International Journal of Clinical Monitoring and Computing 7: 45-57, 1990. (~) 1990 Kluwer Academic Publishers. Printed in the Netherlands.
Development of software for pulse oximeter and investigation of its realtime response in clinical environment S. Meiyappan & Omar Prakash
Erasmus University, Postbus 1738, 3000 DR Rotterdam, The Netherlands Accepted 15 January1990
Key words: pulse oximetry, plethysmograph, SaO2 software, oxyhaemoglobin, realtime data collection, computerized charting Abstract
We developed a pulse oximeter software at Thorax Centre, Erasmus University as a joint project with an industry and evaluated it in the clinical environment of thorax anaesthesia using a computerized protocol for realtime data collection during routine clinical procedures. This paper gives an account of the results we obtained from the development project and the clinical study. The paper consists of two parts. First part describes different components of the software module and their influence on different aspects of the clinical behaviour of the oximeter. The second part describes the results of realtime response investigation. The investigation was carried out using a personal computer to collect the data continuously during anaesthesia, surgery and post-operative periods. Two other industry standard oximeters, Nellcor and Ohmeda were also included in our study. We collected data on more than fifty patients on an average of eight hours per patient over a period of four months with major emphasis on low-saturation occurrences. The interpretation of the data was focused more on the realtime response anomalies on random cases than on ensemble statistical data evaluation. We found, that there are few factors in clinical environment which often influences the measurement of a pulse oximeter very strongly. Most often the anomalies were found during low saturation measurement. The main objective of this paper is to make the results available to practising clinicians so that it may be useful to identify these occurrences during routine clinical usage.
Part I: Software for a pulse oximeter
Non-invasive pulse oximeter falls into the category of a new generation of medical monitoring equipments in which the software logic plays a very sensitive role for its clinical validation. Today, in the clinical environment more and more software driven equipments and systems are finding application. Clinical evaluation and validation of such systems will be a major activity than the development of the system itself!. This paper discusses about one such project we undertook at our centre. The project was software development and clinical vali-
dation for a prototype pulse oximeter hardware designed by Siemens. Later, the success of the software in clinical environment proved its clinical usefulness and Siemens after validating against clinical and industry standards, accepted the software, made necessary changes for manufacturing requirements and currently this module is available along with their monitors as non-invasive Sa02 cartridge.
46 INFRARED ABSORPTION PATTERN
Absorption Spectrum of Infrared
I
] [ 1..~1
RED ABSORPTION PATTERN
Absorption Spectra
~
/
] ]
i,=,7
/
Processed
--i
-
Absorption
'
'
I I t / / / l l l l / Absorption spectrum of red
/_ /
Absorption Ratio Computation points
I "eart -----
T--~me
"
~
Time
Fig. 1. Absorption ratio computation analysis.
1) Design approach
2) The absorption spectrum
The fundamental approach in the software design of a system will determine the simplicity or complexity of the end product. Simplicity should be the goal, as it will help to produce a well behaved, easily controllable software. Simplicity, in this context, is on a relative scale between different approaches. In our software design, we applied an unformatted thinking into the development approaches with strong emphasis on simplicity so that new possibilities could emerge. To our satisfaction, we found few new possibilities in the measurement techniques that showed their merits during the clinical evaluation. The following paragraphs give a concise account of these techniques.
The principle of pulse oximetry is well known. The physics and engineering aspects of pulse oximeter have been in presented in sufficient detail by Kevin K. Tremper and Steven J. Barker [1]. The saturation is measured using the selective absorption of red and infrared light by reduced and oxygenated haemoglobin in blood. The absorption consists of two components - a pulsating component due to absorption by pulsatile arterial blood flow and a time invariant component due to absorption by tissues, venous and capillary blood flow. It is the pulsating component which is very important for the measurement of SaO2. From the pulsatile component an absorption ratio is computed between red and infrared absorption curves. This ratio is used for the calculation of SaO2. It is the measurement of absorption ratio that determines the final accuracy of the results. There are many ways
47 %Sa02 93 92 I n f l u e n c e of s p i k e s on t h e a v e r a g e
was 'lost' on averaging
91 90
II
I
L i t t l e i n f l u e n c e of s p i k e s on m e d i a n t r e n d
Jl
) //f
89
Response delay
\
88
'1 |'
averaging
- I n f l u e n c e of j u s t
/
o n e low v a l u e
, ,i':":'
i'|
d u e to
on t h e average.
87 86
........
Instantaneous
values
M e d i a n of l a t e s t 5 v a l u e s A v e r a g e of l a t e s t 5 v a l u e s
Median t r e n d follows a s s u r e d changes quickly
85 Time
Fig. 2. Realtimetransientcharacteristicscomparison: medianmethodvs averagingmethod.
to determine the absorption ratio and simultaneously filtering the influence of artifacts in the signal. For instance, computing continuously along the signal different absorption values as many as possible and then by finding an appropriate ensemble average could be one solution. More number of absorption values could mean it is more probable that the average value would be closer to the actual value. However, if one would consider the fundamental fact that contributes for the absorption, probably computation of few values which correspond to strong absorption events would give better result than computing at too many different instances. To help find those true instances, there is no general rule of how the absorption curve should look like, it depends on pigmentation, blood circulation, age and many other factors. The only thing that is common in the signal is a strong drop that appears in the plethysmographic signal soon after every systolic point in the arterial blood pressure curve. The absorption is
stronger than at other points in one full period of the pressure pulse curve. This could be chosen as an appropriate instance for computation of the absorption ratio. But the major problem is how to identify this instance successfully always.
4) Functional group analysis A classical approach could be to apply certain digital signal processing techniques on the red and infrared absorption signal to find out the appropriate instances. But simpler, reliable technique, easier to develop using today's powerful programming languages like 'C', running on faster microprocessors, could be functional group analysis. The basic idea is that when there are more than one signal that is attributed for a particular phenomena then instead of analyzing them individually, analyzing them together as a group, will be greatly useful. This scheme will substantially re-
48 %Sa02 100
_
90_
~@..~..Small
~
80_
error region: " S m a l l d i f f e r e n c e s in a b s o r p t i o n r a t i o d u c e s s m a l l d i f f e r e n c e s i n %Sa02"
70_ ~
60_
~
L
a
r
g
e
error region: "Small differences in absorption ratio %Sa02"
50_ 40_ 30_
Approximate Relation A - ratio
20_ 10_
%Sa02 =
• 100% B - C • ratio
Typical values:
h = 3.400 B = 3.1074 C = 0.3983
0 0.400
Fig. 3.
0,800
1.200
1.600
2.000
2.400
2.800
3.200
3.600 Absorption ratio (= R e d / I n f r a r e d )
E m p i r i c a l r e l a t i o n s h i p b e t w e e n % S a O 2 a n d a b s o r p t i o n ratio.
duce the possibility of artifact interference and inaccurate results. In our approach, the absorption signal for red and infrared is transformed into two absorption spectra - time sequence of absorption values. The selection of the appropriate instances in the absorption spectra are made using a size weighted time congruent analysis between the two absorption spectra. This is illustrated in Fig. 1. After this processing, a sequence of simple rule based filter techniques like neighbour distance rule, can be cascaded to improve the rejection of inappropriate instances. The net result of such processing and filtering operations are that very good instances for the calculation of absorption ratio are obtained. Even though they are few in number they produce consistent results which are not scattered. Reproducibility is a measure of reliability. We found that even without applying usual statistical techniques the values calculated at these instances gave good and stable results.
