Comparison of nurse and computer charting of physiological variables in an intensive care unit

International Journal of Clinical Monitoring and Computing 13: 235-241, 1997. 235 (~) 1997 Kluwer Academic Publishers. Printed in the Netherlands.

Comparison of nurse and computer charting of physiological variables in an intensive care unit

Steven Cunningham, Sarah Deere, Robert A. Elton* & Neil McIntosh Department of Child Life and Health, University of Edinburgh, 20 Sylvan Place, Edinburgh, EH9 1UW,, UK; * Medical Statistics Unit, University of Edinburgh, Medical School, Teviot Place, Edinburgh, EH8 9AG, UK

Received 15 April 1996; accepted in final form 24 October 1996

Key words: computers, intensive care, neonate, artifact, trend

Abstract

Objectives. To compare charting of physiological parameters manually by nurses and automatically by computer and so decide whether this task could accurately be performed by computer. Subjects~setting. 101 consecutive patients admitted for tertiary neonatal intensive care. Design. Direct comparison of 48 hours of data collected by both methods. Computer data stored each second and the hourly median compared with the single hourly value noted by the nurse. Methods. All patients were monitored by standard patient monitor and a computer. Four physiological parameters were compared between nurse and computer (both derived information from the standard patient monitor): heart rate, transcutaneous oxygen, mean blood pressure, central temperature. A random 51% of patients had the computer data displayed as trends at the cotside. Comparison of the hourly nurse observation and a computer hourly median value. Computer data was compared before and after the removal of artifact. In addition, the effects on nursing observations of either display or non display of the computer trend data was assessed. Results. Nurse and computer observations were statistically significantly different (p < 0.001), though these were not clinically important. Nurses tended to note a higher figure than the computer median. The cotside display of computer data improved consistency between the nurse and computer observations. Artifact present in the data had little influence on the accuracy of the computer median value. Conclusions. Computer systems can accurately chart physiological data, providing a more flexible record with a minimal risk to data reliability from artifact.

Introduction

Computer collection of physiological parameters already occurs in 4.8% of USA hospitals [1] and in a survey of UK neonatal intensive care units, 50% either already used or wished to develop computerised collection of physiological variables [2]. Yet clinical practice however continues to rely predominantly on handwritten nursing observation, even in many of those centres with computer collection of data. Handwritten data is time consuming, inflexible, open to inaccuracy and limited in its sampled frequency, especially at times when the patient is most unstable [3, 4]. With an increasing workload placed on fewer nurses, the automated recording of physiological parameters could liberate nurses from this routine task, increasing time available for them to perform other duties.

The paucity of published studies validating this method of data collection in intensive care has been noted [5]. Computer recorded intracranial pressure (720 data points each hour) and a single nurse hourly observation were found to be comparable [6]. Three other studies have demonstrated an improvement in detection of particular incidents in intensive care areas with computer monitoring [4, 7, 8], but did not compare routine collection of physiological data by nurses and computers.

The rapid development of computer technology in intensive care areas and the deep mistrust of this technology [9], warrant a careful evaluation. We have compared the collection of physiological data by nurse and computer assessing three areas (a) how comparable are the two methods of data collection, (b) can computer display of physiological information influence nursing

236

observations and, (c) is the inclusion of artifact in computer collated data an important consideration for the reliability of this form of data collection.

Patients and methods

A network of cotside computers links all 12 intensive cots of the tertiary Neonatal Unit at the Simpson Memorial Maternity Pavilion, Edinburgh. This com- mercially available system (Mary Neonatal Monitor- ing, Loughborough, UK) receives information each second on up to 32 channels from standard neonatal physiological monitors (Hewlett Packard 78834A) and displays this information at the cotside as a maximum of five continuously updated trends [10]. The choice of channels displayed and their time scale (22 variants) can be varied by the nurse depending on patient stability. One minute averages of data are permanently stored onto a central server.

