Systematic review, meta-analysis and economic modelling of diagnostic strategies for suspected acute coronary syndrome

Electronic databases

All searches were undertaken by an information specialist (PE) in November 2010. Studies were identified by searching the following electronic databases:

• MEDLINE (via Ovid SP) 1950–

• MEDLINE In-Process & Other Non-Indexed Citations (via Ovid SP) 1950–

• Cumulative Index of Nursing and Allied Health Literature (CINAHL) (via EBSCO) 1981–

• EMBASE (via Ovid SP) 1980–

• Web of Science (WoS) (includes Science Citation Index and Conference Proceedings Citation Index) [via Web of Knowledge (WoK)] 1899–

• Cochrane Central Register of Controlled Trials (CENTRAL)

• Cochrane Database of Systematic Reviews (CDSR)

• NHS Database of Abstracts of Reviews of Effects (DARE)

• Health Technology Assessment database (HTA).

Sensitive keyword strategies using free text and, where available, thesaurus terms using Boolean operators and database-specific syntax were developed to search the electronic databases. For the biochemical markers reviews, synonyms relating to the population (e.g. chest pain, ACS or MI) were combined with terms for the biochemical markers of interest, and the reference standard (troponin), and a search filter aimed at restricting results to studies of either diagnostic accuracy or prognosis (used in the searches of MEDLINE, CINAHL and EMBASE). For the CTCA and exercise ECG review, synonyms relating to the population (e.g. chest pain, ACS or MI) were combined with terms for the diagnostic tests, and the reference standard (coronary angiography) or outcomes (e.g. MACE), and a search filter aimed at restricting results to studies of either diagnostic accuracy or prognosis (used in the searches of MEDLINE, CINAHL and EMBASE). Date limits or language restrictions were not used on any database for either review. All resources were searched from 1985 (CTCA review) or 1995 (biomarkers review) to November 2010. Examples of the MEDLINE search strategy for each review is provided in Appendix 1.

Other resources

To identify additional published, unpublished and ongoing studies, the reference lists of all relevant studies (including existing systematic reviews) were checked. In addition, key experts in the field were approached to identify any relevant citations missed by the search methods applied.

All identified citations from the electronic searches and other resources were imported into and managed using the Reference Manager bibliographic software (version 12.0; Thomson Reuters, Philadelphia, PA, USA).

Study design

All prospective diagnostic cohort studies comparing a relevant index test (biochemical markers or CTCA/ exercise ECG) to the required reference standard for the relevant outcome (MI or CAD) were included in their relevant review. All studies examining the prognostic value of a relevant index text (biochemical markers or CTCA) for at least 30 days' follow-up for MACEs were included, regardless of the reference standard used. Case–control studies (i.e. studies in which patients were selected on the basis of the results of their reference standard test) were excluded.

Population

To be included, a study had to assess adults presenting with suspected ACS. Studies were excluded if patients were selected on the basis of having a clinical diagnosis of ACS (rather than a clinical suspicion of ACS) or positive diagnostic test for ACS, such as ST deviation on the ECG or an elevated biomarker. Studies of patients selected on the basis of a negative diagnostic test were included [e.g. studies that excluded patients with ST elevation myocardial infarction (STEMI)].

Index tests

For the biochemical markers review, the index test included any test assessing the following markers individually or in combination:

• adrenomedullin

• BNP or NT-pro-BNP

• copeptin

• CRP

• galectin-15

• H-FABP

• interleukin 33

• IMA

• matrix metalloproteinase 9 (MMP9)

• MPO

• myoglobin

• PAPP-A

• ST-2

• TnI or TnT.

Studies were only included in the diagnostic accuracy review if the index test was measured at or before patient arrival at hospital. We excluded prognostic studies that only evaluated troponin (or other biomarkers not included in the review, such as CK and CK-MB).

For the second diagnostic review, the index test was either CTCA, regardless of sensitivity (e.g. 64 or 16 slices) or exercise ECG.