5) The statistical median Many different possibilities for artifact interference will show up in a clinical environment. It is extremely difficult and in some cases it is impossible to eliminate all the artifact effects, if they have completely covered up the actual signal. To be able to give dependable readings during this short period, one should resort to statistical techniques. This is an interesting problem by itself. The problem is finding a suitable mathematical technique which, given a continuous set of time varying input values, produces output values which closely matches the input values, reproduces the transient changes in the input and most importantly when the input is incorrect or non existent for a short time, the output will be still consistent and dependable. In this application, the input values are the absorption ratio computed in the previous stage. The difficulty arises when following specifications are framed for this mathematical function: it should not depend too much on the past data and that it should reproduce the true changes as quickly
49
HARDWARE
Front-end signal processing hardware
/
Determines sensitivity, m e a s u r e m e n t of weak signal and influence of a m b i e n t light source,
L.>
SOFTWARE
Filters for Artifacts Identification
Group Analysis of Absorption spectra
Statistical Techniques
1
1
Determines the trend behaviour during sudden
Controls the influence of m o t i o n artifacts and other f o r m s of i n h e r e n t absorption characteristic abnormalities.
Controls accuracy of a b s o r p t i o n ratio measurement, Determines consistency of measurement.
changes in s a t u r a t i o n and Consistency of display.
Determines the accuracy of %Sao2 at v a r i o u s l e v e l s of
saturation.
/
"--I %Sa02 - HR
I
~
I"
I
Empirical Relation / Table l o o k - u p for Sa02
1
Fig. 4. Clinical significance of various components of a pulse oximeter.
as possible, at the same time reasonably accurate value should be produced when no or absurd input value is collected intermittently. A simple technique will be taking an average of fixed number of latest input values. The size of the samples would determine the properties. Though, at first sight this may sound usable, soon one will find out many different drawbacks of this technique. The major drawback is, averaging will offer high resistance to sudden short period changes. Sometimes the changes may be unnoticed, as the changes may be smoothed out. Certainly this will not be ideal for an application where one is always interested in responses to quick changes. Also in the averaging method the result is not an actually measured value but a calculated value. This would mean that if there are abnormal values (very small or very large) they would influence the results badly. To avoid this, usually, the minimum and maximum values will be ignored and the rest will be averaged, with the expectation that next minimum or maxi-
mum values are not abnormal. As we were figuring out for a reasonable alternative statistical method for time varying input, we tried out statistical median which is not usually implemented in this type of applications, probably due to speed consideration. In the median method, every time a new input value is entered a set of fixed number of latest input values (corresponding to a given time period) will be sorted to get an ascending or a descending sequence of values, and the median of the sorted range will be found (the middle value). This median can be taken to be a reliable instantaneous representation of the time varying input. The primary difference, in this case, is that the median is a truly measured value and not a computed value. Secondly, the median value will follow the input changes quickly if the changes are not haphazard (i.e. the changes follow an increasing or decreasing pattern). In this case, the new trend will reach the median point of the sorted list in proportion to the rate at which the actual input changes. This is de-
50 %Sa02 ,',
100 4 2 6 8 - ~
, ,,--,
PATIENT'S PROFILE AGE: 72 Years Diagnosis: CAB 2x State: PreoperativeInduction
J
Sudden hypoxia due to airway o b s t r u c t i o n
Legends: - D e v e l o p m e n t Module - 0hmeda . . . . . . . . Nellcor IL282 C O - o x i m e t e r
Undershoot during sudden hypoxia.
I' 2, , ,4 ,
0
6
8
i , ,2, , 4
10
6
8
i , ,2 , , 4i , ,6, ~8
20
I 2
4
6
30 Time in m i n u t e s
8
40
i
2 46
i
i
8
i
50
t
24
t
i
68
t
i
////
60
Fig. 5. Realtime impulse response characteristics.
picted in Fig. 2. The only difficulty we expect with the median method is that if the successive input value is oscillating between extreme limits (the minimum and maximum) then the median value may not show this oscillation!. The chances for this high frequency oscillation is very remote. It can also be interpreted that the oscillation itself is a strange behaviour, perhaps a disturbance, therefore not showing it can be viewed as an acceptable property of the method. As we found that the benefit of this method are worth the speed compromise, we implemented a quick sort technique for finding the median of the range and used it for final output. It gave very good results in our clinical study (described in Part II), especially during measurement of sudden changes.
6) The empirical relation Once the absorption ratio is computed, in the final
stage, the ratio is converted to % SaO2 using an empirical relationship. We found that it would be appropriate, considering the mathematics of the empirical relation, to convert the ratio into % SaO2 in the final stage just before display and not in the intermediate stages prior to statistical techniques. This will help to eliminate the errors due to precision of calculations. This is illustrated in the Fig. 3. The mathematical formulae is based on an empirical relation. This is where the dependency of the accuracy of the % SaO2 on its percentage level shows up. The relationship is excellent in the high range of % SaO2 but not very accurate in the low range of % SaO2 due to lack of sufficient clinical data for calibration [2]. This could be refined as more and more clinical trials are conducted. Also by applying probe related corrections so that various errors are reduced to a minimum. This would help to improve the results in the lower range. To be able to measure accurately at lower range
51 %Sa02
100 - 8-
PATIENT'S
6
PROFILE
4AGE: 7 4 Y e a r s
290--
Diagnosis:
8-
CAB 4 x
64-
State:
2-
PreoperativeInduction
S h o r t d e l a y in r e s p o n s e
80-864-
On V e n t i l a t i o n ( F i 0 2 = 100%)
Normal breathing ( F i 0 2 = 22%)
2-
Legends: - - -
70-8-
......
6-
Development 0hmeda
Module
Nelleor IL282 CO-oximeter
4260--
Occasional short breaks during measurement
r162 2,
4,
6,
8,
t 10
2,
4,
6,
8,
i 20
2,
4,
8,
Time
Fig. 6.
8,
I 2 4, 30 in minutes
6,
8,
i 40
2,
4,
6,
8,
i 50
2
, 8, j ~S /
6
60
R e a l t i m e transient r e s p o n s e characteristics.
is one requirement but to be able to show changes from normal values quickly is another requirement. Both of these would determine the desirable qualities of a pulse oximeter. The former depends mainly on an empirical relation while the later depends mainly on accuracy of measurement and the efficiency of the statistical techniques used. A general block diagram of various components of the prototype SaO2 module and the clinical significance of the individual components are presented in Fig. 4.
Conclusion We found, after an extensive clinical verification, that our independent original approaches in the software design on two major aspects- the absorption analysis and the statistical technique gave very good results. We paid equally careful attention in our design to maintain software engineering dis-
ciplines on modularity, flexibility and documentation to facilitate future enhancements. Presently, the software is used in Siemens SaO2 cartridge which is a plug-in module for Siemens monitors.
