| JOURNAL HOME | CME HOME | THIS MONTH | PAST ISSUES | ETOC | COLLECTIONS |
| AUTHORS | REVIEWERS | EDITORIAL BOARD | FEEDBACK | RSS | HELP |
| ||||||||||||||
| ||||||||||||||
|
|
|
|
||||||||||||
|
||||||||||||||
From the Masimo Corporation, Irvine, California.
Address correspondence and reprint requests to Michael T. Petterson, RRT, Masimo Corporation, 40 Parker, Irvine, CA 92618. Address e-mail to mpetters{at}masimo.com.
Abstract
Pulse oximetry is an important diagnostic and patient monitoring tool. However, motion can induce considerable error into pulse oximetry accuracy, resulting in loss of data, inaccurate readings, and false alarms. We will discuss how motion artifact affects pulse oximetry accuracy, the clinical consequences of motion artifact, and the methods used by various technologies to minimize the impact of the motion noise.
Pulse oximetry is an important diagnostic and patient monitoring tool widely used throughout the hospital environment. Although pulse oximetry is highly reliable when used on motionless and well-perfused patients, anything that affects the detection or processing of the biological signals on which it depends can affect the accuracy. Numerous factors have been shown to negatively affect the absorbance characteristics or signal-to-noise ratio of pulse oximetry. Factors affecting the absorbance characteristics include dyshemoglobins (carboxyhemoglobin and methemoglobin), dark skin pigmentation, nail polish, and dyes (1–7). A low signal-to-noise ratio leading to inaccurate readings can be caused by a low pulsatile signal (low perfusion), high noise (bright lights, electromagnetic interference, or motion), or a combination of the two (such as motion occurring during low perfusion) (5–12). Of all of these factors, motion artifact has been the most clinically troublesome, resulting in loss of data, inaccurate readings, and false alarms (5–7,9,11). Voluntary and involuntary movement, such as movement during transport, tapping, rubbing, scratching, waving, shivering, and seizures in adult and pediatric patients; and kicking, stretching, crying, flexing, and imposed motion in neonates, are some common sources of patient motion in the clinical setting (13, and see footnote1). Motion artifact can either cause the pulse oximeter to interpret motion as the true signal or obscure the true signal with noise, leading to inaccurate readings, false alarms, and most importantly, missed true alarms. Problems with pulse oximetry reliability, in turn, can result in an increase in caregiver workload, stress, and need for patient handling, all of which can lead to decreased patient safety and increased cost of care (14–17). Various methods have been used to attempt to minimize the impact of the motion noise, all for false-alarm management, including, most recently, algorithm-based motion rejection methods.
PRINCIPLES OF OXIMETRY
To understand how patient motion affects the accuracy and reliability of pulse oximetry, one must first understand the general principles of the technology. Takuo Aoyagi, the inventor of pulse oximetry, based his original technology on two physical principles: 1) the light absorbance of oxygenated hemoglobin is different from that of reduced hemoglobin at the two wavelengths used in pulse oximetry (red and infrared) and; 2) the absorbances at both wavelengths have a pulsatile, or oscillating (AC) component, which is the result of the volume change, normally from arterial blood, occurring between the emitter and the detector of the sensor (18). This volume change is most commonly due to pulsations from the cardiac cycle. There is also a nonpulsatile, or stable, component (DC) that results from light attenuated by skin, fingernails, tissue, bone, and static (or nonpulsating) blood. This nonpulsating blood is composed of variable amounts of arterial, venous, and capillary blood. All pulse oximeters have been empirically calibrated by desaturating healthy volunteers in an oxyhemoglobin range of 100% to 70%. By measuring oxyhemoglobin at numerous stable points within this range, a calibration curve can be generated by collecting absorption data for both the red and infrared wavelengths. These raw data are converted to a "ratio of ratios" (r) (Eq. 1), which is then associated with a specific SaO2 reading during the desaturation. For any computed r, therefore, there is an associated, estimated SaO2 reading from the calibration curve. The measurement of SaO2 by two-wavelength pulse oximetry has been termed Spo2 (19).
|
|
BASIC ASSUMPTIONS IN PULSE OXIMETRY
This empirically calibrated pulse oximeter is highly accurate when these basic assumptions are met:
As mentioned above, there are numerous factors that alter these assumptions, and therefore will decrease the accuracy of the device. The effects of most of these factors have been adequately described elsewhere (2–12). Here, we examine in more detail, the effect of motion on the accuracy of pulse oximetry.
