Do changes in pulse oximeter oxygen saturation predict equivalent changes in arterial oxygen saturation?
Critical Care volume 7, Article number: R67 (2003)
This study investigates the relation between changes in pulse oximeter oxygen saturation (SpO2) and changes in arterial oxygen saturation (SaO2) in the critically ill, and the effects of acidosis and anaemia on precision of using pulse oximetry to predict SaO2.
Patients and methods
Forty-one consecutive patients were recruited from a nine-bed general intensive care unit into a 2-month study. Patients with significant jaundice (bilirubin >40 μmol/l) or inadequate pulse oximetry tracing were excluded.
A total of 1085 paired readings demonstrated only moderate correlation (r= 0.606; P < 0.01) between changes in SpO2 and those in SaO2, and the pulse oximeter tended to overestimate actual changes in SaO2. Anaemia increased the degree of positive bias whereas acidosis reduced it. However, the magnitude of these changes was small.
Changes in SpO2 do not reliably predict equivalent changes in SaO2 in the critically ill. Neither anaemia nor acidosis alters the relation between SpO2 and SaO2 to any clinically important extent.
Pulse oximetry is used almost universally in the management of critically ill patients in the intensive care unit and operating theatre . Its uses include the detection of hypoxia , avoidance of hyperoxia , reduction in the frequency of blood gas analysis , titration of fractional inspired oxygen  and for weaning from mechanical ventilation .
An arterial oxygen saturation (SaO2) of 90% has been proposed as a target for adequate oxygenation during mechanical ventilation . Previous studies investigating the use of the pulse oximeter oxygen saturation (SpO2) in intensive care patients have reported that the minimum SpO2 levels to maintain SaO2 at 90% range between 92% and 96 % [4, 6, 7]. However, these studies have not answered the question of whether, after achieving a target SaO2, a subsequent change in SpO2 predicts a corresponding change in SaO2 in the critically ill.
Some studies have reported that anaemia reduces the precision of pulse oximetry  by increasing the signal to noise ratio with low haemoglobin concentrations, whereas others failed to demonstrate this phenomenon [9, 10]. Acidosis may also influence the relation between SpO2 and SaO2. The in vitro method employed by the carbon monoxide (CO)-oximeter requires red blood cell lysis, whereas the pulse oximeter analyzes haemoglobin in whole blood . The difference between intracellular and extracellular hydrogen ion concentrations under normal physiological conditions has been incorporated into the pulse oximeter algorithms. However, the robustness of this adjustment has not been evaluated in the critically ill and acidotic patient.
We therefore conducted a prospective observational study to test the hypothesis that a change in SpO2 would predict an equivalent change in SaO2. Such a relation, if it exists, would be invaluable in deciding when to titrate fractional inspired oxygen and/or repeat arterial blood gases in the individual patient. Furthermore, we examined the effects of anaemia and acidosis on the precision of using the pulse oximeter to predict the SaO2 in a heterogeneous group of critically ill patients.
Patients and methods
This study was considered by the local research and ethics committee and the need for informed consent was waived in view of the observational nature of the study.
During a 2-month period all patients admitted to our intensive care unit (ICU) who had an arterial line for the measurement of blood gases and who were being monitored by continuous pulse oximetry were recruited. The following patients were excluded: those with significant jaundice (bilirubin >40 μmol/l) or a history of smoke inhalation; those with an inadequate SpO2 trace (as determined by visual analysis of a flat, absent, or irregular signal waveform); and those in whom fewer than two arterial blood gas readings were taken.
Serial arterial blood gas samples were taken after 5 ml blood had been discarded when indicated as part of routine clinical care. No samples were taken solely for the study nor was any attempt made to vary inspired oxygen concentration or mechanical ventilation for the purposes of the study. The samples were analyzed in a standardized manner within 2 min of sampling. Arterial blood gas samples were analyzed using a CO-oximeter (ABL725, Radiometer, Copenhagen, Denmark) that was calibrated daily by laboratory staff and has a 2-hourly automatic internal calibration sequence. Haemoglobin concentration (g/dl), hydrogen ion concentration (nmol/l), and percentage SaO2 were recorded for each sample. Precision and accuracy of a whole blood sample for SaO2, hydrogen ion concentration and haemoglobin concentration are 0.3 and 0%, 0.034 and 0.008 nmol/l, and 0.12 and 0.4 g/dl, respectively, within a haemoglobin range of 5–20 g/dl.
