Author + information
- Received October 5, 2016
- Revision received December 7, 2016
- Accepted December 15, 2016
- Published online April 3, 2017.
- Ernest Spitzer, MDa,b,
- Ton de Vries, MSca,
- Rafael Cavalcante, MD, PhDb,
- Marieke Tuinman, MSca,
- Tessa Rademaker-Havinga, MSca,
- Maaike Alkema, MSca,
- Marie-Angele Morel, MSca,
- Osama I. Soliman, MD, PhDa,b,
- Yoshinobu Onuma, MD, PhDa,b,
- Gerrit-Anne van Es, PhDc,
- Jan G.P. Tijssen, PhDc,
- Eugene McFadden, MDa,d and
- Patrick W. Serruys, MD, PhDe,∗ ()
- aCardialysis Core Laboratories and Clinical Trial Management, Rotterdam, the Netherlands
- bDepartment of Cardiology, Thoraxcenter, Erasmus University Medical Center, Rotterdam, the Netherlands
- cEuropean Cardiovascular Research Institute, Rotterdam, the Netherlands
- dDepartment of Cardiology, Cork University Hospital, Cork, Ireland
- eInternational Centre for Circulatory Health, National Heart and Lung Institute, Imperial College London, London, United Kingdom
- ↵∗Address for correspondence:
Dr. Patrick W. Serruys, International Centre for Circulatory Health, National Heart and Lung Institute, Imperial College London, 59 North Wharf Road, London W2 1LA, United Kingdom.
Objectives This study sought to investigate the differences in detecting (e.g., triggering) periprocedural myocardial infarction (PMI) among 3 current definitions.
Background PMI is a frequent component of primary endpoints in coronary device trials. Identification of all potential suspected events is critical for accurate event ascertainment. Automatic triggers based on study databases prevent underreporting of events.
Methods We generated automated algorithms to trigger PMI based on each definition and compared results using data from the RESOLUTE all comers trial.
Results The operationalization of current PMI definitions was achieved by defining programmable algorithms used to interrogate the study database. From a total of 636 PMI triggers, we identified 234 for the World Health Organization extended definition, 382 for the Third Universal definition, and 216 for the Society for Cardiovascular Angiography and Interventions definition. Differences among the biomarkers used, different cutoff values, and in the hierarchy among biomarkers within definitions, yielded a different number of triggers, and identified unique triggers for each definition. Only 38 triggers were consistently identified by all definitions. Availability of ECG data, eCRF data on clinical presentation, and the reporting of >2 post-procedural values of the same biomarker influenced considerably the number of PMI triggers identified.
Conclusions PMI definitions are not interchangeable. The number of triggers identified and consequently the potential number of events varies significantly, highlighting the importance of rigorous methodology when PMI is a component of a powered endpoint. Emphasis on collection of biomarkers, ECG data, and clinical status at baseline may improve the correct identification of PMI triggers.
Periprocedural myocardial infarction (PMI) is a component of the composite primary endpoint in coronary device trials and trials that compare interventional with surgical revascularization strategies. Multiple different definitions of PMI have been adopted, although 3 are currently preferred. Detection of all potential suspected events for adjudication by an independent clinical event committee (CEC) is critical to ensure complete ascertainment of endpoints in individual trials and to enable comparisons among trials. Where biomarker values are available, automatic triggering based on biomarker and other data in study databases can aid in preventing underreporting of events. Independent adjudication of clinical events is a critical component of the overall quality of randomized trials (1), allowing for a standard, nonbiased, adjudication of events (2). The identification of potential events (i.e., triggers) and the collection of relevant source documents is generally entrusted to clinical research organizations (Figure 1) (3).
Event triggers may have multiple sources including: 1) potential events reported by investigators, generally through electronic case report forms (eCRF); 2) potential (unreported by the investigator) events detected by monitoring of source documents; 3) eCRF-derived programmable event triggers; 4) core laboratory-derived event triggers, such as the detection of potential stent thromboses; and 5) event triggers identified by the CEC members during review of source documentation related to other events. eCRF-derived programmable event triggers are identified through algorithms with specific cutoff values for biomarkers, clinical parameters, and findings on diagnostic techniques. This process presupposes structured integration of the eCRF and the CEC database, and the core laboratory databases when available (e.g., electrocardiogram [ECG], angiography).