101 consecutive admissions to the intensive care unit were assessed. We retrospectively compared four physiological variables simultaneously recorded (from the same physiological monitor) both by nurses on to paper observation charts and also automatically by the cotside computer. The four physiological parameters assessed were: heart rate, transcutaneous oxygen, mean blood pressure and central temperature. The study was performed retrospectively, no prior instruc- tions were given to the nurses as this might have altered their usual behaviour. The nurses noted a single value each hour, whilst the computer accessed data each second (subsequently averaged to one minute). The computer data was assessed in hourly time periods for 48 hours from birth. A median value was calculated by the computer for each hourly time period. The median value was considered optimal as it would be minimally affected by the temporary instability caused by handling etc. We then compared the corresponding hourly nursing observation and the calculated hourly computer median for each of the four parameters (Table 1), plotting the date according to Bland and Altman and including a ca, Is ulation of the limits of agreement (Figure 1).

In addition, a ~ part of a larger study on the effects of computerise,J monitoring on outcome, infants born < 32 weeks gestation, or > 32 weeks and requiring ven- tilation, were randomly assigned to have the computer monitored data either not displayed (A) or, displayed (B) on the cotside screen. Randomisation was by open- ing of a sealed sequentially numbered randomisation

envelope. To assess whether the presence of computer monitoring influenced the hourly nurse observations, we aggregated the mean and standard deviation of the nurse-computer differences for each infant in each of the four parameters. The two groups of infants (A and B) were then compared using a two sample t test.

The presence of artifact in clinical physiological data is inevitable. Some of our artifact was removed by excluding values outside extreme physiological limits (heart rate < 5 or > 250 beats per minute; transcutaneous oxygen < 0.7 or > 20.0 kPa; mean blood pressure < 5 or > 250 mmHg; central temperature < 22.5 or > 45 ~ Removal of artifact within these limits involved a review of the data by two trained observers [11]. In the analysis of comparability of nurse and computer (above), all artifact was removed. However, manual removal of artifact is time consuming, subjec- tive and not possible on a real time basis. We therefore wished to assess how much the manual removal of data influenced the computer median, and if so, was this influence subtle or extreme. A subset of the first 34 infants recruited had the computer median compared with and without the manual removal of artifact in each of the four parameters (Figure 2).

Results

Difference between nurse and computer recordings

Of the 101 infants recruited, two infants had no computer recorded data and two had missing nurse observation charts. The analysis therefore concerns the remain- ing 97 infants, for whom a maximum total of 4656 hours of data could have been collected (Table 1). Missing data points for heart rate and transcutaneous oxygen were predominantly due to infants having early transfer from the intensive care unit to the special care baby unit. Mean blood pressure and temperature measurements were only made continuously in the sickest patients, consequently they have far fewer data points than heart rate or transcutaneous oxygen.

The difference between the hourly nurse reading and the computer hourly median was calculated for every hour observed and plotted as Bland-Altman plots against the average (Figure 1) [12]. The figure demonstrates a close relationship between nurse and computer recordings. In all cases the nurses predominantly recorded higher than the computer. The temperature graph demonstrates a close agreement between the

Table 1. Differences between nurse and computer

237

Number of

comparisons

Mean difference between

nurse and computer

for all patients

mean sd

95% confidence

intervals

within each hour

Mean of all computer

sd's for minute data

Limits of agreement

Heart rate 3442 3.76

(bpm)

Transeutaneous 3141 0.20

oxygen (kPa)

Mean blood 2010 0.67

pressure (mmHg)

Central 1129 0.42

temperature ( o C)

8.92 3.46, 4.06

1.21 0.16, 0.24

4.11 0.49, 0.85

0.55 0.39, 0.45

7.16 - 14.08, + 21.60

0.88 - 2.22, + 2.63

2.67 - 7.55, + 8.89

0.20 - 0.68, + 1.52

nurse and computer when the temperature was > 36 ~ C, but poorer agreement below this.

Table 1 shows the mean and standard deviation of the differences between nurse and computer together with their 'limits of agreement' (mean + 2 sd) and confidence intervals. The mean difference indicates the average bias between the nurse and computer. All are statistically significantly different from zero (expected if there were no difference) (t test p < 0.001 for all four parameters). The differences are small enough to be clinically unimportant. The standard deviation of the nurse-computer differences is a measure of precision, indicating how variable the differences are from hour to hour. For comparison, Table i shows the mean standard deviation of the minute by minute computer readings for each hour. These should be similar to the standard deviation of the nurse-computer differences if the nurses were accurately recording a random minutes' reading; the standard deviation of the computer recordings are actually somewhat smaller, con- firming that other sources of inaccuracy exist between nurse and computer.