Target condition

The target conditions or outcomes of the reviews of biochemical markers were:

• Diagnostic review Acute MI defined according to the universal definition. 7

• Prognostic review MACE, defined as including at least cardiac death and non-fatal MI (individually or as a composite).

The target conditions or outcomes of the review of CTCA and exercise ECG were:

• Diagnostic review CAD identified on ICA.

• Prognostic review MACE, defined as including at least cardiac death and non-fatal MI (individually or as a composite).

Reference standards

Acute MI was defined according to the universal definition and required TnI or TnT measurement for at least 80% of the population at least 6 hours after symptom onset. If the reference standard was any biomarker other than troponin the study was excluded from the diagnostic review. Many studies reported composite diagnostic reference standards or a diagnostic standard of ACS, which included clinically diagnosed ACS, development of ECG changes or a subsequent MACE. Where possible we attempted to extract data for MI according to our definition. If this was not possible we made a judgement whether or not the reference standard approximated to our definition of MI. We included studies that used only new diagnostic ECG changes or outcome-based MACE (e.g. death, non-fatal MI or life-threatening arrhythmia) alongside a troponin-based reference standard. We excluded studies that used clinically diagnosed ACS (i.e. by history and examination findings alone), undefined or any ECG changes, or process-based MACE (e.g. coronary reperfusion) in the reference standard.

Coronary artery disease was determined by ICA and defined in accordance with the primary study. Studies were excluded if coronary angiography was performed only in selected patients, such as those with positive CTCA or exercise ECG. The definition of MACEs required that at least 80% of the cohort be followed for at least 30 days and that a MACE included, at least, cardiac death and non-fatal MI.

Outcomes

Sufficient data were required to construct tables of diagnostic test performance, i.e. numbers of true-positives (TPs), false-negatives (FNs), false-positives (FPs) and true-negatives (TNs). If raw numbers were not reported we attempted to calculate these data from sensitivity and specificity, using prevalence and total number analysed to calculate the denominators. Studies were excluded from analysis as ‘unable to extract data’ if these calculations were not possible or yielded markedly inconsistent data.

Overview of biomarker studies included in the diagnostic review

Table 2 lists all the studies included in the diagnostic accuracy review and the biomarkers that were evaluated with extractable data. We were not able to extract data for all the biomarkers reported in each study. Table 2 lists only the biomarkers with extractable data.

Description of diagnostic studies of presentation troponin

We identified 21 diagnostic studies19,20,48–50,52–57,59,62–64,66,70,72,74,77,81 of presentation TnI and 11 studies19,46,58,60,62,67,71,73,76,78,82 of TnT for inclusion in the review. Two studies19,62 evaluated TnI and T. The characteristics of the study populations are outlined in Tables 3 and 4, whereas details of the index and reference standard test definitions are provided in Tables 5 and 6. Some studies evaluated more than one assay, so assays are reported separately in Tables 3 and 4. Reporting of inclusion and exclusion criteria were variable and several studies excluded patients with a diagnostic ECG. Prevalence of MI varied from 5% to 73% and was relatively high, suggesting that patient cohorts may have been subject to implicit selection processes. Time delay from symptoms to presentation varied from 1.2 hours (mean) to 6 hours (median). Several studies reported data using different diagnostic thresholds for the index test. Where this was done we extracted data for threshold based on the 99th percentile, 10% coefficient of variation (CV) and limit of detection (LoD). In accordance with our inclusion criteria, all studies used the universal definition of MI as the reference standard, and most reported using some form of adjudication, taking into account the results of troponin testing. In most cases the troponin used for the reference standard was a standard (i.e. not high sensitivity) assay using the 10% CV or 99th percentile as a diagnostic threshold. However, the study by Christ et al. 60 reported the use of a reference standard based on high-sensitivity TnT (HsTnT) alongside a reference standard based on the standard assay. For this study we extracted data based on the standard assay reference standard.

