Nomogram outperforms gradient boosting machine for prognostic prediction of laryngeal squamous cell carcinoma: a...

The incidence of laryngeal squamous cell carcinoma (LSCC) remains persistently high, necessitating accurate prognostic prediction for clinical treatment guidance. By comparing the performance of LSCC models based on the Surveillance, Epidemiology, and End Results (SEER) database and those constructed from single-center datasets, this study provides an effective tool for clinical prognosis evaluation. Data from 526 patients with LSCC were extracted from the SEER database. Univariate and multivariate Cox regression analyses were performed to identify independent predictors of overall survival (OS) in LSCC patients. Subsequently, patients were randomly assigned in a 7:3 ratio to the modeling group and the test group. Based on the modeling group data, nomograms and gradient boosting machine (GBM) models were constructed using R software (version 4.4.1) and their performance was evaluated. The testing cohort was utilized to assess the predictive accuracy of the model. In addition, 207 LSCC patients diagnosed at The First Affiliated Hospital of Yangtze University from February 2020 to April 2024 were retrospectively selected as an external validation cohort. Univariate and Multivariate Cox regression analyses determined that age (60-75 years: HR=1.333, P=0.085; >75 years: HR=2.726, P<0.001), tumor size (HR=1.013, P=0.035), radiation (HR=7.555, P<0.001), cause of death (COD, HR=3.996, P<0.001), marital status at diagnosis (HR=1.444, P=0.006), and T stage (HR=1.652, P=0.017) were independent predictive indicators affecting the OS of LSCC patients (P<0.05). On this basis, nomogram and GBM models were constructed. The ROC curve showed that the GBM model had an AUC of 0.747, 0.763, and 0.785 at 1-, 2- and 3-years, respectively. For nomogram model, the 1-, 2- and 3-year AUC values reached 0.809, 0.782 and 0.811, respectively. Delong test showed that the AUC values of nomogram were all higher than those of the GBM model (P<0.05). Next, we used nomogram models for verification. AUC values in the verification cohort were 0.783, 0.786, and 0.801, respectively. The AUC values for the external validation cohort were 0.795, 0.760, and 0.783, respectively. The calibration curve shows that the predicted value is basically consistent with the real value. The nomogram model has robust prediction ability and reliable calibration, and its performance is better than that of GBM model.