- Email: [email protected]
Are routine non-invasive tests useful in prediction of outcome after myocardial infarction in elderly people?
Are routine non-invasive tests useful in prediction of outcome after myocardial infarction in elderly people?
Summary
Introduction
Many patients with acute myocardial infarction undergo tests identify ischaemia, left-ventricular dysfunction, and arrhythmias. We examined the usefulness of these tests in clinical practice by comparing the ability of a cardiologist to
Many patients who experience myocardial infarction have a recurrence or die from cardiac causes within the following year. Factors contributing to the increased risk include the extent of myocardial damage, the presence of residual ischaemia, and the development of electrical instability. Non-invasive tests aimed at quantifying these factors are widely used by physicians to stratify patients into high and low risk categories.l,2 Several clinical variables are also associated with a poor outlook, including signs of leftventricular dysfunction, advanced age, and persisting angina after the infarction.1-6
to
predict outcome at 1 year after infarction with and without knowledge of the results of an exercise test, radionuclide angiogram, and 24 h Holter electrocardiographic (ECG) recording. The study was limited to patients older than 65 years, who have a greater risk of cardiovascular sequelae and undergo fewer interventional procedures. The patient’s own cardiologist predicted outcome on a standard rating scale, based on clinical findings and routine hospital tests. He then made a second prediction after seeing the non-invasive test results. Two other cardiologists not involved in the care of the patient independently made similar predictions. Success in predicting outcome was assessed by comparison of differences between the first and second predictions in the area under receiver operating characteristic curves. During 1 year’s follow-up there were 24 cardiovascular deaths and 3 recurrent myocardial infarctions among the 147 patients. There were no significant differences in mean curve the first and second predictions for the patients’ own cardiologist (0·62 [SE 0·06] vs 0 60 [0·06]) or the other cardiologists (0 63 [0·06] vs 0·64 [0·06] and 0·61 [0 06] vs 0·65 [0·06]). All predictions better than chance. were significantly (p < 0·05) Prediction of outcome in older patients after myocardial infarction is not improved by knowledge of the results of an exercise test, radionuclide angiogram, or 24 h Holter ECG areas between
recording. Lancet 1993; 342: 1069-72
Division of Cardiology, Sunnybrook Health Science Centre and Department of Medicine, University of Toronto, Toronto, Ontario, Canada (M G Myers FRCPC, R S Baigrie FRCPC, M L Charlat FRCPC, C D Morgan FRCPC)
Correspondence to: Dr Martin G Myers, Division of Cardiology, Sunnybrook Health Science Centre, A-222, 2075 Bayview Avenue, Toronto, Ontario M4N 3M5, Canada
Several studies have used mathematical models to compare the usefulness of predictions based on clinical findings with risk stratification based on data from noninvasive tests.4-1O However, these studies either have involved theoretical cases or have used a retrospective approach to an existing data base. Thus, we still do not know whether the results of non-invasive tests improve the ability of a cardiologist to predict outcome after myocardial infarction in comparison with clinical information only. This study aimed to provide the answer. The study population was limited to patients aged 65 years and older since they comprise a substantial proportion of hospital admissions for myocardial infarction and are less likely than younger patients to undergo interventions, such as coronary artery angioplasty or surgery, which could alter
prognosis. Patients and methods Eligible patients were aged 65 years and older and were admitted to the coronary care unit of an urban, university-affiliated hospital. The diagnosis of myocardial infarction was defined as the presence of
following: appearance of new Q-waves on electrocardiogram (ECG); typical ischaemic chest pain of at least 30 min duration; and serum creatinine kinase (CK) at least twice normal with a minimum 5% myocardial isoenzyme (CK-MB) fraction. All survivors discharged from hospital were eligible for entry to the study unless coronary artery bypass surgery or angioplasty was done during the initial hospital stay or there was any co-existing non-cardiac disease likely to limit survival to less than 12 months. All participants gave written informed consent and the study was approved by the hospital’s institutional review at
least
two
of the
board. Between July, 1988, and January, 1991, 288 patients aged 65 years and older satisifed the study criteria for a diagnosis of myocardial infarction. 141 patients were excluded: 50 died before discharge from hospital; 26 underwent echocardiography for various indications and the results were available to their cardiologists; 22 underwent urgent coronary angiography; 20 refused to take part; 10 were not expected to live longer than 12 months; 7 lived too far away; and 6 had language difficulties. After the early part of the study, precautions were taken to limit the number of echocardiograms ordered and to prevent information on
1069
I
ROC
Figure: in patient care Curves
are
curves
for outcome predictions
