Accuracy of Pulse Oximetry for Long-Term Oxygen Therapy Assessment in Chronic Obstructive Pulmonary Disease

Rationale: Landmark studies of long-term oxygen therapy (LTOT) in patients with chronic obstructive pulmonary disease (COPD) used arterial oxygen pressure (Pa_O) to define severe hypoxemia; however, oxygen saturation as measured by pulse oximetry (Sp_O) is commonly used instead. The Global Initiative for Chronic Obstructive Lung Disease (GOLD) guidelines recommend evaluation with arterial blood gas (ABG) analysis if Sp_O is ⩽92%. This recommendation has not been evaluated in stable outpatients with COPD undergoing testing for LTOT. Objectives: To evaluate the performance of Sp_O compared with ABG analysis of Pa_O and arterial oxygen saturation (Sa_O) to detect severe resting hypoxemia in patients with COPD. Methods: Retrospective analysis of paired Sp_O and ABG values from stable outpatients with COPD who underwent LTOT assessment in a single center. We calculated false negatives (FNs) as an Sp_O >88% or >89% in the presence of pulmonary hypertension with a Pa_O ⩽55 mm Hg or ⩽59 mm Hg in the presence of pulmonary hypertension. Test performance was assessed using receiver operating characteristic (ROC) analysis, intraclass correlation coefficient (ICC), test bias, precision, and accuracy root-mean-square (A_rms). An adjusted multivariate analysis was used to evaluate factors affecting Sp_O bias. Results: Of 518 patients, the prevalence of severe resting hypoxemia was 74 (14.3%), with 52 missed by Sp_O (FN, 10%), including 13 (2.5%) with an Sp_O > 92% (occult hypoxemia). FNs and occult hypoxemia in Black patients were 9% and 1.5%, respectively, and were 13% and 5%, respectively, among active smokers. The correlation between Sp_O and Sa_O was acceptable (ICC = 0.78; 95% confidence interval, 0.74-0.81); and the bias of Sp_O was 0.45%, with a precision of 2.6 (-4.65 to +5.55%) and A_rms of 2.59. These measurements were similar in Black patients, but in active smokers, correlation was lower and bias showed greater overestimation of Sp_O. ROC analysis suggests that the optimal Sp_O cutoff to warrant LTOT evaluation by ABG analysis is ⩽94%. Conclusions: Sp_O as the only measure of oxygenation carries a high FN rate in detecting severe resting hypoxemia in patients with COPD undergoing evaluation for LTOT. Reflex measurement of Pa_O by ABG analysis should be used as recommended by GOLD, ideally at a cutoff higher than an Sp_O ⩽92%, especially in active smokers.