Monitoring of respiratory rate in postoperative care ... - Springer Link

MONITORING OF RESPIRATORY RATE IN POSTOPERATIVE CARE USING A NEW PHOTOPLETHYSMOGRAPHIC TECHNIQUE

Lena Nilsson, MD,1 Anders Johansson, PhD 2 and Sigga Kalman MD, PhD 1

Nilsson L, Johansson A, Kalman S. Monitoring of respiratory rate in postoperative care using a new photoplethysmographic technique. J Clin Monit 2000; 16: 309^315

ABSTRACT. Objective. Photoplethysmography (PPG) is a non-invasive optical technique that measures variations in skin blood volume and perfusion. The PPG signal contains components that are synchronous with respiratory and cardiac rhythms. We undertook this study to evaluate PPG for monitoring patients' respiratory rate in the postoperative care unit, using a new prototype device. We compared it with the established technique, transthoracic impedance (TTI). Methods. PPG signals from 16 patients (ASA classes 1^2, mean age 43 years) who were recovering from general anaesthesia after routine operations were recorded continuously for 60 minutes/patient. The respiratory synchronous part of the PPG signal was extracted by using a bandpass ¢lter. Detection of breaths in the ¢ltered PPG signals was done both visually and by using an automated algorithm. In both procedures, the detected breaths were compared with the breaths detected in the TTI reference. Results. A total of 10.661 breaths were recorded, and the mean SD respiratory rate was 12.3 3.5 breaths/minute. When compared with TTI, the rates of false positive and false negative breaths detected by PPG (visual procedure) were 4.6 4.5% and 5.8 6.5%, respectively. When using the algorithm for breath detection from PPG, the rates of false positive and false negative breaths were 11.1 9.7% and 3.7 3.8%, respectively, when compared to TTI. Lower respiratory rates increased the occurrence of false-positive breaths that were detected by the PPG using visual identi¢cation (p < 0.05). The same tendency was seen with the automated PPG procedure (p < 0.10). Conclusions. Our results indicate that PPG has the potential to be useful for monitoring respiratory rate in the postoperative period. KEY WORDS. Photoplethysmography, non-invasive measurement, anaesthesia recovery period, respiratory rate.

INTRODUCTION

From the Departments of 1Anaesthesiology and Intensive Care, and 2 Biomedical Engineering, Linko«ping University, Linko«ping, Sweden. Received May 1, 2000, and in revised form Nov 9, 2000. Accepted for publication Nov 14, 2000. Address correspondence to Lena Nilsson, MD, Department of Anaesthesiology and Intensive Care, Linko«ping University Hospital, S-581 85 Linko«ping, Sweden. E-mail: [email protected] Journal of Clinical Monitoring and Computing 16: 309^315, 2000. ß 2001 Kluwer Academic Publishers. Printed in the Netherlands.

Continuous supervision of respiration is of great importance in postoperative care because of the risk of central or obstructive apnoea, caused by opioid analgesics, neuromuscular blocking agents, and disturbances of respiratory mechanics after abdominal operations [1, 2]. It is also important to detect postoperative hypoxemia, as this is thought to contribute to events such as myocardial ischemia and mental confusion [3, 4]. Respiratory rate in spontaneously breathing patients is usually monitored by intermittent manual counting, or continuously by transthoracic impedance (TTI) via ECG electrodes [5]. Devices based on air£ow detection [6, 7] or chest girth [8, 9] have also been used. Arterial blood oxygen saturation is usually monitored by pulse

