- Email: [email protected]
Selection of patients for coronary arteriography with Thallium-201: Operating position on the receiver operating characteristic curve
Werkin PrsJresst-----Int. J. NW/. Med. & Btol. Vol. 8. pp. 204 lo 208. 1981 Pergamon Press Ltd. Printed in Great Britam 0047.0140/81/080204-05~02.00/0
review reported results of exercise-redistribution 201Tl myocardial perfusion imaging as a diagnostic test by analyzing its sensitivity, specificity, accuracy, and predictive values; (2) construct a receiver operating characteristic (ROC) curve based on reported data and on the results of our own series: (3) to define, from qualitative cost estimates, prevalence data, and information content analysis. the appropriate operating position on the ROC curve; and (4) to indicate from these results the appropriate role of thallium myocardial perfusion imaging as a diagnostic test.
!Mection of Patients for Coronary Arteriography with Thallium-201 : Operating Position on the Receiver Operating Characteristic Curve MARTIN L. NUSYNOWITZ,* ROBERTE. SONNEMAKER* and JOHN L. FLOYD* Nuclear Medicine Service, William Beaumount Army Medical Center, El Paso, TX 79920 and Division of Nuclear Medicine, University of Texas Health Science Center San Antonio, San Antonio, TX 78284, U.S.A.
Introduction IF ONE assumes that coronary bypass surgery improves the quality or quantity of life in patients with myocardial ischemia, and given that coronary arteriography is necessary in the selection of such patients for these surgical procedures, it becomes imperative to develop a strategy for screening for coronary arteriography. Such a strategy is necessary because coronary arteriography is expensive and has an associated morbidity and mortality. Heretofore, symptomatic patients with possible myocardial ischemic disease were screened on the basis of history, physical examination, routine laboratory testing and exercise stress electrocardiography. With these modalities, the selection process was excellent, since 75% of the usual population of patients studied arteriographically are found to have significant coronary artery disease. Recently efforts have been expended to define the role of 201TI myocardial perfusion imaging in this selection process. The purpose of this paper is to: (1)
Materials and Methods Five reported series consisting of thallium myocardial perfusion imaging and coronary arteriographic results in patients evaluated for myocardial ischemia were analyzed!1x5) The numbers of patients faliing into each of four categories-true positive, false positive, false negative and true negative-were tabulated using the arteriographic results as the standard. Although the criteria for arteriographic a$normalities varied somewhat in the various series (ranging from greater than 5OPi to greater than 70?:, narrowing of coronary artery diameter), decisions as to normality or abnormality of results for either arteriography or thallium myocardial perfusion imaging remained as reported. A similar tabulation was performed for the results of 36 consecutive patients studied by our group, in which processing of images by a continuously increasing background subtraction technique resulted in the enhancement of lesions such that sensitivity was 100%.‘6’ The data from the five reported series were analyzed both individually and as a group for sensitivity, specificity, accuracy and prevalence of coronary artery disease in the population. Bayes’ theorem was used to determine positive and negative predictive values at a prevalence of 750,; (the mean prevalence for the pooled data) using the following equations:
Positive Predictive
Value:
[P(D+)][Sensitivity] ‘@+
IT+)
= [P(D+)][Sensitivity] Negative
+ [l - P(D+)][l
Predictive
- Specificity]
Value:
[l - P(D+)] [Specificity] P(D-‘T-)
+ [P(D+)][l
= [l - P(D+)][Specificity]
* Present address: Martin L. Nusynowitz, Division of Nuclear Medicine, University of Texas Health Science Center San Antonio, 7703 Floyd Curl Drive, San Antonio, TX 78284, Robert E. Sonnemaker, Division of Nuclear Medicine, St Luke’s Episcopal, Texas Children’s Hospital, Houston, Texas, and- John L. Floyd, Nuclear Medicine Division, Presbyterian Hospital, Dallas, Texas.
