Author + information
- Received June 7, 1996
- Revision received February 19, 1997
- Accepted February 26, 1997
- Published online June 1, 1997.
- John A Rumberger, PhD, MD, FACCA,* (, )
- Patrick F Sheedy, MDB,
- Jerome F Breen, MDB and
- Robert S Schwartz, MD, FACCA
- ↵*Dr. John A. Rumberger, Department of Cardiovascular Diseases, Mayo Clinic and Foundation, 200 First Street SW, Rochester, Minnesota 55905.
Objectives. We sought to determine a range of cutpoints for coronary calcium scores measured by electron beam computed tomography (EBCT) in predicting the likely severity of associated angiographic coronary artery stenoses.
Background. EBCT can quantify coronary calcium and allow the estimation of atherosclerotic plaque burden, but use of the calcium score to define lumen narrowing is controversial.
Methods. A total of 213 patients (mean [±SD] age 50 ± 9 years) underwent coronary angiography and EBCT. Maximal percent diameter stenosis in any artery was paired with total coronary calcium score. Receiver operating characteristic (ROC) curve analysis was done using definitions of “disease” for maximal stenosis from ≥20% to 100%, and the corresponding score cutpoints were determined for 90% sensitivity, 90% specificity or “optimal” sensitivity and specificity.
Results. ROC curve areas ranged from a mean (±SE) of 0.91 ± 0.02 for ≥20% stenosis to 0.83 ± 0.03 for 100% stenosis. Optimal calcium score cutpoints consisted of nonoverlapping values ranging from 15 for ≥20% stenosis to 327 for 100% stenosis, whereas sensitivities and specificities ranged from 78% to 84%, depending on maximal stenosis severity. Calcium score cutpoints for 90% sensitivity and 90% specificity were also nonoverlapping and ranged from 3 and 27, respectively, for ≥20% stenosis to 154 and 945, respectively, for 100% stenosis; corresponding specificities and sensitivities ranged from 40% to 78%.
Conclusions. These data define the ranges for EBCT coronary calcium score cutpoints that predict the likely severity of associated maximal angiographic stenosis severity to a high sensitivity, high specificity or optimal sensitivity/specificity. These cutpoints potentially can be used in conjunction with clinical variables to predict the severity of lumen narrowing in patients undergoing assessment for coronary artery disease.
(J Am Coll Cardiol 1997;29:1542–8)
Coronary artery calcium is closely associated with mural atheromatous plaque ([1–4]), and electron beam computed tomography (EBCT) has been extensively evaluated regarding its noninvasive identification and quantification. A direct relation has been established between coronary artery calcium area, as measured by EBCT or histologic study, or both, and atherosclerotic plaque area on a heart by heart, vessel by vessel and coronary segment by segment basis ([5–7]). However, guidelines regarding measurements or estimates of coronary mural plaque as a standard for clinical disease severity have not been established. Currently, most clinicians gauge the extent of coronary artery disease according to measures determined at angiography, or they use surrogate estimates of disease severity based on results of noninvasive stress testing. Previous studies have shown a positive relation between the amount of calcified coronary plaque measured by EBCT (either using the coronary “calcium area” or “calcium score” []) and the anatomic severity of coronary lumen narrowing ([9–11]). However, there is not a one-to-one relation between plaque and percent stenosis. Clinicians are then faced with a dilemma when attempting to relate the “calcium score” with the likely severity of any associated coronary lumen stenoses. Therefore, application of positive results from EBCT coronary artery studies indicating how calcium scores may relate to the anatomic extent of coronary artery narrowing in the clinical setting has been controversial.
The purpose of the current study was to discover whether specific ranges of values, or “cutpoints,” regarding EBCT coronary artery calcium scores could be determined to predict with a high sensitivity or specificity, or both, the likely severity of associated coronary stenoses, as compared with direct coronary angiography. To accomplish this, results from 213 consecutive patients who had no previously documented coronary artery disease but who had undergone both angiography and EBCT were analyzed using receiver operating characteristic (ROC) curve analysis.