Pulse oximetry readings were recorded simultaneously with blood gas sampling using a Nellcor (Puritan Bennett, Pleasanton, NJ, USA) finger probe attached to a Hewlett Packard (Palo Alto, CA, USA) Merlin monitor. The pulse oximeter displays an average SpO2 from the preceding 5-s beat by beat analysis. The measurements between healthy individuals (n= 12) had a coefficient of variation of 0.4% at a SpO2 of 97%. Probes were attached to a finger, choosing the digit that gave the best trace and but not necessarily on the arm from which the arterial blood gas sample was drawn. However, the same probe was used for all measurements from the same patient.
Data were stored using Microsoft Excel 97 and analyzed using SigmaStat for Windows 95 (SPSS Inc. Chicago, IL, USA) and GLIM (Generalized Linear Interactive Modeling) version 4, update 8 (Royal Statistical Society, London, UK), running on a DEC Alpha AXP mainframe computer under the Ultirx operating system (OSF/1). The changes in residuals were tested for normality and found to be normally distributed. The linear relations between differences in two successive measurements of SpO2 and SaO2 in all patients were analyzed using Pearson correlation coefficient (r), linear regression and goodness-of-fit (adjusted R2). The variations between and within the patients were examined using comparisons of the residual standard deviations (SDs) between a single line from a common slope through all the changes for all 41 patients and a separate line to each patient.
The effects of anaemia and acidosis on the agreement between the two measurement techniques were examined using a Bland–Altman plot  in which the difference between SpO2 and SaO2 was plotted against their average . Bias and the limits of agreement were calculated. Bias was calculated as the mean of the differences between the CO-oximeter and pulse oximeter readings (SaO2–SpO2) . Positive bias indicated that the pulse oximeter underestimated the SaO2, whereas negative bias indicated that the pulse oximeter was overestimating the SaO2. The limits of agreement were taken as the bias ± (1.96 × SD) [6, 13].
Approximately 95% of data fell within the haemoglobin concentration range 8–11.9 g/dl and the hydrogen ion concentration range 25–62.9 nmol/l (pH7.2–7.6). Therefore, haemoglobin concentrations ≤ 7.9 or ≥ 12 g/dl or hydrogen ion concentrations ≥ 63 nmol/l were regarded in the study as extremes. The differences of biases between these three groups were analyzed using one-way, repeated measure analysis of variance. P ≤ 0.05 was considered statistically significant.
Forty-one (22 male) patients (age [mean ± SD] 70 ± 14 years) were recruited into the study. A total of 1132 simultaneous arterial blood gas and pulse oximeter readings were taken (mean [range] 27 [3–91] readings per patient). Sequential readings in each patient were grouped together into pairs, which gave 1085 paired readings (47 readings were excluded because they were either not paired or unidentifiable to a particular patient, or the patient had fewer than two readings taken). These data were analyzed to determine the relation between changes in SpO2 (ΔSpO2) and changes in SaO2 (ΔSaO2).
The mean ± SD for SpO2 was 94.6 ± 2.7% and the mean for Sao2 was 95.9 ± 2.4%. In terms of predicting ΔSao2 from ΔSpo2, fitting a single line from all the 41 patients, gives a residual SD of 1.303 and fitting a separate line to each patient gives a residual SD of 1.288 (P = 0.999). Therefore, there was no significant difference in residual SD within patients overall. Although we found moderate correlation between SpΔO2 and SaΔO2 (r = 0.606; P < 0.01; Fig. 1), only 36.7% of the variation in this relation was due to the association of changes in SpO2 with changes in SaO2 (adjusted R2 = 0.367). The prediction of ΔSaO2 from ΔSpO2 (ΔSaO2 = 0.003 + 0.477 ΔSpO2) demonstrates that the pulse oximeter overestimates actual changes in SaO2.
