The human eye is poor at recognizing hypoxemia. Even under idealconditions, skilled observers cannot consistently detect hypoxemia until theoxygen (O₂) saturation is below 80% [1]. Thedifficulty that physicians have in detecting hypoxemia was recently exemplifiedin a study of over 14000 patients being evaluated at the UCLA EmergencyDepartment [2]. Patients were monitored by oximetry butrecordings were given to physicians only after they completed their initialassessment. Changes in diagnostic testing and treatment were most likely at anO₂ saturation of 89%, and changes were actually less common at lowersaturations, probably because the physicians were able to detect evidence ofhypoxemia without requiring a pulse oximeter.

With the proliferation of pulse oximeters in different locations ofthe hospital throughout the 1980s, several investigators demonstrated thatepisodic hypoxemia is much more common than previously suspected with anincidence ranging from 20-82% [3,4,5] (Fig. 1). Thesignificance of episodic desaturation on patient outcome is largely unknown[6]. In patients admitted to a general medical service,Bowton et al. [7] found that O₂saturation < 90% of at least 5 min duration occurred in 26% of the patients.On follow-up over the next 4-7 months, those patients experiencing hypoxemiaduring the first 24 h of hospitalization had more than a threefold highermortality than patients who did not desaturate. Although episodic desaturationmay simply be a marker of increased risk rather than the direct cause ofdecreased survival, an increased mortality rate was still observed in patientswith episodic hypoxemia when the investigators corrected for severity ofillness. Whether or not the early detection and treatment of episodic hypoxemiacan affect patient outcome remains unknown.

Pulse oximetry is based on two physical principles: (a) the presenceof a pulsatile signal generated by arterial blood, which is relativelyindependent of non-pulsatile arterial blood, venous and capillary blood, andother tissues; and (b) the fact that oxyhemoglobin (O₂Hb) andreduced hemoglobin (Hb) have different absorption spectra [8]. Currently available oximeters use two light-emitting diodes (LEDs) that emit light at the 660 nm (red) and the 940 nm (infrared) wavelengths.These two wavelengths are used because O₂Hb and Hb have differentabsorption spectra at these particular wavelengths. In the red region,O₂Hb absorbs less light than Hb, while the reverse occurs in theinfrared region. The ratio of absorbencies at these two wavelengths iscalibrated empirically against direct measurements of arterial blood oxygensaturation (S_aO₂) in volunteers, and theresulting calibration algorithm is stored in a digital microprocessor withinthe pulse oximeter. During subsequent use, the calibration curve is used togenerate the pulse oximeter's estimate of arterial saturation(S_pO₂) [9,10] (Fig. 2). In addition to the digitalreadout of O₂ saturation, most pulse oximeters display aplethysmographic waveform which can help clinicians distinguish an artifactualsignal from the true signal (Fig. 3).

The accuracy of commercially available oximeters differ widely,probably because of the different algorithms employed in signal processing[8]. These algorithms are limited by the range ofsaturations that can be safely obtained in volunteers, and also the accuracy ofthe measurement standard (CO-oximeter) [11]. Comparisonof pulse oximetry with direct CO-oximeter measurements should be reported interms of the mean difference between the two techniques (bias) and the standarddeviation of the differences (precision).

In healthy volunteers, oximeters commonly have a mean difference(bias) of < 2% and a standard deviation (precision) of < 3% whenS_aO₂ is 90% or above [12,13]. Comparable results have also been obtained in critically ill patients with good arterial perfusion [14,15]. Accuracy of pulseoximetersdeteriorates when S_aO₂ falls to 80% or less. Inhealthy volunteers under hypoxic conditions, bias of pulse oximetry varies from-15.0 to 13.1 while the precision ranges from 1.0 to 16.0 [12,16,17,18]. In a study in critically ill patients, eight out of 13oximeters had a bias ≥ ± 5% when S_aO₂was < 80% [14]. In a study of 54 ventilator-dependentpatients, the accuracy of oximetry deteriorated significantly at lowS_aO₂ values. Bias ± precision was 1.7 ± 1.2% for S_aO₂ values > 90%, and it increased to 5.1 ± 2.7% when S_aO₂ was ≤ 90% [19].

