glasgow 12-lead ecg analysis program - physio-control · important information version history...

For use with the

LIFEPAK® 15 MONITOR/DEFIBRILLATOR

STATEMENT OF VALIDATION AND ACCURACY

Glasgow 12-lead ECG Analysis Program

Important Information

Version HistoryGlasgow 12-lead ECG Analysis Program Version 27 is used in LIFEPAK 15 software version 3207410-006 and later versions.

!USA RxOnly

LIFEPAK is a registered trademark of Physio-Control, Inc. Specifications are subject to change without notice.

© 2009 Physio-Control, Inc. All rights reserved.

Publication Date: 3/2009 GDR 3302436_A

© 2009 Physio-Control, Inc. Statement of Validation and Accuracy for the Glasgow 12-Lead ECG Analysis Program iii

CONTENTS

1 Origin and DevelopmentA Brief History of the Development of the Glasgow Program............................................. 1-1References ......................................................................................................................... 1-3

2 Intended UseDiagnostic Application ........................................................................................................2-1Intended Population............................................................................................................ 2-1Intended Location ............................................................................................................... 2-1Diagnostic Accuracy ........................................................................................................... 2-2Diagnostic Approach .......................................................................................................... 2-2

3 Measurement AccuracyAccuracy of the Measurement Algorithm............................................................................ 3-1

4 Accuracy of Diagnostic StatementsECG Classification..............................................................................................................4-1

Type A ................................................................................................................ 4-1Type B ................................................................................................................ 4-1Type C................................................................................................................ 4-1

Definition of Statistics ......................................................................................................... 4-1Type A Statements ............................................................................................................. 4-2

CSE Database.................................................................................................... 4-4Acute Myocardial Infarction ................................................................................................ 4-6

Tucson AMI Database........................................................................................ 4-7ST Elevation Myocardial Infarction ..................................................................................... 4-7

Tucson STEMI Database ................................................................................... 4-7Type B Statements ............................................................................................................. 4-8

Glasgow 1000 ECG Database ........................................................................... 4-8Glasgow Adult Normal Database .......................................................................4-9Database of Additional Cases of Atrial Fibrillation ...........................................4-10Pacemaker ECG Database ..............................................................................4-12

Type C Statements in Adults ............................................................................................4-12Type A Statements in Children.........................................................................................4-13

Glasgow Pediatric ECG Database ...................................................................4-13Type B Statements in Children.........................................................................................4-14Type C Statements in Children.........................................................................................4-14

© 2009 Physio-Control, Inc. Statement of Validation and Accuracy for the Glasgow 12-Lead ECG Analysis Program 1-1

1Origin and Development

A Brief History of the Development of the Glasgow Program

The Glasgow 12-lead ECG Analysis Program, available from Physio-Control, is the product of decades of research and continuous improvement by Professor Peter W. Macfarlane, D.Sc, FESC, and colleagues. The continuous improvements have been made by Dr. Macfarlane’s research team at the University of Glasgow.

Investigations into the use of computers to analyze electrocardiograms began in the 1960s.1,2

Dr. Macfarlane began his electrocardiology research at the University of Glasgow in Scotland in 1964, leveraging these studies. His thesis research sought to determine whether computers could be used for routine interpretation of ECGs in hospitals, an innovation that promised time savings and improvements in patient care.3, 4, 5, 6

Over the decades that followed, Glasgow researchers continued to incrementally advance the science of computerized electrocardiology.7, 8, 9, 10, 11, 12, 13

Dr. Macfarlane and his team addressed a serious shortcoming of early computerized 12-lead ECG interpretation, the inability to interpret pediatric ECGs, by developing a 12-lead pediatric ECG interpretation program.14 In 1996 this program was approved by the FDA for use in the United States, while the adult software had been approved several years previously.

Dr. Macfarlane and colleagues have continuously advanced the science of computerized electrocardiology:

• Overall accuracy of atrial fibrillation reporting improved as a result of studies of neural networks, though conventional criteria proved more effective.15, 16

• Marked improvements in the repeatability of ECG analysis came from advances in the use of continuous scoring functions.14, 17, 18, 19

• Extensive studies were made into the normal limits of ECG in varied adult and pediatric populations, resulting in extensive contributions to the 3-volume reference “Comprehensive Electrocardiology,” authored by Macfarlane and Lawrie in 1989.20

• The Glasgow program pioneered the use of age and gender to improve diagnostic accuracy for myocardial infarction and other cardiac abnormalities.21

A study published in the New England Journal of Medicine establishes that the Glasgow ECG analysis program is among the best in class compared to other ECG analysis programs.22 Versions of the Glasgow ECG analysis program have been adopted commercially initially by Siemens Elema, based in Stockholm, Sweden (now Draeger Medical, Andover, Massachusetts, USA), by Burdick of Deerfield, Wisconsin, USA (now owned by Cardiac Science Corporation of Seattle, Washington, USA), and by Spacelabs Healthcare, of Issaquah, Washington, USA.

1-2 Statement of Validation and Accuracy for the Glasgow 12-Lead ECG Analysis Program

Improvements to the Glasgow ECG Analysis Program since 2000 include the following:

• The sensitivity for detecting atrial flutter was significantly enhanced while maintaining high specificity.23

• A further report of the effect of age and gender on ECG measurements was published, showing, for example, that normal limits of ST amplitude in V2 were twice as high in males as in females.24

• In collaboration with Physio-Control, the Glasgow program has been updated to introduce a unique message for acute ST elevation myocardial infarction (STEMI) with age and gender accounted for in the criteria.25, 26, 27 The STEMI statement is, ***MEETS ST ELEVATION MI CRITERIA***. Testing was done using a database of prehospital ECGs from patients with chest pain.

