A total of 231 patients who were diagnosed with HNSCC in the Department of Otorhinolaryngology Head and Neck, First Affiliated Hospital of Xi’an Jiaotong University from January 2008 to January 2018, were enrolled retrospectively for development cohort to construct nomograms. The inclusion criteria were as follows: (1)patients who underwent radical tumor resection surgery and had been pathologically diagnosed as HNSCC; (2) HNSCC had to be primary; and (3) patients with complete clinicopathological and follow-up records. The exclusion criteria were as follows: (1) patients with other malignant diseases previously diagnosed; (2) patients with a history of inflammatory, autoimmune disease, and hematological disease; and (3) patients who had a postoperative survival time <1 month.

In addition, 50 HNSCC patients who only received radiotherapy or chemoradiotherapy after diagnosis between March 2008 and December 2017 in the same hospital were supplemented as a RT/CRT cohort to test the general utility of constructed nomograms based on PIV. Patients who underwent surgery died within 1 month after therapy were excluded. Other inclusion and exclusion criteria were the same with the development cohort.

All data collection was conducted in accordance with the Declaration of Helsinki and approved by the ethics committee of First Affiliated Hospital of Xi’an Jiaotong University (No. 2022–321). All participants signed an informed consent.

Smoking index was calculated as the average number of cigarettes smoked per day multiplied by the number of years smoking (28). The TNM stage was in compliance with the 8th AJCC TNM stage system (8). Blood samples were collected and analyzed within 7 days before surgery or the start date of RT/CRT. We defined normal level of fibrinogen (FIB), albumin (ALB), and total bilirubin levels (TBIL) as 2–4 g/L, 40–55 g/L, and 3.4–17.1 µmol/L, respectively. Immune-inflammatory indices were calculated as NLR = neutrophil count (10⁹/L)/lymphocyte count (10⁹/L) (13); PLR = platelet count (10⁹/L)/lymphocyte count (10⁹/L) (14); LMR = lymphocyte count (10⁹/L)/monocyte count (10⁹/L) (15); and PIV was calculated as previously described: [neutrophil count (10⁹/L) × platelet count (10⁹/L) × monocyte count (10⁹/L)]/lymphocyte count (10⁹/L) (18–22).

For the sake of collecting clinical outcomes information, periodical follow-up evaluations were conducted for all patients until death. After surgery or chemoradiotherapy, patients were rechecked for recurrence by physical examination, blood test, and fiberoptic pharyngorhinoscopy every 2 months in the first 2 years, every 6 months for 3–5 years, and thereafter once a year. CT scans of the neck and chest were performed every 6 months for 3 years and once a year thereafter. Hypopharyngeal cancer patients were additionally monitored by esophagogastroduodenoscopy once a year. Biopsy or fine needle aspiration cytology was carried out if there was suspicious of local or regional recurrences.

The primary endpoint for development cohort was disease-free survival (DFS), which was defined as the interval time (in months) between the radical surgery and the date of death or recurrence, or final follow-up (on 1 January 2023), whichever came first. Progression-free survival (PFS), the primary endpoint for the RT/CRT cohort, was measured in months from the start date of RT/CRT to the date of disease progression or death from any cause or the censoring date of last follow-up (on 30 September 2023). The secondary endpoint, overall survival (OS), was calculated as the time (in months) from the first definite diagnosis to the date of death or final follow-up, whichever came first.

All data were statistically analyzed using IBM SPSS 26.0 and R software (version 4.2.2) with the assistance of R studio (version 2022.07.2 + 576). Restricted cubic spline (RCS) regression was performed to evaluate the association between DFS and PIV as a continuous variable. The cutoff point was determined using maximally selected rank statistics. The differences in the clinicopathological characteristics between the low and high PIV groups were tested by one-way analysis of variance (ANOVA; normal distribution), Wilcoxon rank sum test (skewed distribution), or Chi-square test (categorical variables). Spearman correlation analysis was used to test correlations.

