Chapter  26:  Evaluation of Technologies for Identifying Acute Cardiac Ischemia in Emergency Departments: Evidence Report/Technology Assessment Number 26

Applicability

We grouped the populations and settings of the studies using a four-category scale to help interpret the results. We also collected data about the prevalence of ACI or AMI to assist the interpretation. The four defined population categories are:

• Category I -- Studies that included all patients with signs and symptoms suggestive of ACI, such as chest pain, shortness of breath, jaw pain, acute pulmonary edema, and so forth. This is the most inclusive category.

• Category II -- Studies that used chest pain as the inclusion criteria. Most studies belong to this group. Category II is a subset of category I.

• Category III -- Studies that included patients with chest pain but excluded those with clinical or ECG findings of AMI. Many studies, especially studies of stress cardiac imaging or testing, belong to this group. Category III is a subset of category II.

• Category IV -- Studies in which all patients were hospitalized or which used additional criteria that enrolled highly selected subpopulations. We also placed retrospective studies in this category.

Test Performance Studies

When there were sufficient data for a technology, we used three complementary methods of synthesizing data across several studies to report on its test performance: summary receiver operating characteristics (SROC) analysis, separately combined sensitivity and specificity values using a random effects model, and the summary diagnostic odds ratios using a random effects model.

We defined a three-level methodologic quality scale for test performance studies graded as follows:

• A (least bias) -- such as a study that adheres to the traditionally held concepts of high-quality diagnostic evaluation, including clear descriptions of the population and setting; clear descriptions of the reference standard, the test under investigation, and the diagnostic criteria; masked interpretation of the reference and the test under investigation; verification of the diagnoses in all or most of the patients with negative results; and no significant reporting errors that are likely to result in substantial bias.

• B (susceptible to some bias) -- a study that does not meet all the criteria in category A but whose deficiencies are unlikely to cause major bias.

• C (likely to have significant bias) -- a study with significant design or reporting flaws that cannot preclude major bias. This category includes studies in which verification bias could be a major issue and studies that have significant amounts of missing information or discrepancies in their reporting.

Clinical Impact Studies

In the few instances where there are sufficient data reported by clinical impact studies, dichotomous outcomes expressed as risk ratio or continuous outcomes were combined using a random effects model.

We defined a three-level methodologic quality scale for clinical impact studies graded as follows:

• A (least biased) -- such as prospective controlled trials.

• B (susceptible to some bias) -- such as prospective cohort studies.

• C (likely to have significant bias) -- other designs or studies with significant conduct or reporting problems that could lead to large bias.

Study Size

The study (sample) size is used as a measure of the weight of the evidence. In general, a large study provides a more precise estimate of the treatment effect or test performance. Size alone, however, does not guarantee generalizability. A study that enrolled a large number of selected patients may be less generalizable than several smaller studies that included a broad spectrum of patient populations.

Applicability (population categories)

Figure 2. Population categories

   Figure 2. Population categories

Circles I, II, and III depict the relationships of the generally used inclusion criteria for patients in prospective studies in the ED examined in this evidence report. The circles are used to illustrate the relationships only, and the size of the circles is not meant to represent the actual quantitative relationship between the groups. Most of the studies examined in this evidence report belong to the settings "II" and "III." Circle IV represents a few reviewed studies that used additional selection criteria, are retrospective, or admitted all patients to the CCU.

Figure 2

Category I -- studies that included all patients with symptoms and signs suggestive of ACI, such as chest pain, shortness of breath, jaw pain, acute pulmonary edema, and so on. This is the most inclusive category. Few studies met category I criteria.

Category II -- studies that included possible ACI patients with a chief complaint of chest pain. Most studies belong to this group. Category II is a subset of category I.

Category III -- studies that included patients with chest pain but that excluded those with clinical or ECG findings of AMI. Many studies, especially studies of stress cardiac imaging or testing, belong to this group. The subjects in these studies were considered to be at "low risk" for AMI or ACI. Category III is a subset of category II.

Figure 2

Applicability (disease prevalence)

The prevalence of AMI or ACI is the most objective measure of the similarity of the study populations among the studies. Because different prevalence rates of the diagnoses may reflect the actual enrollment of different distributions of the disease spectrum, varying prevalence rates reported among studies may be related to the variability of reported diagnostic performance and clinical impact results. We recorded this information to assist the interpretation of the results.

Estimates of Diagnostic Test Performance

We used three complementary methods for assessing diagnostic test performance: SROC analysis, independently combined sensitivity and specificity values, and diagnostic odds ratios. Details about these methods are provided in the meta-analysis section later in this chapter.

Estimates of Clinical Impact

Several types of clinical outcomes were reported by the studies. Dichotomous data include overall mortality, number of ACI cases missed, and number of unnecessary hospital or CCU admissions avoided. Continuous data include the mean time to thrombolysis and the mean ejection fractions of intervention and control groups. As was the case for the earlier Working Group report, there is still a paucity of clinical impact studies. When clinical impact information is available, we summarized each of the clinical outcomes independently. Details of these methods are also provided in the meta-analysis section.

Quality Assessment of Internal Validity of Diagnostic Performance Studies

Internal validity refers to the design, conduct, and reporting of the clinical study. Proposals for evaluating the methodological quality of diagnostic test evaluation have been developed (Mulrow, Linn, Gual, et al., 1989; Irwig, Tosteson, Gatsonis, et al., 1994), but they have not been empirically evaluated. A recent article (Lijmer, Mol, Heisterkamp, et al., 1999) found that some of the traditional items of quality (e.g., unmasked interpretation, patient verification) did not have a large influence on the relative diagnostic odds ratio. The results of several other studies that evaluated quality scales also call into question the value of a single quality scale, including scales that may have been "validated" (Juni, Witschi, Bloch, et al., 1999; Clark, Wells, Huet, et al., 1999). Clearly, much more research is needed in quality assessment before a reliable tool becomes available. For the purpose of this evidence report, given the above caveats, we used a three-category scale to provide some indication of the methodological quality of the studies summarized:

Grade A (least bias) -- a study that mostly adheres to the traditionally held concepts of high quality diagnostic evaluation, including: clear description of the population and setting; clear description of the reference standard, the test under investigation, and the diagnostic criteria; masked interpretation of the reference test and the test under investigation; verification of the diagnoses in all or most of the patients with negative results; and no reporting errors that might hide substantial bias.

Grade B (susceptible to some bias) -- a study that does not meet all the criteria in category A. It has some deficiencies but none likely to cause major bias.

Grade C (likely to have significant bias) -- a study with significant design or reporting errors that cannot preclude major bias. This category includes studies in which verification bias could be a large issue and studies that have large amounts of missing information or discrepancies in reporting.

Quality Assessment of Internal Validity of Clinical Impact Studies

The internal validity of clinical impact studies refers to the soundness of the design, conduct, and reporting of the clinical trial. Some of the features of this dimension have been widely used in various "quality" scales, which usually include items such as concealment of random allocation, treatment masking, and the handling of dropouts. Clinical impact studies encountered in our report consist of both randomized trials and nonrandomized prospective studies. In this evidence report, we defined three categories of quality as follows:

Grade A (least bias) -- a controlled clinical trial (randomized or quasi-randomized) with only minor methodological problems and no reporting errors likely to hide substantial bias.

Grade B (susceptible to some bias) -- a well-designed and conducted prospective nonexperimental study design or a controlled trial with some methodological and reporting problems that may hide moderate bias.

Grade C (likely to have large bias) -- a study with major methodological or reporting problems that are likely to hide significant bias. This category includes studies with large amounts of missing information.

Presentation of Results for Each of the Technologies

In Chapter 3, the evidence for each of the technologies is presented in the format described below. We followed the structure of the NHAAP Working Group report and presented results of the prospective studies on test performance, then studies of clinical impact, and then data from other studies (such as those conducted in the CCU). When results are available, appropriate summary tables are included in each section. The summary table is preceded by a narrative description of the included studies. The tables list the qualifying studies, the number of patients, the study population category (as defined earlier), the prevalence of AMI or ACI in each study, the test performance (sensitivity and specificity) or clinical impact, and the methodological quality of the study. The overall results of test performance or clinical impact derived from meta-analyses are also shown. Data used for specific meta-analyses are described in the meta-analysis section of the evidence report. Studies conducted in other settings (e.g., the CCU) are described and summarized in the Data From Other Clinical Studies section.

Diagnostic Test Performance

We used three different methods to summarize the test performance of the diagnostic technologies: SROC curve analysis, separately averaged sensitivity and specificity values across studies, and the diagnostic odds ratio.

The SROC method assumes that the variability in the reported sensitivity and specificity values from different studies is due to different cutoff values being applied (Moses, Shapiro, and Littenberg, 1993). Each study provides a pair of sensitivity and specificity values to the analysis. It uses a regression method to fit a curve that best describes the data in the ROC space. We used the unweighted SROC method because it is probably less biased than the weighted regression method (Irwig, MacAskill, Glasziou, et al., 1995). If multiple thresholds are available for individual diagnostic test studies, ROC curves can be constructed and the areas under the curves can be estimated. The area under the curve provides an assessment of the overall accuracy of the test and allows comparisons with other tests. However, few studies provided results using multiple cutpoints.

The areas under different SROC curves can also be calculated and compared across technologies. However, the range of sensitivity and specificity values from studies in a meta-analysis of diagnostic tests is often limited, and extrapolation of the SROC analysis beyond the values of actual data is not reliable. For example, the specificity values reported by the CK-MB studies are typically between 90 and 100 percent. Thus, the SROC curve that can be constructed with actual data is limited to about the first 10 percent of the SROC space, and extrapolating the SROC curve to the entire SROC space to calculate the area under the curve would not be reasonable. Most of the technologies we examined have narrow reported ranges of sensitivity or specificity values. Therefore, we did not calculate the area under the SROC curve for any of the technologies.

When there is little variability in the test results -- studies appeared to be operating at similar thresholds and reported similar results -- SROC analysis provided little additional information. In this case, separately averaged sensitivity and specificity values across studies will give similarly useful summary information.

We combined the sensitivity and specificity values of the tests across studies using a random effects model to estimate the average values. A random effects model incorporates both the within-study variation (sampling error) and between-study variation (true treatment-effect differences) into the overall treatment estimate. It gives a wider confidence interval than the fixed effects model (which considers only within-study variability) when estimates are based on heterogeneous results.

When each is combined separately, sensitivity and specificity tend to underestimate the true test sensitivity and specificity. They are nonetheless useful estimates of the average test performance and provide an indication of the approximate test operating point for most of the studies. The appropriateness of this method can be verified by inspecting the location of the combined estimates and noting the distance of the estimates from the SROC curve. In our experience, the random effects-averaged sensitivity and specificity results are close to the unweighted SROC curve and well within the confidence intervals of each other. Average sensitivity and specificity results also serve as useful baseline test performance values for the decision and cost-effectiveness analysis.

The diagnostic odds ratio for a diagnostic test is defined as: [sensitivity/(1 - sensitivity)/(1 - specificity)/specificity] (Irwig, MacAskill, Glasziou, et al., 1995). A high diagnostic odds ratio typically has either a high sensitivity or a high specificity value, or both. The higher the odds ratio, in general, the greater the test accuracy. The summary diagnostic odds ratio obtained by combining the odds ratios of individual studies using a random effects model can provide a single summary value that is useful (with other summary information about the test) for comparing technologies. The diagnostic odds ratio is equivalent to the intercept coefficient of the SROC method assuming a zero slope.

Statistical analyses using the SROC curve method and combining sensitivity and specificity using the random effects model was performed using "Meta-Test" version 0.6. Summary diagnostic odds ratios were calculated using "Meta-Analyst" version 0.991. Both of these computer programs were developed by the EPC director (Dr. Lau) and are available to the public. Where necessary, statistical analysis algorithms implemented in MathCAD 7.0™ were also used. We report 95 percent confidence intervals (CIs) along with all estimates.

Clinical Impact Studies

Studies included in meta-analyses of clinical impact outcomes were combined using a random effects model (Laird and Mosteller, 1990). The risk ratio of the outcome was used to combine dichotomous outcome data, such as mortality. A random effects model was also used to combine continuous outcomes, such as differences in the mean time to thrombolysis. Summary risk ratios were calculated using "Meta-Analyst" version 0.991. Continuous outcomes were combined using the random effects model implemented in MathCAD 7.0.™ We report 95 percent CIs with all estimates.

All Studies

Figure 8

The setting of the study also does not appear to yield a consistent prevalence level. However, visually inspecting the settings in categories II and III studies (which have the most hospital- and CCU-restricted studies) appears to show that hospital-restricted studies have a greater prevalence than fully ED-based studies, and CCU-restricted studies have yet a greater prevalence.

However, within category II studies, differences between mean prevalence levels of different settings are all nonsignificant. ED-based studies had a mean prevalence of 24 percent (95 percent CI, 19-29 percent), whereas hospital-restricted studies had a mean prevalence of 29 percent (95 percent CI, 22-37 percent) (p=0.5, ED versus hospital) and CCU-restricted studies had a mean prevalence of 38 percent (95 percent CI, 0-92 percent) (p=0.2, ED versus CCU, and p=0.6, hospital versus CCU). Studies with prehospital settings have somewhat greater prevalence levels (32 percent, 95 percent CI, 20-45 percent) as ED-based studies (p=0.5).

A similar pattern is seen within category III studies. ED-based studies had a mean prevalence of 8.9 percent (95 percent CI, 4.1-14 percent); hospital-restricted studies had a mean prevalence of 21 percent (95 percent CI, 8.7-33 percent) (p=0.1, ED versus hospital); and CCU-restricted studies had a mean prevalence of 36 percent (95 percent CI, 0-100 percent) (p=0.006, ED versus CCU, and p=0.2, hospital versus CCU).

Biomarker Studies

Biomarker studies have a distribution of populations, settings, and prevalence levels similar to all the studies together. Although the information reported by these studies allowed them to meet the entry criteria for this report (studies of technologies tested in the ED for ACI that are not of highly selected patients), it is clear that many of these studies do, in fact, have fairly to very selected patient samples with high prevalence of AMI. In fact, 22 percent of studies showed levels of AMI prevalence lower than 10 percent; 44 percent had prevalence levels between 10 percent and 30 percent; 25 percent had prevalence levels between 30 percent and 50 percent; and 9 percent had prevalence levels above 50 percent.

Other Technologies

Among studies of protocols, predictive instruments, and decision aids, the large majority of studies were ED-based. Notably, with one exception (Gibler, Runyon, Levy, et al., 1995) which apparently excluded a large number of patients with AMI who bypassed the "Heart Emergency Room," all hospital-restricted studies reported AMI prevalence levels greater than similar ED-based studies. The two serial ECG studies, which were both ED-based and excluded patients with initial diagnostic ECGs, showed low prevalence. In contrast, the nonstandard ECG studies were all hospital- or CCU-based and generally showed greater prevalences. The prehospital 12-lead ECG studies found a wide range of prevalence levels, though those with broader inclusion criteria generally found lower prevalences of AMI.

Echocardiography and sestamibi studies reported generally lower prevalence levels. The two exceptions (Sasaki, Charuzi, Beeder, et al., 1986; Peels, Visser, Kupper, et al., 1990) excluded patients with previous AMI, coronary artery disease, or valvular or wall motion abnormalities. Studies of exercise tolerance tests, as would be expected, all found very low levels of prevalence as high-risk patients were excluded. However, the only dobutamine stress echocardiography study (Trippi, Lee, Kopp, et al., 1997), which was restricted to CCU patients, found a considerably greater prevalence of AMI.

Summary

In summary, among studies with generally similar descriptions of inclusion criteria, there remains a large variation as to which patients are included in study samples, as evidenced by the heterogeneity of AMI prevalence rates. Even though there was little evidence in the biomarker studies for a correlation between the prevalence of AMI and test performance, the large prevalence variation raises the question of unreported selection biases within the studies. It is frequently not clear when articles are reviewed why the prevalences of AMI for given studies vary. It is necessary for more articles to report the peculiarities of the populations from which patient samples are drawn and to better report additional biases in subject selection.

There is also some evidence that studies that restricted subjects to patients admitted to the hospital or, more specifically, to the CCU, do in fact report greater prevalences of AMI and, thus, may not be representative of all ED patients seen for the possibility of ACI.

Data From Prospective Clinical Studies in the ED Setting

Studies of test sensitivity and specificity

Table 5. Prehospital 12-lead ECG: Diagnostic test performance (more...)

Table 5. Prehospital 12-lead ECG: Diagnostic test performance studies

Study, year	Study size	Population category	Prevalence of disease (%)		Test performance		Study quality
					Sensitivity (%)	Specificity (%)
Aufderheide 1990	151	I	ACI AMI	40.3 15.9	90 54.2	53 99.2	A
Kudenchuk 1991	1,189	I	ACI AMI	46.1 32.9	- 66.2	- 94.8	A
Dalzell 1991	94	II	ACI AMI	91.5 51.1	78 77.1	88 97.8	B
Bertini 1991	605	II	ACI AMI	52.4 26.0	95 76.4	93 93.5	B
Aufderheide1 1992a, 1992b	439	I	ACI AMI	54.4 20.7	43 41.8	87 99.7	B
Arntz 1992	1,226	II	ACI AMI	- 34.0	- 64.3	- 99.5	B
Otto 1994	428	I	ACI AMI	60.0 23.4	- 60.0	- 80.8	B
Foster 1994	155	II	ACI AMI	- 13.5	- 81.0	- 100	B
Millar-Craig 1997	162	II	ACI AMI	66.0 44.4	- 98.6	- 41.1	B
Brown 1997	32	II	ACI AMI	- 21.9	- 100	- 80.0	C
Kudenchuk 1998	3,027	II	ACI AMI	53.3 38.0	45.92 -	- 96.02	A
Overall	7,508	I/II	ACI AMI	46-92 14-51	76 (54-89) 3 Odds ratio 68 (59-76) 3 Odds ratio	88 (67-96) 3 23.3(6.3-85) 3 97 (89-92) 3 104 (48-224) 3	B

1

Data from two publications of the same study.

2

ACI criteria - ST-segment elevation. Other criteria available; see study.

3

Results from meta-analyses using random effects calculations.

A total of 11 reports qualified for inclusion in the analysis of diagnostic accuracy. It should be noted that there were overlapping reports for two studies (Table 5). Three reports were derived from the same study population (Aufderheide, Keelan, Hendley, et al., 1992a; Aufderheide, Hendley, Woo, et al., 1992b; Otto and Aufderheide, 1994); the first two offer complementary interpretation on overall accuracy, and the third is a retrospective evaluation of the diagnostic accuracy of specific ECG changes. Another two reports stemmed from the Myocardial Infarction Triage and Intervention (MITI) trial (Kudenchuk, Maynard, Cobb, et al., 1998; Kudenchuk, Ho, Weaver, et al., 1991). The latter study contained data only on early screened patients, but the two reports provided complementary data on different diagnostic criteria and outcomes. Therefore eight studies are considered in the synthesis of the results.

The common characteristic of these eight studies was that they targeted populations with chest pain who required an ECG and did not have any significant exclusion criteria. Nevertheless, in Bertini, Rostagno, Taddei, et al. (1991), 8 percent of the hospitalized and 24 percent of the nonhospitalized patients had no data available and without specific reasons given. Similarly in Aufderheide, Keelan, Hendley, et al. (1992a) and Aufderheide, Hendley, Woo, et al. (1992b), of 680 enrolled patients, 149 had unsuccessful transmission of ECG by phone, 72 were transported to nonparticipating facilities, and 20 patients had "unavailable medical records." Finally, in the study by Millar-Craig, Joy, Adamowicz, et al. (1997), two phases were planned. During phase I, paramedics were trained and the prehospital 12-lead ECG was not used for decisionmaking, whereas in phase II, only those paramedics with accuracy over 80 percent in their phase I ECG interpretations participated. In phase II, the prehospital 12-lead ECG was used to decide whether the patient should be directly admitted to the CCU.

