Background Heart-type fatty acid binding protein (H-FABP) has been proposed as an early biomarker of myocardial infarction (MI). The authors aimed to undertake a systematic review and meta-analysis to estimate the early sensitivity and specificity of quantitative and qualitative H-FABP assays.
Methods The authors undertook a systematic search using electronic databases, citation lists and expert contacts to identify all diagnostic cohort studies of patients presenting with suspected acute coronary syndrome that compared H-FABP at presentation to a reference standard based on the Universal definition of MI. Study quality was assessed using the Quality Assessment of Diagnostic Accuracy Studies tool. Meta-analysis was conducted using Bayesian Markov chain Monte Carlo simulation.
Results The authors included eight studies of quantitative H-FABP and nine studies of qualitative H-FABP. The summary estimates of sensitivity and specificity were 81% (95% prediction interval 50% to 95%) and 80% (26% to 98%) respectively for the quantitative assays and 68% (11% to 97%) and 92% (20% to 100%) respectively for the qualitative assays. Four studies reported the sensitivity of troponin and H-FABP at presentation in which the combination was considered positive if either test was positive. The addition of H-FABP to troponin increased sensitivity from 42–75% to 76–97% but decreased specificity from 94–100% to 65–93%.
Conclusion H-FABP has modest sensitivity and specificity for MI at presentation but estimates are subject to substantial uncertainty and primary data are subject to substantial heterogeneity. H-FABP may have a role alongside troponin in improving early sensitivity but comparison with high sensitivity troponin assays is required.
- Myocardial infarction
- cardiac markers
- thromboembolic disease
- cost effectiveness
- acute coronary syndrome
- heart failure
Statistics from Altmetric.com
- Myocardial infarction
- cardiac markers
- thromboembolic disease
- cost effectiveness
- acute coronary syndrome
- heart failure
Chest pain is responsible for around 6% of adult emergency department attendances in the UK.1 In these patients, accurate early diagnosis of acute myocardial infarction (MI) is essential to avoid inadvertent discharge home with MI or admission with benign chest pain. The diagnosis of MI is based upon troponin measurement2 and current guidelines recommend measurement 10–12 h after symptom onset to allow for suboptimal early sensitivity of troponin.3 Since patients typically present around two hours after symptom onset4 this delay often necessitates hospital admission. As a result, there has been substantial interest in developing alternative or additional early markers of myocardial necrosis.
Heart-type fatty acid binding protein (H-FABP) is a relatively small (15 kDa) protein found in the cytoplasm, which because of its small molecular size would be released very early in myocardial ischaemia or infarction. The pharmacokinetic properties of H-FABP that involve early release and detectability in the peripheral circulation have led to interest in its use as an early diagnostic tool for MI. It has been developed as a quantitative and qualitative assay and evaluated in a substantial number of diagnostic cohort studies. Our objective in this meta-analysis was to estimate the diagnostic accuracy of H-FABP for MI when used at patient presentation to hospital.
The systematic review and meta-analysis was undertaken in accordance with the guidelines published by the Centre for Reviews and Dissemination for undertaking systematic reviews5 and the Cochrane Diagnostic Test Accuracy Working Group on the meta-analysis of diagnostic tests.6 This study was undertaken as part of a wider review of diagnostic tests for suspected acute coronary syndrome (ACS)7 and used a combined search strategy. A search of the following bibliographic databases was undertaken in November 2010 (inception dates for each database given in brackets): MEDLINE (1950), MEDLINE In-Process & Other Non-Indexed Citations (1950), Cumulative Index of Nursing and Allied Health Literature (CINAHL) (1981), EMBASE (1980), Web of Science (WoS) (includes Science Citation Index and Conference Proceedings Citation Index) (1899), Cochrane Central Database of Controlled Trials (CENTRAL), Cochrane Database of Systematic Reviews (CDSR), NHS Database of Abstracts of Reviews of Effectiveness (DARE) and the Health Technology Assessment (HTA) database.
The search combined terms relating to the population (eg, chest pain, ACS or MI) with terms for H-FABP and the reference standard (troponin), and a search filter aimed at restricting results to diagnostic accuracy studies (used in the searches of MEDLINE, CINAHL and EMBASE). No date or language restrictions were applied.
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 search were imported into and managed using the Reference Manager bibliographic software V.12.0 (Thomson Reuters, Philadelphia, PA).
The citations retrieved were divided between two reviewers (CC, SG) and screened for relevance to the review. This study selection process was only completed after the two reviewers had first performed an inter-rater reliability test on a sample of 700 titles and abstracts (k=0.71). Studies were selected for review if they met the following criteria:
Design: diagnostic cohort study
Population: 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 (eg, studies that excluded patients with ST-elevation MI).
Index test: H-FABP
Reference standard test: acute MI defined according to the universal definition2 using troponin I or T measurement for a minimum of 80% of the population at least 6 h after symptom onset.