Part II: Clinical evaluation The protocol
To obtain a complete description of the clinical behaviour of our pulse oximeter software in comparison to other industry standard pulse oximeters, we conducted an extensive clinical evaluation during routine clinical procedures. Study of realtime response behaviour was the major objective in the evaluation. This required an automated system for data collection and for plotting the data as graphs. We developed a software on a personal computer IBM PC/AT to collect data from various oximeters and display them conveniently on screen (updated
52 %Sa02 100 ~ PATIENT'S PROFILE AGE: 2 Y e a r s
2-
Diagnosis: Fortran Large d e v i a t i o n s f r o m IL282 Sa02
State: PreoperativeInduction
Legends: - - ........
"Hiccup"
D e v e l o p m e n t Module Ohmeda Nellcor IL282 C 0 - o x i m e t e r
I I r ] 2 4 6 8 0
I
I 10
I I I 2 4 6 8
I
[ 20
I I 2 46
I
J
I
I
8
I
I 24 68 30 T i m e in m i n u t e s
I
F
I
]
I
I
I 2
4
6
8
40
//
I
12 50
4
6 8
//
60
Fig. 7. C o n s i s t e n c y d u r i n g l o w s a t u r a t i o n m e a s u r e m e n t o n a r a n d o m case study.
every second) while at the same time store them in the hard disc. The files stored in the hard disc can be retrieved and plotted. The software was developed to support interface for the following oximeters connected through serial link to the IBM PC/AT: 1) The prototype pulse oximeter with our software using a disposable probe 2) Nellcor pulse oximeter (N-200) with a disposable probe and 3) Ohmeda pulse oximeter (Revision K) with a non-disposable probe. All the probes used were for the finger. To evaluate against laboratory blood-gas reports, blood samples were taken at selected instances and analyzed in IL282 CO-oximeter, the results were stored along with the data collected from the oximeters corresponding to the instance when the blood sample was taken. On a day-to-day basis, the measurement protocol was started when the patient entered the in-
duction room and it was continued during the following phases: normal breathing before anaesthesia, during induction, before operation and for a period of at least four hours in the post-operative care. On an average the total duration of measurement per patient was around eight hours. The measurement was carried out on more than fifty patients. Later, using a Hewlett-Packard multi-colour pen plotter, we plotted on a minute scale, the continuous data collected from the pulse oximeters in different colours to facilitate quick comparison. These plots were extremely useful to describe the behaviour of the three pulse oximeters, under following clinical circumstances: 1) The transient response to sudden and gradual changes. 2) The consistency of the measurement at various % SaO2 levels. 3) The behaviour with respect to motion artifacts and hypothermia.
53 %Sa02 100 PATIENT'S PROFILE
~
~
2i
- i 8' " ' " AGE: 72 Years
9
Diagnosis: CAB
6 4 2 80864270-864260--
2x
State: Postoperative
Legends: - D e v e l o p m e n t Module - Ohmeda . . . . . . . . Nelleor
,-.....
IL282 C O - o x i m e t e r hnamolous breaks during measurement
r I
2
6
lO
4
I
6
8
I
I
I
I
I
I
I
[
2 46 8 I 24 6 8 ' 2 4 6 8 24 20 30 40 50 Time in m i n u t e s
68
60
Fig. 8. R e s p o n s e to small degree involuntary probe m o v e m e n t s with poor peripheral circulation.
4) The influence of ambient light sources in operating theatre. 5) Accuracy of measurement on infants. As space may not permit us to include all the salient graphs we made, we have limited ourself to the most important graphs among them. However, we will give a consolidated report on various aspects of a pulse oximeter we found from our study that may be clinically useful. The laboratory analyzer IL 282 CO oximeter used as reference in our study has been reported to have very good accuracy for the complete range of % SaO2 [1]. The blood-gas reports for the patients whose measurement data appear in this paper, indicated that methaemoglobin and carboxyhaemoglobin levels together were less than 3.5%.
1) Impulse response Impulse response refers to response of the pulse
oximeter to sudden changes in saturation. Comparison of responses of various oximeters to sudden changes in SaO2 has been extensively investigated by J.W. Severinghaus and K.H. Naifeh [2]. Also on a statistical basis the responses were studied by D.M. Kagle and others [3]. The impulse response of oximeters are strongly characterized not by its hardware or probe but by its software, which attempts to do certain intelligent processing of the measurement for consistency of % SaO2 display. In our study we found that Ohmeda quickly responded to those changes, our prototype module followed Ohmeda after few seconds (less than ten seconds) and Nellcor showed relatively slow response, Nellcor software appears to attempt for an over confirmation (by a statistical procedure) before it shows the change in the display. Also we noticed that frequently Nellcor oximeter the data transmitted to the computer lags the data actually displayed on the unit by few seconds. Fig. 5 and
54 %Sa02 100 -8.
PATIENT'S PROFILE
6. 4-
AGE: 5 M o n t h s
2-
Diagnosis: Fallot Tetralogy Less i n f l u e n c e d u e to c l o s e d probe housing
State: Post-anesthesia Pre-sternotomy
Legends: - D e v e l o p m e n t Module - Ohmeda ........ Nellcor IL282 CO- o x i m e t e r
S e v e r e i n f l u e n c e of o p e r a t i n g l a m p s on m e a s u r e m e n t while using disposable probes I I I 1 2 4 6 8 0
Fig. 9.
I
I z4
10
68
20
2 4 6 8
2 4 6 8
30 T i m e in m i n u t e s
40
2 4 6 8
50
24
,, 6
/ 8
60
I n t e r f e r e n c e o n d i r e c t e x p o s u r e to light s o u r c e s in o p e r a t i n g r o o m .
Fig. 6 illustrates typical transient response to impulse and steady-state changes respectively.
2) The consistency of measurement during steady-state The consistency refers to stability of the values displayed at a given percent SaO2 level. The stability is affected mostly by the empirical relation for SaO2 and by the statistical technique employed. The actual measured quantity in a pulse oximeter is not % SaO2 but absorption ratio (described in Part I). The absorption ratio is mapped to SaO2 using an empirical relation or a table look-up procedure. This is were the instability shows up, the relation between the absorption ratio and % SaO2 is not linear, consequently small variations or errors in the absorption ratio may give large changes in % SaO2 depending on the measured region of % SaO2. In the relation this distortion grows high-
er as the % SaO2 drops lower than 85%. In practice, we found that all the three oximeters shows fluctuation below 75%. In our study, Ohmeda showed large fluctuation as compared to our prototype or Nellcor while the mean of the fluctuation was closer to the reference % SaO2 when compared with the other two. Ohmeda's large fluctuation may be due its quick response to changes. NeUcor and our prototype showed fair stability in lower ranges. The stability exhibited by our prototype is primarily due to the statistical median method employed which is insensitive to the magnitude of small and large values of the population. We also observed sometimes Nellcor withdraws the value from display then it restarts the measurement after a while. The lack of accuracy at low saturation has been well known. [1, 2, 4]. It may be due to lack of sufficient clinical data in the low range of % SaO2 for the empirical relation. Industry standard oximeters are reported to produce errors as large as 10% at saturation levels between 70% and 50% [4].
