Study objectives Desaturation area-based parameters derived from oximetry have emerged as novel predictors of cardiovascular disease mortality. Existing algorithms estimate the area under the oxygen desaturation curve but differ in computational aspects due to variations in baseline, sampling-window, and sleep event choice. These differences result in varying computational complexity and predictive performance. This study systematically characterizes published desaturation area-based algorithms to identify the most effective method for predicting cardiovascular disease-related (CVD) mortality and addressed the influence of computational discrepancy in the prediction. Methods and Results This study utilized data from the Sleep Heart Health Study, including corresponding CVD mortality outcomes and covariates. A total of 4,483 participants (53.4% female; mean age: 64.32 years) were analyzed. Fifteen desaturation area methods were implemented based on variations of 3 published algorithms (hypoxic burden, desaturation severity, respiratory event desaturation transient area). The predictive performance of each method was assessed using Cox proportional hazards regression analysis, with adjustments for relevant covariates. A variation based on the hypoxic burden algorithm that used a record-specific baseline was the best-performing method for predicting CVD mortality outcomes. In the fully adjusted model, it demonstrated the strongest predictive performance, with a hazard ratio of 1.79 and a 95% confidence interval of 1.00–3.19 for predicting CVD mortality (p 0.05). Conclusion Computational discrepancies, particularly the choice of sleep event annotations, were found to have a substantial impact on the predictive ability of desaturation area–based parameters for CVD mortality. Among all evaluated methods, the approach based on hypoxic burden with a record-specific baseline demonstrated the strongest predictive performance.
He et al. (Mon,) studied this question.