Effect of screen display on nurse-computer difference

Ninety-two of the ninety-seven infants assessed for comparability of nurse/computer data were eligible to be enrolled into the study comparing the display/non display of computer trend data. Of these, 43 infants were randomised to no screen display and 49 to screen display. The two sets of infants did not have significant differences in either birthweight or gestation (Table 2). The mean and standard deviation of the differences between the nurse and computer observations were compared in the two groups of infants. Of the

eight comparisons tested, two demonstrated a statistically significant reduction in the difference between the nurse and computer observations when the computer information was displayed, together with a trend for less difference in a further four comparisons. This is more than one might expect by chance.

Effect of physiological artifact on computer-nurse agreement

A total of 6.9% of data was excluded as artifact from the hourly computer medians prior to the comparison with nursing observations. 0.6% of data was excluded automatically by the computer as being outside prede- termined upper and lower limits. A further 6.3% of data was excluded manually by the trained observers. The computer median was little influenced by the manual removal of artifact, although as expected the results showed less precision when artifacts were included. Occasional large differences were seen before and after artifact removal, though these represent only a small percentage of observations (Figure 2). The outliers in heart rate are predominantly due to electrical lead fault and those in transcutaneous oxygen by air between the skin and probe. Mean blood pressure artifact is caused by a blocked or damped arterial line, and the single outlier in temperature occurred when the temperature probe fell off the skin.

Discussion

Computers are increasingly prevalent in all areas of medicine, especially areas of intensive therapy. The ability of computers to summarise and expand stored

238

(o1

I !

"6

8

0

8

�9 �9 . ..'.:" "..-~i" ;',.'. �9 e l �9 �9 �9 �9

I . . . . . . . . . . . . . . . . . . . . . ' ; ~ l t ~ . ~ ' ~ , ~. "V . . . . . . . . . . . . - . ' " ~ . . . . . . . . . . . . . . . . . . . . . . �9 . . / . , , , ' . . : . . �9 .

I I ! I I I I I

80 tO0 120 140 160 1~0 200 220

Ivlean ol hea~ r (l~ts/min)

Ibl

t 1 o

.:..,.,.,?.. �9

." t~'.~ .~"~.,{r " ' :.~,.,..~.~ ............................ .............. " . ' :~.~/C,~"[r ~ . ~ �9 7 . ' . . .

�9 . . ' ' . . . ~ .~ . . : . , . �9 " . . ..

- ' - ~ ~ ' -"4 , ...................

I I I I I l

4 6 8 10 12 14

Mean ol banscutaneou8 oxyoerm (KPa)

(cl

1 " ! o

�9 .~ t �9 '. . " . " ............ . ................ L~.'. .......... ~,.l.'..,~.L....:-'- ......................... , ............

, , ~ : ? . . . ~ l I I ~ / = I j k - ~ . �9 . , . .

................. :.:;.~ -__i~,i~.~._ia.L'_.._.:__, ..................................... �9 9 e . , , � 9 1 4 9 �9 �9 �9

I I I I f I

I0 2o 30 40 60 60 7o

Mean ol ~ pree~ree (mmXO)

Idl ~ - [ %

O

�9 �9

�9 �9 ee �9 e i~

�9 o : �9 o ~ l

. . . "'.?.~T'_ "

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . " .............. . - - - : v , " ~ ......................................... �9 - - . - -

I I I I I

34 3S 36 37 38

Mean ol ~ m l ~ (degemr e . e l ~ l

Figure 1. Difference between nurse and computer observations plotted against the mean value (limits of agreement shown as broken lines).

data for varying levels of scrutiny is of great poten- tial benefit, in our increasingly complex intensive care areas. The single hourly physiological values currently noted by nurses may be poorly representative of the patient during that' hour and may not occur at all if the patient was too unstable. Computer systems could therefore potentially provide an automatic, labour sav- ing and hopefully more accurate record of patient details [13].