The 1085 simultaneous arterial blood gas and pulse oximeter readings from the 41 patients were analyzed to determine the effects of anaemia and acidosis on bias and limits of agreement. For the data altogether, the bias was 1.34 and the limits of agreement were -2.29 and +4.97 (Fig. 2). There were only small changes in bias with anaemia (+2.09) and acidosis (+0.38), as shown in Table 1. The difference in bias between hydrogen ion concentrations of 25–63 nmol/l and ≥ 63 nmol/l (P < 0.01), and between haemoglobin concentrations of <8 g/dl, 8–11.9 g/dl and >12 g/dl (P < 0.01) all achieved statistical significance. The bias was not significantly different between haemoglobin concentrations of <7.9 g/dl and 8–11.9 g/dl (P= 0.24). There were insufficient numbers in the group with hydrogen ion concentration <24.9 nmol/l (n= 10) for analysis to be done.
The present study shows that changes in SpO2 do not reliably predict equivalent changes in SaO2, with the pulse oximeter tending to overestimate actual changes in SaO2. We also showed that SpO2 underestimates SaO2 to a greater extent with progressive anaemia, whereas acidosis increases the SpO2 estimate of SaO2. However, the clinical significance of these changes is small.
The titration of fractional inspired oxygen during weaning from mechanical ventilation is frequently adjusted with the goal of maintaining a target SpO2 value. Jubran , in a study involving 54 ICU patients, reported that levels of SpO2 of 92% in white patients and 95% in black patients maintained arterial oxygen tension at 8 kPa or greater in 92% and 85% of patients, respectively. Seguin  defined a minimum SpO2 of 96% to ensure that no patients had a SaO2 below 90%. This approach avoided hypoxia, but 15% of patients had a SaO2 of 98% or greater.
Although target values can be helpful, it would be valuable to know whether a change in SpO2 would predict a similar change in SaO2 in critically ill patients over time. Hypothetically, the relatively static patient factors that interfere with pulse oximetry (skin colour, finger size, carboxyhaemoglobin, methaemoglobin) do not change, and so the correlation between changes in SaO2 and SpO2 might be expected to be closer than that between absolute values from a mixed patient population. This could allow individualized target SpO2 to be set, based on a single, one-off SaO2 reading. Only one small study has attempted to address this question in the intensive care setting. In a series of 45 patients (135 measurements), Van de Louw and coworkers  recently reported that changes in SpO2 could not accurately predict changes in SaO2. Our larger study supports and extends this early finding. The prediction of SaΔO2 from SpΔO2 (ΔSaO2 = 0.003 + 0.477ΔSpO2) demonstrates that, on average, the pulse oximeter overestimates actual changes in SaO2. This suggests that a similar degree of caution is required in interpreting changes in pulse oximetry in the critically ill as in one-off readings.
Progressive reductions in haemoglobin concentration may reduce the precision of the pulse oximeter as the signal : noise ratio from surrounding tissue increases . Early studies examining the effects of anaemia on the precision of the pulse oximeter found reduced precision in association with anaemia. Lee and coworkers  demonstrated a deterioration in bias and precision in dogs with a haematocrit below 10%, and Severinghaus and coworkers  reported increased error in anaemic humans when the SaO2 was less than 75%. In contrast, case reports have described cases in which the pulse oximeter remained precise at haemoglobin concentrations of 2.7 g/dl  and 3.0 g/dl . A subsequent case series of 17 patients with acute anaemia due to haemorrhage (haemoglobin concentration 2.3–8.7 g/dl) did not detect any deterioration in the accuracy of measurements using the pulse oximeter in the absence of hypoxia . Our study did not include sufficient numbers with hypoxia (SpO2 < 90%) for the influence of anaemia on bias and precision to be studied in this patient group. However, under normal physiological conditions (SpO2 > 90%) our results support and extend previous findings in demonstrating that anaemia has only a minor impact on the precision of measurements using the pulse oximeter.
Our data show that, in the presence of acidosis, the degree to which SpO2 underestimates SaO2 was reduced. One possible explanation for this finding may relate to the differences in the techniques used for measuring oxygen saturation. The pulse oximeter analyzes haemoglobin saturation in whole blood in vivo , whereas SaO2 measured by CO-oximetry requires red blood cell lysis prior to analysis. Under normal physiological conditions, algorithms incorporated in the pulse oximeter account for this , although the validity of this adjustment has not been tested outside normal physiological ranges. Alternatively, the effects of the complex interactions between cardiac output , systemic vascular resistance , temperature  and vasoactive drugs [14, 21] on precision of measurements using the pulse oximeter might have contributed to this finding. A further study looking at the precise contribution of each of these factors would be required to elucidate the aetiology of these findings definitively.