EFFECTS OF MOTION ON PULSE OXIMETRY READINGS
Motion can induce considerable error into pulse oximetry accuracy. The mechanism of this has been poorly understood. Before the invention of read-through motion pulse oximetry, conventional wisdom was that motion was transmitted to the sensor introducing noise equally to both the red and infrared components of the signal, thus obscuring the biological signal. Some investigators even considered the forces required to dislodge the pulse oximetry sensor from the digit. Langton and Hanning, for example, correlated the force required to displace the sensor with the degree of motion-induced artifact (20). According to this theory, the motion-induced noise affected both the red and infrared wavelengths, and when the amplitude of the motion noise was large enough, it obscured the biological signal. The motion noise, therefore, resulted in an r ratio of approximately one, which corresponded to a saturation value between 82% and 85% in the empirical calibration curve (21,22). The problem with this theory is that false desaturations below 50% were commonly observed. It was eventually theorized that patient motion can cause movement of the venous blood, as well as other normally nonpulsatile components (such as tissue fluid in edematous patients), along with the arterial blood (21–24). The pulsatile components (AC components) of the signal, then, are composed of more than just arterial blood, which may lead to falsely low saturation readings (due to lower venous saturation). If motion is combined with low perfusion at the sensor site, then the venous blood makes a more significant contribution to the pulsatile component and drives the Spo2 reading even lower. Low perfusion is common in critically ill neonatal, pediatric, and adult patients, all of whom require accurate, continuous oxygenation monitoring. Low perfusion in the patient can corrupt the Spo2 reading in three ways: First, with low perfusion there is an increase in the ratio of venous blood to arterial blood at the measuring site. Second, the decreased perfusion results in increased oxygen extraction and, ultimately, a lower venous saturation. Third, the lower perfusion level is associated with a lower pulse amplitude or AC component, so the noise of motion can have a greater effect when combined with a small biologic signal. Therefore, small amounts of motion will cause the Spo2 to be composed of a larger venous component with a lower venous saturation resulting in lower Spo2 values.
CLINICAL IMPACT OF MOTION ON CONVENTIONAL OXIMETRY
Motion artifact can reduce the perceived clinical significance of pulse oximetry alarms by causing false alarms and data dropouts when the signal processing is overwhelmed by the motion noise. This "Cry Wolf" phenomenon was shown in a study designed to test the clinical significance of patient monitoring alarms in the pediatric intensive care unit (PICU). Lawless found that of the 957 pulse oximetry alarms that occurred in a 3-day trial period, 71% were false and only 7% were clinically significant (16). In another study conducted on 123 patients recovering from general or regional anesthesia, Wiklund et al. (25) found that of the 1516 pulse oximeter alarms during the 207 h of observation, only 23% were true. Fletcher et al. (26) found that in a group of preterm and term infants, motion artifact was present in all infant studies, comprising 19% of the monitored time during quiet sleep, 49% of active sleep, 49% of indeterminate sleep, and 91% of the time during wakefulness. In addition to affecting Spo2 measurements, motion artifact has been shown to affect the heart rate measurement by pulse oximetry as well. Barrington et al. (27) demonstrated in the neonatal intensive care unit (NICU) that the heart rate parameter was in error (error >10 bpm) as much as 25% to 29% of the time, and that this error was specifically related to the patient's movement of the extremities. The false alarms and loss of data that occur with motion artifact can have the effect of increasing caregiver workload by requiring them to repeatedly and needlessly check on the patient and the equipment. Understandably, this can cause caregivers to distrust the significance of the alarms, and may result in alarms being ignored or even turned off (16,28–30). Thus, motion artifact can significantly reduce the value of pulse oximetry as a tool to help clinicians maintain vigilance over their critically ill patients.
COLLECTING AND CHARACTERIZING MOTION DATA
In an effort to characterize the types of motion, which lead to oximetry errors for algorithm development, Tobin et al. (13) at Datex-Ohmeda (Louisville, CO) collected 14.5 h of data on 35 patients who exhibited motion in the different critical care areas of the hospital including the intensive care unit, PICU, NICU, operating room, and ambulance. They found that, although a wide range of motion types led to oximetry error, most errors were generated by intense, aperiodic, random movements that lasted 30 s or less. However, in approximately 5% of the cases, motion exceeded 1 min. Infant patients demonstrated these types of motion more than adults and, in fact, were much more likely to be observed moving compared with adult patients (31% compared with 7%, respectively). Masimo's clinical research data collection set, which includes thousands of hours of data collected from more than 1000 patients in all clinical areas, is consistent with these findings, showing that patient motion tends to be composed of discrete, aperiodic episodes occurring closely together, so that their effect on pulse oximetry would appear as continuous motion, often lasting more than several minutes at a time. In addition, the Masimo clinical data set contains numerous files of continuous motion exceeding 30 min such as shivering patients and those with imposed motion during ground or air transport. Therefore, motion-resistant pulse oximetry has to be able to work during short periods of motion, as well as for long continuous periods of both aperiodic and periodic motion.