55 %Sa02 100 i ]
PATIENT'S PROFILE AGE: II M o n t h s
9~
:',
Diagnosis: Right Blalock Shunt
Short sporadic problems during measurement.
6 4 2 808 64-270-864260--
State: PostanesthesiaPresternotomy
Legends: dreifts d u r i n g
- D e v e l o p m e n t Module - 0hmeda . . . . . . . Nellcor IL282 C O - o x i m e t e r
S 0
2 4 6 8
10
2 4 6 8
20
2 4 6 8 2 4 6 8 2 4 6 8 2 4 6 8 30 40 50 60 Time in m i n u t e s
Fig, 10. C o m m o n problems during m e a s u r e m e n t on children with congenital heart diseases.
The trend pattern during this circumstances usually appears as shown in Fig. 7.
ermia condition produces abnormalities in the measurement as depicted in the Fig. 8.
3) The influence of motion artifacts and hypothermia
4) The influence of ambient light
Most of the time the motion artifact is created by continuous disturbance of the probe especially due to shivering. Under this circumstances Ohmeda frequently showed different readings and warnings while Nellcor and our prototype module showed considerable stability. However, when the body temperature is very low (temperatures below 33 degree Celsius) Ohmeda could display reading consistently while Nellcor and the prototype model failed to measure at times. This may be probably to do with the sensitivity of the probe. Small movements, though under normal circumstances don't produce erroneous results, under certain combination of poor peripheral circulation and/or hypoth-
During our clinical trials we found that all the oximeters readings were badly affected when the probes were directly exposed to operating lamps. We observed that infrared plethysmograph signal was badly disturbed. However, when the probe was covered with a cloth, the disturbance was drastically reduced. Interference due to infrared heat lamps was reported by Brooks TD and others [5]. Under the influence of the light source, Nellcor was severely affected, our prototype module was moderately affected (the reduced influence was due to measurement at selected instances in the signal, as described in Part I) and Ohmeda was less influenced (probably because of the sealed design of its non-disposable probe). It should be noted that nor-
56 mal lamps in the room don't affect the measurement as claimed by the manufacturers. SaO2 measurements under direct exposure to operating lamps generally appear as shown in Fig. 9.
5) Accuracy of measurement on infants Paediatric surgery on infants with congenital heart diseases are performed frequently at our centre. We carried out our evaluation of oximeters on such paediatric patients. We used disposable children's probe for Nellcor and for our prototype module while for Ohmeda we used a non-disposable probe. All the probes used were for the finger. Evaluation of the oximeters with infants were very tough. Especially finding proper probe placement was quite difficult. Due to difficulty in placing the probe over finger and to reduce motion artifact we placed the probes on toe. For healthy subjects the readings were stable and correlated with laboratory measurements for all the three oximeters, while for weak subjects and infants with congenital heart problems the readings were often inaccurate. Ohmeda showed occasionally close values to laboratory results but frequently its readings drifted around. Nellcor and our module in this cases showed values higher than the actual value consistently but reproducing the trend shapes well in agreement with laboratory readings. These effects are illustrated in Fig. 10. The importance for accuracy in measurements on premature infants when SaO2 may hold a constant level of desaturation has been strongly emphasized by Severinghaus and others [2]. Also it has been reported that fairly good measurements are possible in children even while using adult's probe by C.J. Cote and others
[6]. Possible tanning of the skin under the continuous light from the probe is another major factor with respect to measurement on infants. We found in our study, that Ohmeda probes cause more tanning of the skin than the other two oximeters. Using adult's probe on infants may cause more severe tanning in a shorter time.
6) Influence of intravenously administered dyes As we were conducting our evaluation during routine clinical procedure, we did not carry out studies on the influence of iv-administered dyes on SaO2 measurement. However, it has been well reported by M.S. Scheller [7] and others [8]. The result of their investigation is that intravenously administered dyes influences SaO2 readings strongly to produce very low false values. This is well supported by the fact that the absorption of plethysmograph signal by the dyes is quite strong at red wavelength which is used in pulse oximeters. Among the dyes methylene blue shows strong influence, indocyanine green shows moderate influence and indigo carmine shows less influence [7].
Conclusion
The clinical usefulness of a pulse oximeter is proved beyond doubt [1, 4, 9, 10] and has been extensively evaluated by Severinghaus, Chapman and others [2, 3, 4]. In our investigation as we found that most our findings on the overall behaviour of the pulse oximeters were well in agreement with the results published in various clinical research literatures, we concentrated on peculiarities that may be encountered during routine usage in clinical environment. The computerized protocol which we implemented for our study of realtime data collection and multi-colour plotting of the trend curves in a graphical form proved to be very useful and made our clinical investigation easier. Some of the charts which will be of clinical interest were presented in this paper to serve as illustrations. While most of the problems exhibited by the oximeters during our random clinical evaluation study may be solved by the manufacturer themselves, few of them may remain unsolved. In practice, these problems may need to be identified and corrected whenever possible by practising clinicians or nurses.
57
References 1. Tremper KK, Barker SJ. Pulse Oximetry. Anesthesiology 1989; 70: 98-108. 2. Severinghaus JW, Naiefeh KH. Accuracy of Response of Six Pulse Oximeters to Profound Hypoxia. Anesthesiology 1987; 67: 551-558. 3. Kagle DM, Alexander CM, Berko RS, Giuffre M, Gross JB. Evaluation of the Ohmeda 3700 Pulse Oximeter: Steady-state and Transient Response Characteristics: Anesthesiology 1987; 66: 376-380. 4. Chapman KR, Liu FLW, Watson RM, Rebuck AS. Range of Accuracy of Two Wavelength Oximetry. Chest/89/4/ 540-542, April, 1986. 5. Brooks RD, Paulus DA, Winkle WE. Infrared heat lamps interfere with pulse oximeters. Anesthesiology 1984; 61: 630. 6. Cot6 CJ, Goldsteing EA, Cot6 MA, Hoaglin DC, Ryan JF. A Single-blind Study of Pulse Oximetry in Children. Anesthesiology 1988; 68: 184-188.
7. Scheller MS, Unger RJ, Kelner MJ. Effectsoflntravenously Administered Dyes on Pulse Oximetry Readings. Anesthesiology 1986; 65: 550-552. 8. Kessler MR, Eide T, Humayun B, Poppers Pj. Spurious Pulse Oximeter Desaturation with Methylene Blue Injection. Anesthesiology 1986; 65: 435436. 9. Cohen DE, Downes JJ, Raphaely RC. Editorial Views: What Difference Does Pulse Oximetry Make? Anesthesiology 1988; 68: 181-183. 10. Fairley HB. Editorial Views: Changing Perspectives in Monitoring Oxygenation. Anesthesiology 1989; 70: 2-4.
