A comparison of hypoxic burden algorithms using three different methods for calculating baseline oxygen saturation for predicting cardiovascular death in the Sleep Heart Health Study

Respiratory event related oxygen desaturation area measures have recently shown merit as novel predictors of cardiovascular disease (CVD) outcomes. In this study, we investigate one such measure (hypoxic burden (HB)) and investigate how three different ways of calculating the SpO2 baseline of the HB algorithm impact its ability to predict cardiovascular mortality. The three baseline estimation steps include a pre-event baseline, a record-based baseline, and a fixed baseline. Pulse oximetry signals from the Sleep Heart Health Study and the corresponding CVD outcomes were analyzed. The performance of each baseline method was compared using adjusted Cox proportional hazard regression analysis. Results show that HB with the record-based baseline method returned the best performing results with a hazard ratio (HR) of 1.83 (95% CI: 1.03-3.27, p<0.05) in the fully adjusted model, compared to HB with the pre-event baseline method (HR: 1.60, 95%CI: 0.86-3.00, p>0.05) and HB with the fixed baseline method (HR: 1.73, 95%CI: 0.93-3.22, p>0.05).