Application of computerized exercise ECG digitization

Application of Computerized ExerciseECG Digitization Interpretation in Large Clinical Trials

Dennis G. Caralis, MD, MPH, Leslee Shaw, MA, Brian Bilgere, BS, Liwa Younis, MD, PhD, Karen Stocke, BS, Robert D. Wiens, MD, and Bernard R. Chaitman, MD

Abstract: The authors report on a semiautomated

program that incorporates both visual identification of fiducial points and digital determination of the STsegment at 60 ms and 80 ms from the J point, ST slope, changes in R wave, and baseline drift. The off-line program can enhance the accuracy of detecting electrocardiographic (ECG) changes, as well as reproducibility of the exercise and postexercise ECG, as a marker of myocardial ischemia. The analysis program is written in Microsoft QuickBASIC 2.0 for an IBM personal computer interfaced to a Summagraphics mm1201 microgrid II digitizer. The program consists of the following components: (1) alphanumeric data entry, (2) ECG wave form digitization, (2) calculation of test results, (4) physician overread, and (5) editor function for remeasurements. This computerized exercise ECG digitization-interpretation program is accurate and reproducible for the quantitative assessment of ST changes and requires minimal time allotment for physician overread. The program is suitable for analysis and interpretation of large volumes of exercise tests in multicenter clinical trials and is currently utilized in the TIM1 II, TIM1 III, and BAR1 studies sponsored by the National Institutes of Health. Key words: exercise testing, exercise electrocardiography, computer ECG analysis.

Exercise electrocardiography (ECG) is used not only to diagnose coronary artery disease and its ex-

tent, but also to predict subsequent

cardiac events in patients with ischemic heart disease. I Exercise electrocardiography has also been used to determine prognosis in large multicenter clinical trials of asymptomatic subjects such as in the Myocardial Risk Factor Intervention Trial (MRFIT)2 and in symptomatic populations such as in the Coronary Artery Surgery Study (CASS) .3 A nearly fourfold increase in 7-year coronary mortality was reported among men with an abnormal ECG response to exercise compared to those with a normal ST-segment response during exercise testing in MRFIT.2 The ST-

From the Department of Internal Medicine, Division of Cardiology, St. Louis University School of Medicine, St. Louis, Missouri. Presented in part at the 39th Annual Scientific Session of the American College of Cardiology, March 18, 1990, New Orleans, Louisiana. Supported in part by NIH ROI HL42145-03. Bethesda, Maryland and by a grant from the David B. Lichtenstein Foundation, St. Louis, Missouri. Reprint requests: Dennis G. Caralis, MD, MPH, St. Louis University Hospital, Division of Cardiology, 3635 Vista at Grand, P.O. Box 15250, St. Louis, MO 63110-0250.

101

102

Journal of Electrocardiology

Vol. 25 No. 2 April 1992

segment is the most diagnostically relevant measure of the ECG taken during exercise. Conventional visual analysis of ECG changes at rest or during exercise has increased variability, even among experienced observers, with respect to determination of absolute test results, as well as the magnitude of ST-segment deviation.4-6 Computer-assisted interpretation of the exercise ECG has been advocated to improve the accuracy of diagnosing coronary artery disease, providing sensitivity and specificity in the range of TO84% in largely asymptomatic populations.’ Ideally, the same computer analysis program should be used by all participating centers in large clinical trials in order to analyze the ECG records during the preexercise period, during the last 30 seconds of peak exercise, and during recovery. The same interactive computer graphics terminal should be used to verify the quality of all exercise ECG data, whereby an operator could easily check computer measurements and reject leads with excessive residual noise. Computerized on-line analysis of ST-segment change by means of the same commercially available systems in all participating centers has a major disadvantage, that is, cost. It should also be recognized that the algorithms among the different manufacturers are not uniform, and equipment varies from medical center to medical center. Furthermore, computerized identification of QRS onset and offset in the presence of baseline wander and motion artifact is problematic.‘-12 Evaluation of exercise test data during the course of a large clinical trial requires a system for efficient processing of data and storage for subsequent statistical analyses. Use of a semiautomated system that incorporates both visual identification of fiducial points and digital determination of the ST-segment at 60 and 80 ms from the J point may enhance the accuracy of detecting ECG changes as well as the reproducibility of the test as a marker for myocardial ischemia. We have designed a computer-assisted digitization program for the evaluation of rest and exercise ECG changes that has been, and is currently, utilized in the TIM1 II, BARI, and TIM1 III multicenter clinical trials for analysis and interpretation of over 10,000 exercise tests.

