- Email: [email protected]
An evaluation of pulse oximeters in dogs, cats and horses
Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
R E S E A R C H PA P E R
An evaluation of pulse oximeters in dogs, cats and horses Nora S Matthews
DVM, Diplomate ACVA,
Sherrie Hartke,
DVM
y
and John C Allen Jr.
MS
z
Department of Small Animal Medicine and Surgery,Texas A&M University, College Station,TX,
y
North Shore Animal League ^ America,The Alex Lewyt Medical Center, Lewyt Street, Port Washington, NY, z Tyco Healthcare, Mallinckrodt Inc., Hazelwood, MO, USA
Correspondence: Nora S Matthews, Department of Small Animal Medicine and Surgery,Texas A&M University, College Station,TX 77843-4474, USA. E-mail: [email protected].
Abstract Objective Evaluation of ¢ve pulse oximeters in dogs, cats and horses with sensors placed at ¢ve sites and hemoglobin saturation at three plateaus. Study design Prospective randomized multispecies experimental trial. Animals Five healthy dogs, cats and horses. Methods Animals were anesthetized and instrumented with ECG leads and arterial catheters. Five pulse oximeters (Nellcor Puritan Bennett-395, NPB190, NPB-290, NPB-40 and Surgi-Vet V3304) with sensors at ¢ve sites were studied in a 5 5 Latin square design. Ten readings (SpO2) were taken at each of three hemoglobin saturation plateaus (98, 85 and 72%) in each animal. Arterial samples were drawn concurrently and hemoglobin saturation was measured with a co-oximeter. Accuracy of saturation measurements was calculated as the root mean squared di¡erence (RMSD), a composite of bias and precision, for each model tested in each species. Results Accuracy varied widely. In dogs, the RMSD for the NPB-395, NPB-190, NPB-290, NPB-40 and V3304 were 2.7, 2.2, 2.4, 1.7 and 2.7% respectively. Failure to produce readings for the NPB-395, NPB190, NPB-290, NPB-40 and V3304 were 0, 0, 0.7, 0, and 20%, respectively.The Pearson correlation coe⁄cients for the tongue, toe, ear, lip and prepuce or vulva were 0.95, 0.97, 0.69, 0.87 and 0.95, respectively. In horses, the RMSD for the NPB-395, NPB-190,
NPB-290, NPB-40 and V3304 were 3.1, 3.0, 4.7, 3.3 and 2.1%, respectively while rates of failure to produce readings were10, 21,0,17 and 60%, respectively. The Pearson correlation coe⁄cients for the tongue, nostril, ear, lip and prepuce or vulva were 0.98, 0.94, 0.88, 0.93 and 0.94, respectively. In cats, the RMSD for all data for the NPB-395, NPB-190, NPB-290, NPB-40 and V3304 were 5.9, 5.6, 7.9, 7.9 and 10.7%, respectively while failure rates were 0, 0.7, 0, 20 and 32%, respectively. The correlation coe⁄cients for the tongue, rear paw, ear, lip and front paw were 0.54, 0.79,.0.64, 0.49 and 0.57, respectively. For saturations above 90% in cats, the RMSD for the NPB-395, NPB190, NPB-290, NPB-40 and V3304 were 2.6, 4.4, 4.0, 3.5 and 4.8%, respectively, while failure rates were 0, 1.7, 0, 25 and 43%, respectively. Conclusions and clinical relevance Accuracy and failure rates (failure to produce a reading) varied widely from model to model and from species to species. Generally, among the models tested in the clinically relevant range (90^100%) RMSD ranged from 2^5% while failure rates were highest in theV3304. Keywords pulse oximetry, saturation, dogs, cats, horses.
Introduction Pulse oximetry is a common method of monitoring the saturation of hemoglobin with oxygen noninvasively in veterinary patients. Various probe sites and di¡erent models of pulse oximeters have been evaluated in dogs (Fairman 1993; Huss et al. 1995; 3
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
Matthews et al. 1995; Barton et al. 1996; Grosenbaugh & Muir 1998), horses (Cha⁄n et al. 1996) and birds (Schmitt et al.1998). However, newer models of pulse oximeters have become available and no information is published about their accuracy at varying sites and during desaturation. Therefore, the purpose of this study was to evaluate several newer models of pulse oximeters when probes were placed at various sites in three species (dogs, cats and horses) over a range of hemoglobin saturation.
assigned the ¢ve pulse oximeters to the ¢ve probe sites. Two calibrated oximeters of each model were tested on each animal in one of the ¢ve probe sites. Sensor application sites, except the tongue, were shaved prior to sensor placement and sensors were covered, after they were in place, to reduce ambient light. Anesthesia, monitoring and blood sample collection Dogs
Materials and methods Study design All procedures were approved by theTexas A&M University LaboratoryAnimal Care Committee. Five animals of each species (dogs, cats and horses) were anesthetized and instrumented with a pulse oximeter probe at ¢ve di¡erent sites, each attached to one of ¢ve di¡erent pulse oximeters. A catheter was placed into a peripheral artery to permit periodic blood sampling. Arterial samples were drawn at estimated plateaus of 98, 85 and 72% saturation. The saturation plateau was maintained for 10 minutes before drawing samples and taking readings. At 1minute intervals and over at least one complete respiratory cycle, ¢ve arterial samples were drawn at each plateau at the same time that pulse oximeter readings and heart rate were recorded. Animals were resaturated while probes were changed to duplicate monitors and the procedure was repeated. Functional hemoglobin saturation was determined in duplicate, with a co-oximeter (Radiometer OSM-3 Hemoximeter. Radiometer America, Inc., Westlake, OH) which was set in the appropriate mode for each species and calibrated before each animal was anesthetized. Functional saturation was calculated (by the co-oximeter) using the standard formula: % functional SaO2 % O2 Hb 100 ¼ ð100 ð% metHb þ % CoHbÞÞ Following the second period of desaturation and readings, all animals were resaturated, then recovered from anesthesia. Randomization and sensor placement A computer-generated random number scheme, based on a Latin square experimental design, 4
One male and four female mixed-breed dogs, ranging from 6 to11 years in age (mean ¼ 8.8 years) and from 12 to 25 kg (mean ¼ 16 kg) in weight were used in the study. The dogs were healthy (based on physical examination and regular hematology and biochemical testing) and were heartworm negative. Food, but not water, was withheld the night before anesthesia. No pre-anesthetic medications were given and anesthesia was induced with propofol (2.8^7.3 mg kg^1 IV, mean ¼ 5.2) (Propo£o, Abbott Laboratories, Chicago, IL, USA). Following endotracheal intubation, dogs were positioned in lateral recumbency, connected to a circle system (North American Drager Small Animal Anesthesia machine, Telford, PA, USA), and maintained with iso£urane (Iso£o. Abbott Laboratories, Chicago, IL, USA) in 100% oxygen, at approximately1.5 MAC (measured at the proximal end of the endotracheal tube) (Capnomac Ultima. Datex Medical Instrumentation, Tewksbury, MA, USA). Dogs were ventilated at a rate of 11^14 breaths minute ^1 and a tidal volume (8^15 mL kg^1, mean ¼ 11 mL kg^1) needed to maintain PaCO2 in the range of 35^45 mm Hg (4.7^6 kPa). A blood gas sample was drawn at the beginning (after ventilation was established) and end of anesthesia and analyzed immediately (IRMA, Diametrics Medical, St. Paul, MN, USA). Blood pressures were measured noninvasively (Dinamap Veterinary Blood Pressure Monitor 8300, Critikon Inc., Tampa, FL, USA) with the cu¡ placed on a forearm and a lead II ECG was monitored (Nellcor-PB 3940. Mallinckrodt Inc., Pleasanton, CA, USA). An arterial catheter (20 SWG, 3-cm Te£on catheter) was placed percutaneously in a dorsal metatarsal artery for collection of blood samples. All dogs received lactated Ringer’s solution (10 mL kg^1 hour^1) administered through a catheter placed in a cephalic vein and body temperature (measured rectally) was maintained with a forced air warming blanket (Bair Hugger Warming Unit, Model 505. Augustine Medical Inc., Eden Prairie, MN, USA). Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
The pulse oximeters studied were the NPB-40, NPB-190, NPB-290, and NPB-395 (Nellcor oximetry. Mallinckrodt Inc., Pleasanton, CA, USA) connected to a veterinary sensor (Vet Sat. Nellcor oximetry. Mallinckrodt Inc., Pleasanton, CA, USA) and the V3304 (Surgi-Vet Inc., Waukesha, WI, USA) connected to a veterinary sensor (V1703, Surgi-Vet Inc., Waukesha, WI, USA). The ¢ve probe sites were the tongue, lip, ear (placed on the lateral edge and close to the base), toe, and prepuce or vulva. Following instrumentation, the inspired oxygen concentration was reduced (by mixing nitrogen into the circuit through a separate £owmeter) until the 98% saturation plateau was reached, based on readings from the pulse oximeters. One-milliliter blood samples were drawn into heparinized syringes, at 1minute intervals as the pulse oximeter readings and heart rates were recorded simultaneously. Blood samples were drawn over one complete respiratory cycle, waste blood was drawn before sampling and the catheter was £ushed after sampling. The oxygen^nitrogen mixture was then adjusted to reach the 85 and 72% plateaus (again based on pulse oximeter readings) and allowed to stabilize for 10 minutes before the next set of readings and blood samples were taken. Dogs were then resaturated with 100% oxygen while the pulse oximeter probes were connected to the second set of monitors; then the desaturation and sampling process was repeated. All dogs were recovered from anesthesia. Horses Two geldings and three mares of various breeds, ranging from 5 to 15 years (mean ¼ 9.8 years) and from 455 to 554 kg (mean ¼ 493 kg) were studied. The horses were healthy (based on physical exam and measurement of hematocrit and total proteins) and food, but not water, was withheld the night before anesthesia. Horses were premedicated with xylazine (1.0 mg kg^1) given IV through a catheter (14 SWG, 14 cm Te£on catheter) pre-placed in a jugular vein. Anesthesia was induced with diazepam (0.03 mg kg^1) and ketamine (2.2 mg kg^1) given IV. Following endotracheal intubation, horses were positioned in left lateral recumbency on 30-cm thick foam pads and maintained on a circle system with iso£urane (1.25^1.5 MAC) in 100% oxygen and ventilated (North American Drager Large Animal Control Center, Telford, PA, USA) at 8 breaths minute^1 with a tidal volume of 8^10 mL kg^1 (mean ¼ 9 mL kg^1) to maintain PaCO2 in the range of 45^55 mm Hg (6^7.3 kPa). An arterial sample was drawn at the Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