(I)
- Sensitivity]’
(2)
From the sensitivity and specificity data, a receiver operating characteristic curve was drawn by visual interpolation through points defined by the sensitivity and specificity of each study and for the pooled data. Information content for each point was calculated by the method of Metz et. ~1.“) The ratios: positive predictive value to prevalence P(D+ ) T + )/P(D+ ) and the probability of disease, given a negative result, compared to prevalence P(D + I T - )/P(D + ), were also computed for evaluation of utility of each result. 204
Work
in proyrvss
RI2SUltS The calculated value for each parameter in each of the reported series and for the pooled data is given in Table 1. The prevalence of angiographically demonstrable significant coronary artery disease for the pooled data is 75% and is remarkably constant among the series (range 6878%). The overall sensitivity for the five pooled series is 727; (range 5693%); the specificity is 937; and the accuracy is 77%. The positive predictive value at a prevalence of 75 is 977; (range 91-lOO”/A), but the negative predictive value is only 525;. By comparison, the sensitivity of the grouped data for electrocardiography is only 51x, the specificity 88”~;,and the accuracy only 60% (Table 2). The positive predictive value with prevalence of 75 is 93% and the negative predictive value is 37%. These data are very similar to those found in eight other exercise ECG-coronary arteriographic studies reviewed by FORTUIN& WEISS.@) The ROC curve for the “‘Tl myocardial perfusion imaging data from the five series, the pooled value, and our series is plotted as Fig. 1. Note that four of the five series and the group value fall in the low sensitivity, high specificity (low false positive) region of the curve. In contrast to these results, the processed lesionenhanced results of our series showed a sensitivity of lob”,;, a specificity of 56% and the highest resultant accuracy of 89%. More importantly, although the positive predictive value is slightly lower than that of other studies (873&), the negative predictive value is loons& indicating that a normal lesion-enhanced thallium myocardial perfusion imaging study virtually excludes the possibility of a positive coronary arteriogram. The overall average of the ratio P(D+ I T+)/ P(D + ) for the combined ” ‘Tl data was 1.29, showing a slight gain of utility (29%) at a prevalence of 75%; the corresponding value for the authors’ series was 1.16. The ratio P(D + 1T - )/P(D + ) for the combined series was 0.63, differing from unity by 37% and indicating that test utility was greater when used to ex-
0
I
I
2
.4 False
positive
.6
.8
1.0
rote
FIG. 1. Receiver operating characteristic curve for 201Tl myocardial perfusion imaging. The open circle represents the mean values from 5 reported series.
205
206
Work in progress
Table 2. Results of stress ECG in detection
Reference
No. patients
Sensitivity
of myocardial
Specificity
Accuracy
ischemia Positive predictive value*
Negative predictive value*
Authors’ series
36
44
89
56
92
35
I:; (4) (5)
137 31 83 22
92 52 38 40
71 83 loo 88
87 59 53 55
91 90 100 89
37 74 35 32
309
51%
Combined * At prevalence
88%
93%
37%
= 75%.
