Computer versus manual measurement of ST-segment deviation

Journal of Electrocardiology Vol. 29 Supplement

Computer versus Manual Measurement ST-Segment Deviation

of

Michele M. Pelter, RN, MS, Mary G. Adams, RN, MS, and Barbara J. Drew, RN, PhD, FAAN

A b s t r a c t : Total ST scores (sum of absolute deviations in all 12 electrocardio-

graphic [ECG] leads) h a v e b e e n used for research purposes to estimate total ischemic b u r d e n and to predict reperfusion after t h r o m b o l y t i c therapy. C o m puterized m o n i t o r i n g systems are capable of m e a s u r i n g ST deviation to the 10-btV level, w h e r e a s h u m a n s are incapable of such precise resolution. The purpose of this study was to c o m p a r e c o m p u t e r versus m a n u a l ST scores in 12-lead ECGs exhibiting ischemia and to c o m p a r e interrater reliability of m a n u a l m e a s u r e m e n t s b e t w e e n two experts. A total of 58 subjects w i t h 100 btV or m o r e ST deviation in one or m o r e leads during p e r c u t a n e o u s transluminal coronary angioplasty balloon inflation w e r e selected for analysis. ST m e a s u r e m e n t s w e r e m a d e at J + 80 ms, using the isoelectric line as a reference, a n d s u m m e d across all 12 leads. M a n u a l m e a s u r e m e n t s w e r e m a d e to a m i n i m u m of 50 gV by two i n d e p e n d e n t r e v i e w e r s blinded to the c o m p u t e r scores. Total ST scores w e r e c o m p a r e d using paired t-tests, and Pearson coefficients w e r e used to test the correlations. A high correlation was observed b e t w e e n the m a n u a l and c o m p u t e r m e a s u r e m e n t s (r = .96, P < .00) and b e t w e e n the two r e v i e w e r s (r = .96, P < .00). A high degree of interrater reliability is possible w i t h m a n u a l m e a s u r e m e n t s of ST deviation. C o m p u t e r m e a s u r e m e n t s are consistently greater t h a n m a n u a l m e a s u r e m e n t s , p r e s u m a b l y because h u m a n s " r o u n d d o w n " to the nearest 50 btV. As such, c o m p u t e r s m a y detect ischemia that is missed by h u m a n s . However, c o m p u t e r and m a n u a l m e a s u r e m e n t s of ST deviation should n o t be m i x e d w h e n used as a variable for research. K e y w o r d s : c o m p u t e r m e a s u r e m e n t , m a n u a l m e a s u r e m e n t , ST deviation, ischemia, interrater reliability.

C o n t i n u o u s 12-lead ST-segment m o n i t o r i n g has b e e n used to detect s u d d e n reocclusion following percutan e o u s t r a n s l u m i n a l c o r o n a r y angioplasty (PTCA) (1,2)

and to predict c o r o n a r y p a t e n c y after t h r o m b o l y t i c therapy (3-6). Using an ST deviation score provides a m e t h o d to q u a n t i f y myocardial ischemia and to c o m p a r e serial electrocardiograms (ECG) o v e r time. This is accomplished by m e a s u r i n g ST-segment deviation across all 12 leads, in microvolts or millimeters, and t h e n s u m m i n g all absolute ST-segment deviations. Visual i n t e r p r e t a t i o n of ST-segment deviation is limited, as described in the literature, by high interrater variability (6,7), extensive time for analysis (8), and failure to detect ischemic ECG changes too subtle to be observed w i t h the h u m a n eye (9). C o m p u t e r i z e d ECG

From the University of California, San Francisco, San Francisco, California. Supported by an award from the National Institute of Nursing Research, National Institutes of Health (1 RO1NR03436), Bethesda, MD. Reprint requests: Michele M. Pelter, RN, MS, University of California, San Francisco, School of Nursing, N611, San Francisco, CA 94143.