by patients’ own cardiologists
given for predictions based on clinical data alone and on
and
results of non-invasive
left-ventricular function being given to the patient’s cardiologist. Instead, the specific reason for ordering each echocardiogram was addressed, such as the presence of a left-ventricular thrombus or the cause of a systolic murmur. The clinical characteristics of the 147 patients who entered the study are given in the table; patients are classified by whether a clinical endpoint (recurrent myocardial infarction or cardiac death) occurred during follow-up. A summary of each clinical course in hospital was prepared by the study nurse. It included clinically important abnormalities in the medical history, physical examination, routine laboratory tests, ECG, and chest radiograph. The table lists the factors thought to be most relevant to the diagnosis of myocardial infarction and possible adverse outcomes. The cardiologist mainly responsible for the care of the patient was asked to predict outcome at 1 year on a standard rating scale from 0 to 10 (least to greatest likelihood of an event occurring) on the basis of the clinical findings and routine tests. Endpoints of interest were recurrent myocardial infarction or cardiac death. Immediately after making this initial "clinical" prediction, the primary cardiologist was shown data from an exercise test, radionuclide ventriculogram, and 24 h Holter ECG recording, which were done 7-14 days after admission with special procedures to make sure that the patient’s own cardiologist was not informed of the results of these tests before the primary clinical prediction had been made. A second prediction of outcome was then made on the same rating scale. Two other cardiologists (two of the authors) who were not involved in the care of the patient were also asked to make predictions of outcome based initially on clinical information prepared by the study nurse and then after being shown the results of the three non-invasive tests. Treadmill exercise testing was done with a Quinton 3000 exercise system (Seattle, USA). The treadmill speed was maintained at 2-7 km per h (1-7 miles per h) with the incline increasing from 0° to 5° and 10° with 3 min intervals. The test was stopped either at the onset of symptoms (eg, angina, dyspnoea, leg fatigue) or after 9 min. A positive test required at least 1 mm ST segment depression from baseline in 2 contiguous ECG leads. If the test could not be done because of severe cardiac disability, the result was taken to be positive; the inability to carry out an exercise test has been associated with increased cardiovascular morbidity and mortality in this patient population. Equivocal results were taken to be negative. 30 patients were unable to do treadmill exercise for reasons unrelated to cardiac disease. A summary of the findings of each exercise test as outlined in the panel was prepared for the cardiologists. 24 h Holter ECG recording was done with a Space Labs (Redmond, Washington, USA) model 90205 recorder and model 90103 full disclosure analyser. Technically adequate Holter ECG recordings were obtained in all 147 patients. Clinically important
1070
iatc
CtlOW-VU01LIV"-
by two other cardiologists not Involved diagnostic tests.
abnormalities
(panel) were summarised for presentation to the cardiologist. Clinically important ventricular ectopy was deemed to be present if the mean ventricular premature complex rate was 10 per h or greater or if ventricular tachycardia occurred. The radionuclide ventriculogram was done between 5 and 10 days after admission with a General Electric (Horsholm, Denmark) Starcam 300A gamma camera. The left-ventricular ejection fraction
was
calculated and the presence of any wall motion
abnormality was noted (panel). A satisfactory radionuclide ventriculogram was obtained in 139 subjects. Patients were followed up for 12 months with regular telephone contact to document any hospital admission, serious illness, or death. A pre-signed release of information form was used to obtain details of any events, including hospital admission, copies of ECGs, cardiac enzyme results, or primary cause of death. At the end of the study, the areas under the receiver operating characteristic (ROC) curve for predictions of outcome with and without knowledge of the non-invasive test results were compared by the method of McNeil and Hanley.11 The area under the ROC curve represents the probability of selecting a patient with an outcome from a randomly selected pair of patients, one with and one without the outcome. Stepwise logistic regression analysis was used
cardiovascular variables predictive of outcome at 1 Categorical variables were analysed by Fisher’s exact probability test. The minimum level of statistical significance with to assess
year.
two-sided
tests was set at
Panel : Information *
p < 0-05.
provided to the cardiologists
Exercise test
or other symptoms suggestive of ischaemia Electrocardiographic evidence of ischaemia Unable to carry out exercise test (including reasons): Failure to complete the exercise test Arrhythmias during or after exercise Hypotension with exercise
Chest pain
.
24 h Holter ECG
recording
Number of ventricular premature complexes per 24 h Ventricular couplets Ventncular tachycardia ( 3 3 complexes -> 120 per min) Sustained atnal arrhythmias Heart block greater than first degree Volunteered symptoms .
