Hematological Biomarkers of the Obstructive Sleep Apnea Syndrome: A Machine Learning-Based Diagnostic and Prognostic Model

Abstract

Objectives: To investigate the diagnostic and prognostic utility of systemic inflammatory biomarkers—including C-reactive protein (CRP), systemic immune-inflammation index (SII), and Fibrinogen—in patients with obstructive sleep apnea syndrome (OSAS), and to develop a machine learning-based stratification model for disease severity and treatment response. Study Design: Prospective observational cohort study. Setting: Single tertiary referral sleep and otolaryngology center. Methods: Adult OSAS patients (n = 195) diagnosed via polysomnography were treated with either CPAP or surgery and reassessed after ~4 months (16–20 weeks). Hematologic biomarkers were measured pre- and post-treatment. OSAS severity was staged using a composite polysomnography (PSG)-based index. Statistical analyses included mixed linear modeling, ROC analysis, unsupervised clustering, and machine learning (Random Forest) to evaluate biomarker utility. Results: CRP demonstrated the highest diagnostic accuracy for severe OSAS (AUC = 0.91, sensitivity = 88.2%, specificity = 85.7%). Fibrinogen showed the strongest correlation with disease severity (ρ = 0.81) and the largest post-treatment reduction (Cohen’s d = 1.41). SII also correlated with PSG stage and declined significantly after treatment. Machine learning confirmed CRP, SII, and Fibrinogen as top predictors of severity. Clustering analysis revealed three distinct inflammatory phenotypes of OSAS with differential biomarker responsiveness. Conclusions: CRP, SII, and fibrinogen may support risk stratification and follow-up in OSAS but require prospective validation before clinical use. These findings should be viewed as exploratory and hypothesis-generating. Larger multicenter studies with external validation are needed before these biomarkers or the machine-learning model are applied in routine practice.

Topics & Concepts