A brief mention should be made at this point of two excluded studies that also addressed the accuracy of chest pain protocols by paramedics. One study described 25 patients fast-tracked by paramedics to the CCU on the basis of prehospital 12-lead ECG findings (Banerjee and Rhoden, 1998). The diagnosis of AMI was verified subsequently in 14 of 25 patients, but no information is given on how many patients who were not fast-tracked to the CCU had AMI; therefore, the diagnostic characteristics cannot be determined. Another study (Wuerz and Meador, 1995) addressed the diagnostic accuracy of initiation of a chest protocol by paramedics (ECG strip monitoring, intravenous [IV] access, sublingual nitroglycerin, IV morphine) and found a sensitivity of 69 percent (260/376) and specificity of 87 percent (2,386/2,746) for ischemic heart disease. This report offers useful information of the diagnostic accuracy of the paramedics' first clinical impression, but it does not pertain to the accuracy of prehospital 12-lead ECG since only single-strip ECG was obtained.

Not considered for data synthesis are studies pertaining to chest pain protocols or computerized interpretations of ECG in the prehospital setting. This includes two studies where there was a computerized interpretation of the prehospital 12-lead ECG (Grijseels, Deckers, Hoes, et al.,1996; Aufderheide, Rowlandson, Lawrence, et al., 1996) as part of chest pain algorithms. The ACI-TIPI interpretation in Aufderheide, Rowlandson, Lawrence, et al. (1996) was performed retrospectively on 439 patients. Patients with low ACI-TIPI probability (0 to 9 percent) had a low incidence of angina (2.3 percent) and no AMI or life-threatening events. Grijseels, Deckers, Hoes, et al. (1995) considered a computer interpretation of a prehospital 12-lead ECG along with other parameters testing a number of hospital-developed algorithms in the prehospital setting and trying to develop a new algorithm suitable for prehospital triage. Details about this study can be found in the computer-based decision aids section.

Data on diagnostic accuracy for AMI on the basis of the overall ECG interpretation were available in all eight qualifying studies, and data on diagnostic accuracy for ischemia (including AMI and angina) were available from five of the eight qualifying studies. Three studies also provided diagnostic accuracy data for specific observed ECG changes. To avoid duplication of patients in the summary calculations, when more than one set of diagnostic accuracy data using different ECG criteria are reported in a study, only one set is used in the summary calculations, generally the one best approaching the definition of other studies.

Figure 9. SROC analysis of prehospital 12-lead ECG (more...)

   Figure 9. SROC analysis of prehospital 12-lead ECG to diagnose ACI

Table 5 summarizes the results of the prehospital 12-lead ECG studies considered for diagnostic performance. Figure 9 (see meta-analyses section) displays the SROC results of five studies to assess diagnostic performance of prehospital 12-lead ECG for ACI. Compared with AMI outcome, the diagnostic accuracy is inferior. Because of different criteria in the definition of coronary ischemia and different criteria in the definition of an abnormal ECG, there was significant heterogeneity on the sensitivity and specificity estimates in the five studies.

Figure 10. SROC analysis of prehospital 12-lead ECG (more...)

   Figure 10. SROC analysis of prehospital 12-lead ECG to diagnose AMI

Figure 10 (see meta-analyses section) displays the SROC result of the eight studies to assess the diagnostic performance of prehospital 12-lead ECG for AMI. As it can be shown in the figure, the Millar-Craig (1997) study operated at different sensitivity and specificity values compared with the other studies because it used very soft criteria for determining the need for direct admission to the CCU rather than for diagnosing AMI. The heterogeneity in the reported sensitivity may be due to different criteria being used in the studies used to define an abnormal ECG.

Three studies addressed the diagnostic accuracy of specific ECG changes. One study (Otto and Aufderheide, 1994) suggested that consideration of reciprocal changes in addition to ST elevation may be necessary to increase the positive predictive value to acceptable range (>90 percent) for consideration of early prehospital thrombolysis. The study by Kudenchuk, Maynard, Cobb, et al. (1998) also gathered data on the improved diagnostic performance of serial prehospital/hospital ECGs (discussed in the serial ECG section). Both studies show the anticipated tradeoff between sensitivity and specificity as more stringent ECG criteria are utilized. Dalzell, Purvis, and Adgey (1991) also provide data on specific ECG changes, but this seems to be a highly selected population with a very high prevalence of AMI.

Finally, one study (Kudenchuk, Ho, Weaver, et al., 1991) compared the performances of computer-interpreted and physician-interpreted prehospital 12-lead ECG. Computer-interpreted ECG in this study had better specificity (98 percent vs. 95 percent) and worse sensitivity (52 percent vs. 66 percent) than physician-interpreted ECG, and the overall performance was acceptable. Of course, these estimates depend on the stringency of the criteria set to recognize acute injury.

Prospective nonrandomized evidence

Prospective nonrandomized evidence was available from five different teams. The Israeli team that pioneered prehospital thrombolysis published successive publications on their experience of prehospital vs. hospital administration of thrombolysis with streptokinase. These include the following: Koren, Weiss, Hasin, et al. (1985); Weiss, Fine, Applebaum, et al. (1987); Rozenman, Gotsman, Weiss, et al. (1994, 1995); and Weiss, Leitersdorf, Gotsman, et al. (1998). For synthesis of the evidence, we used the latest update on hard endpoints, which is the 1995 report by Rozenman and colleagues. Compared with earlier reports, this 1995 report also included patients with significant systemic hypertension and patients older than 75 years. The Weiss, Leitersdorf, Gotsman, et al. (1998) report does not contain any data on the prespecified endpoints, but simply gives data on long-term evaluation of the ejection fraction in a subset of 362 of the accrued patients (prehospital n=68, hospital n=294), showing fewer symptoms of heart failure (including dyspnea, fatigue, orthopnea, nocturnal dyspnea, nocturia, peripheral edema, and episodes of pulmonary edema) over a followup of 4 years after presentation. This report, as well as the early reports, is not discussed again here.

One of the five studies we examined (Gibler, Kereiakes, Dean, et al., 1991) reported on only four patients who had received prehospital thrombolysis, and inferences are limited. Another study, the Cincinnati Heart Project (reported in Kereiakes, Weaver, Anderson, et al., 1990), addressed the impact of obtaining a prehospital 12-lead ECG on the time from reaching the hospital until initiation of thrombolytic therapy. Thrombolysis was not given in the prehospital setting. The 13 patients enrolled in the project had average delays of 36.3 (standard deviation [SD] 11.3) minutes, as compared with 62.9 (SD 14.7) minutes for 196 patients seen in the same facilities where the prehospital 12-lead ECG protocol was not applied, and 88.8 (SD 54.4) minutes in 211 historic controls from the same facilities. No data are available on time from onset of symptoms to thrombolysis.

All three nonrandomized studies that addressed the impact of prehospital thrombolysis used time from symptom onset to treatment as an endpoint and provided short-term mortality data. Two of the three (Roth, Barbash, Hod, et al., 1990; Rozenman, Gotsman, Weiss, et al., 1995) also addressed ejection fraction in the short term.

Table 6. Time to thrombolysis for prehospital 12-lead ECG: (more...)

Table 6. Time to thrombolysis for prehospital 12-lead ECG: Nonrandomized studies¹

Study, year	Study size		Time to thrombolysis (minutes [SD])		Study quality
	Prehospital	Hospital	Prehospital	Hospital
Roth 1990	74	44	94 (35)	137 (45)	B
Bouten 1992	226	220	100 (56)	166 (56)	B
Rozenman 1995	114	646	84 (48)	126 (60)	B
Overall	1,324		Mean time difference 1 −50.1 (−67.3, −34.2)		B

1

Results from meta-analysis using random effects calculations, 95 percent CI.

The results for the time from onset of symptoms to treatment are summarized in Table 6. Overall, in the over 400 patients given thrombolysis in the prehospital setting, the time gain was approximately 50 minutes compared with their respective controls receiving hospital thrombolysis.

Table 7. Ejection fraction outcome for prehospital 12-lead (more...)

Table 7. Ejection fraction outcome for prehospital 12-lead ECG/thrombolysis: Nonrandomized studies

Study, year	Study size	Population category	Ejection fraction percent (SD)		Study quality
			Prehospital	Hospital
Study team 1
Rozenman 1995	760	III	58 (13)	54 (15)	B
Weiss 1987	113	III	62 (8)	55 (14)	B
Koren 1985	51	III	No data	No data	B
Kereiakes 1990	209	III	No data	No data	B
Roth 1990	116	III	49 (17)	45 (19)	B
Bouten 1992	446	I	No data	No data	B
Overall	1,695	III	49-62 2	45-55 2	B

1

Study team based in Israel.

2

Range of reported values.

Table 8. Mortality outcome for prehospital 12-lead ECG/thrombolysis: (more...)

Table 8. Mortality outcome for prehospital 12-lead ECG/thrombolysis: Nonrandomized studies

Study, year	Study size	Population category	Mortality		Study quality
			Prehospital	Hospital
Study team 1
Rozenman 1995	760	III	5/114	11/645	B
Weiss 1987	113	III	3/34	No data	B
Koren 19852	51	III	-	2/53	B
Bouten 1992	446	I	0/226	6/220	B
Kereiakes 1990	209	III	No data	No data	B
Roth 1990	116	III	4/72	3/44	B
Overall	1,695	III	Risk ratio 3 0.84 (0.17, 4.15)		B

1

Study team based in Israel.

2

Cohort study.

3

Results from meta-analysis using random effects calculations, 95 percent CI.

Ejection fraction in the short term (in 1 week after admission in the 1995 study by Rozenman and colleagues) or upon discharge in the 1990 study by Roth and colleagues was better in the prehospital groups, but reached statistical significance only in the larger 1995 study. The magnitude of the difference was identical in the smaller 1990 study (4 percent). These data are based on 616 patients in the study by Rozenman and coworkers (1995) and 108 patients in the study by Roth and coworkers (1990) (Table 7). In-hospital mortality did not differ significantly between the two groups in any of the three studies, but data were very limited and some heterogeneity may be present. Summarized clinical outcomes are shown in Table 8 along with summary estimates based on random effects estimates. Random effects may be strongly indicated for the mortality outcome, since there was significant heterogeneity among the three studies.

Randomized evidence on clinical impact

A total of 14 randomized trials pertaining to the comparison of prehospital against hospital strategies were identified. One small trial (McNeill, Cunningham, Flannery, et al., 1989) was excluded, since the comparison was between thrombolysis in the CCU vs. thrombolysis in the ED or at home. The latter arm did not include only prehospital thrombolysis (several patients were treated in the ED). The results of this study are consistent with those of the other trials and would not affect the results overall. Eleven of the 13 qualifying trials compared prehospital vs. hospital administration of thrombolysis and typically considered populations of patients with short duration of symptoms (up to anywhere between 3 and 6 hours, when stated), age under 75, and no contraindications to thrombolysis, as perceived by each trial's investigators. The other two trials evaluated whether obtaining a prehospital 12-lead ECG might increase the time spent in the field (Karagounis, Ipsen, Jessop, et al., 1990) or may affect the time to administration of thrombolysis after admission to the hospital (Kereiakes, Gibler, Martin, et al., 1992). Karagounis and coworkers (1990) found that the 34 patients who were randomized to have a prehospital 12-lead ECG had identical in-field times to the 37 patients randomized not to have a prehospital 12-lead ECG (16.4 vs. 16.1 minutes on average). Prehospital thrombolysis was administered in six patients in the first group and thrombolysis was also given to six patients in the second group. Kereiakes and coworkers (1992) found in 11 patients randomized to have a prehospital 12-lead ECG a 15-minute reduction in the in-hospital time to thrombolytic treatment compared with 9 patients who did not have a prehospital 12-lead ECG.

Table 9. Time to thrombolysis outcome for prehospital 12-lead (more...)

Table 9. Time to thrombolysis outcome for prehospital 12-lead ECG: Randomized trials

Study, year	Study size		Time to thrombolysis (minutes)				Study quality
	Prehospital	Hospital	Mean (SD)		Median
			Prehospital	Hospital	Prehospital	Hospital
Castaigne 1989	57	43	ND	ND	131	180	A
Barbash 1990	43	44	96 (36)	132 (42)	ND	ND	A
Karagounis 1990	6	6	48 (12)	68 (29)	ND	ND	A
Schofer 1990	40	38	85 (51)	137 (50)	ND	ND	A
McAleer 1992	43	102	138	172	ND	ND	A
GREAT 19921	163	148	ND	ND	101	240	A
EMIP 1993	2,750	2,719	ND	ND	130	190	A
Weaver (MITI) 1993	175	185	92 (58)	120 (49)	77	110	A
Overall	6,562		Mean Time Difference 2 −33.2 −44.1, 22.3)				A

ND=no data. GREAT=Grampian Region Early Anistreplase Trial

1

Data from three publications of the same study.

2

Results from meta-analysis using random effects calculations of four studies that provided data, 95 percent CI.

The time from onset of symptoms to thrombolysis was mentioned as an outcome in eight trials and the results are summarized in Table 9.

As is evident, although it is not possible to arrive at exact summary estimates because of the heterogeneity in the way data are reported, nevertheless all studies show clearly that a significant reduction in the time to treatment is achieved, ranging between 20 and 60 minutes. The summary estimate would probably be closer to the (by far) largest study, the European Myocardial Infarction Project (EMIP, 1993), where the difference in the median time values was 1 hour. These results are in agreement with the data from the nonrandomized studies.

Table 10. Ejection fraction outcome for prehospital 12-lead (more...)

Table 10. Ejection fraction outcome for prehospital 12-lead ECG/thrombolysis: Randomized trials

Study, year	Study size	Population category	Ejection fraction percent (SD)		Study quality
			Prehospital	Hospital
Castaigne 1989	100	III	57 (ND)	53 (ND)	A
Barbash 1990	87	III	48 (15)	48 (15)	A
Karagounis 1990	71	III	ND	ND	A
Schofer 1990	78	III	51 (10) 3	53 (14) 3	A
GREAT1 1992	311	III	ND	ND	A
McAleer 1992	145	III	57 2	52 2	A
Kereiakes 1992	22	II	ND	ND	A
Weaver (MITI) 1993	360	III	53 (12)	54 (12)	A
EMIP 1993	5,469	III	ND	ND	A
Overall	6,643	III	48-57 4	48-54 4	A

ND=no data

1

Data from three publications of the same study.

2

Evaluated only in 45 percent of the enrolled patients (not stated how selected). In other trials, all patients had evaluation of ejection fraction.

3

Evaluated in 28 patients in each arm.

4

Range of reported values. Meta-analysis was not performed as it is obvious that there is no difference.

Table 11. Mortality outcome for prehospital 12-lead ECG/thrombolysis: (more...)

Table 11. Mortality outcome for prehospital 12-lead ECG/thrombolysis: Randomized trials

Study, year	Study size	Population category	Mortality		Study quality
			Prehospital	Hospital
Castaigne 1989	100	III	3/57	2/43	A
Barbash 1990	87	III	1/43	3/43	A
Karagounis 1990	71	III	No data	No data	A
Schofer 1990	78	III	1/40	2/38	A
GREAT1 1992	311	III	11/163	17/148	A
McAleer 1992	145	III	1/43	12/102	A
Kereiakes 1992	22	II	No data	No data	A
Weaver (MITI) 1993	360	III	10/175	5/185	A
EMIP 1993	5,469	III	266/2,750	303/2,719	A
Overall	6,643	III	Risk ratio 2 0.84 (0.73, 0.98)		A

1

Data from three publications of the same study.

2

Results of meta-analysis using random effects calculations, 95 percent CI.

Short-term effects on the left ventricular ejection fraction were reported in five trials. As shown in Table 10, in contrast to the results of the nonrandomized studies, there was no significant difference noted in any of these five trials, and a favorable trend was seen only in two trials. On the contrary, there were trends for reduced early mortality (in-hospital or up to 60 days) in six of the seven trials that addressed mortality, and the summary estimate shows a statistically significant 16-percent reduction in the risk of death with no heterogeneity between the seven trials (Table 11). The overall estimate of effect is practically identical to that obtained in the nonrandomized studies.

Long-term impact on mortality was assessed in four trial populations (Weaver, Cerqueira, Hallstrom, et al., 1993; Barbash, Roth, Hod, et al., 1990; McAleer, Ruane, Burke, et al., 1992; Grampian Region Early Anistreplase Trial [GREAT], 1992). There was substantial heterogeneity in the presented results. In the study by McAleer and coworkers, 1-year mortality rates were 6.1 percent vs. 20 percent and 2-year mortality rates were 9.1 percent and 30.6 percent, respectively, in the prehospital and hospital arms. Followup publications on the GREAT study (Rawles, 1994, 1997) showed also a survival benefit with mortality rates at 1 year of 10.4 percent vs. 21.6 percent, and at 5 years of 25 percent vs. 36 percent. On the contrary, 2-year survival rates in the much larger MITI trial were 89 percent and 91 percent for the prehospital and hospital groups, respectively (Weaver, Cerqueira, Hallstrom, et al., 1993). Three patients in each arm of the trial conducted by Barbash and coworkers (1990) died over 2 years of followup.

Data From Other Clinical Studies

No data are reported.

Data From Prospective Clinical Studies in the ED Setting

Studies of test sensitivity and specificity

Table 12. Continuous/serial ECG: Diagnostic performance studies (more...)

Table 12. Continuous/serial ECG: Diagnostic performance studies

Study, year	Study size	Population category	Prevalence of disease (%)		Test performance		Study quality
					Sensitivity (%)	Specificity (%)
Hedges 1992	261	III	ACI AMI	40.2 10.7	25 1 39.3	92 1 88.4	B
Gibler 1995	1,010	IV	ACI AMI	4.3 1.2	21.2 -	99.4 -	C
Overall	1,271	III/IV	ACI	4.3-40.2	21-25 2	92-99.4 2	C
					Odds ratio 3.8-45
			AMI	1.2-10.7	39.3 3	88.4 3
					Odds ratio 4.9

1

Only summary results were available for the ACI outcome.

2

Range of values reported. Meta-analysis was not possible because of inadequate data reporting.

3

Point estimate from single study.

Two studies in the population category III that examined the diagnostic performance of serial ECGs were included in this summary. Both of these studies were included in the earlier Working Group report and are summarized in Table 12. No qualifying new study was identified for this update.

Hedges, Young, Henkel, et al. (1992) tested the diagnostic performance of serial ECG testing in patients of whom 68 percent were a population of veterans. Thirty-eight percent (159/420) of the initial patients were not included because of incomplete data. It was unknown whether the interpretation of the reference standard was blinded to the results of the diagnostic test. Because of the lack of reporting of events in the ACI category, the reported summary sensitivity and specificity results could not be verified.

Gibler, Runyon, Levy, et al. (1995) reported on a retrospective study of a 9-hour protocol involving a series of diagnostic tests. The initial population of 1,010 consecutive patients completed both serial CK-MB and serial 12-lead ECG tests. The serial ECG was conducted at 20-second intervals over a 9-hour period. A computer monitor automatically compared subsequent ECGs with the initial baseline ECG. The definition of ischemia included angina as well as unstable angina. Other potential bias factors include 28 patients who were released against medical advice and 113 cocaine users, of whom 10 were admitted by serial ECG detection. Other tests in the 9-hour protocol, echocardiography and graded exercise testing, are discussed in the algorithm/computer-aided diagnostic section.

Data From Other Clinical Studies

No data are reported.

Data From Prospective Clinical Studies in the ED Setting

Data From Other Clinical Studies

Four diagnostic test performance studies met the inclusion criteria of nonstandard lead ECGs. Although the testing was conducted in the ED setting for all studies, in three studies, all patients enrolled were subsequently admitted to the hospital (one study specifically stated cardiac monitored beds). In Justis and Hession (1992), of the 188 patients who had complete data, 163 were admitted. There were no common technologies between the four studies that could be synthesized. They include 15-, 18-, 22-lead ECGs, and a 24-lead variance cardiography.

Justis and Hession (1992) tested the diagnostic performance of the 22-lead ECG within 3 hours of presentation on a prospective cohort in an urban hospital. The 22-lead ECG uses digital cardiac electrogram technology, with each lead providing QRS data. Final diagnosis for AMI was based on CK-MB levels. An index of greater than or equal to 85 was used for positive test results. Because of discrepancies in the reporting of the data in the text as well as in the tables, sensitivity and specificity could not be verified. Enrollment consisted of 212, of whom 24 had incomplete 22-lead ECG data. One hundred eighty-eight patients, not 163 as reported, were the evaluable study size. Twenty-five patients were discharged, 24 with noncardiac chest pain or at low risk of AMI and 1 returned and had AMI confirmed. All 163 patients were admitted for serum enzymes and angiography for AMI and coronary artery disease (CAD), respectively.

