Estimation of Total Sleep Time From Respiratory Event Intervals in Sleep Disordered Breathing

The correct identification of respiratory events and estimation of total sleep time (TST), done in a sleep clinic with polysomnography (PSG), are central elements in the diagnostic process of sleep-disordered breathing and in the quantification of its severity with the apnoea-hypopnea index (AHI). However, total sleep time may not be available in home-based diagnostic procedures with polygraphy (PG), so the respiratory event index (REI), based on total recording time (TRT), is used in place of AHI. Because of that, severity can be greatly under-estimated in people with high sleep fragmentation or long wake bouts. We propose a device-agnostic method to estimate TST using only the time intervals between respiratory events and the TRT. We produce an initial estimate of TST as a function of the REI uncertainty, quantified through bootstrap sampling of respiratory event intervals. Subsequently, we refine this estimate using a regression model that integrates demographic factors and other TST-independent sleep metrics. We tested the approach on 2037 recordings from the MESA cohort. We obtained a moderately good estimation of TST (Correlation 0.84, agreement 2.0  $\pm$  44.5 min), and better TST-based REI than TRT-based REI (70.3% error reduction, ×8 less false negatives) also in presence of low sleep efficiency. Estimates were comparable with state-of-the-art methods based on actigraphy, and further improved by their combined usage, but underperformed against methods using cardio-respiratory signals. It is possible to have a moderate-to-good estimate of the TST employing only minimal information from respiratory events and TRT, improving existing REI estimates without additional signals.