Radionuclide ventriculogram
Left-ventncular ejection fraction Presence of wall motion abnormalities
excluded from the stepwise logistic regression analysis. 20 patients received thrombolytic treatment after admission to hospital. Of the clinical variables examined, only a history of previous myocardial infarction or pre-existing heart failure was significantly more frequent in the patients who had endpoints during follow-up than in the other patients
were
(table). Of the 140
ineligible patients with a diagnosis of 91 survived until hospital discharge. infarction, myocardial 12 of these were lost to follow-up. 9 of the remaining 79 patients (11%) died of cardiac causes within the next year. Data on myocardial infarction in this group could not be obtained because no follow-up ECGs were available. Discussion of non-invasive tests to improve predictions of after myocardial infarction is intuitively attractive. Cardiologists tend to do one or more noninvasive tests before choosing between medical treatment and a more aggressive interventional approach. This apparently logical and simple algorithm involves several assumptions, including the belief that the results of non-invasive tests improve the cardiologist’s ability to predict outcome. Such a belief seems reasonable, since post-hoc statistical analysis shows that the individual non-invasive tests can predict outcome. Our study is an initial attempt to assess the real contribution of specialised diagnostic tests to outcome prediction prospectively in the clinical setting after myocardial infarction. Our results show that the three most commonly used non-invasive tests do not improve a cardiologist’s ability to predict outcome in patients aged 65 years and older. Although the cardiologists were able to estimate prognosis significantly better than chance alone on clinical data and simple laboratory tests, their predictions were inaccurate. Statistical analysis of the left-ventricular ejection fraction in these older patients showed that this variable is better at predicting outcome than the cardiologists. Thus, non-invasive tests may identify highrisk patients in post-hoc statistical models, but they do not necessarily add incrementally to the accuracy of predictions of outcome when prospectively incorporated into a cardiologist’s decision-making process in the clinical
The
use
outcome
LVEF=left-ventricular ejection fraction; M = myocardial infarction. *p < 01, tp< 0 05, tp<0 001 for comparison of patients with and without endpoints.
Table: Clinical
findings and results of three non-invasive tests
Results
During the 1-year follow-up there were 24 cardiovascular deaths and 3 recurrent myocardial infarctions among the 147 patients. Predictions based on clinical data alone were better than chance (figure). significantly (p<0-05) the non-invasive test results did not of However, knowledge the increase accuracy of the three sets of significantly predictions in comparison with those based on clinical data only. ROC curve areas with and without knowledge of the test results for the primary and other cardiologists were 0-62 (SE 0-06) and 0-60 (0-06), 0-63 (0-06) and 0-64 (0-06), and 0-61 (0-06) andO-65 (0-06), respectively (figure). The sample size would allow us to detect an increase in the ROC curve area of up to 0. 14 (alpha.=0-05, beta = 0-2, one-tailed." Stepwise logistic regression analysis of the non-invasive test results showed only the ejection fraction to be predictive of outcome (p 0-008). ROC curve analysis with the ejection fraction to predict outcome gave an area of 0-68 (006), a value higher than any of the cardiologist’s predictions, although not significantly so. Ventricular ectopy in the 24 h Holter ECG recording did not correlate with outcome. 30 of the 147 patients did not undertake treadmill exercise for reasons unrelated to cardiac disease, including impaired musculoskeletal function (12), intermittent claudication (5), poor general health (4), lung disease (1), and patient refusal (8). 6 of these 30 patients experienced endpoints during follow-up. These 30 patients =
setting. The ROC curve areas in this study ranged from 0-60 to 0-65. For a diagnostic test to be useful in the clinical setting, the area should exceed 080; lower values lead to a high false-positive prediction rate.6,12 The sample size of this study was sufficiently large for any clinically meaningful changes to be detected, if the information provided by the diagnostic tests had improved the ability of the cardiologists
predict outcome. Why were the predictions of the cardiologist not improved by knowledge of the non-invasive test results? Clinical findings alone may provide adequate estimates of left-ventricular function and residual ischaemia, so further information would be of little value. Knowledge of a patient’s clinical course in hospital, including the presence of congestive heart failure and post-infarction angina, probably identifies many of the same higher-risk patients as do the radionuclide ventriculogram and treadmill exercise
to
test.
risk stratification after myocardial infarction in the elderly is based primarily on data derived from younger patients. 13,14 The extrapolation The
current
approach
to
1071
of these data to the elderly may not be justified. Older people who have myocardial infarctions are more likely than younger people to have atypical symptoms,15 to receive less aggressive therapy,16 to have ventricular ectopy, 17 to be less able to undertake treadmill exercise,18 and to die.19-21 Moreover, most studies supporting recommended approaches to risk stratification were done before use of anti-platelet and beta-blocker prophylaxis became routine. Thus, our study may provide new insight into the status of older patients with myocardial infarction, since each patient received aspirin and beta-blocker treatment unless specifically contraindicated. The beneficial effects of these drugs on reinfarction and cardiovascular death rates could explain why the exercise test was not a useful predictor of outcome in the study population. Stepwise logistic regression analysis can identify clinical variables or diagnostic tests that are the best predictors of subsequent events in a specific group of patients, but it provides less useful information for the individual patient. Diamond2l has emphasised that one can improve the accuracy of a prediction by incorporating more variables (clinical or diagnostic tests) into a mathematical model, but in doing so, the variance associated with the individual
prediction increases-"powerful predictors are not very common and common predictors are not very powerful". This reality partly explains why three cardiologists, including the patient’s own physician, failed to predict outcome accurately. The ROC curve analysis showed the cardiologist’s predictions to be only slightly better (albeit significantly so) than those made by chance alone. These findings may be discomforting for physicians who rely on non-invasive tests for decisions about outcome and the need for intervention in patients with acute myocardial infarction. Because of the important clinical and economic implications of routine use of these tests, further studies are needed to clarify the limitatons of existing approaches to assessing risk after acute myocardial infarction. We thank Ms Pauline Parsons, Dr John P Szalai, and Mr Marko Katic, for important contributions to the study. The study was supported by a grant-in-aid from the Heart and Stroke Foundation of Ontario.