310

Journal of Clinical Monitoring and Computing Vol 16 No 4 2000

oximetry [10], the value of which has been established in several studies in the postoperative period [11, 12] and is recommended for postoperative care [13]. However, if a patient is given oxygen, desaturation can be a late sign of apnoea. It is therefore essential to monitor not only oxygen saturation but also respiratory rate. Changes in the respiratory rate can also be a sensitive marker of impending respiratory dysfunction. Photoplethysmography (PPG) is a well-known optical technique that is used to follow variations in skin blood volume and perfusion. Illumination of the skin results in absorption and scattering of the incident light. The intensity of the transmitted or re£ected light, detected by a photodetector, measures blood content and £ow in the target skin volume [14]. This technique is widely used in pulse oximetry. However, the PPG signal contains components synchronous not only with cardiac rhythm, but also with respiratory rhythm. The respiratory component, sometimes referred to as RIIV (respiratory induced intensity variation), is caused by respiratory synchronous blood volume variations, probably predominately on the venous side of the circulation [15^17]. By extracting the cardiac- and respiratory related components, and by applying a mathematical algorithm, we have developed a new PPG device to monitor heart and respiratory rate simultaneously [18^20]. The autonomic nervous system modulates skin blood perfusion and therefore the PPG signal [21]. Conditions such as pain, hypovolemia, hypothermia, and the residual e¡ect of anaesthetic agents in£uence the autonomic nervous system. These conditions are common in patients after surgery. The PPG method for respiratory rate monitoring has given good results in healthy resting adults under experimental conditions [20], but it is also essential to evaluate its use in postoperative patients who may be under the in£uence of the conditions mentioned above. If PPG proves to be a useful means for monitoring respiratory rate in postoperative patients it may be possible to design a single non-invasive sensor to monitor respiratory rate, heart rate, and arterial oxygen saturation simultaneously. The purpose of the present investigation was to determine the accuracy of the new PPG device for monitoring respiratory rate in postoperative patients after routine operations. METHODS AND MATERIALS Sixteen patients (10 women, 6 men, classi¢ed according to American Association of Anesthesiologists as healthy or having a mild systemic disease, ASA group 1 or 2),

Fig. 1. The optical sensor used to generate the PPG signal.

aged 17^78 (median 40, mean 43) years were studied in the postoperative care unit at Linko«ping University Hospital, Sweden. All patients gave informed consent to participate in the study, and the Ethics Committee, Linko«ping University Hospital, approved the study. The patients were randomly chosen and had had various types of routine operations, mostly orthopaedic or abdominal. No patient included in the study was later excluded from the analysis. All patients had received general anaesthesia, using mainly inhalation anaesthetics. Fourteen patients were given muscle relaxants and all were given opioids during anaesthesia. Continuous measurements were made for 60 minutes in each patient during the postoperative period. The ambient room temperature was 22^24 ³C. Patients were cared for by the regular nursing sta¡ and all interventions needed in nursing were permitted during the measurements, so as to evaluate the device in conditions that were as realistic as possible. We used a non-commercial PPG device with optical sensor and electronics constructed in our laboratory [20, 22]. The optical sensor (Figure 1) emitted a single infrared wavelength (880 nm). The sensor was ¢xed with double-sided adhesive tape on the medial side of the forearm (right arm when possible). Skin temperature close to the sensor site was measured (Linear Laboratories, C-600M, U.S.A.). The signal from the optical sensor was ampli¢ed after subtraction of its baseline level (DC) representing non-pulsatile re£ection [20]. The subtraction level was manually set whenever the ampli¢er output got out of range (DC adjustment). The PPG-signal was converted from analogue to digital form (100 Hz, DAQCard-1200, National Instruments, U.S.A.) and stored in a laptop computer (Acer 970CX, Taiwan). The TTI surface electrodes of the monitoring system (Hewlett Packard, model 54S, U.S.A.) were positioned according to the manufacturer's recommen-

Nilsson et al: Monitoring of Respiratory Rate in Postoperative Care

311

dations. The analogue TTI output of the monitor was recorded and stored together with the PPG signal. LabView software (v 4.1) was used for data acquisition.