EBCT, also known as ultrafast computed tomography (CT) (Imatron C-100, C-150) and cine-CT, uses a stationary source/detector pair whereby X-rays are produced as a rotating electron beam is swept across a series of one to four semicircular tungsten targets. There are no moving parts to the imaging chain, and scanning through or at preset times during the cardiac cycle is possible in rapid succession. For identification of coronary artery calcium, consecutive, single-slice, 3-mm thick, 100-ms scans are performed. In this “high resolution” volume scanning mode, 40 sequential tomograms of the heart in synchrony with the heart rate during late diastole can be completed in as many cardiac cycles. Depending on the pulse rate, two serial breath holds may be required to acquire scans from the root of the aorta and the origin of the left main coronary artery through distal portions of the right coronary artery. In-plane spatial resolution (using a 300-mm field of view) is 0.4 mm2. Identification of coronary artery calcium requires no intravenous contrast injection because calcification of arterial walls demonstrates relatively high Hounsfield (H) densities, 2- to 10-fold greater than the surrounding soft tissue. Thus, the appearance of high density intramural coronary calcium deposits adjacent to low density soft tissue and surrounding fat made visual identification of calcium for determination of calcium content of the coronary arteries by EBCT relatively straightforward.
1.2 Patient Group.
To make a direct comparison between EBCT-derived coronary artery calcium and the likely severity of coronary angiographic disease, results from patients who underwent both EBCT scanning and direct coronary angiography at the Mayo Clinic were analyzed. The angiographic/EBCT study was approved by the local Institutional Review Board. The patient group consisted of men and women who had no previously documented history of obstructive coronary artery disease and were scheduled by their cardiologist for a diagnostic coronary angiogram. No patient had unstable angina or underwent angioplasty at the time of angiography. Results from subsets of this patient group have been published previously ([9, 10, 12]).
Individual patients were approached an average of 1 day after angiography for consent to perform an EBCT scan. Two-hundred thirteen consecutive patients (152 men and 51 women, average age 50 ± 9 years) were included in the analysis. One hundred sixty patients (75%) underwent angiography for assessment of chest pain; 25 (12%) had abnormal stress tests; 17 (8%) had unexplained heart failure; and 11 (5%) had a questionable history of previous infarction, pericarditis or unexplained dyspnea. Coronary angiography was performed by the Judkins technique with a minimum of five views of the left coronary system and two views of the right coronary system. The absence or presence of discrete coronary artery stenoses was visually assessed by two angiographers, and the maximal percent diameter stenosis in any epicardial coronary artery was defined as 0% to <20% (none or trivial), ≥20%, ≥30%, ≥40%, ≥50%, ≥60%, ≥70%, ≥80%, ≥90% or 100% stenosis. In case of a disagreement in the interpretation of a given coronary angiogram, it was read by a third cardiologist, and the maximal degree of stenosis was arbitrated.
1.3 EBCT Scan Analysis.
A number of studies have confirmed that coronary artery calcium detected by EBCT is a very sensitive test for coronary atherosclerotic plaque () and for “significant” coronary lumen disease (i.e., ≥50% diameter or ≥75% area stenoses) ([9, 10, 13]), with a negative predictive value of 90% or better ([9, 10, 13, 14]). However, the specificity of a positive test (any coronary calcification) for obstructive coronary artery disease has been considered too poor to advocate EBCT as a primary test to predict the severity of lumen disease. However, the coronary artery calcium content, or “score,” originally introduced by Agatston et al. (), is a continuous variable that may be more helpful in predicting the likely severity of lumen narrowing than merely the absence or presence of coronary calcium.
Each EBCT scan was analyzed in a fashion as previously discussed ([9, 10, 12]). For each study, a calcium score was determined using the methods of Agatston et al. (). This algorithm has been widely used in research and clinical studies. The calcium score is the product of the area of calcification (at least two contiguous pixels []) per coronary segment and a factor rated 1 through 4 dictated by the maximal calcium Hounsfield density within that segment. A calcium score was calculated for each epicardial coronary segment and recorded as a composite (i.e., total or summed) score for the entire epicardial coronary system (left main, left anterior descending, left circumflex and right coronary arteries).
ROC curve analysis was used as an extension of traditional sensitivity and specificity analyses ([9, 13, 15, 16]) to establish relations between total coronary calcium score measured by EBCT and maximal severity of coronary artery stenosis measured by angiography. Based on the analysis, an ROC curve was constructed that plotted the “true positive rate” (i.e., sensitivity) as the dependent y-variable for all measured calcium scores and the “false positive rate” (i.e., 1 − Specificity) as the independent x-variable at any given definition of disease (i.e., maximal angiographic lumen narrowing). ROC analyses were repeated with the same data set using a variable definition of disease based on maximal angiographic stenosis versus all values of total EBCT coronary calcium scores in the data set. Thresholds for angiographic lumen disease were ≥20%, ≥30%, ≥40%, ≥50%, ≥60%, ≥70%, ≥80%, ≥90% and 100% maximal stenoses.