There are several potential confounding variables that were not controlled for in the study design. First, like in other studies [4, 6], we did not analyze the influence of carboxyhaemoglobin and methaemoglobin concentrations on bias and precision. The pulse oximeter is unable to distinguish between these two forms of haemoglobin and oxyhaemoglobin, leading it to overestimate the actual SaO2 if significant concentrations of either are present [22, 23]. We excluded patients with a history of smoke inhalation, in whom carboxyhaemoglobin levels may be high. In nonsmokers carboxyhaemoglobin levels are normally less than 2% and methaemoglobin levels are less than 1%  – levels that are already accounted for by the built-in algorithms of pulse oximeters. In cigarette smokers carboxyhaemoglobin is initially elevated (average 4.78%) but falls over time (half life 5–6 hours) . The clearance of carboxyhaemoglobin is also accelerated by ventilation . Because most patients had been ventilated for several hours before entry into the study, this is unlikely to have significantly confounded the results. We excluded patients with significant jaundice – a group known to have high carboxyhaemoglobin levels  – in order to minimize this potential error, and no patients were admitted following smoke inhalation during the study period. Anaemia and acidosis have not been found to influence car-boxyhaemoglobin or methaemoglobin concentrations. Although we believe that the influence of carboxyhaemoglobin levels in the study was minimal, we are unable to rule it out as a potential confounding variable.
Second, we did not classify patients according to skin colour or race, which may impact on accuracy of the pulse oximeter . Because skin colour is constant, comparisons of changes in SpO2 are unlikely to have been affected. Data for the assessments for bias and precision caused at the extremes of anaemia and acidosis were collected from 19 and 14 patients, respectively, and there did not appear to be any systematic difference in the groups' racial composition from that in the overall study population. No patients to our knowledge had sickle cell anaemia/trait , although this was not specifically tested for.
Third, the mean SpO2 reading for the total data was 94.6%, with a corresponding SaO2 value of 95.9%. This is consistent with previous investigators' recommendations for minimal target values for SpO2 during mechanical ventilation. However, less than 5% of data fell in the range of SpO2 levels below 90%. Fig. 2 shows increasing positive bias and greater variation as saturations fall. This is consistent with a worsening of bias and precision with pulse oximetry when the SaO2 is less than 90% . At lower saturations the effects of anaemia and acidosis may become more prominent, and our results should therefore be applied with caution in this situation.
Finally, the pulse oximeter presents SpO2 data as integers whereas the CO-oximeter presents SaO2 data to 1 decimal place. With over 1000 data points, it is likely that the oximeter rounded up and rounded down a similar number of times, and so these differences will most likely cancel each other out. At most, the maximum differences due to the measurement of SpO2 in integers will account for less than 1% of the observed bias as compared with SaO2.
In conclusion, in a heterogeneous group of ICU patients, we showed that changes in pulse oximetry do not reliably predict equivalent changes in SaO2. We also demonstrated that neither anaemia nor acidosis alters the precision of measurements between the Nellcor pulse oximeter and CO-oximeter to any clinically important extent. The pulse oximeter remains a valuable tool in the care of intensive care patients, but an awareness of its limitations is an important component of enhancing the quality of care.
Changes in SpO2 do not reliably predict equivalent changes in SaO2 in the critically ill
Anaemia and acidosis have only a minor influence on the precision of measurements of SpO2 and SaO2
intensive care unit
arterial oxygen saturation
pulse oximeter oxygen saturation.
Jubran A: Advances in respiratory monitoring during mechanical ventilation. Chest 1999, 116: 1416-1425. 10.1378/chest.116.5.1416
Paky F, Koeck CM: Pulse oximetry in ventilated preterm newborns: reliability of detection of hyperoxaemia and hypoxaemia, and feasibility of alarm settings. Acta Paediatr 1995, 84: 613-616.
Inman KJ, Sibbald WJ, Rutledge FS, Speechley M, Martin CM, Clark BJ: Does implementing pulse oximetry in a critical care unit result in substantial arterial blood gas savings? Chest 1993, 104: 542-546.
Jubran A, Tobin MJ: Reliability of pulse oximetry in titrating supplemental oxygen therapy in ventilator-dependent patients. Chest 1990, 97: 1420-1425.