elude coronary artery disease. The corresponding value from the authors’ series was 0.00, differing from unity by lOO%, indicating the greatest test utility. Discussion Diagnostic tests may be classified as tests of discovery, exclusion, or confirmation.t9) Discovery tests should be highly sensitive, even at the cost of specificity and inexpensive in both financial and health costs, so that a disease may be found if present. The stress electrocardiogram has long been employed as the basic discovery test for ischemic myocardial disease, but its shortcomings in this regard are well appreciated.t”’ Simply stated, abnormal results are good predictors of coronary artery disease (high positive predictive value), but normal or indeterminate results of stress electrocardiography as usually employed are meaningless. Because of its low sensitivity, normal findings of exercise electrocardiography are often associated with coronary artery disease. Even when a more liberal criterion of abnormality is employed with exercise electrocardiography (0.5 mm ST depression), sensitivity remains unacceptably low (63%) and specificity falls.” ‘) Thus, stress electrocardiographic testing does not fulfill the requirements for a discovery test. Thallium myocardial perfusion imaging has been studied as a screening test for coronary artery disease. It is obvious that the sensitivity of the test as generally employed, while far exceeding that of exercise electrocardiography, is still too low for use of the modality as a discovery test. If thallium myocardial perfusion imaging is interpreted in conjunction with an exercise electrocardiographic test so that an abnormal result is defined as either modality being positive, sensitivity rises to 90Y0,“’ a value approaching that of a useful discovery test in many circumstances, but marginal in the population ordinarily evaluated where prevalence of disease is already 75%. Bayes’ theorem indicates that the negative predictive value of a test is dependent on its sensitivity; the negative predictive value approaches 10% as sensitivity approaches lOOo/, (equation 2). Therefore, the
ideal test of exclusion possesses a sensitivity of 100%; in this case, a normal
60%
test excludes
the possibility
of
disease. In a confirmation test the cardinal feature is value specificity, since the positive predictive
approaches lOOD/, as specificity approaches 100% (equation 1). Coronary arteriography with its high specificity thus is the confirmation test for myocardial ischemic disease. Does thallium myocardial perfusion imaging have a role, then, in screening patients for coronary arteriography? The answer lies in the analysis of several factors. Firstly, ROC curve data show that most reported series operate on the left-hand, low sensitivity, high specificity region of the ROC curve. As a consequence, thallium myocardial perfusion imaging results are only a slightly better than exercise electrocardiographic findings; both are excellent in predicting arteriographically demonstrable disease when the test is positive, because of high specificity, but both are poor in excluding disease when negative. It has been reported that when both stress electrocardiography and thallium myocardial perfusion imaging are used in conjunction, the sensitivity increased to 90% approaching the values desired for discovery tests, but the negative predictive value is still too low for use of the combined procedure as an exclusion test. On the other hand, it can be demonstrated that “rT1 myocardial perfusion imaging is most useful if one chooses instead an ROC operating point so as to maximize sensitivity, even at a cost in specificity. MCNEIL ef u/.(‘~’ have emphasized that the optimal position on the ROC curve is that region where the slope K=
additional
cost FP
additional
cost FN
x-.
P(D-) P(D+)
It is readily seen that the slope is given by the product of two factors, one related to costs and the other to prevalence. The prevalence of coronary artery disease is remarkably constant at about 75% in the usual population referred for arteriography; thus, P(D-)/P(D+) = 0.25/0.75 = 0.33, and this fact indicates that the operating position should be in the low slope (high sensitivity, low specificity) portion of the curve. Furthermore, the cost of a false positive preliminary disgnostic test is relatively low compared to that of a false negative, since the former will result in an unnecessary coronary arteriogram whereas the latter, by preventing the performance of a necessary coronary arteriogram, might result in the diminution of the quality and/or quantity of life if bypass surgery
Work in progress
-t 0
I2
34
5
.6
7
8
.9
IO
207
I 0
I
I 2
I 3
I 4
I 5
I .6
I’I 7
0
I’ 9
I
3
P (D+)
P(D+)
FIG. 2. Stress ECG and “‘Tl myocardial perfusion imaging predictive values as a function of prevalence of coronary artery disease. The left hand graph shows positive predictive values and the right hand graph the negative predictive values. The arrow indicates a prevalence of 750.