• Sgarbossa's criteria were added to the program to allow detection of acute myocardial infarction in some patients who also have left bundle branch block.28 The criteria improve sensitivity for STEMI while maintaining high specificity.29

• An updated description of the Glasgow ECG analysis program has been published.30

Dr. Macfarlane worked collaboratively with other electrocardiology researchers:

• Dr. Macfarlane was a member of the steering committee of a project entitled Common Standards for Quantitative Electrocardiography (the CSE project) that ran from 1978 until 1990. Many collaborative publications were produced as a result of this work, and several databases were established. The CSE group published a paper on recommendations for measurement standards in quantitative electrocardiography.31 The CSE group gathered a set of ECGs which are available as a reference database against which the accuracy of measurement of ECG waveforms by ECG programs can be tested.32

• As a member of an ad hoc subcommittee of the American Heart Association Committee on Electrocardiography, Dr. Macfarlane assisted in publishing recommendations for signal processing of electrocardiograms.33

• Dr. Macfarlane’s laboratory contributed a large percentage of the recordings in the CSE group’s diagnostic database, used to test the accuracy of ECG programs.22, 34

• Dr. Macfarlane was part of a group of electrocardiography experts who pointed out areas where ECG standards could be improved for diagnosis of myocardial ischemia/infarction in patients with acute coronary syndromes.35

• Dr. Macfarlane also contributed to the AHA 2005 Guidelines related to automated ECG analysis.36

• Dr. Macfarlane’s work on the influence of age and gender on normal limits of the ECG has influenced the international ECG criteria for STEMI in the 2007 definition of myocardial infarction. 37 For example, criteria for lead V1 are no longer grouped with those for V2 and V3.

1


References

The references that follow are for this chapter only.

1. Pipberger HV, Arms RJ, Stallmann FW. Automatic screening of normal and abnormal electrocardiograms by means of digital electronic computer. Proc Soc Exp Biol Med. 1961;106:130-132.

2. Caceres CA, Steinberg CA, Abraham S, Carbery WJ, McBride JM, Tolles WE, Rikli AE. Computer extraction of electrocardiographic parameters. Circulation. 1962;25:356-362.

3. Macfarlane PW. A modified axial lead system for orthogonal lead electrocardiography. Cardiovasc Res. 1969;3:510-515.

4. Macfarlane PW. ECG waveform identification by digital computer. Cardiovasc Res. 1971;5:141-146.

5. Macfarlane PW, Lorimer AR, Lawrie TD. Normal ranges of modified axial lead system electrocardiogram parameters. Br Heart J. 1971;33:258-265.

6. Macfarlane PW, Lorimer AR, Lawrie TD. 3 and 12 lead electrocardiogram interpretation by computer. A comparison on 1093 patients. Br Heart J. 1971;33:266-274.

7. Macfarlane PW, Cawood HT, Taylor TP, Lawrie TD. Routine automated electrocardiogram interpretation. Biomed Eng. 1972;7:176-180.

8. Macfarlane PW, Watts MP, Peden J, Lennox G, Lawrie TD. Computer-assisted ECG interpretation. Br J Clin Equip. 1976;1:61-70.

9. Taylor TP, Macfarlane PW, Lawrie TD. Arrhythmia interpretation by digital computer. In: Schubert E, ed. Neue Ergebnisse der Electrocardiologie II. Berlin: Humboldt University; 1974:243-245.

10.Macfarlane PW, Peden A, Podolski M, Lawrie TD. A new 12 lead ECG diagnostic computer program. In: Ueda H, et al, eds. Recent Advances in Electrocardiology. Jpn Heart J. 1982;23(suppl.1):667-670.

11.Macfarlane PW, Podolski M, Watts MP, Shoat D, Macfarlane DK, Lawrie TD. The new Glasgow system. In: Willems JL, van Bemmel JH, Zyweitz C, eds. Computer ECG Analysis: Towards Standardization. Amsterdam: North Holland; 1986:31-36.

12.Macfarlane PW. Computer interpretation of cardiac rhythm. In: Willems JL, van Bemmel JH, Zyweitz C, eds. Computer ECG Analysis: Towards Standardization. Amsterdam: North Holland; 1986:279-284.

13.Macfarlane PW, Devine B, Latif S, McLaughlin S, Shoat DB, Watts MP. Methodology of ECG interpretation in the Glasgow program. Meth Inf Med. 1990;29:354-361.

14.Macfarlane PW, Coleman EN, Devine B, Houston A, McLaughlin S, Aitchison TC, Pomphery EO. A new 12-lead pediatric ECG interpretation program. J Electrocardiol. 1990;23(suppl):76-81.

15.Yang TF, Devine B, Macfarlane PW. Artificial neural networks for the diagnosis of atrial fibrillation. Med Biol Eng Computing. 1994;32:615-619.


16.Yang TF, Devine B, Macfarlane PW. Deterministic logic versus software-based artificial neural networks in the diagnosis of atrial fibrillation. J. Electrocardiol. 1993;26(suppl):90-94.

17.McLaughlin SC, Aitchison TC, Macfarlane PW. Improved repeatability of 12-lead ECG analysis using continuous scoring techniques. J Electrocardiol. 1993;26(suppl):101-107.

18.McLaughlin SC, Aitchison TC, Macfarlane PW. Methods for improving the repeatability of automated ECG analysis. Methods Inf Med. 1995;34:272-282.

19.McLaughlin SC, Aitchison TC, Macfarlane PW. The value of the coefficient of variation in assessing repeat variation in ECG measurements. Eur Heart J. 1998;19:342-351.

20.Macfarlane PW, Lawrie TD, eds. Comprehensive Electrocardiology. New York: Pergamon Press; 1989.

21.Macfarlane PW, McLaughlin SC, Devine BD, Yang TF. Effects of age, sex and race on ECG interval measurements. J Electrocardiol. 1994;27(suppl):14-19.

22.Willems JL, Abreu-Lima C, Arnaud P, et al, incl. Macfarlane PW. The diagnostic performance of computer programs for the interpretation of electrocardiograms. New Engl J Med. 1991;325:1767-1173.

23.Morrison S, Macfarlane PW. Computer detection of atrial flutter. Annals of Noninvasive Electrocardiology. 2000;5:358-364.

24.Macfarlane PW. Age and sex related normal limits of ST amplitude. J Electrocardiol. 2001;34(suppl):235-241.

25.Macfarlane PW, Browne D, Devine B, Clark E, Miller E, Seyal J, Hampton D. Modification of ACC/ESC criteria for acute myocardial infarction. J Electrocardiol. 2004;37(suppl):98-103.