Three commonly used definitions of PMI are encountered in current percutaneous coronary intervention (PCI) trials. First, the World Health Organization (WHO) extended definition, which was developed in the context of the RAC (RESOLUTE All-Comers) trial. The RAC trial originally aimed to use the WHO historical definition (4), because of the preference of the industry at that time aiming to ensure comparability and poolability of data among trials (5,6). However, given the need for adjudication of event triggers in which biomarkers were not available, the WHO extended definition, which is the oldest considered in this analysis, introduced a hierarchy that would allow adjudication even in the absence of complete biomarker data. Second, the Third Universal definition (TUD) of the Joint European Society of Cardiology/American College of Cardiology Foundation/American Heart Association/World Heart Federation Task Force (7), which has troponin as the preferred biomarker always in conjunction with ancillary criteria, such as symptoms or ECG findings consistent with ischemia or angiographic complications. Finally and most recently, the Society for Cardiovascular Angiography and Interventions (SCAI) provided recommendations for the diagnosis of PMI based on creatine kinase-myocardial band (CK-MB) as the preferred biomarker and with a markedly higher threshold based on evidence suggesting that this threshold, irrespective of other criteria, is associated with clinically relevant PMI (8).
To accurately identify event triggers based on these definitions, we aimed to operationalize them into programmable algorithms. For validation purposes, we used the study database of the RAC trial, an investigation sponsored by Medtronic CardioVascular, which compared drug-eluting stents releasing either zotarolimus or everolimus (9).
PMI definitions were derived from the original sources (5,7,8); algorithms drafted; and discussed in several meetings of cardiologists, statisticians, and scientists with relevant expertise in the topic to reach a consensus that would permit reliable identification of PMI triggers while minimizing false positives. Operationalization of definitions was primarily based on biomarker values. However, when ECG findings or clinical details recorded in the eCRF could help to better characterize the triggers, they were also integrated.
Six specific issues were identified and discussed up-front because these would affect the structure of the algorithms. First, the interchangeability of units reported as multiples by the investigational sites because in most trials these are not uniformly reported. Thus, upper reference limit was interchangeably used together with upper limit of normal. Second, all 3 definitions classify the status of the biomarkers as stable, rising, or falling, based theoretically on pre-procedure biomarkers. However, when only 1 pre-procedure value of a specific biomarker was available, we allowed comparisons between pre- and post-PCI values to achieve this classification. Third, the presence of MI at baseline (i.e., before the procedure) was defined as the presence of ST-segment elevation myocardial infarction/non-ST-segment elevation myocardial infarction (STEMI/NSTEMI) at baseline according to the investigator, CK-MB, or cardiac troponin >1 upper limit of normal/upper reference limit (for all 3 definitions), and also if CK >1 upper limit of normal/upper reference limit for WHO extended, ignoring the hierarchies. When both baseline biomarkers and clinical diagnosis were missing the most sensitive path (i.e., the one with the lowest threshold for generating a trigger) was chosen. Fourth, ECG data were included in the algorithms for WHO extended and SCAI, because this would improve specificity without affecting sensitivity (i.e., biomarker cutoffs change according to ECG findings). The presence of new Q waves (or left bundle branch block) after the procedure in patients in stable condition at baseline, or the presence of significant changes after the procedure as proxy, would determine the path in the algorithm. When ECG data were missing, ECG changes were presumed present. Fifth, the hierarchical use of biomarkers was implemented for all definitions. WHO extended uses first CK, then CK-MB, and then troponin, with 1 exception (5). TUD uses first troponin, then CK-MB. SCAI uses first CK-MB, and then troponin. CK is not used in TUD or SCAI. This hierarchy is used regardless of whether pre-procedure information is available for the selected biomarker. If the pre-procedure value is not available, a worst case approach is used (value 0 is assumed). Sixth, when no post-procedure biomarkers were available, patients were further classified according to the presence of MI at baseline, and if ECG findings were reported.
Once algorithms for all definitions were completed, they were programmed and executed in the study dataset of the RAC trial (9).
The algorithms used for the operationalization of each definition are shown in Figures 2, 3, 4, and 5. The availability in the RAC trial of parameters used in the algorithms is presented in Online Table 1.