We performed the Kaplan–Meier analysis and log-rank test to plot the survival curves. Subgroup analysis and interaction terms were used to confirm whether there were any correlations between PIV and the various clinical parameters. Univariate and multivariate Cox regression using stepwise backward LR method and the least absolute shrinkage and selection operator (LASSO) regression model were used to search for independent prognostic factors of convincing nomograms in HNSCC patients. The LASSO coefficient profiles of the 19 variables for development cohort were constructed from the log (λ) sequence, which shrank exactly to zero to be used for variable selection. A string of λ values covering the entire range, which were determined by 1,000-fold bootstrapping resampling, was elected to calculate the cross-validation error and search out prognostic factors of the nomogram. The concordance index (C-index) based on Harrell C statistics and receiver operating characteristic (ROC) curve with AUC were used to measure the accuracy of the nomograms. The relationship between the predicted and actual risks for outcomes of the nomograms was graphically displayed via calibration plots, while decision curve analysis (DCA) was used to evaluate the clinical benefits and utility of the nomogram for predicting prognosis. Trend tests were performed by modeling PIV as a continuous variable, dividing it by cutoff point or into quartiles; Wald tests were used to assess statistical significance. A p-value <0.05 from the two-sided test was considered statistically significant.

A total of 231 HNSCC patients after radical surgery were screened for eligibility, 161 of whom [151 (93.8%) men and 10 (6.2%) women] were eventually enrolled in development cohort. In these patients, 91 (56.5%) were over 60 years at diagnosis and 105 (65.2%) had a smoking index of <650. The primary tumors of the vast majority were located in the larynx [137 (85.1%)] and nearly half [74 (46%)] were well differentiated. Additionally, the distribution of the TNM stage at diagnosis was 0/I, 43.5%; II, 14.9%; III, 17.4%; and IV, 24.2%, respectively. It is worth noting that, compared to those with stage 0/I, HNSCC patients with higher TNM stages harbored significantly higher PIV and log PIV (0/I vs. II, p<0.001; 0/I vs. III, p<0.01; 0/I vs. IV, p<0.001) ( Supplementary Figure 1 ). Besides radical tumor resection, 56 (34.8%) patients received PORT or POCRT and others received radical surgery only. The proportion of patients with normal levels of FIB, ALB, and TBIL were 80.7%, 50.9%, and 86.3%, respectively. The median values of the NLR, PLR, and LMR for all patients were 1.94, 97.26, and 4.43, respectively. Details are given in Table 1 .

When analyzed as a continuous variable, the restricted cubic spline plot showed that PIV had a positive dose–response association with the mortality and recurrence risk in development cohort, and the risk increased rapidly when PIV value exceeded 123.3, suggesting its potential to conjecture patients’ survival ( Supplementary Figure 2A ). Considering the continuity of PIV and its association with DFS, we performed a bilateral outcome-oriented maximally selected rank statistics test, indicating that the optimal cutoff value for PIV associated with DFS was 123.3 ( Supplementary Figure 2B ). All patients in the development cohort were then stratified into two groups based on the cutoff of PIV.

Ultimately, 76 patients in PIV-Low group and 85 patients in PIV-High group were analyzed. As shown in Table 1 , higher PIV was significantly associated with higher T stage, more advanced TNM stage, receiving PORT/POCRT, higher NLR, higher PLR, and lower LMR (p<0.001, p<0.001, p=0.033, p<0.001, p<0.001, and p<0.001, respectively), while no statistically relevant correlations were noted in other characteristics. In addition, Spearman rank correlation test revealed that FIB level had a strong positive association with PIV when grouped by age (<60 years: R=0.321, p=0.007; ≥60 years: R=0.466, p<0.001), however, other clinical parameters did not exhibit this feature ( Supplementary Figure 3 ).

At data cutoff of the development cohort, the median follow-up period was 60 months (range, 3–143 months), during which 79 patients died and 18 patients experienced tumor recurrence, and the median DFS and OS were 59 months and 60 months, respectively. The DFS rates at 1 year, 3 years, and 5 years were 88.2%, 60.9%, and 49.1% as compared with 90.7%, 60.9%, and 50.3% of the OS rates, respectively. In terms of Kaplan–Meier curves and log-rank test results, the 5-year DFS rate of PIV-Low group was statistically higher than that of PIV-High group (73.7% vs. 23.5%, HR: 5.00, 95% CI: 3.23–7.71, p<0.0001, Figure 1 ), which was similar to that of the 5-year OS rate (76.3% vs. 27.1%, HR: 5.29, 95% CI: 3.38–8.28, p<0.0001, Supplementary Figure 4 ). All these results indicated that PIV-High group had a worse prognosis compared with PIV-Low group.