26.Macfarlane PW, Browne D, Devine B, Clark E, Miller E, Seyal J, Hampton D. Effect of age and gender on diagnostic accuracy of ECG diagnosis of acute myocardial infarction. In: A.Murray, ed. Computers in Cardiology. 2004;31:165-168.

27.Macfarlane PW, Hampton DR, Clark E, Devine B, Jayne CP. Computer and cardiologist diagnosis of ST-elevation myocardial infarction. J Electrocardiol. 2007;40(suppl 1):S32-S33.

28.Sgarbossa EB, Pinski SL, Barbagelata A, Underwood DA, Gates KB, Topal EJ, Califf RM, Wagner GS. Electrocardiographic diagnosis of evolving acute myocardial infarction in the presence of left bundle-branch block. N Engl J Med. 1996;334:481-487.

29.Tabas JA, Rodriguez RM, Seligman HK, Goldschlager NF. Electrocardiographic criteria for detecting acute myocardial infarction in patients with left bundle branch block: a meta-analysis. Ann Emerg Med. 2008;52:329-336.

30.Macfarlane PW, Devine B, Clark E. The University of Glasgow (Uni-G) ECG alalysis program. Computers in Cardiology. 2005;32:451-454.

31.The CSE Working Party (incl. Macfarlane PW). Recommendations for measurement standards in quantitative electrocardiography. Eur Heart J. 1985;6:815-825.

1


32.Bailey JJ, Berson AS, Garson A, Horan LG, Macfarlane PW, Mortara D, Zywietz C. Recommendations for standardization and specifications in automated electrocardiography: bandwidth and digital signal processing. Circulation. 1990;81:730-739.

33.Willems JL, Arnaud P, van Bemmel JH, et al, incl. Macfarlane PW. A reference database for multilead electrocardiographic computer measurement programs. J Am Coll Cardiol. 1987; 10:1313-1321.

34.Willems JL, Arnaud P, van Bemmel JH, Degani R, Macfarlane PW, Zywietz C, for the CSE Working Party. Common standards for quantitative electrocardiography: goals and main results. Methods Inf Med. 1990;29:263-271.

35.Wagner G, Lim T, Gettes L, Gorgels A, Josephson M, Wellens H, Anderson S, Childers R, Clemmensen P, Kligfield P, Macfarlane P, Pahlm O, Selvester R. Consideration of pitfalls in and omissions from the current ECG standards for diagnosis of myocardial ischemia/infarction in patients who have acute coronary syndromes. Cardiol Clin. 2006;24:331-342.

36.Kligfield P, et al, incl. Macfarlane PW. Guidelines for the standardization and interpretation of the electrocardiogram, part 1: the electrocardiogram and its technology. Circulation. 2007;115:1306-1324.

37.Thygesen K, Alpert JS, White HD; Joint ESC/ACCF/AHA/WHF Task Force for the Redefinition of Myocardial Infarction. Universal definition of myocardial infarction. Circulation. 2007;116:2634-2653. Also in: J Am Col Cardiol. 2007;50:2173-2195. Also in: Eur Heart J. 2007;28:2525-2538.


2Intended Use

The Glasgow 12-lead ECG analysis program is intended to be used as follows.

Diagnostic Application

The Glasgow program is intended to provide an interpretation of the resting 12-lead ECG in all patient care situations, whether this be in a hospital or primary care setting. It is capable of diagnosing all commonly recognized ECG abnormalities such as myocardial infarction (MI), including acute MI, ventricular hypertrophy, ST-T abnormalities and common abnormalities of rhythm. Conduction defects and other abnormalities such as prolonged QT interval are also reported.

The software has been widely used in clinical trials, e.g. the West of Scotland Coronary Prevention Study,1 and hence has had wide exposure to recording of electrocardiograms in all commonly required situations.

Intended Population

The Glasgow program is intended for use in adults and children of any age from birth upwards. The program makes significant use of the patient's age and gender and indeed operates at the level of days in the case of neonates.2, 3 It is believed to be the only program that is based on normal limits derived using the algorithm itself, with this applying to criteria for subjects of all ages, including neonates. It is known that other 12-lead ECG programs utilize the Glasgow normal limits.

Intended Location

The Glasgow program is intended to be used in hospital or in a general physician’s office, or in locations outside of the hospital, such as an ambulance or patient's home. The program is able to accept the patient's age and gender, and it automatically invokes the appropriate criteria and routines, such as special logic for acute cardiac ischemia, when necessary.

1 Shepherd J, Cobbe SM, Ford I, et al, incl. Macfarlane PW. Prevention of coronary heart disease with Pravastatin in men with hypercholesterolemia. New Engl J Med. 1995;333:1301-1307.

2 Macfarlane PW, Coleman EN, Pomphrey EO, McLaughlin S, Houston A, Aitchison TC. Normal limits of the high-fidelity pediatric ECG. Preliminary observations. J Electrocardiol. 1989;22(suppl):162-168.

3 Macfarlane PW, Budgett S, Devine B, Aitchison TC. Paediatric ECG analysis - the Glasgow approach. In: Liebman J, ed. Electrocardiology 96. 1997:451-460.


Diagnostic Accuracy

The program is designed to be as accurate as possible, with the emphasis towards a high specificity, given that the criteria are based on the normal limits already described.

Nonetheless, the program has high sensitivity for detecting all cardiac abnormalities, as is evidenced by the results presented in Section 4, Accuracy of Diagnostic Statements.

The program aims for the highest sensitivity at a high specificity, although there is always a trade off between one and the other.

Diagnostic Approach

The Glasgow program first derives a comprehensive set of measurements of the 12-lead ECG, then performs an interpretation based on the measurements and other inputs. For further information see the Physician’s Guide to the Glasgow 12-lead ECG Analysis Program, available from your Physio-Control representative or by calling 1-800-442-1142.


3

Measurement Accuracy

Accuracy of the Measurement Algorithm

The software measurement algorithm has been evaluated against the sixteen IEC calibration ECGs.1 The results show that the software meets all of the tolerance limits set out by the IEC standard. In addition,100 CSE measurement ECGs have been processed as required for the test using the Glasgow program, and the results meet all of the tolerance limits specified by the IEC standard. Disclosed changes of measurements caused by noise on ECGs are presented in the following table.

