This section describes the methods for the systematic review of the evidence relating platelet function testing to the risk of adverse clinical outcome(s) in patients on aspirin therapy with established cardiovascular disease or CVD, or diabetes.
The review will specifically target studies which relate platelet function testing to clinical outcome in patients with established cardiovascular disease or CVD or diabetes who are being treated with aspirin. Analysis will consider whether or not PFTs have prognostic ability in that they are able to distinguish between groups of patients with different average outcome risks. If demonstrable, analysis will subsequently consider diagnostic/predictive ability, i.e. whether or not given tests have sufficiently high diagnostic/predictive utility to accurately distinguish those individual patients who will have an adverse outcome from those who will not.
A standard systematic review approach was used and is described below.
Selection criteria
Two broad types of study were considered relevant for this review: those studies that provide information on the prognostic or diagnostic/predictive utility of PFTs and those that report prognostic models, in which a PFT is one of multiple prognostic factors predicting clinical outcomes in a population of interest. The selection criteria for each are outlined below.
Prognostic utility and diagnostic utility studies
Types of study
Any prospective primary studies, or systematic reviews of such studies, assessing PFT(s) in relation to clinical outcomes.
Types of participants
Patients aged ≥ 18 years on aspirin (as monotherapy or in combination with other antiplatelet agents), with established cardiovascular disease or CVD, or diabetes. Studies with mixed populations were included as long as data for relevant patients were extractable. Studies with patients on aspirin for peripheral vascular disease were noted.
Setting
Studies in any setting were included.
Technology
Either a COX-1-specific PFT (which measures aspirin response specifically) or a global PFT in patients receiving aspirin as the only antiplatelet therapy. The selection process was guided by the information in Table 2.
Outcomes
Clinical outcomes, such as vascular events [non-fatal and fatal ischaemic stroke, TIA, systemic embolism (pulmonary embolism, peripheral arterial embolism), MI, revascularisation procedures]; haemorrhagic events; all-cause mortality; mortality due to vascular events; composite outcomes containing the above [e.g. major adverse cardiac events (MACEs)].
Timing
Reported outcomes had to occur after the undertaking of a PFT and the post-test follow-up period had to be 7 days or longer. Thus, studies performing platelet function testing after clinical events, with no further follow-up after the testing, were excluded (unless the testing was undertaken on stored samples retrieved prior to the clinical event, as this retains the temporal relationship between testing and subsequent outcome occurrence).
Prognostic model studies
Studies reporting prognostic models, in which a PFT was one of multiple prognostic factors predicting clinical outcomes in a population of interest, were eligible for review, in order to examine the contribution of the PFT to the overall performance of the prognostic model, and to establish whether or not predictive accuracy of clinical outcomes was improved by combining test results with other prognostic factors. The following criteria were used to select such studies:
Was a statistical model outlined to predict a relevant clinical outcome outlined above?
Did the model include a factor for PFT result or aspirin resistance?
Was the model developed for use in patients aged ≥ 18 years and on aspirin (alone or in combination with another therapy) for established cardiovascular disease or CVD or diabetes?
Searches
The following bibliographic databases were searched:
The Cochrane Library (Wiley) (issue 4 of 12) [including the Cochrane Database of Systematic Reviews, Database of Abstracts of Reviews of Effects, HTA Database, NHS Economic Evaluation Database (NHS EED) and Cochrane Central Register of Controlled Trials] to April 2012, MEDLINE (Ovid) from 1950 to 2012, MEDLINE In-Process & Other Non-Indexed Citations (Ovid) to 25 April 2012 and EMBASE (Ovid) from 1980 to 2012.
Search strategies combined index and text words encompassing the technologies (platelet function testing) and the patient group (cardiovascular disease, CVD and diabetes), as well as focusing on aspirin resistance. The Zetoc database (The British Library), Conference Proceedings Citation Index and Science Citation Index (Web of Science) were searched for conference proceedings. ClinicalTrials.gov, the UK Clinical Research Network Study Portfolio Database, the World Health Organization International Clinical Trials Registry Platform and the metaRegister of Controlled Trials were also searched for ongoing studies.