ROC curve analysis was done using True Epistat () installed on a desktop computer. Data for individual maximal coronary stenoses determined by angiography and corresponding total EBCT calcium score for each patient were entered into parallel columns using a spreadsheet program and then imported in analysis of variance format into True Epistat. The statistical analyses were repeated for variable definitions of disease (i.e., maximal angiographic stenosis) as the reference standard with any coronary calcium score ≥0 as the test variable. ROC curve areas are reported as mean value ±SE (i.e., the estimated standard “error” of the calculated ROC curve area [trapezoidal rule] from the area that would have been obtained with an infinite sample []). The significance of any ROC curve area (A) from that of “chance” (area of 0.5 [[9, 13, 15, 16]]) involved calculation of the z statistic: (z = [A − 0.5]/SE) (); the corresponding p value was then determined from a look-up table.
Angiography demonstrated maximal coronary diameter stenoses from 0% to 100%, with 24% of the patients having normal coronary arteries (or at most trivial disease) and another 23% with mild to moderate disease (maximal stenosis ≥20% but <50%). Thus, the angiographic spectrum of patients was 47% with, at most, “nonobstructive” disease and 53% with ≥50% diameter stenosis. The distribution of results as a function of maximal coronary stenosis in any vessel and the number of patients within each angiographic category are given in Table 1. The arithmetic mean maximal diameter stenosis (±SD) was 52 ± 39%, and the median stenosis was 60%; the 25th percentile was 20% stenosis and the 75th percentile was 90% stenosis. Calcium scores for the 213 patients ranged from 0 to 4,091. The arithmetic mean calcium score was 440 ± 756, and the median score was 107; the 25th percentile was a score of 2 and the 75th percentile was a score of 541.
2.1 ROC Curves.
Individual ROC curves were generated for each decade of maximal angiographic disease from ≥20% to 100% stenosis in any segment of the major epicardial coronary arteries (Fig. 1). Each data point corresponds to the sensitivity and specificity of a particular calcium score for the specified definition of disease. An area under the ROC curve of 0.5 would imply equity between true positive and false positive test results (the equivalent of a random coin flip or chance). Curve areas >0.5 represent tests with increasingly greater diagnostic accuracy up to the “perfect” test (100% sensitive, 100% specific) at a curve area of 1.0. Individual ROC curve areas (±SE) are given in Table 2. All curve areas were statistically different (p < 0.000001) from a curve area of 0.5 (i.e., for the indeterminate test), suggesting a cogent application for coronary calcium scoring as a means to predict the likely severity of angiographic disease across all degrees of maximal lumen narrowing examined. Although all curves were consistent with a statistically valid use of EBCT calcium scores to predict variable angiographic disease, the individual z statistics (Table 2) showed that the strongest application was for predicting the presence of at least ≥20% stenosis at angiography (z = 20.50), and the weakest was for predicting the presence of at least 100% stenosis at angiography (z = 11.30).
2.2 “Optimal” Sensitivity and Specificity of Calcium Scores for Angiographic Disease.
The ROC curve data list sensitivity and specificity as a function of EBCT calcium score. The general shape of the ROC curves is such that sensitivity and specificity were inversely related; that is, for use of the EBCT coronary calcium score, regardless of the severity of corresponding angiographic stenoses, the test was most sensitive but least specific when the calcium score was low and least sensitive but most specific when the calcium score was high. “Optimal” calcium score, which corresponds to the point at which sensitivity andspecificity are maximum for a given degree of angiographic stenosis, can be defined as the maximal value of the sum of sensitivity and specificity across all calcium scores. Fig. 2demonstrates the graphic equivalent of this calculation. Plotted are individual calcium scores on the horizontal axis and corresponding sensitivities and specificities on the vertical axis. Fig. 2(top) plots part of the sensitivity and specificity data for maximal angiographic stenosis of ≥40%, whereas Fig. 2(bottom) plots a portion of the sensitivity and specificity data for angiographic stenosis of ≥70%. The calcium score at which the sensitivity and specificity data cross defines the “optimal” calcium score for predicting the corresponding associated degree of angiographic lumen narrowing. This calculation was performed for all tabulated ROC data, and Table 3gives the optimal calcium scores and associated sensitivities and specificities for each decade of angiographic narrowing examined in this patient group. Values ranged from 74% to 85%.
