Revisiting level II sleep studies in the era of COVID-19: a theoretical economic decision model in patients with suspected obstructive sleep apnea

Abstract

Background The recent pandemic has made it more challenging to assess patients with suspected obstructive sleep apnea (OSA) with in laboratory polysomnography (PSG) due to concerns of patient and staff safety. The purpose of this study was to assess how Level II sleep studies (LII, full PSG in the home) might be utilized in diagnostic algorithms of suspected OSA using a theoretical decision model. Methods We examined four diagnostic algorithms for suspected OSA: an initial PSG approach, an initial LII approach, an initial Level III approach (LIII, limited channel home sleep study) followed by PSG if needed, and an initial LIII approach followed by LII if needed. Costs per patient assessed was calculated as a function of pretest OSA probability and a variety of other variables (e.g. costs of tests, failure rate of LIII/LII, sensitivity/specificity of LIII). The situation in British Columbia was used as a case study. Results The variation in cost per test was calculated for each algorithm as a function of the above variables. For British Columbia, initial LII was the least costly across a broad range of pretest OSA probabilities (< 0.80) while initial LIII followed by LII as needed was least costly at very high pretest probability (> 0.8). In patients with a pretest OSA probability of 0.5, costs per patient for initial PSG, initial LII, initial LIII followed by PSG, and initial LIII followed by LII were: $588, $417, $607, and $481 respectively. Conclusions Using a theoretical decision model, we developed a preliminary cost framework to assess the potential role of LII studies in OSA assessment. Across a broad range of patient pretest probabilities, initial LII studies may provide substantial cost advantages. LII studies might be especially useful during pandemics as they combine the extensive physiologic information characteristic of PSG with the ability to avoid in-laboratory stays. More empiric studies need to be done to test these different algorithms.