Table 13. Nonstandard lead ECG: Diagnostic performance studies (more...)

Table 13. Nonstandard lead ECG: Diagnostic performance studies

Study, year	Study size	Population category 1	Prevalence of disease (%)		Test performance		Study quality
					Sensitivity (%)	Specificity (%)
Justis 1992	163	IV	AMI	21.8	83	76	B
Zalenski 1993	149	IV	AMI	22.8	58.8	93.0	B
Spadafore 1996	52	IV	ACI AMI	48.1 25	96.0 -	40.7 -	B
Zalenski 1997b	533	IV	AMI	64.7	66.1	84.0	A
Overall	897	IV	ACI	48	96 2	41 2	B
					Odds ratio 17 2
			AMI	22-65	59-83 3	76-93 3
					Odds ratio 10-19 3

1

All patients admitted.

2

Point estimate from single study.

3

Range of values reported. Meta-analysis is inappropriate because of the different technology used in each study.

Spadafore, Lieber, and Vasilenko (1996) tested the performance of 24-lead variance cardiography (VC) in a community hospital with a prospective cohort selected for hospital admission. The VC is a 12-lead ECG with 12 auxiliary leads capable of inputting computer data, calculating QRS morphologic variability. The derived index of greater than or equal to 75 was selected as positive for ACI. The VC was tested in the ED with the physicians blinded to the VC results. The final diagnosis for ACI included characteristic clinical presentation, ECG findings, plus CK and CK-MB levels. Several limitations to be considered, as noted by the authors, are: VC does not appear to be able to differentiate between AMI and UAP patients, all AMI patients were male, and patients positive on their initial 12-lead ECGS were also included in the study.

Zalenski, Rydman, Sloan, et al. (1997b) tested an 18-lead ECG in seven emergency departments of public and private hospitals, including teaching hospitals. The 18-lead ECG included 12-lead ECG plus 3 posterior leads, V₇, V₈, V₉, and 3 right ventricular leads, V_4R, V_5R, V_6R. The single cardiologist reading the initial 12-lead ECG for final diagnosis and 18-lead ECG was blinded to the test results and the patient's outcome. Final diagnosis included CK and CK-MB results or pathologic Q waves. Analysis between the included and excluded patients (66 of 70) showed no difference for gender and race, but the excluded patients were younger, 61.1 vs. 65.9 years.

Data From Prospective Clinical Studies in the ED Setting

One new study (Kirk, Turnipseed, Lewis, et al., 1998) and two reports (Tsakonis, Shesser, Rosenthal, et al., 1991; Kerns, Shaub, and Fontanarosa, 1993) previously used by the Working Group are included in our evidence summary. However, two studies (Gibler, Runyon, Levy, et al., 1995; Zalenski, Roberts, Das, et al., 1994) mentioned in the Working Group report are not included in the current report. The study by Gibler and colleagues (1995) is a retrospective study that enrolled 1,010 patients at the outset; the protocol of this study also resulted in a highly selected patient population that received exercise stress ECG. The study by Zalenski and colleagues (1994) is an abstract that has not yet been published as a full article and therefore not used in our report.

Studies of test sensitivity and specificity

Table 14. Exercise stress ECG to diagnose ACI: Diagnostic performance (more...)

Table 14. Exercise stress ECG to diagnose ACI: Diagnostic performance studies

Study, year	Study size	Population category	Prevalence of disease (%)		Test performance		Study quality
					Sensitivity (%)	Specificity (%)
Kirk 1998	212	III	ACI AMI	6.1 1.4	100 -	92.5 -	A
Lewis 1999	100	III (ED/CPU)	ACI 10	AMI 2	70 -	82.2 -	C
Overall	312	III	ACI	6-10	70-100 1	82-93 1	B
					Odds ratio 11-? 2

CPU=chest pain unit.

1

Range of values reported.

2

Upper range cannot be estimated because of study with 100 percent sensitivity.

Kirk, Turnipseed, Lewis, et al. (1998) studied a prospective cohort to determine the safety and utility of exercise stress testing (Table 14). The intent was to include a large, heterogeneous, low-risk population. Two hundred and twelve patients were enrolled and underwent the modified Bruce protocol. Of the 28 patients testing positive, 3 had myocardial infarction and 10 had unstable angina. There were 59 nondiagnostic and 125 negative results for a specificity of 92.5 percent.

Lewis, Amsterdam, Turnipseed, et al. (1999) studied patients with known cardiac arterial disease. The modified Bruce protocol was performed on 100 patients with a sensitivity of 70 percent and specificity of 82.2 percent. Two concerns that the study design may result in bias are first, that the study sample did not consist of consecutive patients but was based on physician referral, and second, that for the diagnosis of unstable angina, the test under study was used as part of the criteria.

Studies of the clinical impact of the test's actual use

Table 15. Exercise stress ECG: Clinical impact studies (more...)

Table 15. Exercise stress ECG: Clinical impact studies

Study, year	Study size	Population category	Prevalence of disease (%)		Clinical impact	Study quality
Tsakonis 1991	28	III	AMI ACI	0 0	No cardiac event	C
Kerns 1993	32	III	AMI ACI	0 0	No cardiac event	C
Kirk 1998	212	III	ACI AMI	6.1 1.4	Unknown 1	B
Overall	272	III	ACI AMI	0-6.1 0-1.4	Unknown	Not applicable 2

1

This study has no comparison arm.

2

Not applicable because of unknown clinical impact.

Tsakonis, Shessee, Rosenthal, et al. (1991) looked at the feasibility and safety of emergency cardiac treadmill exercise stress testing (Table 15). A convenience sample of 28 patients with normal ECG results was enrolled. They were also likely candidates for hospitalization for coronary artery disease. The patients underwent a modified Bruce protocol, of which 23 tested negative. Four positive test results were concluded to be false positives, and the one patient with a positive test refused further workup and was lost to followup. A 1- to 12-month followup found no cardiac events.

Kerns, Shaub, and Fontanarosa (1993) looked at the feasibility and safety of cardiac treadmill exercise stress testing. Compared were two groups of patients -- prospective patients presenting to the emergency department and retrospective inpatient controls. Both groups underwent the Bruce protocol stress test with normal test results. The controls had a "primary discharge diagnosis of atypical or noncardiac chest pain." The emergency patients were followed for 6 months, and no cardiac events were experienced.

In the Kirk, Turnipseed, Lewis, et al. (1998) study described in the last section, of the 212 patients, 125 (59.0 percent) tested negative and were discharged, and 59 patients (27.8 percent) had nondiagnostic tests. Of the 28 patients who tested positive, 3 had AMI and 10 had unstable angina (UA). There was no mortality or morbidity for 95 percent (205) of patients followed for 30 days.

Data From Other Clinical Studies

No data are reported.

Data From Prospective Clinical Studies in the ED Setting

A total of 47 studies were assessed for the diagnostic accuracy of either CK or CK-MB in the diagnosis of AMI in the ED. Of these, 24 have been published since 1995. Twenty-one studies included all eligible ED patients and qualified for inclusion in meta-analysis. Of these 21 studies, 16 have been published since 1995. Ten included studies addressed CK as a single test upon admission to the ED; only two addressed serial testing of CK. Ten included studies addressed CK-MB as a single test upon admission to the ED; seven addressed serial testing of CK-MB. A number of studies analyzed both CK and CK-MB and/or presentation and serial testing. Fourteen additional studies included patients evaluated in the ED but excluded patients who were discharged from the ED.

The assessments of CK and CK-MB are presented in the following order:

• Single CK upon presentation to ED to diagnose AMI or ACI.

• Serial CK upon presentation to ED to diagnose AMI or ACI.

• Single CK-MB upon presentation to ED to diagnose AMI or ACI.

• Serial CK-MB upon presentation to ED to diagnose AMI or ACI.

Single CK Upon Presentation to ED to Diagnose AMI

Studies of test sensitivity and specificity

Table 16. Presentation creatine kinase to diagnose AMI: (more...)

Table 16. Presentation creatine kinase to diagnose AMI: Diagnostic performance studies (ED studies)

Study, year	Study size	Population category	Prevalence of AMI (%)	Test performance		Study quality
				Sensitivity (%)	Specificity (%)
Eisenberg 1979	80	II	16	54	82	C
Roxin 1984	305	I	26	28	96	C
Lee 1987	639	II	16	38	80	C
Hedges 1987	773	II	6.6	55	65	B
Viskin 1987	252	II	30	38	93	C
Mair 1991a	96	II	24	35	89	C
Mair 1991b	126	II	40	43	83	C
Thomson 1995	383	II	18	42	93	C
Gornall 1996	98	III	41	28	86	B
Hetland 1996	133	II	34	7.1	93	B
Laurino 19971	115	III	22	16 1	ND	C
Overall2	2,885	I/II/III	6.6-41	36 (29-44) 3	88 (80-93) 3	C
				Odds ratio 3.7 (2.5-5.4) 3

1

Study not included in meta-analysis.

2

Laurino (1997) not included in overall summary because the specificity was not reported.

3

Results from meta-analyses using random effects calculations.

Table 17. Presentation creatine kinase to diagnose AMI: (more...)

Table 17. Presentation creatine kinase to diagnose AMI: Diagnostic performance studies (all patients admitted)

Study, year	Study size	Population category	Prevalence of AMI (%)	Test performance		Study quality
				Sensitivity (%)	Specificity (%)
Tucker 1994	133	IV (Hospital admission)	29	51	82	C
Tucker 1997	177	IV (Hospital admission)	15	30	84	B
Young 19931	222	IV (CCU admission)	19	40	No data	C
Overall	532	IV	15-29	30-51 1	82-84 1	C
				Odds ratio 4.7-12 1

1

Range of reported values.

A total of 18 studies pertaining to the study of CK as a single test upon presentation to the ED for the diagnosis of AMI were retrieved. One study (Mair, Artner-Dworzak, Lechleitner, et al., 1992) presented the same data set as a previously published study. Two studies (Young and Green, 1993; Laurino, Pelletier, Eadry, et al., 1997) reported only sensitivity results. These two studies are presented in Tables 16 and 17 and Evidence Tables 5a and 5b. One study (Mair, Genser, Morandell, et al., 1996) reported only sensitivity results from a population that includes patients selected for having AMI. Two studies (Katz, Irwig, Vinen, et al., 1998; Ohman, Casey, Bengston, et al., 1990) analyzed their data with logistic regression using a continuous CK variable. Thus 12 data sets were available for meta-analysis. Two studies included only patients who were admitted to the hospital after ED evaluation; thus, only 10 of these studies (published between 1979 and 1996) included patients seen in the ED, including patients both admitted to the hospital and discharged from the ED.

The study by Eisenberg, Horowitz, Busch, et al. (1979) included patients over 30 years with chest pain seen at an urban ED. Patients without AMI were underrepresented as no data were presented on 29 additional patients who met the enrollment criteria but did not return for followup evaluation. No demographic information was reported. Of note, no description of test measurement or definition of a positive CK test was given.

The study by Roxin, Cullhed, Groth, et al. (1984) included patients "referred to [an urban] hospital on suspicion of AMI." No information was given on loss to followup. Of note, of the 305 patients included, 147 were not included in the reported test performance analysis because 22 had a final diagnosis of "possible AMI" (with atypical ECG changes or elevated aspartate amino transferase [AST] levels in only one blood sample) and for 125, AMI "could not be entirely excluded" (patients had chest pain, shock, or pulmonary edema with no other cause, but no ECG changes or elevations of AST). For the purposes of meta-analysis, all patients not meeting WHO criteria (including those with possible or "cannot rule out" AMI) were classified as not having AMI.

The study by Lee, Weisberg, Cook, et al. (1987) included patients over 30 years seen in an urban ED with a chief complaint of chest pain. Patients without AMI were underrepresented, since discharged patients who did not have followup testing were excluded from analysis. In total, 416 of 1,055 patients were excluded.

The study by Hedges, Rouan, Toltzis, et al. (1987) included patients over 30 years seen in an urban ED with chest pain. The study reported that 97 of 861 patients were excluded because a CK at presentation was missing; however, 773 patients were included in the analysis. Description of followup of discharged patients implied no loss to followup.

The study by Viskin, Heller, Gheva, et al. (1987) included patients over 25 years seen in an urban ED with chest pain. Of 300 eligible patients, 48 were excluded primarily because of incomplete ED or followup data.

One study by Mair, Artner-Dworzak, Lechleitner, et al. (1991a) included patients seen in an urban ED. No data were provided as to inclusion or exclusion criteria, loss to followup, or demographics.

Another study by Mair, Artner-Dworzak, Dienstl, et al. (1991b) included patients seen in an urban ED with a chief complaint of chest pain. No data were provided for loss to followup or demographics.

The study by Thomson, Gibbons, Smars, et al. (1995) included patients over 20 years seen in an urban ED with anterior or left lateral chest pain. Patients with recent infection, steroid use, cancer, gastrointestinal bleed, surgery, hemodialysis, or cardiopulmonary resuscitation were excluded. Only 3 of 20 patients who were discharged from the ED had followup tests and were evaluated.

The study by Gornall and Roth (1996) included patients seen in a suburban ED with chest pain. Patients with ECGs diagnostic of AMI at presentation or who had noncardiac diseases diagnosed by history and physical at presentation were excluded. No data were reported on possible loss to followup.

The study by Hetland and Dickstein (1996) included patients seen in an urban ED with "chest discomfort suggestive of AMI" for fewer than 6 hours. Patients who arrived in the ED from 10 p.m. to 8 a.m. were not included. No data were reported on possible loss to followup.

In summary, few reports gave detailed descriptions of inclusion or exclusion criteria. None described the ED or hospital setting. Eight studies primarily included only patients with chest pain. The prevalence of AMI in these studies ranged from 7 percent to 34 percent. One study had more restrictive inclusion criteria (Gornall and Roth, 1996), which excluded patients who clearly did not have ACI, and had the highest AMI prevalence, at 41 percent. One study (Mair, Artner-Dworzak, Dienstl, et al., 1991a) did not report inclusion criteria, but apparently included highly selected patients, as the prevalence of AMI was relatively high at 40 percent.

In general, AMI was defined by WHO criteria or variations thereof. In most, serial CK or CK-MB was used to define AMI. Only one (Roxin, Cullhed, Groth, et al., 1984) explicitly did not use CK to define AMI, but used serial AST instead. The studies used a variety of thresholds for abnormal CK, ranging from 80 U/L to 336 U/L for men and 70 U/L to 326.5 U/L for women. One study (Eisenberg, Horowitz, Busch, et al., 1979) did not report the threshold used.

Figure 11. SROC analysis of presentation CK to diagnose (more...)

   Figure 11. SROC analysis of presentation CK to diagnose AMI (ED only studies)

Table 16 and Figure 11 (see meta-analysis section) show a summary of included articles, results of meta-analysis, and unweighted ROC analysis. Heterogeneity among studies (primarily differences in test sensitivity levels) was not explained by CK thresholds (definition of test positivity), reported differences in eligibility criteria, study setting, or prevalence of AMI. Across studies, there were insufficient data to allow full analysis of symptom duration prior to testing; however, those studies that analyzed test performance by symptom duration (Hetland and Dickstein, 1996; Laurino, Pelletier, Eadry, et al., 1997; Lee, Weisberg, Cook, et al., 1987; Mair, Artner-Dworzak, Dienstl, et al., 1991b; Viskin, Heller, Gheva, et al., 1987) all found increased sensitivity of presentation CK in patients with longer duration of symptoms.

Studies of clinical impact of the test's actual use

No studies are reported.

Data From Other Clinical Studies

Two studies included patients evaluated in the ED who were subsequently admitted to the hospital, thus excluding patients discharged from the ED.

The first study by Tucker, Collins, Anderson, et al. (1994) included patients who were seen in a large urban community ED with chest discomfort or other symptoms prompting ordering cardiac enzymes and who were subsequently admitted to the hospital. Patients were excluded if their symptoms had persisted for more than 24 hours. Of 193 patients presenting to the ED, 30 were excluded because they were discharged from the ED; 17 had more than 24 hours of symptoms; 6 had improper laboratory test collection; 6 had no consent; and 1 was transferred. No information was reported on method of diagnosing AMI, level of abnormal CK, or patient demographics.

The second study by Tucker, Collins, Anderson, et al. (1997) included patients who were seen in an urban ED with chest discomfort of less than 24 hours and who were subsequently hospitalized. Patients with initial diagnostic ECGs or who received cardiopulmonary resuscitation were excluded. No data were reported on patients not analyzed. CK was defined as abnormal if activity was greater than 170 U/L in men or 135 U/L in women.

In summary, these two studies had similar AMI prevalence levels (at 29 percent and 15 percent, respectively) as the 10 ED studies. Like most ED studies, the second study by Tucker and colleagues (1997) included about two-thirds men. No demographic data were reported in the first study by Tucker and colleagues (1994).

Tucker and coworkers (1997) used WHO criteria with serial CK and CK-MB to diagnose AMI; and Tucker and coworkers (1994) did not report method of AMI diagnosis. The definition of abnormal CK was similar to that in the ED studies by Tucker and coworkers in 1997. Again, Tucker and coworkers (1994) did not report the definition of abnormal CK.

Including these two studies in the analysis yielded 3,195 evaluable patients, addressing the diagnostic accuracy of CK as a single test done at presentation to the ED. The random effects model yielded a pooled sensitivity of 37 percent (95 percent CI 31 percent, 44 percent) and a pooled specificity of 87 percent (95 percent CI 80 percent, 91 percent). The random effects model odds ratio is 3.9 (95 percent CI 2.7, 5.7).

Serial CK Upon Presentation to ED to Diagnose AMI

Studies of test sensitivity and specificity

Four studies pertaining to the study of CK in serial testing in the ED setting for the diagnosis of AMI were retrieved. One study (Tucker, Collins, Anderson, et al., 1997) did not report sufficient data as to the numbers of patients included at each time point to allow for complete data extraction. In addition, it was not clear from the text whether the presented test performance values over time represented cumulative, serial data. One study (Katz, Irwig, Vinen, et al., 1998) analyzed the data with logistic regression using a continuous CK variable. Thus only two studies were available for meta-analysis; however, because the number of studies available was small, meta-analysis was not performed.

Table 18. Serial creatine kinase to diagnose AMI: Diagnostic (more...)

Table 18. Serial creatine kinase to diagnose AMI: Diagnostic performance studies (ED studies)

Study, Year	Study size	Population category	Prevalence of AMI (%)	Times of blood draws evaluated	Test performance		Study quality
					Sensitivity (%)	Specificity (%)
Gerhardt 1982	481	I	43	Hours 10, 16 of chest pain	99	68	C
Roxin 1984	305	I	26	0, 2, 4 hours	69	84	C
Overall	786	I	26-43	-	69-99 1	68-84 1	C
					Odds ratio 12-220 1

1

Range of reported values.

The study by Gerhardt, Waldenstrom, Horder, et al. (1982) included patients seen in an ED with "symptoms indicative of their having had an AMI within the previous 24 hours" (Table 18). The test of interest was CK-B subunit, which was corrected for CK-BB. The test was considered abnormal if activity was greater than 12 U/L. Blood was drawn 10 and 16 hours after presentation. No data were reported on patients not analyzed or on demographics.

The study by Roxin, Cullhed, Groth, et al. (1984) included patients "referred to [an urban] hospital on suspicion of AMI" (Table 18). No information was given on loss to followup. CK samples were drawn at presentation and 2 and 4 hours later. Of note, of the 305 patients included, 147 were not included in the reported test performance analysis because 22 had a final diagnosis of "possible AMI" (with atypical ECG changes or elevated AST levels in only one blood sample) and for 125, AMI "could not be entirely excluded" (patients had chest pain, shock, or pulmonary edema with no other cause, but no ECG changes or elevations of AST).

Figure 13. Studies of serial CK to diagnose AMI1

1

Due

(more...)

