How Accurate is Pulse Oximetry in the ICU?

Abstract & Commentary

Van de louw and colleagues at Henri Mondor hospital in Paris used carefully standardized techniques to make simultaneous measurements of arterial blood oxyhemoglobin saturation by both blood gas analysis (SaO2, the reference standard) and pulse oximetry (SpO2, the experimental measurement) in all patients admitted to their medical ICU during a 5-month period for whom an arterial blood gas was ordered. Repeated data collections were carried out in all patients who had multiple blood gases drawn. Bland-Altman analysis was performed on all paired measurements in order to determine bias and precision of SaO2 vs. SpO2.

One hundred two patients had 323 blood gas specimens drawn during the study period, yielding an equal number of paired SaO2 and SpO2 measurements. The mean difference between SaO2 and SpO2 was -0.02%, and the standard deviation of the differences was 2.1%. From one sample to another, the fluctuations in the differences between SaO2 and SpO2 indicated that SaO2 could not reliably be predicted from SpO2 after a single arterial blood gas. To detect a SaO2 of 90% or less, a SpO2 of 90% or less had a sensitivity below 70%. To achieve a negative predictive value of 99%, a SpO2 threshold of 94% was necessary to detect a SaO2 of 90% or less. Statistically significant differences were also observed between the 3 brands of pulse oximeter used (Hewlett-Packard, Nellcor, and Ohmeda) with respect to how well the SpO2 reading reflected the true SaO2. Agreement between SaO2 and SpO2 was poorer in patients with hypoxemia and in those who required vasoactive drugs (Van de Louw A, et al. Accuracy of pulse oximetry in the intensive care unit. Intensive Care Med. 2001; 27[10]:1606-1613).

Comment by David J. Pierson, MD

I thought the findings of this study were remarkable for 2 reasons, only one of which was commented on by Van de Louw et al. This latter was the fact that the clinical accuracy of SpO2 is nowhere near as good as we tend to assume it is in managing patients in the ICU. In the 102 medical ICU patients in whom the measurements were made, a "real" saturation (that is, SaO2 as measured on a blood gas machine in the laboratory) of 90% could be associated with a SpO2 of anywhere from 86% to 94%. As Van de Louw et al point out, this range of SpO2 values could well lead to different responses from those managing the patient in terms of diagnostic procedures or ventilator changes.

The range of differences between SaO2 and SpO2 found in this study is within the manufacturers’ specifications for their pulse oximeters. The devices were behaving as they were intended. However, our tendency as clinicians is to accept a digitally displayed value as shown. When the pulse oximeter says 94%, we assume the patients oxygenation is all right and when it says 86% we worry. This study emphasizes the fact that the same oxygenation in the patient can be associated with both of those SpO2 values, and with anything in between.

The other thing I thought was remarkable about the results of this study was that the 102 patients had only 323 arterial blood gases drawn while they were in the ICU. There were only 123 SpO2 data points from the 62 mechanically ventilated patients in the study. This means that, in Van de Louw et al’s ICU, SpO2 readings are actually used to manage the patients, and that relatively few arterial blood gases are drawn. This is in keeping with what has been shown in several studies, cited by Van de Louw et al, that the use of pulse oximetry can decrease the number of arterial specimens drawn and thus save money and resources. However, in the ICUs in which I work I am afraid this has not occurred in actual practice. Every ventilated patient gets a daily "routine" blood gas before morning rounds, another one every time the ventilator settings are changed, and usually another one if the indicated SpO2 drops below a given value (typically 90%). I would bet that the average ventilated patient in my hospital gets several times the average of 2 blood gas specimens observed in this study—and I doubt that my hospital is unique in this respect.

Given the results of the present study, maybe this practice is not so unreasonable. In order to guarantee a true arterial saturation of 90% or more, the SpO2 threshold should be 94%. In a critically ill patient, a SpO2 of less than 90% may occur in the presence of a SaO2 well above 90%, but then again it may be an accurate reflection of clinically important hypoxemia. In the management of hypoxemic patients or those requiring vasoactive agents, especially, a pulse oximeter may not be as helpful as we have tended to assume it is.

Dr. Pierson is Professor of Medicine, University of Washington, Medical Director, Respitory Care, Harborview Medical Center, Seattle, WA.