To further investigate the survival characteristics in specific TNM stages, we divided the development cohort into four groups and found that higher PIV was still correlated with poorer disease-free survival in stage 0/I (93.6% vs. 34.8%, HR: 14.98, 95% CI: 5.16–43.53, p<0.0001) and III (50.0% vs. 16.7%, HR: 3.22, 95% CI: 1.34–7.73, p=0.013, Supplementary Figures 5A, C ) patients. Despite that similar trends were observed in stage II and IV patients, the results did not reach statistical significance (stage II: 71.4% vs. 29.4%, HR: 3.59, 95% CI: 1.22–10.60, p=0.069; stage IV: 16.7% vs. 14.8%, HR: 1.86, 95% CI: 0.94–3.69, p=0.083, Supplementary Figures 5B, D ). As shown in Supplementary Figure 6 , the result was analogously applicable to OS in stage 0/I (95.7% vs. 43.5%, HR: 18.47, 95% CI: 5.83–58.54, p<0.0001), III (50.0% vs. 16.7%, HR: 3.22, 95% CI: 1.34–7.73, p=0.013), and IV (33.3% vs. 14.8%, HR: 2.20, 95% CI: 1.08–4.47, p=0.041) patients, while it was a pity that the result was not statistically significant in stage II patients (57.1% vs. 35.3%, HR: 2.49, 95% CI: 0.86–7.25, p=0.140).

Further stratified analyses were conducted, disclosing the associations between PIV and HR for DFS and OS in various subgroups according to sex, age, smoking index, T stage, N stage, TNM stage, PORT/POCRT, FIB, ALB, and TBIL ( Supplementary Tables 1 , 2 ). Interaction analyses indicated that N stage (p=0.004), TNM stage (p=0.035), and PORT/POCRT (p=0.006) were able to interact with PIV for DFS, while no interactions between PIV and other clinicopathological characteristics were observed ( Supplementary Table 1 ). Specifically, stage N1, stage 0/I, and receiving no PORT/POCRT were associated with higher risk for death and recurrence. The significant interactions in subgroups defined by N stage (p=0.031) and PORT/POCRT (p=0.024) were also detected for OS ( Supplementary Table 2 ).

Univariate and multivariate Cox regression analyses were first performed to investigate potential prognostic factors for DFS and OS in development cohort ( Table 2 ; Supplementary Table 3 ). According to univariate analysis, the following variables were proved statistically affecting both DFS and OS: age, smoking index, T stage, N stage, TNM stage, PORT/POCRT, FIB, ALB, NLR, PLR, LMR, and PIV. The multivariate analysis that included all factors whose p <0.1 in the univariate analysis indicated that higher age (DFS: p=0.039; OS: p=0.010), higher TNM stage (DFS: p<0.001; OS: p<0.001), higher NLR (DFS: p=0.009; OS: p=0.007), and higher PIV (DFS: p=0.002; OS: p=0.005) were identified as independent prognostic factors for DFS and OS. Stage III and IV were stronger prognostic factors than PIV. Except for this, the HR of higher PIV for DFS (HR: 3.605, p<0.001) and OS (HR: 3.600, p<0.001) increased markedly, conforming that PIV was an important prognostic factor for DFS and OS except for TNM stage.

Furthermore, LASSO regression model was also applied to identify potential prognostic factors for DFS of the development cohort ( Figure 2 ). Based on the purpose of constructing a handy prognosis model, the λ that provided the most parsimonious model within one standard error of the optimum value (λ.1se) was slated ( Figure 2A ). Ultimately, the cross-validation screened out three indicators based on λ.1se (0.154) including TNM stage, LMR, and PIV, which could be used for prognosis prediction ( Figure 2B ).

In order to precisely predict 1-year, 3-year, and 5-year DFS for HNSCC patients after radical resection, Model A and Model B were developed based on the results of the multivariate Cox regression analysis ( Figure 3A ) and LASSO regression model ( Figure 3B ), respectively. With each variable scored using the developed model, the 1-year, 3-year, and 5-year probability of DFS for each patient could be estimated by calculating the total score and drawing a vertical line to the probability scale axes.

Several criteria regarding discrimination and calibration were used to thoroughly evaluate and validate prediction performance of Model A and Model B internally. The C-indexes of the two models both performed better than that of traditional AJCC TNM stage in internal validation (Model A, 0.795; Model B, 0.788; TNM stage, 0.691). The calibration curves for the two models showed satisfactory consistency between actual observations and nomogram predictions for 1-year, 3-year, and 5-year DFS ( Figures 4A–C ) and OS ( Supplementary Figures 7A–C ) in internal validation. In addition, the ROC curves for Model A and Model B in internal validation showed superior sensitivity and specificity for DFS ( Figures 4D–F ) and OS ( Supplementary Figures 7D–F ), reflecting good discrimination ability compared to AJCC TNM staging alone. Moreover, according to the results of DCA curves in internal validation, the clinical net benefits for DFS ( Figures 4G–I ) and OS ( Supplementary Figures 7G–I ) of Model A and Model B were outstanding compared with AJCC TNM staging, demonstrating the feasibility of making more valuable judgments in clinical application. To sum up, all results internally validated the excellent predictive ability and accuracy of Model A and Model B in comparison with traditional AJCC TNM stage system.

