Validation of an artificial intelligence based single-channel photoplethysmography home sleep apnea test.

STUDY

OBJECTIVES

To validate EnsoSleep photoplethysmography (ePPG) software as a medical device (SaMD) for home sleep apnea testing compared to polysomnography.

METHODS

Simultaneous polysomnography (PSG) and photoplethysmography (PPG) signals were recorded for a single night in 225 individuals at risk for obstructive sleep apnea (OSA). PSG was scored by three registered polysomnographic technologists and reviewed by a board-certified sleep specialist for fidelity. The ePPG studies were scored via a machine learning (ML) based scoring module. McNemar’s test, Deming regression, Bland-Altman plots, and sensitivity and specificity were used to compare ePPG to PSG. In addition to AHI, we also assessed total sleep time, sleep efficiency, sleep latency, wake after sleep onset, and sleep architecture.

RESULTS

McNemar’s test resulted in a p-value of 0.511, suggesting no systematic differences between the measurements. For ePPG, the mean difference in 3% AHI, compared to PSG, was 0.74 higher (95% CI: 0.43, 1.06) apneas and hypopneas per hour of sleep. The OSA classification 3% AHI sensitivity was 90.9% (95% CI: 86.3%, 95.4%), with a specificity of 72.3% (95% CI: 62.5%, 81.9%). The mean difference between PSG and ePPG was − 2.94 min (95% CI: -4.44, -1.62) for total sleep time, -9.9 min for sleep latency (95% CI: -10.80, -8.52), 9.42 min (95% CI: 8.34, 10.62) for wake after sleep onset, and − 0.6% (95% CI: -1.0%, -0.3%) for sleep efficiency.

CONCLUSIONS

This research demonstrated that ePPG SaMD is non-inferior to other home sleep apnea test (HSAT) options as an aid in the diagnosis of OSA.