The mean differences between nurse and computer in our study were clinically of little significance, however the nurses had a consistent tendency to note

a higher value and appeared to be influenced in their choice of data point (Table 1). We feel that the nurses subconsciously attempted to present a more stable picture of their patient. Procedures and instability fre- quently produce a decrease (hypoxia [14], bradycardia [15], hypothermia [16]) or large fluctuation-blood pressure [17]) of values in intensive care infants. A nurse may recognise that temporary changes in vital signs are untypical for that patient and might choose to ignore them and either await a period of stability or make an educated estimate for that hour. All three of these possibilities (random choice, await stability,

(a)

i i 0

I I ! I I

0 5o I00 150 200

HeArt ta le w ~ ( ~ t a~' far ( b e a ~ / m l n )

Ib)

i i o

i n

239

/" �9 A A A

o 5 Io 15 20 25

Tran~utaneous oxyWm w~th~t a~act (KPa)

It)

J / I I I I

20 40 60 80

Blood Ixessure w~houl a~facl (mmXo)

|all

]

I' i' I I I I

25 30 35 40

Temperalure without mli~acl (degrees Celsius)

Figure 2. Hourly computer medians with artifact plotted against those with artifact removed.

estimate) will occur to an extent in handwritten observations depending on the level of training and expertise of the nurse. It is impossible when viewing handwritten trends to determine how the hourly value has been derived [18]. The computer is more consistent, it is not subject to observer variation or the level of nurse training. In addition, our randomisation of infants to display or non display of physiological trend data, did appear to have some influence on nursing judgement (Table 2), with the computers trends possibly acting as an aide memoir to the physiological progress of their patient during that hour [19].

A well trained nurse is able to identify erroneous data and can ignore this when making decisions about the infant. We therefore also evaluated how much the presence of artifact within physiological limits would influence a computer hourly median value (Figure 2). Of the few values affected by artifact, many would be clinically obvious. As a consequence we do not feel that artifact within present physiological limits, significantly reduces the reliability of real time physiological data. Expert computer systems should in the future be able to identify and reject artifact [20].

240

Table 2. Comparison of randomised groups for baseline characteristics and means and standard deviation of nurse computer differences. Figures shown are mean (sd) of values for infants in each group

No screen display Screen display 95% confidence Significance

interval

Number of infants 43

Gestation (weeks) 32 (5)

Birthweight (g) 1886 (889)

Heart rate Mean 4.61 (3.16)

(beats per minute) Sd 8.58 (2.90)

Transcutaneous oxygen Mean 0.23 (0.54)

(kPa) Sd 1.25 (0.43)

Mean blood pressure Mean 0.74 (1.61)

(mmHg) Sd 4.22 (2.31)

Central temperature Mean 0.61 (0.71)

( ~ Sd 0.38 (0.20)

49

31 (4) ns

1640 (828) ns

3.87 (4.57) - 0.94, + 2.42 ns

8.72 (6.21) - 1.96, + 2.24 ns

0.22 (0.37) - 0.19, + 0.21 ns

1.04 (0.28) + 0.05, + 0.37 p < 0.01

0.30 (1.43) - 0.35, + 1.23 ns

3.02 (1.05) + 2.26, + 2.14 p < 0.05

0.52 (0.66) - 0.35, + 0.53 ns

0.47 (0.36) - 0.28, + 0.09 ns

Reducing routine paperwork would allow nurses more time for direct patient care [21], though frequent- ly time spent on nurse recordings are not reduced by computers [22, 23]. The computer system we assessed required minimal input from nursing staff to obtain detailed physiological trends. We have demonstrated that collection of physiological data by computer is an alternative to nurse charting of physiological parameters, and that the presence of artifact (within physiological limits) is of little practical significance.

Acknowledgements

Dr. Cunningham and S.N. Deere were funded by the Scottish Home and Health Department. Our thanks to the staff of the Neonatal Intensive Care Unit, Simpson Memorial Maternity Pavilion, Edinburgh for their help and support with this project.