Paper ECG Tracings From Site

Fig. 1. Data flow. CEL = Core ECG lab facility; DCC = data coordinating center.

lated to each clinical unit. The exercise ECGs are set to print unprocessed, unfiltered ECG wave forms. In the rare patient where noisy tracings occur in spite of adequate skin preparation, a high-pass filter (40 Hz) is accepted. Prior to the start of the clinica trial, exercise ECG personnel at individual clinical units are required to send five consecutive exercise tests to the core laboratory facility for certification. If the tracings do not pass inspection, additional tracings are required until certification standards are met. The clinical units are instructed to send the entire exercise ECG when possible, or, alternatively, the 12-lead ECGs obtained at standing rest, peak exercise, immediate recovery, and 3-minute recovery or the last postexercise ECG that shows return of the ECG to baseline (Fig. 1). The original paper ECG tracings (photocopies are discouraged) are mailed from the site and digitized at the core ECG facility by experienced coders. Selected digitized data is transmitted to the Data Coordinating Center using Bitnet or a diskette. Four IBM or compatible work stations and digitizing boards are used to measure and store data. The data are subsequently downloaded to a MicroVAX II or DEC system 5100. The network is illustrated in Figure 2.

A

Station#2

MicmVax II

Materials and Methods In each clinical trial, an appropriate exercise protocol is established, lead placement is determined, and a manual of operations describing the technique of skin preparation, settings for the exercise ECG, calibration of exercise treadmill speed, etc. is circu-

Fig. 2.

Computer network at core laboratory facility. A = SummaGraphics MM1201 Microgrid II digitizer; B = IBM or compatible computer.

Computerized Exercise ECG

l

Caralis

et al.

103

A random 2% sample of 12-lead exercise ECGs is circulated for recoding at selected intervals. The samples are enriched with 2 5 % of tests showing ischemic change. Reproducibility for the maximum number of abnormal leads is assessed using Kappa statistics. The Kappa statistics range from 0.74 to 0.9 1 for several recent 25 exercise ECG sets.

Program Description

The program determines exercise-induced changes in R wave, J point, ST-segment at 60 and 80 ms, ST-segment slope, and baseline drift and incorporates physician and nonphysician interpretation with final approval designated by the physician staff. A summary of exercise measurements and abnormal response is provided. The analysis program is written in Microsoft QuickBASIC 2.0 for an IBM personal computer interfaced to a Summagraphics MM 120 1 micrdgrid II digitizer and 16-button bitpad cursor with an illuminated 3-power magnification lens and precision cross hair cursor (Summagraphics Corporation, Fairfield, CT) (Fig. 3). The digitizing tablet has a resolution of 1,016 dots per inch and a repeatability of -t 0.01 inch. Fifty repeat measures of 1, 5, and 10 mm produced digitized measurements of 1 5 0.1, 5 k 0.1, and 10 t 0.15 mm (? SD), respectively, using one of the digitizing boards in our laboratory. The program consists of the following components: (1) alphanumeric data entry, (2) ECG wave form digitization (Fig. 4), (3) calculation of test results, (4) physician overread, and (5) editor function for remeasurements. Routine 12-lead or rhythm strip

Fig. 3. Digitizing board, bum In bitpad cursor,

16-

and IBM -compatible computer.

form, magnified as visualized through the cursor. The cross hair is located at the J point.

Fig. 4. ECG wave

ECG tracings may be used with this program. A monitor is used to aid the operator in determining measurements through the use of a 1: I grid displayed on the screen. Patient *me and test date entry results in a database inquiry in order to prevent entering of du-