beginning and end of anesthesia and analyzed immediately. ECG (leads in base-apex con¢guration), rectal temperature, and direct blood pressures (18 SWG, 4-cm Te£on catheter in the dorsal metatarsal artery) (Propaq 106 EL, Protocol Inc., Beaverton, OR, USA) were monitored continuously. Lactated Ringer’s solution (10 mL kg^1 hour^1) was administered and an infusion of dobutamine was used as necessary to maintain mean blood pressure above 65 mm Hg. The same monitors and probes were used; large animal clips were placed on all probes (Vet Sat large clip; V-1704, Surgi-Vet). Probe sites were the tongue, lip, ear (positioned as in the dogs), nostril (on the lamina of the alar cartilage) and, prepuce or vulva. The procedure for sample collection was the same as in the dogs. Following desaturation and sampling, horses were resaturated with 100% oxygen, disconnected from the anesthesia machine, then allowed to self-recover in a darkened recovery stall. The endotracheal tube was left in position and oxygen was insu¥ated through it (10 L minute ^1) until the horses assumed sternal recumbency. Cats Five, female, domestic short-haired cats, ranging in age from 10 to 14 years (mean ¼ 12 years) and in weight from 3.2 to 4.8 kg (mean ¼ 4 kg) were studied. Cats were healthy, based on physical exam, regular hematology and biochemical testing, including testing for feline immunode¢ciency and leukemia viruses. Food, but not water, was withheld the night before anesthesia. Cats were premedicated with glycopyrrolate (0.01 mg kg^1), oxymorphone (0.1 mg kg^1) and diazepam (0.3 mg kg^1) given intramuscularly. A catheter (20 SWG, 3-cm Te£on) was placed in a cephalic vein and anesthesia was induced with propofol (mean dose,5 mg kg^1 IV) in two cats and etomidate (mean dose, 0.5 mg kg^1 IV) in three cats. Following the second cat, the induction protocol was changed to attempt to facilitate placement of the arterial catheters; we thought that an etomidate induction would produce less alteration in heart rate and blood pressure. Following endotracheal intubation, cats were positioned in lateral recumbency and maintained on approximately 1.5 MAC iso£urane in 100% oxygen with a Bain breathing system. Four of ¢ve cats were ventilated (Penlon Neonatal Pediatric Anesthesia Ventilator. Series 300. Bear Medical Systems Inc., Riverside, CA, USA) at 10^12 breaths minute ^1 at a tidal volume of 20 mL kg^1 to maintain normal PaCO2, while the ¢fth cat was allowed to breathe spontaneously. A blood gas sample was 5
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
drawn once the arterial catheter was in place and at the end of the procedure and analyzed immediately. Cats were instrumented to record a lead II ECG (Nellcor 3940 Patient Monitor, Mallinckrodt Inc., Pleasanton, CA), rectal temperature and noninvasive blood pressure (with cu¡ placed on a forearm) (Dinamap Veterinary Blood Pressure Monitor 8300, Critikon Inc., Tampa, FL). An arterial catheter (22SWG, 3-cm Te£on) was percutaneously placed in the dorsal metatarsal artery (four cats) or ventral coccygeal artery (one cat). Lactated Ringer’s solution (10 mL kg^1hour^1) was infused throughout the procedure and a forced air warming blanket (Bair Hugger) was used to maintain normal body temperature. Pulse oximeters and probes studied were the same as for the dogs. Probe sites used were the tongue, ear, lip, front metacarpus and hind metatarsus. The procedure for desaturation, pulse oximeter readings and sample collection were similar, but blood sample volume was reduced to 0.3 mL per sample to avoid signi¢cant blood loss. The arterial blood samples were drawn into syringes containing heparin disks to minimize dilution from heparin (Self-¢lling PICO 70 Samplers, Radiometer America Inc., Westlake, OH, USA). Samples were drawn at 1-minute intervals at the 98 and 85% plateaus, but the sampling interval was reduced to 20 seconds at the 72% plateau because of di⁄culty in trying to maintain a stable plateau and concern for the cats. Following completion of the desaturation sequence, cats were resaturated with100% oxygen and recovered from anesthesia. Statistical methods and analysis Accuracy of saturation (SpO2) and pulse rate (PR) measurements for ¢ve pulse oximeter models was evaluated against co-oximetry and ECG references, respectively, in three animal species: dog, horse, and cat. Models tested were the N-395, NPB-190, NPB290, NPB-40, and V3304. Extreme values identi¢ed by the study monitor as clinical outliers with an assignable cause were excluded. Exclusion of extreme values as statistical outliers was based upon four statistical criteria (Belsey et al.1980; Kleinbaum et al.1988): 1 jJackknife residualj > 2, where the Jackknife residual is a studentized residual with the ith observation deleted; 2 Leverage > 2p/n, where leverage is the geometric distance of each individual ith predictor point from the center point of the predictor space; 6
3 COVRATIO ^ 1j 3p/n, where the COVRATIO statistic is the change in the determinant of covariance matrix of the estimates caused by deleting the ith observation; pffiffiffiffiffiffiffi 4 DFFITSj > 2 p=n, where the DFFITS statistic is a scaled measure of the change in the predicted value with the ith observation deleted; where n denotes the number of observations and p is the number of parameters. If at least three of the four criteria were met, an observation was excluded from the analysis data set. In addition, occasions in which a device failed to produce a reading were counted as missing values. Accuracy was calculated as the root mean squared di¡erence (RMSD):
RMSD ¼
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 Xn 1 Xn 2 ðyi xi Þ2 ¼ d ; ð1Þ i¼1 i¼1 i n n
where n is the number of measurements taken, yi is the SpO2 or PR readings from the pulse oximeter, xi is the corresponding measurement from co-oximetry or EKG, respectively, and di ¼ yi ^ xi. Units associated with RMSD are either saturation percentage points (SpO2) or beats minute ^1 (PR). For each animal species, the accuracy of each pulse oximeter model was evaluated for SpO2 and pulse rate, respectively. The pulse oximeter SpO2 value was compared to the corresponding average value of the two SaO2 measurements obtained by co-oximetry from arterial blood samples, and pulse oximeter pulse rate was compared against the corresponding measurement obtained from the ECG. For all analyses, the following results are reported: 1 Intercept and slope of regression line. 2 Bias and Precision: in general, the accuracy of devices can be assessed by two major components: bias and precision. Bias quanti¢es the systematic departure from the line of identity yi ¼ xi and can be estimated by the mean value of the di¡erences. Precision quanti¢es reproducibility of measurements, and is characterized by the estimated standard deviation of the di¡erences calculated using n in the denominatorOinstead of n ^ 1. 3 RMSD: accuracy is a composite measure appropriately represented by the RMSD (%).When squared, RMSD can be partitioned into components re£ecting bias and precision: 1 Xn 1 Xn 2 ðy xi Þ2 ¼ d i¼1 i i¼1 i n n 2 2 2 2 ^ ^ ¼ d þ ^d ¼ Bias þ Precision ;
RMSD2 ¼
ð2Þ
Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
where d ¼ 1 n
Xn
d i¼1 i
and ^2d ¼
1 Xn ðd dÞ2 i¼1 i n
4 Device failure rate: the device failure rate was calculated as the number of times the device failed to produce a reading divided by the total number of observations. In evaluating performance, the percentage of times a device failed to produce a reading was considered jointly with RMSD. Pearson correlation coe⁄cients were calculated for the di¡erent probe sites for each species. The R2, co-coe⁄cient and intercept coe⁄cient are presented.
Results Results from dogs are presented in Table 1. The mean sampling time was 130 minutes. RMSD (%) for saturations ranged from 1.74 (NPB-40) to 2.72 (V3304). The NPB-40, NPB-190 and NPB-395 gave readings on all sites on all dogs while the NPB-290 failed to produce one reading and theV3304 failed to produce readings in 30 instances. RMSD for heart rates ranged from 0.95 to 1.14. Variability of SpO2 within plateaus was 1.1%, while variability between models was 3.6%. Variability within the co-oximeter was 0.3%. Mean values for carboxyhemoglobin, methemoglobin and total hemoglobin reported by the co-oximeter were 1.4, 0.4 and 10.4 g dL1. Results from horses are presented inTable 2. Mean sampling time was 106 minutes. RMSD for saturations ranged from 2.13 (V3304) to 4.70 (NPB-290). All monitors except the NPB-290 failed to produce readings some of the time; failure rates ranged from 10% (NPB-395) to 60% (V3304). RMSD for heart rates ranged from 0.63 to 0.91. Variability of SpO2 within plateaus was 1.6%, while variability between models was 3.5%. Variability within the co-oximeter was 0.1%. Mean values for carboxyhemoglobin, methemoglobin and total hemoglobin reported by the co-oximeter were1.4,1.2 and12.0 g dL1. Results from cats are presented in Tables 3a and 3b. Mean sampling time was 84 minutes. Table 3a presents all data while Table 3b summarizes data for SaO2 > 90%. For all data, RMSD for saturations ranged from 5.56 (NPB-190) to 10.65 (V3304) while failure to produce a reading ranged from rates of 0% (NPB-395 and NPB-290) to 31.5% (V3304). For SaO2 > 90% in cats, RMSD ranged from 2.55 (NPB395) to 4.82 (V3304) while failure to produce readings Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
ranged from 0% (NPB-395 and NPB-290) to 43% (V3304). RMSD for heart rates for all data ranged from 2.8 to 3.6 while RMSD for heart rates with SaO2 > 90% ranged from 2.6 to 4.9. Variability of SpO2 within plateaus was 1.9%, while variability between models was 5.7%. Variability within the co-oximeter was 0.6%, while mean values for carboxyhemoglobin, methemoglobin and total hemoglobin were1.3, 0.7 and 8.1 g dL1. Correlation coe⁄cients (R2), and intercepts for the various probe sites for each species are shown in Table 4. In horses, correlation for the tongue, nostril, prepuce/vulva, and lip were fairly high (>0.93) while the ear appeared to be least accurate. In dogs, correlation for the tongue, toe, prepuce/vulva were good (>0.95) while the lip and ear were less accurate. In cats, correlation for all sites was low; the rear paw had the highest correlation (0.79).