Data analysis The evaluation procedure was divided into ¢ve parts: 1. Extraction of the PPG respiratory signals (RIIV) The recording from each patient (60 minutes) was divided into 30 two-minute blocks. Each block was ¢ltered with a bandpass ¢lter (0.13^0.48 Hz, 16th degree Bessel ¢lter), which suppressed the cardiac-related variations and the frequencies below the respiratory frequency in the PPG signal. A characteristic example of a PPG signal before and after the ¢ltering process is shown together with the corresponding TTI signal in Figure 2. 2. Identi¢cation of poor quality blocks of TTI and PPG Blocks in which it was di¤cult to identify single breaths in the TTI signal could not be used as a reference. Such blocks were identi¢ed visually, counted, and removed from further analysis. Blocks containing adjustments of the baseline level of the PPG signal were also counted and removed, as it is not possible to evaluate the signal during these events. 3. Visual comparison of theTTI and RIIV signals In each of the remaining blocks, the breaths of the TTI and RIIV signals were identi¢ed. Those that were detected in the TTI but not in the RIIV signal were recorded as missed (false negatives), while breaths detected in the RIIV but not in the TTI, were regarded as overdetected (false positives). 4. Automated comparison of theTTI and RIIV signals To illustrate the performance of a prospective automated system, we tested a simple breath detection algorithm. The extreme values (peaks) of each curve were used to identify the breaths. To suppress the cardiac artefact of the TTI signal, the same bandpass ¢lter as described above was used to ¢lter the signal. The number of breaths in each block and in both signals was counted and the di¡erence between them was noted as the number of missed or overdetected breaths. 5. Statistics The rates of missed and overdetected breaths are given as mean SD. Furthermore, the e¡ect of respiratory rate and age on these error rates was examined by linear regression. Separate regressions were performed for each of the two variables and for each (mean) error rate Yi:

Fig. 2. A characteristic block of signals: TTI signal (top), PPG signal (middle) and RIIV signal (bottom). RIIV signal is obtained by bandpass ¢ltering of the PPG signal. Upward de£ections indicate inspiration.The vertical axis is in arbitrary units.

Y1: The rate of overdetected breaths using visual detection Y2: The rate of overdetected breaths using automated detection Y3: The rate of missed breaths using visual detection Y4: The rate of missed breaths using automated detection Eight regression analyses were performed to evaluate the impact of respiratory rate and age on each of the error rates. Yi Ai Bi RR "i

n 16

1

Yi Ci Di AGE "i

n 16

2

RR is the patient's mean respiratory rate during the measurement and AGE is the patient's age. The residuals ("i ) were considered normally distributed in both models. A, B, C and D are the regression coe¤cients. The in£uence of the variable sex on the error rates (Yi ) was also investigated (Student's t-test). Di¡erences were considered signi¢cant at p < 0.05. RESULTS Of the 2-minute blocks, 19 20% (mean SD) contained disturbances of the TTI signal, and 7 9% DC adjustments of the PPG signal. These blocks, sometimes coexisting, were excluded, resulting in 76 20 % of the blocks being used for comparison. The number of breaths detected in the TTI reference signal in these blocks was 10.661, equivalent to a mean respiratory rate of 12.3 3.5 breaths/minute. The false positive and false negative rates of the visual and the automated evaluation procedures are shown in Table 1. Skin temperature was 31.1 1.3 ³C.

312


In£uence of respiratory rate The result of the regression analysis investigating the PPG accuracy at di¡erent respiratory rates is shown in Table 2. There was a signi¢cant tendency for PPG to overdetect breaths at lower respiratory rates with the visual assessment of the waveforms compared with the TTI reference (p < 0.05). The same tendency was seen with the automated PPG procedure (p < 0.10).

In£uence of age and sex Regression analysis showed that age did not exert any signi¢cant in£uence on overdetected or missed breaths (pi > 0.20). Sex did not in£uence the results (pi > 0.19). DISCUSSION Measurement of arterial oxygen saturation is not su¤cient to detect all respiratory disturbances after anaesthesia and surgery. Monitoring of the respiratory rate is also important. The possibility of combining these measurements in one skin sensor makes PPG attractive as a method for postoperative monitoring. PPG has been used experimentally to measure respiratory rate in resting healthy subjects [20]. In the present study it was evaluated for the ¢rst time postoperatively in adult patients. We found 5.8 6.5% false negative recordings of breaths and 4.6 4.5% false positive recordings. From a clinical point of view overdetection of breaths is more alarming than missed detection. The e¡ect of respiratory rate on the incidence of false positive breath detection is disturbing, as a low respiratory rate is one of the conditions that is important to detect postoperatively. One reason for overdetection in the RIIV signals was the frequent occurrence of waves with double peaks resulting from only one breath (Figure 3). This phenomenon might be the result of detecting respiratory variation from both the arterial and venous sides of the circulation where a phase di¡erence might indicate two breaths. Another possible explanation is disturbances from lower frequency variations in the blood perfusion, re£ecting brainstem activity or a resonance phenomenon within the barore£ex loop [23]. We showed that age, and sex had no in£uence, but further studies are needed to investigate possible in£uences of these and other physiological variables. We restricted the duration of the blocks that we analysed to two minutes. This gave 30 blocks for each patient. In the visual analysis the duration of the blocks was of no importance. For the automated analysis there