1 IEC 60601-2-51:2003-02, Medical electrical equipment — Part 2-51: Particular requirements for safety, including essential performance, of recording and analysing single channel and multichannel electrocardiographs.

Table 3-1 Changes of measurements cause by noise on ECGs

Global Measurement Type of Added NoiseDisclosed Differences

Mean (ms) Standard Deviation (ms)

P Duration High frequency 0.600 2.503

P Duration Line frequency (50Hz) -0.400 1.265

P Duration Line frequency (60Hz) 0.000 2.494

P Duration Base-line -0.400 1.265

QRS Duration High frequency 0.600 3.534

QRS Duration Line frequency (50Hz) 0.400 2.951

QRS Duration Line frequency (60Hz) -0.400 2.797

QRS Duration Base-line 0.200 3.048

QT Interval High frequency -2.200 3.048

QT Interval Line frequency (50Hz) -0.200 0.632

QT Interval Line frequency (60Hz) -0.200 1.476

QT Interval Base-line -0.800 1.398


4Accuracy of Diagnostic Statements

ECG Classification

A task force of the American College of Cardiology established a classification system for ECG abnormalities on the basis of the types of statements that could be made. These are as follows:

Type A

An ECG abnormality which can be confirmed by non electrocardiographic means, e.g. ventricular hypertrophy that can be confirmed by echocardiography, or recent myocardial infarction confirmed by a rise in biomarkers.

Type B

An ECG abnormality basically detected by the ECG itself, e.g arrhythmias or conduction abnormalities such as bundle branch block.

Type C

An ECG abnormality that is essentially descriptive, e.g. axis deviation, moderate ST elevation, etc.

Definition of Statistics

TRUE POSITIVE (TP) A correct report of an abnormality being present

TRUE NEGATIVE (TN) A correct report of an abnormality being absent

FALSE POSITIVE (FP) An incorrect report of an abnormality being present

FALSE NEGATIVE (FN) An incorrect report of an abnormality being absent

SENSITIVITY (SENS) TP / (TP + FN )

SPECIFICITY (SPEC) TN / (TN + FP )

POSITIVE PREDICTIVE VALUE (PPV) TP / (TP + FP )

NEGATIVE PREDICTIVE VALUE (NPV) TN / (TN + FN )

PREVALENCE (PREV)

TOTAL ACCURACY

Number of occurrences of an abnormality

Total number of cases in the database

Total number of cases in the database

Total number of cases correctly classified


Type A Statements

CSE Abbreviations

The following abbreviations are used in Table 4-1, Table 4-2, and Table 4-3.

LVH = Left ventricular hypertrophy

RVH = Right ventricular hypertrophy

BVH = Left and right ventricular hypertrophy

AMI = Anterior myocardial infarction

IMI = Inferior myocardial infarction

MIX = Anterior and inferior myocardial infarction

VH+MI = Ventricular hypertrophy and myocardial infarction

OTHER = Cardiologist defined abnormality excluding above definitions

The Table 4-1 results are derived from an analysis of the CSE database in March 2007. The gold standard ("truth") was derived from the clinical data.

Total accuracy: 73.7%, partially correct: 75.7%, (both on 1220 cases)

Table abbreviations are defined in the previous section, "CSE Abbreviations".

Table 4-1 Type A Statementsa

a The CSE database does not allow a meaningful interpretation of statistics on statements involving "possible" and "probable" qualifiers. They are taken into account in determining the sensitivity etc of the various diagnoses as the statement with the highest likelihood, where definite > probable > possible, is given most weight in handling a specific interpretation

SENS (%) SPEC (%) PPV (%) NPV (%) PREV

D

IAG

NO

ST

IC C

AT

EG

OR

Y

NORMAL 97 77 b

b Specificity and positive predictive value for 'NORMAL' should be interpreted carefully. A report of 'NORMAL' in a case of 'MYOCARDIAL INFARCTION' or 'hypertrophy' contributes to decreased specificity for 'NORMAL' (even though the ECG may appear 'NORMAL'). In the CSE study, an ECG report stating only 'MYOCARDIAL ISCHEMIA' was mapped to 'NORMAL' even if the true answer was 'INFARCTION', thereby also contributing to decreased specificity for 'NORMAL'

66 b 98 382/1220

LVH 57 98 82 93 183/1220

RVH 44 100 92 97 55/1220

BVH 47 99 78 98 53/1220

AMI 72 98 88 96 170/1220

IMI 71 99 93 92 273/1220

MIX 63 98 69 98 73/1220

4


The Table 4-2 results are derived from an analysis of the CSE database in March 2007. The gold standard ("truth") was derived from the clinical data. In this table, there is a more detailed breakdown of the reports, e.g. 1.0% of ECGs from individuals regarded as normal were reported by the program as LVH. On the other hand, 29.5% of ECGs from patients with clinical evidence of LVH were reported as normal.

Total accuracy: 73.7 %, partially correct: 75.7 %, (both on 1220 cases)

Table abbreviations are defined in "CSE Abbreviations" on page 4-2

Table 4-2 Type A Statements - Program Versus Gold Standard from Clinical Data

GOLD STANDARD (TRUTH) FROM CLINICAL DATA

NORMAL (%) LVH (%) RVH (%) BVH (%) AMI (%) IMI (%) MIX (%) VH+MI (%)

PR

OG

RA

M

NORMAL (%) 97.1 29.5 40.0 13.2 15.9 25.6 8.2 25.8

LVH (%) 1.0 56.8 7.3 0.0 4.4 1.3 5.5 0.0

RVH (%) 0.0 0.5 43.6 0.0 0.0 0.4 0.0 0.0

BVH (%) 0.0 2.2 3.6 46.7 0.6 0.0 0.0 0.0

AMI (%) 1.6 3.6 0.0 7.5 71.8 0.0 0.0 0.0

IMI (%) 0.3 6.0 1.8 1.9 0.3 70.5 0.0 0.0

MIX (%) 0.0 0.8 1.8 0.0 7.1 2.2 63.4 0.0

VH+MI (%) 0.0 0.0 0.0 0.0 0.0 0.0 0.0 46.8

OTHER (%) 0.0 0.5 1.8 30.7 0.0 0.0 22.9 27.4

TOTAL (%) 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0

PREV 382/1220 183/1220 55/1220 53/1220 170/1220 273/1220 73/1220 31/1220


The Table 4-3 shows distributions of the CSE March 2007 computer interpretations with respect to the consensus opinion of the 8 cardiologists. Prevalence totals change compared to Table 4-1 and Table 4-2 because the gold standard has changed.