Quality assessments of diagnostic studies of presentation troponin

Figures 3 and 4 show the quality assessments for studies of TnI and TnT, respectively, whereas Figures 5 and 6 show the methodological quality summaries. The studies were generally high quality, perhaps reflecting exclusion of lower-quality studies by our selection criteria. Presentation troponin measurement is obviously not independent of a troponin-based reference standard, so our assessment of verification bias focused on whether or not the index and reference standard troponin were measured on different samples. There was some uncertainty about whether index and reference standard tests were assessed blind. This is not likely to have influenced reporting of the index test as in most cases this was a mechanised process producing a quantitative result. However, bias could have resulted if reference standard adjudicators were aware of the presentation troponin result (detection bias). The only other possible issue was the timing of the reference standard, which was not always explicit.

FIGURE 3.

Quality assessment of diagnostic studies of TnI.

FIGURE 4.

Quality assessment of diagnostic studies of TnT.

FIGURE 5.

Methodological quality summary of diagnostic studies of TnI.

FIGURE 6.

Methodological quality summary of diagnostic studies of TnT.

Analysis of diagnostic accuracy studies of presentation troponin

Tables 7 and 8 show the reported sensitivity and specificity of each assay at key thresholds in each study of TnI and TnT, respectively. The studies used a variety of different assays and thresholds for positivity. In consequence, there is a wide range of reported values for sensitivity and specificity.

We did not undertake meta-analysis across all studies because of variation in the assays and thresholds used, with some studies using high thresholds with no clear basis. Instead, we undertook separate analyses for TnI and TnT using the 99th percentile or 10% CV threshold, when these data were reported (Figures 7–10). The studies by Christ and Popp60 and Reichlin et al. 19 reported data for more than one assay in each potential analysis, so we selected data from one assay in each analysis. We also analysed the following high-sensitivity assays using the 99th percentile (Figures 11–13):

1. ADVIA Centaur Ultra troponin I (Siemens Healthcare, Basel, Switzerland)

2. Abbott Architect troponin I (Abbott Laboratories, IL, USA)

3. Roche hsTnT. (Roche Diagnostics, Basel, Switzerland).

FIGURE 7.

Meta-analysis of studies of Tnl using the 99th percentile.

FIGURE 8.

Meta-analysis of studies of Tnl using 10% CV.

FIGURE 9.

Meta-analysis of studies of TnT using the 99th percentile.

FIGURE 10.

Meta-analysis of studies of TnT using 10% CV.

FIGURE 11.

Meta-analysis of studies of ADVIA Centaur Ultra Tnl.

FIGURE 12.

Meta-analysis of studies of Abbott Architect Tnl.

FIGURE 13.

Meta-analysis of studies of Roche HsTnT.

The results show that sensitivity and specificity for TnI were 77% (95% predictive interval 29–96%) and 93% (95% predictive interval 46–100%), respectively, when the 99th percentile was used and 82% (95% predictive interval 40–97%) and 93% (95% predictive interval 74–98%) when the 10% CV was used. This apparently counterintuitive finding (lower sensitivity at a lower diagnostic threshold) is probably explained by either random error or differences in the study populations or assays included in the two analyses. The corresponding results for TnT were 80% (95% predictive interval 33–97%) and 91% (95% predictive interval 53–99%) when the 99th percentile was used and 74% (95% predictive interval 34–94%) and 96% (95% predictive interval 76–99%) when the 10% CV was used, suggesting that different thresholds provide a trade-off between sensitivity and specificity. The differences between point estimates of sensitivity and specificity for TnI and TnT probably reflect differences in the assays or thresholds evaluated, the constituent study populations or random error, rather than a systematic difference between TnI and TnT. The credible ranges for the estimates differed markedly depending upon whether the pooled effect or predictive effect was estimated, reflecting the marked heterogeneity between the studies. The predictive distribution is likely to provide the most appropriate reflection of uncertainty and is used in the cost-effectiveness modelling.