References 1 2 3
DeBusk RF. Specialized testing after recent acute myocardial infarction. Ann Intern Med 1989; 110: 470-81. Moss AJ, Benhorin J. Prognosis and management after a first myocardial infarction. N Engl J Med 1990; 322: 743-53. DeBusk RF, Kramer HC, Nash E. Stepwise risk stratification soon after acute myocardial infarction. Am J Cardiol 1983; 52: 116166.
1072
Williams WL, Nair RC, Higginson LAJ, Baird MG, Allan K, Beanlands DS. Comparison of clinical and treadmill variables for the prediction of outcome after myocardial infarction. J Am Coll Cardiol 1984; 4: 477-86. 5 Fioretti P, Brower RW, Simoons ML, et al. Relative value of clinical variables, bicycle ergometry, rest radionuclide ventriculography and 24 hour ambulatory electrocardiographic monitoring at discharge to predict 1 year survival after myocardial infarction. J Am Coll Cardiol 1986; 8: 40-49. 6 Griffin BP, Shah PK, Diamond GA, Berman DS, Ferguson JG. Incremental prognostic accuracy of clinical, radionuclide and hemodynamic data in acute myocardial infarction. Am J Cardiol 1991; 68: 707-12. 7 Fioretti P, Brower RW, Simoons ML, et al. Prediction of mortality during the first year after acute myocardial infarction from variables and stress test at hospital discharge. Am J Cardiol 1985; 55: 1313-18. 8 Olmstead WL, Groden DL, Silverman ME. Prognosis in survivors of acute myocardial infarction occurring at age 70 years or older. Am J Cardiol 1987; 60: 971-75. 9 Pollock SG, Abbott RD, Boucher CA, Beller GA, Kaul S. Independent and incremental prognostic value of tests performed in hierarchical order to evaluate patients with suspected coronary artery disease. Circulation 1992; 85: 237-48. 10 Hlatky M, Botvinick E, Brundage B. Diagnostic accuracy of cardiologists compared with probability calculations using Bayes’ rule. Am J Cardiol 1982; 49: 1927-31. 11 McNeil BJ, Hanley JA. Statistical approaches to the analysis of receiver operating characteristic (ROC) curves. Med Decis Making 1984; 4: 137-50. 12 Diamond GA. Future imperfect: the limitations of clinical prediction models and the limits of clinical prediction. J Am Coll Cardiol 1989; 14: 12A-22A. 13 American College of Physicians. Evaluation of patients after recent acute myocardial infarction. Ann Intern Med 1989; 110: 485-88. 14 Fallen EL, Armstrong P, Cairns J, et al. Report of the Canadian Cardiovascular Society’s Consensus Conference on the Management of the Post-Myocardial Infarction Patient. Can Med Assoc J 1991; 144: 1015-25. 15 Solomon CG, Lee TH, Cook EF, et al. Comparison of clinical presentation of acute myocardial infarction in patients older than 65 years of age to younger patients: the Multicentre Chest Pain Study Experience. Am J Cardiol 1989; 63: 772-76. 16 Montague TJ, Wong RY, Burton JR, Bay KS, Catellier DJ, Teo KK. Changes in acute myocardial infarction risk and patterns of practice for patients older and younger than 70 years, 1987-90. Can J Cardiol 1992; 8: 596-600. 17 Crow RS, Prineas RJ, Dias V, Taylor HL, Jacobs D, Blackburn H. Ventricular premature beats in a population sample. Circulation 1975; 51 (suppl III): 211-20. 18 Deckers JW, Fioretti P, Simoons ML, Baardman T, Hugenholtz PG. Ineligibility for pre-discharge exercise testing after myocardial infarction in the elderly: implications for prognosis. Eur Heart J 1984; 5 (suppl E): 97-100. 19 Hoit BD, Gilpin EA, Henning H, et al. Myocardial infarction in young patients: an analysis by age subsets. Circulation 1986; 74: 712-21. 20 Cairns JA, Singer J, Gent M, et al. One year mortality outcomes of all coronary and intensive care unit patients with acute myocardial infarction, unstable angina or other chest pain in Hamilton, Ontario, a city of 375,000 people. Can J Cardiol 1989; 5: 239-46. 21 Montague TJ, Ikuta RM, Wong RY, Bay KS, Teo KK, Davies NJ. Comparison of risk and patterns of practice in patients older and younger than 70 years with acute myocardial infarction in a 2 year period (1987-1989). Am J Cardiol 1991; 68: 843-47.