104

Journal of Electrocardiology

Vol. 25 No. 2 April 1992

plicate test material and provide editing options for data remeasurement when required. Double entry of all screen-entered information is required to reduce data entry error. Additional demographic and test information includes: ( 1) rest and peak exercise heart rate, (2) usage of digitalis, (3) notation of original copy vs photocopy, and (4) ST-segment recovery time. Information on ventricular arrhythmia is documented on screen 2. Ventricular arrhythmias are classified as premature ventricular beats, ventricular bigeminy, ventricular couplets, nonsustained ventricular tachycardia, sustained ventricular tachycardia. and ventricular fibrillation. Ventricular arrhythmias are recorded at rest, during exercise, and in the recovery phase. The third screen reports ECG data that affect exercise test interpretation, including left ventricular hypertrophy, left bundle branch block at rest or during exercise, supraventricular tachycardia, ventricular preexcitation, ventricular pacemaker, excessive motion artifact, or missing ECG data. A right bundle branch block option permits analysis of leads Vs and Vg. If no leads have an ST-segment shift of 2 kO.05 mV, the data are saved as a no change option. Following entry of measurement exclusion options, the exercise ECG is considered noninterpretable, a data output form is generated, the package is assigned for physician overread, and the program recycles for the next patient. The rest and peak exercise or immediate recovery ECGs are selected for measurement. All leads with 2 -+ 0.05 mV of ST-segment change are selected for digitizing. Leads with < 20.05 mV of ST-segment change are entered as exhibiting no change. The operator chooses three consecutive complexes with a stable baseline for measurement. Excessive motion artifact or baseline drift is noted. All leads with 2 kO.05 mV of ST-segment deviation are selected for measurement. Following entry of a right bundle branch block or lead measuring option, the fourth screen initiates the measurement program. The leads from the rest ECG are measured first, followed by leads from the peak exercise or recovery ECG that exhibits maximal ST-segment change. Four bitpad buttons control menu options and allow for exiting the program and lead remeasurement. The remaining 12 buttons correspond to each of the 12 ECG leads. Any lead chosen for analysis may be measured by depressing one of the 12 ECG lead buttons on the bitpad. The program prompts the coder to mark the ECG complex at the following fiducial points: the PQ junction, the peak of the R wave, and the J point. After indicating the J point, the ST-segment is traced and measurements are determined automatically at

60 and 80 ms. An electronic beep indicates when 80 ms after the J point has been reached. Three consecutive complexes with a stable baseline are chosen. Baseline drift is determined from the difference, in millivolts, between consecutive PQ junctions. Leads with 20.3 mV of artifact are discarded. All measurements are made from the top of the line tracing. The following changes from the rest tracing are required to designate an abnormal lead: 1. Downsloping ST-segment depression: J point depression 2 0.1 mV, ST-segment depression at 80 ms 10. I mV, and a negative ST slope > 0.1 mV/s. 2. Horizontal ST-segment depression: J point depression 20.1 mV, ST-segment depression at 80 ms 20.1 mV, and ST slope & 0.1 mV/s. 3. Upsloping ST-segment depression: J point depression 20.16 mV, ST-segment depression at 80 ms ~0.15 mV, and ST slope > + 0.1 mV/s. 4. ST-segment elevation in a non-Q wave, noninfarct lead: J point and ST-80 change 2 + 0.1 mV. 5. ST-segment elevation in a Q wave, infarct lead: J point and ST-80 change 2 + 0.1 mV. Following measurement of each requested lead, all rest and peak, J point, ST-segment, ST slope, and R wave height values, as well as the calculated net difference, are printed. A summary of abnormal lead information is printed, based upon the net difference in ST-segment changes. The number of leads with abnormal ST-segment depression or elevation in the anterior, inferior, or lateral lead groups is included in this section, as well as the maximal depth or height of change and sum of ST-segment change.

Results Output Page #I The first output page is illustrated in live sections (Fig. 5). ST-segment values can be measured 60 and 80 ms after the J point. In this example, all ST values are 80 ms after the J point. The initial five lines of data indicate alphanumeric patient identification, study date, whether the ECG wave forms were raw data or average beats from a commercially available exercise ECG, proper calibration scale, and presence or absence of digitalis usage, etc. The measurements are presented in tabular form for each lead, with three sections indicating resting values, exercise values, and the delta difference between rest and exercise for J point, ST-80 measurements, slope, and R wave amplitude. Baseline drift measured between