The high-sensitivity assays unsurprisingly had higher sensitivity but lower specificity, although with considerable uncertainty reflected in the wide predictive intervals for their estimates. The Roche HsTnT assay had a sensitivity of 96% (95% predictive interval 27–100%) and a specificity of 72% (95% predictive interval 3–96%). The ADVIA Centaur Ultra-TnI assay had a sensitivity of 86% (95% predictive interval 22–96%) and a specificity of 89% (95% predictive interval 40–97%). The Abbot Architect troponin I assay had a sensitivity of 83% (95% predictive interval 58–95%) and a specificity of 95% (95% predictive interval 67–100%).

These analyses compared high-sensitivity troponin index tests with a reference standard based on a standard troponin assay. We identified one study60 that compared HsTnT at presentation with a reference standard based on HsTnT, as well as a standard TnT assay. The sensitivity and specificity of HsTnT were 95.0% and 61.5%, respectively, compared with a reference standard based on the standard assay, and were 94.3% and 69.6%, respectively, compared with a reference standard based on the high-sensitivity troponin assay. These findings suggest that the higher sensitivity and lower specificity of high-sensitivity assays compared with standard troponin assays are not simply due to different assays being used for index test and reference standard, but represent a genuine improvement in early sensitivity at the expense of specificity for a final diagnosis of MI. The lower specificity of high-sensitivity assays may be due to a greater ability to detect myocardial injury secondary to other clinical conditions. 83,84

Description of diagnostic studies of heart-type fatty acid-binding protein

We identified 17 diagnostic studies49,50,57–59,63,64,67,68,72–78,82 of H-FABP for inclusion in the review. Table 9 shows the population characteristics and Table 10 shows the index and reference standard test characteristics. As with the troponin studies, reporting of exclusion criteria were variable, with some studies excluding patients with diagnostic ECG changes. The prevalence of MI varied from 15% to 73% and was relatively high, suggesting some selection of higher risk cases. The time from symptom onset to sampling varied from 1.2 hours (mean) to 5.9 hours (median). Around half of the studies evaluated qualitative assays, most specifying that this was the CardioDetect assay with a diagnostic threshold of 7 µg/l. The threshold used by quantitative assay was variable. Most of the studies used reference standards based on a standard modern troponin assay, using the 10% CV or 99th percentile as a diagnostic threshold, although not all gave details of the assay and threshold.

Quality assessment of diagnostic studies of heart-type fatty acid-binding protein

Figure 14 shows the quality assessments for studies of H-FABP, while Figure 15 shows the methodological quality summary. The overall quality and the issues raised were similar to those for the studies of troponin.

FIGURE 14.

Quality assessment of diagnostic studies of H-FABP.

FIGURE 15.

Methodological quality summary of diagnostic studies of H-FABP.

Analysis of diagnostic studies of heart-type fatty acid-binding protein

Figure 16 shows the meta-analysis of the studies of quantitative H-FABP and Figure 17 shows the meta-analysis of qualitative assays. The summary estimates of sensitivity and specificity were 81% (95% predictive interval 50–95%) and 80% (95% predictive interval 26–98%), respectively, for the quantitative assays and 68% (95% predictive interval 11–97%) and 92% (95% predictive interval 20–100%), respectively, for the qualitative assays.

FIGURE 16.

Meta-analysis of quantitative H-FABP assays.

FIGURE 17.

Meta-analysis of qualitative H-FABP assays.

Description of diagnostic studies of ischaemia-modified albumin

We identified four studies48,61,68,70 that were eligible for inclusion in the review (Tables 11 and 12). A number of other studies of IMA were excluded because the reference standard was ACS, based on clinical criteria, and cases with MI were not reported separately. Two studies restricted recruitment to patients presenting within 370 and 8 hours61 of symptom onset. Only one study48 reported the median time delay from symptom onset. Thresholds of between 75 and 91 were used for IMA. Three studies used a modern standard troponin assay for the reference standard, whereas the older study from Christensen et al. 61 inevitably used an older troponin reference standard with a higher threshold for positivity.

Quality assessment of diagnostic studies of ischaemia-modified albumin

Figure 18 shows the quality assessments for studies of IMA, whereas Figure 19 shows the methodological quality summary.