   Figure 13. Studies of serial CK to diagnose AMI¹

Table 18 and Figure 13 (see meta-analysis section) show a summary of included articles. Both studies reported broad eligibility criteria; however, these two studies varied significantly in their methods. The study by Gerhardt and coworkers (1982) based timing of serial testing on symptom duration, and the study by Roxin and coworkers (1984) based timing of serial testing on time since arrival to the ED. The study by Gerhardt and colleagues (1982) also reported an AMI prevalence at 43 percent, considerably greater than the prevalence at 26 percent reported in the study by Roxin and colleagues (1984).

Studies of clinical impact of the test's actual use

No studies are reported.

Data From Other Clinical Studies

No data are reported.

Single CK-MB Upon Presentation to ED to Diagnose AMI

Studies of test sensitivity and specificity

Table 19. Presentation creatine kinase-MB to diagnose AMI: (more...)

Table 19. Presentation creatine kinase-MB to diagnose AMI: Diagnostic performance studies (ED studies)

Study, year	Study size	Population category	Prevalence of AMI (%)	Test performance		Study quality
				Sensitivity (%)	Specificity (%)
Mair 1991b	126	II	40	57	92	C
Collins 1993	195	II	42	52	97	B
Brogan 1994	189	I	12	23	99	B
Castaldo 1994	157	II	37	22	98	C
Montague 1995	89	II	28	36	98	C
Thomson 1995	375	II	18	56	93	C
Hetland 1996	133	II	34	44	98	B
Mair 1996	100	II	39	59	89	B
Gornall 1996	98	III	41	25	98	B
Hedges 1996	1,042	III	6.4	57	96	A
Laurino1 1997	115	III	22	20	ND	C
Overall2	2,504	I/II/III	6.4-42	44 (35-53) 3	96 (94-97) 3	B
				Odds ratio 23 (17-32) 3

1

Study not included in meta-analysis because data were incomplete.

2

Laurino (1997) not included in overall summary.

3

Results from meta-analysis using random effects calculations.

Table 20. Presentation creatine kinase-MB to diagnose AMI: (more...)

Table 20. Presentation creatine kinase-MB to diagnose AMI: Diagnostic performance studies (all patients admitted)

Study, year	Study size	Population category (setting)	Prevalence of AMI (%)	Test performance		Study quality
				Sensitivity (%)	Specificity (%)
Gibler 1987	59	IV (hospital admission)	36	14	97	C
Gibler 1990	183	IV (hospital admission)	17	55	97	A
Marin 1992	313	IV (hospital admission)	22	41	95	C
Tucker 1994	133	IV (hospital admission)	29	33	99	C
Apple 1995	98	IV (hospital admission)	6.1	100	86	B
Tucker 1997	177	IV (hospital admission)	15	26	97	B
Kontos 1997a	101	IV (CCU admission)	20	30	100	B
Fesmire 1998	764	IV (hospital admission)	20	48	97	C
Kontos 1999a	2,093	IV (CCU admission)	8.9	46	99	C
Young 19931	222	IV (CCU admission)	19	40	No data	C
Overall2	3,921	IV	6.1-36	40 (33-47) 3	97 (94-99) 3	C
				Odds ratio 28 (14-54) 3

1

Study not included in meta-analysis.

2

Laurino (1997) not included in overall summary.

3

Results from meta-analysis using random effects calculations (studies including admitted patients only).

A total of 31 studies pertaining to the evaluation of CK-MB as a single test upon presentation to the ED for the diagnosis of AMI were retrieved. Three studies (Mair, Artner-Dworzak, Lechleitner, et al., 1992; Hedges, Young, Henkel, et al., 1994; Young, Gibler, Hedges, et al., 1997) presented the same data sets as previously published studies and were thus excluded. Two studies (Laurino, Pelletier, Eadry, et al., 1997; Young and Green, 1993) reported only sensitivity results. These two studies are presented in Tables 19 and 20 and Evidence Tables 7a and 7b but are not included in meta-analysis. One study (D'Costa, Fleming, and Patterson, 1997) compared CK-MB data with troponin I data and did not report test performance data for AMI. Two studies (Katz, Irwig, Vinen, et al., 1998; Ohman, Casey, Bengston, et al., 1990) analyzed their data with logistic regression using a continuous CK-MB variable. One study (Zalenski, McCarren, Roberts, et al., 1997a) reported on a highly selected population sample of low-risk patients who were hospitalized. The study was excluded because 57 percent of the patients had symptoms for more than 24 hours (40 percent for more than 48 hours) and thus were not representative of typical ED patients. One report (Mach, Lovis, Chevrolet, et al., 1995) was excluded because it studied a highly selected sample of patients in whom AMI was suspected by criteria of the Imminent Myocardial Infarction Rotterdam Study; notably, the AMI prevalence was 78 percent. Two studies (de Winter, Koster, Sturk, et al., 1995; Laurino, Bender, Kessimian, et al., 1996) reported data only by onset of symptoms in such a way that data on presentation testing could not be extracted. Therefore, 19 data sets were available for analysis of CK-MB to diagnose AMI. Of these, seven studies included only patients who were admitted to the hospital after ED evaluation. Two others included only patients admitted to the CCU after ED evaluation. Thus, 10 studies (published between 1991 and 1996) that analyzed patients seen in the ED, including patients both admitted to the hospital and discharged from the ED, were included in meta-analysis.

One study by Mair, Artner-Dworzak, Dienstl, et al. (1991b) included patients seen in an urban ED with a chief complaint of chest pain. CK-MB was considered abnormal if mass was greater than 7 ng/ml. No data were reported on subjects not analyzed or on demographics.

The study by Collins, Wright, Rinsler, et al. (1993) included patients seen in a suburban community hospital with chest pain "suggestive of AMI" and referred with acute symptoms for cardiology assessment. CK-MB was defined as abnormal if mass was greater than 7 ng/ml; however, data were available for multiple CK-MB thresholds. No data were reported on subjects not analyzed or on demographics.

The study by Brogan, Friedman, McCuskey, et al. (1994) included patients seen in a rural university-affiliated ED with a broad range of chief complaints consistent with ACI. Patients whose symptoms were of more than 12 hours duration or who had renal failure or muscular dystrophy were excluded. CK-MB was defined as abnormal if mass was greater than 9 ng/ml and the index was greater than 2 percent. No data were reported on subjects not analyzed or on gender.

The study by Castaldo, Ercolini, Forino, et al. (1994) included patients with less than 2 hours of chest pain and were thought likely to be having AMI. Although patients had measurements at 3, 6, and 9 hours after onset of chest pain (rather than specifically at presentation), as they presented within 2 hours of onset of chest pain, the 3-hour data were used as admission measurements. CK-MB was defined as abnormal if index was greater than 5 percent. No data were reported on patients not analyzed.

The study by Thomson, Gibbons, Smars, et al. (1995) included patients over 20 years old seen in an urban ED with anterior or left lateral chest pain. Patients with recent infection, steroid use, cancer, gastrointestinal bleed, surgery, hemodialysis, or cardiopulmonary resuscitation were excluded (n=87). Only 3 of 20 patients who were discharged from the ED had followup tests and were evaluated. Fourteen patients had multiple ED visits, each of which was included in the analysis. "Exclusion of the 17 repeat visits [resulted in] only a 1 percent to 2 percent increase in the sensitivity." CK-MB was defined as abnormal if mass was greater than 4.7 ng/ml.

The study by Montague and Kircher (1995) included patients having chest pain or "suspected AMI" seen in a suburban ED. Patients with renal insufficiency (n=6) were excluded. Thirty-seven eligible patients were not evaluated in any analyses because only one blood sample was drawn. CK-MB was defined as abnormal if mass was greater than 5 ng/ml.

The second study by Mair, Genser, Morandell, et al. (1996) included patients seen in an urban ED with a chief complaint of chest pain. No data were reported on subjects not analyzed; however, one patient is unaccounted for in the CK-MB analysis. CK-MB was defined as abnormal if mass was greater than 5 ng/ml.

The study by Hedges, Gibler, Young, et al. (1996) included patients seen in a university-affiliated ED with chest discomfort or clinical suspicion of AMI. Patients with diagnostic ECG, who were unstable, or who had recent cardioversion were excluded. Thirteen patients were excluded because data were incomplete. CK-MB was defined as abnormal if mass was greater than 7 ng/ml.

The study by Gornall and Roth (1996) included patients seen in a suburban ED with chest pain. Patients with ECGs diagnostic of AMI at presentation or who had noncardiac diseases diagnosed by history and physical at presentation were excluded. No data were reported on subjects not analyzed. CK-MB was defined as abnormal if mass was greater than 8 ng/ml.

The study by Hetland and Dickstein (1996) included patients seen in an urban ED with "chest discomfort suggestive of AMI" for less than 6 hours. Patients who arrived in the ED from 10 p.m. to 8 a.m. were not included. No data were reported on subjects not analyzed. CK-MB was defined as abnormal if mass was greater than 5 ng/ml.

In general, subjects were patients who presented to the ED with chest pain or discomfort. One study (Brogan, Friedman, McCuskey, et al., 1994) included all patients with symptoms consistent with ACI. One study (Hedges, Gibler, Young, et al., 1996) excluded patients with diagnostic ECG. One study (Gornall and Roth, 1996) excluded patients with diagnostic ECGs and patients with easily diagnosed noncardiac diseases. One study (Hetland and Dickstein, 1996) excluded patients with symptoms lasting more than 6 hours. One study (Castaldo, Ercolini, Forino, et al., 1994) excluded patients with chest pain lasting less than 2 hours. One study (Thomson, Gibbons, Smars, et al., 1995) excluded patients with a variety of recent noncardiac conditions, and one (Montague and Kircher, 1995) excluded patients with renal insufficiency. Very little information was available about eligible patients who were not evaluated.

In general, AMI was defined by WHO criteria or variations thereof. In all, serial CK or CK-MB was used to define AMI. Only two (Mair, Artner-Dworzak, Dienstl, et al., 1991b; Brogan, Friedman, McCuskey, et al., 1994) explicitly did not use CK-MB to define AMI (apparently using CK only). One (Thomson, Gibbons, Smars, et al., 1995) used only serial CK-MB and sudden death to define AMI. Almost all studies used mass measurements of CK-MB (ng/ml) ranging from 4.7 ng/ml to 8 ng/ml. One (Brogan, Friedman, McCuskey, et al., 1994) used a combination of mass (9 ng/ml) and index (2 percent MB). One study (Castaldo, Ercolini, Forino, et al., 1994) used only CK-MB index (5 percent MB).

Figure 12. SROC analysis of presentation CK-MB to diagnose (more...)

   Figure 12. SROC analysis of presentation CK-MB to diagnose AMI (ED only studies)

Table 19 and Figure 12 (see meta-analysis section) show a summary of included articles, results of meta-analysis, and unweighted ROC analysis. As with presentation CK, heterogeneity among studies could not be explained by CK-MB thresholds, reported differences in eligibility criteria, study setting, or prevalence of AMI. There was not a clear correlation between mean or median symptom duration and test sensitivity in the nine studies that reported this information. Those studies that analyzed test performance by symptom duration (Mair and coworkers, 1991b, 1996; Collins and coworkers, 1993; Brogan and coworkers, 1994; de Winter and coworkers, 1995; Hetland and Dickstein, 1996; Laurino and coworkers, 1996, 1997) all found increased sensitivity of presentation CK-MB in patients with longer duration of symptoms.

Studies of clinical impact on the test's actual use

No studies are reported.

Data From Other Clinical Studies

Nine articles included patients evaluated in the ED who were subsequently admitted to the hospital, thus excluding patients discharged from the ED.

The study by Gibler, Gibler, Weinshenker, et al. (1987) included patients seen in a community hospital ED with chest pain who were subsequently hospitalized. Fourteen subjects were not included in analysis because of missing laboratory results or lack of consent. The definition of abnormal CK-MB was not reported. No demographic data were provided.

A second study by Gibler, Lewis, Erb, et al. (1990) included patients in the ED with chest pain who were subsequently hospitalized. Three patients were not included in the analysis: one patient had incomplete data, one was already hospitalized prior to chest pain, one had an initial CK greater than 4,000 IU/L. CK-MB was defined as abnormal if mass was greater than 7.5 ng/ml.

The study by Marin and Teichman (1992) included patients seen in the ED with chest pain consistent with AMI or ischemia for less than 12 hours who were subsequently hospitalized. Patients were excluded if they were transferred to another hospital for admission; 62 patients were not included in the analysis because no final diagnosis could be made because of such events as surgery or death, 22 patients had chest pain for more than 12 hours, and 3 were missing data. CK-MB was defined as abnormal if mass was greater than 7.5 ng/ml.

The study by Tucker, Collins, Anderson, et al. (1994) included patients seen in the ED with chest discomfort for less than 24 hours that prompted a physician to order cardiac enzymes and who were subsequently hospitalized. Sixty patients were not included in the analysis: 30 were discharged from the ED, 17 had symptoms for more than 24 hours, 6 had improper laboratory test collection, 6 had no consent, and 1 was transferred to another facility. CK-MB was defined as abnormal if mass was greater than 5 ng/ml. No data were reported on demographics.

The study by Apple, Voss, Lund, et al. (1995) included patients seen in the ED with chest pain who were subsequently hospitalized. Patients had a mean duration of chest pain of 14 hours. No information was reported on patients not included in the analysis or on subject demographics. CK-MB was defined as abnormal if mass was greater than 5 ng/ml.

The second study by Tucker, Collins, Anderson, et al. (1997) included patients seen in an urban ED with chest discomfort of less than 24 hours who were subsequently hospitalized. Patients with initial diagnostic ECGs or who received cardiopulmonary resuscitation were excluded. No data were reported on patients not analyzed. CK-MB was defined as abnormal if mass was greater than 5 ng/ml.

The study by Fesmire, Percy, Bardoner, et al. (1998) included patients seen in the ED of a university teaching hospital with chest pain who were subsequently hospitalized. Patients with recent cocaine use, tachyarrhythmias, pulmonary edema, or demand pacemakers were excluded. Of 1,000 eligible patients, 236 failed to have a second CK-MB level drawn and thus were not analyzed. CK-MB was defined as abnormal if mass was greater than 6 ng/ml.

Two articles (Kontos, Anderson, Hanbury, et al., 1997a; Kontos, Anderson, Schmidt, et al., 1999a) included patients evaluated in the ED who were subsequently admitted to the coronary care unit, thus excluding patients discharged from the ED or admitted to a noncoronary care unit bed.

The 1997a study by Kontos and colleagues included patients seen in an ED with chest pain who were subsequently admitted to the CCU. No specific exclusion criteria were reported; nine patients were not included in the analysis because of missing laboratory data. CK-MB was defined as abnormal if mass was greater than either 4.7 ng/ml or 8 ng/ml, depending on the analyzer used.

The 1999a article by Kontos and colleagues included patients initially seen in an urban ED who were admitted to the coronary care unit with suspected AMI. Patients were excluded if their initial ECG was diagnostic for AMI or if they had cardiac or respiratory arrest in the ED; 241 patients with diagnostic ECGs, 44 patients with cardiac or respiratory arrest, and 406 patients with missing laboratory data were excluded. CK-MB was defined as abnormal if mass was greater than 8 ng/ml and index was greater than 4 percent.

In summary, the range of prevalence within the studies of hospitalized patients was similar to that within ED-based studies. Likewise, the symptom inclusion and exclusion criteria and patient demographics were similar overall. The range of sensitivity and specificity values reported was also within the same range as in the ED-based studies. The one exception to this was the study by Apple and coworkers (1995), which reported 100 percent sensitivity in a sample of patients whose average symptom duration was 14 hours, compared with 2 to 6 hours in other studies. These studies were published between 1987 and 1999.

Analyzing all 19 studies addressing the diagnostic accuracy of CK-MB as a single test done at presentation to the ED yielded 6,425 patients. The random effects model yielded a pooled sensitivity of 42 percent (95 percent CI=36 to 48 percent) and a pooled specificity of 97 percent (95 percent CI=95 to 98 percent). The random effects model odds ratio is 25 (95 percent CI=18 to 36 percent).

Single CK-MB Upon Presentation to ED to Diagnose ACI

Table 21. Presentation creatine kinase-MB to diagnose ACI: (more...)

Table 21. Presentation creatine kinase-MB to diagnose ACI: Diagnostic performance studies

Study, year	Study size	Population category	Prevalence of ACI (%)	Test performance		Study quality
				Sensitivity (%)	Specificity (%)
Hedges 1996	1,042	III	20.4 (AMI 6.4)	23	96	C
				Odds ratio=7.2

Only one study evaluated CK-MB as a single test upon presentation to the ED to diagnose both AMI and UAP (Table 21).

Hedges, Gibler, Young, et al. (1996) included patients seen in a university-affiliated ED with chest discomfort or clinical suspicion of AMI. Patients with diagnostic ECGs, who were unstable, or who had recent cardioversion were excluded. CK-MB was defined as abnormal if mass was greater than 7 ng/ml. AMI was defined by WHO criteria with echocardiography and cardiac catheterization used to diagnose some patients. No definition of UAP was given. Of 1,042 patients 67 (6.4 percent) were diagnosed with AMI and 146 (14 percent) were diagnosed with UAP.

Data From Prospective Clinical Studies in the ED Setting

Studies of test sensitivity and specificity

Table 22. Serial creatine kinase-MB to diagnose AMI: Diagnostic (more...)

Table 22. Serial creatine kinase-MB to diagnose AMI: Diagnostic performance studies (ED studies)

Study, year	Study size	Population category	Prevalence of AMI (%)	Times of blood draws evaluated	Test performance		Study quality
					Sensitivity (%)	Specificity (%)
Gerhardt 1982	481	I	43	Hours 10, 16 of chest pain	100	97	C
Hedges 1992	261	III	11	0, 1, 2, 3 hours	68	95	C
Brogan 1994	189	I	12	0, 1 hours	41	99	B
Castaldo 1994	157 1	II	37	Hours 3, 6 of chest pain	62	95	C
Montague 1995	89	II	28	0, 2 hours	68	92	C
Gibler 1995	1,010	III	1.2	0, 3, 6, 9 hours	100	98	C
Hedges 1996	1,042	III	6.4	0, 3 hours	88	95	A
Sayre 19982	473	I	5.1 (34/667) 2	0, 3 hours 3	40 3	99.8 3	B
Overall4	3,149	I/II/III	1.2-53	-	800 (61-91) 5	96 (94-98) 5	C
					Odds ratio 130 (40-400) 5

1

Not all subjects had two laboratory samples.

2

AMI prevalence information reported for total sample only. No extractable data reported for subset of population (n=473) included in 3-hour test performance data. Therefore, study not included in meta-analysis.

3

Includes up to 128 of 473 patients with only a single blood draw.

4

Sayre (1998) not included in overall summary.

5

Results from meta-analysis using random effects calculations.

A total of 23 studies pertaining to the evaluation of CK-MB used in serial testing in the ED for the diagnosis of AMI were retrieved. Two articles (Young, Gibler, Hedges, et al., 1997; Hedges, Young, Henkel, et al., 1994) presented the same data sets as previously published articles and were thus excluded. One article (Puleo, Meyer, Wathen, et al., 1994) was excluded because it provided data on subforms of CK-MB only. Two articles (Brogan, Vuori, Friedman, et al., 1996; Brogan, Bock, McCuskey, et al., 1997) reported data only by symptom duration, which could not be extracted for meta-analysis or serial testing. One study (Katz, Irwig, Vinen, et al., 1998) analyzed the data with logistic regression using a continuous CK variable. One study (Mach, Lovis, Chevrolet, et al., 1995) was excluded because it studied a highly selected sample of patients in whom AMI was suspected by criteria of the Imminent Myocardial Infarction Rotterdam Study; notably, the AMI prevalence was 78 percent. One study (de Winter, Koster, Sturk, et al., 1995) reported data only by onset of symptoms in such a way that data on serial testing could not be extracted. Sayre, Kaufmann, Chen, et al. (1998) reported test performance data on 473 of 667 patients who presented to the ED with symptoms suggestive of ACI and had blood drawn within 3 hours of presentation to the ED. A large, but unreported, number of these patients had only one blood sample drawn. In addition, the prevalence of AMI in this subpopulation was not reported and thus meta-analysis could not be performed. However, this study is included in Table 22 and Evidence Table 8A. Sevenof the remaining 14 articles (published between 1982 and 1996) included patients seen in the ED or a chest pain evaluation unit that included patients both admitted to the hospital and discharged from the ED.