4
Summary
Introduction
Many patients with acute myocardial infarction undergo tests identify ischaemia, left-ventricular dysfunction, and arrhythmias. We examined the usefulness of these tests in clinical practice by comparing the ability of a cardiologist to
Many patients who experience myocardial infarction have a recurrence or die from cardiac causes within the following year. Factors contributing to the increased risk include the extent of myocardial damage, the presence of residual ischaemia, and the development of electrical instability. Non-invasive tests aimed at quantifying these factors are widely used by physicians to stratify patients into high and low risk categories.l,2 Several clinical variables are also associated with a poor outlook, including signs of leftventricular dysfunction, advanced age, and persisting angina after the infarction.1-6
to
predict outcome at 1 year after infarction with and without knowledge of the results of an exercise test, radionuclide angiogram, and 24 h Holter electrocardiographic (ECG) recording. The study was limited to patients older than 65 years, who have a greater risk of cardiovascular sequelae and undergo fewer interventional procedures. The patient’s own cardiologist predicted outcome on a standard rating scale, based on clinical findings and routine hospital tests. He then made a second prediction after seeing the non-invasive test results. Two other cardiologists not involved in the care of the patient independently made similar predictions. Success in predicting outcome was assessed by comparison of differences between the first and second predictions in the area under receiver operating characteristic curves. During 1 year’s follow-up there were 24 cardiovascular deaths and 3 recurrent myocardial infarctions among the 147 patients. There were no significant differences in mean curve the first and second predictions for the patients’ own cardiologist (0·62 [SE 0·06] vs 0 60 [0·06]) or the other cardiologists (0 63 [0·06] vs 0·64 [0·06] and 0·61 [0 06] vs 0·65 [0·06]). All predictions better than chance. were significantly (p < 0·05) Prediction of outcome in older patients after myocardial infarction is not improved by knowledge of the results of an exercise test, radionuclide angiogram, or 24 h Holter ECG areas between
recording. Lancet 1993; 342: 1069-72
Division of Cardiology, Sunnybrook Health Science Centre and Department of Medicine, University of Toronto, Toronto, Ontario, Canada (M G Myers FRCPC, R S Baigrie FRCPC, M L Charlat FRCPC, C D Morgan FRCPC)
Correspondence to: Dr Martin G Myers, Division of Cardiology, Sunnybrook Health Science Centre, A-222, 2075 Bayview Avenue, Toronto, Ontario M4N 3M5, Canada
Several studies have used mathematical models to compare the usefulness of predictions based on clinical findings with risk stratification based on data from noninvasive tests.4-1O However, these studies either have involved theoretical cases or have used a retrospective approach to an existing data base. Thus, we still do not know whether the results of non-invasive tests improve the ability of a cardiologist to predict outcome after myocardial infarction in comparison with clinical information only. This study aimed to provide the answer. The study population was limited to patients aged 65 years and older since they comprise a substantial proportion of hospital admissions for myocardial infarction and are less likely than younger patients to undergo interventions, such as coronary artery angioplasty or surgery, which could alter
prognosis. Patients and methods Eligible patients were aged 65 years and older and were admitted to the coronary care unit of an urban, university-affiliated hospital. The diagnosis of myocardial infarction was defined as the presence of
following: appearance of new Q-waves on electrocardiogram (ECG); typical ischaemic chest pain of at least 30 min duration; and serum creatinine kinase (CK) at least twice normal with a minimum 5% myocardial isoenzyme (CK-MB) fraction. All survivors discharged from hospital were eligible for entry to the study unless coronary artery bypass surgery or angioplasty was done during the initial hospital stay or there was any co-existing non-cardiac disease likely to limit survival to less than 12 months. All participants gave written informed consent and the study was approved by the hospital’s institutional review at
least
two
of the
board. Between July, 1988, and January, 1991, 288 patients aged 65 years and older satisifed the study criteria for a diagnosis of myocardial infarction. 141 patients were excluded: 50 died before discharge from hospital; 26 underwent echocardiography for various indications and the results were available to their cardiologists; 22 underwent urgent coronary angiography; 20 refused to take part; 10 were not expected to live longer than 12 months; 7 lived too far away; and 6 had language difficulties. After the early part of the study, precautions were taken to limit the number of echocardiograms ordered and to prevent information on
1069
I
ROC
Figure: in patient care Curves
are
curves
for outcome predictions
by patients’ own cardiologists
given for predictions based on clinical data alone and on
and
results of non-invasive
