Pulse oximetry noninvasively and continuously monitors arterial oxygen saturation (SpO2) by shining red and near-infrared light through a blood-perfused tissue bed and analyzing detected light signals that modulate with the cardiac cycle. Two light-emitting diodes (LEDs) in the sensor provide the light the system uses to resolve relative amounts of oxy- and deoxyhemoglobin (O2Hb and HHb, respectively). Accordingly, manufacturers design and calibrate their readings to correlate with the functional hemoglobin oxygen saturation as measured from invasively sampled arterial blood (SaO2, by definition).
Additional chromophores in the blood can interfere with reading accuracy if they absorb light in the wavelength regions used by the pulse oximeter, i.e., if their optical absorption spectra overlap those of O2Hb and HHb at one or both of the LED wavelengths used in the sensor. Two such examples, indocyanine green and methylene blue, have been documented to interfere with SpO2 readings.1,2 Methemoglobin (MetHb) also biases SpO2 readings when its concentration exceeds levels normally found in the blood.3,4 (This interference is unrelated to the confusion between functional saturation [the percentage of O2Hb relative to the hemoglobin available to bind oxygen] and fractional saturation [the percentage of O2Hb relative to total hemoglobin]. In the presence of dysfunctional hemoglobin, functional saturation is higher than fractional saturation, resulting in a “gap” between the 2 values, because of their definitions. The presence of MetHb may lead the pulse oximeter to read lower than measured functional saturation because of interference and not because the device is tracking fractional saturation.5,6)
Using 2 distinct wavelengths of light limits pulse oximeters to resolving the relative concentrations of 2 blood constituents and is indeed one of the basic assumptions of their simple and easy-to-use design. In a vast majority of monitored patients, O2Hb and HHb are the principal visible and near-infrared light absorbers present in the arterial blood. In certain clinical situations, however, dysfunctional hemoglobins may be present at more significant levels than normal. Increased carboxyhemoglobin (COHb) levels result from patient exposure to carbon monoxide, and MetHb can be created from contact with and/or ingestion of certain chemicals or drugs including dapsone, or in some cases, the use of topical anesthetics such as benzocaine.4,7,8
For patients suspected to have an elevated dysfunctional hemoglobin level, blood samples are obtained for analysis using a laboratory multiwavelength oximeter. These bench-top devices measure precisely and simultaneously the concentrations of O2Hb, HHb, MetHb, and COHb. Recently, an alternative pulse oximeter-like device has become available for the continuous and noninvasive measurement of the percentage of COHb (%COHb) and MetHb (%MetHb) content in the arterial blood (Masimo Rainbow SET® Radical-7 Pulse CO-Oximeter, Masimo Corp., Irvine, CA). This system uses 7 or more LED wavelengths in the sensor to estimate SaO2, %COHb, and %MetHb, reporting values of SpO2, SpCO®, and SpMet™, respectively.9
A similar device to the Radical-7, the Rad-57 Pulse CO-Oximeter (Masimo Corp.), has been reported to accurately track %COHb and %MetHb in healthy adults under controlled laboratory conditions.10 Enrolled subjects breathed room air to maintain normal functional saturation while their COHb or MetHb levels were increased artificially. The observations were thus limited to conditions of only 2 hemoglobin species being present simultaneously in the arterial blood (i.e., O2Hb and COHb, or O2Hb and MetHb). The ability of the device to resolve dysfunctional hemoglobins during hypoxia, or vice versa, was not studied (i.e., 3 or more species). The Radical-7 Operator's Manual notes that “very low SpO2 levels may cause inaccurate SpCO and SpMet measurements,”9 but it does not provide further details or explanation.
In clinical practice, elevated dysfunctional hemoglobin may be present in addition to, rather than instead of, hypoxemia. Thus, it becomes useful to understand the interactions that exist in the measurement of the various hemoglobin species. The Rainbow system's reliability in tracking %COHb under more dynamic conditions has been questioned elsewhere,11 and its degraded accuracy during hypoxia has been observed.12 However, the system's ability to track %MetHb accurately has not been evaluated independently.