is thereby denied. Since the ratio of costs is low, the low slope region of the ROC curve should again be employed. Another indication of operating position comes from comparison of information content of each operating point. At a prevalence of 0.75, the maximum information content occurs in using a high sensitivity (but low specificity) operating point. All other points give values for information content that are lower (Table 1). Therefore, the utility of thallium myocardial perfusion imaging from considerations of sensitivity and specificity as compared to stress electrocardiology, cost analysis, information content analysis and prevalence considerations lies in its value as a test of high sensitivity and thereby as an effective exclusion test. and The values of the ratios P(D+ 1T+)/P(D+) P(D + / T - )/P(D + ) provide additional confirmation. This high sensitivity can be achieved by processing manuevers designed to enhance lesions, such as those described.@) Employed in this manner, a normal thallium myocardial perfusion imaging result virtually excludes the diagnosis of coronary artery disease. In our series, 65”/, of patients with negative or indeterminate stress electrocardiograms had angiographically demonstrable coronary artery disease; the processed thallium images were positive in each of these. Not one patient with a negative thallium myocardial perfusion image under the conditions of interpretation described had a positive coronary arteriogram. Had thallium myocardial perfusion imaging been used as a test of exclusion, almost 25% of patients with a negative stress electrocardiogram could have been spared the arteriographic procedure. Considering the relative costs in both financial and risk terms of coronary arteriography and myocardial perfusion imaging, considerable savings could be rendered if the imaging technique under high sensitivity conditions were used as a test of exclusion. If the prevalence of coronary artery disease in the population under study were lower, as might occur with application of screening techniques to population groups less likely to have coronary artery disease, then thallium myocardial perfusion imaging would
have an even more important role as a test of exclusion. For example, if prevalence of disease were 50”/;;,the positive predictive value of thallium myocardial imaging perfusion as ordinarily performed drops to 91”& whereas the negative predictive value rises only to 80%. On the other hand, when operating on the high sensitivity portion of the ROC curve, the negative predictive value at a prevalence of 50”/:, is still extraordinarily high (Fig. 2). In conclusion, the analysis indicates that the maximal utility for thallium myocardial perfusion imaging results from its application as a test of exclusion. This may be accomplished by the use of techniques which increase the sensitivity of the procedure, such as continuous background subtraction to enhance lesions. Under these circumstances, patients with an abnormal stress electrocardiogram would be referred for coronary arteriography. Patients who have normal or indeterminate stress electrocardiographic examinations would undergo thallium myocardial perfusion imaging testing. If the results of this test are normal, coronary arteriography is not indicated. In this sense, thallium myocardial perfusion imaging studies also act as a discovery test in the patients with normal or indeterminate stress electrocardiographic results. Considerable savings in financial and health costs result.
References HAMILTON G., TROBAUGH G., RITCHIE J. and WILLIAMS D. Cirdution 55, Suppl. 3, 140 (1977). OKADA R. D., RAE,Y,LER K. L., W~~LFENDEN J. M., PATTON D. D., GROVES B. M., HAGER W. D. and GOLDMAN S. Circulation 55, Suppl 3,
140 (1977). BLEND D. K., MCCARTHY D. M., SCIACCA R. R. and CANNON P. J. Circulation 55, Suppl 3, 229
(1977). BAILEY I. K., GRIFFITH L. S. C., ROULEAU J., STRAUSS W. and PITT B. Circuhtion 55, 79-87 (1977).
Work in progress
208 5.
ROSENBLAT? A., L~WENSTEIN J., KERM W. and HANDMAKER H. Am. Heart J. 94, 463470
(1977). 6.
7. 8.
10.
SONNEMAKER R. E., FLOYD J. L., NUSYNOWITZ M. L., BODE R. F., SPICER M. J. and WALISZEWSKI J. A. Radiology 131, 199-203 (1979). METZ C. E., ODENOUGH D. J., ROSSMANNK.
Radiohyy
9.
11.
109, 297-304 (1973).
FORTUIN N. J.and
699-712 (1977).
WEISS J. L. Circulation 56,
12.
R. Clin. Biostat. p. 221. C.V. Mosby, St Louis (1977). BORER J. S., BRENSIKE J. F., REDWOOD D. R., ITSCOITZ S. K., PASSAMANI E. R., STONE N. J..
FEINSTEIN, A.
RICHARDSONJ. M., LEVY R. I. and EPSTEIN S. E. N. Engl. J. Med. 293, 367-371 (1975). MCCONAHAY D. R.. MCCALLISTER B. D. and SMITH R. E. Am. J. Cardiol. 28, I-9 (1971). MCNEIL B. J., KEELER E. and ADELSTEIN S. J. N.