Computerized

consecutive PQ junctions indicates a relatively stable baseline in the rest and exercise tracings. RXindicates measurements made during the recovery phase of exercise. Slope is determined from the J point to ST60 or ST-80, depending on the set point for digitized measurements. The last column indicating slope compares the rest to exercise slope. The slope* measurement 5 indicates that the ST-segment slope was horizontal at rest as well as during exercise. The exclusion code of 1 for leads I, Vr , VZ, and Vj indicates no significant change in the four leads, and they are not measured. Ventricular arrhythmia data indicates that the operator detected no ventricular arrhythmias during the exercise test. The rest exercise ECG (Fig. 6) and exercise ECG (Fig. 7) measurements are illustrated on output page # 1 (Fig. 5), and the difference in absolute measurements is shown in the column labeled difference. For example, in lead V6 the J point was depressed 1.O mm and the ST 80 was depressed 0.7 mm. When the J point or ST-80 .measurement is elevated at rest, the isoelectric point for the rest tracing is automatically set at 0 for comparison with the exercise ECG measurements. The exercise slope in lead Vg was 0.4 mV per second. Thus, lead V6 does not meet criteria for an ischemic response because the ST-80 measurement was only 0.7 mm. Similarly, although lead Vs shows important J point depression, the ST-segment slope is positive, resulting in an ST-SO measurement cl.0 mm, which does not meet the criteria for an ischemic response as described in the methodology.

Output Page #2 The output on page 2 (Fig. 8) indicates abnormal lead information. The output for this patient indicates that two leads met criteria for abnormality; there were no resting ST-segment abnormalities; there were no Q waves at rest in the recovery phase; the maximum depth of ST-segment depression for individual leads was at the ST-80 point; no ST-segment elevation was observed; and the abnormal lead met criteria 2, which indicates the presence of horizontal ST-segment depression. The program permits measurement of bipolar leads as illustrated. The lower half of output page 2 contains summary data on abnormalities, with respect to region of interest, and sums up the extent of ST-segment depression in the individual territories, as well as total information. Thus, in this patient with two abnormal leads, the maximum depth of ST-segment depression in any lead was 0.15 mV, and the sum of ST-segment depression for both abnormal leads was 0.27 mV

Exercise

ECG

l

Caralis

et al.

105

(Fig. 9). Similar information can be provided for STsegment elevation, which was not observed in this patient. A third output page, which is not illustrated, indicates the presence of persisting ECG abnormalities in the recovery phase, which lead group they are in, and maximum depth of ST-segment depression.

Discussion Improved technological advances in the field of digital ECG signal processing have aided in the reduction of noise and improved precision of the ECG signal and data compression.r3-I8 Storage of exercise measurements has allowed for various computer algorithms to be written and validated, based upon sensitive and specific cutpoints for enhanced detection of coronary artery disease extent and severity, and has aided in the prediction of future cardiac events. Although major advances have been made in the area of computerized exercise ECG equipment, largescale on-line analysis of exercise tests from numerous medical centers across the nation for clinical trials is difficult because of the significant cost to replace or upgrade existing equipment, incompatibility between certain exercise ECG carts and treadmills, and individual physician preferences. For these reasons, we investigated the utility of an off-line exercise ECG processing system.” The system requires high-quality exercise ECG acquisition at participating clinical units, ascertained through strict quality control requirements set by the central ECG laboratory. The new system we report permits precision measurements of exercise-induced ST-segment changes, allowing accurate data collection for large-scale clinical trials without requiring capital equipment expenses at individual clinical centers. The cost can be considerable and, in a recent large acute infarct study (TIM1 II), there were 50 hospitals at which patients underwent exercise testing. On-line analysis programs have utilized various mathematical constructs such as cubic spline and filtering systems to eliminate baseline wander and motion artifact. ” Within our off-line system,” three successive ECG complexes were used in order to minimize the effect of baseline noise and excessive wander. Data recorded from leads exhibiting measured artifact exceeding 0.3 mV are discarded. Measurement values of J point and ST-80 are averaged for each lead to minimize the effect of outlying values. The suggested data sampling window during on-

106

Journal of Electrocardiology

BkRI Study

Test

Vol. 25 No. 2 April 1992

Patient ID: Date: 12/06/89 Interval: Baseline

Coding

Date:

Computer

02/11/91 ST

Rest

LEFlD II II

0.3 0.3

III-.

