Exploration of prognostic models for chronic rhinosinusitis with nasal polyps based on machine learning

Abstract

Objective: To analysis the molecular characteristics of chronic rhinosinusitis with nasal polyps (CRSwNP), to unravel its pathophysiological mechanisms, and to develop a prognostic model capable of effectively predicting postoperative recurrence. Methods: The data from three datasets (GSE198950, GSE179265, and GSE136825) were integrated, comprising 39 control cases, 16 cases of chronic rhinosinusitis without nasal polyps, and 89 cases of CRSwNP. Differential expression genes (DEGs) were identified based on adjusted P<0.05 and Log2FC>1. KEGG and GO enrichment analyses, as well as STRING node scoring, were conducted. Variable selection was performed using random forest and least absolute shrinkage and selection operator regression (LASSO), with key nodes identified through intersection analysis. Mann-Whitney U test was applied, and variables with P<0.05 were included in the model. A prognostic model for CRSwNP was constructed using logistic regression, externally validated using RNA-seq data, and evaluated with receiver operating characteristic (ROC) curve analysis to calculate the area under the curve (AUC). Results: This research illustrated both upregulated and downregulated DEGs in CRSwNP, activating pathways like neuroactive ligand-receptor interaction and IL-17 signaling, while inhibiting calcium signaling and gap junctions. Key nodes identified through random forest and LASSO, including G protein subunit γ4 (U=3.00 P=0.028), Cholecystokinin (U=0.50, P=0.006), Epidermal growth factor (U=1.00 P=0.008), and Neurexin-1 (U=0.00, P=0.004), showing statistical significance in external validation. The prognostic model, visualized in a line graph, exhibited high reliability (C-index=0.875,AUC=0.866). The ROC curve in external validation indicated its effectiveness in predicting postoperative recurrence (AUC=0.859). Conclusions: This study integrates multiple datasets on CRSwNP to provide a comprehensive description of its molecular features. The prognostic model, built upon key nodes identified through random forest and LASSO analyses, demonstrates high accuracy in both internal and external validations, thus providing robust support for the development of personalized treatment strategies for CRSwNP.