Enyl. J. Med. 293, 211-215 (1975).
review reported results of exercise-redistribution 201Tl myocardial perfusion imaging as a diagnostic test by analyzing its sensitivity, specificity, accuracy, and predictive values; (2) construct a receiver operating characteristic (ROC) curve based on reported data and on the results of our own series: (3) to define, from qualitative cost estimates, prevalence data, and information content analysis. the appropriate operating position on the ROC curve; and (4) to indicate from these results the appropriate role of thallium myocardial perfusion imaging as a diagnostic test.
!Mection of Patients for Coronary Arteriography with Thallium-201 : Operating Position on the Receiver Operating Characteristic Curve MARTIN L. NUSYNOWITZ,* ROBERTE. SONNEMAKER* and JOHN L. FLOYD* Nuclear Medicine Service, William Beaumount Army Medical Center, El Paso, TX 79920 and Division of Nuclear Medicine, University of Texas Health Science Center San Antonio, San Antonio, TX 78284, U.S.A.
Introduction IF ONE assumes that coronary bypass surgery improves the quality or quantity of life in patients with myocardial ischemia, and given that coronary arteriography is necessary in the selection of such patients for these surgical procedures, it becomes imperative to develop a strategy for screening for coronary arteriography. Such a strategy is necessary because coronary arteriography is expensive and has an associated morbidity and mortality. Heretofore, symptomatic patients with possible myocardial ischemic disease were screened on the basis of history, physical examination, routine laboratory testing and exercise stress electrocardiography. With these modalities, the selection process was excellent, since 75% of the usual population of patients studied arteriographically are found to have significant coronary artery disease. Recently efforts have been expended to define the role of 201TI myocardial perfusion imaging in this selection process. The purpose of this paper is to: (1)
Materials and Methods Five reported series consisting of thallium myocardial perfusion imaging and coronary arteriographic results in patients evaluated for myocardial ischemia were analyzed!1x5) The numbers of patients faliing into each of four categories-true positive, false positive, false negative and true negative-were tabulated using the arteriographic results as the standard. Although the criteria for arteriographic a$normalities varied somewhat in the various series (ranging from greater than 5OPi to greater than 70?:, narrowing of coronary artery diameter), decisions as to normality or abnormality of results for either arteriography or thallium myocardial perfusion imaging remained as reported. A similar tabulation was performed for the results of 36 consecutive patients studied by our group, in which processing of images by a continuously increasing background subtraction technique resulted in the enhancement of lesions such that sensitivity was 100%.‘6’ The data from the five reported series were analyzed both individually and as a group for sensitivity, specificity, accuracy and prevalence of coronary artery disease in the population. Bayes’ theorem was used to determine positive and negative predictive values at a prevalence of 750,; (the mean prevalence for the pooled data) using the following equations:
Positive Predictive
Value:
[P(D+)][Sensitivity] ‘@+
IT+)
= [P(D+)][Sensitivity] Negative
+ [l - P(D+)][l
Predictive
- Specificity]
Value:
[l - P(D+)] [Specificity] P(D-‘T-)
+ [P(D+)][l
= [l - P(D+)][Specificity]
* Present address: Martin L. Nusynowitz, Division of Nuclear Medicine, University of Texas Health Science Center San Antonio, 7703 Floyd Curl Drive, San Antonio, TX 78284, Robert E. Sonnemaker, Division of Nuclear Medicine, St Luke’s Episcopal, Texas Children’s Hospital, Houston, Texas, and- John L. Floyd, Nuclear Medicine Division, Presbyterian Hospital, Dallas, Texas.