-0.2

aVRaVLaVFv4 vs V5 - R+ V6 -

-0.1 0.0 0.2 0.1 0.1 0.1 0.3

BaseLine I =n/a

Drift: aVR=O/O aVL=O/O aVF=O/l

=0/O

111=0/O II

-R=O/O

values

V5

R mm

JPt mm

0.5 0.5

14.7 14.7

0.0

0.3

10.6

-0.3 0.1 0.6 0.5 0.4 0.4 0.4

-0.2 0.0 0.5 0.5 0.3 0.3 0.1

0.0 1.0 12.6 8.7 10.8 10.8 10.6

-2.2 -1.0 -1.6 0.7 0.8 -1.7 -1.0 71.1 TO.7 -1.0

0.7 0.7

R/E VI V2 V3

=n/a =n/a =n/a

v4 v5 V6

J

!

-1.5 -0.6 -1.2 0.1 0.6 -0.8 -0.4 -0.7 -0.6 -0.7

0‘.9 0.5 0.5 -0.7 -0.2 1.2 0.7 0.5 0.1 0.4

Exclusion I =I

II = 0 III= 0 *Slope u-u=1 U-H=2 U-D=3

UNKN

0 Leads 4

point. 151 DIFFERENCE STChg R mm mm

R mm

JChg mm

15.0 16.9 13.7 0.0 0.0 15.5 12.8 14.1 12.0 11.1

-2.2 -1.0 -1.4 0.7 0.8 -1.7 -1.0 -1.1 -0.7 -1.0

Code: aVR= 0 aVL= 0 aVF= 0 Legend: H-U=4 H-H=5 H-D=6

VI v2 v3

= = =

1 1 1

-1.5 -0.6 -1.2 0.1 0.5 -0.8 -0.4 -0.7 -0.6 -0.7

v4 v5 V6

= = =

-0.3 -2.2 -3.1 0.0 -1.0 2.9 -4.1 -3.3 -1.2 -0.5

Slope* 5 5 5 5 5 4 5 5 5 5

0 0 0

D-U=7 D-H=8

D-D=9 lR=RECOVERY

2=RBBB S=WPW 8=FIRTIFACT Il=V TFICH/FIB

REST INDICFlTED

after

Heart Rate: EXERCISE ST80 Slope mV/sec mm

=0/o =0/o =0/O

Ventricular

‘

80ms

-R=O/O

EXCLUSION LEGEND: f=NO CH#NGES '4=LVH 7=EX RBBB lO=PoCEMAKER 13=CIVR EXCLUSION

OPERATOR

are Exer.

Heart Rate: 60 REST JPt ST80 Slope mm mm mV/sec

- R*

II

Digitalis: Photo/Orig: Standard

fiveraged: NO Min Recov: 3 Calib Scale: 1 .O

NO

3=LBBB 6=EX LBBB 9=MISSING lE=SVT

Arrhythmia

DFITCI

Information RECOVERY

EXERCISE

ARRHYTHMIAS

Fig. 5. Output

line analysis, as determined by Bhargava et a1.,13 is approximately lo-20 beats. Review of the standard ECG allows for review of 2.5 seconds of data per lead, approximately 5-10 ECG complexes. -In order to allow for optimal review of data, it may be recommended that a 5-second per lead format be utilized within this system. Optimal classification of normal and abnormal STsegment results depends upon accurate detection of primary fiducial points. Computerized detection of primary points on the QRS complex has proven to be problematic. 14,20,21Various detection algorithms that yield a wide variety of results under differing conditions have been written. The more recent ver-

page # 1.

sions of the Marquette CASE system provide a QRS template with tic marks indicating QRS onset and offset, which are automatically determined. If the physician or technician disagrees with the onset or offset, the tic marks can be moved through visual identification and the measurements recalculated. Similarly, our system allows for enhanced detection of the isoelectric and J points by means of visual identification by experienced ECG coders, along with computerized measurement of the change in J, ST60 or ST-80, and slope values. Following data storage of measurement values, customized software has been written to classify normal and abnormal ECG responses to exercise with

Computerized

-- -.

:

Exercise ECG

,f-----y

l

“Li-------

i

aVF

Fig. 6. Rest ECG tracing

corresponding

to output

page #I

Caralis et al.

(Fig. 5).

’

Fig. 7. Exercise

ECG tracing

corresponding

to output

page #l

(Fig. 5).