Solutions to the Noise Problem
As previously discussed, continuous monitoring with pulse oximetry is necessary for all acute care patients, but use of conventional oximetry often results in too many false alarms, diminishing its predictive value and impairing patient safety by distracting caregivers. In order for pulse oximetry to be more help than hindrance to the health care provider, the motion artifact problem needed to be improved. Medical device engineers have used various techniques to address motion artifact in pulse oximetry. Commonly used strategies include averaging the saturation data over a longer period of time, holding data until clean data are qualified, then averaging that with the previous clean data and implementing alarm delays. Additionally, with the advent of read-through motion and other kinds of motion-resistant technologies, sophisticated algorithms, some using parallel processing with multiple algorithms, have been used to isolate the biological signal in the presence of noise.
Data Averaging, Alarm Delay, and Holding Data
Technologies that smooth data with longer averaging algorithms to reduce the effect of motion artifact can result in fewer false alarms. Figure 1 demonstrates the theoretic effect of increasing averaging time. Rheineck-Leyssius and Kalkman tested the effect of data averaging on false-alarm rates in a study conducted on 200 postsurgical patients (31). Spo2 values for each patient were recorded with a laptop computer, and then analyzed to identify episodes of low Spo2 and motion artifact. The data were then subjected to an algorithm that simulated data averaging settings of nine different durations between 10 and 90 s. The study found that, when Spo2 values were subjected to 10 s averaging (default 3 s averaging), alarms were reduced by almost 50%. When the data were averaged for 42 s, the alarm rate was reduced further by 82%, but 6 of the 73 true severe hypoxemic episodes were not detected. In a more recent study, designed to compare the sensitivity and specificity of several pulse oximetry technologies during motion and low perfusion, Barker found that technologies that had longer default averaging times of 10 and 12 s had a decreased ability to detect rapidly occurring hypoxemias, compared with those with 5 to 8 s averaging (32). Data averaging, therefore, has two effects: first, it increases the amount of time needed to report an initial value. Second, it has the effect of smoothing any resultant changes in oxyhemoglobin saturation, which, in turn, can directly impact the diagnosis and treatment of patients. The diagnosis of obstructive sleep apnea (OSA), for example, is dependent on the tracking of transient desaturations in the patient. Figure 2 shows data from an infant diagnosed with OSA of prematurity. Frequent OSA episodes result in rapid transient desaturations, the magnitudes of which are clearly detected when the pulse oximeter is set for 2-s averaging. These transient desaturations are almost completely lost when the device is set for 16-s averaging.
|
|
Rheineck-Leyssius and Kalkman formally investigated the effect of eliminating alarms for transient episodes of low Spo2 readings by introducing a delay between the onset of the alarm condition (a reading of <90% Spo2) and the triggering of the alarm. This time delay strategy was compared with the effect of decreasing the alarm condition to 85% to determine which was more effective at eliminating false alarms. The researchers found that decreasing the alarm threshold from 90% to 85% would eliminate the same number of false alarms as delaying the triggering of the alarm 15 s from alarm condition onset (31). However, the incidence of hypoxemias lasting longer than 1 min in patients for whom the alarm threshold was set at 85% was higher than in patients for whom the alarm threshold was set at 90%, suggesting that the time delay strategy was the safer method of reducing false alarms. Neither method, however, provides a safety net for patients suffering from rapid desaturations. Poets and Southall found that preterm infants can desaturate at a rate of up 12.6% per second (33). Thus, an alarm delay of even 15 s could prevent a patient from receiving time-critical, lifesaving intervention.
Manufacturers have also used data holding strategies to decrease false alarms. With this technique, pulses are qualified, and only those fitting certain criteria are used in the calculation of Spo2. When motion is sufficient to affect the Spo2 reading, the current Spo2 is displayed until new clean data are qualified and added to the previous buffer of good data. The holding of data may continue for up to 50 s if good data are not found (34). This technique has been shown to miss significant desaturations in both volunteer motion studies and clinical studies (32,35,36). By using pulse oximeters that use any of these strategies, clinicians are left to pay for the reduction in false alarms at the cost of lost information on their patients' oxygenation status. It is important to note that current International Organization for Standardization (ISO) standards for pulse oximeters require data to be updated at least every 30 s (37).