Different probes that are used with a pulse oximeter can also affectthe accuracy of S_pO₂ measurements. Inpatientswith poor peripheral perfusion as a consequence of car-diopulmonarybypass surgery, finger probes had lower precision and more readings within 3%of the reference (CO-oximeter) than the other probes. Overall rankings weresignificantly better for the finger probes than probes on other sites (Fig.4) [20]. The response time ofoximeter probes was assessed by Severinghaus and Naifeh [17] who induced 30-60s hypoxic plateaus between anS_aO₂ of 40 and 70% in healthy volunteers.Oximeter probes placed on the ear generally had a much faster response to asudden decrease in fractional inspired oxygen concentration(F_iO₂) than did the finger probes (10-20 versus24-35s, repectively). Employing hypobaric facility to induce hypoxia in normalvolunteers, Young et al. [21] also observedthat the response time of the finger probes were slower than the ear probes inresponse to either a decrease or increase in O₂ saturation.

Oximeters have a number of limitations which may lead to inaccuratereadings (Table 1). Pulse oximeters measureS_aO₂ that is physiologically related to arterialoxygen tension (P_aO₂) according to the O2Hbdissociation curve. Because the O₂Hb dissociation curve has asigmoid shape, oximetry is relatively insensitive in detecting the developmentof hypoxemia in patients with high baseline levels ofP_aO₂ [11,22].

Pulse oximeters employ only two wavelengths of light and, thus, candistinguish only two substances, Hb and O₂Hb. When carboxyhemoglobin(COHb) and methemoglobin (MetHb) are also present, four wavelengths arerequired to determine the 'fractional SaO₂': i.e.,(O₂Hb × 100)/(Hb + O₂Hb + COHb + MetHb). In the presence ofelevated COHb levels, oximetry consistently over- estimated the trueS_aO₂ [23,24] by the amount of COHb present. Elevated MetHb levels also may cause inaccurate oximetry readings [25,26]. Anemia does not appear to affect the accuracy of pulse oximetry: in non-hypoxemic patients with acute anemia (mean Hb, 5.2 ± 0.3(SE) g/dl), pulse oximetry was accurate in measuring O₂ saturationwith abias of only 0.53% [27]. However, in patients withsickle cell anemia presenting with acute vaso-occlusive crisis [28], mean bias of pulse oximetry was 4.5% (in some patients it was as high as 8%), which was significantly greater than in a control group ofpatients without sickle cell anemia. Severe hyperbilirubinemia (mean bilirubin,30.6 mg/dl) does not effect the accuracy of pulse oximetry [29].

Intravenous dyes such as methylene blue, indocyaninegreen, and indigocarmine can cause falsely low S_pO₂ readings[30], an effect that persists for up to 20 min [31]. Nail polish, if blue, green or black, causes inaccurate S_pO₂ readings [32],whereas acrylic nails do not interfere with pulse oximetry readings [33]. Falsely low and high S_pO₂ readings occur with fluorescent and xenon arc surgical lamps [34].

Motion artifact continues to be a significant source of error andfalse alarms [35,36,37,38]. In a recent, prospective study in an intensive care unit (ICU) setting, S_pO₂signals accounted for almost half of a total of 2525 false alarms [39] (Fig. 5). In 123 patients recoveringfrom general or spinal-epidural anesthesia, 77% of pulse oximeter alarms werefalse in nature, which the investigators attributed to sensor displacement,motion artifact, and a decrease in skin perfusion [40].In this study, the alarm threshold was set at anS_pO₂ of 90% and it is not clear if a minimumduration was specified. A recent study in 647 patients in the recovery roomcompared the influence of two pulse oximeter lower alarm limit settings(S_pO₂ 90% = group 90 andS_pO₂ 85% = group 85) on the incidence ofhypoxemia[41]. Although the number of audible alarms waslower in group 85, hypoxic episodes (defined asS_pO₂ ≤ 90% lasting > 1 min) weremore common in group 90 than in group 85 (11 versus 6%, respectively). Theinvestigators concluded that decreasing the alarm limit to reduce false alarmsmay lead to increase in more relevant episodes of hypoxemia.

Various methods have been employed to reject motion artifact but havemet with little success [8,42,43]. An innovative technological approach, termed Masimo signal extraction technology (SET^™;Masimo Corporation, Mission Viejo, California, USA), was recently introduced toextract the true signal from artifact due to noise and low perfusion [44].This technique incorporates new algorithms for processing the pulseoximeter's red and infrared light signals that enable the noisecomponent, which is common to the two wavelengths, to be measured andsubtracted. When tested in healthy volunteers during standardized motion,Masimo SET^™ exhibited much lower error rates (defined aspercentage of time that the oximeter error exceeded 5%, 7%, and 10%) anddropout rates (defined as the percentage of time that the oximeter provided noS_pO₂ data) than did the Nellcor N-200 andNellcor N-3000 oximeters (Nellcor Puritan Bennett, Pleasanton, California, USA)for all test conditions [45]. The lowest performanceindex (defined as the percentage of time that the oximeter's value waswithin 7% of the control S_pO₂ value) was 97% forMasimo SET^™ comparedwith 47% for the N-3000 and 68% for theN-200. In 50 postoperative patients, Dumas et al [46] observed that a pulse oximeter's alarm frequency wasdecreased twofold with a Masimo SET^™ system versus aconventional oximeter (Nellcor N-200). Improved performance was particularlystriking during conditions of gross (non-rhythmic) motion and tremor, when a22-fold reduction in signal loss over time was observed (Fig. 6).