78

Measurement of ST-Segment Deviation systems that provide ST-segment measurements may lessen or eliminate these limitations. For example, the measurements would be free of h u m a n bias, microvolt measurements are measured to a precision of 10 bt¥, STsegment measurements are immediately available for analysis, and the storage and preservation of hundreds of ECGs for one patient is possible. Previous researchers have addressed interrater reliability w h e n measuring ST-segment deviation and the use of computerized ST-segment measurements in patients undergoing exercise testing (7,8,10,11). However, a comparison of manual versus computerized ST-segment measurements during continuous ST-segment monitoring of hospitalized patients has not been reported previously. The purpose of this study was (1) to compare manual versus computer ST scores in 12-lead ECGs exhibiting ischemia and (2) to compare the interrater reliability of manual measurements between two experts.

Patient Population The sample was selected from patients enrolled in an ongoing study assessing the value of continuous ST-segm e n t monitoring, using a derived 12-lead ECG (12), for the detection of ischemia in patients with unstable angina or acute myocardial infarction (MI) (2). Subjects included in this study were all patients w h o had ECG evidence of ischemia during a n o n e m e r g e n t PTCA procedure for unstable angina. Excluded were patients with paced ventricular rhythms or left bundle branch block, since interpretation of ST-segment deviation is difficult in these patients.

Equipment All 12-lead ECG data collected in the adult cardiac catheterization laboratory was obtained using the Mortara ELI 100 ST Monitor (STM) (Mortara Instruments, Milwaukee, WI). The ELI I00 STM is a real-time digital electrocardiograph, which analyzes all 12 ECG leads every 20 seconds. At every 20-second interval, a "median beat" is generated for each of the 12 leads, which serves as the template from which subsequent ECGs are compared. The median beat ECG is updated with every 20second analysis. The ELI 100 STM was programmed to store a 12-lead ECG every 10 minutes, or if an ST-segment change of 100 btV or more was detected in at least one of the ECG leads lasting longer than 60 seconds. The ELI 100 STM uses the PR segment as the isoelectric reference point and was configured to measure ST-segment deviation at the J point plus 80 (ms). After ECG data are collected in the adult cardiac catheterization laboratory, all stored ECGs are transmitted to an ST Review Station computer (Mortara Instruments). The ST Review Station computer software trends the 12-lead ECG data, allowing analysis of each 12-lead

•

Pelter et al.

79

ECG, as well as displaying the microvolt values for every 12-lead ECG. Computer microvolt values are measured from the median beat. It is possible to print the medianbeat ECG from the ST Review Station; thus, this was the 12-Iead ECG both of the reviewers measured.

Methods For each patient, an ischemic 12-lead ECG was chosen at the end of the first PTCA balloon inflation. Ischemia was defined as transient ST-segment deviation of 100 #V or more in one or more leads of the 12-lead ECG. Two reviewers, both with extensive training and experience in ST analysis, manually measured each patients' ischemic 12-lead ECG. Both reviewers used a transparent ruler and calipers for measuring and each was blinded to the other's measurements, as well as to the computer values. The PR segment served as the isoelectric reference point, and ST-segment deviation (elevation or depression) was measured from the J point plus 80 ms to the nearest 50 ~V. Absolute microvolt values of ST-segment deviation were then summed across all 12 leads of the ECG; this ST score served as the unit of analysis for each patient.

Statistics Paired t-tests were used to determine w h e t h e r differences in ST scores between the two reviewers and b et w een the computer and reviewers were statistically significant. Pearson coefficients were used to determine the correlation b et w een the two reviewers and between each reviewer and the computer, with 1,O representing a perfect correlation.