Address for offprints: S. Meiyappan, c/o Omar Prakash, Thorax Centre, Erasmus University, Postbus 1738, 3000 DR Rotterdam, The Netherlands
Development of software for pulse oximeter and investigation of its realtime response in clinical environment S. Meiyappan & Omar Prakash
Erasmus University, Postbus 1738, 3000 DR Rotterdam, The Netherlands Accepted 15 January1990
Key words: pulse oximetry, plethysmograph, SaO2 software, oxyhaemoglobin, realtime data collection, computerized charting Abstract
We developed a pulse oximeter software at Thorax Centre, Erasmus University as a joint project with an industry and evaluated it in the clinical environment of thorax anaesthesia using a computerized protocol for realtime data collection during routine clinical procedures. This paper gives an account of the results we obtained from the development project and the clinical study. The paper consists of two parts. First part describes different components of the software module and their influence on different aspects of the clinical behaviour of the oximeter. The second part describes the results of realtime response investigation. The investigation was carried out using a personal computer to collect the data continuously during anaesthesia, surgery and post-operative periods. Two other industry standard oximeters, Nellcor and Ohmeda were also included in our study. We collected data on more than fifty patients on an average of eight hours per patient over a period of four months with major emphasis on low-saturation occurrences. The interpretation of the data was focused more on the realtime response anomalies on random cases than on ensemble statistical data evaluation. We found, that there are few factors in clinical environment which often influences the measurement of a pulse oximeter very strongly. Most often the anomalies were found during low saturation measurement. The main objective of this paper is to make the results available to practising clinicians so that it may be useful to identify these occurrences during routine clinical usage.
Part I: Software for a pulse oximeter
Non-invasive pulse oximeter falls into the category of a new generation of medical monitoring equipments in which the software logic plays a very sensitive role for its clinical validation. Today, in the clinical environment more and more software driven equipments and systems are finding application. Clinical evaluation and validation of such systems will be a major activity than the development of the system itself!. This paper discusses about one such project we undertook at our centre. The project was software development and clinical vali-
dation for a prototype pulse oximeter hardware designed by Siemens. Later, the success of the software in clinical environment proved its clinical usefulness and Siemens after validating against clinical and industry standards, accepted the software, made necessary changes for manufacturing requirements and currently this module is available along with their monitors as non-invasive Sa02 cartridge.
46 INFRARED ABSORPTION PATTERN
Absorption Spectrum of Infrared
I
] [ 1..~1
RED ABSORPTION PATTERN
Absorption Spectra
~
/
] ]
i,=,7
/
Processed
--i
-
Absorption
'
'
I I t / / / l l l l / Absorption spectrum of red
/_ /
Absorption Ratio Computation points
I "eart -----
T--~me
"
~
Time
Fig. 1. Absorption ratio computation analysis.
1) Design approach
2) The absorption spectrum
The fundamental approach in the software design of a system will determine the simplicity or complexity of the end product. Simplicity should be the goal, as it will help to produce a well behaved, easily controllable software. Simplicity, in this context, is on a relative scale between different approaches. In our software design, we applied an unformatted thinking into the development approaches with strong emphasis on simplicity so that new possibilities could emerge. To our satisfaction, we found few new possibilities in the measurement techniques that showed their merits during the clinical evaluation. The following paragraphs give a concise account of these techniques.
The principle of pulse oximetry is well known. The physics and engineering aspects of pulse oximeter have been in presented in sufficient detail by Kevin K. Tremper and Steven J. Barker [1]. The saturation is measured using the selective absorption of red and infrared light by reduced and oxygenated haemoglobin in blood. The absorption consists of two components - a pulsating component due to absorption by pulsatile arterial blood flow and a time invariant component due to absorption by tissues, venous and capillary blood flow. It is the pulsating component which is very important for the measurement of SaO2. From the pulsatile component an absorption ratio is computed between red and infrared absorption curves. This ratio is used for the calculation of SaO2. It is the measurement of absorption ratio that determines the final accuracy of the results. There are many ways
47 %Sa02 93 92 I n f l u e n c e of s p i k e s on t h e a v e r a g e
was 'lost' on averaging
91 90
II
I
L i t t l e i n f l u e n c e of s p i k e s on m e d i a n t r e n d
Jl
) //f
89
Response delay
\
88
'1 |'
averaging
- I n f l u e n c e of j u s t
/
o n e low v a l u e
, ,i':":'
i'|
d u e to
on t h e average.
87 86
........
Instantaneous
values
M e d i a n of l a t e s t 5 v a l u e s A v e r a g e of l a t e s t 5 v a l u e s
Median t r e n d follows a s s u r e d changes quickly
85 Time
Fig. 2. Realtimetransientcharacteristicscomparison: medianmethodvs averagingmethod.
to determine the absorption ratio and simultaneously filtering the influence of artifacts in the signal. For instance, computing continuously along the signal different absorption values as many as possible and then by finding an appropriate ensemble average could be one solution. More number of absorption values could mean it is more probable that the average value would be closer to the actual value. However, if one would consider the fundamental fact that contributes for the absorption, probably computation of few values which correspond to strong absorption events would give better result than computing at too many different instances. To help find those true instances, there is no general rule of how the absorption curve should look like, it depends on pigmentation, blood circulation, age and many other factors. The only thing that is common in the signal is a strong drop that appears in the plethysmographic signal soon after every systolic point in the arterial blood pressure curve. The absorption is
stronger than at other points in one full period of the pressure pulse curve. This could be chosen as an appropriate instance for computation of the absorption ratio. But the major problem is how to identify this instance successfully always.
4) Functional group analysis A classical approach could be to apply certain digital signal processing techniques on the red and infrared absorption signal to find out the appropriate instances. But simpler, reliable technique, easier to develop using today's powerful programming languages like 'C', running on faster microprocessors, could be functional group analysis. The basic idea is that when there are more than one signal that is attributed for a particular phenomena then instead of analyzing them individually, analyzing them together as a group, will be greatly useful. This scheme will substantially re-
48 %Sa02 100
_
90_
~@..~..Small
~
80_
error region: " S m a l l d i f f e r e n c e s in a b s o r p t i o n r a t i o d u c e s s m a l l d i f f e r e n c e s i n %Sa02"
70_ ~
60_
~
L
a
r
g
e
error region: "Small differences in absorption ratio %Sa02"
50_ 40_ 30_
Approximate Relation A - ratio
20_ 10_
%Sa02 =
• 100% B - C • ratio
Typical values:
h = 3.400 B = 3.1074 C = 0.3983
0 0.400
Fig. 3.
0,800
1.200
1.600
2.000
2.400
2.800
3.200
3.600 Absorption ratio (= R e d / I n f r a r e d )
E m p i r i c a l r e l a t i o n s h i p b e t w e e n % S a O 2 a n d a b s o r p t i o n ratio.
duce the possibility of artifact interference and inaccurate results. In our approach, the absorption signal for red and infrared is transformed into two absorption spectra - time sequence of absorption values. The selection of the appropriate instances in the absorption spectra are made using a size weighted time congruent analysis between the two absorption spectra. This is illustrated in Fig. 1. After this processing, a sequence of simple rule based filter techniques like neighbour distance rule, can be cascaded to improve the rejection of inappropriate instances. The net result of such processing and filtering operations are that very good instances for the calculation of absorption ratio are obtained. Even though they are few in number they produce consistent results which are not scattered. Reproducibility is a measure of reliability. We found that even without applying usual statistical techniques the values calculated at these instances gave good and stable results.