The goal of our study was to determine the ability of Masimo's pulse CO-oximeter to estimate %MetHb accurately during hypoxia. We also wished to determine whether the pulse CO-oximeter differs from a standard 2-wavelength pulse oximeter in estimating the SaO2 during normoxia and hypoxia when MetHb is present. Previous work with laboratory animals demonstrated that the ability of 2-wavelength oximeters to detect hypoxemia is impaired, with SpO2 readings approaching 85% with increasing MetHb, independent of SaO2.3
This study was approved by the University of California at San Francisco Committee on Human Research, and informed consent was obtained from all subjects. In the first portion of our study, 8 healthy adult subjects were included with an equal number of men and women and spanning a range of skin pigmentation. Adult nondisposable clip-on-type Rainbow sensors (model DCI, sensor lot 20,924-A7M136) were placed on the index and middle fingers of each subject and connected to 2 Radical-7 oximeters (SET software V188.8.131.52). A radial arterial cannula was placed in either the left or right wrist of each subject. Blood gas analysis to determine SaO2 and %MetHb levels was performed on a multiwavelength optical blood analyzer (OSM™ 3 Hemoximeter™, Radiometer Medical A/S, Copenhagen, Denmark).
Each subject had 2 blood samples drawn while breathing room air at the beginning of each experiment. Elevated MetHb was then induced by IV administration of approximately 300 mg sodium nitrite to produce a target %MetHb level of 7% to 8%, followed by another 2 blood samples. Hypoxia was then induced to different targeted SaO2 levels (between 70% and 100%, based on end-tidal gas analysis) by having subjects breathe mixtures of nitrogen, air, and carbon dioxide according to a protocol that we have described in detail previously.13 Each SaO2 plateau level was maintained for at least 30 seconds and until pulse oximeter readings stabilized, at which point 2 arterial blood samples were obtained approximately 30 seconds apart. SpO2 from the oximeters was recorded at 2 Hz on a computer. SpMet values were recorded manually.
A second group of 6 subjects (2 men and 4 women, spanning light to dark pigmentation) was studied following a similar procedure, with %MetHb elevated to a target level of 15%. A conventional 2-wavelength pulse oximeter was added (Avant® 9700 with clip-on finger probe, Nonin Medical Inc., Plymouth, MN), and the Radical-7 SpCO and SpMet readings were recorded during each blood draw. Blood samples were obtained while subjects breathed room air, and an additional single sample was taken at a target level of 80% SaO2. A Radiometer ABL800 hemoximeter was used for blood analysis with the second study group.
SpMet performance was analyzed by observing the incidence of excessive reading bias at the various levels of hypoxia. Significance was assessed using the Fisher exact test. Positive (PPV) and negative predictive values for detecting methemoglobinemia were calculated from the observed data. SpO2 reading bias, precision, and accuracy calculated as the root mean square error (Arms) were determined for the data collected at the elevated %MetHb levels. Statistical significance was evaluated using the Student t test. For all statistical tests, P < 0.05 was considered significant.
In the first part of the study, 135 blood draws were obtained (12–20 per subject), covering a span of 66.2% to 99% SaO2 and 0.6% to 8.3% %MetHb. Three to 8 draws from each subject were available in the second group (35 total), spanning 76.4% to 98.8% SaO2, 0.6% to 14.4% %MetHb, and normal %COHb values ranging from 0.2% to 1.3%. None of the subjects was anemic (hemoglobin ≤10 gm/dL).
SpMet Reading Accuracy
Figure 1 shows the differences between pulse CO-oximeter and blood gas analyzer measurements of MetHb concentration as a function of SaO2. Table 1 summarizes the SpMet reading characteristics over ranges of progressively decreasing SaO2. A “reading error” is defined here as a displayed SpMet value with an absolute bias in excess of 5% compared with the measured %MetHb. SpMet reading accuracy appeared best when SaO2 was >95%, with an increasing trend to overestimate %MetHb as the SaO2 decreased (P < 0.05).
SpO2 Reading Accuracy
Figure 2 shows the pooled SpO2 readings and reading differences (bias, SpO2 − SaO2) plotted versus SaO2 for data collected with %MetHb >4% from both subject groups. Table 2 summarizes the Rainbow's SpO2 reading biases versus SaO2 level from the first group. For the highest SaO2 range (95%–100%), the SpO2 was 6.3% ± 3.0% lower than SaO2 (Arms = 7.0%). This reading bias was smaller in magnitude for each of the successively lower ranges (P < 0.05) and best in the 70% to 80% range (−1.1% ± 1.7%) with an Arms of 2.1%. Figure 3 shows the SpO2 bias plotted versus %MetHb for the highest SaO2 range (95%–100%).