Results The sample included 58 patients. The PTCA vessels dilated included 30 in the left anterior descending coronary artery, 12 in the right coronary artery, and 16 in the circumflex artery. All of the 58 subjects were in sinus rhythm, 1 had left ventricular hypertrophy, and 4 had right bundle branch block. The range of ST scores for reviewer 1 was from 400 to 4,850 ~tV; for reviewer 2, it was from 400 to 4,150 btV; and for the computer, it was from 601 to 5,303 btV. The lowest ST score of all 58 patients was during left anterior descending PTCA balloon inflation. Reviewer 1 measured an ST score of 400 btV, reviewer 2 measured an ST score of 400 btV, and the computer measured an ST score of 601 btV (Fig. 1). The m a x i m u m ST score of all 58 patients was also during left anterior descending PTCA balloon inflation. Reviewer 1 measured an ST score of 3,800 ~tV, reviewer 2 measured an ST score of 3,900 [,tV, and the computer measured an ST score 5,303 btV (Fig. 2).

80

Journal of Electrocardiology Vol. 29 Supplement

Fig. 1. Lowest ST score of all 58 patients during left anterior descending percutaneous transluminal coronary angioplasty.

aVL

Computer Reviewer #1 Reviewer #2

0 0 0

I

-aVR

II

aVF

-8 0 0

10 0 0

-8 0 0

-5 0 0

Mean ST scores were determined for all 58 patients. The mean ST score for reviewer 1 was 1,306 ~tV, reviewer 2 was 1,350 ~tV, and the computer was 1,674 ftV. A high correlation (r = .96) was observed between the two reviewers. Figure 3 shows a scatter plot demonstrating the variance between the two reviewers. On average, the two reviewers' ST scores differed by 44 #V (not significant). Figures 4 and 5 are diagrams using a line of equality to demonstrate the variance between each reviewer and the computer measurements (13). On average, the computer ST scores were 368 ~tV higher than the ST scores for reviewer 1 (P < .00). Comparing the ST score for reviewer 2 to the computer, on average, measured 324 ~tV higher (P < .00). In all 58 patients, the computer ST scores were higher than both of the reviewers ST scores. Table 1 illustrates m e a n ST scores comparing each reviewer to the other and then each reviewer to the computer measurements.

The first conclusion from this study is that a high degree of interrater reliability is possible with m a n u a l measurements of 12-lead ECGs demonstrating ST-segm e n t deviation. This was demonstrated by the high correlation between the two reviewers (r = .96) and the small difference (44 mV) in m e a n ST scores. Similar to our findings, Caralis et al. demonstrated a high interrater reliability between cardiologists measuring ST-segment changes observed during exercise testing (8). In their study, a total of four cardiologists measured ST-segment deviation during exercise testing in 20 subjects. The correlation between two cardiologists, using a computerassisted measurement program, was .90. The correlation for the other two cardiologists measuring ST-segment deviation without the computer-assisted measurement program was .76. Although the two observers using the

Computer Reviewer #1 Reviewer #2

V1

V2

V3

V4

V5

V6

ST Score

55 50 50

80 50 50

180 150 150

135 100 t00

95 50 50

-20 0 0

601 (pV) 400~V) 400(,uV)

computer-assisted ST measuring program demonstrated a higher correlation than the observers who did not, both sets correlated well with one another. Both Caralis and co-workers' study and our study demonstrate that it is possible for two reviewers, presumably with similar training and experience, to measure ST-segment deviation closely and consistently with one another. In contrast to our findings, Blackburn et al. reported results to the contrary (7). In their study, 14 cardiologists were asked to interpret the exercise ECGs of 38 asymptomatic subjects. The observers were to code the ECG as normal, abnormal, borderline, or technicalIy uninterpretable. A high degree of interrater variability was reported when assessing an exercise ECG as abnormal (from 5 to 58%). This represented a I0-fold interobserver difference. These researchers did find that interrater reliability improved, to nearly acceptable levels, w h e n a subgroup of their observers (6) were asked to use quantitative criteria w h e n assessing the exercise ECGs, suggesting perhaps that quantitative measures of ST-segment deviation produce Iess variability between interpreters than categorical measures. The second conclusion that can be drawn from this study is that computer measurements are consistently and significantly higher than h u m a n measurements. The most likely explanation is that computers can analyze ST-segment deviation to the nearest 10-~tV level, whereas h u m a n s are incapable of such precise resolution. The two reviewers in this study measured ST-segment deviation to the nearest 50 ~tV, and it is our contention that h u m a n s are more likely to "round down" w h e n measuring ST-segment deviation, erring more conservatively. The computer is capable of more precise measurements and, thus, higher total ST scores, since microvolt values are summed across all 12 ECG leads. For example, the lowest ST score in our sample demonstrated that the computer ST score was 200 ~tV higher than each of the reviewers ST scores, appearing to be clinically significant. However, each was within at

