Abstract
Background: Obstructive sleep apnea (OSA) is one of the most prevalent yet underdiagnosed sleep disorders. We evaluated the diagnostic accuracy of radar-based systems and ranked frequency bands for the non-contact detection of OSA. Methods: A systematic search of six databases was conducted from inception to May 23, 2025. Eligible studies included adults assessed for OSA using radar-based systems compared to polysomnography. Hierarchical SROC modeling, threshold-based meta-analyses, and frequency band-stratified network meta-analysis were performed. Certainty of evidence was assessed using GRADE. The PROSPERO registration number is CRD420251059236. Results: We identified 23,906 records and included 20 studies involving 1540 participants. The primary outcome included a high area under the curve (AUC) of approximately 0.91, an optimal apnea-hypopnea index (AHI) cutoff of ≥22 with a sensitivity of 0.8155 (95% confidence interval (CI): 0.6862-0.8993) and specificity of 0.8819 (95% CI: 0.7799-0.9402). At an AHI threshold of 30, X-band dual radar performed the best, followed by K-band, which yielded significant but more variable results. C-bands consistently showed lower diagnostic values. Conclusions: This study provides a novel radar band comparison for OSA detection, highlighting clinically relevant thresholds. Key limitations are indirect comparisons and limited, varied samples. Radar-based systems show high sensitivity for OSA detection, optimized by frequency, radar type, artificial intelligence support, and dual sensors within 0.2-1.5 m. Future work should expand the frequency analysis, standardize AHI thresholds, and validate results in specific subgroups.