The study by Gerhardt, Waldenstrom, Horder, et al. (1982) included patients seen in an ED with "symptoms indicative of their having had an AMI within the previous 24 hours." The test of interest was CK-B subunit, which was corrected for CK-BB. The test was considered abnormal if activity was greater than 12 U/L. Blood was drawn at 10 and 16 hours after presentation. No data were reported on patients not analyzed or on demographics.

The study by Hedges, Young, Henkel, et al. (1992) included patients with chest discomfort seen at either an urban university ED or a Veterans' Administration ED. Patients were excluded if they had ECGs diagnostic for AMI, shock, anemia, or recent cardioversion; 159 patients were not included because of incomplete data. CK-MB was defined as abnormal if mass was greater than 8 ng/ml. Blood was drawn hourly from presentation through 3 hours.

The study by Brogan, Friedman, McCuskey, et al. (1994) included patients seen in a rural university-affiliated ED with a broad range of chief complaints consistent with ACI. Patients whose symptoms were of more than 12 hours duration or who had renal failure or muscular dystrophy were excluded. CK-MB was defined as abnormal if mass was greater than 9 ng/ml and the index was greater than 2 percent. Blood was drawn at presentation and at 1 hour. No data were reported on patients not analyzed or on gender.

The study by Castaldo, Ercolini, Forino, et al. (1994) included patients with less than 2 hours of chest pain and were thought likely to be having AMI. Although patients had measurements at 3, 6, and 9 hours after onset of chest pain (rather than specifically at presentation), as they presented within 2 hours of onset of chest pain, the 3- and 6- hour data were used as serial measurements. CK-MB was defined as abnormal if index was greater than 5 percent. No data were reported on patients not analyzed.

The study by Gibler, Runyon, Levy, et al. (1995) included patients seen in a "Heart Emergency Room" with chest pain. Those with diagnostic ECG, hypotension, or history of ACI, AMI, or CAD were excluded. Notably, those with symptoms suggestive of UAP were also excluded and patients admitted directly to the CCU who were not evaluated in the Heart Emergency Room were not included. CK-MB was defined as abnormal if greater than 6 ng/ml or 7 ng/ml (depending on the analyzer used) with an index greater than 5 percent. Blood was drawn every 3 hours from presentation through 9 hours. No data were reported on patients not analyzed.

The study by Montague and Kircher (1995) included patients with chest pain or "suspected AMI" seen in a suburban ED. Patients with renal insufficiency (n=6) were excluded. Thirty-seven eligible patients were not evaluated in any analyses because only one blood sample was drawn. CK-MB was defined as abnormal if mass was greater than 5 ng/ml. Blood was drawn at presentation and hour 2.

The study by Hedges, Gibler, Young, et al. (1996) included patients with chest discomfort or clinical suspicion of AMI seen in a university-affiliated ED. Patients with diagnostic ECG, who were unstable, or who had recent cardioversion were excluded. Thirteen patients were excluded because of incomplete data. CK-MB was defined as abnormal if mass was greater than 7 ng/ml. Blood was drawn at presentation and after 3 hours.

In summary, six studies evaluated all eligible patients seen in the ED (Gerhardt, Waldenstrom, Horder, et al., 1982; Hedges, Young, Henkel, et al., 1992; Brogan, Friedman, McCuskey, et al., 1994; Castaldo, Ercolini, Forino, et al., 1994; Montague and Kircher, 1995; Hedges, Gibler, Young, et al., 1996). One study evaluated patients seen in a chest pain evaluation unit (Gibler, Runyon, Levy, et al., 1995).

The prevalence for AMI in these studies ranged from 1.2 percent to 43 percent. In general, subjects were patients who presented to the ED with chest pain. Two studies (Gerhardt, Waldenstrom, Horder, et al., 1982; Brogan, Friedman, McCuskey, et al., 1994) included all patients with symptoms consistent with ACI. Two studies also included patients with suspected AMI (Montague and Kircher, 1995; Hedges, Gibler, Young, et al., 1996). Two studies excluded patients with more than 12 hours of symptoms (Brogan, Friedman, McCuskey, et al., 1994; de Winter, Koster, Sturk, et al., 1995); one (Castaldo, Ercolini, Forino, et al., 1994) excluded those with less than 2 hours of chest pain. Three studies excluded patients who required cardioversion or were otherwise unstable (Hedges, Young, Henkel, et al., 1992; Gibler, Runyon, Levy, et al., 1995; Hedges, Gibler, Young, et al., 1996). Three trials excluded patients with diagnostic ECG (Hedges, Young, Henkel, et al., 1992; Gibler, Runyon, Levy, et al., 1995; Hedges, Gibler, Young, et al., 1996). One study also excluded patients with a history of coronary artery disease or "a clinical syndrome. . .consistent with unstable angina" (Gibler, Runyon, Levy, et al., 1995). All but one study (Gerhardt, Waldenstrom, Horder, et al., 1982) excluded patients with trauma or whose chest pain was explained by chest radiography.

In general, AMI was defined by WHO criteria; however, one study (Gibler, Runyon, Levy, et al., 1995) did not define AMI. All studies that reported which cardiac enzymes were used to define AMI used CK and/or CK-MB except for one study that used AST and lactate dehydrogenase (LDH) to define AMI, although this study also used CK-MB when diagnosis was doubtful (Gerhardt, Waldenstrom, Horder, et al., 1982). One also defined AMI by angiography (Hedges, Gibler, Young, et al., 1996).

Four studies used mass (ng/ml) measurements for CK-MB, using thresholds that ranged from 5 ng/ml to 8 ng/ml. One study used a combination of mass (ng/ml) and index (percent MB) measurements for CK-MB (Brogan, Friedman, McCuskey, et al., 1994), using a threshold of 9 ng/ml and 2 percent CK-MB. One used activity (U/L) measurements (Gerhardt, Waldenstrom, Horder, et al., 1982), using a threshold of 12 U/L. One used only an index of 5 percent (Castaldo, Ercolini, Forino, et al., 1994). Two studies (Gerhardt, Waldenstrom, Horder, et al., 1982; Castaldo, Ercolini, Forino, et al., 1994) drew laboratory samples according to symptom duration at either 10 and 16 hours after onset of symptoms or at 3, 6, and 9 hours after onset of symptoms. In the rest of the studies, samples were drawn at ED presentation and at various times from 1 to 9 hours after presentation.

Figure 14. SROC analysis of serial CK-MB to diagnose (more...)

   Figure 14. SROC analysis of serial CK-MB to diagnose AMI (ED only studies)¹

Table 22 and Figure 14 show a summary of included articles, results of meta-analysis, and unweighted ROC analysis. Heterogeneity was best explained by timing of serial testing. Studies that performed serial testing within 2 hours of presentation had test sensitivity levels below 70; those that drew serum samples at presentation and 3 or 4 hours had sensitivities from 68 percent to 90 percent. Test sensitivity was 100 percent in the two studies that performed serial testing at 6 or more hours after presentation. Likewise, the study that performed serial testing up to 16 hours after onset of symptoms had substantially higher sensitivity than the study that performed serial testing up to 6 hours after symptom onset.

Studies of clinical impact on the test's actual use

Table 23. Serial creatine kinase-MB to diagnose ACI: clinical (more...)

Table 23. Serial creatine kinase-MB to diagnose ACI: clinical impact study

Study, year	Study size	Populationcategory	Prevalence of ACI (%)	Clinical impact		Study quality
Hedges 1996	1,042	III	AMI 6.4 UAP 14	Additional admission to hospital of ACI patients:	4 (1.9%) 3 AMI, 1 UAP	C
				Additional discharge from ED of ACI patients:	0
				Additional admission to hospital of non-ACI patients:	5 (0.6%)
				Additional discharge from ED of non-ACI patients:	27 (3.3%)

One study (Hedges, Gibler, Young, et al., 1996) evaluated CK-MB and changes in decisionmaking by ED physicians on whether patients required hospital or CCU admission (Table 23). The study was a single-armed, prospective, observational study of 1,042 patients seen in seven EDs with chest discomfort and a clinical suspicion of AMI. Patients were excluded who had diagnostic ECGs, clinical instability, recent cardioversion, chest trauma, or diagnostic chest radiographs.

The analysis pertained primarily to patient-specific rankings by the physicians of the importance attributed to CK-MB to clinical decisionmaking. However, overall in the 67 patients with AMI, the decision to admit to the hospital was changed from "no" to "yes" after the CK-MB results for 3 (4.5 percent) patients were reviewed. For no patient with AMI was the decision to admit changed to not admit after review of the CK-MB levels. Likewise, in the 67 patients with AMI, the decision to admit to the CCU was changed from "no" to "yes" for 13 (19.4 percent) patients. The decision was changed to not admit to the CCU for two (3.0 percent) patients.

Of the 146 patients with UAP, the decision to admit to the hospital was changed from "no" to "yes" for one (0.7 percent) patient. No patients with UAP were discharged from the ED. Of the 829 patients without ACI, 5 (0.6 percent) additional patients were admitted to the hospital after review of the test results; 27 (3.3 percent) additional patients were discharged.

Data From Other Clinical Studies

Five articles included patients evaluated in the ED who were subsequently admitted to the hospital, thus excluding patients discharged from the ED.

Table 24. Serial creatine kinase-MB to diagnose AMI: Diagnostic (more...)

Table 24. Serial creatine kinase-MB to diagnose AMI: Diagnostic performance studies (all patients admitted)

Study, year	Study size	Population category	Prevalence of AMI (%)	Times of blood draws evaluated	Test performance		Study quality
					Sensitivity (%)	Specificity (%)
Gibler 1987	59	IV (hospital admission)	45	0, 3, 6 hours	100	92	C
Gibler 1992	577	IV (hospital admission)	12	0, 1, 2, 3 hours	80	94	B
Marin 1992	313	IV (hospital admission)	20	0, 1, 2, 3 hours	90	93	C
Hoekstra 1994	5,063	IV (hospital admission)	7.3	0, 2 hours	63	95	B
Levitt 1996	190	IV (hospital admission)	11	0, 3 hours	81	99	A
Kontos 1997a	101	IV (CCU admission)	20	0, 4 hours	90	100	B
Kontos 1999a	2,093	IV (CCU admission)	8.9	0, 3 hours	78	99	C
Overall	8,396	IV (hospital/CCU admission)	7.3-45	-	81 (71-89) 1	97 (94-98) 1
					Odds ratio 171 (52-565) 1

1

Results from meta-analysis using random effects calculations.

Another study (Gibler, Young, Hedges, et al., 1992) included patients seen in an urban ED with a chief complaint of chest pain who were subsequently hospitalized. Patients with diagnostic ECGs were excluded (52 patients). CK-MB was defined as abnormal if mass was greater than 8 ng/ml. Blood was drawn at presentation and hourly for 3 hours. No data on subject demographics were reported.

The study by Marin and Teichman (1992) included patients seen in the ED with chest pain consistent with AMI or ischemia for less than 12 hours who were subsequently hospitalized. Patients were excluded if they were transferred to another hospital for admission; 62 patients were not included in the analysis because no final diagnosis could be made due to such events as surgery or death, 22 patients had chest pain for more than 12 hours, and 3 had missing data. CK-MB was defined as abnormal if mass was greater than 7.5 ng/ml. Blood was drawn at presentation and hourly for 3 hours.

The study by Hoekstra, Hedges, Gibler, et al. (1994) included patients seen in a community hospital ED with a chief complaint of chest pain or discomfort consistent with AMI who were subsequently hospitalized. Patients with initial diagnostic ECGs were excluded. No data were reported on patients not included in the analysis. The definition of abnormal CK-MB varied depending on the analyzer used, ranging from 5.6 ng/ml to 7.5 ng/ml. Blood was drawn at presentation and at 2 hours. No data were reported on demographics.

The study by Levitt, Promes, Bullock, et al. (1996) included patients seen in an urban ED with chest pain consistent with acute cardiac ischemia who were subsequently hospitalized. Patients with diagnostic ECGs were excluded. No data were reported on patients not analyzed. CK-MB was defined as abnormal if mass was greater than 11.9 ng/ml. This level was chosen post hoc after examination of a receiver operator curve. Blood was drawn at presentation and at 3 hours.

Two articles included patients evaluated in the ED who were subsequently admitted to the coronary care unit, thus excluding patients discharged from the ED or admitted to a noncoronary care unit bed (Kontos Anderson, Hanbury, et al., 1997a; Kontos, Anderson, Schmidt, et al., 1999a). These two articles were described above. (See Single CK-MB Upon Presentation to ED to Diagnose AMI -- Data from other clinical studies.) In the 1997a study by Kontos and colleagues, blood was drawn at presentation and at 4 hours. In the 1999a study by Kontos and coworkers, blood was drawn at presentation to the ED and at 3 hours.

In summary, seven additional studies, which included only hospitalized patients, addressed the diagnostic accuracy of CK-MB as a serial test done in the ED. Two of these evaluated only patients seen in the ED subsequently admitted to the coronary care unit (Kontos, Anderson, Hanbury, et al., 1997a; Kontos, Anderson, Schmidt, et al., 1999a). These studies were published between 1987 and 1999.

The prevalence for AMI among these studies ranged from 7 percent to 45 percent. Three studies included patients who presented to the ED with chest pain (Gibler, Gibler, Weinshenker, et al., 1987; Marin and Teichman, 1992; Kontos, Anderson, Hanbury, et al., 1997a); the other four excluded patients with diagnostic ECGs. One study excluded patients with more than 12 hours of symptoms (Marin and Teichman, 1992). One study excluded patients who required cardioversion or were otherwise unstable (Kontos, Anderson, Schmidt, et al., 1999a).

In general, AMI was defined by WHO criteria or variants. One study also defined AMI by angiography (Hoekstra, Hedges, Gibler, et al., 1994). All studies that reported which cardiac enzymes were used to define AMI used CK and/or CK-MB. One study (Kontos, Anderson, Schmidt, et al., 1999a) also used troponin I in some patients.

Five studies used mass (ng/ml) measurements for CK-MB, using thresholds that ranged from 4.7 ng/ml to 11.9 ng/ml. One study used a combination of mass (ng/ml) and index (percent MB) measurements for CK-MB (Kontos, Anderson, Schmidt, et al., 1999a), using a threshold of 8 ng/ml and 4 percent CK-MB. One study did not report a method for measuring CK-MB or a threshold (Gibler, Gibler, Weinshenker, et al., 1987). Each study had a different protocol for timing serial testing, drawing samples at ED presentation and at various times from 2 to 6 hours after presentation.

Analyzing all 14 studies addressing the diagnostic accuracy of serial CK-MB done in the ED yielded 11,625 patients. The random effects model yielded a pooled sensitivity of 79 percent (95 percent CI=71 to 86 percent) and a pooled specificity of 96 percent (95 percent CI=95 to 97 percent). The random effects model odds ratio is 140 (95 percent CI=65 to 310).

Serial CK-MB Upon Presentation to ED to Diagnose ACI

Table 25. Serial creatine kinase-MB to diagnose ACI: Diagnostic (more...)

Table 25. Serial creatine kinase-MB to diagnose ACI: Diagnostic performance study

Study, year	Study size	Population category	Prevalence of ACI (%)	Times of blood draws evaluated	Test performance		Study quality
					Sensitivity (%)	Specificity (%)
Hedges 1996	1,042	III	ACI 20.4 (AMI 6.4)	0, 3 hours	31	95	C 1

1

No definition of UAP was reported in study.

Only one study (Hedges, Gibler, Young, et al., 1996) pertained to the evaluation of serial CK-MB in the ED to diagnose both AMI and UAP (Table 25). That study included patients seen in a university-affiliated ED with chest discomfort or clinical suspicion of AMI. Patients with diagnostic ECGs, who were unstable, or who had recent cardioversion were excluded. CK-MB was defined as abnormal if mass was greater than 7 ng/ml. AMI was defined by WHO criteria with echocardiography and cardiac catheterization used to diagnose some patients. No definition of UAP was given.

Data From Prospective Clinical Studies in the ED Setting

Studies of test sensitivity and specificity

A total of 27 reports were evaluated for the assessment of the diagnostic accuracy of myoglobin in the diagnosis of AMI and ACI. Ten articles studied presentation myoglobin in the ED setting. We are uncertain whether there is overlap between the Tucker, Collins, Anderson, et al. (1994) and Tucker, Collins, Anderson, et al. (1997) reports and between the Kontos, Anderson, Hanbury, et al. (1997a) and Kontos, Anderson, Schmidt, et al. (1999a) reports, but protocols seem to have differences, even if the investigators are the same. Sensitivity analyses including only one of the two studies yield identical results.

The included patients could have had variable duration of symptoms upon presentation, which might have affected diagnostic performance. One study (Castaldo, Ercolini, Forino, et al., 1994) provides information defined according to time from onset of symptoms. Patients had measurements at 3, 6, and 9 hours after onset of chest pain, and all patients presented within 2 hours of onset of chest pain. In the quantitative synthesis, the 3-hour data are used as admission measurements and 6-hour data are used as measurements 2 to 4 hours after admission.

Castaldo and colleagues (1994) limited enrollment to patients with less than 2 hours of symptoms; Kennedy, Harrison, Burton, et al. (1997) to less than 4 hours of symptoms; Hetland and Dickstein (1996) to less than 6 hours of symptoms; and Brogan, Friedman, McCuskey, et al. (1994) to less than 12 hours of symptoms. Brogan and coworkers (1994) give a mean of 3.2 hours from onset of symptoms to presentation; Mair, Artner-Dworzak, Lechleitner, et al. (1992) reported that the median time from onset of symptoms was 2.3 hours; and Hetland and Dickstein (1996) reported median times from onset of symptoms for patients with AMI of 2 to 4 hours and for those without AMI of 3 hours. Other reports lack this crucial information, but it may be assumed that the distribution of this measurement is likely to be typical of unselected populations presenting with chest pain in the ED. Delays of blood draw from the time of ED arrival are not mentioned typically.

Two studies (Gornall and Roth, 1996; Kennedy, Harrison, Burton, et al., 1997) restricted the analysis to patients who had nondiagnostic ECG. This may limit their generalizability. Other reports did not seem to have substantial restrictions. Exclusion criteria related to conditions that may cause spurious elevations in myoglobin (trauma, renal insufficiency, known muscular dystrophy or other muscular disease, recent intramuscular injections, and recent prehospital thrombolysis) were mentioned to various extents and combinations in the included reports -- none of these exclusions is likely to limit the study populations substantially, although exact data on screened-out patients are sparse. Lack of samples or inadequate samples seemed to be a negligible problem, when this information was recorded and reported.

Table 26. Presentation myoglobin to diagnose AMI: Diagnostic (more...)

Table 26. Presentation myoglobin to diagnose AMI: Diagnostic performance studies (ED studies)

Study, year	Study size	Population category	Prevalence of AMI (%)	Test performance		Study quality
				Sensitivity (%)	Specificity (%)
Gilkeson 1978	71	I	18	38	93	C
Roxin 1984	305	I	26	71	83	C
Mair 1992	126	I	40	45	91	B
Castaldo 1994	157	II	37	38	100	C
Brogan 1994	189	I	12	55	98	A
Montague 1995	89	I	28	56	81	B
Mair 1996	101	I	39	46	89	B
Gornall 1996	98	III	41	43	98	B
Hetland 1996	133	II	34	51	94	C
Kennedy 1997	86	III	23	38	96	C
Overall	1,395	I/II/III	12-41	49 (41-57) 1	93 (88-96) 1	B/C
				Odds ratio 13 (7.9-21) 1

1

Results from meta-analysis using random effects model.

Figure 16. SROC analysis of presentation myoglobin (more...)