Table 1. Rates of overdetected and missed breaths for the visual and the automated procedures. 76 20% of the blocks were considered (n = 16)

Visual detection Automated algorithm

Overdetected breaths

Missed breaths

Mean (%)

SD

Mean (%)

SD

4.6

4.5

5.8

6.5

11.1

9.7

3.7

3.8

Table 2. The accuracy of PPG at di¡erent respiratory rates; summary of the regression analysis. The rates of overdetecting and missing breaths by using the two ways of PPG breath detection (visual/automatic) are compared to the respiratory rate, as calculated from the transthoracic impedance reference (TTI). The signi¢cant e¡ect (p < 0.05) is indicated by boldface lettering. A and B are the regression coe¤cients of model 1 p

R

A

B

Overdetected breaths Visual detection Automated algorithm

0.03 0.09

0.55 0.43

12.5 24.7

^0.71 ^1.22

Missed breaths Visual detection Automated algorithm

0.68 0.68

0.11 0.11

3.50 2.32

0.21 0.12

Fig. 3. Characteristic artefact of double peaks in the RIIV signal. This e¡ect increases the number of overdetected breaths particularly with automated breath detection and at lower respiratory rates. TTI signal (top), PPG signal (middle) and RIIV signal (bottom).

was a risk that false negative and false positive breaths could occur in the same two-minute block, thus giving a better net result of the number of breaths counted. A block that is too short would have made the data


di¤cult to handle, and the duration that we chose was a compromise. Our experience was, however, that false negative and false positive breaths did not often occur together in a period of two minutes. The PPG device that we used has recently been evaluated in newborn infants for monitoring heart and respiratory rates [18]. Those authors found a lower rate of false negative (2.7 1.1%) and false positive (1.5 0.4%) recordings of breaths than we did. In their study, however, only 29% of the recorded time included representative TTI signals. In our study, as much as 76% of the recorded two-minute blocks, could be used for comparison. We also found large variations in the accuracy of breath detection between patients, which explains the large SDs. Small babies have a respiratory rate of about 30^60 breaths/minute, whereas the adult respiratory rate is around 10^20 breaths/minute. Disturbances of the TTI reference signal as a result of movement seems more frequent in babies. In adults on the other hand, many TTI signals could be used as reference signals, even though partly disturbed by movement. When we removed PPG blocks that were clearly a¡ected by movement or lower frequency perfusion variations (which are well known problems in pulse oximetry) it resulted in a reduction in the rate of false negative and false positive recordings of breaths (visually) to 3.4 4.5% and 2.7 2.9%, respectively. These are in the same range as those reported when PPG was used in babies. We aim to reduce the in£uence of these factors by improvement of the optical sensor and the digital ¢lter [24]. The simple algorithm for breath detection we used had a tendency to overdetect breaths (Table 1). A more advanced algorithm for automatic detection of breaths based on neural networks has been evaluated [Johansson, personal communication]. We used infrared light, which reaches a large volume of blood because of low absorption, thus improving the RIIV signal. Previous experimental studies in our laboratory have shown that the RIIV signal is usually strongly expressed when an optical re£ection mode sensor is positioned on the forearm. This is probably because cutaneous arterial pulses have less in£uence at this location [25], and the respiratory-induced variations in the venous circulation become more pronounced. In more peripheral parts of the body such as ¢ngers and toes the sympathetic out£ow is greater [26]. Sympathetic activity in£uences the PPG signal by a¡ecting blood perfusion, and if strongly expressed, there might be a risk of interference of the RIIV signal. Combining RIIV recording and pulse oximetry, the latter based on arterial pulsation, seems possible as both techniques are based on PPG. In small babies it was possible to record respiratory rate and heart rate simul-