Discussion We believe this is the ¢rst multi-species desaturation study on pulse oximeters and the only study to provide information on cats. Previous studies have tended to compare two or three models of pulse oximeters with one or two probe types. Also, in the past analysis has generally consisted of bias, precision and regression analysis. Since RMSD is now considered to be the industry standard (J.C. Allen, personal communication 2000), we thought it would be most appropriate. It is interesting that monitors appeared to perform di¡erently on di¡erent species while the techniques and sampling methods were similar. For instance, for the NPB-40, RMSD was1.74 in dogs,3.25 in horses and 7.86 in cats. As the spectral properties of hemoglobin have been shown not to vary signi¢cantly from species to species (Grosenbaugh & Muir 1997) we are not sure how to account for this species variability. It is possible that the algorithms in di¡erent models are a¡ected by variations in heart rate or signal strength to varying degrees. Ambient light has been reported as a source of erroneous pulse oximeter readings (Huss et al. 1995). However, in this study all probes were covered, so extraneous light is not a likely source of error. Although the co-oximeter is considered to be the gold standard for comparison of SaO2 and SpO2, it should probably be scrutinized as a source of error. We calibrated the co-oximeter prior to each animal (low, normal, high and total Hb calibrating solutions) and used the correct species setting for each animal. 7
No. of observations
Accuracy analysis after excluding device failures and outliers
Outliers
Regression
Pulse oximeter model
n
Saturation (SpO2)
N-395 NPB-190 NPB-290 NPB-40 V3304
149 149 149 149 149
6 0 0 0 0
Pulse rate (PR)
N-395 NPB-190 NPB-290 NPB-40 V3304
149 149 149 149 149
0 0 0 0 0
Parameters
Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
a
nd
intercept
Slope
Biase
Precision accuracyf
RMSD
(4.0%) (0.0%) (0.0%) (0.0%) (0.0%)
1 5 5 6 3
(0.7%) (3.4%) (3.4%) (4.0%) (2.0%)
7 5 5 6 3
(4.7%) (3.4%) (3.4%) (4.0%) (2.0%)
142 144 143 143 116
2.47 0.07 0.17 0.39 18.34
1.02 0.98 0.99 0.99 0.80
0.80 1.52 1.26 0.61 0.83
2.55 1.62 2.04 1.64 2.60
2.67 2.22 2.39 1.74 2.72
0 0 1 0 30
(0.0%) (0.0%) (0.7%) (0.0%) (20.1%)
(0.0%) (0.0%) (0.0%) (0.0%) (0.0%)
7 7 8 7 9
(4.7%) (4.7%) (5.4%) (4.7%) (6.0%)
7 7 8 7 9
(4.7%) (4.7%) (5.4%) (4.7%) (6.0%)
142 142 140 142 110
0.79 0.31 0.13 0.32 0.59
1.01 1.00 1.00 1.00 1.00
0.04 0.05 0.15 0.11 0.55
1.15 1.12 1.04 0.94 0.98
1.14 1.12 1.05 0.95 1.12
0 0 1 0 30
(0.0%) (0.0%) (0.7%) (0.0%) (20.1%)
Dog data were combined from the following locations: ear, lip, toe, tongue, prepuce, and vulva. The clinical outliers were justified based on review by the investigator. c Identification of statistical outliers was based upon meeting at least three of four statistical criteria forextreme values. d Sample size device failures and outliers. Xafterexcluding n e d ¼ ð1=nÞ i¼1 di, where di ¼ yi xi ¼ SpO2 ðSaO2i þ SaO22i Þ=2: b
f
^d ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Xn 2: ð1=nÞ ðdi dÞ i¼1
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi g 2 2 þ ^2d ¼ Bias ^ ^ 2 þ Precision RMSD ¼ d :
Device failures to produce reading
Statisticalc Total
clinicalb
g
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
8 Table 1 Accuracy analysis of pulse oximeters in dogs (n ¼ 5)a
No. of observations
Accuracy analysis after excluding device failures and outliers
Outliers
Regression
Pulse oximeter model
n
Saturation (SpO2)
N-395 NPB-190 NPB-290 NPB-40 V3304
150 150 150 150 150
0 0 3 3 0
Pulse rate (PR)
N-395 NPB-190 NPB-290 NPB-40 V3304
150 150 150 150 150
0 0 6 0 0
Parameters
a
clinicalb
Statisticalc Total
nd
intercept
Slope
(0.0%) (0.0%) (2.0%) (2.0%) (0.0%)
2 5 3 3 1
(1.3%) (3.3%) (2.0%) (2.0%) (0.7%)
2 5 6 6 1
(1.3%) (3.3%) (4.0%) (4.0%) (0.7%)
133 113 144 119 59
11.92 9.67 0.39 7.15 7.22
0.84 0.87 0.96 1.05 0.9
(0.0%) (0.0%) (4.0%) (0.0%) (0.0%)
5 4 3 4 1
(3.3%) (2.7%) (2.0%) (2.7%) (0.7%)
5 4 9 4 1
(3.3%) (2.7%) (6.0%) (2.7%) (0.7%)
130 114 141 121 59
0.66 0.33 0.90 0.93 0.58
1.01 1.00 1.01 1.01 1.00
Horse data were combined from the followinglocations: ear, lip, nostril, tongue, prepuce, and vulva. The clinical outliers were justified based on clinical review. c Identification of statistical outliers was based upon meeting at least three of four statistical criteria forextreme values. d Sample size afterexcluding device failures and outliers. X n e d ¼ ð1=nÞ i¼1 di, where di ¼ yi xi ¼ SpO2i ðSaO21i þ SaO22i Þ=2: b
f
g
^d ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Xn 2: ð1=nÞ i¼1 ðdi dÞ
RMSD ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2 þ ^2 ¼ Bias ^ ^ 2 þ Precision d : d
Device failures to produce reading
Precision accuracy
RMSD
2.13 1.94 3.23 2.52 1.65
2.25 2.30 3.42 2.07 1.36
3.09 3.00 4.70 3.25 2.13
15 (10.0%) 32 (21.3%) 0 (0.0%) 25 (16.7%) 90 (60.0%)
0.18 0.21 0.21 0.23 0.66
0.66 0.64 0.89 0.59 0.63
0.68 0.68 0.91 0.63 0.91
15 (10.0%) 32 (21.3%) 0 (0.0%) 25 (16.7%) 90 (60.0%)
Biase
g
9
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
Table 2 Accuracy analysis of pulse oximeters in horses (n ¼ 5)a
No. of observations
Accuracy analysis after excluding device failures and outliers
Outliers
Regression
Pulse oximeter model
n
Saturation (SpO2)
N-395 NPB-190 NPB-290 NPB-40 V3304
149 149 149 149 149
5 0 0 0 0
Pulse rate (PR)
N-395 NPB-190 NPB-290 NPB-40 V3304
149 149 149 149 149
0 1 0 0 0
Parameters
Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
a
nd
intercept
Slope
Biase
Precision accuracy
RMSD
(3.4%) (0.0%) (0.0%) (0.0%) (0.0%)
6 7 3 1 5
(4.0%) (4.7%) (2.0%) (0.7%) (3.4%)
138 141 146 118 97
17.07 6.77 27.44 36.55 49.41
0.82 1.05 0.69 0.60 0.46
1.44 2.32 0.94 3.17 4.26
5.79 5.07 7.82 7.22 9.81
5.94 5.56 7.85 7.86 10.65
0 1 0 30 47
(0.0%) (0.7%) (0.0%) (20.1%) (31.5%)
(0.0%) (0.7%) (0.0%) (0.0%) (0.0%)
5 6 6 3 3
(3.4%) (4.0%) (4.0%) (2.0%) (2.0%)
96 93 95 72 61
1.77 0.24 1.11 4.49 3.02
1.01 1.00 1.00 0.97 0.98
0.58 0.10 0.46 0.15 0.44
3.46 3.57 3.47 2.88 2.82
3.49 3.55 3.48 2.86 2.83
48 49 48 74 85
(32.2%) (32.9%) (32.2%) (49.7%) (57.0%)
11 (7.4%) 7 (4.7%) 3 (2.0%) 1 (0.7%) 5 (3.4%) 5 7 6 3 3
(3.4%) (4.7%) (4.0%) (2.0%) (2.0%)
Cat data were combined from the following locations: ear, metacarpus, lip, metatarsus, and tongue. The clinical outliers were justified based on clinical review. c Identification of statistical outliers was based upon meeting at least three of four statistical criteria forextreme values. d Sample size afterexcluding device failures and outliers. X n e d ¼ ð1=nÞ i¼1 di, where di ¼ yi xi ¼ SpO2i ðSaO21i þ SaO22i Þ=2: b
f
g
^d ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Xn 2: ð1=nÞ i¼1 ðdi dÞ
RMSD ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2 þ ^2 ¼ Bias ^ ^ 2 þ Precision d : d
Device failures to produce reading
Statisticalc Total
clinicalb
g
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
10 Table 3a Accuracy analysis of pulse oximeters in cats (n ¼ 5)a
No. of observations
Accuracy analysis after excluding device failures and outliers
Outliers
Regression
Pulse oximeter model
n
Saturation (SpO2)
N-395 NPB-190 NPB-290 NPB-40 V3304
60 60 60 60 60
0 0 0 0 0
Pulse rate (PR)
N-395 NPB-190 NPB-290 NPB-40 V3304
60 60 60 60 60
0 0 0 0 0
Parameters
a
clinicalb
f
^d ¼
g
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Xn 2: ð1=nÞ i¼1 ðdi dÞ
RMSD ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2 þ ^2 ¼ Bias ^ ^ 2 þ Precision d : d
RMSD
0.88 2.39 2.00 0.59 1.33
2.42 3.71 3.45 3.45 4.7
2.55 4.39 3.96 3.46 4.82
0 1 0 15 26
(0.0%) (1.7%) (0.0%) (25.0%) (43.3%)
0.40 0.09 0.44 0.33 0.27
2.95 4.93 2.91 2.57 2.69
2.95 4.88 2.91 2.57 2.65
14 15 14 29 37
(23.3%) (25.0%) (23.3%) (48.3%) (61.7%)
nd
intercept
Slope
Biase
(0.0%) (0.0%) (0.0%) (0.0%) (0.0%)
1 2 2 3 0
(1.7%) (3.3%) (3.3%) (5.0%) (0.0%)
1 2 2 3 0
(1.7%) (3.3%) (3.3%) (5.0%) (0.0%)
59 57 58 42 34
21.78 8.84 36.55 18.44 141.95
0.76 0.88 1.36 1.19 0.49
(0.0%) (0.0%) (0.0%) (0.0%) (0.0%)
3 2 3 1 1
(5.0%) (3.3%) (5.0%) (1.7%) (1.7%)
3 2 3 1 1
(5.0%) (3.3%) (5.0%) (1.7%) (1.7%)
43 43 43 30 22
0.58 3.10 0.87 6.80 12.25
1.00 0.98 1.00 0.95 0.91
Cat data were combined from the followinglocations: ear, metacarpus, lip, metatarsus, and tongue. The clinical outliers were justified based on clinical review. c Identification of statistical outliers was based upon meeting at least three of four statistical criteria forextreme values. d Sample size device failures and outliers. Xafterexcluding n e d ¼ ð1=nÞ i¼1 di, where di ¼ yi xi ¼ SpO2i ðSaO21i þ SaO22i Þ=2: b
Device failures to produce reading
Precision accuracy
Statisticalc Total
g
11
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
Table 3b Accuracy analysis of pulse oximeters in catsa when SaO2 90% (n ¼ 5)a
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
Dog
Horse
Cat
Tongue Toe Prepuce/vulva Lip Ear Tongue Nostril Prepuce/vulva Lip Ear Tongue Front paw Rear paw Lip Ear
R2
Co-coefficient, m
Intercept coefficient, b
0.95 0.97 0.95 0.87 0.69 0.98 0.94 0.94 0.93 0.88 0.54 0.57 0.79 0.49 0.64
0.92 0.96 0.97 0.99 0.99 0.88 1.1 0.83 0.84 0.87 0.62 0.60 1.11 0.75 0.69
3.1 1.9 0.2 1.3 1.3 6.6 10.2 9.9 9.2 5.2 33.9 34.9 13.5 17.8 27.9
Table 4 Correlation coe⁄cients for pulse oximeter sites ^ all models included in calculations for each site
Y ¼ mx þ b.