5) The statistical median Many different possibilities for artifact interference will show up in a clinical environment. It is extremely difficult and in some cases it is impossible to eliminate all the artifact effects, if they have completely covered up the actual signal. To be able to give dependable readings during this short period, one should resort to statistical techniques. This is an interesting problem by itself. The problem is finding a suitable mathematical technique which, given a continuous set of time varying input values, produces output values which closely matches the input values, reproduces the transient changes in the input and most importantly when the input is incorrect or non existent for a short time, the output will be still consistent and dependable. In this application, the input values are the absorption ratio computed in the previous stage. The difficulty arises when following specifications are framed for this mathematical function: it should not depend too much on the past data and that it should reproduce the true changes as quickly
49
HARDWARE
Front-end signal processing hardware
/
Determines sensitivity, m e a s u r e m e n t of weak signal and influence of a m b i e n t light source,
L.>
SOFTWARE
Filters for Artifacts Identification
Group Analysis of Absorption spectra
Statistical Techniques
1
1
Determines the trend behaviour during sudden
Controls the influence of m o t i o n artifacts and other f o r m s of i n h e r e n t absorption characteristic abnormalities.
Controls accuracy of a b s o r p t i o n ratio measurement, Determines consistency of measurement.
changes in s a t u r a t i o n and Consistency of display.
Determines the accuracy of %Sao2 at v a r i o u s l e v e l s of
saturation.
/
"--I %Sa02 - HR
I
~
I"
I
Empirical Relation / Table l o o k - u p for Sa02
1
Fig. 4. Clinical significance of various components of a pulse oximeter.
as possible, at the same time reasonably accurate value should be produced when no or absurd input value is collected intermittently. A simple technique will be taking an average of fixed number of latest input values. The size of the samples would determine the properties. Though, at first sight this may sound usable, soon one will find out many different drawbacks of this technique. The major drawback is, averaging will offer high resistance to sudden short period changes. Sometimes the changes may be unnoticed, as the changes may be smoothed out. Certainly this will not be ideal for an application where one is always interested in responses to quick changes. Also in the averaging method the result is not an actually measured value but a calculated value. This would mean that if there are abnormal values (very small or very large) they would influence the results badly. To avoid this, usually, the minimum and maximum values will be ignored and the rest will be averaged, with the expectation that next minimum or maxi-
mum values are not abnormal. As we were figuring out for a reasonable alternative statistical method for time varying input, we tried out statistical median which is not usually implemented in this type of applications, probably due to speed consideration. In the median method, every time a new input value is entered a set of fixed number of latest input values (corresponding to a given time period) will be sorted to get an ascending or a descending sequence of values, and the median of the sorted range will be found (the middle value). This median can be taken to be a reliable instantaneous representation of the time varying input. The primary difference, in this case, is that the median is a truly measured value and not a computed value. Secondly, the median value will follow the input changes quickly if the changes are not haphazard (i.e. the changes follow an increasing or decreasing pattern). In this case, the new trend will reach the median point of the sorted list in proportion to the rate at which the actual input changes. This is de-
50 %Sa02 ,',
100 4 2 6 8 - ~
, ,,--,
PATIENT'S PROFILE AGE: 72 Years Diagnosis: CAB 2x State: PreoperativeInduction
J
Sudden hypoxia due to airway o b s t r u c t i o n
Legends: - D e v e l o p m e n t Module - 0hmeda . . . . . . . . Nellcor IL282 C O - o x i m e t e r
Undershoot during sudden hypoxia.
I' 2, , ,4 ,
0
6
8
i , ,2, , 4
10
6
8
i , ,2 , , 4i , ,6, ~8
20
I 2
4
6
30 Time in m i n u t e s
8
40
i
2 46
i
i
8
i
50
t
24
t
i
68
t
i
////
60
Fig. 5. Realtime impulse response characteristics.
picted in Fig. 2. The only difficulty we expect with the median method is that if the successive input value is oscillating between extreme limits (the minimum and maximum) then the median value may not show this oscillation!. The chances for this high frequency oscillation is very remote. It can also be interpreted that the oscillation itself is a strange behaviour, perhaps a disturbance, therefore not showing it can be viewed as an acceptable property of the method. As we found that the benefit of this method are worth the speed compromise, we implemented a quick sort technique for finding the median of the range and used it for final output. It gave very good results in our clinical study (described in Part II), especially during measurement of sudden changes.
6) The empirical relation Once the absorption ratio is computed, in the final
stage, the ratio is converted to % SaO2 using an empirical relationship. We found that it would be appropriate, considering the mathematics of the empirical relation, to convert the ratio into % SaO2 in the final stage just before display and not in the intermediate stages prior to statistical techniques. This will help to eliminate the errors due to precision of calculations. This is illustrated in the Fig. 3. The mathematical formulae is based on an empirical relation. This is where the dependency of the accuracy of the % SaO2 on its percentage level shows up. The relationship is excellent in the high range of % SaO2 but not very accurate in the low range of % SaO2 due to lack of sufficient clinical data for calibration [2]. This could be refined as more and more clinical trials are conducted. Also by applying probe related corrections so that various errors are reduced to a minimum. This would help to improve the results in the lower range. To be able to measure accurately at lower range
51 %Sa02
100 - 8-
PATIENT'S
6
PROFILE
4AGE: 7 4 Y e a r s
290--
Diagnosis:
8-
CAB 4 x
64-
State:
2-
PreoperativeInduction
S h o r t d e l a y in r e s p o n s e
80-864-
On V e n t i l a t i o n ( F i 0 2 = 100%)
Normal breathing ( F i 0 2 = 22%)
2-
Legends: - - -
70-8-
......
6-
Development 0hmeda
Module
Nelleor IL282 CO-oximeter
4260--
Occasional short breaks during measurement
r162 2,
4,
6,
8,
t 10
2,
4,
6,
8,
i 20
2,
4,
8,
Time
Fig. 6.
8,
I 2 4, 30 in minutes
6,
8,
i 40
2,
4,
6,
8,
i 50
2
, 8, j ~S /
6
60
R e a l t i m e transient r e s p o n s e characteristics.
is one requirement but to be able to show changes from normal values quickly is another requirement. Both of these would determine the desirable qualities of a pulse oximeter. The former depends mainly on an empirical relation while the later depends mainly on accuracy of measurement and the efficiency of the statistical techniques used. A general block diagram of various components of the prototype SaO2 module and the clinical significance of the individual components are presented in Fig. 4.
Conclusion We found, after an extensive clinical verification, that our independent original approaches in the software design on two major aspects- the absorption analysis and the statistical technique gave very good results. We paid equally careful attention in our design to maintain software engineering dis-
ciplines on modularity, flexibility and documentation to facilitate future enhancements. Presently, the software is used in Siemens SaO2 cartridge which is a plug-in module for Siemens monitors.
