RFS for AUCs of 0.752 (95% CI = 0.678-0.826) and 0.701 (95% CI = 0.624-0.777) ( Figure 4B ), which confirmed that the SIIN score is a reliable predictor of long-term outcomes in this population. When using time-ROC analysis in the validation set, ROC curve analysis demonstrated similar results. The AUCs for 3- and 5-year OS were 0.692 (95% CI: 0.568-0.815) and 0.651 (95% CI: 0.524-0.778) ( Figure 4C ), while the AUCs for RFS were 0.683 (95% CI: 0.557-0.808) and 0.651 (95% CI: 0.526-0.776) ( Figure 4D ). These findings suggest that the SIIN score retains its predictive accuracy in different patient subsets with HNSCC. Subsequently, to optimize clinical utility, patients were categorized into low- and high-risk groups based on a SIIN score cutoff of 4.857, derived from the ROC curve analysis ( Supplementary Figure S1 ). This stratification enabled a clearer delineation of prognosis, allowing for more targeted clinical interventions for those at higher risk.

Next, we compared the SIIN score’s performance with other established prognostic biomarkers. The SIIN score exhibited superior predictive power, with significantly higher AUC values than those of previous comprehensive biomarkers including PNI, NLR, SII, ALBI and PLR for OS (AUC = 0.708) and RFS (AUC = 0.685) in the training set ( Figures 5A, B ), and showed consistent efficiency for OS (AUC = 0.642) and RFS (AUC = 0.599) in the validation set ( Figures 5C, D ). Meanwhile, individual blood markers including creatinine, total bilirubin, albumin, alanine aminotransferase and monocyte were verified to have lower efficiency than SIIN score across the training set (OS: AUC = 0.708; RFS: AUC = 0.685) ( Supplementary Figures S2A, B ) and the validation set (OS: AUC = 0.642; RFS: AUC = 0.599) ( Supplementary Figures S2C, D ). This highlights the SIIN score’s potential as a more precise prognostic tool in clinical practice.

Kaplan-Meier survival analysis provided further validation. Patients with higher SIIN scores had significantly poorer survival outcomes in both the training (OS: p < 0.0001; RFS: p < 0.0001) and validation sets (OS: p < 0.0001; RFS: p < 0.0001) ( Figure 6 ). These results emphasize the SIIN score’s role in effectively stratifying patients based on survival risk, underscoring its value in guiding clinical decisions.

To solidify the SIIN score’s standing as an independent prognostic factor, we performed univariate and multivariate Cox regression analyses. The univariate analysis demonstrated that the SIIN score was significantly associated with OS and RFS in both the training ( Table 2 ; Supplementary Table S1 ) and validation sets ( Table 3 ; Supplementary Table S2 ), with p-values of less than 0.001 for all the endpoints. In the multivariate analysis, we examined clinical and pathological variables not directly linked to the SIIN score, including age, sex, smoking index, AJCC TNM stage, tumor differentiation, tumor type, and treatment modalities (RT/CRT). The multivariate results confirmed that the SIIN score remained an independent prognostic factor for OS in both the training cohort (HR 1.14, 95% CI: 1.09-1.18, p < 0.001) and the validation cohort (HR 1.04, 95% CI: 1.00-1.08, p = 0.026). Similarly, for RFS, the SIIN score continued to be a significant predictor in the training set (HR 1.13, 95% CI: 1.09-1.18, p < 0.001) and the validation set (HR 1.04, 95% CI: 1.00-1.08, p = 0.028).

Calibration curves for both training and validation sets showed strong concordance between predicted and observed outcomes, indicating excellent model calibration ( Figures 7A–F ; Supplementary Figures S3A–F ). Furthermore, decision curve analysis (DCA) was used to assess the clinical utility of the SIIN-based nomogram model at various threshold probabilities ( Figures 7G–L ; Supplementary Figures S3G–L ). The results demonstrated substantial net benefit across a range of threshold probabilities, confirming that the SIIN-based nomogram provides robust discrimination and valuable clinical guidance in predicting patient outcomes.