313

taneously by using PPG [18]. There should be no technical obstacles in combining the systems, but probably sensor placement on other parts of the body than the forearm will prove superior. Experiments with the sensor on the forehead are promising in this respect [Johansson, personal communication]. A preliminary report was published recently, which showed that respiratory rate derived from a pulse oximeter placed on a ¢nger, was as accurate as capnography and visual observation and better than TTI [27]. TTI was chosen as the reference method for detection of breaths. This is the technique that is in routine use in our hospital and represents an often-used method in postoperative care. Under experimental conditions the PPG method detected an almost identical number of breaths as TTI [20]. As we wanted to evaluate the PPG technique and not the TTI method, we took care to establish as good TTI recording conditions as possible for each patient before starting the measurements. All data containing TTI records of low quality were excluded from further analysis. This was a subjective judgement by one of the authors (AJ) and almost one ¢fth of the blocks were withdrawn for this reason. If we had not divided the recording time into blocks, 10 13% of the registration time would have included low quality TTI reference signals. Our problems with TTI recording were similar to those reported in a study by Drummond et al. in 1996, in which TTI was used to measure respiration in postoperative patients after major abdominal operations [5]. In their study the TTI waveform was considered poor for a mean of 16% of the time in each patient, with great individual di¡erences. In another study in a postoperative care unit, 28% of TTI alarms were considered false [28]. Bandaging after some operations makes accurate positioning of the ECG electrodes used to record TTI impossible and it can sometimes be di¤cult to establish good skin/electrode contact. ECG monitoring alone has a high incidence of false alarms [28] and is of limited sensitivity for many postoperative disorders. The new PPG device is noninvasive and the optical sensor is small and can usually be positioned easily. It can also be developed to include measurements of arterial oxygen saturation. This combination is probably more useful in postoperative care than the usual combination of ECG and TTI. The present study represents a ¢rst clinical evaluation of a prototype device. Further development in our laboratory is necessary. A lesser part of the data (7%) included manual DC adjustment and was removed. In a forthcoming prototype, DC adjustment will be automatically performed and an automatic gain control system will minimise the occurrence of this event. Disturbances by patient movement are well known

314


problems in PPG, but methods for reduction are underway [24]. In order to document the accuracy of the PPG prototype during motion, evaluation using an air£ow reference (capnography, pneumotachography) is needed. Further studies are also required to evaluate the in£uence of the autonomic nervous system as well as the e¡ect of blood volume status (hypovolemia, hypervolemia) on the RIIV signal.

10. 11. 12. 13.

CONCLUSION This study in unselected adult patients after operation indicates that the PPG method has potential as a tool for monitoring respiratory rate. The possibility of combining respiratory rate measurements and pulse oximetry is a favourable aspect of the new technique. We expect that improvements in the sensor and automated algorithm will make the device more accurate and useful clinically. The Swedish National Centre of Excellence for NonInvasive Medical Measurements (NIMED) gave support to this work.

REFERENCES 1. Jones JG, Sapsford DJ, Wheatley RG. Postoperative hypoxaemia: mechanisms and time course. Anaesthesia 1990; 45: 566^573 2. Nimmo AF, Drummond GB. Respiratory mechanics after abdominal surgery measured with continuous analysis of pressure, £ow and volume signals. Br J Anaesth 1996; 77: 317^326 3. Gill NP, Wright B, Reilly CS. Relationship between hypoxaemic and cardiac ischaemic events in the perioperative period. Br J Anaesth 1992; 68: 471^473 4. Rosenberg J, Kehlet H. Postoperative mental confusion association with postoperative hypoxemia. Surgery 1993; 114: 76^81 5. Drummond GB, Nimmo AF, Elton RA. Thoracic impedance used for measuring chest wall movement in postoperative patients. Br J Anaesth 1996; 77: 327^332 6. Cyna AM, Kulkarni V, Tunstall ME, et al. Aura: A new respiratory monitor and apnoea alarm for spontaneously breathing patients. Br J Anaesth 1991; 67: 341^345 7. Gordh T, Rawal N, Stro«m S, Ho«k B. Respiratory monitoring during postoperative analgesia. J Clin Monit 1995; 11: 365^372 8. Cohn MA, Rao ASV, Broudy M, et al. The respiratory inductive plethysmograph: A non-invasive monitor of respiration. Bull Eur Physiopathol Respir 1982; 18: 643^658 9. Anderson W, Brock-Utne AJ, Brock-Utne JG, Brodsky

14. 15.

