Purpose: Using quality-of-life measures and pulse oximetry, this study developed a two-tiered prediction algorithm with an aim to prioritize sleep-disordered breathing patients for polysomnography.
Methods: Data from 355 patients were evaluated to obtain their clinical information, Chinese version of Epworth sleepiness scale, and snore outcomes survey scores against respiratory disturbance index (RDI). In the first-tier screening, receiver-operating characteristics were calculated with an initial strategy of choosing optimal prediction sensitivity. The second-tier strategy investigated the association between pulse oximetry data (desaturation index of 3%) against RDI to optimize prediction specificity.
Results: The "SOS score of 55 and ESS score of 9" was the optimal combination that yielded the highest sensitivity (0.603) in the first-tier screening. The strategy can includ 94.93% possible patients (probability = 0.6) with positive predictive value of 0.997. The area under the curve (AUC) was 0.88 (p < 0.001). Desaturation index of 3% would optimized specificity (0.966, probability = 0.5) in the second-tier screening to exclude 54% of innocent patients, with negative predictive values of 0.93 and AUC of 0.951 (p < 0.001). The two-tier screening model jointly excluded 4.8% of innocent subjects and prioritized 40% of severe patients for polysomnography.
Conclusions: The prediction model is sufficiently accurate and feasible for large-scale population screening.