Reference lists of relevant articles, particularly systematic reviews, were hand-searched to identify other potentially relevant articles. Furthermore, a subject expert was used to identify any studies which may not be identified using standard methods.
Restrictions on publication language and date were not applied to the searches.
Copies of the search strategies used in electronic databases can be found in Appendix 1.
In addition, abstracts from the following national and international proceedings were hand-searched from 2009 onwards:
platelet conferences (Platelets International Symposium)
cardiology conferences (British Cardiovascular Society, American College of Cardiology, European Society of Cardiology, American Heart Association, American College of Chest Physicians)
stroke conferences (International Stroke Conference, American Stroke Association)
haematology conferences (British Society for Haematology, International Society on Thrombosis and Haemostasis, International Society for Laboratory Haematology).
Abstracts that were identified were considered for relevance in a similar way to fully published studies/articles.
Search results were entered into reference management software [Reference Manager version 11 (Thomson ResearchSoft, San Francisco, CA, USA)]. Duplicate records were removed by built-in algorithms and subsequent manual checking.
The searches of electronic databases were undertaken in April 2012 and were not updated after this time. A note was made of any additional relevant studies published subsequently that came to the attention of the authors of this report. These studies were not reviewed to avoid bias. A brief comment is made about these studies in Chapter 5, Relevant studies identified after the search cut-off dates.
Study selection
Study selection was undertaken as a two-step process. Titles (and abstracts where available) in records were initially screened by two reviewers, using prespecified screening criteria. These criteria were kept necessarily broad as it was anticipated that not all relevant information would necessarily be presented in an abstract, and thus the use of stricter criteria was likely to lead to the exclusion of relevant articles at this screen stage. These criteria were based on whether or not the records indicated that articles were about, or likely to be about, platelet function testing; reported, or were likely to report, clinical outcomes measured after a PFT; and were about patients who had or were likely to have cardiovascular/cerebrovascular or diabetic disease and were receiving aspirin therapy.
An additional criterion for conference abstracts was that these needed to be published from 2009 onwards to be retained. Letters to journals were not automatically classed as irrelevant, because often new results relevant to this field are made available through this medium.
Full texts of any potentially relevant articles or those where a decision could not be made were sought. In the second part of the two-step selection process, full-text articles were assessed against the full inclusion criteria by two reviewers independently. Any discrepancies between reviewers were resolved by discussion or by referral to a third reviewer. A copy of the selection form used for this process is available on request.
Both stages of the selection process were piloted prior to full implementation.
At title and abstract screening and for full-text screening, appropriate portions of non-English-language articles were translated where necessary to aid the selection process.
A record was kept of all decisions made, the reason for exclusion from the review at the full-text screening stage, articles that were not obtainable even by The British Library and also cases where decisions could not be made owing to missing information in a paper or abstract. In the case of this last scenario, an e-mail was sent to an author requesting further information.
During the selection process, any study identified that was thought to be of relevance to the cost-effectiveness review was cross-checked against the search results for that review to ensure comprehensiveness.
Assessment of risk of bias
Risk of bias was assessed by one reviewer and independently checked by a second reviewer. Disagreements were resolved by discussion.
Prognostic and diagnostic/predictive utility
As the review involved assessment of both prognostic and diagnostic/predictive utility, the quality assessment strategy involved using criteria of relevance from both the Quality Assessment of Diagnostic Accuracy Studies (revised tool) (QUADAS-2) guidelines79 for diagnostic test studies and criteria for checking the quality of prognostic studies suggested by Hayden et al.80
These criteria were compiled under the five domains outlined below with their corresponding assessment questions.
Prognostic models
If any prognostic models were included, the quality criteria described by Altman81 were to be used in addition to those of Hayden et al.80 Specific elements to be considered were:
methods of model development (selection of candidate risk variables, relative weighting, handling of continuous variables)
internal and external model validations
study design (prospective/retrospective)
sample size (considered a priori)
missing data (quantity, and how missing data were handled in the statistical analysis)
criteria for inclusion of prognostic factors into the model (adequately described, and whether or not well-known prognostic factors were included regardless of significance).