(I)
- Sensitivity]’
(2)
From the sensitivity and specificity data, a receiver operating characteristic curve was drawn by visual interpolation through points defined by the sensitivity and specificity of each study and for the pooled data. Information content for each point was calculated by the method of Metz et. ~1.“) The ratios: positive predictive value to prevalence P(D+ ) T + )/P(D+ ) and the probability of disease, given a negative result, compared to prevalence P(D + I T - )/P(D + ), were also computed for evaluation of utility of each result. 204
Work
in proyrvss
RI2SUltS The calculated value for each parameter in each of the reported series and for the pooled data is given in Table 1. The prevalence of angiographically demonstrable significant coronary artery disease for the pooled data is 75% and is remarkably constant among the series (range 6878%). The overall sensitivity for the five pooled series is 727; (range 5693%); the specificity is 937; and the accuracy is 77%. The positive predictive value at a prevalence of 75 is 977; (range 91-lOO”/A), but the negative predictive value is only 525;. By comparison, the sensitivity of the grouped data for electrocardiography is only 51x, the specificity 88”~;,and the accuracy only 60% (Table 2). The positive predictive value with prevalence of 75 is 93% and the negative predictive value is 37%. These data are very similar to those found in eight other exercise ECG-coronary arteriographic studies reviewed by FORTUIN& WEISS.@) The ROC curve for the “‘Tl myocardial perfusion imaging data from the five series, the pooled value, and our series is plotted as Fig. 1. Note that four of the five series and the group value fall in the low sensitivity, high specificity (low false positive) region of the curve. In contrast to these results, the processed lesionenhanced results of our series showed a sensitivity of lob”,;, a specificity of 56% and the highest resultant accuracy of 89%. More importantly, although the positive predictive value is slightly lower than that of other studies (873&), the negative predictive value is loons& indicating that a normal lesion-enhanced thallium myocardial perfusion imaging study virtually excludes the possibility of a positive coronary arteriogram. The overall average of the ratio P(D+ I T+)/ P(D + ) for the combined ” ‘Tl data was 1.29, showing a slight gain of utility (29%) at a prevalence of 75%; the corresponding value for the authors’ series was 1.16. The ratio P(D + 1T - )/P(D + ) for the combined series was 0.63, differing from unity by 37% and indicating that test utility was greater when used to ex-
0
I
I
2
.4 False
positive
.6
.8
1.0
rote
FIG. 1. Receiver operating characteristic curve for 201Tl myocardial perfusion imaging. The open circle represents the mean values from 5 reported series.
205
206
Work in progress
Table 2. Results of stress ECG in detection
Reference
No. patients
Sensitivity
of myocardial
Specificity
Accuracy
ischemia Positive predictive value*
Negative predictive value*
Authors’ series
36
44
89
56
92
35
I:; (4) (5)
137 31 83 22
92 52 38 40
71 83 loo 88
87 59 53 55
91 90 100 89
37 74 35 32
309
51%
Combined * At prevalence
88%
93%
37%
= 75%.