   Figure 16. SROC analysis of presentation myoglobin to diagnose AMI (ED only studies)

The included reports use variable cutoffs for the definition of an abnormal myoglobin value, but the differences are small. Cutoffs range from 70 to 100 ng/ml. Mair, Genser, Morandell, et al. (1996) and Hetland and Dickstein (1996) also provide estimates for other thresholds that could generate ROC curves (reported area under curve 0.65 in Mair and coworkers (1996) and 0.80 in Hetland and Dickstein (1996). Shown in Table 26 and Figure 16 is the unweighted SROC analysis which suggests modest diagnostic performance. There is substantial heterogeneity in the sensitivity estimates which may be explained by the inclusion of studies with different distributions of the time from onset of symptoms. Those studies that analyzed test performance by symptom duration (Brogan, Friedman, McCuskey, et al., 1994; de Winter, Koster, Sturk, et al., 1995; Hetland and Dickstein, 1996; Laurino, Pelletier, Eadry, et al., 1997; Mair, Genser, Morandell, et al., 1996) all found increased sensitivity of presentation myoglobin in patients with longer duration of symptoms.

Only Kennedy, Harrison, Burton, et al. (1997) offered some data on the diagnostic performance of myoglobin for the diagnosis of more broadly defined coronary ischemia (AMI or angina).

Serial myoglobin in ED to diagnose AMI

Table 27. Serial myoglobin to diagnose AMI: Diagnostic (more...)

Table 27. Serial myoglobin to diagnose AMI: Diagnostic performance studies (ED studies)

Study, year	Study size	Population category	Prevalence of AMI (%)	Test performance		Study quality
				Sensitivity (%)	Specificity (%)
Simple numerical thresholds
Roxin 1984	305	I	26	98	83	C
Brogan 1994	189	I	12	73	96	A
Castaldo 1994	157 1	II	37	90	100	C
Montague 1995	89	I	28	100	77	B
Kennedy 1997	91	III	23	86	96 2	C
Overall	831	I/II/III	23-37	90 (78-96)	90 (78-96)	A/B/C
				Odds ratio 140 (66-300)
Simple numerical threshold or increase in level
Brogan 1994	189	I	12	91	96	A
Gornall 1996	98	II	41	93	100	B
Doubling of myoglobin value only
Montague 1995	89	I	28	64	98	B

1

Not all subjects had two laboratory samples.

2

Text unclear. Specificity not stated clearly for serial myoglobin. Data included here may be overestimation of specificity.

Figure 17. SROC analysis of serial myoglobin to diagnose (more...)

   Figure 17. SROC analysis of serial myoglobin to diagnose AMI (ED only studies)

Eleven studies provided data on myoglobin measurements obtained some time after the initial ED/hospital presentation (Table 27; Figure 17). Ten reports evaluated use of simple numerical thresholds for myoglobin. Two reports (Brogan, Friedman, McCuskey, et al., 1994; Gornall and Roth, 1996) studied the diagnostic value of using a combination of numerical thresholds and increases (of 40 ng/ml, 50 percent, or 100 percent) from the initial serum myoglobin value. One report (Montague and Kircher, 1995) also evaluated the value doubling only within 2 hours after presentation.

As with serial CK-MB, heterogeneity was best explained by timing of serial testing. Overall, as time between serial tests was increased, the sensitivity of the test also increased. However, two small studies that performed serial testing at 2 hours after presentation reported test sensitivity of 100 percent.

Data From Prospective Clinical Studies in the ED Setting

Studies of test sensitivity and specificity

Clinical trials of troponin I and troponin T include the following:

Troponin I upon presentation

Mair, Genser, Morandell, et al. (1996) selected patients for enrollment including patients with AMI to test the diagnostic performance of an immunoassay for troponin I in the ED. There were no data on patient demographics or inclusion/exclusion criteria. Final diagnosis by cardiologist was made blinded to test results. Average onset of pain to ED presentation was 2.6 hours with a range of 0.3 to 48 hours.

Table 28. Presentation troponin I to diagnose AMI: Diagnostic (more...)

Table 28. Presentation troponin I to diagnose AMI: Diagnostic performance studies (ED studies)

Study, year	Study Size	Population category	Prevalence of AMI (%)	Test performance		Study quality
				Sensitivity (%)	Specificity (%)
Mair 1996	101	II	39	23	95	C
Hamm 1997	773	III	6	66	89	A
Overall	874	II/III	6-39	23-66 1	89-95 1	A

1

Range of reported values.

Hamm, Goldmann, Heeschen, et al. (1997) evaluated the diagnostic test performance of troponin I and troponin T in the emergency department on 773 consecutive patients. Serum samples were taken at presentation and repeated 4 hours later. Patients presenting under 2 hours after onset of pain had an additional sample drawn so that all patients had tests performed at least 6 hours after onset of pain. Outside laboratory results for troponin I were made blinded to patient outcome and compared with final diagnosis by CK and CK-MB levels (see sections on serial troponin I, serial troponin T, and presentation troponin T for additional sensitivity/specificity data). The results of the two studies are shown in Table 28. The limited data suggest that the test performance is similar to that of presentation troponin T.

Other troponin I clinical data

Table 29. Presentation troponin I to diagnose AMI: Diagnostic (more...)

Table 29. Presentation troponin I to diagnose AMI: Diagnostic performance studies (all patients admitted)

Study, Year	Study size	Population category	Prevalence of AMI (%)	AMI test performance		Study quality
				Sensitivity (%)	Specificity (%)
Apple 1995	98	IV	6	100	92	C
Tucker 1997	177	IV	15	3.7	98	B
D'Costa 1997	316	II	20	79	ND	C
Laurino 1997	115	I	22	32	ND	C

There were two studies evaluating the test performance of immunoassays for detecting troponin I for patient presentation to the ED, but the patients were subsequently admitted (Table 29).

Apple, Voss, Lund, et al. (1995) analyzed the diagnostic performance of troponin I from a sample drawn at presentation in the ED from 98 consecutive patients, who were then admitted to rule out AMI. There were no data on blinding of test interpretation or patient characteristics. The mean sampling time from onset of chest pain for all patients was 14 hours and 7.4 hours for patients with AMI.

Tucker, Collins, Anderson, et al. (1997) tested the diagnostic performance of troponin I in a cohort of consecutively enrolled patients with nondiagnostic ECGs. All patients were subsequently admitted with the admitting physician blinded to the test results. The mean time from onset of chest pain for all patients was 4.0 hours and the median 2.2 hours.

Two studies also reported only sensitivity data on troponin I. D'Costa, Fleming, and Patterson (1997) included patients with less than 24 hours of chest pain, excluding patients with trauma or renal insufficiency. Data on presentation troponin I are presented only for the 62 patients with AMI who had troponin I levels above or below 1.0 ng/ml. Laurino, Pelletier, Eadry, et al. (1997) analyzed patients with a chief complaint of chest pain or symptoms consistent with AMI or ischemia for less than 6 hours. Patients receiving thrombolytics were excluded. Data on presentation troponin I are presented only for the 25 patients with AMI who had troponin I levels above or below 0.6 ng/ml.

Serial troponin I to diagnose AMI

Table 30. Serial troponin I to diagnose AMI: Diagnostic (more...)

Table 30. Serial troponin I to diagnose AMI: Diagnostic performance study

Study, year	Study size	Population category	Prevalence of AMI (%)	Test performance		Study quality
				Sensitivity (%)	Specificity (%)
Hamm 1997	773	III	6	100	83	A

One study evaluated the test performance of serial immunoassays for detecting troponin I for patient presentation to the ED (Table 30). Hamm, Goldmann, Heeschen, et al. (1997) evaluated the diagnostic performance of serial troponin I at presentation and at 6 hours after onset of pain for 773 consecutive patients (see section on presentation troponin I for details of study).

Table 31. Serial troponin I to diagnose AMI: Diagnostic (more...)

Table 31. Serial troponin I to diagnose AMI: Diagnostic performance study (all patients admitted)

Study, year	Study size	Population category	Prevalence of AMI (%)	Test performance		Study quality
				Sensitivity (%)	Specificity (%)
Kontos 1999b	620	IV	10	90	96	B

One study (Kontos, Jesse, Anderson, et al., 1999b) reviewed the diagnostic performance of serial troponin I over an 8-hour period in the CCU with 620 patients considered at low to moderate risk for ACI (Table 31). Patients were admitted through the ED; however, laboratory samples were drawn in the CCU. The test results were made available to the treating physician, and there were no data on the blood sampling interval other than references to sampling periods similar to other studies that used chest pain evaluation protocols.

Troponin T upon presentation

Table 32. Presentation troponin T to diagnose AMI: Diagnostic (more...)

Table 32. Presentation troponin T to diagnose AMI: Diagnostic performance studies (ED studies)

Study, year	Study size	Population category	Prevalence of AMI(%)	Test performance		Study quality
				Sensitivity (%)	Specificity (%)
Mair 1991a	96	II	24	57	96	C
de Winter 19951	309	III	53	34	91	B
Mair 1996	101	II	39	28	92	C
Hetland 1996	133	II	34	53	86	C
Hamm 1997	773	III	6	51	92	A
Gust 1998	68	II 2	24	25	98	B
Overall	1,171	II/III	6-39	44 (32-56) 3	92 (88-95) 3	C
				Odds ratio 10 (5.9-18) 3

1

Not included in meta-analysis. No data on numbers of subjects included in analysis of troponin T. Does not include all patients presenting to the ED.

2

Serum sample drawn prior to arrival in ED.

3

Results from meta-analysis using random effects model.

There are four eligible ED studies and one study in which laboratory samples were drawn prior to arrival in the ED (Table 32). It should be noted that different studies use variable cutoffs for the definition of an abnormal troponin T value. Cutoffs range from 0.05 ng/ml to 0.5 ng/ml, but all except two studies have cutoffs that are in the range of 0.1 to 0.2 ng/ml. Mair, Genser, Morandell, et al. (1996) and Hetland and Dickstein (1996) also provide estimates for the area under the ROC curve based on a range of thresholds (estimates of the areas reported are 0.60 and 0.76, respectively).

Mair, Artner-Dworzak, Lechleitner, et al. (1991) enrolled 96 ED patients from the hospital's department of internal medicine. There were no data on enrollment criteria or time from onset of pain to presentation. The cardiologist was blinded to the test results. The cutoff for a positive test was 0.5 ng/ml.

Mair, Gener, Morandell, et al. (1996) selected patients for enrollment including patients with AMI to test the diagnostic performance of an immunoassay for troponin T. There were no data on patient demographics or inclusion/exclusion criteria. The range for onset of pain to presentation was 0.3 to 48 hours with a median delay of 2.6 hours. Final diagnosis was made blinded to the test results.

Hetland and Dickstein (1996) evaluated 133 consecutive patients with a maximum of 6 hours of pain at presentation to the ED. The median time of pain onset to ED presentation was 3 hours for non-AMI patients, 2 hours for AMI female patients, and 4 hours for AMI male patients. Final diagnosis was determined with knowledge of the test results.

Hamm, Goldmann, Heeschen, et al. (1997) evaluated the diagnostic performance of troponin T at presentation and at 6 hours after onset of pain for 773 consecutive patients (see section on presentation troponin I for details of study).

Gust, Gust, Bottiger, et al. (1998) evaluated the diagnostic performance of troponin T drawn by an emergency doctor traveling with the emergency medical service prior to arrival in the ED. The 68 patients had radiating chest pain which was not relieved by rest or sublingual nitroglycerin. Patients with recent AMI, thrombolytic treatment, or angioplasty were excluded. Patients had a mean of 4.3 hours of symptoms (range 0.5 to 10.0 hours) prior to arrival of the ambulance. A rapid bedside immunoassay was used.

One study (de Winter, Koster, Sturk, et al., 1995) did not provide sufficient information to allow complete data extraction. It was therefore not included in meta-analysis, although it is included in Table 32. de Winter and colleagues (1995) studied 309 consecutive patients considered at low risk for AMI. Sensitivity and specificity estimates according to time from onset of symptoms were reported as there were multiple blood samples taken from the onset of chest pain, hourly from 3rd to 8th, and every 4th hour thereafter to 24 hours. The prevalence of AMI appears to be relatively high at 53 percent.

The overall specificity appears excellent, although the sensitivity seems to vary substantially among studies. It should be noted that the included patients could have had variable duration of symptoms upon presentation and typically these studies offer little information on this parameter and how it might have affected diagnostic performance.

Serial troponin T to diagnose AMI

Table 33. Serial troponin T (up to 6 hours) to diagnose (more...)

Table 33. Serial troponin T (up to 6 hours) to diagnose AMI: Diagnostic performance studies (ED studies)

Study, year	Study size	Population category	Prevalence of AMI (%)	Test performance		Study quality
				Sensitivity (%)	Specificity (%)
Hamm 1997	773	III	6	90 1	80 1	A
Sayre 1998	546 2	II	5	65 3	93 3	C

1

Data from 0 to 4 hours used; other measurements available.

2

546 of 667 patients had blood drawn within 3 hours of presentation to ED.

3

Includes patients with 1, 2, or 3 blood draws within 3 hours of presentation to ED. Up to 128 patients had only blood draw.

4

Range of reported values.

Two studies (Sayre, Kaufmann, Chen, et al., 1998) and (Hamm, Goldmann, Heeschen, et al., 1997) reported information on the diagnostic performance of multiple serial measurements obtained within less than 6 hours from ED presentation (Table 33). These two studies suggest that the sensitivity of the assay increases with longer symptom duration. In the study by Sayre and coworkers (1998), the sensitivity improved from 65 percent at 0 to 3 hours after admission to 79 percent at 3 to 6 hours after admission, whereas specificity remained essentially unchanged at 93 percent using the same cutoff of 0.2 ng/ml. However this analysis of "serial" data includes up to 128 patients with only a single blood draw. Based on measurements up to 6 hours after presentation, Hamm and colleagues (1997) recorded a sensitivity of 94 percent (44/47) for a specificity of 89 percent (647/726) using a cutoff of 0.18 ng/ml. These data suggest that serial measurements for 6 hours may improve the sensitivity from the 50 percent at presentation to 80 to 90 percent, at the same time maintaining the specificity at around 90 percent.

Data From Other Clinical Studies

Table 34. Presentation troponin T to diagnose AMI: (more...)

Table 34. Presentation troponin T to diagnose AMI: Diagnostic performance studies (other studies)

Study, year	Study size	Population category	Prevalence of AMI (%)	Test performance		Study quality
				Sensitivity (%)	Specificity (%)
Mach 1995	32	IV	78	40	"80" 1	C
Tucker 1997	68	IV	15	15	97	B

1

Reported specificity value not possible with seven patients.

Tucker, Collins, Anderson, et al. (1997) tested the diagnostic performance of troponin T in a cohort of consecutively enrolled patients with nondiagnostic ECGs. However, all patients were subsequently admitted. The mean time from onset of chest pain for all patients was 4.0, median 2.2 hours.

Table 35. Serial troponin T to diagnose AMI: Diagnostic performance (more...)

Table 35. Serial troponin T to diagnose AMI: Diagnostic performance studies (other studies)

Study, year	Study size	Population category	Prevalence of AMI (%)	Test performance		Study quality
				Sensitivity (%)	Specificity (%)
Mach 1995	32	IV	78	100	86	C
Mohler 1998	100	IV	6	94	89	B

Only two studies offered diagnostic performance data for more broadly defined coronary outcomes. Mohler and colleagues (1998) provided data on AMI or unstable angina and showed a sensitivity of 58 percent (36/62) and specificity of 97 percent (37/38). Green, Beaudreau, Chan, et al. (1998) evaluated AMI or any adverse cardiac event as the outcome and found a sensitivity of 22 percent (10/45) and specificity of 98 percent (242/247). These data, although limited, suggest excellent specificity, but very poor sensitivity, for the diagnosis of more broadly defined cardiac coronary outcomes.

Data From Prospective Clinical Studies in the ED Setting

Four studies reported on the combination of CK-MB and myoglobin. Three of these reported data on the two tests drawn in one serum sample; two reported data on serial samples. No other combinations of biomarker tests were reported by other studies. Only one of these studies evaluated all ED patients.

Combination CK-MB and myoglobin upon presentation

Table 36. Presentation CK-MB and myoglobin to diagnose AMI: (more...)

Table 36. Presentation CK-MB and myoglobin to diagnose AMI: Diagnostic performance studies

Study, year	Study size	Population category	Prevalence of AMI (%)	Test performance		Study quality
				Sensitivity (%)	Specificity (%)
Montague 1995	89	II	28	100 1	72 1	C
Kontos 1997a	101	IV	20	85	80	B
Kontos 1999a	2,093	IV	9	62	89	C
Overall	2,283	IV	9-28	83 (81-96) 2	82 (68-90) 2	B/C
				Odds ratio 17 (7.6-40) 2

1

Results from sample drawn at 2 hours after presentation to ED.

2

Results from meta-analysis using random effects model.

Two studies reported data on combination CK-MB and myoglobin drawn at patient presentation to the ED (Kontos, Anderson, Hanbury, et al. 1997a; Kontos, Anderson, Schmidt, et al., 1999a) (Table 36). A third study reported only combination data from the serum sample drawn 2 hours after presentation (Montague and Kircher, 1995). Eligibility criteria and test thresholds were the same as for evaluation of the individual tests. Kontos and colleagues (1997a, 1999a) included only patients admitted to the CCU; Montague and Kircher (1995) included all eligible ED patients. A positive combination test was defined as either a positive CK-MB or a positive myoglobin.

Although the test performance of the combination test appears better than that for each of the individual tests, it should be noted that it is likely that the decision to report the combination result was made a posteriori. It is unclear if the small number of studies that actually reported the combination data is a biased sample (in that studies with better combination performance than individual test performance may be more likely to report the data). In addition, the study that reported only 2-hour data had a substantially higher test sensitivity of 100 percent.

Serial combination CK-MB and myoglobin to diagnose AMI

Table 37. Serial CK-MB and myoglobin to diagnose AMI: Diagnostic (more...)

Table 37. Serial CK-MB and myoglobin to diagnose AMI: Diagnostic performance studies

Study, year	Study size	Population Category	Prevalence of AMI (%)	Test performance		Study Quality
				Sensitivity (%)	Specificity (%)
Levitt 1996	190	IV	11	100	91	A
Kontos 1997a	101	IV	20	100	75	B

Only two studies reported on combination CK-MB and myoglobin drawn in serial serum samples (Levitt, Promes, Bullock, et al., 1996; Kontos, Anderson, Hanbury, et al., 1997a) (Table 37). Eligibility criteria and test thresholds were the same as for evaluation of the individual tests. The studies included only patients admitted to either the hospital or the CCU. A positive combination test was defined as either a positive CK-MB or a positive myoglobin in either of the serum samples drawn.

With such a small sample, it is difficult to compare individual with combination serial testing, although the test performance appears to be at least similar.

Data from Prospective Clinical Trials in the ED Setting

Studies of test sensitivity and specificity

Table 38. Presentation P-selectin: Diagnostic performance study (more...)

Table 38. Presentation P-selectin: Diagnostic performance study

Study, year	Study size	Population category	Test	Prevalence of disease (%)	Test performance		Study quality
					Sensitivity	Specificity(%)
Hollander 1999	263	II	Soluble P-selectin	AMI 8.4 ACI 33	45 (27-69) 35 (25-46)	76 (70-81) 79 (72-85)	B
			Membrane-bound P-selectin	AMI ACI	32 (15-54) 30 (21-41)	71 (65-77) 71 (64-78)
			P-selectin index 1	AMI ACI	59 (41-82) 55 (45-66)	54 (48-61) 57 (50-65)

1

P-selectin index required elevation in either soluble or membrane-bound P-selectin.

A single study (Hollander, Muttreja, Dalesandro, et al., 1999) studied the diagnostic test performance of P-selectin to diagnose ACI and AMI in the ED (Table 38). Consecutive patients over age 18 who presented to the ED of an urban, tertiary care center who had a chief complaint of chest pain for less than 6 hours were included. Patients whose symptoms were clearly of noncardiac origin were excluded. Approximately 10 percent of patients were not analyzed because of refusal to participate or inadequate blood samples. AMI was defined by WHO criteria; UAP was classified according to the AHCPR (now AHRQ) risk stratification scheme, in addition to requiring demonstration of CAD, exercise-induced ischemia, or CK-MB levels between 5 and 10 ng/ml. Subjects had presented to the ED with a median of 3 hours of symptoms.