Read-Through Motion and Motion Tolerant Algorithms
Several manufacturers of pulse oximetry systems reference the use of algorithms that are motion tolerant in their marketing literature. Because these algorithms are highly proprietary, the details on how each manufacturer's technology identifies and processes the incoming signals are generally not available. Signal-processing algorithms for conventional pulse oximeters were typically time-domain based, using analog filtering and moving average techniques to identify and process biological signals (4,23,24). Masimo SET, a technology that reads through motion, uses several signal processing techniques. Masimo's proprietary Discrete Saturation Transform (DST), one of the several algorithms used in Masimo SET, is an example of a time domain algorithm. The DST algorithm was developed specifically to identify the arterial signal in the presence of the nonarterial (venous) signal that occurs during motion (22,23,38,39). The DST comprises a reference signal generator, an adaptive filter and a peak picker, which work in concert to determine the most likely Spo2 value based on the incoming signals. The reference signal generator uses a combination of the red to infrared ratios and the detected signals to produce a series of noise reference signals. The noise reference signals (serially from 0% to 100% Spo2) are then compared with the incoming signal using an adaptive filter. A power spectrum is created by plotting the power output as a function of Spo2, from 0% to 100%. The right-most peak (identified by the peak picker) of the power spectrum plot during motion corresponds to the highest saturation value, which should be the arterial saturation, because arterial saturation is higher than venous saturation. During nonmotion conditions, there is only one peak. The output of one or more of the other algorithms may also be evaluated by a confidence-based arbitrator algorithm, which produces a measure of confidence about the quality of the incoming data.
Philips' FAST Spo2 (Fourier Artifact Suppression Technology) is a technology that makes motion-tolerant claims and depends on a frequency-based algorithm (40). According to a published description, the FAST-Spo2, algorithm first identifies the frequency components of the pulse rate and compares those to the frequency components of the incoming signal to select the component that is at the pulse rate for both the red and infrared wavelengths. This component, or value, is then used to calculate the oxygen saturation after several additional "checks" are made. These include determining if the frequency components are multiples of the selected pulse rate, if there is a good correlation between the red and infrared spectra at the selected frequency and if the calculated Spo2 value is reasonable and close to the last three accepted values (41). Nellcor's Oximax N-600, on the other hand, which uses the "Variable Cardiac Gated Averaging" algorithm, appears to be time-domain-based, in that it attenuates incoming signals that do not occur synchronously with the average rhythm of the pulse rate and allows the parts of the waveform that are synchronous with the heart rate to remain unattenuated and thus contribute more to the calculated Spo2 (42).
Evaluation of Motion-Resistant Pulse Oximetry Technologies
There are many published studies comparing the performance of motion-resistant pulse oximetry technologies to conventional technologies and to each other. These studies can be categorized into three types; 1) those conducted in a laboratory setting on healthy volunteers, 2) clinical studies of hospitalized adults, pediatric, or neonatal patients usually in intensive care units, and 3) OSA studies.
Laboratory studies tend to use either "hand motion machines" or depend on subject-generated motion. Protocols that use machine-generated motion have the advantage of testing devices with standardized, highly consistent, and reproducible motions. The pros and cons of testing devices with subject-generated versus machine-generated motion have been debated (43–46). Some researchers have studied tapping and rubbing motions using both techniques and found there to be little or no differences in the results (46–48). All laboratory studies have shown that read-through motion pulse oximeters are superior to conventional pulse oximetry (49). A search for published literature on laboratory pulse oximetry motion studies (excluding abstracts more than 3-yr-old, manufacturer supported or produced studies, reviews and case studies) yields 12 abstracts (47–58) and five papers (59–63) that specifically compare next generation technologies for motion performance. Twelve of the studies, which come from two research laboratories, use similar protocols and show similar results with one manufacturer's technology consistently outperforming other technologies on variables such as failure rate (3 studies), missed events (5 studies), false alarms (4 studies), sensitivity (3 studies), and specificity (3 studies). The results of the remaining five studies, which used unique protocols, were generally mixed with no single technology performing the best on all variables. It should be noted that not all manufacturers were equally represented in the studies. Seventeen of the studies compared some version of Masimo SET, eight studies compared some version of a Nellcor, eight compared Datex Ohmeda devices, and six compared Philips devices.