WHO extended definition
The WHO extended definition results are found in Figure 2 and Online Figure 1. Out of 2,509 procedures included in this analysis (2,292 index and 217 staged), 234 yielded a PMI trigger; 161 were related to the index procedure, and 73 to staged procedures; 114 procedures followed a different algorithm (Figure 5, Online Figure 2) because there were no post-procedure biomarkers available. Overall, 916 presented with a STEMI/NSTEMI at baseline, whereas 1,479 were procedures for patients in stable condition. From the former, 845 had >1 value of the same biomarker post-PCI, whereas 71 only had 1 value. When >1 value was available, only 8.0% were classified as triggers, whereas when only 1 value was available, 22.5% were so classified. Patients in stable condition were further categorized according to the presence of clinically relevant changes in the ECG at discharge relative to baseline. A total of 98 of 278 (35.3%) with ECG changes (8) or missing ECG data (270) were classified as triggers, whereas 52 of 1,201 (4.3%) without ECG changes were so classified.
Third universal definition
The TUD results are found in Figure 3 and Online Figure 3. Overall, 382 procedures were associated with a PMI trigger; 318 were related to the index procedure, and 64 to staged procedures; 150 procedures followed a different algorithm (Figure 5, Online Figure 2) because there were no post-procedure biomarkers available; 833 presented with a STEMI/NSTEMI at baseline, whereas 1,526 were procedures for patients in stable condition. From the former, 756 had >1 value of the same biomarker post-PCI, whereas 77 only had 1 value. When >1 value was available, 11.8% were classified as triggers, whereas when only 1 value was available, 48.1% were so classified. No subcategorization according to ECG findings was performed because for the TUD, the ECG findings do not mandate different cutoff values for the biomarkers.
The SCAI definition results are found in Figure 4 and Online Figure 4. Overall, 216 procedures were associated with a PMI trigger; 204 were related to the index procedure and 12 to staged procedures; 150 procedures followed a different algorithm (Figure 5, Online Figure 2) because there were no post-procedure biomarkers available; 833 presented with a STEMI/NSTEMI at baseline, whereas 1,526 were procedures for patients in stable condition. From the former, 756 had >1 value of the same biomarker post-PCI, whereas 77 only had 1 value. When >1 value was available, 185 (24.5%) were classified as triggers, whereas when only 1 value was available, 9 (11.7%) were so classified. Patients in stable condition were further categorized according to the presence of clinically relevant changes in the ECG at discharge relative to baseline; 14 of 276 (5.1%) with ECG changes (10) or missing ECG data (266), were classified as triggers; 8 of 1,250 (0.6%) without ECG changes were so classified.
Comparison among definitions
When all definitions were combined, 636 PMI triggers were identified corresponding to 26.6% of all procedures. However, only 38 triggers (6.0%) were identified based on all definitions. A Venn diagram showing the distribution of triggers is presented in Figure 6A. A total of 91 (14.3%), 240 (37.7%), and 147 (23.1%) triggers were uniquely identified using the WHO extended, TUD, and SCAI definitions, respectively. Concordance among 2 definitions was observed in 89 (WHO extended and TUD, 14.0%), 15 (TUD and SCAI, 2.4%), and 16 (WHO extended and SCAI, 2.5%) event triggers.
Calculations by modifying the algorithm
To further assess the relevance of different parameters for triggering, we repeated the analysis excluding clinical presentation (Figure 6B), ECG data (Figure 6C), or both (Figure 6D) from the algorithms. We observed a consistent reduction in the number of triggers obtained with all 3 definitions when clinical presentation data were available, and when ECG data were available. When both were excluded from the algorithm, we observed an almost 3-fold increase in the number of triggers according to the WHO extended definition (from 234 to 626 triggers), and an overall 40% increase for all combined definitions. When clinical presentation was excluded, 15% of triggers were detected consistently by all definitions; 7.6% when ECG data were excluded, and, 11.7% when both were excluded (Figures 6B to 6D).
Calculations using the peak value only
When using only the peak post-procedural value of a biomarker, instead of all post-procedural values available, the number of triggers increased from 636 to 867. With the SCAI definition using only 1 biomarker (214 with peak value vs. 216 with all values available) there was no material effect on the number of triggers. There were few cases only identified when using peak values (5) or all values (3). However, based on the WHO extended and TUD, we observed an increase in the number of triggers from 234 to 470, and from 382 to 782, respectively.