Total agreement: 81.48%

Table abbreviations are defined in "CSE Abbreviations" on page 4-2

CSE Database

Table 4-1 through Table 4-3 of this section provide the results of analyzing the 1220 ECGs in the Common Standards for Quantitative Electrocardiography (CSE) database using the Glasgow Program. A brief explanation of the study and the outputs follows.

The CSE database1 was constructed by acquiring ECGs from 1220 individuals (831 men, 389 women, mean age 52 ± 13 years). The ECGs were acquired in five different European centers using a variety of equipment, but signals were sampled at 500 samples/sec and all leads were recorded simultaneously.

Individual centers in the study processed the ECGs in their own local laboratory and submitted the interpretations, mapped to an agreed scheme (e.g. LVH was 21A), to a central lab in Leuven, Belgium where data on sensitivity and other statistics were calculated. The true classification of the cases was known only to the core lab, and in practical terms this meant that the classifications were effectively stored inside software used to determine the accuracy of

Table 4-3 Type A Statements - Program Versus Gold Standard from Cardiologists

REFERENCE (FROM CARDIOLOGISTS CONSENSUS)

NORMAL (%) LVH (%) RVH (%) BVH (%) AMI (%) IMI (%) MIX (%) VH+MI (%) OTHER (%) TOTAL (%)

PR

OG

RA

M

NORMAL (%) 92.9 18.0 28.3 3.4 9.4 14.0 11.8 20.5 6.3 46.3

LVH (%) 2.2 72.0 0.0 10.3 2.5 1.1 0.7 4.5 3.1 10.4

RVH (%) 0.7 0.0 66.7 5.2 0.0 0.4 0.0 0.0 0.0 2.1

BVH (%) 0.5 2.8 0.0 70.7 1.3 0.2 0.0 0.0 7.8 2.6

AMI (%) 1.0 0.3 1.7 3.4 80.2 0.0 2.8 0.0 6.3 11.4

IMI (%) 1.2 4.8 0.0 0.0 0.0 82.9 4.9 4.5 0.0 17.0

MIX (%) 0.1 0.7 3.3 0.0 6.6 1.3 71.5 0.0 3.1 5.6

VH+MI (%) 0.0 0.0 0.0 0.0 0.0 0.0 0.0 61.4 3.1 1.2

OTHER (%) 1.4 1.4 0.0 6.9 0.0 0.0 8.3 9.1 70.3 3.4

TOTAL (%) 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0

PREV 503/1220 145/1220 30/1220 29/1220 159/1220 228/1220 72/1220 22/1220 32/1220 1220

1 Willems JL, Abreu-Lima C, Arnaud P, et al, incl. Macfarlane PW. The diagnostic performance of computer programs for the interpretation of electrocardiograms. New Engl J Med. 1991;325:1767-1773.

4


individual programs. This is still the case today, but following the untimely death of Professor Jos Willems, who directed the lab in Leuven, the responsibility for maintaining the secrecy of the classifications and for providing further assessments of accuracy of software has transferred to the lab of Professor Paul Rubel, based in Lyon, France.

The composition of the CSE database includes 286 individuals who were apparently healthy and 96 patients referred for cardiological investigation but found to have no cardiac abnormality, who together make up a group of 382 controls. The remaining 838 subjects had known clinical conditions, such as myocardial infarction or valvular heart disease.

Patients are classified as having ventricular hypertrophy, myocardial infarction or "no structural abnormality" on the basis of clinical information. This could have included echocardiographic data, cardiac enzyme data, and in some cases, a knowledge of intra-cardiac pressures determined at cardiac catheterization.

Three cardiologists from different European countries reviewed the clinical data and agreed on the classification. Table 4-1 and Table 4-2 are based on this information. Thus, a sensitivity of 56.8% for left ventricular hypertrophy (LVH) is with respect to a clinical classification that is expected to accompany such an abnormality.

It is also important to understand that ST-T abnormalities in isolation were mapped to NORMAL (or more strictly, NO STRUCTURAL ABNORMALITY). Thus, if a patient had an inferior myocardial infarction, and a program reported ST-T abnormalities suggestive of myocardial ischemia, the corresponding ECG would be regarded as false negative and placed in the normal/inferior MI box, i.e. in the computer report of normal column in the row entry for inferior MI, where the percentage is 25.6%.

Some patients had multiple abnormalities such as left ventricular hypertrophy and inferior myocardial infarction. A complicated scoring system allowed for such combinations to be considered and some of the outputs therefore state that there was "additional mixing" in the CSE test center. In general terms, this mixing gave credit for both abnormalities in such a patient being reported by a program. Thus, the use of mixing enhances the accuracy of total results.

Separately from this form of classification, which was not always acceptable to members of the CSE Working Group, was another classification produced by a set of 8 cardiologists. In turn, the accuracy of the cardiologists was assessed against the clinical data, but their interpretations were combined to produce a so-called "cardiologist interpretation" or "referee consensus" with respect to which programs were also evaluated. There was not much detail presented on this aspect of the CSE study in the original paper1 in 1991, although outcomes were lodged separately with the publisher.

As might be expected, a completely different set of results (Table 4-3) is obtained when the cardiologist is used as the gold standard. Consider the following example by way of explanation.


A patient may well have LVH by echo but a normal ECG. With respect to the clinical database, a program reporting a normal ECG in this patient would be regarded as providing a false negative result. On the other hand, the cardiologists’ combined opinion in this case would also be a normal ECG, similar to the computer. In this case, the program would be regarded as providing the correct interpretation. Thus, the same ECG may be correct with respect to one gold standard, but incorrect with respect to another.