left-ventricular function being given to the patient’s cardiologist. Instead, the specific reason for ordering each echocardiogram was addressed, such as the presence of a left-ventricular thrombus or the cause of a systolic murmur. The clinical characteristics of the 147 patients who entered the study are given in the table; patients are classified by whether a clinical endpoint (recurrent myocardial infarction or cardiac death) occurred during follow-up. A summary of each clinical course in hospital was prepared by the study nurse. It included clinically important abnormalities in the medical history, physical examination, routine laboratory tests, ECG, and chest radiograph. The table lists the factors thought to be most relevant to the diagnosis of myocardial infarction and possible adverse outcomes. The cardiologist mainly responsible for the care of the patient was asked to predict outcome at 1 year on a standard rating scale from 0 to 10 (least to greatest likelihood of an event occurring) on the basis of the clinical findings and routine tests. Endpoints of interest were recurrent myocardial infarction or cardiac death. Immediately after making this initial "clinical" prediction, the primary cardiologist was shown data from an exercise test, radionuclide ventriculogram, and 24 h Holter ECG recording, which were done 7-14 days after admission with special procedures to make sure that the patient’s own cardiologist was not informed of the results of these tests before the primary clinical prediction had been made. A second prediction of outcome was then made on the same rating scale. Two other cardiologists (two of the authors) who were not involved in the care of the patient were also asked to make predictions of outcome based initially on clinical information prepared by the study nurse and then after being shown the results of the three non-invasive tests. Treadmill exercise testing was done with a Quinton 3000 exercise system (Seattle, USA). The treadmill speed was maintained at 2-7 km per h (1-7 miles per h) with the incline increasing from 0° to 5° and 10° with 3 min intervals. The test was stopped either at the onset of symptoms (eg, angina, dyspnoea, leg fatigue) or after 9 min. A positive test required at least 1 mm ST segment depression from baseline in 2 contiguous ECG leads. If the test could not be done because of severe cardiac disability, the result was taken to be positive; the inability to carry out an exercise test has been associated with increased cardiovascular morbidity and mortality in this patient population. Equivocal results were taken to be negative. 30 patients were unable to do treadmill exercise for reasons unrelated to cardiac disease. A summary of the findings of each exercise test as outlined in the panel was prepared for the cardiologists. 24 h Holter ECG recording was done with a Space Labs (Redmond, Washington, USA) model 90205 recorder and model 90103 full disclosure analyser. Technically adequate Holter ECG recordings were obtained in all 147 patients. Clinically important
1070
iatc
CtlOW-VU01LIV"-
by two other cardiologists not Involved diagnostic tests.
abnormalities
(panel) were summarised for presentation to the cardiologist. Clinically important ventricular ectopy was deemed to be present if the mean ventricular premature complex rate was 10 per h or greater or if ventricular tachycardia occurred. The radionuclide ventriculogram was done between 5 and 10 days after admission with a General Electric (Horsholm, Denmark) Starcam 300A gamma camera. The left-ventricular ejection fraction
was
calculated and the presence of any wall motion
abnormality was noted (panel). A satisfactory radionuclide ventriculogram was obtained in 139 subjects. Patients were followed up for 12 months with regular telephone contact to document any hospital admission, serious illness, or death. A pre-signed release of information form was used to obtain details of any events, including hospital admission, copies of ECGs, cardiac enzyme results, or primary cause of death. At the end of the study, the areas under the receiver operating characteristic (ROC) curve for predictions of outcome with and without knowledge of the non-invasive test results were compared by the method of McNeil and Hanley.11 The area under the ROC curve represents the probability of selecting a patient with an outcome from a randomly selected pair of patients, one with and one without the outcome. Stepwise logistic regression analysis was used
cardiovascular variables predictive of outcome at 1 Categorical variables were analysed by Fisher’s exact probability test. The minimum level of statistical significance with to assess
year.
two-sided
tests was set at
Panel : Information *
p < 0-05.
provided to the cardiologists
Exercise test
or other symptoms suggestive of ischaemia Electrocardiographic evidence of ischaemia Unable to carry out exercise test (including reasons): Failure to complete the exercise test Arrhythmias during or after exercise Hypotension with exercise
Chest pain
.
24 h Holter ECG
recording
Number of ventricular premature complexes per 24 h Ventricular couplets Ventncular tachycardia ( 3 3 complexes -> 120 per min) Sustained atnal arrhythmias Heart block greater than first degree Volunteered symptoms .