SpCO Reading Accuracy
The Rainbow's SpCO readings in the second study group are shown in Figure 4. Beginning with %MetHb at a normal level and SaO2 >95%, reading bias increased systematically as %MetHb increased. For %MetHb levels more than approximately 7%, SpCO readings stabilized at or near a displayed value of 50% (%COHb levels in all subjects were <1.3%).
“Gold-standard” bench-top laboratory oximeters determine the levels of the 4 predominant hemoglobins * in vitro by hemolyzing a small volume of sampled blood to remove the erythrocytes' optical light-scattering effects, shining at least 4 wavelengths of collimated light through a clear glass or plastic sample cuvette with known thickness, and calculating the optical absorbance at each wavelength from the detected light levels. These absorbance values are used according to the Lambert-Beer law to ascertain the absolute concentrations of each constituent.14 With a single and known path length through the clear hemoglobin solution, bench-top oximeters can resolve precisely each of the hemoglobin species independent of their individual concentrations, i.e., without “cross-talk” where the presence of one species interferes with the measurement of another.
In contrast, noninvasive in vivo systems collect light that passes through living tissues, a light scattering, heterogeneous optical medium with patient-to-patient and digit-to-digit size variability. The probe's resulting detected light signals come from a distribution of optical path lengths that vary within and among monitored patients and differ for each of the wavelengths of light used.15 The Lambert-Beer law no longer strictly applies and, consequently, these systems (including pulse oximeters) are empirically calibrated in a study group under controlled conditions to best match the observed computed values to those obtained from blood samples. In a light-scattering environment, a measurement system's ability to resolve the various hemoglobin species, without cross-talk, becomes a more challenging task.
Our results suggest that the Rainbow system exhibits cross-talk when >2 hemoglobin species are present. We observed SpMet readings for %MetHb <8% that were consistent with the manufacturer's accuracy specifications (Arms = 1%) when subjects were very well oxygenated (SaO2 ≥98%), although increasing the HHb content in the pulsing arterial blood (hypoxia) caused accuracy to diminish significantly. The SpMet − %MetHb reading bias became progressively larger as the SaO2 decreased in each subject, overestimating %MetHb by at least 10% when SaO2 was <75%. The relative number of readings with an error in excess of 5% MetHb increased over each of the upper 5-point spans of decreasing SaO2, suggesting that the growing amount of HHb in the arterial blood degrades SpMet performance (all SpMet readings were in error by >5% with SaO2 <80%). The reading precision also degraded for SaO2 <95%, in some instances resulting in %MetHb overestimations by 10% to as much as 45%. The larger reading errors were not the same value in the 2 simultaneously placed sensors, suggesting that because the SaO2 and MetHb levels would presumably be the same in the subject's 2 monitored fingers, additional factors beyond cross-talk seem to be involved as well.
Overestimating %MetHb levels in the blood can increase the occurrence of false alarms for methemoglobinemia. Normal %MetHb levels are <2%, and although an increase to the 5% to 8% range is noteworthy, it would represent minimal impact on blood oxygen-carrying capacity and in most cases not be of clinical concern. However, as MetHb levels increase beyond 10%, the decrease in arterial blood oxygen content and delivery become more meaningful because of the offsetting decrease in available functional hemoglobin and the left shift of the oxygen dissociation curve.16 Accordingly, we computed the predictive values in our combined 14-subject data set for detecting methemoglobinemia considering a threshold %MetHb level of 10% (Fig. 5 and Table 3). A PPV close to 100% indicates high correspondence between measured SpMet >10% and true %MetHb >10%, regardless of either of their actual readings. The Rainbow's SpMet PPV seems to suffer from its tendency to greatly overestimate %MetHb as the SaO2 decreases. In the normal SaO2 >95% range, PPV was 90% but decreases to <10% in the 85% to 95% and <85% SaO2 ranges. The PPV improves in our data for detecting abnormal %MetHb levels when a more conservative 7% %MetHb threshold is used. Regardless of the threshold choice, very high SpMet reading bias could be mistaken as a severe methemoglobinemia and affect the choice of intervention. This type of false positive would not be a patient safety issue, whereas a false negative would be of greater concern because treatable methemoglobinemia might be missed. Obtaining a blood sample would help to clarify the true conditions, and would likely be done in a patient with significant hypoxemia.