The main findings of the present investigation regarding triggering PMI based on 3 current definitions are as follows:
1. Operationalization of PMI definitions (WHO extended, TUD, and SCAI) for event triggering was accomplished by the development of 3 definition-specific algorithms;
2. Different PMI definitions (WHO extended, TUD, and SCAI) yielded markedly different numbers of PMI triggers and, not uncommonly, were mutually exclusive;
3. Differences among the number of triggers related to the different definitions are explained by the biomarkers used, the hierarchies in biomarker selection, and the cutoff values;
4. The careful collection of pre-procedural biomarkers (ideally 2 for the TUD), at least 2 sets of post-procedural biomarkers, the documentation of new ischemic ECG changes at discharge relative to baseline (or relative to post-index for staged procedures), and recording of clinical presentation at baseline (i.e., STEMI/NSTEMI vs. others) improves identification of PMI triggers, potentially reducing adjudication costs.
Randomized trials use event rates to compare the safety and efficacy of 2 or more treatment strategies, devices, or pharmacological treatments. Consequently, the meticulous detection of events is crucial to guarantee a fair comparison (10). In large multicenter trials this endeavor is undertaken by a joint collaboration among the investigational sites through accurate and comprehensive reporting of events, clinical research associates acting as site monitors with the responsibility of reviewing charts of patients enrolled and prevent underreporting of events, core laboratory-derived event triggers (i.e., angiographic or ECG findings), and a programmed review of the database searching for abnormal values (1,11). Multiple event triggers that may result in the adjudication of an endpoint need to be identified and the relevant source documents collected before presentation to the CEC (Figure 1).
Exhaustive detection of triggers involves a step-wise process that can be customized for a trial. The main source of triggers is those reported by the investigator. Although these triggers are clinically accepted and represent the expert opinion of the treating physician or local investigator, noise may be introduced because of variable compliance with protocol-mandated definitions. A local or remote monitor who reviews source documents to verify completeness of reporting should perform an initial validation. Relevant considerations for the monitoring activities include the independence of the monitors and the percentage of source documents that are monitored. Additional elements of validation include the eCRF (highly dependent on the appropriateness of the questions and the completeness of data), the core laboratories (for biomarker assessment, ECG interpretation, or angiographic analysis), and the in-depth review of source documents by the adjudicators, who may identify additional, unreported events.
eCRF-derived event triggers are defined by rules, such as cutoff values of troponin for PMI, or cutoff values of hemoglobin for anemia (3). These rules need to be as inclusive as possible so as not to miss potential events (false negatives) while minimizing the inclusion of nonevents (false positives). The development of trigger rules involves the strict implementation of the definitions delineated in the investigational protocols and should properly balance a trade-off between ensuring quality while minimizing cost. Once an event trigger is generated, it adds cost to the trial (source document collection, digitalization, dossier preparation, and adjudication, among others); thus, event triggers are of significant relevance for investigators and sponsors while designing a trial.
PMI definitions have changed over time. The continuous progress in cardiovascular medicine has brought a shift from nonspecific enzymes, such as CK and lactate dehydrogenase, toward more specific biomarkers, such as cardiac CK-MB, and ultimately cardiac troponins. These changes have had a significant impact in definitions. For instance, CK is no longer used in the SCAI or TUD. Moreover, whereas troponin is the preferred biomarker for TUD, SCAI strongly recommends the use of CK-MB as the preferred biomarker based on robust historical data.
Our results demonstrate that the TUD provides the highest number of event triggers given the preferential use of troponin, and the lower cutoffs as compared with SCAI. Nonetheless, the lack of systematic collection in the eCRF of ancillary criteria is problematic for consistent event adjudication. The WHO extended definition, although scientifically interesting for the present analysis, may progressively be abandoned and is mostly used to allow data comparability. The SCAI definition results in a lower number of triggers. The rationale proposed for this definition is that it identifies patients with clinically relevant PMI and is therefore more appropriate as a component of a composite primary endpoint (12). It is noteworthy that TUD may be preferred for trials in which a higher rate of events is needed because of statistical power and when mechanistic comparisons are aimed. These include stent versus stent comparisons (e.g., different strut thickness or stent design) and comparisons among drugs administered before PCI. The SCAI definition may be preferred for large randomized controlled trials because of its simplicity and less costs involved, and in strategy trials comparing PCI with coronary artery bypass graft. However, the relative merits of the Universal MI definition and the SCAI definition of PMI are not addressed in the current investigation.