elude coronary artery disease. The corresponding value from the authors’ series was 0.00, differing from unity by lOO%, indicating the greatest test utility. Discussion Diagnostic tests may be classified as tests of discovery, exclusion, or confirmation.t9) Discovery tests should be highly sensitive, even at the cost of specificity and inexpensive in both financial and health costs, so that a disease may be found if present. The stress electrocardiogram has long been employed as the basic discovery test for ischemic myocardial disease, but its shortcomings in this regard are well appreciated.t”’ Simply stated, abnormal results are good predictors of coronary artery disease (high positive predictive value), but normal or indeterminate results of stress electrocardiography as usually employed are meaningless. Because of its low sensitivity, normal findings of exercise electrocardiography are often associated with coronary artery disease. Even when a more liberal criterion of abnormality is employed with exercise electrocardiography (0.5 mm ST depression), sensitivity remains unacceptably low (63%) and specificity falls.” ‘) Thus, stress electrocardiographic testing does not fulfill the requirements for a discovery test. Thallium myocardial perfusion imaging has been studied as a screening test for coronary artery disease. It is obvious that the sensitivity of the test as generally employed, while far exceeding that of exercise electrocardiography, is still too low for use of the modality as a discovery test. If thallium myocardial perfusion imaging is interpreted in conjunction with an exercise electrocardiographic test so that an abnormal result is defined as either modality being positive, sensitivity rises to 90Y0,“’ a value approaching that of a useful discovery test in many circumstances, but marginal in the population ordinarily evaluated where prevalence of disease is already 75%. Bayes’ theorem indicates that the negative predictive value of a test is dependent on its sensitivity; the negative predictive value approaches 10% as sensitivity approaches lOOo/, (equation 2). Therefore, the
ideal test of exclusion possesses a sensitivity of 100%; in this case, a normal
60%
test excludes
the possibility
of
disease. In a confirmation test the cardinal feature is value specificity, since the positive predictive
approaches lOOD/, as specificity approaches 100% (equation 1). Coronary arteriography with its high specificity thus is the confirmation test for myocardial ischemic disease. Does thallium myocardial perfusion imaging have a role, then, in screening patients for coronary arteriography? The answer lies in the analysis of several factors. Firstly, ROC curve data show that most reported series operate on the left-hand, low sensitivity, high specificity region of the ROC curve. As a consequence, thallium myocardial perfusion imaging results are only a slightly better than exercise electrocardiographic findings; both are excellent in predicting arteriographically demonstrable disease when the test is positive, because of high specificity, but both are poor in excluding disease when negative. It has been reported that when both stress electrocardiography and thallium myocardial perfusion imaging are used in conjunction, the sensitivity increased to 90% approaching the values desired for discovery tests, but the negative predictive value is still too low for use of the combined procedure as an exclusion test. On the other hand, it can be demonstrated that “rT1 myocardial perfusion imaging is most useful if one chooses instead an ROC operating point so as to maximize sensitivity, even at a cost in specificity. MCNEIL ef u/.(‘~’ have emphasized that the optimal position on the ROC curve is that region where the slope K=
additional
cost FP
additional
cost FN
x-.
P(D-) P(D+)
It is readily seen that the slope is given by the product of two factors, one related to costs and the other to prevalence. The prevalence of coronary artery disease is remarkably constant at about 75% in the usual population referred for arteriography; thus, P(D-)/P(D+) = 0.25/0.75 = 0.33, and this fact indicates that the operating position should be in the low slope (high sensitivity, low specificity) portion of the curve. Furthermore, the cost of a false positive preliminary disgnostic test is relatively low compared to that of a false negative, since the former will result in an unnecessary coronary arteriogram whereas the latter, by preventing the performance of a necessary coronary arteriogram, might result in the diminution of the quality and/or quantity of life if bypass surgery
Work in progress
-t 0
I2
34
5
.6
7
8
.9
IO
207
I 0
I
I 2
I 3
I 4
I 5
I .6
I’I 7
0
I’ 9
I
3
P (D+)
P(D+)
FIG. 2. Stress ECG and “‘Tl myocardial perfusion imaging predictive values as a function of prevalence of coronary artery disease. The left hand graph shows positive predictive values and the right hand graph the negative predictive values. The arrow indicates a prevalence of 750.