Discussion

Fig. 2. Highest ST score of all 58 patients during left anterior descending percutaneous transluminal coronary angioplasty.

III -5 0 0

aVL

I

-aVR

0 0 0

-5 0 0

-8 0 0

II -10 0 0

aVF

III

Vl

V2

V3

V4

V5

V6

-8 0 0

-5 0 0

55 S0 50

80 50 50

180 150 150

135 100 100

95 50 50

-20 0 0

ST Score 601(,uV) 400@V) 400~V)

Measurement of ST-Segment Deviation n=58 5000

-

--

1

Pelter et al.

81

r = 0.96

I

6000

I

400o

n=58

r=0.96

•

I

©

5000

/ /

~" ~ 4000

~

3000

8 ~ 2000

2000

Iooo~ .....

0

,~-.

.... .

0

~.

1000

.......

2000

,

I

t

i -

3000

~ -

ooo

4000

i

-

5000

ST SCORES(#V)FORREVIEWER #1

0 0

1000

2000

3000

4000

5000

6000

ST SCORES (~tV) FOR COMPUTER

Fig. 3. Comparison of ST scores b e t w e e n the two reviewers.

Fig. 5. Comparison of ST scores b et w een the computer and reviewer 2 using a line of equality, n = 58.

least 45 btV of the computer measurements in all 12 leads, and in several of the leads, the difference was fewer than 10 btV. These microvolt differences, w h e n assessing each lead individually, do not represent clinically significant discrepancies. Therefore, the 200-I,tV difference b et ween the computer ST score and the reviewers' score is accounted for w h e n absolute microvolt values are s u m m e d across all 12 ECG leads. However, our data also demonstrated that differences in ST scores b et wee n the computer and the reviewers can also be attributed to individual ECG leads and not as a result of minimal microvolt distribution across all 12 leads. For example, the ]2-lead ECG exhibiting the highest ST score in our sample showed that 4 of the 12 ECG leads, specifically leads V2 through Vs, accounted for the large discrepancy in ST scores between reviewer 1 (ST score = 3,800 btV) and reviewer 2 (ST score = 3,900 btV) compared to the computer (ST score = 5,303 btV). In the other eight leads, both reviewers measured to within 100 BV of the computer measurements. As such, the differences in the ST scores could be attributed to precordial leads Vx-V5 and were not distributed evenly across all 12 ECG leads. However, the total ST scores have been used by numerous researchers in assessing the value of continuous ST-segment monitoring, confirming that this is an important variable to assess.

During both minimal ST-segment deviation as well as maximal ST-segment deviation, computer ST scores were higher in every case compared to the two reviewers. This suggests that the computer may detect ST-segment changes that are too subtle to be observed by the h u m a n eye. This sensitivity becomes significant w h e n only small ST-segment changes occur during ischemia. In our study and in numerous other research studies assessing the value of continuous ST-segment monitoring, ischemia is conservatively defined as an ST-segment change of 100 btV or more in any one of the 12 ECG leads, so that the detection of minimal ST-segment change is important. However, it should be mentioned that this study was conducted in a very controlled environment with patients supine and often sedated, so that the ECGs are high quality. In most cases, our ECG tracings are free of wandering baseline, 60-cycle interference, and patient body position changes, which can significantly alter the interpretation of ST-segment deviation. Therefore, the idea that computer measurements may detect ischemia missed by humans may not be applicable to patients outside of this controlled environment. Our data demonstrate that w h e n comparing m e a n ST scores between the computer and h u m a n reviewers during both minimal and maximal ST-segment deviation, computer mean ST scores are always higher. This suggests that the two m easu r em en t methods, manual versus computerized, can produce significantly different m e a n ST deviation scores.