Thresholds for P-selectin were chosen post hoc after ROC curve analysis. Blood was drawn at presentation and at 1 hour. To diagnose AMI, presentation soluble P-selectin was found to have a sensitivity of 45 percent and a specificity of 76 percent with a threshold of 138.8 ng/ml. Presentation membrane-bound P-selectin had a slightly lower test performance, with a sensitivity of 32 percent and a specificity of 71 percent with a threshold of 4.8 percent. In addition, a "P-selectin index" (requiring elevation in either soluble or membrane-bound P-selectin) had a sensitivity of 59 percent and a specificity of 54 percent. Serial testing results at 1 hour were not presented, although it was stated that they did not significantly affect test performance.

To diagnose ACI, presentation soluble P-selectin had a sensitivity of 35 percent and a specificity of 79 percent; membrane-bound P-selectin, a sensitivity of 30 percent and a specificity of 71 percent; and the P-selectin index, a sensitivity of 55 percent and a specificity of 57 percent. Again, serial results were not presented.

Overall, in this single study, P-selectin at presentation (and reportedly with serial testing at presentation and 1 hour) had poor to moderate sensitivity and moderate selectivity.

Data From Other Clinical Studies

Malondialdehyde (MDA)-modified low-density lipoprotein (LDL) is a biomarker beginning to be evaluated for its role in diagnosing or categorizing ACI. It may reflect endothelial injury or plaque instability (Holvoet, Collen, and Van de Werf, 1999). This is the only study to date that investigates the clinical utility of this biomarker. The study included 104 patients with ACI (42 with UAP, 62 with AMI) who presented to the ED with chest pain and ECG changes, and 64 patients with stable, angiographically documented coronary artery disease. Blood samples were drawn for patients with ACI within 6 to 8 hours of onset of chest pain. Plasma levels of MDA-modified LDL were 2.6- to 2.9-fold higher in patients with AMI or UAP than with stable coronary artery disease. Plasma levels were similar in patients with AMI and UAP. ROC analysis also revealed that MDA-modified LDL discriminated between stable CAD and UAP. At a threshold of 0.70 mg/dl (0.02 mmol/L) MDA-modified LDL had a sensitivity of 95 percent for UAP and 95 percent for AMI. The specificity in this highly selected sample of patients with stable or unstable coronary artery disease was 95 percent.

Data From Prospective Clinical Trials in the ED Setting

A total of 11 reports were considered for the assessment of the diagnostic accuracy of echocardiography (Horowitz, Morganroth, Parrotto, et al., 1982; Sasaki, Charuzi, Beeder, et al., 1986; Peels, Visser, Kupper, et al., 1990; Sabia, Afrookteh, Touchstone, et al., 1991; Levitt, Promes, Bullock, et al., 1996; Mohler, Ryan, Segar, et al., 1996, 1998; Trippi, Kopp, Lee, et al., 1996; Trippi, Lee, Kopp, et al., 1997; Kontos, Arrowood, Jesse, et al., 1998; Gibler, Runyon, Levy, et al., 1995). Eight of these reports dealt with rest echocardiograms, and two dealt with both rest and dobutamine stress echocardiography (Trippi, Kopp, Lee, et al., 1996; Trippi, Lee, Kopp, et al., 1997). Both of the articles by Trippi and colleagues pertain to the same study and the earlier one contains the first 26 patients of the later publication; however, complementary information is provided. One study compared video vs. digital recording of echocardiograms (Mohler, Ryan, Segar, et al., 1996).

Studies of test sensitivity and specificity

Table 39. Rest echocardiography to diagnose AMI: Diagnostic (more...)

Table 39. Rest echocardiography to diagnose AMI: Diagnostic performance studies (ED studies)

Study, year	Study size	Population category	Prevalence of AMI (%)	Test performance		Study quality
				Sensitivity (%)	Specificity (%)
Peels 1990	43	III	30.2	92.3	53.3	C
Sabia 1991	169	I	16.7	93.1	57.1	B
Kontos 1998	185	III	3.2	100	82.1	B
Overall	397	III	3-30	93 (81-97) 1	66 (43-83) 1	B
				Odds ratio 20 (6.5-62) 1

1

Results from meta-analyses using random effects calculations.

Only three rest echocardiography studies met the inclusion criteria of being exclusively in the ED setting (Peels, Visser, Kupper, et al., 1990; Sabia, Afrookteh, Touchstone, et al., 1991; Kontos, Arrowood, Jesse, et al., 1998) and their details are presented in Evidence Table 14a. The sensitivity values reported by the three studies showed very little variation. Therefore, the SROC method was not particularly useful to summarize the data for rest echocardiography. The random effects model sensitivity and specificity results calculated for these three studies are shown in Table 39.

Two studies (Peels, Visser, Kupper, et al., 1990; Kontos, Arrowood, Jesse, et al., 1998) (Table 39) provided data for a broad ACI definition for diagnostic performance; the random effect diagnostic odds ratio is similar to that of AMI outcome, 16.6 (95 percent CI, 7.1 to 38.9).

Data From Other Clinical Studies

Several studies that do not meet the strict ED only inclusion criteria are discussed here. With the exception of studies by Mohler, Ryan, Segar, et al. (1998) and Horowitz, Morganroth, Parratto, et al. (1982), all other studies required normal or nondiagnostic ECGs as an inclusion criterion. In addition, Trippi, Lee, Kopp, et al. (1997) and Gibler, Runyon, Levy, et al. (1995) also required normal enzymes. Gibler and coworkers (1995) performed echocardiograms only after serial enzymes and serial ECGs had been normal for 9 hours and not all patients underwent echocardiography. It is thus a very selected population. A past history of AMI was an exclusion criterion for Trippi and colleagues (1997), Gibler and colleagues (1995), Sasaki, Charuzi, Beeder, et al. (1986), and Horowitz and colleagues (1982), whereas at least the first three studies also excluded patients with any history of CAD or ACI -- no similar statement is clarified in Sasaki and coworkers (1986) and Horowitz and coworkers (1982). Almost all the patients in the study by Mohler and colleagues (1998) had previous echocardiograms for comparison to avoid misinterpretation of old abnormalities. The results from Levitt, Promes, Bullock, et al. (1996) are expressed as mean and standard deviation of wall motion scores, and no difference was reported in echocardiographic scores between the AMI and non-AMI groups (mean 16.9 vs. 15.3, p=0.32).

The diagnosis of AMI was based on WHO criteria or variants thereof, but several studies seemed to depend entirely on enzymes. More broadly defined coronary syndromes were studied in Sasaki and coworkers (1986) (AMI or CAD by angiography or stress test), Trippi and coworkers (1997) (AMI or CAD by catheterization or telephone survey), and Gibler and coworkers (1995) (any cardiac disease). When the ED studies and non-ED studies are combined, the random effects pooled sensitivity is 77 percent (95 percent CI=51 to 92 percent) and the specificity is 85 percent (95 percent CI=74 to 91 percent) using this broad coronary syndrome definition. The use of diverse and suboptimal reference standards may affect the estimate of diagnostic performance of echocardiography. The study by Mohler, Ryan, Segar, et al. (1996) suggests that there is excellent concordance between video and digital evaluation of echocardiograms.

Data From Prospective Clinical Trials in the ED Setting

Data From Other Clinical studies

Table 40. Dobutamine stress echocardiography -- broaden (more...)

Table 40. Dobutamine stress echocardiography -- broaden ACI definition: Diagnostic performance study

Study, year	Study size	Population category	Prevalence of disease (%)		Test performance		Study quality
					Sensitivity (%)	Specificity (%)
Trippi 1997	139	III (ED/CCU)	ACI AMI	11.6-13.7 1 3.7-4.3 1	- 89.5 2	- 88.9 2	C

1

Depends on which of the three possible numbers reported was used in the calculations.

2

Point estimate from single study.

Data From Prospective Clinical Trials in the ED Setting

Studies of test sensitivity and specificity

Nine reports pertaining to the study of Tc-99m sestamibi imaging in the ED setting were retrieved. Two teams had produced three reports, each of which overlapped to a substantial extent; and another team produced two overlapping reports. First, the Virginia Commonwealth University team produced three overlapping reports. Kontos, Arrowood, Jesse, et al. (1998) presented information on 185 patients who had both Tc-99m sestamibi and echocardiography during the period of August 1994 to December 1994; and Tatum, Jesse, Kontos, et al. (1997) presented information on 438 patients (regardless of whether they had echocardiography as well or not) during the period June 1, 1994, to October 26, 1994, at the same institution based on a rule-out ACI protocol which used Tc-99m sestamibi for patients at low to moderate risk for AMI and ACI.

The overarching report by Kontos, Jesse, Schmidt, et al. (1997b) contains information on patients who had Tc-99m sestamibi imaging at the same institutions between June 1994 and August 1995. Therefore only this report, of the three, was used in the evidence synthesis.

Similarly, three reports originated from the William Beaumont Hospital. Weissman, Dickinson, Dworkin, et al. (1996) report on the original experience of 50 scanned patients. The data were used to generate cost-effectiveness estimates, but the report does not include diagnostic accuracy data which were previously presented in a 1995 abstract in the Journal of Nuclear Medicine. The Stewart, Dickinson, Weissman, et al. (1996) report contains an update with a total of 68 patients, and therefore only this report was considered in the evidence synthesis.

It was verified that the paper by Hilton, Fulmer, Abuan, et al. (1996) included the same patient population as an earlier report. Therefore, the report by Hilton, Thompson, Williams, et al. (1994) is used in the evidence synthesis.

The study by Varetto, Cantalupi, Altieri, et al. (1993) was mentioned in the original Working Group report, but it was also noted that only patients admitted to the CCU were studied. This study was not included in our analysis.

Table 41. Rest sestamibi to diagnose AMI: diagnostic performance (more...)

Table 41. Rest sestamibi to diagnose AMI: diagnostic performance studies

Study, year	Study size	Population category	Prevalence of disease (%)		Test performance		Study quality
					Sensitivity (%)	Specificity (%)
Hilton 1994	102	III	ACI AMI	13.7 11.8	93.3 100	79.3 77.8	B
Stewart 1996	68	III	ACI AMI	8.8 1.5	100 100	53.2 49.3	B
Kontos 1997b1	532	III	ACI AMI	17.1 5.3	79.8 92.9	80.9 71.2	A
Kontos 19981	185	III	ACI AMI	- 3.2	- 100	- 84.4	A
Overall	702 1	III	AC	9-17	81 (74-87) 2	73 (56-85) 2	B
					Odds ratio 30 (10-92) 2
			AMI	2-12	92 (78-98) 2	67 (52-79) 2
					Odds ratio 26 (6-113) 2

1

There are some overlaps in the patient populations in these three reports. Communication with the authors of these reports revealed that 53 patients in the Kontos (1997b) study were included in the Kontos (1998) study.

2

Results from meta-analysis using random effects calculations.

With these clarifications, four studies with a total of 765 evaluable patients addressed the diagnostic accuracy of Tc-99-sestamibi scan in ED patients with chest pain (Table 41). These studies were published between 1993 and 1997 and, with one exception (Stewart, Dickinson, Weissman, et al. 1996), used only rest scan. A subgroup of patients in the Stewart, Dickinson, Weissman, et al. (1996) study (36/68 patients) also underwent stress scan, if the rest scan was unrevealing. Demographic and reference standard peculiarities are shown in the evidence tables.

Characteristically, all studies addressed populations of patients where there was a presumed low or moderate at the most risk of AMI and ACI and the ECG was normal or nondiagnostic. Also, patients with a history of AMI were excluded in three of the four studies to avoid difficulty in interpreting segmental wall motion abnormalities. The Kontos, Jesse, Schmidt, et al. (1997b) report included patients regardless of a prior history of MI, but separate data are also provided for the subgroup of patients without such a history.

The overall rate of confirmed AMI in these studies was only 7.1 percent (54/765), and the rate of more broadly defined coronary syndromes (defined with somehow different definitions across studies -- see Evidence Table 15) was 21.3 percent (163/765). In this regard, these studies generally targeted highly selected populations: In one study where data were available, only 64 of 274 consecutive patients with chest pain qualified for inclusion (Varetto, Cantalupi, Altieri, et al., 1993); in another study (Tatum, Jesse, Kontos, et al., 1997, a subset of Kontos, Jesse, Schmidt, et al., 1997b), the qualifying rate was 442/1,187; in Stewart, Dickinson, Weissman, et al. (1996), the study population represented only 7 percent of all patients evaluated for chest pain syndromes not felt to be AMI; Hilton, Thompson, Williams, et al. (1994) did not provide information on the respective percentage. Unsuccessful imaging was uncommon in the two studies that provided these data (less than 2 percent for both combined).

The overall test performance suggests excellent sensitivity, but only modest specificity for AMI. The accuracy results for more broadly defined coronary disease syndromes should be interpreted with caution, since the definitions of coronary disease were heterogeneous.

Data From Prospective Clinical Trials in the ED Setting

Studies of test sensitivity and specificity

Table 42. ACI-TIPI to diagnose ACI: Diagnostic performance (more...)

Table 42. ACI-TIPI to diagnose ACI: Diagnostic performance studies

Study, year	Study size	Population category	Prevalence of disease (%)		Overall physician performance		Study quality
					Sensitivity (%)	Specificity (%)
Pozen 1980	856 Test 401 Control 455	I	ACI Test Control	16.5 17.4	86.4	91.6	A
Pozen 1984	2,320	I	ACI Test Control	32 29	94.5	78.1	A
Davison 1990	232	II	ACI AMI	29.7 20.7	No data 1 -	No data 1 -	A
Selker 1991	2,320	I	ACI AMI	34.2 17.6	Not applicable		A
Overall	5,496	I	ACI AM	16.5-34.2 17.6-20.7	86-95 2 -	78-92 2 -	A

1

See study for SROC curves.

2

Range of values reported.

Table 43. ACI predictive instrument/ACI-TIPI to diagnose ACI: (more...)

Table 43. ACI predictive instrument/ACI-TIPI to diagnose ACI: clinical impact studies

Study, year	Study size		Population category	Prevalence of disease (%)		30-day mortality(%)		Study quality
Pozen 1980	856		I	ACI		No data		A
	Test Control	401 455		Test Control	16.5 17.4
Pozen 1984	2,320		I	ACI		No data		A
				Test Control	32 29
Corey 1987	56		I	ACI - No data		No data		C
Sarasin 1994	529		I 1	ACI AMI	59.0 31.9	Test Control	8 9	B
	Test Control	294 235
Selker 1998	10,689		I	ACI 2 Test Control	24 23	2.6 2.4		A
	Test Control	4,738 5,951		AMI Test Control	8 8
Overall	14,450		I	ACI AMI	(17-59.0) (8-31.9)	2.4-9		A

1

Study conducted in Switzerland, where patients may have been referred by primary care physician.

2

Includes angina, unstable angina pectoris, and myocardial infarction.

Four studies met the inclusion criteria for the diagnostic performance of the original ACI predictive instrument and the ACI-TIPI (Table 42).

Pozen, D'Agostino, Mitchell, et al. (1980) tested the diagnostic performance of the original ACI predictive instrument in a single, urban, teaching hospital. The 10-month study consisted of alternating periods of experimental versus control months. During the experimental months, the instrument's probability of diagnosing ACI was made available to the ED physician who had discretion whether to utilize the additional information for patient diagnosis. Sensitivity dropped from 89.9 percent during the control period to 86.4 percent during the experimental period, whereas the specificity improved from 80.1 percent to 91.6 percent, respectively. The followup rate was 92 percent, and variables revealing differences between the two groups were adjusted for. The monthly rotation schedule of the ED clinicians effectively controlled for bias or eliminated the maturation effect. The generalizability of the results from this single, urban setting is uncertain.

Pozen, D'Agostino, Selker, et al. (1984) tested the diagnostic performance of the ACI predictive instrument with slight modifications in the number of variables used compared with the number in the 1980 study by Pozen and coworkers. Six hospitals were involved, consisting of two hospitals per three different teaching-status levels with two study designs employed: (1) an alternating 11-month schedule of an experimental versus a control month, and (2) a time-series consisting of 6 experimental months followed by 5 control months. The study was basically conducted in the same manner as the previous study in terms of diagnostic and inclusion criteria, retrospective review and blinding, and the ED physician autonomy to incorporate the instrument's probability in the diagnosis. The difference in sensitivity from 95.3 percent during the control period to 94.5 percent during the experimental period was nonsignificant, whereas the specificity improved from 73.2 percent to 78.1 percent (p=0.002), respectively. That the incidence of disease was greater in the second paper than in the first paper could be a reflection of a wider and larger general population, since the first paper was set in a single, urban setting with a relatively large minority representation of 45 percent to 51 percent. Followup rate was 89 percent. Reported prevalence and sensitivity/specificity data could not be verified because of the lack of reporting of actual numbers.

Davison, Suchman, and Goldstein (1990) tested three decision aids in a small study of 235 patients. Specific variables to each of the decision aids were incorporated into a data collection form filled out by the ED residents who were also blinded to the prediction of the aids and to the purpose of the study. It is not known whether the final diagnosis was made blinded to the predictions of the decision aids. There were minor inconsistencies between the number of enrolled patients and patients analyzed. SROC was reported using nine cut points to distinguish normal from abnormal rather than a continuous curve as intended by the design of the instrument. For the evaluation patients for AMI, none of the protocols could reduce the false-positive rate without reduction of the true-positive rate.

Selker, Griffith, and D'Agostino (1991) tested the ACI Time-Insensitive Predictive Instrument at the same time the 1984 study by Pozen and colleagues was being conducted, utilizing the same data on the same population at the same six hospitals. The diagnostic performance of the probability scale and the risk stratification groupings was tested. The TIPI SROC was compared to the original predictive instrument's SROC, 0.88 to 0.89, respectively. The mean TIPI probability was 59 percent for patients with ischemia and 21 percent for patients without ischemia (p=0.0001), regardless of the triage decision. In the test performance of risk stratification, of the four risk probability groups, 1.6 percent had ACI (0.7 percent AMI) in the low-risk group, 12.0 percent (4.4 percent AMI) had ACI in the lower moderate risk group, 36.7 percent (12.3 percent AMI) had ACI in the higher moderate risk group, and 81.6 percent had ACI (53.3 AMI) in the high-risk group. Physician reviewers assigning diagnosis were blinded to the study's purpose and instrument components.

Data From Other Clinical Studies

No data are reported.

Data From Prospective Clinical Trials in the ED Setting

Studies of test sensitivity and specificity

Table 44. Goldman chest pain protocol to diagnose AMI: diagnostic (more...)

Table 44. Goldman chest pain protocol to diagnose AMI: diagnostic performance studies

Study, year	Study size	Population category	Prevalence of disease (%)		Test performance		Study quality
					Sensitivity (%)	Specificity (%)
Goldman 1982	357	II	ACI AMI	- 15.4	- 90.9	- 69.9	A
Goldman 1988	4,770	II	ACI AMI	26.6 12.1	- 88.0	- 74.2	A
Davison 1990	232	II	ACI AMI	29.7 20.7	ND 1 -	ND 1 -	A
Overall	5,359	II	ACI AMI	26.6-29.7 12.1-20.7	- (88.0-90.9)	- (69.9-4.2)	A

1

See study for SROC curves.

Three studies met the inclusion criteria for the diagnostic performance of the Goldman chest pain protocol (Table 44). Goldman, Weinberg, Weisberg, et al. (1982) tested a computer-derived model using recursive partitioning analysis to predict myocardial infarction in patients with chest pain. Two different patient populations and settings were included. The ED set consisted of 357 patients with varying chest pain unexplained by trauma or radiographs. It was not clear what the inclusion criteria were for the admitted set of 111 patients. The admitted set had a higher prevalence of illness, 27.0 percent to the ED rate of 15.4 percent. Although the diagnostic criteria for unstable angina were well defined, there were no further diagnostic results for this condition reported in the study.

Goldman, Cook, Brand, et al. (1988) tested a modified version of the original Goldman protocol with an added provision to accommodate patients with and without history of cardiac ischemia. The study was evaluated at two university hospitals and four community hospitals. One hundred twenty-eight patients who enrolled were not included in analysis because followup was insufficient for final diagnosis, but in addition, 595 patients were not included because of incomplete forms or insufficient clinical data to complete assignment into subgrouping.