Part II: Clinical evaluation The protocol
To obtain a complete description of the clinical behaviour of our pulse oximeter software in comparison to other industry standard pulse oximeters, we conducted an extensive clinical evaluation during routine clinical procedures. Study of realtime response behaviour was the major objective in the evaluation. This required an automated system for data collection and for plotting the data as graphs. We developed a software on a personal computer IBM PC/AT to collect data from various oximeters and display them conveniently on screen (updated
52 %Sa02 100 ~ PATIENT'S PROFILE AGE: 2 Y e a r s
2-
Diagnosis: Fortran Large d e v i a t i o n s f r o m IL282 Sa02
State: PreoperativeInduction
Legends: - - ........
"Hiccup"
D e v e l o p m e n t Module Ohmeda Nellcor IL282 C 0 - o x i m e t e r
I I r ] 2 4 6 8 0
I
I 10
I I I 2 4 6 8
I
[ 20
I I 2 46
I
J
I
I
8
I
I 24 68 30 T i m e in m i n u t e s
I
F
I
]
I
I
I 2
4
6
8
40
//
I
12 50
4
6 8
//
60
Fig. 7. C o n s i s t e n c y d u r i n g l o w s a t u r a t i o n m e a s u r e m e n t o n a r a n d o m case study.
every second) while at the same time store them in the hard disc. The files stored in the hard disc can be retrieved and plotted. The software was developed to support interface for the following oximeters connected through serial link to the IBM PC/AT: 1) The prototype pulse oximeter with our software using a disposable probe 2) Nellcor pulse oximeter (N-200) with a disposable probe and 3) Ohmeda pulse oximeter (Revision K) with a non-disposable probe. All the probes used were for the finger. To evaluate against laboratory blood-gas reports, blood samples were taken at selected instances and analyzed in IL282 CO-oximeter, the results were stored along with the data collected from the oximeters corresponding to the instance when the blood sample was taken. On a day-to-day basis, the measurement protocol was started when the patient entered the in-
duction room and it was continued during the following phases: normal breathing before anaesthesia, during induction, before operation and for a period of at least four hours in the post-operative care. On an average the total duration of measurement per patient was around eight hours. The measurement was carried out on more than fifty patients. Later, using a Hewlett-Packard multi-colour pen plotter, we plotted on a minute scale, the continuous data collected from the pulse oximeters in different colours to facilitate quick comparison. These plots were extremely useful to describe the behaviour of the three pulse oximeters, under following clinical circumstances: 1) The transient response to sudden and gradual changes. 2) The consistency of the measurement at various % SaO2 levels. 3) The behaviour with respect to motion artifacts and hypothermia.
53 %Sa02 100 PATIENT'S PROFILE
~
~
2i
- i 8' " ' " AGE: 72 Years
9
Diagnosis: CAB
6 4 2 80864270-864260--
2x
State: Postoperative
Legends: - D e v e l o p m e n t Module - Ohmeda . . . . . . . . Nelleor
,-.....
IL282 C O - o x i m e t e r hnamolous breaks during measurement
r I
2
6
lO
4
I
6
8
I
I
I
I
I
I
I
[
2 46 8 I 24 6 8 ' 2 4 6 8 24 20 30 40 50 Time in m i n u t e s
68
60
Fig. 8. R e s p o n s e to small degree involuntary probe m o v e m e n t s with poor peripheral circulation.
4) The influence of ambient light sources in operating theatre. 5) Accuracy of measurement on infants. As space may not permit us to include all the salient graphs we made, we have limited ourself to the most important graphs among them. However, we will give a consolidated report on various aspects of a pulse oximeter we found from our study that may be clinically useful. The laboratory analyzer IL 282 CO oximeter used as reference in our study has been reported to have very good accuracy for the complete range of % SaO2 [1]. The blood-gas reports for the patients whose measurement data appear in this paper, indicated that methaemoglobin and carboxyhaemoglobin levels together were less than 3.5%.
1) Impulse response Impulse response refers to response of the pulse
oximeter to sudden changes in saturation. Comparison of responses of various oximeters to sudden changes in SaO2 has been extensively investigated by J.W. Severinghaus and K.H. Naifeh [2]. Also on a statistical basis the responses were studied by D.M. Kagle and others [3]. The impulse response of oximeters are strongly characterized not by its hardware or probe but by its software, which attempts to do certain intelligent processing of the measurement for consistency of % SaO2 display. In our study we found that Ohmeda quickly responded to those changes, our prototype module followed Ohmeda after few seconds (less than ten seconds) and Nellcor showed relatively slow response, Nellcor software appears to attempt for an over confirmation (by a statistical procedure) before it shows the change in the display. Also we noticed that frequently Nellcor oximeter the data transmitted to the computer lags the data actually displayed on the unit by few seconds. Fig. 5 and
54 %Sa02 100 -8.
PATIENT'S PROFILE
6. 4-
AGE: 5 M o n t h s
2-
Diagnosis: Fallot Tetralogy Less i n f l u e n c e d u e to c l o s e d probe housing
State: Post-anesthesia Pre-sternotomy
Legends: - D e v e l o p m e n t Module - Ohmeda ........ Nellcor IL282 CO- o x i m e t e r
S e v e r e i n f l u e n c e of o p e r a t i n g l a m p s on m e a s u r e m e n t while using disposable probes I I I 1 2 4 6 8 0
Fig. 9.
I
I z4
10
68
20
2 4 6 8
2 4 6 8
30 T i m e in m i n u t e s
40
2 4 6 8
50
24
,, 6
/ 8
60
I n t e r f e r e n c e o n d i r e c t e x p o s u r e to light s o u r c e s in o p e r a t i n g r o o m .
Fig. 6 illustrates typical transient response to impulse and steady-state changes respectively.
2) The consistency of measurement during steady-state The consistency refers to stability of the values displayed at a given percent SaO2 level. The stability is affected mostly by the empirical relation for SaO2 and by the statistical technique employed. The actual measured quantity in a pulse oximeter is not % SaO2 but absorption ratio (described in Part I). The absorption ratio is mapped to SaO2 using an empirical relation or a table look-up procedure. This is were the instability shows up, the relation between the absorption ratio and % SaO2 is not linear, consequently small variations or errors in the absorption ratio may give large changes in % SaO2 depending on the measured region of % SaO2. In the relation this distortion grows high-
er as the % SaO2 drops lower than 85%. In practice, we found that all the three oximeters shows fluctuation below 75%. In our study, Ohmeda showed large fluctuation as compared to our prototype or Nellcor while the mean of the fluctuation was closer to the reference % SaO2 when compared with the other two. Ohmeda's large fluctuation may be due its quick response to changes. NeUcor and our prototype showed fair stability in lower ranges. The stability exhibited by our prototype is primarily due to the statistical median method employed which is insensitive to the magnitude of small and large values of the population. We also observed sometimes Nellcor withdraws the value from display then it restarts the measurement after a while. The lack of accuracy at low saturation has been well known. [1, 2, 4]. It may be due to lack of sufficient clinical data in the low range of % SaO2 for the empirical relation. Industry standard oximeters are reported to produce errors as large as 10% at saturation levels between 70% and 50% [4].