2.3 Variable Sensitivity and Specificity of Calcium Scores for Angiographic Disease.
As benchmarks to predict the likely severity of angiographic disease, the ROC tabular data were evaluated to determine ranges of EBCT calcium scores corresponding to 90% sensitivity (and associated specificity) and those scores corresponding to 90% specificity (and associated sensitivity). These data are given in Table 4.
Sensitivityand specificityare common terms used to define the applicability of noninvasive testing performed during the evaluation of cardiac patients. Mostly these refer to application of various “stress” tests, with or without adjunctive cardiac imaging, to predict the presence of at least “significant” coronary artery disease generally verified by direct comparison with maximal stenoses determined by coronary angiography. These tests are valid in many situations to predict the absence or presence of coronary stenoses of ≥50% or ≥70%, depending on the study, but are not reliable in predicting the absence or presence of maximal coronary lumen narrowing either below or above these thresholds. Based on the current analysis, the following conclusions can be stated regarding analogous use of quantitative EBCT coronary calcium scores. First, ROC curve analysis demonstrated that there exist ranges of values for EBCT calcium scores applicable to assessment of maximal angiographic coronary lumen narrowing from ≥20% stenosis (mild disease) through 100% stenosis. ROC curve areas ranged from a high of 0.91 for mild narrowing to a low of 0.83 for occlusive disease, but all were significantly different from chance. Second, a range of values for “optimal” coronary calcium scores can be determined as nonoverlapping cutpoints, which are associated with a high sensitivity andspecificity (>80% for both) for a range of coronary artery disease from mild, to moderate, to significant, and with modest sensitivities and specificities (74% to 78%) for critical and obstructive disease. Third, specific nonoverlapping cutpoint values for coronary calcium scores can also be determined, which are either highly sensitive (90%) or highly specific (90%) for a predefined range of associated angiographic lumen narrowing. Thus, based on the current analysis, quantitative assessment of coronary artery calcification and calcium “scoring” using EBCT can be used to predict, with variable but generally high sensitivity and specificity, the likely severity of anatomic coronary artery disease across a broad spectrum from nonobstructive to obstructive and occlusive coronary artery disease.
3.1 Clinical Interpretation and Application of EBCT Calcium Scores.
A negative EBCT scan (i.e., no detectable mural calcification) does not establish that coronary plaque is absent (), nor can it exclude the presence of a lipid-rich but unstable plaque. In fact, the area of calcium quantified by EBCT is roughly only one-fifth that of the total area of histologic plaque (). However, the negative predictive value of EBCT for fixed “significant” coronary lumen obstructive disease is very high, approaching 95% in some series ([9, 11]); moreover, up to two-thirds of men and women who have angiographically normal arteries have no calcium on EBCT examination (). A positive EBCT study (i.e., presence of quantifiable mural coronary calcium), in contrast, is 100% specific for atheromatous coronary plaque ([5, 7, 19]). The current investigation has attempted to expand these causal relations into calcium score guidelines for positive studies that may be used to predict coronary artery disease severity on an individual basis, using the more commonly used clinical estimation of disease severity based on maximal percent angiographic lumen stenosis.
It is our opinion that an EBCT scan for quantification of coronary artery calcium should function only as an adjunct to traditional evaluations or risk assessment, or both, and requires interpretation in the clinical context of the patient’s presentation. Only in this way will the absence, presence and amount of coronary artery calcification allow for a potential means to aid in the treatment of patients at risk for premature coronary artery disease () or with chest pain syndromes. Individuals without established or documented coronary artery disease can present for evaluation in three broad areas: 1) patients with new onset chest pain seen in the emergency department; 2) patients with symptoms of chest pain or unexplained dyspnea in the outpatient clinic; and 3) asymptomatic but “high risk” patients who may or may not have premature atherosclerotic disease. Each group presents a different diagnostic challenge. In particular, the patient in the emergency department may or may not have “significant” coronary artery disease, but urgent triage to the home or hospital or for further testing is required. In this case, a test that is very sensitive to the presence of advanced or significant coronary obstructive disease may be of value. The patient who has, at most, a stable angina syndrome and who is otherwise ambulatory may require testing that maximizes or “optimizes” both the sensitivity and specificity for significant disease. The asymptomatic but high risk patient, termed so because of the presence of two or more cardiovascular risk factors, would most likely be a candidate for a test with a very high specificity (i.e., a low false positive rate) in predicting the presence of moderate to severe coronary artery disease. EBCT assessments of coronary artery calcium, depending on the absence of calcium or the magnitude of the calcium score as examined in the current study, could be of value to the clinician in each of these patient groups, using ranges for coronary calcium scores that are highly sensitive or maximize sensitivity and specificity for obstructive disease, or are highly specific for moderate to severe lumen disease.