6000

~/ i

5000 >

1

4"

!

4000

3000

/,

/

i

J

~,,

iI,

Conclusion and Implications

I •

~2000

I 0

1000

2000

i 3000

4000

5000

6000

ST SCORES (~V)FOR COMPUTER

Fig. 4. Comparison of ST scores between the computer and reviewer 1 using a line of equality, n = 58.

The following conclusions can be drawn: (1) a high degree of interrater reliability is possible with manual measurements of ST-segment deviation, and (2) computer measurements are consistently greater than manual measurements. Computers may detect ischemia that is undetected by humans, and computer and manual measurements of ST-segment deviation should not be mixed w h e n used as a variable for research.

82

Journal of Electrocardiology Vol. 29 Supplement Table 1. Comparison of M e a n ST Scores Reviewer I: 1,306 p.V Reviewer 2:1,350 {tV Difference: 44 ~LV T-test: NS

Computer: 1,674 ftV Reviewer 1:1,306 IxV Difference: 368 ~V P< .0I

References 1. Krucoff M: Identification of high-risk patients with silent myocardial ischemia after percutaneous transluminal coronary angioplasty by multilead monitoring. Am J Cardiol 61:29E 1988 2. Drew BJ, Adams MG, Pelter MM, Wung SF: ST segment monitoring with a derived 12-lead electrocardiogram is superior to routine CCU monitoring. Am J Crit Care 5:198, i996 3. Krucoff MW, Wagner BA, Pope JE et ah The portable programmable microprocessor-driven real-time 12lead electrocardiographic monitor: a preliminary report of a n e w device for the noninvasive detection of successful reperfusion or silent coronary reocclusion. Am J Cardiol 65:143, i990 4. Kwon K, Benedict E Wilcox I e t ah The unstable ST segment early after thrombolysis for acute infarction and its usefulness as a marker of recurrent coronary occlusion. Am J Cardiol 67:109, 1991 5. Doevendans PA, Gorgels AP, van der Zee R et ah Electrocardiographic diagnosis of reperfusion during thrombolytic therapy in acute myocardial infarction. Am J Cardiol 75:1206, 1995 6. Segall HN: The electrocardiogram and its interpretation: a study of reports by 20 physicians on a set

7.

8.

9.

10.

11.

12.

! 3.

Computer: 1,674 [xV Reviewer 2:1,350 ~tV Difference: 324 ~tV P< .01

of 100 electrocardiograms. Can Med Assoc J 82:2, 1960 Blackburn H, Blomqvist G, Freiman A et ah The exercise electrocardiogram: differences in interpretation. Report of a technical group on exercise electrocardiography. Am J Cardiol 21:871, 1968 Caralis DG, Wiens G, Shaw L, Younis LT et al: An off-line digital system for reproducible interpretation of the exercise ECG. J Electrocardiol 23:285, 1990 Krucoff MW, Pope JE, Bottner RK et al: Computerassisted ST segment monitoring: Experience during and after brief coronary occlusion. J Electrocardiol 20:15, 1987 Wolthuis RA, Fischer JR, Hopkirk A et ah New criteria for computer interpretation of exercise electrocardiograms in a largely asymptomatic population. Int J Cardiol 2:203, 1982 Sievanen H, Karhumaki L, Vuori I, Malmivou J: Improved diagnosis performance of the exercise ECG test by computerized mulitvariate ST-segment/heart rate analysis. J Electrocardiol 24:129, 1991 Dower GE, Yakush A, Nazzal SB et al: Deriving the 12-lead electrocardiogram from four (EASI) electrodes. J Electrocardiol 7(suppl): 182, 1988 Bland JM, Altman DG: Statistical methods for assessing agreement between two methods of clinical measurement. Lancet I:307, 1986