Slutsky AS: Consensus conference on mechanical ventilation: January 28–30, 1993 at Northbrook, Illinois, USA. Part I. European Society of Intensive Care Medicine, the ACCP and the SCCM. Intensive Care Med 1994, 20: 64-79.
Seguin P, Le Rouzo A, Tanguy M, Guillou YM, Feuillu A, Malledant Y: Evidence for the need of bedside accuracy of pulse oxime-try in an intensive care unit. Crit Care Med 2000, 28: 703-706. 10.1097/00003246-200003000-00017
Rotello LC, Warren J, Jastremski MS, Milewski A: A nurse-directed protocol using pulse oximetry to wean mechanically ventilated patients from toxic oxygen concentrations. Chest 1992, 102: 1833-1835.
Lee S, Tremper KK, Barker SJ: Effects of anemia on pulse oximetry and continuous mixed venous hemoglobin saturation monitoring in dogs. Anesthesiology 1991, 75: 118-122.
Jay GD, Hughes L, Renzi FP: Pulse oximetry is accurate in acute anemia from hemorrhage. Ann Emerg Med 1994, 24: 32-35.
Ramsing T, Rosenberg J: Pulse oximetry in severe anaemia. Intensive Care Med 1992, 18: 125-126.
Ralston AC, Webb RK, Runciman WB: Potential errors in pulse oximetry. I. Pulse oximeter evaluation. Anaesthesia 1991, 46: 202-206.
Bland JM, Altman DG: Comparing methods of measurement: why plotting difference against standard method is misleading. Lancet 1995, 346: 1085-1087. 10.1016/S0140-6736(95)91748-9
Bland JM, Altman DG: Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986, 1: 307-310.
Van de Louw A, Cracco A, Cert C, Harf A, Duvaldesin P, Lemaire F, Brochard L: Accuracy of pulse oximetry in the intensive care unit. Intensive Care Med 2001, 27: 1606-1613. 10.1007/s001340101064
Wukitsch MW, Petterson MT, Tobler DR, Pologe JA: Pulse oximetry: analysis of theory, technology, and practice. J Clin Monit 1988, 4: 290-301.
Severinghaus JW, Koh SO: Effect of anemia on pulse oximeter accuracy at low saturation. J Clin Monit 1990, 6: 85-88.
Ortiz FO, Aldrich TK, Nagel RL, Benjamin LJ: Accuracy of pulse oximetry in sickle cell disease. Am J Respir Crit Care Med 1999, 159: 447-451.
The Optical System In Reference Manual for ABL™700 Series. Copenhagen: Radiometer 2001.
Palve H, Vuori A: Accuracy of three pulse oximeters at low cardiac index and peripheral temperature. Crit Care Med 1991, 19: 560-562.
Secker C, Spiers P: Accuracy of pulse oximetry in patients with low systemic vascular resistance. Anaesthesia 1997, 52: 127-130. 10.1111/j.1365-2044.1997.32-az0062.x
Ibanez J, Velasco J, Raurich JM: The accuracy of the Biox 3700 pulse oximeter in patients receiving vasoactive therapy. Intensive Care Med 1991, 17: 484-486.
Watcha MF, Connor MT, Hing AV: Pulse oximetry in methemoglobinemia. Am J Dis Child 1989, 143: 845-847.
Bozeman WP, Myers RA, Barish RA: Confirmation of the pulse oximetry gap in carbon monoxide poisoning. Ann Emerg Med 1997, 30: 608-611.
Turner JA, McNicol MW, Sillett RW: Distribution of carboxyhaemoglobin concentrations in smokers and non-smokers. Thorax 1986, 41: 25-27.
Rucker J, Vesley A, Takeuchi A: Effect of ventilation on carbon monoxide clearance in humans. Am J Respir Crit Care Med 1999, 159: A767.
Veyckemans F, Baele P, Guillaume JE, Willems E, Robert A, Clerbaux T: Hyperbilirubinemia does not interfere with hemoglobin saturation measured by pulse oximetry. Anesthesiology 1989, 70: 118-122.
We thank Professor WW Mapleson for advice on statistics.
About this article
Cite this article
Perkins, G.D., McAuley, D.F., Giles, S. et al. Do changes in pulse oximeter oxygen saturation predict equivalent changes in arterial oxygen saturation?. Crit Care 7, R67 (2003). https://doi.org/10.1186/cc2339