I

,

1

107

108

Journal of Electrocardiology

BAR1 Study

Patient ID: Date: 12/06/89 Interval: Baseline

Test Coding

Date:

02/l

Vol. 25 No. 2 April 1992 &vet-aged: NO Min Recov: 3 Calib Scale: 1.0

l/91

Computer Abnormal

Lead

Rest

Abn

QWave Recovery Max STDep Max STEle Criteria Criteria Criteria Criteria

Criteria Criteria Criteria Criteria

Criteria

II

III

NO NO NO -1.5

NO NO NO -1.2

2

2

1: J point >= l-CM5: J point 2: J point >= 2-CM5:J point 3: J point >= 3-CM5: J point 4: J point >= 5: J point >=

Lead

ST Abn. Leads : Depth STDep
lead):

ST No. Max Sum

Abn. Leads : Height STEleCany ST

Elevation:

UNKN

0 Leads 4

Information

-1 mm;ST slope > -1 mV/sec;ST80 >= -1 mm >= -1 mm;ST slope > -1 mV/sec;ST80 >= -1 .5 mm -1 mm; -1 mV/sec <= ST slope <= 1 mV/sec; ST80 >= a= -l.Smm;-lmV/sec <= ST slope <= lmV/sec;ST80 )= -1 .6 mm;ST slope > 1 mV/sec;ST80 >= -1 .5 mm >= -2.1 mm;ST slope > 1 mV/sec;ST80 a= -2.0 mm 1 mm in non Q-Wave lead;ST80 >= 1 mm/set 1 mm in Q-Wave lead;ST80 )= 1 mm/set Abnormal

No. Max Sum

Digitalis: Photo/Orig: Standard

lead):

Lead

-1 mm -1.5mm

Information

Segment Depression FINT
LAT
LFIT(1

,I_ ,V6) 0 0.0 0.0

,L ,V6) 0

TOT&L 2 -1.5 -2.7

TOTCIL 0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

Fig. 8. Output page #2.

acceptable diagnostic sensitivity and specificity. Similar to on-line analysis, our program allows for digital representation of key QRS wave points. Although our system has the capability to store more frequent data points from the ECG signal, only those fiducial points that are necessary for determination of STsegment shifts during exercise have been utilized. The slope measurements that we report use the time difference between the J point and ST-60 or ST80 measurement. In some computerized systems, the slope measurements are measured continuously over incremental time bits between the J point and ST60 measurement, with the greatest measurement indicating the slope. In other systems, a finite time interval after the J point compared to the ST-60 measurement indicates slope. In the current system, the J point measurement is determined by the operator and does not require assumptions implicit to on-line systems. We have accepted slope measurements 2 1 mV/s as being consistent with a horizontal ST-segment. Values greater or less than this range indicate

upsloping or downsloping ST-segments. In our experience, this range is often accepted as being consistent with a (horizontal) ST-segment. The precise number indicating a horizontal ST-segment slope is somewhat arbitrary, with some investigators considering a range to.7 mV/s as indicating a horizontal slope. The criteria for J point and ST-80 depression 20.1 mV is well established in the exercise test literature as providing an optimal balance of sensitivity and specificity for diagnostic and prognostic purposes. Decreasing the threshold would increase sensitivity with an unacceptable decrease in specificity. Increasing the depth of ST-segment depression before calling an ischemic response would decrease sensitivity. Further applications are currently being developed to allow for serial comparison of test results, submaximal analysis of data for plotting of change in ST/change in heart rate indices,2’-23 and enhanced graphic capabilities within the existing software. Validation of future developments within our large data

Computerized

Fig. 9. 3-minute obtained

recovery

tracing

from which

leads II and V5 measurements

set of patients from the TIMI and BARI trials of unstable angina and postmyocardial infarction or in patients with multivessel disease will allow for enhanced generalizability of findings and applicability to an increased percentage of patients from the large population of coronary patients. Digitization of ECGs by desktop optical scanners may provide a useful adjunct that could potentially decrease the amount of work required by coders who presently are required to identify QRS onset, offset, and key fiducial points.24 More research is required in this area. The computerized exercise ECG digitization-interpretation method described in this report is accurate and reproducible for quantitative assessment of ST-segment change, requiring minimal time allotment for physician overread.” In a previous report from our laboratory, exercise ECGs of 20 patients were analyzed using this program compared to measurements made by two cardiologists performing independent interpretations. The average amount of time required for the cardiologist to overread the computerized measurements was 2.7 -+ 1.5 minutes per ECG as compared to 20.7 +- 11 minutes for the cardiologist who did not have computer-

Exercise ECG

illustrated

in output

l

Caralis et al.