is thereby denied. Since the ratio of costs is low, the low slope region of the ROC curve should again be employed. Another indication of operating position comes from comparison of information content of each operating point. At a prevalence of 0.75, the maximum information content occurs in using a high sensitivity (but low specificity) operating point. All other points give values for information content that are lower (Table 1). Therefore, the utility of thallium myocardial perfusion imaging from considerations of sensitivity and specificity as compared to stress electrocardiology, cost analysis, information content analysis and prevalence considerations lies in its value as a test of high sensitivity and thereby as an effective exclusion test. and The values of the ratios P(D+ 1T+)/P(D+) P(D + / T - )/P(D + ) provide additional confirmation. This high sensitivity can be achieved by processing manuevers designed to enhance lesions, such as those described.@) Employed in this manner, a normal thallium myocardial perfusion imaging result virtually excludes the diagnosis of coronary artery disease. In our series, 65”/, of patients with negative or indeterminate stress electrocardiograms had angiographically demonstrable coronary artery disease; the processed thallium images were positive in each of these. Not one patient with a negative thallium myocardial perfusion image under the conditions of interpretation described had a positive coronary arteriogram. Had thallium myocardial perfusion imaging been used as a test of exclusion, almost 25% of patients with a negative stress electrocardiogram could have been spared the arteriographic procedure. Considering the relative costs in both financial and risk terms of coronary arteriography and myocardial perfusion imaging, considerable savings could be rendered if the imaging technique under high sensitivity conditions were used as a test of exclusion. If the prevalence of coronary artery disease in the population under study were lower, as might occur with application of screening techniques to population groups less likely to have coronary artery disease, then thallium myocardial perfusion imaging would
have an even more important role as a test of exclusion. For example, if prevalence of disease were 50”/;;,the positive predictive value of thallium myocardial imaging perfusion as ordinarily performed drops to 91”& whereas the negative predictive value rises only to 80%. On the other hand, when operating on the high sensitivity portion of the ROC curve, the negative predictive value at a prevalence of 50”/:, is still extraordinarily high (Fig. 2). In conclusion, the analysis indicates that the maximal utility for thallium myocardial perfusion imaging results from its application as a test of exclusion. This may be accomplished by the use of techniques which increase the sensitivity of the procedure, such as continuous background subtraction to enhance lesions. Under these circumstances, patients with an abnormal stress electrocardiogram would be referred for coronary arteriography. Patients who have normal or indeterminate stress electrocardiographic examinations would undergo thallium myocardial perfusion imaging testing. If the results of this test are normal, coronary arteriography is not indicated. In this sense, thallium myocardial perfusion imaging studies also act as a discovery test in the patients with normal or indeterminate stress electrocardiographic results. Considerable savings in financial and health costs result.
References HAMILTON G., TROBAUGH G., RITCHIE J. and WILLIAMS D. Cirdution 55, Suppl. 3, 140 (1977). OKADA R. D., RAE,Y,LER K. L., W~~LFENDEN J. M., PATTON D. D., GROVES B. M., HAGER W. D. and GOLDMAN S. Circulation 55, Suppl 3,
140 (1977). BLEND D. K., MCCARTHY D. M., SCIACCA R. R. and CANNON P. J. Circulation 55, Suppl 3, 229
(1977). BAILEY I. K., GRIFFITH L. S. C., ROULEAU J., STRAUSS W. and PITT B. Circuhtion 55, 79-87 (1977).
Work in progress
208 5.
ROSENBLAT? A., L~WENSTEIN J., KERM W. and HANDMAKER H. Am. Heart J. 94, 463470
(1977). 6.
7. 8.
10.
SONNEMAKER R. E., FLOYD J. L., NUSYNOWITZ M. L., BODE R. F., SPICER M. J. and WALISZEWSKI J. A. Radiology 131, 199-203 (1979). METZ C. E., ODENOUGH D. J., ROSSMANNK.
Radiohyy
9.
11.
109, 297-304 (1973).
FORTUIN N. J.and
699-712 (1977).
WEISS J. L. Circulation 56,
12.
R. Clin. Biostat. p. 221. C.V. Mosby, St Louis (1977). BORER J. S., BRENSIKE J. F., REDWOOD D. R., ITSCOITZ S. K., PASSAMANI E. R., STONE N. J..
FEINSTEIN, A.
RICHARDSONJ. M., LEVY R. I. and EPSTEIN S. E. N. Engl. J. Med. 293, 367-371 (1975). MCCONAHAY D. R.. MCCALLISTER B. D. and SMITH R. E. Am. J. Cardiol. 28, I-9 (1971). MCNEIL B. J., KEELER E. and ADELSTEIN S. J. N.
Enyl. J. Med. 293, 211-215 (1975).