In general terms, it can be seen that the program has a much higher agreement with cardiologists than with the clinical data. Part of the answer lies in the previous paragraph, e.g. 56.8% correct diagnoses of LVH versus the clinical data and 72.0% versus the cardiologists. Note also that in the group of 382 controls, the Glasgow program agreed in 98.2% of cases with the cardiologists.

In conclusion, it is pleasing to note that the total accuracy remains very high at 73.7%. This is particularly satisfying given the large number of changes which continue to be made to the program compared to the 1991 version, when the total accuracy was 69.7%. Accuracy with respect to the cardiologists continues to remain very high at 81.48%.

Repeatability of interpretation is not assessed by the CSE test set, nor is rhythm analysis or conduction defects.

Acute Myocardial Infarction

The Table 4-4 results are from the Tucson AMI Database, a set of prehospital ECGs from patients seen by emergency medical services for chest pain. The results are based on 248 patients with a clinical discharge diagnosis of acute myocardial infarction while the remainder of the 1220 patients all had chest pain - see Tucson AMI Database in the following section for further details.

AMI = Acute myocardial infarction

For prehospital use, it is important to test the 12-lead ECG program with prehospital ECGs from chest pain patients, because in AMI the ECG can substantially evolve by the time the patient arrives at the hospital. In an AMI patient, the prehospital ECG is more likely than the emergency room ECG to show repolarization abnormalities such as T-wave inversion or hyperacute T waves, and less likely to show Q waves.

It is important to understand that from the 12-lead ECG alone (including patient age and gender), 12-lead ECG programs (including the Glasgow program) and cardiologists tend to be good at detecting acute ST elevation myocardial infarction (STEMI), but insensitive at detecting AMI without ST elevation (NSTEMI).

Patients with NSTEMI are usually identified through enzyme tests that identify markers released into the bloodstream by cardiac cells involved in AMI. The sensitivity for AMI of 51.6% in Table 4-4 is explained by the fact that the Glasgow program detects a high percentage of STEMI ECGs and

Table 4-4 Acute Myocardial Infarction


AMI 51.6 97.6 84.8 88.9 248/1220

4


a low percentage of NSTEMI ECGs, just as cardiologists do when given only the ECG and patient age and gender. Studies typically find that about half of AMI patients have STEMI and the other half NSTEMI.

Tucson AMI Database

The EGG recordings in this database are from a study of prehospital chest pain patients conducted in Tucson, Arizona between 1989 and 1992. The study was undertaken to determine whether prehospital recording of 12-lead ECGs resulted in earlier identification and faster treatment of acute myocardial infarction (AMI) patients when they reached the hospital. Persons over 18 years of age were eligible if 911 had been called to complain of acute, non-traumatic chest pain. Each patient enrolled had a standard 12-lead ECG recorded at the scene by paramedics before transport to the hospital. The diagnostic classification of each patient who was enrolled was established from the hospital discharge diagnosis, in which laboratory measures of cardiac enzyme levels and/or ECG evolution were used to identify the source of the chest pain as AMI, angina, or non-ischemic. Out of 1220 patients, 248 patients (20.3%) had a discharge dignosis of AMI. This database was used to test accuracy for detection of AMI in patients with chest pain.

ST Elevation Myocardial Infarction

To assess the Glasgow program's sensitivity for detecting STEMI, the patients from the Tucson AMI database with a discharge diagnosis of AMI were classified as STEMI or NSTEMI in the Tucson STEMI Database (see below for further details). Of the 217 AMI patients in the database, 113 (52%) were classified by the cardiologists as meeting the criteria for STEMI.

Based on the use of the Tucson STEMI Database, the sensitivity of the Glasgow program for detecting STEMI was 89.4%.

Tucson STEMI Database

The Tucson STEMI database is a subset of the above Tucson AMI database. Initially included were all patients with a discharge diagnosis of AMI. Two cardiologists independently overread the database as to whether or not the ECG met the ACC/ESC definition of STEMI. The cardiologists attempted to reach a consensus on cases where they differed, but they could not reach agreement on all cases. Cases were excluded if STEMI criteria could not be measured (e.g., due to left bundle branch block) or if the cardiologists disagreed as to whether or not the ACC/ESC criteria for STEMI were met. The final database included 113 cases for which the cardiologists agreed on STEMI and 104 cases that were non-STEMI. This database was used to test sensitivity for detection of STEMI in patients with known AMI.


Type B Statements

The Table 4-5 results for type B statements were obtained from the Glasgow 1000 ECG database (explained in the following section, “Glasgow 1000 ECG Database”), from which 73 children's ECGs were removed in order to provide results on adults only. In addition, 31 WPW examples were obtained from a group of 31 patients being investigated by electrophysiological testing. There were therefore 927+31 (958) ECGs available for assessment of type B statements in adults.

RBBB = Right bundle branch block

LBBB = Left bundle branch block

LAFB = Left anterior fascicular block

IVCD = Intraventricular conduction defect

WPW = Wolf Parkinson White

Glasgow 1000 ECG Database

Within the Glasgow lab, a number of databases have been assembled. The 1000 ECG database was selected to provide a wide range of normal and abnormal ECGs, including arrhythmias, conduction defects, and other abnormalities that could be used to test the Glasgow program. It was constructed mainly from ECGs recorded from hospitalized patients or individuals visiting outpatient clinics. The ECGs are not clinically classified. The database is often used as a test set to ensure that a copy of the program compiled in a different environment will produce the same results as in its native development system. There are 506 males (mean age 62 ± 22 years) and 494 females (mean age 68 ± 19 years) in the database with an age range of 5 days to 96 years. There are 73 subjects with an age of 16 years or under. The racial distribution of this database is unknown, but given that it is used for assessing arrhythmias and Type B statements, this is of no relevance.