Radionuclide ventriculogram
Left-ventncular ejection fraction Presence of wall motion abnormalities
excluded from the stepwise logistic regression analysis. 20 patients received thrombolytic treatment after admission to hospital. Of the clinical variables examined, only a history of previous myocardial infarction or pre-existing heart failure was significantly more frequent in the patients who had endpoints during follow-up than in the other patients
were
(table). Of the 140
ineligible patients with a diagnosis of 91 survived until hospital discharge. infarction, myocardial 12 of these were lost to follow-up. 9 of the remaining 79 patients (11%) died of cardiac causes within the next year. Data on myocardial infarction in this group could not be obtained because no follow-up ECGs were available. Discussion of non-invasive tests to improve predictions of after myocardial infarction is intuitively attractive. Cardiologists tend to do one or more noninvasive tests before choosing between medical treatment and a more aggressive interventional approach. This apparently logical and simple algorithm involves several assumptions, including the belief that the results of non-invasive tests improve the cardiologist’s ability to predict outcome. Such a belief seems reasonable, since post-hoc statistical analysis shows that the individual non-invasive tests can predict outcome. Our study is an initial attempt to assess the real contribution of specialised diagnostic tests to outcome prediction prospectively in the clinical setting after myocardial infarction. Our results show that the three most commonly used non-invasive tests do not improve a cardiologist’s ability to predict outcome in patients aged 65 years and older. Although the cardiologists were able to estimate prognosis significantly better than chance alone on clinical data and simple laboratory tests, their predictions were inaccurate. Statistical analysis of the left-ventricular ejection fraction in these older patients showed that this variable is better at predicting outcome than the cardiologists. Thus, non-invasive tests may identify highrisk patients in post-hoc statistical models, but they do not necessarily add incrementally to the accuracy of predictions of outcome when prospectively incorporated into a cardiologist’s decision-making process in the clinical
The
use
outcome
LVEF=left-ventricular ejection fraction; M = myocardial infarction. *p < 01, tp< 0 05, tp<0 001 for comparison of patients with and without endpoints.
Table: Clinical
findings and results of three non-invasive tests
Results
During the 1-year follow-up there were 24 cardiovascular deaths and 3 recurrent myocardial infarctions among the 147 patients. Predictions based on clinical data alone were better than chance (figure). significantly (p<0-05) the non-invasive test results did not of However, knowledge the increase accuracy of the three sets of significantly predictions in comparison with those based on clinical data only. ROC curve areas with and without knowledge of the test results for the primary and other cardiologists were 0-62 (SE 0-06) and 0-60 (0-06), 0-63 (0-06) and 0-64 (0-06), and 0-61 (0-06) andO-65 (0-06), respectively (figure). The sample size would allow us to detect an increase in the ROC curve area of up to 0. 14 (alpha.=0-05, beta = 0-2, one-tailed." Stepwise logistic regression analysis of the non-invasive test results showed only the ejection fraction to be predictive of outcome (p 0-008). ROC curve analysis with the ejection fraction to predict outcome gave an area of 0-68 (006), a value higher than any of the cardiologist’s predictions, although not significantly so. Ventricular ectopy in the 24 h Holter ECG recording did not correlate with outcome. 30 of the 147 patients did not undertake treadmill exercise for reasons unrelated to cardiac disease, including impaired musculoskeletal function (12), intermittent claudication (5), poor general health (4), lung disease (1), and patient refusal (8). 6 of these 30 patients experienced endpoints during follow-up. These 30 patients =
setting. The ROC curve areas in this study ranged from 0-60 to 0-65. For a diagnostic test to be useful in the clinical setting, the area should exceed 080; lower values lead to a high false-positive prediction rate.6,12 The sample size of this study was sufficiently large for any clinically meaningful changes to be detected, if the information provided by the diagnostic tests had improved the ability of the cardiologists
predict outcome. Why were the predictions of the cardiologist not improved by knowledge of the non-invasive test results? Clinical findings alone may provide adequate estimates of left-ventricular function and residual ischaemia, so further information would be of little value. Knowledge of a patient’s clinical course in hospital, including the presence of congestive heart failure and post-infarction angina, probably identifies many of the same higher-risk patients as do the radionuclide ventriculogram and treadmill exercise
to
test.
risk stratification after myocardial infarction in the elderly is based primarily on data derived from younger patients. 13,14 The extrapolation The
current
approach
to
1071
of these data to the elderly may not be justified. Older people who have myocardial infarctions are more likely than younger people to have atypical symptoms,15 to receive less aggressive therapy,16 to have ventricular ectopy, 17 to be less able to undertake treadmill exercise,18 and to die.19-21 Moreover, most studies supporting recommended approaches to risk stratification were done before use of anti-platelet and beta-blocker prophylaxis became routine. Thus, our study may provide new insight into the status of older patients with myocardial infarction, since each patient received aspirin and beta-blocker treatment unless specifically contraindicated. The beneficial effects of these drugs on reinfarction and cardiovascular death rates could explain why the exercise test was not a useful predictor of outcome in the study population. Stepwise logistic regression analysis can identify clinical variables or diagnostic tests that are the best predictors of subsequent events in a specific group of patients, but it provides less useful information for the individual patient. Diamond2l has emphasised that one can improve the accuracy of a prediction by incorporating more variables (clinical or diagnostic tests) into a mathematical model, but in doing so, the variance associated with the individual
prediction increases-"powerful predictors are not very common and common predictors are not very powerful". This reality partly explains why three cardiologists, including the patient’s own physician, failed to predict outcome accurately. The ROC curve analysis showed the cardiologist’s predictions to be only slightly better (albeit significantly so) than those made by chance alone. These findings may be discomforting for physicians who rely on non-invasive tests for decisions about outcome and the need for intervention in patients with acute myocardial infarction. Because of the important clinical and economic implications of routine use of these tests, further studies are needed to clarify the limitatons of existing approaches to assessing risk after acute myocardial infarction. We thank Ms Pauline Parsons, Dr John P Szalai, and Mr Marko Katic, for important contributions to the study. The study was supported by a grant-in-aid from the Heart and Stroke Foundation of Ontario.