16.

17. 18.

19.

20. 21.

22.

23. 24. 25.

26.

JB. Evaluation of a respiratory rate monitor in postsurgical patients. J Clin Anesth 1992; 4: 289^291 Tremper KK, Barker SJ. Pulse oximetry. Anesthesiology 1989; 70: 98^108 Mo«ller JT, Johannessen NW, Espersen K, et al. Randomized evaluation of pulse oximetry in 20,802 patients: II. Anesthesiology 1993; 78: 445^453 Dumas C, Wahr JA, Tremper KK. Clinical evaluation of a prototype motion artifact resistant pulse oximeter in the recovery room. Anesth Analg 1996; 83: 269^272 Standards of the American Society of Anesthesiologists. Standards for postanesthesia care. Last amended on October 19, 1994; http://www.asahg.org Kamal AAR, Harness JB, Irving G, Mearns AJ. Skin photoplethysmography ^ a review. Comp Methods Prog Biomed 1989; 28: 257^269 Nilsson L, Johansson A, Svanerudh J, Kalman S. Is the respiratory component of the photoplethysmographic signal of venous origin? Med Biol Eng Comput 1999; 37 (Suppl 2): 912^913 Ugnell H. The respiratory synchronous photoplethysmographic signal. Its dependence on light wavelength and sample volume. Med Biol Eng Comput 1996; 34 (Suppl 1, pt1): 275^276 Dorlas JC, Nijboer JA. Photo-electric plethysmography as a monitoring device in anaesthesia. Br J Anaesth 1985; 57: 524^530 Johansson A, Oëberg PAî, Sedin G. Monitoring of heart and respiratory rates in newborn infants using a new photoplethysmographic technique. J Clin Monit 1999; 15: 461^467 Johansson A, Nilsson L, Kalman S, Oëberg PAî. Monitoring of Respiratory Rates using Photoplethysmography, 20th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, Hong Kong, 1998, Report No. 3226 Lindberg L-G, Ugnell H, Oëberg PAî. Monitoring of respiratory and heart rates using a ¢bre-optic sensor. Med Biol Eng Comput 1992; 30: 533^537 Larsen PD, Harty M, Thiruchelvam M, Galletly DC. Spectral analysis of AC and DC components of the pulse photoplethysmograph at rest and during induction of anaesthesia. Int J Clin Monit Comput 1997; 14: 89^95 Johansson A, Oëberg PAî. Estimation of respiratory volumes from the photoplethysmographic signal. Part 1: Experimental results. Med Biol Eng Comput 1999; 37: 42^47 Galletly DC, Williams TB, Robinson BJ. Periodic cardiovascular and ventilatory activity during midazolam sedation. Br J Anaesth 1996; 76: 503^507 Hayes MJ, Smith PR. Artifact reduction in photoplethysmography. Applied Optics 1998; 37: 7437^7446 Tur E, Tur M, Maibach HI, Guy RH. Basal perfusion of the cutaneous microcirculation: Measurements as a function of anatomic position. J Invest Dermatol 1983; 81: 442^446 Ziege S, Schmid-Scho«nbein H, Grebe R, Martin E. Long-term registration of cutaneous microcirculation during general anesthesia. Int J Microcirc Clin Exp 1997; 17: 385^394


27. Hall JM, Lampotang S, Thoman J, et al. A continuous respiratory rate monitor derived from the optoplethysmogram of a pulse oximeter: Clinical evaluation. Anesthesiology 1998; 89 (3A): A 971 28. Wiklund L, Ho«k B, Sta®hl K, Jordeby-Jo«nsson A. Postanesthesia monitoring revisited: frequency of true and false alarms from di¡erent monitoring devices. J Clin Anesth 1994; 6: 182^188

315