References

Summers S, Ratcliff C, Becker A, Resler M. Nurses utilisation of computers to document nursing care in the hospital setting. In: Kingsland LC III ed. Symposium on computer applications in medical care; 1989 ~qov 5-8. Washington DC: IEEE Computer Society Press 1989:834-838

Cunningham S. Computerised physiological monitoring in neonatal intensive care [Ph.D. thesis]. Edinburgh: University of Edinburgh 1993

Gardner RM, Shabot MM. Computerised ICU data management: pitfalls and promises. Int J Clin Mort Comp 1990; 7:99-105

Kari A, Ruokonen E, Takala J. Comparison of acceptance and per- formance of automated and manual data management systems in intensive care. Int J Clin Mon Comp 1990; 7:157-162

Sittig DF, Gardner RM. Summary of the 10th annual international symposium on computers in critical care, pulmonary medicine and anaesthesia. Int J Clin Mon Comp 1990; 7:33-39

Turner HB, Anderson RL, Ward JD, Young HF, Marmarou A. Com- parison of nurse and computer recording of ICP in head injured patients. J Neurosci Nurs 1988; 20:236-239

Aukburg S J, Ketikidis PH, Kitz DS, Mavrides TG, Matchinsky BB. Automation of physiological data presentation and alarms in the post anaesthesia care unit. In: Kingsland LC III ed. Symposium on computer applications in medical care; 1989 Nov 5-8. Wash- ington DC: IEEE Computer Society Press 1989:580-582

Corrie J, Piper I, Tocher JL, Jones P, Anderson S, Andrews PJD et al. Microcomputer based data recording. Brit J Int Care 1993; 3: 225-233

Goodwin J. The importance of clinical skills. BMJ 1995; 310: 1281- 1282

Cunningham S, Deere S, Elton RA, Mclntosh N. Neonatal physiological trend monitoring by computer. Int J Clin Mon Comp 1992; 9:221-227

Cunningham S, Symon AG, Elton RA, Mclntosh N. The practical management of artifact in physiological data. Int J Clin Mon Comp 1994; 11:211-216

Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986; 307-310

Colvin JR, Kenny GNC. Computerised management of intensive care unit data. In: Coleridge Smith PD, Scurr JH eds. Medical applications of microcomputers. London: Springer Verlag 1988: 51-59

Long JG. Excessive handling as a cause of hypoxaemia. Pediatrics 1980; 65:203-207

Cordero L, Hon EH. Neonatal bradycardia following nasopharyngeal stimulation. J Pediatr 1971; 78:441-447

Mok Q, Bass CA, Ducker DA, Mclntosh N. Temperature instability during nursing procedures in preterm infants. Arch Dis Child 1991; 66:783-786

Perlmann JM, McMenamin JB, Volpe JJ. Fluctuating cerebral blood flow velocity in respiratory distress syndrome. N Engl J Mid 1983; 309:204-209

Gardner RM, Hawley WL, East TD, Oniki TA, Young HFW. Real time data acquisition: recommendations for the Medical Infor- mation Bus (MIB). Int J Clin Mon Comp 1992; 8:251-258

Green CA, Gilhooly KJ, Logie R, Ross DG. Human factors and computerisation in intensive care units: a review. Int J Clin Mon Comp 1991; 8:167-178

Cunningham S, Deere S, Bass CA, Mclntosh N. Approaches to artifact identification and manipulation in a computerised cot monitoring system. In: Lafeber H, ed. Fetal and neonatal physiological measurements. Amsterdam: Elsevier Science Publications 1991: 191-194

Cook M. Using computers to enhance professional practice. Nursing Times 1982:1542-1544

Pryor TA. Computerised nurse charting. Int J Clin Mon Comp 1989; 6:173-179

241

Bradshaw KE, Sittig DF, Gardner RM, Pryor TA, Budd M. Improving efficiency and quality in a computerised ICU. In: Greenes WW, ed. Symposium on computer applications in medical care; 1988 Nov 6-9. Washington DC: IEEE Computer Society Press 1988: 763-767

Address for correspondence: Dr. Steven Cunningham, Dept. of Child Life and Health, University of Edinburgh, 20 Sylvan Place, Edinburgh, EH9 1UW, U.K. e-mail: [email protected], Fax: (0)131 536 0821

Documents

Comparison of nurse and computer charting of physiological variables in an intensive care unit