The value of a database that collects specific items captured in the definitions of PMI has been highlighted by our results. When developing the eCRF investigators should include the clinical presentation; presence of prolonged chest pain (>20 min) during the procedure; information regarding the baseline and pre-discharge ECGs, and more specifically the presence of ischemic ST-segment changes, new pathological Q waves, or new left bundle branch block; and the presence of angiographic periprocedural complications defined as slow-flow or no-flow at the end of the procedure (Thrombolysis In Myocardial Infarction flow grade <3), occlusion of a major coronary artery or a side branch, or distal embolization (13). Imaging demonstration of new loss of viable myocardium or new regional wall motion abnormality in patients in stable condition undergoing PCI should also be included. These elements would be investigator-reported with evident limitations but would provide, with appropriate verification in place, a more rational basis for comparison among trials. Ideally, electrocardiographic and angiographic assessment should be performed by a central core laboratory and the results provided to the CEC (6,14).
Detecting PMI outside the setting of a randomized controlled trial creates additional challenges. Clinical centers may have preferences among definitions, which could even be the case for individual physicians within 1 single institution. Consequently, large prospective registries need standardization of definitions for voluntary reporting. This has proven useful in the National Cardiovascular Database Registry CathPCI Registry where the impact of PMI definitions on observed rate of events has been recently explored (15).
Although we were able to show important differences among PMI triggers according to the 3 different definitions, the sensitivity and specificity of the triggers can only be tested if triggering and adjudication are performed following all 3 definitions. It remains important to elucidate the impact of using different definitions in a single trial, because not only does the number of triggers change, but also the individual procedures potentially associated with a PMI.
PMI definitions are not interchangeable. The identified number of triggers and consequently the potential number of events vary significantly, prompting careful consideration when PMI is a component of a powered endpoint. Rigorous collection of biomarkers, ECG data, and clinical status at baseline may improve the correct identification of PMI triggers.
WHAT IS KNOWN? PMI is a frequent component of primary endpoints in coronary device trials.
WHAT IS NEW? Accurate detection of PMI requires careful interpretation of the definitions, because those may largely affect event rates. Contemporaneous definitions of PMI may be mutually exclusive for detecting potential events.
WHAT IS NEXT? Emphasis on collection of biomarkers, ECG data, and clinical status at baseline may improve the correct identification of PMI triggers.
For supplemental tables and figures, please see the online version of this article.
The RESOLUTE all comers trial was funded by Medtronic CardioVascular. Dr. Onuma is a member of the Advisory Board of Abbott Vascular. Dr. McFadden has received honoraria for clinical event committee work for Abbott Vascular; and a travel grant from Menarini Ireland. Dr. Serruys is a consultant for Abbott Laboratories, AstraZeneca Pharmaceuticals, Biotronik, Medtronic, Volcano Europe BVBA, St. Jude Medical, Stentys France, Svelte Medical Systems, Inc., and Sino Medical Sciences Technology, Inc. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- clinical event committee
- creatine kinase-myocardial band
- electronic case report form
- non-ST-elevation myocardial infarction
- percutaneous coronary intervention
- periprocedural myocardial infarction
- RESOLUTE All-Comers trial
- Society for Cardiovascular Angiography and Interventions
- ST-segment elevation myocardial infarction
- Third Universal definition
- World Health Organization
- Received October 5, 2016.
- Revision received December 7, 2016.
- Accepted December 15, 2016.
- 2017 American College of Cardiology Foundation
- Mahaffey K.W.,
- Harrington R.A.,
- Akkerhuis M.,
- et al.
- ↵(1979) Nomenclature and criteria for diagnosis of ischemic heart disease. Report of the Joint International Society and Federation of Cardiology/World Health Organization task force on standardization of clinical nomenclature. Circulation 59:607–609.
- Cavalcante R.,
- Serruys P.W.
- Thygesen K.,
- Alpert J.S.,
- Jaffe A.S.,
- et al.
- Moussa I.D.,
- Klein L.W.,
- Shah B.,
- et al.
- Mahaffey K.W.,
- Roe M.T.,
- Dyke C.K.,
- et al.
- Park D.W.,
- Kim Y.H.,
- Yun S.C.,
- et al.
- Ishibashi Y.,
- Muramatsu T.,
- Nakatani S.,
- et al.
- Cavalcante R.,
- Sotomi Y.,
- Onuma Y.
- Baker N.C.,
- Lipinski M.J.,
- Escarcega R.O.,
- et al.