Systematic review, meta-analysis and economic modelling of ...

DOI: 10.3310/hta17010 Health Technology Assessment 2013 Vol. 17 No. 1Chapter 5 DiscussionStatement of principal findingsDiagnostic accuracy of presentation biomarkers for myocardial infarctionA large number of studies have estimated the accuracy of troponin at presentation for diagnosing MI,compared with a reference standard based on the universal definition using delayed troponin testing.Many of these are limited by inadequacies of the troponin assay used as the index test or referencestandard, whereas differences in the assays and threshold used limited our ability to compare andsynthesise data from different studies. We restricted meta-analysis to studies using similar or the sameassay at a comparable diagnostic threshold and using a reference standard based on a modern troponinassay with an acceptable diagnostic threshold. Even in these analyses there was substantial heterogeneitybetween results, which is reflected in the wide predictive intervals around each estimate.Our meta-analysis showed that the sensitivity and specificity of TnI at presentation were 77% (95%predictive interval 29% to 96%) and 93% (95% predictive interval 46% to 100%), respectively, when the99th percentile was used and 82% (95% predictive interval 40% to 97%) and 93% (95% predictive interval74% to 98%) when the 10% CV was used. The corresponding results for TnT were 80% (95% predictiveinterval 33% to 97%) and 91% (95% predictive interval 53% to 99%) when the 99th percentile was usedand 74% (95% predictive interval 34% to 94%) and 96% (95% predictive interval 76% to 99%) when the10% CV was used. When analysis was restricted to high-sensitivity assays we found that the Roche HsTnTassay had a sensitivity of 96% (95% predictive interval 27% to 100%) and a specificity 72% (95% predictiveinterval 3% to 96%), the ADVIA Centaur Ultra troponin I assay had a sensitivity of 86% (95% predictiveinterval 22% to 96%) and a specificity 89% (95% predictive interval 40% to 97%), and the AbbottArchitect troponin I assay had a sensitivity of 83% (95% predictive interval 58% to 95%) and a specificity95% (95% predictive interval 67% to 100%).The differences in estimates of sensitivity and specificity for different assays may reflect differences in studymethods and populations, but they suggest that using a lower threshold for positivity and high-sensitivityassay improves sensitivity at the expense of specificity. It is not entirely clear whether this loss of specificityrepresents the expected loss of specificity that is seen whenever the threshold for positivity is loweredfor an imperfect test, or whether the apparent FPs may actually be TPs misclassified by an inadequatereference standard. We identified one study 60 that seemed to suggest the former, but further data arerequired an address this issue. Such data would also determine whether the estimates of sensitivity fortroponin at presentation are lower when compared with a high-sensitivity reference standard.The findings suggest that high-sensitivity assays have sufficient sensitivity at presentation to identifymost cases of MI that would subsequently be identified by a standard 10-hour troponin test, but thereis substantial uncertainty around these estimates and a significant proportion with MI will be missedby presentation troponin testing. Whether or not this means that 10-hour troponin testing should beundertaken depends on the costs and benefits of detecting additional cases and is explored in detail in theeconomic analysis.We also sought studies of other biomarkers measured at presentation to determine their accuracy for MIeither alone or in combination with troponin. Only myoglobin and H-FABP had been evaluated againstan acceptable reference standard in a large number of studies. Our meta-analysis of H-FABP showed thatthe summary estimates of sensitivity and specificity were 81% (95% predictive interval 50 % to 95%)and 80% (95% predictive interval 26% to 98%), respectively, for the quantitative assays and 68% (95%predictive interval 11% to 97%) and 92% (95% predictive interval 20% to 100%), respectively, for thequalitative assays. Our meta-analysis of myoglobin showed that the summary estimates of sensitivity and© Queen’s Printer and Controller of HMSO 2013. This work was produced by Goodacre et al. under the terms of a commissioning contract issued by the Secretary of Statefor Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journalsprovided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should beaddressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton SciencePark, Southampton SO16 7NS, UK.107