Using information from Table 4, the emergency department patient with a calcium score ≥37 would demonstrate at least 90% sensitivity for ≥50% coronary stenosis. Alternatively, as discussed earlier, a zero calcium score would give a negative predictive value of 90% to 95%. Thus, if no calcium was observed, the presence of ≥50% fixed stenosis would be remote. In contrast, a calcium score >0 but <37 may suggest that the patient needs a period of observation (perhaps in a chest pain unit), whereas a calcium score ≥37 might suggest the need for further noninvasive or invasive testing or admission to the hospital.
Using information from Table 3, the ambulatory but symptomatic patient with a calcium score ≥80 would demonstrate 84% sensitivity and 84% specificity for ≥50% stenosis. These values for sensitivity and specificity are comparable to conventional outpatient stress testing methods. However, the advantages of EBCT are that, if available, it can be done quickly (total testing time of 10 to 15 min), it is diagnostic even if the patient has an abnormal rest electrocardiogram or is concurrently taking cardiotonic medications (e.g., beta-blockers, digoxin) and the cost is approximately one-half that of stress echocardiography and one-third that of stress thallium scintigraphy ().
In a 10-year follow-up study of 521 patients who were normal or had, at most, nonobstructive coronary artery disease at the time of initial coronary angiography, Proudfit et al. () found an all-cause mortality rate of 6% for patients with <30% stenosis, compared with a 25% all-cause mortality rate for those with ≥30% but <50% stenosis. In addition, the incidence of any cardiac events (death, arteriographic progression to obstruction, myocardial infarction) over the next 10 years was 2.1% in those with normal angiograms, increasing to 13.8% with <30% stenosis, and to 33% in those with 30% to 50% maximal stenoses. Using information from Table 4, a patient would have 90% specificity for moderate anatomic (nonobstructive) coronary artery disease (i.e., ≥30% to ≥40% stenosis) with a calcium score between 89 and 110. Subject to the limitations discussed subsequently, this range of cutpoints could potentially be applicable for long-range prognostication to the “high risk” patient who had no previous history of disease. Such individuals may be candidates for targeted aggressive risk factor modification, including use of high priced lipid-lowering drugs, if appropriate. Recent investigations by Detrano et al. (), in angiographically studied patients, have suggested a threshold calcium score of >100 as an independent factor representing increased short-term (24 months) risk for a cardiac event, whereas Arad et al. () have shown a score of ≥160 to provide for increased short-term (19 months) risk in asymptomatic patients.
3.2 Study Limitations.
The analyses presented here are from only 213 patients who were highly selected for participation in our original study ([9, 12]), and the data given in Table 3andTable 4should be considered only as estimates of what may be found in a larger population. None of the patients had a previous diagnosis of coronary artery disease, and they were generally middle aged; thus, broad application to other age groups should be done with caution. Two recent reports from our laboratory, one a histopathologic study () and the other a clinical study (), have shown that EBCT coronary calcium has a similar predictive value in men and women when matched for the extent of coronary lumen narrowing (although there are still differences in the extent of disease in men and women at a given age). Thus, cutpoints for calcium scoring, as presented, are considered to be applicable to women as well as to men because women with any particular set of calcium scores tend to have the same amount of angiographic coronary disease as men with similar scores.
