Characteristics of patients
The entire cohort from the SEER database included a total of 2526 LA-HPSCC patients from 2004 to 2015 and met the inclusion criteria; thereafter, patients were randomly stratified into two groups in a ratio of 7:3, with 1768 and 758 patients in the training and validation groups, respectively. There was no statistical difference in the distribution of patients between the training and validation groups (P >0.05). The demographic and clinical characteristics of the patients are shown in Table 1. The majority of the patients were men (n=2084, 82.5%), white race (n=1884, 74.6%), insured (n=1787, 70.7%), and presented with lesions located in the pyriform sinus (n=1377, 54.5%). Concerning treatment modalities, the percentage of patients who received surgery-based, radiotherapy-based, and chemotherapy alone or other treatments were 14.7%, 72.4%, and 12.9%, respectively. Additionally, 276 patients with LA-HPSCC who were treated at the Sichuan Cancer Hospital were enrolled for external validation; their demographics and clinical characteristics are presented inSupplementary Table 1. For the entire population in the SEER database, the median survival time was 20 (18.6-21.3) months, and the median survival was 19 (17.4-20.5) months in the training group and 21(18.9-23.1) months in the validation group. For the external validation group, the median survival time was 19 (15.1-23.1) months.
Screening independent prognostic factors
Univariate Cox regression analysis was performed on the training group to define the clinical parameters significantly associated with OS and cancer-specific survival (CSS) in LA-HPSCC patients. We found that age, race, insurance, marital status, primary site, T stage, N stage, AJCC stage, and treatment were significantly associated with OS and CSS (all P < 0.05, Table 2 and Supplementary Table 2). Furthermore, the Cox multivariate regression analysis determined all these factors to be independent, except the clinical stage (Table 2) because it was a combination of T and N stages and had a significant co-linearity with the TN stage. The OS and CSS curves of the patients grouped by each predictor are displayed in Supplementary Figures 2 and 3.
Construction and validation of the nomogram
The prognostic nomograms integrating all significant independent factors in the training group are shown in Figure 2. The C-indices for OS and CSS prediction of the nomogram models for the training, internal validation, and external validation groups were greater than those of the AJCC stage, suggesting that the models had significantly greater predictive power for the AJCC stage of the disease (Supplementary Table 3).
The calibration plots showed consistency between predictive survival and actual survival. For the training, internal validation, and external validation groups, the nomogram models revealed good accuracy for the 3-year and 5-year OS (Figure 3) and CSS prediction (Supplementary Figure 4). The ROC and DCA analyses both demonstrated that the models for OS (Figures 4, 5) and CSS (Supplementary Figure 5) were superior to the AJCC staging system in prognostic prediction .
Development of an online survival estimate calculator
An online version of our nomograms for OS and CSS in LA-HPSCC patients can be accessed at https://la-hpscc.shinyapps.io/DynNomappHPSCC/ and https://lahpscc.shinyapps.io/DynNomappHPSCCforCSS/ to further assist the researchers and clinicians. The predicted survival probability across time can be easily determined by inputting clinical features and reading output figures and tables generated by the web server.
Comparison between surgery-based and radiotherapy-based interventions and subgroup analysis
The PSM analysis was used to compare the survival differences between the surgery-based and radiotherapy-based treatments. Prior to the matched analysis, we observed that the surgery-based treatment had a better OS and CSS than the radiotherapy-based treatment, with median survival time and 5-year survival rates of 34 vs. 21 months, 42 vs. 26 months, and 35.5% vs. 28.0% and 43.9% vs. 36.8%, respectively (Figures 6A, D, and Supplementary Table 4). After matching for patient characteristics, surgery-based treatment still provided a significant benefit in OS and CSS, with median survival time and 5-year survival rates of 33 vs. 18 months, 40 vs. 22 months, and 35.5% vs. 27.2%, 42.3% vs. 33.7%, respectively (Figure 6B, E, and Supplementary Table 4). In addition, we also used PSM analysis to balance the clinical factors of patients treated with CRT and radiotherapy alone (Supplementary Table 5) and found that adding chemotherapy significantly improved the patient survival rates (Figures 6C, F).
Elder patients (>65 years) generally were excluded from prospective trials and the impact of age on survival of LA-HPSCC is unclear. We found that age group between 65 and 75 had slight but significant increase of death risk than age under 65 (HR 1.17, 95% CI, 1.05-1.30, P=0.004), whereas the risk increased to 1.82 (95% CI, 1.58-2.09, P<0.001) and 3.16 (95% CI, 2.35-4.23, P<0.001) for age between 75 and 85 and 85 beyond, respectively (Figure 6G). Among patients >75 years old, radical surgery (29 patients) provided superior OS (median, 19 months) than radiotherapy-based therapy (45 patients, median, 9 months) as in younger patients but with comparable CSS (median, 19 months vs 12 months, Supplementary Figure 6). However, due to limited patient number and lack of performance score records, the prolonged OS brought by surgery in the elderly needs further validation.
Notably, there was great heterogeneity in the survival rates among the various clinical stages. We found that patients with stage T3N1 disease had a significantly poorer prognosis than those with other stage III diseases, with a median survival time of 15 vs. 35 months (Figure 6H). Patients staged T4aN0 and T1N2 had a better prognosis than other stage IVA patients, with median survival times of 38 and 57 vs. 17 months (Figure 6I), which suggests that these TN groups of HPSCC may need adjustment in staging, upgrade, or downgrade, respectively.