Any prognostic models identified were to be summarised qualitatively (summarising, for example, included variables, calculation of risk score, predictive accuracy and whether or not the model was validated internally and externally) and quantitatively by extracting performance statistics for calibration (such as observed/expected outcomes) and discrimination (such as sensitivity and specificity) of the model. Similarly, where studies reported the incremental value of including PFTs in prognostic models, these data were to be summarised.
Data extraction
Data extraction was conducted by one reviewer using a standardised, piloted data extraction form, and independently checked by a second. Disagreements were resolved through discussion or referral to a third reviewer.
The data extraction process was necessarily complex owing to the nature and variability of the included studies. Data extraction was undertaken directly into a specially created sheet in Microsoft Excel 2010 (Microsoft Corporation, Redmond, WA, USA). Extensive data related to the following domains were extracted: study design and characteristics; patient characteristics; antiplatelet regimens; PFT utilised; outcome measures and length of follow-up; data required for analyses; statistical methods employed and their appropriateness. Studies were grouped according to whether patients were on monotherapy (aspirin only) or dual therapy (with a second antiplatelet agent such as clopidogrel added to aspirin), in order to distinguish between patients in a stable (monotherapy) or acute phase (dual therapy) of thrombotic disease. Patients who have experienced ACS, or who have undergone PCI, will generally have a second agent added to their therapy for up to 1 year before reverting back to monotherapy. Note that for reasons outlined in Presentation of results, only results pertaining to monotherapy studies have been presented in this report.
For further details on data extracted, readers can consult a copy of the database via information presented in Appendix 4.
With regard to the data extracted for analysis, details are given in the following section.
Analysis
Data extraction for potential meta-analysis
A key analytical aim was to conduct meta-analysis for each test in relation to each clinical outcome reported by the individual studies. To do this, relevant data reported by the included studies needed to be extracted. Data extraction was conducted independently by two reviewers, and if necessary any differences were resolved via discussion with a third reviewer. If multiple cut-off levels were considered in a study (e.g. to define test ‘positive’ and test ‘negative’), then results were sought for each cut-off reported. Both unadjusted and adjusted results were extracted, as both were considered to be important. Unadjusted results help ascertain the prognostic ability of a test when it is used in isolation. Adjusted results reveal whether or not a test has prognostic utility over and above other prognostic factors; a true causal factor of poor outcome will retain strong prognostic value even after adjustment, and so this further informs the clinical value of a test.
Two groups of summary results were sought during data extraction, as follows.
Prognostic ability: unadjusted and adjusted odds ratios and hazard ratios
The prognostic ability of each test reveals its association with clinical outcome and provides the relative risk between groups defined by test values; for example, the odds of poor outcome in test-positive patients compared with test-negative patients.
For binary outcomes, the reported unadjusted OR and its 95% CI and p-value were extracted. If these were not available, data were sought to populate a 2 × 2 table, from which the values could be calculated directly. Any adjusted ORs (with CIs and p-values) reported were extracted along with the reported set of adjustment factors that were used.
When the follow-up is longer and/or there are patients lost to follow-up (censored), time-to-event analyses are more appropriate to account for different lengths of follow-up. When time-to-event analyses were reported (e.g. Cox regression analyses, log-rank tests), the unadjusted HR and its 95% CI and p-value were sought and extracted. If these were not provided directly, then the methods of Parmar et al.82 to indirectly estimate them from other available data were used. If these were not possible, and a 2 × 2 table was available for a particular time point, the method of Perneger83 was used; this method assumes that all patients are followed up for the same length of time. Any adjusted HRs (with CIs and p-values) reported and the set of adjustment factors that were used were also extracted. For studies using Cox regression, whether or not the proportional hazards assumption had been checked and was considered valid was recorded.
If studies reported results according to the test on its continuous scale, that prognostic result was extracted directly (and so did not force a categorisation). If results were presented for the test categorised into three or more groups (e.g. according to tertiles or quartiles), results for each comparison presented were extracted, but where possible the groups were collapsed down to a binary comparison (to be most comparable with other studies, which generally used a dichotomisation). This collapsing was only possible for calculating unadjusted ORs or unadjusted HRs when 2 × 2 tables could be derived; it was not possible for adjusted results.