FIGURE 18.

Quality assessment of diagnostic studies of IMA.

FIGURE 19.

Methodological quality summary of diagnostic studies of IMA.

Analysis of diagnostic studies of ischaemia-modified albumin

Figure 20 shows the results of meta-analysis of studies of IMA. The summary estimates of sensitivity and specificity were 77% (95% predictive interval 19–98%) and 39% (95% predictive interval 2–95%), respectively.

FIGURE 20.

Meta-analysis of studies of IMA.

Description of diagnostic studies of myoglobin

We identified 13 diagnostic studies18,49,52,57,58,63,64,68,69,71,74,77,79 of myoglobin for inclusion in the review. Table 13 shows the population characteristics and Table 14 shows the index and reference standard test characteristics. Reporting of exclusion criteria was variable and some studies excluded patients with diagnostic ECG changes. The prevalence of MI was generally high and varied from 18% to 73%. The median time from symptom onset to sampling varied from 2.2 to 7.8 hours. There was no consistency in the diagnostic threshold used. It ranged from 51 to 150 µg/l and 5 out of 13 studies used different thresholds for men and women, whereas 8 out of 13 studies used the same threshold. Several studies used relatively high thresholds for positivity for the reference standard troponin or did not report the timing of sampling or the assay used.

Quality assessment of diagnostic studies of myoglobin

Figure 21 shows the quality assessments for studies of myoglobin, whereas Figure 22 shows the methodological quality summary. The quality assessment of acceptability of the reference standard was limited to whether or not the reference standard criteria were reported clearly and met the universal definition. Although most studies had an acceptable reference standard by these criteria, it is debatable whether the troponin assays and threshold used represented best current practice in some cases.

FIGURE 21.

Quality assessment of diagnostic studies of myoglobin.

FIGURE 22.

Methodological quality summary of diagnostic studies of myoglobin.

Analysis of diagnostic studies of myoglobin

Figure 23 shows the results of meta-analysis of studies of myoglobin. The summary estimates of sensitivity and specificity were 62% (35–83%) and 83% (35–98%), respectively.

FIGURE 23.

Meta-analysis of studies of myoglobin.

Description of diagnostic studies of other biomarkers

We identified 10 diagnostic studies46,47,51,55–57,65,71,80,81 of other biomarkers. Table 15 shows the population characteristics and Table 16 shows the index and reference standard test characteristics. The prevalence of MI was lower than the studies of troponin, H-FABP and myoglobin, and varied from 5% to 29%. The median time from symptom onset to sampling varied from 2 to 4.5 hours. Most of the studies used a modern troponin assay with an acceptable threshold for the reference standard.

Quality assessment of diagnostic studies of other biomarkers

Figure 24 shows the quality assessments for studies of other biomarkers, whereas Figure 25 shows the methodological quality summary.

FIGURE 24.

Quality assessment of diagnostic studies of other biomarkers.

FIGURE 25.

Methodological quality summary of diagnostic studies of other biomarkers.

Analysis of diagnostic studies of other biomarkers

The studies reported four analyses of MPO,55,57,65,81 two each of BNP,55,56 CD40L,46,55 copeptin71,80 and CRP,51,55 and one each of MMP9,55 NT-pro-BNP55 and PAPP-A. 46 No two analyses evaluated the same biomarker at the same threshold. The data were therefore insufficient for meaningful meta-analysis. Sensitivity and specificity of each biomarker in each analysis are reported in Table 17. Overall, diagnostic accuracy was modest. Sensitivity exceeding 0.8 was achieved only at the expense of specificity. None of these analyses suggests that the biomarker in question could be used as a single test for early diagnosis of MI.