Additional cross-talk behavior was seen with the Radical-7's SpCO readings in the limited data available from 4 of the second set of subjects studied. Inaccuracy caused by elevated %MetHb >1.5% is noted in the device's operator's manual,9 although its magnitude and direction are not cited. We found estimates of the subjects' normal %COHb levels (<1.5%) to be consistent with the Rainbow's stated reading accuracy (Arms = 3%) before increasing their MetHb. When %MetHb increased beyond the normal level, SpCO greatly overestimated %COHb, displaying a value at or near 50% when %MetHb exceeded about 7% (Fig. 4).
SpO2 generally underestimated SaO2 in our study when MetHb was present beyond normal levels. A tendency for SpO2 to underestimate SaO2 at high saturation and overestimate it at low saturation as %MetHb increases has been reported previously.3,17Figure 3 shows that both types of pulse oximeter underestimate SaO2 as MetHb levels increase. With the 4% to 8% %MetHb concentrations we induced in the first data group, mean SpO2 reading bias was approximately −6% at high saturation, consistent with that seen in the earlier publications for similar concentrations. In the second data group with SaO2 >95% and 10% to 14.4% %MetHb (13 blood draws), we found that both the Rainbow and the conventional 2-wavelength pulse oximeter (Nonin 9700) exhibited unequal although similar magnitudes of reading bias (−4.5% ± 2.0% and −7.3% ± 1.5%, respectively); the difference perhaps attributable to different LED wavelengths used between the sensors.18Figure 2A plots all of our observed SpO2 reading data for %MetHb >4%. Although the general bias with SaO2 >95% was comparable with what Barker et al.3 found in dogs, the slope of our regressions was closer to unity than they found (estimated by translating the published data to use SaO2 as the independent variable). In this %MetHb range, both the Radical-7 and Nonin pulse oximeters were able to reliably trend the hypoxia with SaO2 <90%.
Interestingly, given the Rainbow's erroneously high readings of SpMet when the SaO2 decreases and MetHb's influence to cause SpO2 readings to underestimate SaO2 during normoxia, the Rainbow system creates an ambiguity in diagnosing methemoglobinemia and/or hypoxia. A standard pulse oximeter will read a low SpO2 value when the patient's (a) SaO2 is low or (b) %MetHb is significantly elevated with a normal SaO2; if MetHb is suspected, a blood gas should be ordered and analyzed with a laboratory multiwavelength oximeter. The Rainbow system's SpO2 readings behave in the same manner as a standard pulse oximeter, but we found that its SpMet readings could also increase in both situations: when, again, either the patient's (a) SaO2 is low or (b) %MetHb is elevated. In clinical use, a low SpO2 reading paired with an elevated SpMet value could be attributable to hypoxia, methemoglobinemia, or both. As in using a pulse oximeter alone, resolving the ambiguity in practice calls for obtaining a blood sample. However, both the Radical-7 and the Nonin 9700 are capable of detecting clinically significant hypoxemia at %MetHb levels from 0% to 15%.
Summarizing the interactions we observed, clinicians interested in using the Rainbow system should be aware of the following when interpreting its readings:
- SpMet readings are most accurate when SaO2 is >95%, with an increasing tendency to overestimate the true %MetHb level as SaO2 decreases <95%.
- SpCO readings overestimate, meaningfully and increasingly, the true %COHb levels when %MetHb increases beyond 2%. (SaO2 <95% also seems to bias SpCO readings high as reported elsewhere.12)
- Elevated %MetHb causes both the Rainbow and Nonin to report SpO2 values that are too low at normal SaO2 levels (high 90s) but tracks SaO2 adequately during desaturation, at least up to %MetHb levels of 15%.
Limitations of our study include a lack of SpMet observations made with normal and up to 4% %MetHb during hypoxia. Thus, we cannot comment about the incidence of the Rainbow system falsely suggesting the presence of methemoglobinemia when true %MetHb levels are <4%. The technology used to detect MetHb and COHb in vivo is still being developed and further studies of accuracy may be necessary as newer software and hardware versions become available.†
In conclusion, we found the Rainbow's SpMet readings to become significantly and at times meaningfully inaccurate when SaO2 was <95%, frequently overestimating true %MetHb by 10% to 40%. Similar to the conventional 2-wavelength pulse oximeter included in our study, the Rainbow SpO2 readings become biased to read low at normal SaO2 levels when %MetHb increases beyond the normal levels. However, both types of oximeter are sufficiently accurate in detecting O2Hb desaturation in the presence of MetHb, unlike the poor sensitivity shown previously in dogs.