If studies provided a 2 × 2 table with one or both groups with a zero cell, then a continuity correction was added to these in order to calculate effect sizes, using the method of Sweeting et al.84 The continuity correction added was 1/(sample size of the opposite group).
Diagnostic/predictive accuracy
If prognostic utility can be demonstrated, an evaluation of diagnostic/predictive utility of aspirin resistance requires assessment of whether or not PFTs are able to determine, for individual patients, if they are at increased risk of clinically important outcomes and thus warrant consideration of treatment modification.
Ordinarily, test accuracy is assessed on ability to distinguish between patients who are subject to a risk factor/carry a marker for disease, etc., and those who are not. However, in the current context of platelet function testing predicting future adverse clinical outcomes, diagnostic utility requires the test to identify the risk factor, and then the risk factor has to be intrinsically linked to the outcome. Thus, the diagnostic utility contains elements of the accuracy of the test in measuring platelet function and the strength of the association between the platelet function and the outcome. Furthermore, there is no single outcome in the current context and the risk of each possible outcome might vary over time. This means that, prior to assessment of diagnostic utility, it is important to have demonstrable association between the marker and outcome(s).
As will be seen in Chapter 5, no strong association was identified between any PTF and clinical outcome, thus determination of diagnostic utility is mute. However, where data were available to consider an assessment of diagnostic utility, the presence of these data was noted and they were extracted. Speculative analysis of sensitivities and specificities was undertaken and this is presented in Appendix 3 along with a description of the relevant analysis methods.
Meta-analysis methods
Once the summary results were extracted for each study and for each test, the clinical experts and researchers met to identify groups of similar patient groups and clinical outcomes across studies. For each patient group and outcome identified, the possibility for meta-analysis was considered; that is, whether or not suitable data were available from multiple studies for the same clinical outcome and test in relation to prognostic ability (relative risk scale: synthesis of ORs or HRs, taking unadjusted and adjusted results separately) and, speculatively, the diagnostic/predictive ability (absolute risk scale: sensitivity and specificity). Where possible, a separate meta-analysis for each cut-off level was considered. The intended methods for any meta-analyses were outlined in the protocol. As a result of the clinical and methodological heterogeneity between studies, pooling of data was determined to be inappropriate even in subgroups of studies employing the same PFT. However, data are presented in this report in forest plots (without the summary estimate) along with some relevant study characteristics highlighting heterogeneity.
Amendments to protocol
Initially the protocol did not specify that studies of patients on dual/triple antiplatelet therapy [i.e. aspirin with additional antiplatelet agent(s)] had to employ an aspirin-specific PFT, rather than any PFT. This was changed prior to study selection and the pertinent platelet function assays are reflected in Table 2.
It was originally stated that studies which met all of the inclusion criteria except for reporting clinical outcomes would be noted, as these might provide useful information for cost-effectiveness analysis (e.g. uncertainty around the prevalence of those defined as aspirin resistant from specific assays in specific populations). From very early in the study selection process, the protocol was amended to omit this owing to the very large number of studies being identified and limited benefit of identifying these across all the tests and populations.
These amendments were reported to the NIHR and a revised protocol was submitted.
Presentation of results
Throughout the following sections, our aim has been to highlight the heterogeneity between studies with regard to population, PFT, outcomes and analysis of studies.
Results have therefore been separated according to whether patients were receiving only aspirin as antiplatelet therapy (monotherapy) or aspirin and a second antiplatelet agent (dual therapy) at the time of the PFT. There are a number of reasons for this:
Populations receiving monotherapy are potentially likely to differ from those receiving dual therapy (e.g. they are less likely to have very recently had an acute cardiovascular event or to be undergoing non-elective PCI).
The influence of a second antiplatelet agent on an aspirin-specific PFT is unclear.
The second antiplatelet agent is likely to influence occurrence of clinical outcomes, and occurrence of outcome is fundamental to determination of prognostic utility.