Diagnostic studies of biomarkers in combination with troponin

Nine studies48,55,57,67,70,71,76,77,80 reported 11 analyses of the sensitivity and specificity of biomarkers in combination with troponin, compared with troponin alone. These are outlined in Table 18. We did not undertake meta-analysis because no combination was evaluated in more than two studies. In most cases, troponin and the alternative biomarker were combined by classifying the combination as positive if either test was positive. However, the study by Apple et al. 55 classified the combination as positive only if both tests were positive. Thus, in most studies the combination had higher sensitivity and lower specificity than troponin alone, whereas the combinations tested by Apple et al. 55 had lower sensitivity and higher specificity than troponin alone.

These studies show that combining troponin with another biomarker at presentation, with elevation of either biomarker producing a positive test, results in markedly improved sensitivity but with a loss in specificity that can be substantial. None of these analyses uses a high-sensitivity troponin assay. The results of the troponin meta-analysis suggest that a similar improvement in sensitivity at the expense of specificity can be achieved if a lower threshold for troponin positivity is used.

Summary of the findings of the diagnostic biomarker review

The sensitivity and specificity of troponin measurement at presentation depends on the assay used and the threshold for positivity. High-sensitivity assays using the 99th percentile as the threshold for positivity can achieve sensitivity at presentation close to, or exceeding, 90%. However, maximising early sensitivity involves some loss of specificity. Only one study60 compared presentation testing with a high-sensitivity assay with a reference standard based on a high-sensitivity assay and showed that the loss of specificity did not seem to be explained by using a standard troponin assay in the reference standard.

Many other biomarkers have been tested for their ability to detect MI at presentation but of those we set out to investigate only myoglobin and H-FABP have been evaluated against an acceptable reference standard in a large number of studies. In general, the alternative biomarkers had inadequate diagnostic accuracy to act as a single diagnostic test for MI at presentation. When used in combination with troponin a number of biomarkers (H-FABP, copeptin, IMA and myoglobin) improved sensitivity for MI at presentation, but at the expense of loss of specificity. Similar changes in sensitivity and specificity can be achieved with troponin as a single test by using a high-sensitivity assay.

Overview of biomarker studies included in the prognostic review

We identified 44 studies46–51,85–122 for inclusion in the prognostic accuracy review. These are listed along with the relevant biomarkers in Table 19. We have only listed the biomarkers identified for our review. Some studies evaluated additional biomarkers. Five studies46,48–51 reported both prognostic and diagnostic data. Two studies44,45 were subsequently excluded because data could have overlapped with other included studies. 85,86

Description of studies included in the prognostic biomarker review

The characteristics of the included studies are outlined in Table 20. Most studies did not report selection criteria beyond those needed to define acute chest pain or suspected ACS. However, some studies excluded patients with high clinical risk, ECG changes of ischaemia or positive admission troponin or CK-MB. The duration of follow-up ranged from the duration of inpatient stay to 5 years. Definitions of MACEs varied between studies, with some studies predicting only mortality, whereas others predicted a range of outcomes. Where more than one definition of a MACE was used or more than one time point for follow-up was reported, we used the most inclusive definition and the longest duration of follow-up.

Quality assessment of studies included in the prognostic biomarkers review

Table 21 shows the results of quality assessment. Nearly all the studies reported adequately, defined MACEs in the methods section, did not incorporate presenting diagnosis in the definition of a MACE and achieved adequate follow-up. However, only around half undertook analysis that went beyond testing or estimating the association between the biomarker and a MACE, and only a minority tested whether or not the biomarker added prognostic value to that provided by troponin.

Analysis of prognostic biomarker studies

Table 22 shows the main univariate analyses reported in the prognostic biomarker studies, i.e. any analysis that tested or estimated the association between a biomarker and a MACE. There was substantial variation in the analyses reported. Some only used a hypothesis test for the association between a biomarker and a MACE, others estimated parameters [sensitivity, specificity or area under receiver operating characteristic (AUROC)] for discriminating between patients with and without MACEs, and others estimated the odds ratio (OR), RR or hazard ratio (HR) for MACEs for quartiles of the biomarker or a biomarker level above a specified threshold. Many of these analyses report a significant association but they are of limited value because they do not tell us whether or not the biomarker in question provides prognostic information beyond that already available from clinical assessment, ECG and troponin measurement.