However, we have no information about generation of the extinction coe⁄cients for the di¡erent species within the co-oximeter. The co-oximeter uses ¢ve wavelengths of light where the pulse oximeter only uses two. Samples were run through the co-oximeter in duplicate and averaged for the analysis. Ideally, triplicate analysis might have been more accurate, but would have contributed to a longer delay before analysis of each sample, which might have introduced further error due to re-mixing and storage of samples. Calculated variability within the co-oximeter varied from 0.1% in horses to 0.6% in cats, which would seem to indicate that it was less accurate in the cats. Again, this may have been due to the small size of samples collected. Total hemoglobin values reported for the cats were also low compared to dogs and horses, even though all cats were clinically normal; complete blood counts taken within 1 month before and after the study showed total hemoglobin values in the range of 10^14 g dL ^1. This may also indicate inaccuracy in the co-oximeter readings during the study, which could have been due to small sample size or dilutional e¡ects of the heparin pellet. As explained in the statistical methods, outliers were quali¢ed by clinical review or by diagnostic statistics. Clinical review consisted of checking experimental notes for reasons that would contribute to an inaccurate reading not related to the monitor. In the horses, probes were readjusted when one slid o¡ the vulva (NPB-40) and when the motion detector light repeatedly came on (NPB-290). In dogs, repositioning 12
of a probe revealed an indentation of lip tissue (NPB-395) and a probe was repositioned on the vulva (NPB-395). In cats, a probe sliding o¡ a small lip was repeatedly repositioned (NPB-395). Probe placement and sites have been previously shown to contribute to variability between SpO2 and SaO2 (Jacobson et al. 1992; Cha⁄n et al. 1996). Since the objective of this study was to evaluate the overall performance of the pulse oximeter model, we chose not to draw conclusions about the accuracy of a particular site, since one would then have to assume that all monitors performed equally. Variability in sites is probably a large cause of the variability between pulse rate measured by the pulse oximeter and heart rate displayed by the ECG. Signal strength was very likely di¡erent from site to site, which would be a cause for disparities between pulse rate and displayed heart rate. In cats, the small ECG complexes may have a¡ected how well the ECG counted heart rate. For most monitors, accuracy of the saturation reading is linked to accuracy of the pulse rate; if the monitor is not able to detect a good pulse signal, no saturation will be displayed. Heart rates displayed on the ECG could be di¡erent from pulse rates for many reasons, including the time over which ECG is measured and multiplication factors. The overall failure rate of the V3304 to produce readings was much higher than for the other monitors. This may have been because the probe is intended for lingual use only (Surgi-VetV3304 Operation Manual 1999,Version 3, pp. 4^9), although it did function on nonlingual sites. Failure to produce any Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
readings occurred in the cat when the probe was placed on the metacarpal or metatarsal region, in the dog when the probe was placed on the toe, and in the horse when the probe was placed on the lip, ear or vulva. A previous desaturation study in horses reported a bias of 0.55 and precision of 2.57, which is similar to the precision reported in this study of1.4^3.4%. However, the previous study evaluated only one site (the tongue) with one monitor (Martinez et al. 1999) and only 60 readings were taken compared to 150 readings in this study. In the present study, two monitors failed to produce readings during the ¢rst 45^ 60 minutes after induction in horses. In one horse, the NPB-40 (probe on ear) and NPB-395 (probe on nostril) did not produce readings during the ¢rst desaturation period, but produced readings during the second set of desaturations without touching or repositioning the probes. In a second horse, the NPB-40 (probe on prepuce) did not produce readings during the ¢rst desaturation period, but gave readings during the second desaturation period. The NPB-190 (probe on nostril) failed to produce readings at any time on this horse. Both of these horses had deeply pigmented (i.e. black) skin. Dark pigmentation has been previously suggested to be a source of failures and inaccuracies with pulse oximeters in animals (Jacobson et al. 1992), but we speculate that local vasoconstriction produced by the induction drugs may have caused the temporal failures. Alternatively, the combination of dark pigmentation with local vasoconstriction may have produced the failures. Half (13 of 25) of the sites used were pigmented black; of these, four sites failed to produce readings, with three of the four being attached to the same monitor. All nonblack sites (including brown, grey and speckled pigmentation) gave readings. The 60% failure rate seen with the V3304 in the horse was higher than for other monitors, and did not change over time. Again, this may have been due to probe placement; the monitor did produce readings 100% of the time when the probe was placed on the tongue (as is the manufacturers recommendation). Hypotension, associated with poor peripheral perfusion, has also been shown to a¡ect pulse oximeter readings (Reich et al.1996). However, we maintained a mean blood pressure >60 mm Hg in the horses by infusing dobutamine and blood pressure was maintained in the dogs and cats by adjusting anesthetic depth and/or increasing £uid administration. It might have been useful to measure blood pressure directly in the dogs and cats, since the accuracy of Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
this method is better than indirect measurement. However, we would not have been able to get readings during the times when samples were drawn (which would have been the times of most interest). An increase in blood pressures was seen at the lower saturations (as would be expected with hypoxia), however, since these blood pressures were not recorded we have not presented them as data.We do not believe that hypotension produced pulse oximeter failures in this study. We cannot explain why the accuracy of all monitors was so much poorer in cats than in dogs or horses, when the same techniques and methodology were used for sampling. It is possible that varying the induction drugs may have introduced some error into the cat data. Saturation values in the cats also appeared to be more unstable at the lower saturation plateaus than the dogs or horses presumably because of their smaller size. This instability would contribute to error. It is also possible that the smaller blood samples drawn from the cats contributed to errors in the co-oximeter. Although the co-oximeter requires only 35 mL of blood and our sample size was 0.3 mL, mixing and dilutional errors may have occurred. Arterial catheters were placed peripherally since these could be placed percutaneously without requiring a ldquor;cut-down and with the thought that the pulse oximeter probes were also being placed peripherally. However, given the large variability, we wonder if cats may vasoconstrict peripheral sites at varying rates. Changes in sympathetic tone associated with decreased oxygen tension may have a¡ected blood £ow to peripheral sites. This would also explain why the di¡erences between pulse rates from the pulse oximeter were di¡erent from ECG heart rates in the cats; weak pulses might not be counted by the pulse oximeter. The higher heart rates seen in the cats might also contribute to greater error between pulse rates and ECG heart rates. However, the errors appeared to be greater in cats than those observed in dogs and horses. The NPB-40 failed to produce readings on one cat (probe placed on the metatarsal site). Percutaneous placement of the arterial catheter in the dorsal pedal artery had been unsuccessful in that foot so it is possible that blood £ow may have been disrupted su⁄ciently to cause the failures. Pulse oximeter failures have been reported to be useful for evaluating compromised circulation (Dorsch & Dorsch 1999). However, no perfusion problems were observed after recovery of the cat from anesthesia. 13
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
It is interesting that the accuracy of all monitors was reasonably good ( 3^5%) for cats when SaO2 was greater than 90%, although unfortunate that none of the models were very accurate for lower saturations. Since this is the range most commonly seen in clinical patients, it seems to provide important information about use of these models in cats. It would seem that further studies of the accuracy of pulse oximeters in cats should be performed.
Conclusions Accuracy of the monitors evaluated varied from species to species. In the dog, the RMSD for all monitors was 2^3%; the best monitor appeared to be the NPB-40, followed by the NPB-190, 290, 395 and V3304. In horses, the RMSD for all monitors was 2^5%; if the V3304 is excluded due to the high failure rate, the best monitor appeared to be the NPB190 followed by the NPB-395, NPB-40 and NPB290. In cats, accuracy over the entire range of saturations evaluated (70^100%) was not particularly good; RMSD for all monitors was 6^11%. In cats, the best monitor appeared to be the NPB-190, followed by NPB-395, NPB-290, NPB-40 and V3304. In the clinically normal range (i.e. saturations >90%), accuracy was similar to that seen in the dogs and horses; RMSD for all monitors for this range was 3^5%. However, based on this study, caution is advised when interpreting pulse oximeter values in cats. When the V3304 gave readings, they were reasonably accurate, but the monitor failed to produce readings 20% of the time in dogs, 30% of the time in cats and 60% of the time in horses, which was much higher than for other monitors.
Acknowledgements Funding and equipment for this study were provided by Mallinckrodt Inc., Pleasanton, CA, USA.
References Barton LJ, DeveyJJ, Gorski S et al. (1996). Evaluation of transmittance and re£ectance pulse oximetry in a canine
14
model of hypotension and desaturation. J Vet Emerg Crit Care6, 21^28. Belsey DA, Kuh E,Welsch RE (1980) Regression Diagnostics. JohnWiley & Sons, NewYork, NY. Cha⁄n MK, Matthews NS, Cohen N et al. (1996) Evaluation of pulse oximetry in anaesthetised foals using multiple combinations of transducer type and transducer attachment site. EqVet J28, 437^445. DorschJA, Dorsch SE (1999) Pulse oximetry. In: Understanding Anesthesia Equipment, 4th edn. Dorsch JA, Dorsch SE (eds). Williams & Wilkins, Philadelphia, PA, USA, pp. 811^847. Fairman NB (1993) Evaluation of pulse oximetry as a continuous monitoring technique in critically ill dogs in the small animal intensive care unit. J Vet Emerg Crit Care2, 50^56. Grosenbaugh DA, Muir WW (1997) Absorbance spectra of inter-species hemoglobins in the visible and near infrared regions. J Vet Emerg Crit Care 7,36^42. Grosenbaugh DA, MuirWW (1998) Accuracy of noninvasive oxyhemoglobin saturation, end-tidal carbon dioxide concentration, and blood pressure monitoring during experimentally induced hypoxemia, hypotension, or hypertension in anesthetized dogs. Am J Vet Res 59, 205^212. Huss BT Anderson MA, Branson KR et al. (1995) Evaluation of pulse oximeter probes and probe placement in healthy dogs. J Am Anim Hosp Assoc 31,9^14. Jacobson JD, Miller MW, Matthews NS et al. (1992) Evaluation of accuracy of pulse oximetry in dogs. Am J Vet Res 53,537^540. Kleinbaum DG, Kupper LL, Muller KE (1988). Applied Regression Analysis and other Multivariate Methods. PWS-KENT Publishing Co., Boston, MA. Martinez EA, Carroll GL, Harts¢eld SM (1999). Evaluation of the Nonin 8600V veterinary pulse oximeter in anesthetized horses. J Vet Emerg Crit Care 9,13^17. Matthews NS, Sanders EA, Harts¢eld SM et al (1995) A comparison of 2 pulse oximeters in dogs. J Vet Emerg Crit Care 5,116^120. Reich DL, Timcenko A, Bodian CA et al. (1996) Predictors of pulse oximetry data failure. Anesthesiology 84, 859^864. Schmitt PM, Gobel T,Trautvetter E (1998) Evaluation of pulse oximetry as a monitoring method in avian anesthesia. J Avian Med Surg12,91^99. Received 30 July 2001; accepted 8 July 2002.
Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
R E S E A R C H PA P E R
An evaluation of pulse oximeters in dogs, cats and horses Nora S Matthews
DVM, Diplomate ACVA,
Sherrie Hartke,
DVM
y
and John C Allen Jr.
MS
z
Department of Small Animal Medicine and Surgery,Texas A&M University, College Station,TX,
y
North Shore Animal League ^ America,The Alex Lewyt Medical Center, Lewyt Street, Port Washington, NY, z Tyco Healthcare, Mallinckrodt Inc., Hazelwood, MO, USA
Correspondence: Nora S Matthews, Department of Small Animal Medicine and Surgery,Texas A&M University, College Station,TX 77843-4474, USA. E-mail: [email protected].
Abstract Objective Evaluation of ¢ve pulse oximeters in dogs, cats and horses with sensors placed at ¢ve sites and hemoglobin saturation at three plateaus. Study design Prospective randomized multispecies experimental trial. Animals Five healthy dogs, cats and horses. Methods Animals were anesthetized and instrumented with ECG leads and arterial catheters. Five pulse oximeters (Nellcor Puritan Bennett-395, NPB190, NPB-290, NPB-40 and Surgi-Vet V3304) with sensors at ¢ve sites were studied in a 5 5 Latin square design. Ten readings (SpO2) were taken at each of three hemoglobin saturation plateaus (98, 85 and 72%) in each animal. Arterial samples were drawn concurrently and hemoglobin saturation was measured with a co-oximeter. Accuracy of saturation measurements was calculated as the root mean squared di¡erence (RMSD), a composite of bias and precision, for each model tested in each species. Results Accuracy varied widely. In dogs, the RMSD for the NPB-395, NPB-190, NPB-290, NPB-40 and V3304 were 2.7, 2.2, 2.4, 1.7 and 2.7% respectively. Failure to produce readings for the NPB-395, NPB190, NPB-290, NPB-40 and V3304 were 0, 0, 0.7, 0, and 20%, respectively.The Pearson correlation coe⁄cients for the tongue, toe, ear, lip and prepuce or vulva were 0.95, 0.97, 0.69, 0.87 and 0.95, respectively. In horses, the RMSD for the NPB-395, NPB-190,
NPB-290, NPB-40 and V3304 were 3.1, 3.0, 4.7, 3.3 and 2.1%, respectively while rates of failure to produce readings were10, 21,0,17 and 60%, respectively. The Pearson correlation coe⁄cients for the tongue, nostril, ear, lip and prepuce or vulva were 0.98, 0.94, 0.88, 0.93 and 0.94, respectively. In cats, the RMSD for all data for the NPB-395, NPB-190, NPB-290, NPB-40 and V3304 were 5.9, 5.6, 7.9, 7.9 and 10.7%, respectively while failure rates were 0, 0.7, 0, 20 and 32%, respectively. The correlation coe⁄cients for the tongue, rear paw, ear, lip and front paw were 0.54, 0.79,.0.64, 0.49 and 0.57, respectively. For saturations above 90% in cats, the RMSD for the NPB-395, NPB190, NPB-290, NPB-40 and V3304 were 2.6, 4.4, 4.0, 3.5 and 4.8%, respectively, while failure rates were 0, 1.7, 0, 25 and 43%, respectively. Conclusions and clinical relevance Accuracy and failure rates (failure to produce a reading) varied widely from model to model and from species to species. Generally, among the models tested in the clinically relevant range (90^100%) RMSD ranged from 2^5% while failure rates were highest in theV3304. Keywords pulse oximetry, saturation, dogs, cats, horses.
Introduction Pulse oximetry is a common method of monitoring the saturation of hemoglobin with oxygen noninvasively in veterinary patients. Various probe sites and di¡erent models of pulse oximeters have been evaluated in dogs (Fairman 1993; Huss et al. 1995; 3
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
Matthews et al. 1995; Barton et al. 1996; Grosenbaugh & Muir 1998), horses (Cha⁄n et al. 1996) and birds (Schmitt et al.1998). However, newer models of pulse oximeters have become available and no information is published about their accuracy at varying sites and during desaturation. Therefore, the purpose of this study was to evaluate several newer models of pulse oximeters when probes were placed at various sites in three species (dogs, cats and horses) over a range of hemoglobin saturation.
assigned the ¢ve pulse oximeters to the ¢ve probe sites. Two calibrated oximeters of each model were tested on each animal in one of the ¢ve probe sites. Sensor application sites, except the tongue, were shaved prior to sensor placement and sensors were covered, after they were in place, to reduce ambient light. Anesthesia, monitoring and blood sample collection Dogs
Materials and methods Study design All procedures were approved by theTexas A&M University LaboratoryAnimal Care Committee. Five animals of each species (dogs, cats and horses) were anesthetized and instrumented with a pulse oximeter probe at ¢ve di¡erent sites, each attached to one of ¢ve di¡erent pulse oximeters. A catheter was placed into a peripheral artery to permit periodic blood sampling. Arterial samples were drawn at estimated plateaus of 98, 85 and 72% saturation. The saturation plateau was maintained for 10 minutes before drawing samples and taking readings. At 1minute intervals and over at least one complete respiratory cycle, ¢ve arterial samples were drawn at each plateau at the same time that pulse oximeter readings and heart rate were recorded. Animals were resaturated while probes were changed to duplicate monitors and the procedure was repeated. Functional hemoglobin saturation was determined in duplicate, with a co-oximeter (Radiometer OSM-3 Hemoximeter. Radiometer America, Inc., Westlake, OH) which was set in the appropriate mode for each species and calibrated before each animal was anesthetized. Functional saturation was calculated (by the co-oximeter) using the standard formula: % functional SaO2 % O2 Hb 100 ¼ ð100 ð% metHb þ % CoHbÞÞ Following the second period of desaturation and readings, all animals were resaturated, then recovered from anesthesia. Randomization and sensor placement A computer-generated random number scheme, based on a Latin square experimental design, 4
One male and four female mixed-breed dogs, ranging from 6 to11 years in age (mean ¼ 8.8 years) and from 12 to 25 kg (mean ¼ 16 kg) in weight were used in the study. The dogs were healthy (based on physical examination and regular hematology and biochemical testing) and were heartworm negative. Food, but not water, was withheld the night before anesthesia. No pre-anesthetic medications were given and anesthesia was induced with propofol (2.8^7.3 mg kg^1 IV, mean ¼ 5.2) (Propo£o, Abbott Laboratories, Chicago, IL, USA). Following endotracheal intubation, dogs were positioned in lateral recumbency, connected to a circle system (North American Drager Small Animal Anesthesia machine, Telford, PA, USA), and maintained with iso£urane (Iso£o. Abbott Laboratories, Chicago, IL, USA) in 100% oxygen, at approximately1.5 MAC (measured at the proximal end of the endotracheal tube) (Capnomac Ultima. Datex Medical Instrumentation, Tewksbury, MA, USA). Dogs were ventilated at a rate of 11^14 breaths minute ^1 and a tidal volume (8^15 mL kg^1, mean ¼ 11 mL kg^1) needed to maintain PaCO2 in the range of 35^45 mm Hg (4.7^6 kPa). A blood gas sample was drawn at the beginning (after ventilation was established) and end of anesthesia and analyzed immediately (IRMA, Diametrics Medical, St. Paul, MN, USA). Blood pressures were measured noninvasively (Dinamap Veterinary Blood Pressure Monitor 8300, Critikon Inc., Tampa, FL, USA) with the cu¡ placed on a forearm and a lead II ECG was monitored (Nellcor-PB 3940. Mallinckrodt Inc., Pleasanton, CA, USA). An arterial catheter (20 SWG, 3-cm Te£on catheter) was placed percutaneously in a dorsal metatarsal artery for collection of blood samples. All dogs received lactated Ringer’s solution (10 mL kg^1 hour^1) administered through a catheter placed in a cephalic vein and body temperature (measured rectally) was maintained with a forced air warming blanket (Bair Hugger Warming Unit, Model 505. Augustine Medical Inc., Eden Prairie, MN, USA). Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
The pulse oximeters studied were the NPB-40, NPB-190, NPB-290, and NPB-395 (Nellcor oximetry. Mallinckrodt Inc., Pleasanton, CA, USA) connected to a veterinary sensor (Vet Sat. Nellcor oximetry. Mallinckrodt Inc., Pleasanton, CA, USA) and the V3304 (Surgi-Vet Inc., Waukesha, WI, USA) connected to a veterinary sensor (V1703, Surgi-Vet Inc., Waukesha, WI, USA). The ¢ve probe sites were the tongue, lip, ear (placed on the lateral edge and close to the base), toe, and prepuce or vulva. Following instrumentation, the inspired oxygen concentration was reduced (by mixing nitrogen into the circuit through a separate £owmeter) until the 98% saturation plateau was reached, based on readings from the pulse oximeters. One-milliliter blood samples were drawn into heparinized syringes, at 1minute intervals as the pulse oximeter readings and heart rates were recorded simultaneously. Blood samples were drawn over one complete respiratory cycle, waste blood was drawn before sampling and the catheter was £ushed after sampling. The oxygen^nitrogen mixture was then adjusted to reach the 85 and 72% plateaus (again based on pulse oximeter readings) and allowed to stabilize for 10 minutes before the next set of readings and blood samples were taken. Dogs were then resaturated with 100% oxygen while the pulse oximeter probes were connected to the second set of monitors; then the desaturation and sampling process was repeated. All dogs were recovered from anesthesia. Horses Two geldings and three mares of various breeds, ranging from 5 to 15 years (mean ¼ 9.8 years) and from 455 to 554 kg (mean ¼ 493 kg) were studied. The horses were healthy (based on physical exam and measurement of hematocrit and total proteins) and food, but not water, was withheld the night before anesthesia. Horses were premedicated with xylazine (1.0 mg kg^1) given IV through a catheter (14 SWG, 14 cm Te£on catheter) pre-placed in a jugular vein. Anesthesia was induced with diazepam (0.03 mg kg^1) and ketamine (2.2 mg kg^1) given IV. Following endotracheal intubation, horses were positioned in left lateral recumbency on 30-cm thick foam pads and maintained on a circle system with iso£urane (1.25^1.5 MAC) in 100% oxygen and ventilated (North American Drager Large Animal Control Center, Telford, PA, USA) at 8 breaths minute^1 with a tidal volume of 8^10 mL kg^1 (mean ¼ 9 mL kg^1) to maintain PaCO2 in the range of 45^55 mm Hg (6^7.3 kPa). An arterial sample was drawn at the Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