Table 4-5 Type B Statements in Adults - Conduction Defects

SENS (%) SPEC (%) PPV (%) NPV (%) PREV (%)

CO

ND

UC

TIO

N D

EF

EC

T

RBBB 96.0 99.9 96.0 99.9 24/958

LBBB 10.00 100.0 100.0 100.0 16/958

RBBB with LAFB 100.0 100.0 100.0 100.0 9/958

Incomplete RBBB/ rSr’ V1 100.0 99.9 95.6 100.0 22/958

IVCD 100.0 99.7 82.4 100.0 14/958

WPW 88.0 100.0 100.0 99.6 33/958

(Possible) LAFB 100.0 100.0 100.0 10.00 15/958

4


The Table 4-6 shows the results of rhythm interpretations from several databases. The first database is the Glasgow 1000 ECG Database, explained in the previous section. A second database of 1498 ECGs from apparently healthy adults (explained in the following section, “Glasgow Adult Normal Database”) is also incorporated to increase the number of sinus and other common rhythms. These ECGs are supplemented by 72 cases of atrial fibrillation (explained in the following section, “Database of Additional Cases of Atrial Fibrillation”) which are included to augment the number of cases of this arrhythmia.

Glasgow Adult Normal Database

The normal ECG database is composed of ECGs recorded from 1498 apparently healthy individuals who were each examined by a physician and who had no evidence of heart disease or any other condition such as diabetes which might be expected to lead to cardiovascular

Table 4-6 Type B Statements - Rhythm


DO

MIN

AN

T R

HY

TH

M S

TA

TE

ME

NT

S

Sinus rhythm 99.43 98.54 99.31 98.78 1746/2570

Sinus bradycardia 100.00 99.69 97.83 100.00 315/2570

Atrial fibrillation 93.66 99.84 97.08 99.63 142/2570

Sinus arrhythmia 93.14 100.00 100.00 99.72 102/2570

Sinus tachycardia 100.00 99.80 94.25 100.00 82/2570

Sinus bradycardia with sinus arrhythmia 84.44 100.00 100.00 99.72 45/2570

Atrial flutter 92.31 100.00 100.00 99.88 39/2570

Possible atrial flutter 100.00 99.96 97.37 100.00 37/2570

Possible ectopic atrial rhythm 96.15 99.92 92.59 99.96 26/2570

Possible ectopic atrial bradycardia 100.00 100.00 100.00 100.00 15/2570

A-V dissociation 50.00 100.00 100.00 99.92 4/2570

Probable atrial fibrillation 100.00 99.92 60.00 100.00 3/2570

Probable accelerated junctional rhythm 100.00 99.96 75.00 100.00 3/2570

Probable supraventricular tachycardia 100.00 100.00 100.00 100.00 3/2570

Probable sinus tachycardia 100.00 99.96 66.67 100.00 2/2570

Sinus tachycardia with sinus arrhythmia 100.00 100.00 100.00 100.00 1/2570

Irregular ectopic atrial bradycardia 100.00 100.00 100.00 100.00 1/2570

Probable atrial tachycardia 100.00 100.00 100.00 100.00 1/2570

Marked sinus bradycardia 100.00 100.00 100.00 100.00 1/2570

Possible junctional rhythm 100.00 99.96 50.00 100.00 1/2570

Regular supraventricular rhythm 100.00 99.88 25.00 100.00 1/2570


abnormalities. This database has been used extensively in the determination of normal limits of ECGs such as those relating to the QT interval.2 It contains ECGs from 863 males and 638 females with an age range of 18 to 78 years. This cohort was recruited from local government workers in Glasgow plus students from the University and was essentially 100% Caucasian.

Database of Additional Cases of Atrial Fibrillation

In order to supplement the number of cases of atrial fibrillation, an additional 72 cases were added to the database of 1000 ECGs from which rhythm analysis was assessed. There were 48 males (mean age 66.8 ± 15.5 years) and 24 females (mean age 74.7 ± 8.3 years).

2 Macfarlane PW, McLaughlin SC, Rodger JC. Influence of lead selection and population on automated measurement of QT dispersion. Circulation. 1998;98:2160-2167.

4


The Table 4-7 results are for supplementary statements using the same databases as the previous table.

PVCs = Premature ventricular complexes

PACs = Premature atrial complexes

A-V = Atrioventricular

SA = Sinoatrial

Table 4-7 Supplementary Statements


SU

PP

LE

ME

NT

AR

Y S

TA

TE

ME

NT

S

~ with rapid ventricular response 100.00 100.00 100.00 100.00 71/2570

~with PVCs 98.39 99.84 93.85 99.96 62/2570

~ with PACs 96.08 99.09 68.06 99.92 51/2570

~ with borderline 1st degree A-V block 97.87 99.76 88.46 99.96 47/2570

~ with 1st degree A-V block 94.59 99.92 94.59 99.92 37/2570

~ or aberrant ventricular conduction 100.00 99.96 95.00 100.00 19/2570

~ with slow ventricular response 100.00 99.96 92.86 100.00 13/2570

~ with 2:1 A-V block 80.00 100.00 100.00 99.92 10/2570

~with frequent multifocal PVCs 100.00 100.00 100.00 100.00 7/2570

~ with uncontrolled ventricular response 100.00 100.00 100.00 100.00 7/2570

~ with 4:1 A-V block 100.00 100.00 100.00 100.00 5/2570

~ with aberrantly conducted supraventricular complexes

60.00 99.84 42.86 99.92 5/2570

~ with frequent PVCs 100.00 99.92 60.00 100.00 3/2570

~ with frequent PACs 100.00 99.96 75.00 100.00 3/2570

~ with multifocal PVCs 100.00 100.00 100.00 100.00 2/2570

~ with undetermined irregularity 100.00 100.00 100.00 100.00 2/2570

~ with paroxysmal idioventricular rhythm 100.00 100.00 100.00 100.00 1/2570

~ with 2nd degree A-V block, Mobiltz l (Wenckeback)

0.00 99.96 0.00 99.96 1/2570

~ with 3:1 A-V block 100.00 100.00 100.00 100.00 1/2570

~ with complete A-V block 0.00 100.00 0.00 99.96 1/2570

~ with bigeminal PVCs 100.00 100.00 100.00 100.00 1/2570

~ with 2nd degree (Mobitz II) SA block 0.00 99.96 0.00 100.00 0/2570


The Table 4-8 results are for pacing statements derived from 47 cases of paced ECGs, explained in the following section, “Pacemaker ECG Database”.