Some of the studies used multivariate analysis to adjust for known predictors of MACEs and determine whether or not the biomarkers predicted a MACE when other variables were taken into account. These are shown in Table 23. If troponin was included as a covariate then this analysis could potentially show whether the biomarker provided additional prognostic information to troponin. The findings showed some evidence that BNP, NT-pro-BNP, MPO and H-FABP can provide prognostic value when other predictor variables are taken into account, whereas results for CRP and myoglobin were mixed.

Table 24 shows whether or not the biomarker predicts MACEs in troponin-negative patients. This is probably the most useful analysis because troponin measurement is likely to be routine practice in most settings. Unfortunately, only a few studies reported this analysis so it is difficult to draw conclusions. However, there is some evidence that CRP, PAPP-A and H-FABP can predict MACEs in troponin-negative patients.

Summary of the findings of the prognostic biomarker review

A variety of different biomarkers have been studied and an association shown between increased levels and risk of MACEs, but it is not clear in most cases whether or not this adds useful prognostic information beyond that available from clinical assessment, ECG and troponin. There is some evidence that BNP, NT-pro-BNP, MPO and H-FABP can provide additional prognostic value beyond troponin, whereas CRP, PAPP-A and H-FABP can predict MACEs in troponin-negative patients. However, these findings are based on a small number of heterogeneous studies and the utility of this prognostic value is unclear.

Diagnostic studies of computed tomographic coronary angiography

Table 25 shows the characteristics of the diagnostic studies that compared CTCA with a reference standard of ICA for CAD. The studies were relatively small (n = 31 to 113). Mean age varied from 53 to 62 years, and men outnumbered women in all studies. Most studies explicitly excluded patients with diagnostic ECG changes. The threshold for diagnosing obstructive CAD was 50% stenosis in all studies, except for the study of Sato et al. ,129 which used a threshold of 75% for both tests.

Figure 27 shows the quality assessment and Figure 28 the methodological quality summary of diagnostic studies of CTCA. Study quality was generally high, although blinding of interpretation of the index or reference standard test was unclear or absent in around half of the studies.

FIGURE 27.

Quality assessment of diagnostic studies of CTCA.

FIGURE 28.

Methodological quality summary of diagnostic studies of CTCA.

Figure 29 shows the result of meta-analysis of CTCA diagnostic studies. The summary estimates of sensitivity and specificity were 93% (95% predictive interval 61% to 99%) and 87% (16% to 100%), respectively. The highest sensitivity and specificity was achieved in the only study of 64-slice C T. 126 Two studies124,125 reported markedly lower specificity. The variation in specificity may be explained by artifact due to calcification, movement or heart rate, which may be more common or more variable in patients presenting with acute symptoms.

FIGURE 29.

Meta-analysis of CTCA diagnostic studies.

Prognostic studies of computed tomographic coronary angiography

Table 26 shows the characteristics of the seven prognostic studies of CTCA and one study of CAC scoring. Three of the cohorts (four studies) were compared with control groups in a trial,131–134 whereas the others were single cohort studies. All of the CTCA studies used 64-slice CT. The cohorts were generally larger (n = 30–588) and the mean age (46–56 years) younger than the diagnostic studies. This reflects the inclusion criteria that generally selected low- to intermediate-risk patients. Those with ECG changes and positive biomarkers were usually explicitly excluded. The diagnostic classification for CTCA either dichotomised scans into obstructive (> 50% stenosis) or non-obstructive (< 50%), or limited positive scans to those with stenosis > 70% and used an intermediate category for stenosis of 26–69% or 50–70%. Duration of follow-up ranged from 30 days to 2 years. Definitions of MACEs varied, with most studies including revascularisation in the definition but two limiting MACEs to death and MI,133 or death, MI and unstable angina. 131 Most cases of MACEs were revascularisation rather than death or MI.