beginning and end of anesthesia and analyzed immediately. ECG (leads in base-apex con¢guration), rectal temperature, and direct blood pressures (18 SWG, 4-cm Te£on catheter in the dorsal metatarsal artery) (Propaq 106 EL, Protocol Inc., Beaverton, OR, USA) were monitored continuously. Lactated Ringer’s solution (10 mL kg^1 hour^1) was administered and an infusion of dobutamine was used as necessary to maintain mean blood pressure above 65 mm Hg. The same monitors and probes were used; large animal clips were placed on all probes (Vet Sat large clip; V-1704, Surgi-Vet). Probe sites were the tongue, lip, ear (positioned as in the dogs), nostril (on the lamina of the alar cartilage) and, prepuce or vulva. The procedure for sample collection was the same as in the dogs. Following desaturation and sampling, horses were resaturated with 100% oxygen, disconnected from the anesthesia machine, then allowed to self-recover in a darkened recovery stall. The endotracheal tube was left in position and oxygen was insu¥ated through it (10 L minute ^1) until the horses assumed sternal recumbency. Cats Five, female, domestic short-haired cats, ranging in age from 10 to 14 years (mean ¼ 12 years) and in weight from 3.2 to 4.8 kg (mean ¼ 4 kg) were studied. Cats were healthy, based on physical exam, regular hematology and biochemical testing, including testing for feline immunode¢ciency and leukemia viruses. Food, but not water, was withheld the night before anesthesia. Cats were premedicated with glycopyrrolate (0.01 mg kg^1), oxymorphone (0.1 mg kg^1) and diazepam (0.3 mg kg^1) given intramuscularly. A catheter (20 SWG, 3-cm Te£on) was placed in a cephalic vein and anesthesia was induced with propofol (mean dose,5 mg kg^1 IV) in two cats and etomidate (mean dose, 0.5 mg kg^1 IV) in three cats. Following the second cat, the induction protocol was changed to attempt to facilitate placement of the arterial catheters; we thought that an etomidate induction would produce less alteration in heart rate and blood pressure. Following endotracheal intubation, cats were positioned in lateral recumbency and maintained on approximately 1.5 MAC iso£urane in 100% oxygen with a Bain breathing system. Four of ¢ve cats were ventilated (Penlon Neonatal Pediatric Anesthesia Ventilator. Series 300. Bear Medical Systems Inc., Riverside, CA, USA) at 10^12 breaths minute ^1 at a tidal volume of 20 mL kg^1 to maintain normal PaCO2, while the ¢fth cat was allowed to breathe spontaneously. A blood gas sample was 5
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
drawn once the arterial catheter was in place and at the end of the procedure and analyzed immediately. Cats were instrumented to record a lead II ECG (Nellcor 3940 Patient Monitor, Mallinckrodt Inc., Pleasanton, CA), rectal temperature and noninvasive blood pressure (with cu¡ placed on a forearm) (Dinamap Veterinary Blood Pressure Monitor 8300, Critikon Inc., Tampa, FL). An arterial catheter (22SWG, 3-cm Te£on) was percutaneously placed in the dorsal metatarsal artery (four cats) or ventral coccygeal artery (one cat). Lactated Ringer’s solution (10 mL kg^1hour^1) was infused throughout the procedure and a forced air warming blanket (Bair Hugger) was used to maintain normal body temperature. Pulse oximeters and probes studied were the same as for the dogs. Probe sites used were the tongue, ear, lip, front metacarpus and hind metatarsus. The procedure for desaturation, pulse oximeter readings and sample collection were similar, but blood sample volume was reduced to 0.3 mL per sample to avoid signi¢cant blood loss. The arterial blood samples were drawn into syringes containing heparin disks to minimize dilution from heparin (Self-¢lling PICO 70 Samplers, Radiometer America Inc., Westlake, OH, USA). Samples were drawn at 1-minute intervals at the 98 and 85% plateaus, but the sampling interval was reduced to 20 seconds at the 72% plateau because of di⁄culty in trying to maintain a stable plateau and concern for the cats. Following completion of the desaturation sequence, cats were resaturated with100% oxygen and recovered from anesthesia. Statistical methods and analysis Accuracy of saturation (SpO2) and pulse rate (PR) measurements for ¢ve pulse oximeter models was evaluated against co-oximetry and ECG references, respectively, in three animal species: dog, horse, and cat. Models tested were the N-395, NPB-190, NPB290, NPB-40, and V3304. Extreme values identi¢ed by the study monitor as clinical outliers with an assignable cause were excluded. Exclusion of extreme values as statistical outliers was based upon four statistical criteria (Belsey et al.1980; Kleinbaum et al.1988): 1 jJackknife residualj > 2, where the Jackknife residual is a studentized residual with the ith observation deleted; 2 Leverage > 2p/n, where leverage is the geometric distance of each individual ith predictor point from the center point of the predictor space; 6
3 COVRATIO ^ 1j 3p/n, where the COVRATIO statistic is the change in the determinant of covariance matrix of the estimates caused by deleting the ith observation; pffiffiffiffiffiffiffi 4 DFFITSj > 2 p=n, where the DFFITS statistic is a scaled measure of the change in the predicted value with the ith observation deleted; where n denotes the number of observations and p is the number of parameters. If at least three of the four criteria were met, an observation was excluded from the analysis data set. In addition, occasions in which a device failed to produce a reading were counted as missing values. Accuracy was calculated as the root mean squared di¡erence (RMSD):
RMSD ¼
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 Xn 1 Xn 2 ðyi xi Þ2 ¼ d ; ð1Þ i¼1 i¼1 i n n
where n is the number of measurements taken, yi is the SpO2 or PR readings from the pulse oximeter, xi is the corresponding measurement from co-oximetry or EKG, respectively, and di ¼ yi ^ xi. Units associated with RMSD are either saturation percentage points (SpO2) or beats minute ^1 (PR). For each animal species, the accuracy of each pulse oximeter model was evaluated for SpO2 and pulse rate, respectively. The pulse oximeter SpO2 value was compared to the corresponding average value of the two SaO2 measurements obtained by co-oximetry from arterial blood samples, and pulse oximeter pulse rate was compared against the corresponding measurement obtained from the ECG. For all analyses, the following results are reported: 1 Intercept and slope of regression line. 2 Bias and Precision: in general, the accuracy of devices can be assessed by two major components: bias and precision. Bias quanti¢es the systematic departure from the line of identity yi ¼ xi and can be estimated by the mean value of the di¡erences. Precision quanti¢es reproducibility of measurements, and is characterized by the estimated standard deviation of the di¡erences calculated using n in the denominatorOinstead of n ^ 1. 3 RMSD: accuracy is a composite measure appropriately represented by the RMSD (%).When squared, RMSD can be partitioned into components re£ecting bias and precision: 1 Xn 1 Xn 2 ðy xi Þ2 ¼ d i¼1 i i¼1 i n n 2 2 2 2 ^ ^ ¼ d þ ^d ¼ Bias þ Precision ;
RMSD2 ¼
ð2Þ
Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
where d ¼ 1 n
Xn
d i¼1 i
and ^2d ¼
1 Xn ðd dÞ2 i¼1 i n
4 Device failure rate: the device failure rate was calculated as the number of times the device failed to produce a reading divided by the total number of observations. In evaluating performance, the percentage of times a device failed to produce a reading was considered jointly with RMSD. Pearson correlation coe⁄cients were calculated for the di¡erent probe sites for each species. The R2, co-coe⁄cient and intercept coe⁄cient are presented.
Results Results from dogs are presented in Table 1. The mean sampling time was 130 minutes. RMSD (%) for saturations ranged from 1.74 (NPB-40) to 2.72 (V3304). The NPB-40, NPB-190 and NPB-395 gave readings on all sites on all dogs while the NPB-290 failed to produce one reading and theV3304 failed to produce readings in 30 instances. RMSD for heart rates ranged from 0.95 to 1.14. Variability of SpO2 within plateaus was 1.1%, while variability between models was 3.6%. Variability within the co-oximeter was 0.3%. Mean values for carboxyhemoglobin, methemoglobin and total hemoglobin reported by the co-oximeter were 1.4, 0.4 and 10.4 g dL1. Results from horses are presented inTable 2. Mean sampling time was 106 minutes. RMSD for saturations ranged from 2.13 (V3304) to 4.70 (NPB-290). All monitors except the NPB-290 failed to produce readings some of the time; failure rates ranged from 10% (NPB-395) to 60% (V3304). RMSD for heart rates ranged from 0.63 to 0.91. Variability of SpO2 within plateaus was 1.6%, while variability between models was 3.5%. Variability within the co-oximeter was 0.1%. Mean values for carboxyhemoglobin, methemoglobin and total hemoglobin reported by the co-oximeter were1.4,1.2 and12.0 g dL1. Results from cats are presented in Tables 3a and 3b. Mean sampling time was 84 minutes. Table 3a presents all data while Table 3b summarizes data for SaO2 > 90%. For all data, RMSD for saturations ranged from 5.56 (NPB-190) to 10.65 (V3304) while failure to produce a reading ranged from rates of 0% (NPB-395 and NPB-290) to 31.5% (V3304). For SaO2 > 90% in cats, RMSD ranged from 2.55 (NPB395) to 4.82 (V3304) while failure to produce readings Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
ranged from 0% (NPB-395 and NPB-290) to 43% (V3304). RMSD for heart rates for all data ranged from 2.8 to 3.6 while RMSD for heart rates with SaO2 > 90% ranged from 2.6 to 4.9. Variability of SpO2 within plateaus was 1.9%, while variability between models was 5.7%. Variability within the co-oximeter was 0.6%, while mean values for carboxyhemoglobin, methemoglobin and total hemoglobin were1.3, 0.7 and 8.1 g dL1. Correlation coe⁄cients (R2), and intercepts for the various probe sites for each species are shown in Table 4. In horses, correlation for the tongue, nostril, prepuce/vulva, and lip were fairly high (>0.93) while the ear appeared to be least accurate. In dogs, correlation for the tongue, toe, prepuce/vulva were good (>0.95) while the lip and ear were less accurate. In cats, correlation for all sites was low; the rear paw had the highest correlation (0.79).