To confirm the prognostic potential of PIV for HNSCC patients after radical resection, we performed sensitivity analysis ( Supplementary Tables 4 , 5 ) and found that PIV levels were positively associated with the death and recurrence of the development cohort in both Model A and Model B. First, grouped by the cutoff point (<123.3 vs. ≥123.3), the sterling prognostic probability of PIV was confirmed on the basis of HRs for DFS and OS obtained in Model A and Model B. Additionally, whether as continuous or divided into quartiles by PIV level, the results indicated that patients with a higher PIV had significantly worse DFS and OS than that of patients with a lower PIV.

Considering that some HNSCC patients received RT or CRT only because of early stages or other conditions of losing the opportunity for surgery in clinical practice, we supposed that the independent prognostic factors for HNSCC patients after surgery still correlated with prognosis of patients after RT/CRT. For this purpose, we enrolled a RT/CRT cohort composed of 50 HNSCC patients who only received radiotherapy or chemoradiotherapy and applied it for testing the generality of the independent prognostic factors identified in the development cohort. The baseline clinicopathological characteristics of the RT/CRT cohort are listed in Supplementary Table 6 . The primary and secondary endpoints of this cohort were PFS and OS, respectively.

For a start, we divided the cohort into PIV-Low and High groups by predetermined cutoff value and performed Kaplan–Meier analysis and log-rank test to identify the difference between the two groups. The results shown in Supplementary Figure 8 confirm that PIV-High group did have worse PFS (2.8% vs. 78.6%, HR: 10.41, 95% CI: 5.51–19.69, p<0.0001) and OS (11.1% vs. 78.6%, HR: 9.82, 95% CI: 5.06–19.05, p<0.0001) than PIV-Low group.

Furthermore, we carried out univariate and multivariate Cox regression analyses to gain the prognostic factors for PFS and OS of the RT/CRT cohort. As shown in Supplementary Table 7 , the result of multivariate Cox regression analysis verified that higher TNM stage (0/I: Ref, p=0.017; II: HR: 6.219, p=0.018; III: HR: 9.738, p=0.006; IV: HR: 13.945, p=0.001), higher PLR (HR: 1.007, p=0.002), and higher PIV (HR: 4.890, p=0.025) were independent prognostic factors for PFS of the RT/CRT cohort. As for OS, the result showed that higher smoking index (HR: 2.387, p=0.023), higher TNM stage (0/I: Ref, p=0.007; II: HR: 5.203 p=0.039; III: HR: 19.505, p=0.001; IV: HR: 14.773, p=0.001), higher PLR (HR: 1.008, p=0.003), and higher PIV (HR: 5.581, p=0.025) were independent prognostic factors ( Supplementary Table 8 ).

According to the above results, higher TNM stage and higher PIV were important prognostic factors for both patients after radical surgery and patients after RT/CRT. Given that the nomograms, Model A and Model B, which both includes TNM stage and PIV, performed well in internal validation in the development cohort, we wanted to test if the nomograms have the wide application ability of prognosis prediction for patients who only received RT/CRT, in addition to patients received radical surgery.

We applied Model A and Model B for PFS and OS prediction in the RT/CRT cohort and evaluated the performance comprehensively. The C-indexes of the two models were both higher than that of traditional AJCC TNM stage (Model A, 0.815; Model B, 0.793; TNM stage, 0.683). The calibration curves for the two models showed satisfactory consistency between actual observations and nomogram predictions for 1-year, 3-year, and 5-year PFS ( Figures 5A–C ) and OS ( Supplementary Figures 9A–C ). In ROC analysis, the AUC values of Model A and Model B for PFS ( Figures 5D–F ) and OS ( Supplementary Figures 9D–F ) were obviously higher than that of AJCC TNM stage alone. In addition, the results of DCA curves showed that Model A and Model B could gain more clinical net benefits for PFS ( Figures 5G–I ) and OS ( Supplementary Figures 9G–I ) than AJCC TNM stage.

Based on these results, Model A and Model B showed good prediction performance for patients who only received RT/CRT compared with traditional AJCC TNM stage system. It indicated that in clinic, the two nomograms may have the ability of widespread clinical utility for HNSCC patients receiving different treatments including radical surgery or only RT/CRT.