Table 27 shows the quality assessment of the CTCA and CAC scoring prognostic studies. All the studies described patient characteristics in terms of age and sex, but the description of times to presentation was inconsistent. All but one study134 defined MACEs in their methods section. In all studies the identification and definition of MACEs was independent of the index test and, in accordance with the inclusion criteria, MACEs were reported for at least 80% of the cohort. However, only one study136 used multivariate analysis to determine if CTCA provided additional prognostic value beyond routine assessment with ECG and biomarkers.

Table 28 summarises the results of the prognostic studies of CTCA. It was not always clear whether patients with positive CTCA had been followed up and whether there had been any events in these patients. MACE rates were generally very low in patients with a negative CTCA. The only adverse event in a patient with negative CTCA was a death in the long-term follow-up cohort of Hollander et al. 132 However, these low event rates may reflect selection of low-risk patients rather than accurate risk stratification by CTCA. Most of the events reported in patients with positive CTCA findings were process events [i.e. percutaneous coronary intervention (PCI) or coronary artery bypass graft (CABG)], which in an unblinded study may simply reflect physicians acting upon CTCA findings. No patient with positive or intermediate CTCA died on follow-up. There were 2 out of 43 and 12 out of 185 non-fatal MIs among those patients with positive or intermediate CTCA in the cohorts of Rubinshtein135 and Schlett. 136 The cohorts of Miller134 (n = 18), Goldstein131 (n = 32) and Hollander133 (n = 54, follow-up to 30 days) reported no cases of death or non-fatal MIs among those patients with positive CTCA. It could be argued that the process outcomes (PCI and CABG) prevented subsequent death or non-fatal MI in those with positive CTCA, but this is difficult to determine.

In the study of Schlett et al. 136 patients and carers were blind to CTCA findings, so any association between CTCA findings and process events (PCI and CABG) was not simply due to physicians acting on CTCA findings. Schlett et al. found that CTCA predicted MACEs, even after adjustment using a clinical risk score incorporating ECG and troponin measurement. Thus, this study provides the best evidence that CTCA provides independent prognostic value beyond routine clinical assessment.

The study of CT CAC scoring151 reported that 9 out of 91 patients with a CAC score of > 0 had MACEs (two MI and nine PCI), compared with 0 out of 82 with a CAC score = 0.

The results of meta-analysis of CTCA prognostic studies are shown in Figure 30 (positive and intermediate vs negative) and Figure 31 (positive vs intermediate and negative). Only studies that definitely reported data from patients with positive and negative CTCA are included in this analysis. Meta-analysis of the five studies with analysable data showed a RR for MACEs of 3.1 (95% CrI 0.3 to 18.7) for positive and intermediate scans compared with negative scan and 5.8 CrI (95% Crl 0.6 to 24.5) for positive scan compared with intermediate or negative scans. These estimates are subject to considerable uncertainty, with the CrI including one (i.e. no association) for both estimates. Taken alongside the limitations relating to patient selection and process outcomes suggests that there is currently only weak evidence that CTCA provides prognostically useful information in patients with suspected ACS.

FIGURE 30.

Meta-analysis of CTCA prognostic studies: positive and intermediate vs negative.

FIGURE 31.

Meta-analysis of CTCA prognostic studies: positive vs intermediate and negative.

Prognostic studies of exercise electrocardiography

Table 29 shows the characteristics of the prognostic studies of exercise ECG. Sample sizes ranged from 28 to 1000. The mean age (30–60 years) was relatively young, reflecting the selection of low-risk patients in many of the cohorts. Follow-up ranged from 30 days to > 12 months. There was no consistency in the definitions and reporting of MACEs, with some studies reporting composite outcomes only and others reporting outcomes separately with no indication of whether or not some patients had suffered multiple different adverse outcomes.

Table 30 shows the quality assessment of the exercise ECG studies. The population age and sex were always well described but most studies did not clearly define their inclusion criteria. MACEs were defined in the methods section in all but one study and was defined and identified independent to the index test in all studies. No study undertook multivariate analysis to determine the independent prognostic value of exercise ECG.