Discussion We believe this is the ¢rst multi-species desaturation study on pulse oximeters and the only study to provide information on cats. Previous studies have tended to compare two or three models of pulse oximeters with one or two probe types. Also, in the past analysis has generally consisted of bias, precision and regression analysis. Since RMSD is now considered to be the industry standard (J.C. Allen, personal communication 2000), we thought it would be most appropriate. It is interesting that monitors appeared to perform di¡erently on di¡erent species while the techniques and sampling methods were similar. For instance, for the NPB-40, RMSD was1.74 in dogs,3.25 in horses and 7.86 in cats. As the spectral properties of hemoglobin have been shown not to vary signi¢cantly from species to species (Grosenbaugh & Muir 1997) we are not sure how to account for this species variability. It is possible that the algorithms in di¡erent models are a¡ected by variations in heart rate or signal strength to varying degrees. Ambient light has been reported as a source of erroneous pulse oximeter readings (Huss et al. 1995). However, in this study all probes were covered, so extraneous light is not a likely source of error. Although the co-oximeter is considered to be the gold standard for comparison of SaO2 and SpO2, it should probably be scrutinized as a source of error. We calibrated the co-oximeter prior to each animal (low, normal, high and total Hb calibrating solutions) and used the correct species setting for each animal. 7
No. of observations
Accuracy analysis after excluding device failures and outliers
Outliers
Regression
Pulse oximeter model
n
Saturation (SpO2)
N-395 NPB-190 NPB-290 NPB-40 V3304
149 149 149 149 149
6 0 0 0 0
Pulse rate (PR)
N-395 NPB-190 NPB-290 NPB-40 V3304
149 149 149 149 149
0 0 0 0 0
Parameters
Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
a
nd
intercept
Slope
Biase
Precision accuracyf
RMSD
(4.0%) (0.0%) (0.0%) (0.0%) (0.0%)
1 5 5 6 3
(0.7%) (3.4%) (3.4%) (4.0%) (2.0%)
7 5 5 6 3
(4.7%) (3.4%) (3.4%) (4.0%) (2.0%)
142 144 143 143 116
2.47 0.07 0.17 0.39 18.34
1.02 0.98 0.99 0.99 0.80
0.80 1.52 1.26 0.61 0.83
2.55 1.62 2.04 1.64 2.60
2.67 2.22 2.39 1.74 2.72
0 0 1 0 30
(0.0%) (0.0%) (0.7%) (0.0%) (20.1%)
(0.0%) (0.0%) (0.0%) (0.0%) (0.0%)
7 7 8 7 9
(4.7%) (4.7%) (5.4%) (4.7%) (6.0%)
7 7 8 7 9
(4.7%) (4.7%) (5.4%) (4.7%) (6.0%)
142 142 140 142 110
0.79 0.31 0.13 0.32 0.59
1.01 1.00 1.00 1.00 1.00
0.04 0.05 0.15 0.11 0.55
1.15 1.12 1.04 0.94 0.98
1.14 1.12 1.05 0.95 1.12
0 0 1 0 30
(0.0%) (0.0%) (0.7%) (0.0%) (20.1%)
Dog data were combined from the following locations: ear, lip, toe, tongue, prepuce, and vulva. The clinical outliers were justified based on review by the investigator. c Identification of statistical outliers was based upon meeting at least three of four statistical criteria forextreme values. d Sample size device failures and outliers. Xafterexcluding n e d ¼ ð1=nÞ i¼1 di, where di ¼ yi xi ¼ SpO2 ðSaO2i þ SaO22i Þ=2: b
f
^d ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Xn 2: ð1=nÞ ðdi dÞ i¼1
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi g 2 2 þ ^2d ¼ Bias ^ ^ 2 þ Precision RMSD ¼ d :
Device failures to produce reading
Statisticalc Total
clinicalb
g
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
8 Table 1 Accuracy analysis of pulse oximeters in dogs (n ¼ 5)a
No. of observations
Accuracy analysis after excluding device failures and outliers
Outliers
Regression
Pulse oximeter model
n
Saturation (SpO2)
N-395 NPB-190 NPB-290 NPB-40 V3304
150 150 150 150 150
0 0 3 3 0
Pulse rate (PR)
N-395 NPB-190 NPB-290 NPB-40 V3304
150 150 150 150 150
0 0 6 0 0
Parameters
a
clinicalb
Statisticalc Total
nd
intercept
Slope
(0.0%) (0.0%) (2.0%) (2.0%) (0.0%)
2 5 3 3 1
(1.3%) (3.3%) (2.0%) (2.0%) (0.7%)
2 5 6 6 1
(1.3%) (3.3%) (4.0%) (4.0%) (0.7%)
133 113 144 119 59
11.92 9.67 0.39 7.15 7.22
0.84 0.87 0.96 1.05 0.9
(0.0%) (0.0%) (4.0%) (0.0%) (0.0%)
5 4 3 4 1
(3.3%) (2.7%) (2.0%) (2.7%) (0.7%)
5 4 9 4 1
(3.3%) (2.7%) (6.0%) (2.7%) (0.7%)
130 114 141 121 59
0.66 0.33 0.90 0.93 0.58
1.01 1.00 1.01 1.01 1.00
Horse data were combined from the followinglocations: ear, lip, nostril, tongue, prepuce, and vulva. The clinical outliers were justified based on clinical review. c Identification of statistical outliers was based upon meeting at least three of four statistical criteria forextreme values. d Sample size afterexcluding device failures and outliers. X n e d ¼ ð1=nÞ i¼1 di, where di ¼ yi xi ¼ SpO2i ðSaO21i þ SaO22i Þ=2: b
f
g
^d ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Xn 2: ð1=nÞ i¼1 ðdi dÞ
RMSD ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2 þ ^2 ¼ Bias ^ ^ 2 þ Precision d : d
Device failures to produce reading
Precision accuracy
RMSD
2.13 1.94 3.23 2.52 1.65
2.25 2.30 3.42 2.07 1.36
3.09 3.00 4.70 3.25 2.13
15 (10.0%) 32 (21.3%) 0 (0.0%) 25 (16.7%) 90 (60.0%)
0.18 0.21 0.21 0.23 0.66
0.66 0.64 0.89 0.59 0.63
0.68 0.68 0.91 0.63 0.91
15 (10.0%) 32 (21.3%) 0 (0.0%) 25 (16.7%) 90 (60.0%)
Biase
g
9
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
Table 2 Accuracy analysis of pulse oximeters in horses (n ¼ 5)a
No. of observations
Accuracy analysis after excluding device failures and outliers
Outliers
Regression
Pulse oximeter model
n
Saturation (SpO2)
N-395 NPB-190 NPB-290 NPB-40 V3304
149 149 149 149 149
5 0 0 0 0
Pulse rate (PR)
N-395 NPB-190 NPB-290 NPB-40 V3304
149 149 149 149 149
0 1 0 0 0
Parameters
Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
a
nd
intercept
Slope
Biase
Precision accuracy
RMSD
(3.4%) (0.0%) (0.0%) (0.0%) (0.0%)
6 7 3 1 5
(4.0%) (4.7%) (2.0%) (0.7%) (3.4%)
138 141 146 118 97
17.07 6.77 27.44 36.55 49.41
0.82 1.05 0.69 0.60 0.46
1.44 2.32 0.94 3.17 4.26
5.79 5.07 7.82 7.22 9.81
5.94 5.56 7.85 7.86 10.65
0 1 0 30 47
(0.0%) (0.7%) (0.0%) (20.1%) (31.5%)
(0.0%) (0.7%) (0.0%) (0.0%) (0.0%)
5 6 6 3 3
(3.4%) (4.0%) (4.0%) (2.0%) (2.0%)
96 93 95 72 61
1.77 0.24 1.11 4.49 3.02
1.01 1.00 1.00 0.97 0.98
0.58 0.10 0.46 0.15 0.44
3.46 3.57 3.47 2.88 2.82
3.49 3.55 3.48 2.86 2.83
48 49 48 74 85
(32.2%) (32.9%) (32.2%) (49.7%) (57.0%)
11 (7.4%) 7 (4.7%) 3 (2.0%) 1 (0.7%) 5 (3.4%) 5 7 6 3 3
(3.4%) (4.7%) (4.0%) (2.0%) (2.0%)
Cat data were combined from the following locations: ear, metacarpus, lip, metatarsus, and tongue. The clinical outliers were justified based on clinical review. c Identification of statistical outliers was based upon meeting at least three of four statistical criteria forextreme values. d Sample size afterexcluding device failures and outliers. X n e d ¼ ð1=nÞ i¼1 di, where di ¼ yi xi ¼ SpO2i ðSaO21i þ SaO22i Þ=2: b
f
g
^d ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Xn 2: ð1=nÞ i¼1 ðdi dÞ
RMSD ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2 þ ^2 ¼ Bias ^ ^ 2 þ Precision d : d
Device failures to produce reading
Statisticalc Total
clinicalb
g
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
10 Table 3a Accuracy analysis of pulse oximeters in cats (n ¼ 5)a
No. of observations
Accuracy analysis after excluding device failures and outliers
Outliers
Regression
Pulse oximeter model
n
Saturation (SpO2)
N-395 NPB-190 NPB-290 NPB-40 V3304
60 60 60 60 60
0 0 0 0 0
Pulse rate (PR)
N-395 NPB-190 NPB-290 NPB-40 V3304
60 60 60 60 60
0 0 0 0 0
Parameters
a
clinicalb
f
^d ¼
g
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Xn 2: ð1=nÞ i¼1 ðdi dÞ
RMSD ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2 þ ^2 ¼ Bias ^ ^ 2 þ Precision d : d
RMSD
0.88 2.39 2.00 0.59 1.33
2.42 3.71 3.45 3.45 4.7
2.55 4.39 3.96 3.46 4.82
0 1 0 15 26
(0.0%) (1.7%) (0.0%) (25.0%) (43.3%)
0.40 0.09 0.44 0.33 0.27
2.95 4.93 2.91 2.57 2.69
2.95 4.88 2.91 2.57 2.65
14 15 14 29 37
(23.3%) (25.0%) (23.3%) (48.3%) (61.7%)
nd
intercept
Slope
Biase
(0.0%) (0.0%) (0.0%) (0.0%) (0.0%)
1 2 2 3 0
(1.7%) (3.3%) (3.3%) (5.0%) (0.0%)
1 2 2 3 0
(1.7%) (3.3%) (3.3%) (5.0%) (0.0%)
59 57 58 42 34
21.78 8.84 36.55 18.44 141.95
0.76 0.88 1.36 1.19 0.49
(0.0%) (0.0%) (0.0%) (0.0%) (0.0%)
3 2 3 1 1
(5.0%) (3.3%) (5.0%) (1.7%) (1.7%)
3 2 3 1 1
(5.0%) (3.3%) (5.0%) (1.7%) (1.7%)
43 43 43 30 22
0.58 3.10 0.87 6.80 12.25
1.00 0.98 1.00 0.95 0.91
Cat data were combined from the followinglocations: ear, metacarpus, lip, metatarsus, and tongue. The clinical outliers were justified based on clinical review. c Identification of statistical outliers was based upon meeting at least three of four statistical criteria forextreme values. d Sample size device failures and outliers. Xafterexcluding n e d ¼ ð1=nÞ i¼1 di, where di ¼ yi xi ¼ SpO2i ðSaO21i þ SaO22i Þ=2: b
Device failures to produce reading
Precision accuracy
Statisticalc Total
g
11
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
Table 3b Accuracy analysis of pulse oximeters in catsa when SaO2 90% (n ¼ 5)a
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
Dog
Horse
Cat
Tongue Toe Prepuce/vulva Lip Ear Tongue Nostril Prepuce/vulva Lip Ear Tongue Front paw Rear paw Lip Ear
R2
Co-coefficient, m
Intercept coefficient, b
0.95 0.97 0.95 0.87 0.69 0.98 0.94 0.94 0.93 0.88 0.54 0.57 0.79 0.49 0.64
0.92 0.96 0.97 0.99 0.99 0.88 1.1 0.83 0.84 0.87 0.62 0.60 1.11 0.75 0.69
3.1 1.9 0.2 1.3 1.3 6.6 10.2 9.9 9.2 5.2 33.9 34.9 13.5 17.8 27.9
Table 4 Correlation coe⁄cients for pulse oximeter sites ^ all models included in calculations for each site
Y ¼ mx þ b.