55 %Sa02 100 i ]
PATIENT'S PROFILE AGE: II M o n t h s
9~
:',
Diagnosis: Right Blalock Shunt
Short sporadic problems during measurement.
6 4 2 808 64-270-864260--
State: PostanesthesiaPresternotomy
Legends: dreifts d u r i n g
- D e v e l o p m e n t Module - 0hmeda . . . . . . . Nellcor IL282 C O - o x i m e t e r
S 0
2 4 6 8
10
2 4 6 8
20
2 4 6 8 2 4 6 8 2 4 6 8 2 4 6 8 30 40 50 60 Time in m i n u t e s
Fig, 10. C o m m o n problems during m e a s u r e m e n t on children with congenital heart diseases.
The trend pattern during this circumstances usually appears as shown in Fig. 7.
ermia condition produces abnormalities in the measurement as depicted in the Fig. 8.
3) The influence of motion artifacts and hypothermia
4) The influence of ambient light
Most of the time the motion artifact is created by continuous disturbance of the probe especially due to shivering. Under this circumstances Ohmeda frequently showed different readings and warnings while Nellcor and our prototype module showed considerable stability. However, when the body temperature is very low (temperatures below 33 degree Celsius) Ohmeda could display reading consistently while Nellcor and the prototype model failed to measure at times. This may be probably to do with the sensitivity of the probe. Small movements, though under normal circumstances don't produce erroneous results, under certain combination of poor peripheral circulation and/or hypoth-
During our clinical trials we found that all the oximeters readings were badly affected when the probes were directly exposed to operating lamps. We observed that infrared plethysmograph signal was badly disturbed. However, when the probe was covered with a cloth, the disturbance was drastically reduced. Interference due to infrared heat lamps was reported by Brooks TD and others [5]. Under the influence of the light source, Nellcor was severely affected, our prototype module was moderately affected (the reduced influence was due to measurement at selected instances in the signal, as described in Part I) and Ohmeda was less influenced (probably because of the sealed design of its non-disposable probe). It should be noted that nor-
56 mal lamps in the room don't affect the measurement as claimed by the manufacturers. SaO2 measurements under direct exposure to operating lamps generally appear as shown in Fig. 9.
5) Accuracy of measurement on infants Paediatric surgery on infants with congenital heart diseases are performed frequently at our centre. We carried out our evaluation of oximeters on such paediatric patients. We used disposable children's probe for Nellcor and for our prototype module while for Ohmeda we used a non-disposable probe. All the probes used were for the finger. Evaluation of the oximeters with infants were very tough. Especially finding proper probe placement was quite difficult. Due to difficulty in placing the probe over finger and to reduce motion artifact we placed the probes on toe. For healthy subjects the readings were stable and correlated with laboratory measurements for all the three oximeters, while for weak subjects and infants with congenital heart problems the readings were often inaccurate. Ohmeda showed occasionally close values to laboratory results but frequently its readings drifted around. Nellcor and our module in this cases showed values higher than the actual value consistently but reproducing the trend shapes well in agreement with laboratory readings. These effects are illustrated in Fig. 10. The importance for accuracy in measurements on premature infants when SaO2 may hold a constant level of desaturation has been strongly emphasized by Severinghaus and others [2]. Also it has been reported that fairly good measurements are possible in children even while using adult's probe by C.J. Cote and others
[6]. Possible tanning of the skin under the continuous light from the probe is another major factor with respect to measurement on infants. We found in our study, that Ohmeda probes cause more tanning of the skin than the other two oximeters. Using adult's probe on infants may cause more severe tanning in a shorter time.
6) Influence of intravenously administered dyes As we were conducting our evaluation during routine clinical procedure, we did not carry out studies on the influence of iv-administered dyes on SaO2 measurement. However, it has been well reported by M.S. Scheller [7] and others [8]. The result of their investigation is that intravenously administered dyes influences SaO2 readings strongly to produce very low false values. This is well supported by the fact that the absorption of plethysmograph signal by the dyes is quite strong at red wavelength which is used in pulse oximeters. Among the dyes methylene blue shows strong influence, indocyanine green shows moderate influence and indigo carmine shows less influence [7].
Conclusion
The clinical usefulness of a pulse oximeter is proved beyond doubt [1, 4, 9, 10] and has been extensively evaluated by Severinghaus, Chapman and others [2, 3, 4]. In our investigation as we found that most our findings on the overall behaviour of the pulse oximeters were well in agreement with the results published in various clinical research literatures, we concentrated on peculiarities that may be encountered during routine usage in clinical environment. The computerized protocol which we implemented for our study of realtime data collection and multi-colour plotting of the trend curves in a graphical form proved to be very useful and made our clinical investigation easier. Some of the charts which will be of clinical interest were presented in this paper to serve as illustrations. While most of the problems exhibited by the oximeters during our random clinical evaluation study may be solved by the manufacturer themselves, few of them may remain unsolved. In practice, these problems may need to be identified and corrected whenever possible by practising clinicians or nurses.
57
References 1. Tremper KK, Barker SJ. Pulse Oximetry. Anesthesiology 1989; 70: 98-108. 2. Severinghaus JW, Naiefeh KH. Accuracy of Response of Six Pulse Oximeters to Profound Hypoxia. Anesthesiology 1987; 67: 551-558. 3. Kagle DM, Alexander CM, Berko RS, Giuffre M, Gross JB. Evaluation of the Ohmeda 3700 Pulse Oximeter: Steady-state and Transient Response Characteristics: Anesthesiology 1987; 66: 376-380. 4. Chapman KR, Liu FLW, Watson RM, Rebuck AS. Range of Accuracy of Two Wavelength Oximetry. Chest/89/4/ 540-542, April, 1986. 5. Brooks RD, Paulus DA, Winkle WE. Infrared heat lamps interfere with pulse oximeters. Anesthesiology 1984; 61: 630. 6. Cot6 CJ, Goldsteing EA, Cot6 MA, Hoaglin DC, Ryan JF. A Single-blind Study of Pulse Oximetry in Children. Anesthesiology 1988; 68: 184-188.
7. Scheller MS, Unger RJ, Kelner MJ. Effectsoflntravenously Administered Dyes on Pulse Oximetry Readings. Anesthesiology 1986; 65: 550-552. 8. Kessler MR, Eide T, Humayun B, Poppers Pj. Spurious Pulse Oximeter Desaturation with Methylene Blue Injection. Anesthesiology 1986; 65: 435436. 9. Cohen DE, Downes JJ, Raphaely RC. Editorial Views: What Difference Does Pulse Oximetry Make? Anesthesiology 1988; 68: 181-183. 10. Fairley HB. Editorial Views: Changing Perspectives in Monitoring Oxygenation. Anesthesiology 1989; 70: 2-4.
Address for offprints: S. Meiyappan, c/o Omar Prakash, Thorax Centre, Erasmus University, Postbus 1738, 3000 DR Rotterdam, The Netherlands