Data were extracted from each study by one reviewer (CC, MK or JL) into a standardised data extraction form and independently checked for accuracy by a second reviewer (CC, MK or JL). Discrepancies were resolved by discussion between the two reviewers and if agreement could not be reached, the principal investigator was consulted (SG). Where multiple publications of the same study were identified, data were extracted and reported as a single study. Where there was possible overlap between cohorts reported from the same author group or study centre, data were excluded from one of the cohorts to avoid duplication. The following data were extracted from all studies, where reported: study characteristics (author, year of publication, journal, country, study design, setting); participant details (age, sex, presenting condition, inclusion and exclusion criteria); index test details (including (median) time from pain onset to presentation or blood test, diagnostic threshold and assay); reference standard details (including diagnostic threshold and assay); prevalence of MI and outcome data (true positives, false negatives, false positives and true negatives); sensitivity; specificity; and any additional potential relevant citations from the reference list. If raw numbers for outcome data were not reported, where possible, these data were calculated from the reported 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.
The methodological quality of each included study was assessed by one reviewer (CC, MK or JL) but checked by a second (CC, MK or JL) using a modified version of the Quality Assessment of Diagnostic Accuracy Studies tool8 (a generic, validated, quality assessment instrument for diagnostic accuracy studies). In case of doubt, the principal investigator (SG) was consulted.
The analysis was conducted using Markov chain Monte Carlo simulation and was implemented using the freely available software WinBUGS.9 In general, there are advantages of the Bayesian approach over a classical approach, including: (1) the ability to analyse complex models exactly, (2) the ability to incorporate external evidence in addition to sample data, (3) the ability to make probabilistic statements about parameters. The diagnostic test data was analysed using a bivariate normal model for the logits (ie, log odds) of the sensitivities and specificities in each study to allow for correlation between outcomes within studies.10
We let‘∼’ should be read ‘is distributed as’ and ‘N’ is the bivariate normal distribution with specified mean and variance-covariance matrix.
We completed the model by giving the uncertain parameters the following prior distributions:‘IW’ is the inverse-Wishart distribution.
Convergence of the model was assessed using the Gelman-Rubin convergence statistic.11 Convergence occurred after 15 000 iterations. We used a burn-in of 15 000 and generated a further 20 000 iterations to estimate the parameters in the model.
Results are presented as a 95% credible interval for the population sensitivities and specificities, and a 95% prediction interval for a new study.
We identified 17 diagnostic studies of H-FABP for inclusion in the review.12–28 Table 1 shows the population characteristics of these studies. Reporting of exclusion criteria was variable and some studies excluded 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 (mean) to 5.9 (median) hours.
Table 2 shows the index and reference standard test characteristics. Around half 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 the H-FABP quantitative assay was variable. The reference test used was cardiac troponin, although the methods used were contemporary29 rather than high sensitivity assays and not all gave details of the assay and threshold. There was some heterogeneity in the troponin I methods used although most used a 10% coefficient of variation (CV) as the diagnostic discriminant. The position for cardiac troponin T was slightly more complicated. Here the diagnostic discriminant included a value equivalent to the original WHO criteria (0.1 μg/l), the 10% CV (0.03 μg/l) or the detection limit of the assay (0.01 μg/l) which approximates to the 99th percentile value.
Figure 1 shows the quality assessments for the included studies. Quality was generally high, perhaps reflecting exclusion of lower quality studies by our selection criteria. There was some uncertainty about whether index and reference standard tests were assessed blind, so bias could have resulted if reference standard adjudicators were aware of the presentation H-FABP result (detection bias).
Figure 2 shows the meta-analysis of the studies of quantitative H-FABP and figure 3 shows the meta-analysis of qualitative assays. The summary estimates of sensitivity and specificity were 81% (95% prediction interval 50% to 95%) and 80% (26% to 98%) respectively for the quantitative assays and 68% (11% to 97%) and 92% (20% to 100%) respectively for the qualitative assays.
Four studies reported the sensitivity of a combination of troponin and H-FABP at presentation in which the combination was considered positive if either test was positive. The results of these studies are shown in table 3 along with the sensitivity and specificity of the troponin assay alone. The addition of H-FABP to troponin increased sensitivity from 42–75% to 76–97% but decreased specificity from 94–100% to 65–93%.
