Automatic detection of sleep-disordered breathing from a single-channel airflow record

Single-channel airflow monitors developed for screening of sleep-disordered breathing (SDB) have conflicting results for accuracy. It was hypothesised that the analytical algorithm is crucial for the performance and the present authors tried to develop a novel computer algorithm. A total of 399 polysomnography (PSG) records were employed, including a thermal sensor signal. The first 100 records were used in the development of the algorithm and the remainder for validation. In addition, 119 PSG records, including a thermocouple signal and a nasal pressure signal, were used for the validation. The algorithm was designed to obtain a time series (flow-power) using power spectral analysis, which expresses fluctuation in the airflow signal amplitude. From the time series the algorithm detects transient falls of the flow-power and calculates flow-respiratory disturbance index (RDI), defined as the number of falls per hour. In the validation group, the areas under receiver operating characteristic curves for diagnosis of SDB (apnoea/hypopnoea index ≥5) were 0.96, 0.95 and 0.95, for the records of the thermal sensor, thermocouple and nasal pressure system, respectively. The diagnostic sensitivity/specificity ratios of the flow-RDI were 96/76, 88/80 and 97%/77%, respectively. The present results suggest that a single-channel airflow monitor can be used to detect sleep-disordered breathing automatically if the analytic algorithm is optimised.