In a small study of 235 patients, Davison, Suchman, and Goldstein (1990) tested three decision aids, as described in Goldman, Weinberg, Weisberg, et al. (1982); Pozen, D'Agostino, Selker, et al. (1984); and Brush, Brand, Acampora, et al. (1985). Specific variables to each of the decision aids were incorporated into a data collection form filled out by the ED residents who were also blinded to the prediction of the aids and to the purpose of the study. It is not known whether the final diagnosis was made blinded to the predictions of the decision aids. There were minor inconsistencies in the number of enrolled patients and patients analyzed. SROC was reported. For the evaluation patients for AMI, none of the protocols could reduce the false-positive rate without reduction of the true-positive rate.

Studies of clinical impact of the test's actual use

Table 45. Goldman chest pain protocol to diagnose AMI: Clinical (more...)

Table 45. Goldman chest pain protocol to diagnose AMI: Clinical impact study

Study, year	Study size		Population category	Prevalence of AMI (%)		Mortality	Study quality
Lee 1995	1,921		II	AMI	6.6	No data	A
	Intervention	924		Intervention	6.9
	Control	997		Control	6.2

One study evaluated the clinical impact of the Goldman protocol (Table 45). Lee, Pearson, Johnson, et al. (1995) evaluated the clinical impact with "a low-intensity, nonintrusive intervention" study at a teaching hospital ED. The time-series design consisted of 6 cycles of 14 weeks each: 5-week intervention followed by 5-week control period separated by 2-week washout periods. The risk estimates and triage recommendations were made available with no human contact or influence by flowcharts and stickers to physicians. There were no differences between the two groups for outcomes of hospitalization rates, length of stay, and estimated costs. Of the intervention group, 50.5 percent (467/924) were hospitalized with 10 percent (92/924) admitted to the CCU. Correspondingly, the control group had 52.2 percent (520/997) hospitalized with 9.5 percent admitted. A low-risk group, defined as up to 7 percent risk for having AMI, included 1,160 patients. This low-risk population had a 1.6-percent CCU admission rate. Mean length of stay for the intervention group was 4.9 total, 3.5 in the CCU; and for the control group, 4.9 total, 4.1 in the CCU. Mean estimated costs for the intervention group were $7,822 vs. $7,955 for the control group. The protocol was amended after the first cycle to include triage recommendations at the request of physicians for guidance in interpreting quantitative risk predictions. Because outcome data were not reported by cycle, it is difficult to detect whether there may have been an effect because of the change in the protocol after commencement. Noted in the study were the differences in decisionmaking at teaching vs. community hospitals and how the intervention intended for a single decisionmaker, as in a community hospital, may not have an effect in a different setting where several physicians are involved in the decisionmaking. Although the cycles did not correlate with the rotation schedules of the physicians, there was no information as to structure or number of the rotations.

Data From Other Clinical Studies

No data are reported.

Data From Prospective Clinical Trials in the ED Setting

Studies of clinical impact of the test's actual use

Table 46. Algorithm/protocol: Clinical impact studies (more...)

Table 46. Algorithm/protocol: Clinical impact studies

Study, year	Study size	Population category	Prevalence of disease (%)		Mortality	Study quality
Gomez 1996	100	III (CPU)	ACI AMI	6.0 2.0	0/100 at 30 days	A
Mikhail 1997	502	III (CPU)	ACI AMI	8.8 2.0	2/374 at 150 days	B
Overall	1,502	III (CPU)	ACI AMI	6.0-8.8 2.0		B

CPU - chest pain unit

Two studies evaluated the clinical impact of algorithms (Table 46). Gomez, Anderson, Karagounis, et al. (1996) evaluated a rapid rule-out protocol based in the ED chest pain evaluation unit. The patient population included was at low risk for acute myocardial infarction as defined by the Goldman method. One hundred patients were randomized into two groups, the test group to undergo the rapid rule-out protocol and the control group who were assigned to standard care of hospital admission and management as required by their physicians. The length of stay and hospital charges were the primary outcomes of study. Some additional endpoints included missed diagnosis and frequency of final diagnosis of ischemia overall and by the study group. By intention-to-treat analysis of 100 patients, the median length of stay and hospital charges were lower for the rule-out protocol group compared with the control group: median 12.1 versus 22.3 hours, and median $895 vs. $1,488, p=0.0001. The analysis of 92 patients with ischemia ruled out was similar in shorter length of stay and charges: median 11.9 vs. 22.8 hours, and median $893 vs. $1,349, p=0.0001, respectively. There were six patients diagnosed with ischemia, one patient in each group for acute myocardial infarction and four patients for unstable angina in the control group. There were no missed diagnoses at 30-day followup. Because of the study population's low event rate, there was inadequate power to show a difference between the two groups. Though blinding was not appropriate, it may be a factor in influencing the control group physicians' choices concerning triage, evaluations, and length of stay.

Mikhail, Smith, Gray, et al. (1997) evaluated a 23-hour chest pain protocol in the chest pain center in a community hospital. All patients underwent a mandatory stress test unless they had end-stage heart disease, in which case they needed cardiac enzyme tests to rule out AMI. Rule-out tests included CK/CK-MB, myoglobin, and continuous ECG. Of the 502 patients transferred from the emergency department, 420 had stress testing consisting of 247 standard-graded stress tests, 161 stress echocardiography, 9 dobutamine echocardiography, and 3 thallium stress tests. A total of 67 patients (13.3 percent) from the 502 enrolled were admitted from the chest pain center. Final diagnosis included 44 with ischemic heart disease, 10 of whom had acute myocardial infarction. Of the 32 patients admitted with positive stress results, 24 had a final diagnosis of ischemia. There was no mortality or AMI for patients discharged with negative findings from the chest pain center at 14 days. At 150 days, 354 (86.0 percent) of the 435 discharged were followed. Two deaths and one AMI were reported. Of two patients with known CAD and initially lost to followup, one had cardiac arrest 2 weeks after chest pain unit (CPU) discharge, and one had AMI 6 days after discharge. One patient with end-stage heart disease and admitted for rule-out protocol only had cardiac arrest. This population was defined as at low to moderate risk for ischemic heart disease. There were discrepancies with numbers reported.

Data From Other Clinical Studies

Table 47. Algorithm/protocol: Diagnostic performance studies (more...)

Table 47. Algorithm/protocol: Diagnostic performance studies

Study, year	Study size	Population category (setting)	Prevalence of disease (%)		Test performance		Study quality
					Sensitivity (%)	Specificity (%)
Lee 1991	2,684	IV (CCU)	ACI AMI	58.2 30.0	Not applicable		A
	957		Low-risk patients:
			ACI AMI	34.6 7.0
Gibler 1995	1,010	IV	ACI AMI	4.3 1.2	Serial ECG (n=1,010)		A
					21.2	99.4
					Echocardiography (n=901)
					47.4	99.0
					Exercise Stress (n=791)
					28.6	99.4
					Serial CK-MB (n=1,010)
					100	98.3
Zalenski 1997a	317	IV	ACI	9.5	90.0	50.5	A
Overall	4,011	IV	ACI AMI	4.3-58.2 1.2-30.0			A

The Gibler, Runyon, Levy, et al. (1995) article is a retrospective review of 1,010 consecutive patients at an urban, tertiary hospital evaluating a 9-hour heart ED program. The protocol included a series of tests including serial CK-MB at presentation, and at 3, 6, and 9 hours and serial ECG at 20-second intervals for 9 hours. At the end of the 9-hour protocol, if there was no indication of evolving ischemia or necrosis, a history and physical examination was conducted by a cardiologist. A two-dimensional echocardiography at rest was conducted by the cardiologist in the ED. This was followed by a graded exercise (maximal Bruce protocol) test. If the exercise test was negative, the patient was discharged and instructed to return in 24 to 48 hours for followup tests. One hundred fifty-three patients were admitted, of whom 52 were diagnosed with cardiac conditions. Twelve of the admitted patients had acute myocardial infarction, and 31 had angina or unstable angina. From the data reported, it would appear that all 1,010 patients completed the serial CK-MB for a sensitivity of 100 percent, 12 of 29 for AMI, specificity 98.3 percent, and serial ECG for a sensitivity of 21.2 percent, specificity 99.4 percent. In the ED, 901 patients underwent echocardiography for a sensitivity of 47.4 percent (9 of 19 true positive results) and a specificity of 99.0 percent. Exercise stress test was conducted for 791 for a sensitivity of 28.6 percent (4 of 14 true positive results) and a specificity of 99.4 percent. The population was selected for being at "low- to moderate-risk" for a rate of ischemia of 4.3 percent.

In the Zalenski, McCarren, Roberts, et al. (1997) study, a 12-hour protocol to diagnose ACI was executed in the ED setting for 359 patients who were at low risk for AMI as defined by the Goldman protocol, but who were also in need of hospital admission. The tests were conducted as follows: serial CK-MB, serial ECG, and clinical assessments concurrently for 12 hours. Exercise stress ECG was conducted if the previous tests and examinations were negative. All patients were admitted to establish reference diagnoses. The inclusion/exclusion criteria were modified after the first year of the 29-month enrollment period by lowering the age from 30 years to 20 years and eliminating the restriction on duration of pain from less than 30 seconds to 72 hours. It should be noted that 43.5 percent of the 317 evaluable patients had chest pain lasting fewer than 24 hours, 16.6 percent had pain from 24 to 48 hours, and 39.9 percent had pain for more than 48 hours. Thirty-one patients had incomplete enzyme determinations, and three withdrew from the study. Another eight patients had noncardiac diagnoses.

Data From Prospective Clinical Trials in the ED Setting

Studies of test sensitivity and specificity

Table 48. Computer-based decision aids to diagnose (more...)

Table 48. Computer-based decision aids to diagnose AMI: Diagnostic performance studies

Study, year	Study size	Population category	Prevalence of AMI (%)	Test performance		Study quality
				Sensitivity (%)	Specificity (%)
Tierney 1985	540	II	11.5	81	86	C
Baxt 1991	331	II	10.6	97.2	96.3	A
Dilger 1992	122	I/II	32.8	89	86	A
Jonsbu 1993	1,252	II	41.7	98.3	58.9	A
Baxt 1996	1,070	II	7.0	96.0	96.0	A
Kennedy 1997 Center 1: Center 2:	200 91	II (admitted)	33.8 1	91.2	90.2	A
		III (admitted)	23.1	52.4	80.0	A
Overall	3,606	I/II/III	7-42	52-98	59-96	A

1

200 patients validation data set plus 90 patients used in training of artificial neural network at center 1.

There were six diagnostic performance studies that met the inclusion criteria for computer-based decision aids in the ED setting (Table 48).

Tierney, Roth, Psaty, et al. (1985) validated the diagnostic performance of a predictive model on 540 ED patients from an urban hospital. The logistic regression model uses four dichotomous variables obtained by a questionnaire completed by the physician. CK/CK-MB, LDH isoenzyme-1, or ECG results were used for the diagnosis of AMI. The sensitivity and specificity could not be verified with the data reported. Of the 655 enrolled in the study, 115 were lost to followup and omitted from analyses. Of the 540 evaluable patients, 284 were used to derive the model.

Baxt (1991) prospectively validated the diagnostic performance of an artificial neural network on physician-completed questionnaires of 331 patients presenting with anterior chest pain. The information on the questionnaire was based on patient history, symptoms, and ECG results. The final diagnosis of AMI was based on CK/CK-MB, LDH isoenzyme-1, or ECG results or sestamibi imaging results. Twenty-four patients of 355 enrolled could not be followed and were therefore omitted from the study. The followup data inputted were also blinded to the initial data entry.

Dilger, Pietsch-Breitfeld, Stein, et al. (1992) prospectively validated a predictive model based on six variables obtained by physician-completed questionnaires. A population of 122 was evaluated for AMI with the final diagnosis made by one of the investigators based on "anamnestic, electrocardiographic, and enzymatic infarction criteria without knowledge of predictive result by the model." Included as one of the six variables in the predictive model is CK level at presentation, which is also used as part of the final diagnosis. The authors note that, for final diagnoses, the time course of 22 hours of total CK and CK-MB levels were utilized. The sensitivity and specificity could not be verified with the data reported.

Jonsbu, Aase, Rollag, et al. (1993) applied a computer-derived decision support system, which was based on case history data, to identify, from ED patients presenting with chest pain, those in need of CCU admissions and patients who may have AMI. Final diagnoses from ECG results or isoenzymes levels were made blinded to the test under investigation. The sensitivity and specificity could not be verified with the data reported.

Baxt and Skora (1996) tested the diagnostic performance of the artificial neural network as reported in Baxt (1991) on a larger patient population of 1,070 at a major, urban teaching hospital. This was conducted in a similar fashion as the preliminary study, with the individuals involved in the reference and the test diagnoses blinded to the other's results. The reference criteria were also based on CK and CK-MB levels or ECG results. The network does not base its diagnosis of AMI on ECG changes, as 33 of the 75 patients had no such changes and 72 AMI patients were correctly identified by the network. Thirty patients who were diagnosed negative by the network were lost to followup.

Kennedy, Harrison, Burton, et al. (1997) used two centers, the first center to train and prospectively validate an artificial neural network and the second center to test its application and its portability. As noted by the authors, the two patient populations were very different. The patient inclusion/exclusion criteria for the training center were minimal, "non-traumatic chest pain." But this population was also from the department of medicine admissions and may have been at high risk for AMI. The authors considered the test group at the second center to be a difficult group to diagnose because of the relatively strict inclusion/exclusion criteria of chest pain under 4 hours upon presentation and nondiagnostic ECGs. It also appears that the second group was admitted after data were collected in the ED. The artificial neural network is a computer program consisting of a set of processing units with data entered directly into the computer with the aid of data entry screens. The final diagnosis for AMI was based on two of three criteria: clinical history, ECG, or biochemical results. The reviewers were blinded to the results of the network. The differences in sensitivity and specificity between the two groups may be indicative of differences in population risk for AMI. The sensitivity and specificity also could not be verified with the data reported.

Studies of clinical impact of the test's actual use

Table 49. Computer-based decision aids in prehospital setting: (more...)

Table 49. Computer-based decision aids in prehospital setting: Clinical impact study

Study, year	Study size	Population category	Prevalence of disease (%)		Mortality (%)	Study quality
Grijseels 1996	977	III 1 (Prehospital)	ACI AM	47.8 29.9	2.6 at 30 days	A

1

Prehospital subjects preselected by their general practitioner.

Grijseels, Deckers, Hoes, et al. (1996) implemented a decision rule for 1,020 prehospital patients enrolled by their general practitioners in the Netherlands (Table 49). The decision rule recommended 750 for hospitalization, of whom 427 had ACI. Of the 227 not recommended for hospitalization, general practitioners admitted 128, of whom 32 had ACI (25 percent). Ultimately, 8 patients with ACI were missed, and there were 25 deaths total. The general practitioners decided not to admit 19 patients, 3 of whom developed AMI.

Data From Other Clinical Studies

Table 50. Computer-based decision aids in prehospital setting: (more...)

Table 50. Computer-based decision aids in prehospital setting: Diagnostic performance study

Study, year	Study size	Population category	Prevalence of disease (%)		AMI test performance		Study quality
					Sensitivity (%)	Specificity(%)
Grijseels 1996	977	III 1 (Prehospital)	ACI AMI	47.8 29.9	91.4 -	36.7 -	B

1

Prehospital subjects preselected by their general practitioner.

Creatine Kinase (CK), Single and Serial Measurements

Only two studies have evaluated serial CK testing. Both used broad inclusion criteria but enrolled populations in which the prevalence of AMI was moderate to high. Test sensitivity was high (95 to 99 percent) in serial tests performed over about 15 hours after presentation to the ED (or from the onset of symptoms), but was only modest (69 percent) in the one study that drew serial samples for 4 hours. Test specificity was modest in both studies (68 and 84 percent).

As a single test, CK is insensitive and only modestly specific for AMI. Serial testing appears to have a higher sensitivity, although the specificity remains modest. However, the evidence is insufficient to evaluate serial CK measurements over a short time. Because high serum CK levels represent infarcted myocardium, CK has not been evaluated for diagnosing ACI in the ED. There were no clinical impact studies for CK.

Creatine Kinase Subunit (CK-MB), Single and Serial Measurements

As is the case with CK, the total sample size and number of studies on a single CK-MB measurement at presentation to the ED are large. The evidence suggests that the sensitivity of single CK-MB for AMI is low (44 percent), although specificity is high (96 percent). Studies reported a broad range of sensitivity for diagnosing AMI. Again, as is the case for CK, limited evidence suggested that the sensitivity of CK-MB depends on the duration of the patient's symptoms; sensitivity increases with longer symptom duration. In general, studies reported a narrow range (92 percent to 99 percent) of test specificity. Test performance across studies did not appear to vary by type of hospital, inclusion criteria, AMI prevalence, or test threshold.

The total sample size and number of studies of serial tests for CK-MB in the ED setting are large. Overall, serial testing had a modest sensitivity (87 percent) and a high specificity (96 percent) for AMI. However, test sensitivity was strongly related to the timing of serial testing. All studies that performed serial testing for at least 4 hours after presentation to the ED (or until at least 8 hours after symptom onset) found test sensitivity to be greater than 90 percent. Conversely, all studies that performed serial testing to at most 3 hours found test sensitivity to be lower than 90 percent. The pooled sensitivity for serial testing to at least 4 hours was 96 percent; the pooled sensitivity for serial testing until 3 hours was only 81 percent. In general, test specificity was in a narrow range across studies and was above 90 percent.

CK-MB as a single test was only modestly sensitive and specific for AMI; however, serial testing performed over 4 to 9 hours was highly sensitive and highly specific. Because serum CK-MB levels represent infarcted myocardium, CK-MB has not been tested for diagnosing ACI in the ED. There were no clinical impact studies for CK-MB.

Troponin T and Troponin I

The evidence for the diagnostic performance of troponin T is substantial for diagnosing AMI but rather limited for diagnosing ACI. Data for troponin I are limited, but its performance was similar to that of troponin T. The sensitivity of presentation troponin T for diagnosing AMI in the ED was poor, but it improved substantially if serial measurements were obtained for up to 6 hours after ED presentation. Most likely, the sensitivity was better for patients who have had symptoms for longer periods of time. The specificity of troponin T for AMI was in the range of 90 percent.

Myoglobin

The diagnostic performance of myoglobin has been well studied for diagnosing AMI, but not for diagnosing ACI. The performance of myoglobin was similar to that of CK-MB, except that serial testing may have provided adequately high sensitivity earlier. The sensitivity of myoglobin for diagnosing AMI in the ED was poor (49 percent) when a single initial measurement was obtained, but sensitivity improved greatly if a second measurement was obtained 2 to 4 hours after the first one (>86 percent). However, the sensitivity for patients only recently symptomatic was poor, and a second measurement in 2 to 4 hours may still not have been sufficiently sensitive to be useful. Specificity was very good, but not excellent. Its deviation from excellence in the various reports probably depended on the extent to which other reasons for elevated myoglobin were excluded a priori. There is limited evidence to suggest that the failure of myoglobin level to increase over 1 to 2 hours was excellent at excluding AMI (specificity >94 percent).

Other Biomarkers

Studies on P-selectin and malondialdehyde-modified low-density lipoprotein are just beginning to appear. There was only one ED study of P-selectin that reported low sensitivity and low specificity for AMI.

Combination CK-MB and Myoglobin

The only combination of biomarkers to diagnose AMI that has been reported is CK-MB and myoglobin, and this in only three studies. In addition, it appears that the decision to analyze this combination was made post hoc, and thus the findings may have overestimated true test performance (as studies with poor test performance may have been less likely to report their finding). That said, combination CK-MB and myoglobin at ED presentation (where a positive test was defined as either CK-MB or myoglobin being elevated) had good test performance, with both a high sensitivity (83 percent), though poorer specificity (82 percent), as compared with presentation CK-MB or myoglobin alone. Serial combination CK-MB and myoglobin (where a positive test was defined as any of the four serum samples being elevated) performed similarly to the individual serial biomarkers, although sensitivity may be higher (100 percent).