H-FABP has modest sensitivity and specificity for MI at presentation to hospital. Quantitative assays have sensitivity of 81% (95% prediction interval 50% to 95%) and specificity of 80% (26% to 98%), while qualitative assays have sensitivity of 68% (11% to 97%) and specificity of 92% (20% to 100%). The wide predictive intervals for the estimates reflect the heterogeneity in the primary studies. The summary estimates and surrounding uncertainty suggest that H-FABP cannot be used as a lone marker for MI at presentation. However, this is not how we would envisage using H-FABP in practice. Troponin is an established marker for MI and any alternative biomarker would be used alongside troponin. Four of the studies12 18 25 26 reported this combination and results showed that adding H-FABP improved sensitivity compared with troponin alone but reduced specificity. The loss of specificity may be acceptable if the only consequence was the requirement to measure a 10 h troponin, which then acted as the definitive test, but it is debatable whether the sensitivities reported are high enough to allow early discharge. It has been proposed that the combination of a negative troponin value on admission plus a negative heart fatty acid binding protein, when combined with the ECG would allow early discharge. Further studies are required to assess whether this is truly the case.
The measurement of cardiac troponin alone on admission when a sensitive assay is used which is able to truly define the 99th percentile has shown a similar improvement in sensitivity at the expense of specificity.30 31 The additional value of H-FABP is therefore uncertain when compared with a sensitive troponin assay.32 33 Studies are required which use sequential sampling and measurement of troponin using the new highly sensitive assays in combination with heart fatty acid binding protein to see whether troponin alone or in combination is the optimal diagnostic strategy for very early rule out. In addition, the most appropriate diagnostic threshold for heart fatty acid binding protein remains to be defined. It has been suggested that a much lower threshold (2.5–3 μg/l) is appropriate. Hence the cardioDetect system may be inadequately sensitive.
Our systematic review and meta-analysis has a number of differences from a recent meta-analysis of H-FABP.34 We included eight recent studies that have been published since the other meta-analysis was undertaken12–14 17–19 22 23 and excluded six older studies that used the old WHO definition of MI35–40 and one study where we were unable to extract data.41 Thus, only nine studies15 16 20 21 24–28 were included in both reviews. The summary estimates from the previous review were 84% (95% CI 76% to 90%) for sensitivity and 84% (95% CI 76% to 89%) for specificity. Subgroup analysis showed that the sensitivity and specificity were 86% (75% to 93%) and 84% (73% to 91%) respectively for the quantitative assay and 81% (69% to 90%) and 85% (74% to 92%) respectively for the qualitative assay. Our updated analysis based on studies using the current definition of MI suggests that all the parameters except specificity for the qualitative assay are worse than suggested in the previous study. Furthermore, our methods of analysis ensure that the predictive intervals reported for estimates more appropriately reflect the uncertainty in estimates due to heterogeneity between studies. We also report the results of analyses comparing H-FABP with troponin to troponin alone to show the effect of adding H-FABP to routine assessment with troponin.
We limited our review to studies evaluating H-FABP in comparison with a reference standard of MI based on the universal definition.2 We therefore excluded studies that used a wider definition of ACS as a reference standard. This was because widening the definition of ACS typically involves including clinical judgements that may have little empirical basis, may vary between studies and may not be independent of the test under investigation. The potential problem of limiting analysis to studies with MI as the reference standard is that there is some evidence that H-FABP predicts adverse outcome in troponin negative patients with suspected ACS.20 42 Thus false positive cases in our analysis may actually have prognostically significant ACS.
Interpretation of our findings is inevitably limited by uncertainty in the estimates and the substantial heterogeneity between the studies, both clinically in terms of the application of inclusion and exclusion criteria, timing of sampling, and troponin assay and threshold used for the reference standard, and statistically in terms of the variation in sensitivity and specificity. The impact of the heterogeneity is reflected in the wide predictive intervals and demonstrates that the range of sensitivities and specificities for a new study are extremely variable. Some of the heterogeneity may be explained by ‘outlier’ studies with low sensitivity (eg, Charpentier et al14) or specificity (eg, Liao et al).23 It is not clear from examining the study characteristics why these two studies should report markedly different results to the others but it is noteworthy that they respectively reported the second lowest (15%) and highest (73%) prevalence of MI.
This meta-analysis suggests that there is still considerable uncertainty regarding the role of H-FABP in early diagnosis of MI. There is clearly insufficient evidence or practical rationale to support its use as a lone biomarker. It may have a role alongside troponin but further research is required to determine the impact on early sensitivity and specificity of adding H-FABP to high sensitivity troponin. Currently available data do not support the routine use of H-FABP as an early cardiac marker for MI.
The authors thank Pippa Evans for help with the literature searches and Susan Proctor and Kathryn Paulucy for clerical assistance.
Funding This project trial was funded by the National Institute for Health Research Health Technology Assessment Programme (number 09/22/21) and sponsored by the University of Sheffield. The study funders had no role in study design; in the collection, analysis and interpretation of data; in the writing of the report; and in the decision to submit the paper for publication. The researchers were independent of the study funders.
Disclaimer The views and opinions expressed therein are those of the authors and do not necessarily reflect those of the NIHR HTA. All authors, external and internal, had full access to all of the data (including statistical reports and tables) in the study and can take responsibility for the integrity of the data and the accuracy of the data analysis.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.