An additional issue is that comparisons with patients undergoing clinically indicated angiography could favor evaluation of patients with more advanced coronary disease and may not permit estimations of disease severity with all possible coronary calcium scores, owing to a lack of correlations performed across the entire anatomic spectrum of disease. However, in fact, the angiographic distribution of disease in the patients studied ranged from none or trivial coronary disease through nonobstructive, obstructive and occlusive disease. Nearly 50% of the patients had, at most, nonobstructive disease with a broad, albeit nonnormal, distribution across all subgroups above and below this threshold (Table 1). Twenty-four percent of patients had normal or nearly normal coronary arteries and 19% had occlusive (100% maximal stenosis) disease. Thus, the current study group is representative of a comprehensive range of anatomic disease severity, as might be seen in a group of middle-aged patients undergoing evaluation for cardiac disease.
Seventy-five percent of the patients examined had symptoms of “angina” before coronary angiography. Another possible limitation of the study could relate to the validity of predicting the lumen severity of disease using calcium scores from Table 3andTable 4in an asymptomatic rather than a symptomatic patient. However, definitions of coronary anatomy should not be confused with definitions of coronary physiology. The absence or presence of <50% or ≥50% lumen stenosis does not rule in or rule out inducible ischemia, it only indicates that it is less likely with the former and more likely with the latter. The presence of coronary calcium implies the anatomic presence of atherosclerotic plaque; it implies nothing about the physiologic response of the individual to the presence or amount of atherosclerotic plaque. A host of confounding factors can influence the absence or presence of symptoms such as diabetes, left ventricular hypertrophy and various cardiotonic medications. In fact, a recent report by Guerci et al. () has demonstrated a significant correlation between EBCT calcium score and maximal percent lumen stenosis in a small group of asymptomatic patients undergoing diagnostic angiography. Just the same, application of the calcium cutpoints reported in the current investigation for evaluation of asymptomatic patients must be done with caution and only in the context of additional clinical information.
Finally, there are limitations in the application of EBCT and calcium scoring. Although calculation of the total calcium score using EBCT is quantitative ([8–12, 24]) and operator independent (), reproducibility varies from excellent () to moderate (), depending on the laboratory. Thus, it is possible with application of EBCT that calcium scores above or below the values given in Table 3andTable 4may prove to be more appropriate in broad clinical practice. The calcium score values should be interpreted as ranges and not absolutes.
The current study has attempted to define ranges for EBCT-quantified coronary calcium scores that can be used to predict the likely severity of associated anatomic coronary disease as measured by direct coronary angiography. Potential applications of these various cutpoints, as determined from ROC curve analysis, are dependent on the clinical situation and the question that the clinician is trying to answer. These clinical situations then dictate the ranges for EBCT calcium scores employed as being highly sensitive, maximally sensitive and specific or highly specific for a predetermined or expected “disease” severity. The incremental value of EBCT calcium scanning in a given patient cannot be determined from the current data set, but these data form a foundation to pursue this issue to its next logical step. However, it must be emphasized that EBCT coronary calcium scores are not a substitute for coronary angiography, and these guidelines are meant to function only as adjuncts to traditional evaluations of patients suspected of having coronary artery disease.
☆ This study was supported by the Mayo Clinic and Foundation, Rochester, Minnesota and by Grant HL46292 from the National Heart, Lung, and Blood Institute, National Institutes of Health, Bethesda, Maryland.
- X-ray computed tomography
- electron beam computed tomography
- receiver operating characteristic
- Received June 7, 1996.
- Revision received February 19, 1997.
- Accepted February 26, 1997.
- The American College of Cardiology
- Blankenhorn DH,
- Stern D
- McCarthy JH,
- Palmer FJ
- Sangorgi G,
- Srivatsa SS,
- Staab M,
- et al.
- Rumberger JA,
- Simons DB,
- Fitzpatrick LA,
- Sheedy PF,
- Schwartz RS
- Agatston AS,
- Janowitz WR,
- Hildner FJ,
- Zusmer NR,
- Viamonte M,
- Detrano R
- Rumberger JA,
- Sheedy PF,
- Breen JR,
- Schwartz RS
- Budoff MK,
- Georgiou D,
- Brody A,
- et al.
- Simons DB,
- Schwartz RS,
- Edwards WD,
- Sheedy PF,
- Breen JF,
- Rumberger JA
- ↵True Epistat. 5th ed. Richardson, TX: Epistat Services, 1994.
- Beadenkopf WG,
- Daoud AS,
- Love BM
- Proudfit WL,
- Bruschke VG,
- Sones FM Jr..
- Detrano R,
- Tzung H,
- Wang S,
- et al.
- Arad Y,
- Spadaro LA,
- Goodman K,
- et al.