However, we have no information about generation of the extinction coe⁄cients for the di¡erent species within the co-oximeter. The co-oximeter uses ¢ve wavelengths of light where the pulse oximeter only uses two. Samples were run through the co-oximeter in duplicate and averaged for the analysis. Ideally, triplicate analysis might have been more accurate, but would have contributed to a longer delay before analysis of each sample, which might have introduced further error due to re-mixing and storage of samples. Calculated variability within the co-oximeter varied from 0.1% in horses to 0.6% in cats, which would seem to indicate that it was less accurate in the cats. Again, this may have been due to the small size of samples collected. Total hemoglobin values reported for the cats were also low compared to dogs and horses, even though all cats were clinically normal; complete blood counts taken within 1 month before and after the study showed total hemoglobin values in the range of 10^14 g dL ^1. This may also indicate inaccuracy in the co-oximeter readings during the study, which could have been due to small sample size or dilutional e¡ects of the heparin pellet. As explained in the statistical methods, outliers were quali¢ed by clinical review or by diagnostic statistics. Clinical review consisted of checking experimental notes for reasons that would contribute to an inaccurate reading not related to the monitor. In the horses, probes were readjusted when one slid o¡ the vulva (NPB-40) and when the motion detector light repeatedly came on (NPB-290). In dogs, repositioning 12
of a probe revealed an indentation of lip tissue (NPB-395) and a probe was repositioned on the vulva (NPB-395). In cats, a probe sliding o¡ a small lip was repeatedly repositioned (NPB-395). Probe placement and sites have been previously shown to contribute to variability between SpO2 and SaO2 (Jacobson et al. 1992; Cha⁄n et al. 1996). Since the objective of this study was to evaluate the overall performance of the pulse oximeter model, we chose not to draw conclusions about the accuracy of a particular site, since one would then have to assume that all monitors performed equally. Variability in sites is probably a large cause of the variability between pulse rate measured by the pulse oximeter and heart rate displayed by the ECG. Signal strength was very likely di¡erent from site to site, which would be a cause for disparities between pulse rate and displayed heart rate. In cats, the small ECG complexes may have a¡ected how well the ECG counted heart rate. For most monitors, accuracy of the saturation reading is linked to accuracy of the pulse rate; if the monitor is not able to detect a good pulse signal, no saturation will be displayed. Heart rates displayed on the ECG could be di¡erent from pulse rates for many reasons, including the time over which ECG is measured and multiplication factors. The overall failure rate of the V3304 to produce readings was much higher than for the other monitors. This may have been because the probe is intended for lingual use only (Surgi-VetV3304 Operation Manual 1999,Version 3, pp. 4^9), although it did function on nonlingual sites. Failure to produce any Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
readings occurred in the cat when the probe was placed on the metacarpal or metatarsal region, in the dog when the probe was placed on the toe, and in the horse when the probe was placed on the lip, ear or vulva. A previous desaturation study in horses reported a bias of 0.55 and precision of 2.57, which is similar to the precision reported in this study of1.4^3.4%. However, the previous study evaluated only one site (the tongue) with one monitor (Martinez et al. 1999) and only 60 readings were taken compared to 150 readings in this study. In the present study, two monitors failed to produce readings during the ¢rst 45^ 60 minutes after induction in horses. In one horse, the NPB-40 (probe on ear) and NPB-395 (probe on nostril) did not produce readings during the ¢rst desaturation period, but produced readings during the second set of desaturations without touching or repositioning the probes. In a second horse, the NPB-40 (probe on prepuce) did not produce readings during the ¢rst desaturation period, but gave readings during the second desaturation period. The NPB-190 (probe on nostril) failed to produce readings at any time on this horse. Both of these horses had deeply pigmented (i.e. black) skin. Dark pigmentation has been previously suggested to be a source of failures and inaccuracies with pulse oximeters in animals (Jacobson et al. 1992), but we speculate that local vasoconstriction produced by the induction drugs may have caused the temporal failures. Alternatively, the combination of dark pigmentation with local vasoconstriction may have produced the failures. Half (13 of 25) of the sites used were pigmented black; of these, four sites failed to produce readings, with three of the four being attached to the same monitor. All nonblack sites (including brown, grey and speckled pigmentation) gave readings. The 60% failure rate seen with the V3304 in the horse was higher than for other monitors, and did not change over time. Again, this may have been due to probe placement; the monitor did produce readings 100% of the time when the probe was placed on the tongue (as is the manufacturers recommendation). Hypotension, associated with poor peripheral perfusion, has also been shown to a¡ect pulse oximeter readings (Reich et al.1996). However, we maintained a mean blood pressure >60 mm Hg in the horses by infusing dobutamine and blood pressure was maintained in the dogs and cats by adjusting anesthetic depth and/or increasing £uid administration. It might have been useful to measure blood pressure directly in the dogs and cats, since the accuracy of Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14
this method is better than indirect measurement. However, we would not have been able to get readings during the times when samples were drawn (which would have been the times of most interest). An increase in blood pressures was seen at the lower saturations (as would be expected with hypoxia), however, since these blood pressures were not recorded we have not presented them as data.We do not believe that hypotension produced pulse oximeter failures in this study. We cannot explain why the accuracy of all monitors was so much poorer in cats than in dogs or horses, when the same techniques and methodology were used for sampling. It is possible that varying the induction drugs may have introduced some error into the cat data. Saturation values in the cats also appeared to be more unstable at the lower saturation plateaus than the dogs or horses presumably because of their smaller size. This instability would contribute to error. It is also possible that the smaller blood samples drawn from the cats contributed to errors in the co-oximeter. Although the co-oximeter requires only 35 mL of blood and our sample size was 0.3 mL, mixing and dilutional errors may have occurred. Arterial catheters were placed peripherally since these could be placed percutaneously without requiring a ldquor;cut-down and with the thought that the pulse oximeter probes were also being placed peripherally. However, given the large variability, we wonder if cats may vasoconstrict peripheral sites at varying rates. Changes in sympathetic tone associated with decreased oxygen tension may have a¡ected blood £ow to peripheral sites. This would also explain why the di¡erences between pulse rates from the pulse oximeter were di¡erent from ECG heart rates in the cats; weak pulses might not be counted by the pulse oximeter. The higher heart rates seen in the cats might also contribute to greater error between pulse rates and ECG heart rates. However, the errors appeared to be greater in cats than those observed in dogs and horses. The NPB-40 failed to produce readings on one cat (probe placed on the metatarsal site). Percutaneous placement of the arterial catheter in the dorsal pedal artery had been unsuccessful in that foot so it is possible that blood £ow may have been disrupted su⁄ciently to cause the failures. Pulse oximeter failures have been reported to be useful for evaluating compromised circulation (Dorsch & Dorsch 1999). However, no perfusion problems were observed after recovery of the cat from anesthesia. 13
An evaluation of pulse oximeters in dogs, cats and horses NS Matthews et al.
It is interesting that the accuracy of all monitors was reasonably good ( 3^5%) for cats when SaO2 was greater than 90%, although unfortunate that none of the models were very accurate for lower saturations. Since this is the range most commonly seen in clinical patients, it seems to provide important information about use of these models in cats. It would seem that further studies of the accuracy of pulse oximeters in cats should be performed.
Conclusions Accuracy of the monitors evaluated varied from species to species. In the dog, the RMSD for all monitors was 2^3%; the best monitor appeared to be the NPB-40, followed by the NPB-190, 290, 395 and V3304. In horses, the RMSD for all monitors was 2^5%; if the V3304 is excluded due to the high failure rate, the best monitor appeared to be the NPB190 followed by the NPB-395, NPB-40 and NPB290. In cats, accuracy over the entire range of saturations evaluated (70^100%) was not particularly good; RMSD for all monitors was 6^11%. In cats, the best monitor appeared to be the NPB-190, followed by NPB-395, NPB-290, NPB-40 and V3304. In the clinically normal range (i.e. saturations >90%), accuracy was similar to that seen in the dogs and horses; RMSD for all monitors for this range was 3^5%. However, based on this study, caution is advised when interpreting pulse oximeter values in cats. When the V3304 gave readings, they were reasonably accurate, but the monitor failed to produce readings 20% of the time in dogs, 30% of the time in cats and 60% of the time in horses, which was much higher than for other monitors.
Acknowledgements Funding and equipment for this study were provided by Mallinckrodt Inc., Pleasanton, CA, USA.
References Barton LJ, DeveyJJ, Gorski S et al. (1996). Evaluation of transmittance and re£ectance pulse oximetry in a canine
14
model of hypotension and desaturation. J Vet Emerg Crit Care6, 21^28. Belsey DA, Kuh E,Welsch RE (1980) Regression Diagnostics. JohnWiley & Sons, NewYork, NY. Cha⁄n MK, Matthews NS, Cohen N et al. (1996) Evaluation of pulse oximetry in anaesthetised foals using multiple combinations of transducer type and transducer attachment site. EqVet J28, 437^445. DorschJA, Dorsch SE (1999) Pulse oximetry. In: Understanding Anesthesia Equipment, 4th edn. Dorsch JA, Dorsch SE (eds). Williams & Wilkins, Philadelphia, PA, USA, pp. 811^847. Fairman NB (1993) Evaluation of pulse oximetry as a continuous monitoring technique in critically ill dogs in the small animal intensive care unit. J Vet Emerg Crit Care2, 50^56. Grosenbaugh DA, Muir WW (1997) Absorbance spectra of inter-species hemoglobins in the visible and near infrared regions. J Vet Emerg Crit Care 7,36^42. Grosenbaugh DA, MuirWW (1998) Accuracy of noninvasive oxyhemoglobin saturation, end-tidal carbon dioxide concentration, and blood pressure monitoring during experimentally induced hypoxemia, hypotension, or hypertension in anesthetized dogs. Am J Vet Res 59, 205^212. Huss BT Anderson MA, Branson KR et al. (1995) Evaluation of pulse oximeter probes and probe placement in healthy dogs. J Am Anim Hosp Assoc 31,9^14. Jacobson JD, Miller MW, Matthews NS et al. (1992) Evaluation of accuracy of pulse oximetry in dogs. Am J Vet Res 53,537^540. Kleinbaum DG, Kupper LL, Muller KE (1988). Applied Regression Analysis and other Multivariate Methods. PWS-KENT Publishing Co., Boston, MA. Martinez EA, Carroll GL, Harts¢eld SM (1999). Evaluation of the Nonin 8600V veterinary pulse oximeter in anesthetized horses. J Vet Emerg Crit Care 9,13^17. Matthews NS, Sanders EA, Harts¢eld SM et al (1995) A comparison of 2 pulse oximeters in dogs. J Vet Emerg Crit Care 5,116^120. Reich DL, Timcenko A, Bodian CA et al. (1996) Predictors of pulse oximetry data failure. Anesthesiology 84, 859^864. Schmitt PM, Gobel T,Trautvetter E (1998) Evaluation of pulse oximetry as a monitoring method in avian anesthesia. J Avian Med Surg12,91^99. Received 30 July 2001; accepted 8 July 2002.
Veterinary Anaesthesia and Analgesia, 2003, 30, 3^14