Abstract Rationale Body position influences the severity of obstructive sleep apnea (OSA), while respiratory rate (RR) provides complementary information on breathing dynamics. Both are traditionally measured with wired sensors and thoracoabdominal belts, which may affect natural sleep and limit the feasibility of home sleep apnea test (HSAT). This study aimed to validate a HSAT wireless sternal sensor?patch for simultaneous estimation of body position and respiration. Methods Patients with suspected OSA underwent full in-laboratory PSG that included information of body position (EMBLA S7000, Embla Systems, USA). The patients also wore a sternal wireless patch (Bodystar®, Biologix Sistemas S.A., Brazil). The signal was sent via bluetooth to a mobile phone and analyzed using a cloud algorithm for detection of body position (Upright, Supine, Prone, Right, Left) and chest-wall motion for respiration detection. Label of body position was determined at a frequency of 1 Hz by visualization of a synchronized bedroom video by a blinded rater. Recordings were time-aligned and missing data were excluded. Both Bodystar® and PSG position channels were evaluated against video using accuracy, macro-F1, balanced accuracy, and Cohen’s κ (95% CIs via night-level bootstrap, n = 2,000). Paired nightly differences (Bodystar® vs PSG) were tested for superiority with the Wilcoxon signed-rank test. For respiration, triaxial accelerometry was filtered and combined into a respiratory-effort waveform; RR was derived by short-time spectral analysis with adaptive quality control and minute-level aggregation. Agreement with PSG RR was assessed by Mean Absolute Error (MAE), Bland-Altman limits of agreement (LoA), Lin’s Concordance Correlation Coefficient (CCC), and correlation (r). Results Twenty-one patients (age: 55±18 years; 62% male; body mass index: 29±5 kg/m²; apnea-hypopnea index: 17 11-43 events/hour) contributed 501,664 position labeled seconds. The overall time on Supine, Right, Left and Upright was 49.1%, 25.3%, 24.8% and 0.8%, respectively. Bodystar® achieved higher accuracy and agreement with video than the PSG position channel (Table). Paired nightly comparison favored Bodystar® for macro-F1 (Δ = +0.021; p = 0.02). Per-class F1 was Upright 0.93, Supine 0.94, Right 0.95, Left 0.92 for Bodystar® versus 0.08, 0.91, 0.87, 0.96 for PSG, respectively. RR derived from Bodystar® (18 patients; 5,098 minutes) showed an excellent accuracy: MAE 0.31 breaths·min−1, LoA −1.16 to + 1.15, r = 0.965, CCC = 0.965, |Δ|≤2 = 98.5%, |Δ|≤5 = 99.9%. Accuracy remained stable across body positions (MAE ≈ 0.3, LoA ≈ ±1 breaths·min−1). Conclusion The Bodystar® sensor accurately determined body position and RR. This is a patient-friendly belt free tool that increases HSAT capacity of sleep phenotyping. This abstract is funded by: None
Rocha et al. (Fri,) studied this question.