Inaccurate oximetry readings have been observed in pigmented patients,but not by all investigators [8]. In 33 healthy blacksubjects during normoxia and hypoxia, the correlation betweenS_pO₂ and S_aO₂ wasinferior with a Biox IIA oximeter (Ohmeda, Boulder, Colorado, USA) (r= 0.80) than with the older Hewlett-Packard (Waltham, Massachusetts, USA)(non-pulse) oximeter (r = 0.94) [47]. Incritically ill patients [19], bias ± precision wasgreater in black patients, 3.3 ± 2.7%, than in white patients, 2.2 ±1.8%; also, a bias > 4% occurred more frequently in black patients (27%) thanin white patients (11%).

Low perfusion states, such as low cardiac output, vasoconstriction andhypothermia, may impair peripheral perfusion and may make it difficult for asensor to distinguish a true signal from background noise. In cardiac surgerypatients experiencing hypothermia and poor perfusion, only two of 20 oximeters(Criticare CSI 503, Criticare Systems, Inc., Milwaukee, Wisconsin, USA; DatexSatlite, Datex Instrumentarium Corp., Helsinki, Finland) provided measurementswithin ± 4% of the CO-oximeter value [48].Measurements of SpO₂ with a Biox 3700 oximeter had a bias > ± 4% in 37% of patients receiving vasoactive therapy [49].

An under-recognized and worrisome problem with pulse oximetry is thatmany users have a limited understanding of how it functions and theimplications of its measurements. In a recent survey [50], 30% of physicians and 93%of nurses thought that theoximeter measured P_aO₂. Some clinicians alsohave a limited knowledge of the O₂-dissociation curve,and^{they do not recognize that S_pO₂values in the high 80s represent seriously low values ofP_aO₂. In the above survey, some doctors andnurses were not especially worried about patients withS_pO₂ values as low as 80% (equivalent toPaO₂ ≤ 45 torr).}

Bierman et al. [4] reported that fewerarterial blood gas (ABG) samples were obtained in cardiac surgery patientsifS_pO₂ data were available to the caregivers.Interestingly,the availability of oximetry data had no effect on the durationof ICU stay, duration of mechanical ventilation, or the need for supplementalO₂. In an emergency department, a recent report showed that thenumber of unjustified ABGs (as determined by independent experts) over a2-month period decreased from 29% when pulse oximetry was unavailable to 12%when oximetry was available; the number of justified ABGs did not change [58].

Solsona et al. [59] measured thenumber of blood gas measurements in 417 patients admitted to a medical-surgicalICU during a 12-month period in which only two pulse oximeters were available(i.e., control period). They then studied 306 patients admitted over a 9-monthperiod when 12 pulse oximeters were available for the same number of beds(i.e., intervention period). Less frequent use of mechanical ventilation and aslightly lower number of arterial blood samples were observed when pulseoximetry was fully available. Inman et al. [60]examined the effect of implementing pulse oximetry without any specificalgorithm for its appropriate use. They studied 148 patients before theimplementation of oximetry in their ICU and 141 patients after itsimplementation. The number of ABG samples decreased from 7.2 to 6.4 per patientper day, a reduction of only 10.3% compared with average reductions of 39% inthe previous studies [4,61]. Thissuggests that, without explicit guidelines, the pulse oximeter was used inaddition to, rather than instead of, ABG samples.

Pulse oximetry is probably one of the most important advances inrespiratory monitoring. Over the last 15 years, numerous studies have focusedon the technical aspects of pulse oximeters and found that these instrumentshave a reasonable degree of accuracy. This degree of accuracy, coupled with theease of operation of most instruments, has led to the widespread use of pulseoximetry for monitoring patients in the ICU. Perhaps the major challenge facingpulse oximetry is whether this technology can be incorporated effectively intodiagnostic and management algorithms that can improve the efficiency ofclinical management in the intensive care unit.