Prediction of the Apnea-Hypopnea Index from Overnight Pulse Oximetry

Study objectives: To compare the relative usefulness of the different indexes derived from pulse oximetry in the diagnosis of obstructive sleep apnea (OSA), and to determine if a combination of these indexes improves the prediction of the apnea-hypopnea index (AHI) measured by polysomnography. Design: Prediction model developed from 224 patients, validated prospectively in 101 patients from the same center (group 1) and in 191 patients from a different sleep center (group 2). Setting: Two independent sleep clinics run by university sleep specialists. Participants: Patients who underwent polysomnography for suspicion of OSA. Interventions: The following indexes were calculated from pulse oximetry recordings performed simultaneously during polysomnography: (1) Δ index, the average of the absolute differences of oxygen saturation between successive 12-s intervals; (2) desaturation events per hour to 2%, 3%, and 4% levels; and (3) cumulative time spent below 90%, 88%, 86%, 84%, 82%, and 80% saturation. Measurements and results: The best predictor was the Δ index, although desaturation events provided similar levels of diagnostic accuracy. An aggregation of multivariate models using combination of indexes reduced the prediction error (r² = 0.70) significantly (p < 0.05) compared to using the Δ index alone (r ² = 0.60). The proportion of subjects from the validation groups within 95% confidence interval (CI) of the derivation group was 90% (95% CI, 83 to 95%) and 91% (95% CI, 86 to 95%) for groups 1 and 2, respectively. The overall likelihood ratios for the aggregated model in all patient groups were 4.2 (95% CI, 3.3 to 15.3), 3.4 (95% CI, 2.7 to 4.3), 3.0 (95% CI, 2.2 to 4.1), and 6.7 (95% CI, 4.9 to 9.2) for normal (AHI < 5/h), mild (AHI 5 to < 15/h), moderate (AHI 15 to < 30/h), and severe (AHI ≥ 30/h) disease, respectively. Conclusions: The Δ index and oxygen desaturation indexes provided similar levels of diagnostic accuracy. The combination of indexes improved the precision of the predicted AHI and may offer a potentially simpler alternative to polysomnography.