Although clinical studies have variables that may be harder to control compared with laboratory studies, they have the advantage of using real patients in the hospital setting, and therefore may provide a "truer" test of device performance. Of the various clinical studies published on pulse oximeter performance, those conducted on patients in the NICU and PICU are the most relevant to motion testing for reasons previously discussed. Because of the unique variables in clinical studies, however, it is difficult to combine findings across studies to draw conclusions regarding the relative performances of motion-tolerant pulse oximeter technologies. Additionally, many clinical studies did not provide rigorous enough test conditions to distinguish among technologies. One generalization that can be made however is that, as with the laboratory studies, clinical studies designed to compare the performance of conventional oximetry to read-through motion technologies consistently report fewer false alarms and data dropouts with the new technologies (14,15,17,35,36,59,64,65).
Overnight polysomography is the most accurate and comprehensive method for recording respiration during sleep and is the "gold standard" for diagnosing OSA in children and adults. In a study designed to determine how often pulse oximetry readings were corrupted by motion artifact during sleep, Fletcher et al. (26) calculated the percentage of time that motion artifact was present in recordings of sleep in term and preterm infants. The study found that motion artifact affects more than 50% of recorded traces. Although the NICU may have a sicker population, the sleep laboratory has more instrumentation to establish respiratory movement and airflow patterns, which directly affect oxygenation. Thus, the pediatric sleep laboratory would seem to be a perfect laboratory for pulse oximetry testing. Trang et al. (66) compared a conventional pulse oximeter to a read-through motion pulse oximeter to evaluate which was better at detecting sleep desaturations. The study, conducted on 34 children admitted to a sleep clinic for possible sleep disordered breathing, found that the read-through motion pulse oximeter detected far more true desaturations and had far fewer false desaturations than the conventional technology. In a 2002 study, Brouillette et al. (14) compared the performance of read-through motion pulse oximeters with standard conventional pulse oximetry technology during pediatric OSA studies. They found that although both read-through motion devices had far fewer false alarms than the conventional pulse oximeter, one read-through motion technology was superior in tracking actual desaturations than the other. Studies conducted on adult sleep laboratory patients have shown similar results to these infant sleep studies (67). Thus, regardless of patient population, the clinical sleep studies consistently show that the newer read-through motion technologies reduce false alarms and that some also reliably track true desaturations, especially during OSA testing.
Does Improved Technology Make a Difference?
In a landmark study, Durbin and Rostow showed that caregivers learned to use read-through motion technology to improve the care of postoperative open-heart surgery patients. Compared with conventional pulse oximetry, clinicians were able to significantly reduce the number of arterial blood gas analyses and time to wean from 100% to 40% O2 when using read-through motion technology (15). Chow et al. (68) credit read-through motion technology as being instrumental in their protocol to reduce blindness and vision problems from retinopathy of prematurity in preterm neonates. Brouillette et al. (14) found that a read-through motion pulse oximeter was able to reduce workload and improve reliability of desaturation detection compared with conventional pulse oximetry. Two separate studies used different motion-tolerant technologies to accurately screen for congenital heart disease in infants (69,70). One of these studies shows that while a read-through motion pulse oximeter had high sensitivity and specificity, thus making it a very useful tool for this type of needed screening, the false desaturations caused by motion artifact made the conventional technology unusable (69). Routine use of pulse oximetry has proven to be highly valuable and carries very little risk to the patient, except when data dropout and false alarms reduce the quality of care. Motion-tolerant technologies have reduced the dropout and false-alarm rates, thus improving patient care without the risks associated with conventional oximetry.
Summary
A wide variety of studies have been published on both the usefulness and potential problems with pulse oximetry in the clinical environment. Most clinicians would agree that motion artifact and the resulting false alarms have been the most significant drawback to using pulse oximetry in the critical care setting. Motion artifact results in a low signal-to-noise ratio, with Spo2 being driven to lower than actual readings due to venous motion. This venous component is exacerbated by low perfusion. Alarm management in the past tried to deal with these issues by longer averaging, holding data, or alarm delays. Read-through motion and motion-tolerant technologies have dealt with the root cause of motion artifact, thus achieving more reliable readings. To be a useful clinical tool, pulse oximeters should give real-time, continuous, and accurate measurements over a wide range of saturation values, during all types of patient motion, continuous and intermittent, aperiodic and rhythmic, and during low perfusion. There appear to be significant differences in the abilities of the currently available motion-tolerant technologies to achieve these goals, but all motion-tolerant technologies appear to perform better than conventional technologies.
Footnotes
1Masimo's clinical data set contains data for over 1000 patients from all clinical areas. 
Accepted for publication May 7, 2007.
REFERENCES
This article has been cited by other articles:
|
|
 |
|
  M. Gorges, B. A. Markewitz, and D. R. Westenskow Improving Alarm Performance in the Medical Intensive Care Unit Using Delays and Clinical Context Anesth. Analg., May 1, 2009; 108(5): 1546 - 1552. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