The authors thank Masimo and Nonin Medical for the loan of the oximeters used in this study.
* Sulfhemoglobin (SHb) and fetal hemoglobin (FHb) are additionally resolved in some laboratory multiwavelength oximeters that use an extended range of wavelength channels for their measurements.
† In July 2009, Masimo introduced a new sensor that is specific for the measurement of methemoglobin concentration. Masimo believes it addresses the issues reported here. The new sensor is being evaluated in the Hypoxia Research Laboratory in the Department of Anesthesia and Perioperative Care at University of California, San Francisco. Drs. Feiner and Bickler are directly involved in these studies.
1. Sidi A, Paulus DA, Rush W, Gravenstein N, Davis RF. Methylene blue and indocyanine green artifactually lower pulse oximetry readings of oxygen saturation. Studies in dogs. J Clin Monit 1987;3:249–56
2. Scheller MS, Unger RJ, Kelner MJ. Effects of intravenously administered dyes on pulse oximeter readings. Anesthesiology 1986;65:550–2
3. Barker SJ, Tremper KK, Hyatt J. Effects of methemoglobinemia on pulse oximetry and mixed venous oximetry. Anesthesiology 1989;70:112–7
4. Trillo RA, Aukburg S. Dapsone-induced methemoglobinemia and pulse oximetry. Anesthesiology 1992;77:594–6
5. Bozeman WP, Meyers RP, Barish RA. Confirmation of the pulse oximetry gap in carbon monoxide poisoning. Ann Emerg Med 1997;30:608–11
6. Toffaletti J, Zijlstra WG. Misconceptions in reporting oxygen saturation. Anesth Analg 2007;105:S5–9
7. Severinghaus JW, Xu FD, Spellman MJ Jr. Benzocaine and methemoglobin: recommended actions. Anesthesiology 1991; 74:385–7
8. Ash-Bernal R, Wise R, Wright SM. Acquired methemoglobinemia: a retrospective series of 138 cases at 2 teaching hospitals. Medicine (Baltimore) 2004;83:265–73
9. Radical-7 Signal Extraction Pulse CO-Oximeter with Rainbow Technology, Operator's Manual. Irvine, CA: Masimo Corporation, 2008
10. Barker SJ, Curry J, Redford D, Morgan S. Measurement of carboxyhemoglobin and methemoglobin by pulse oximetry. Anesthesiology 2006;105:892–7
11. O'Malley GF. Non-invasive carbon monoxide measurement is not accurate. Ann Emerg Med 2006;48:477–8
12. Bebout DE, Mannheimer PD. Laboratory summary report: performance observations of the Rainbow “Pulse CO-Oximeter.” Respir Ther 2008;3:39–41
13. Feiner JR, Severinghaus JW, Bickler PE. Dark skin decreases the accuracy of pulse oximeters at low oxygen saturation: the effects of oximeter probe type and gender. Anesth Analg 2007;105:S18–23
14. OSM™ 3 Hemoximeter™ Operator's Manual.Copenhagen, Denmark: Radiometer Medical A/S, 1997
15. Mannheimer PD. The light-tissue interaction of pulse oximetry. Anesth Analg 2007;105:S10–7
16. Darling RC, Roughton FJW. The effect of methemoglobin on the equilibrium between oxygen and hemoglobin. Am J Physiol 1942;137:56–68
17. Nijland R, Jongsma HW, Nijhuis JG, Oeseburg B, Zijlstra WG. Notes on the apparent discordance of pulse oximetry and multi-wavelength haemoglobin photometry. Acta Anaesthesiol Scand Suppl 1995;107:49–52
© 2010 International Anesthesia Research Society
18. Reynolds KJ, Palayiwa E, Moyle JTB, Sykes K, Hahn CEW. The effect of dyshemoglobins on pulse oximetry: part I, theoretical approach and part II, experimental results using an in vitro test system. J Clin Monit 1993;9:81–90