109

page #2 (Fig. 8) were

assisted measurements (p < 0.001). The proportion of variance for J point and ST-SO measurements was 0.93 and 0.90 when the same ECG was processed twice by the two cardiologists assisted by computerdetermined measurements, whereas the proportion of variance was 0.73 and 0.76 for the two cardiologists who did not have computer-assisted measurements. I9 Thus, this exercise ECG digitizationinterpretation method is particularly suitable for large clinical trials and high-volume exercise laboratories, providing reproducible measurements with a minimum physician time commitment.

References 1. Weiner

DA, Ryan TJ, McCabe CH et al: The role of exercise testing in identifying patients with improved survival after coronary artery bypass surgery. J Am Cob Cardiol 8:748, 1989 2. Rautahaju PM, Prineas RJ, Eifler WJ et al: Prognostic value of exercise electrocardiogram in men at high risk of future coronary heart disease: multiple risk factor intervention trial experience. J Am Cob Cardiol 8: 1, 1986

110

Journal of Electrocardiology

Vol. 25 No. 2 April 1992

3. Ryan TJ, Weiner DA, McCabe CH et al: Exercise testing in the coronary artery surgery study randomized preparation. Circulation 72: 3 1, 1985 4. Blackbum H: The exercise electrocardiogram: differences in interpretation. Am J Cardiol 2 1:87 1, 1968 5. Davies LG: Observer variation in reports on electrocardiograms. Br Heart J 20: 153, 1958 6. Segall H: The electrocardiogram and its interpretation: a study of reports by 20 physicians on a set of 100 electrocardiograms. Can Med Assoc J 82:2, 1960 7. Wolthuis RA, Fischer JR, Hopkirk A et al: New criteria for computer interpretation of exercise electrocardiograms in a largely asymptomatic population. lnt J Cardiol 2:203, 1982 8. Simoons ML, Boom HB, Smallenburg E: On-line processing of orthogonal exercise electrocardiograms. Comput Biomed Res 8:105, 1975 9. Meyer CR, Keiser HN: Electrocardiographic baseline noise estimation and removal using cubic splines and state-space computation techniques. Comput Biomed Res 10:459, 1977 10. McManus CD, Teppner U, Neubert D, Lobodzinski SM: Estimation and removal of baseline drift in the electrocardiogram. Comput Biomed Res 18: 1, 1985 11. Froning JN, Olson MD, Froelicher VF: Problems and limitations of ECG baseline estimation and removal using a cubic spline technique during exercise ECG testing: recommendations for proper implementation. J Electrocardiol 2 1: 149, 1988 12. Watanabe K, Bhargava V, Froelicher V: Computer analysis of the exercise ECG: a review. Prog Cardiovast Dis 23:423, 1980 13. Bhargava V, Goldberger A: Myocardial infarction diminishes both low and high frequency QRS potentials: power spectrum analysis of lead II. J Electrocardiol 14:57, 1981 14. Simoons ML, Hugenholtz PG, Ascoop CA et al: Quan-

15.

16.

17.

18.

19.

20.

21

22.

23.

24.

titation of exercise electrocardiography. Circulation 63:471, 1981 McHenry PL, Stowe De, Lancaster MC: Computer quantitation of the ST-segment response during maximal treadmill exercise: clinical correlation. Circulation 38:691, 1968 Lester FM, Sheffield LT, Reeves TJ: Factors affecting the maximal heart rate during extreme exercise. Ala J Med Sci 6:42, 1969 Wolf HK, MacInnis PJ, Stock S et al: Computer analysis of rest and exercise electrocardiograms. Comput Biomed Res 5:329, 1972 Winter DA, Rautaharju PM, Wolf HK: Measurement and characteristics of over-all noise content in exercise electrocardiograms. Am Heart J 74:324, 1967 Caralis DG, Wiens G, Shaw L et al: An off-line digital system for reproducible interpretation of the exercise ECG. J Electrocardiol 23~285, 1990 Simoon ML, Van Den Brand M, Hugenholtz PG: Quantitative analysis of exercise electrocardiograms and left ventricular angiocardiograms in patients with abnormal QRS complexes at rest. Circulation 55:55, 1977 Okin PM, Ameison 0, Kligfield P: Recovery-phase patterns of ST segment depression in the heart rate domain identification of coronary artery disease by the rate-recovery loop. Circulation 80:533, 1989 Okin PM, I