Pacemaker ECG Database

The accuracy of statements relating to artificial implanted pacemaker rhythm is entirely dependent on the detection of the pacemaker stimuli by a separate algorithm. The stimulus locations are then passed to the Glasgow program along with the ECG. Thus, for this database, 47 ECGs were selected where the pacemaker stimuli were seen to be correctly detected from inspection of relevant indicators on the ECG printout. Age, gender and racial distribution are of no relevance.

Type C Statements in Adults

The Table 4-9 results were derived from the Glasgow 1000 ECG database, (explained previously in the section, “Glasgow 1000 ECG Database”) with the 73 childrens’ ECGs removed.

LAD = Left axis deviation

RAD = Right axis deviation

Table 4-8 Pacing Statements


PA

CIN

G S

TA

TE

ME

NT

S

Atrial pacing 100.0 97.3 90.9 100.0 10/47

Demand atrial pacing 100.0 100.0 100.0 100.0 5/47

Ventricular pacing 90.9 100.0 100.0 97.3 11/47

A-V sequential pacemaker 90.9 100.0 100.0 97.3 11/47

Demand pacing 100.0 97.3 90.9 100.0 10/47

Table 4-9 Type C Statements in Adults


E

CG

FIN

DIN

GS

LAD 100.0 100.0 100.0 100.0 65/927

Leftward axis 100.0 100.0 100.0 100.0 59/927

RAD 100.0 100.0 100.0 100.0 15/927

Severe RAD 100.0 100.0 100.0 100.0 1/927

Rightward axis 100.0 100.0 100.0 100.0 12/927

Non specific ST ±T changes 100.0 99.8 97.3 100.0 71/927

rSr' - probable normal variant 100.0 99.9 94.4 100.0 17/927

Poor R wave progression 92.8 99.9 96.3 99.8 26/927

4


Type A Statements in Children

The Table 4-10 results were taken from the Glasgow Pediatric ECG Database, explained in the following section, “Glasgow Pediatric ECG Database”.

RVH = Right ventricular hypertrophy

LVH = Left ventricular hypertrophy

BVH = Biventricular hypertrophy

Glasgow Pediatric ECG Database

This database consists of 840 ECGs recorded from neonates, infants and children referred or admitted to hospital for investigation of various problems. There are 436 males (mean age 5.6 ± 5.1 years) and 401 females (mean age 5.6 ± 5.2 years) with a combined age range of 1 day to 18 years. The subject's gender was not recorded in three cases. Race was also not recorded but the population can be assumed to be 100% Caucasian, including children whose parents have immigrated into Scotland from South Asia. The gold standard is the overeader's opinion. Results for RVH and LVH used the combined interpretation of two pediatric cardiologists who were provided with clinical information on a subset of 664 children whose ECGs were being reviewed. The remaining ECGs were reported without knowledge of the clinical history. Results using this database have now been published.3, 4

Table 4-10 Type A Statements in Children


E

CG

FIN

DIN

GS

Normal 97.5 97.9 97.8 97.7 405/840

RVH 53.5 94.4 53.5 94.4 71/664

LVH 44.4 95.8 28.9 97.4 27/664

BVH 100.0 99.1 73.9 100.0 17/840

Possible right atrialabnormality

100.0 99.6 96.7 100.0 89/840

Possible left atrial abnormality 100.0 99.5 73.3 100.0 11/840

Possible biatrialabnormality

100.0 100.0 100.0 100.0 6/840

Abnormal ventricularconduction pathways (Q waves)

92.8 99.9 96.3 99.8 26/840

Borderline high QRS voltage - probable normal variant

100.0 100.0 100.0 100.0 24/840

3 Hamilton RM, Houston AB, McLeod K, Macfarlane PW. Evaluation of pediatric diagnosis of ventricular hypertrophy by computer program compared with cardiologists. Pediatric Cardiology. 2005;26:373-378.

4 Hamilton RM, McLeod K, Houston AB, Macfarlane PW. Inter- and intra-observer variability in LVH and RVH reporting in pediatric ECGs. Annals of Noninvasive Electrocardiology. 2005;10:330-333.


Type B Statements in Children

The Table 4-11 results are based on the type B abnormalities in the Glasgow Pediatric ECG Database, explained in the previous section.

RBBB = Right bundle branch block

LBBB = Left bundle branch block

IVCD = Intra ventricular conduction defect

WPW = Wolf Parkinson White

Type C Statements in Children

The Table 4-12 results are based on 840 ECGs in the Glasgow pediatric ECG database, explained in the previous section, “Glasgow Pediatric ECG Database”.

Table 4-11 Type B Statements in Children


E

CG

FIN

DIN

GS

RBBB 92.5% 99.8% 94.9% 99.6% 40/840

LBBB 100% 100% 100% 100% 1/840

IVCD 100% 99.5% 73.3% 100% 11/840

WPW pattern 50% 100% 50% 99.8% 4/840

Incomplete RBBB 100% 100% 100% 100% 6/840

rSr' (V1) - probable normal variant 100% 100% 100% 100% 14/840

Table 4-12 Type C Abnormalities in Children


EC

G F

IND

ING

S

ST±T changes are non specific 100.0 99.7 98.1 100.0 107/840

ST elevation 100.0 99.8 97.2 100.0 35/840

Right axis deviation 100.0 100.0 100.0 100.0 43/840

Severe right axis deviation 100.0 100.0 100.0 100.0 8/840

QRS axis leftward for age 100.0 99.7 97.8 100.0 46/840

Left axis deviation 100.0 100.0 100.0 100.0 25/840

Indeterminate axis 100.0 100.0 100.0 100.0 3/840

Physio-Control, Inc.11811 Willows Road NERedmond, WA 98052 USATelephone: 425.867.4000Toll Free (USA only): 800.442.1142Fax: 425.867.4121www.physio-control.com

Medtronic B.V.Earl Bakkenstraat 106422 PJ HeerlenThe Netherlands

Publication date:3/2009 GDR 3302436_A

glasgow 12-lead ecg analysis program - physio-control · important information version history...

Documents