References 1 2 3
DeBusk RF. Specialized testing after recent acute myocardial infarction. Ann Intern Med 1989; 110: 470-81. Moss AJ, Benhorin J. Prognosis and management after a first myocardial infarction. N Engl J Med 1990; 322: 743-53. DeBusk RF, Kramer HC, Nash E. Stepwise risk stratification soon after acute myocardial infarction. Am J Cardiol 1983; 52: 116166.
1072
Williams WL, Nair RC, Higginson LAJ, Baird MG, Allan K, Beanlands DS. Comparison of clinical and treadmill variables for the prediction of outcome after myocardial infarction. J Am Coll Cardiol 1984; 4: 477-86. 5 Fioretti P, Brower RW, Simoons ML, et al. Relative value of clinical variables, bicycle ergometry, rest radionuclide ventriculography and 24 hour ambulatory electrocardiographic monitoring at discharge to predict 1 year survival after myocardial infarction. J Am Coll Cardiol 1986; 8: 40-49. 6 Griffin BP, Shah PK, Diamond GA, Berman DS, Ferguson JG. Incremental prognostic accuracy of clinical, radionuclide and hemodynamic data in acute myocardial infarction. Am J Cardiol 1991; 68: 707-12. 7 Fioretti P, Brower RW, Simoons ML, et al. Prediction of mortality during the first year after acute myocardial infarction from variables and stress test at hospital discharge. Am J Cardiol 1985; 55: 1313-18. 8 Olmstead WL, Groden DL, Silverman ME. Prognosis in survivors of acute myocardial infarction occurring at age 70 years or older. Am J Cardiol 1987; 60: 971-75. 9 Pollock SG, Abbott RD, Boucher CA, Beller GA, Kaul S. Independent and incremental prognostic value of tests performed in hierarchical order to evaluate patients with suspected coronary artery disease. Circulation 1992; 85: 237-48. 10 Hlatky M, Botvinick E, Brundage B. Diagnostic accuracy of cardiologists compared with probability calculations using Bayes’ rule. Am J Cardiol 1982; 49: 1927-31. 11 McNeil BJ, Hanley JA. Statistical approaches to the analysis of receiver operating characteristic (ROC) curves. Med Decis Making 1984; 4: 137-50. 12 Diamond GA. Future imperfect: the limitations of clinical prediction models and the limits of clinical prediction. J Am Coll Cardiol 1989; 14: 12A-22A. 13 American College of Physicians. Evaluation of patients after recent acute myocardial infarction. Ann Intern Med 1989; 110: 485-88. 14 Fallen EL, Armstrong P, Cairns J, et al. Report of the Canadian Cardiovascular Society’s Consensus Conference on the Management of the Post-Myocardial Infarction Patient. Can Med Assoc J 1991; 144: 1015-25. 15 Solomon CG, Lee TH, Cook EF, et al. Comparison of clinical presentation of acute myocardial infarction in patients older than 65 years of age to younger patients: the Multicentre Chest Pain Study Experience. Am J Cardiol 1989; 63: 772-76. 16 Montague TJ, Wong RY, Burton JR, Bay KS, Catellier DJ, Teo KK. Changes in acute myocardial infarction risk and patterns of practice for patients older and younger than 70 years, 1987-90. Can J Cardiol 1992; 8: 596-600. 17 Crow RS, Prineas RJ, Dias V, Taylor HL, Jacobs D, Blackburn H. Ventricular premature beats in a population sample. Circulation 1975; 51 (suppl III): 211-20. 18 Deckers JW, Fioretti P, Simoons ML, Baardman T, Hugenholtz PG. Ineligibility for pre-discharge exercise testing after myocardial infarction in the elderly: implications for prognosis. Eur Heart J 1984; 5 (suppl E): 97-100. 19 Hoit BD, Gilpin EA, Henning H, et al. Myocardial infarction in young patients: an analysis by age subsets. Circulation 1986; 74: 712-21. 20 Cairns JA, Singer J, Gent M, et al. One year mortality outcomes of all coronary and intensive care unit patients with acute myocardial infarction, unstable angina or other chest pain in Hamilton, Ontario, a city of 375,000 people. Can J Cardiol 1989; 5: 239-46. 21 Montague TJ, Ikuta RM, Wong RY, Bay KS, Teo KK, Davies NJ. Comparison of risk and patterns of practice in patients older and younger than 70 years with acute myocardial infarction in a 2 year period (1987-1989). Am J Cardiol 1991; 68: 843-47.
4