Resistance to other antiplatelet agents is known, and may affect event rates.
The original intention was to report and analyse studies relating to both patients receiving monotherapy and those receiving dual therapy. It was decided to undertake a stepwise approach to the analysis, starting with monotherapy studies and then moving on to dual-therapy studies; based on the reasons listed above, it is possible that an association between aspirin resistance and clinical outcome may be more apparent within those populations receiving aspirin therapy alone, as it might be more difficult to demonstrate prognostic utility in patients receiving aspirin with additional antiplatelet therapy because of the potential added confounding effect of the other antiplatelet agent.
Furthermore, it is debateable whether or not analysis of studies with dual therapy is warranted in the absence of demonstrated prognostic utility of platelet function testing in patients treated with aspirin as monotherapy. As this criterion was not met (i.e. prognostic utility could not be adequately demonstrated), all results presented in the following sections relate to monotherapy only. However, in the interest of transparency the authors wish for all extracted and analytical data (including those from dual-therapy studies) to be available to readers of this report. The data have been made available through a web portal and further details can be found in Appendix 4, including how to access the data.
Monotherapy studies were further defined as those where all, or the vast majority of, patients were on monotherapy at the time of the PFT, given that treatment strategies may change over time depending on disease progression. Adding a second agent may affect the rate of clinical events, and this may not be independent of the underlying risk, as higher-risk patients are more likely to be receiving or to commence dual therapy. Where studies have clearly specified where a proportion of patients have at some point during the follow-up period switched therapy or received additional therapy, this information has been extracted. It is, however, possible that not all studies have reported this information.
Populations have been broadly classified as having (i) stable CAD, (ii) stable CVD/stroke, (iii) PAD/peripheral vascular disease (PVD) or (iv) unstable angina (UA)/ACS. Where patients are undergoing elective PCI (PCI) or primary PCI (PPCI), this has also been indicated. Where the population comprises several patient groups, this has been classified as miscellaneous. Note that some acute populations have been included where the PFT was undertaken when patients were on monotherapy.
Results have been separated for different PFTs, and where several thresholds or agonists have been used, this has been indicated. Where different PFTs have been used within the same study, results have been presented in Chapter 5, Studies with more than one test.
Outcomes have been classified as (i) death, (ii) MACE, (iii) ischaemic/thrombotic or (iv) haemorrhagic/bleeding. A consistent definition for MACE is not used in the literature;85 for example, it may or may not include stroke. For a composite outcome that includes cerebrovascular complications, the abbreviation MACCE is sometimes used (with the additional ‘C’ indicating the cerebrovascular component), but again, this is not consistent. Rather than devise a definition of what constitutes MACE or MACCE for this report, studies with a composite outcome of adverse cardiovascular events have been grouped together using the abbreviation MACE. Where stroke has been reported as a separate outcome, this has also been highlighted.
Within the categories of MACE/MACCE there are some inconsistencies between studies in how this has been defined; this has been appropriately highlighted where necessary. The category of ‘ischaemic/thrombotic’ events is broad and encompasses a number of different events such as revascularisation, angina, bypass surgery, cardiovascular readmission, graft occlusion, MI, etc.
The different outcome measures used in the studies have been summarised as a first step in deciding whether or not pooling is possible and to give an idea of the range of outcome measures used. They have been grouped according to the following: sensitivity and specificity, unadjusted or adjusted ORs, or unadjusted or adjusted HRs. Where HRs or ORs have not been presented but have been calculated for this report, this has been indicated. Additionally, where groups have been collapsed in order to provide a single threshold, this has also been indicated. Note that where outcomes have been reported for different test characteristics (e.g. different agonist, threshold, etc.), not all results will necessarily have been summarised using the same outcome measures.
Odds ratios and HRs provide information on the usefulness of a PFT as a prognostic risk factor. Adjusted ORs or HRs may take into account differences in clinical characteristics, which are linked to adverse events. At the least informative level, articles have only provided a narrative statement regarding the relationship between PFT results and clinical events.
Quality assessment of studies is also clearly presented to aid interpretation of findings.
Owing to the extensive nature of the data extracted from included studies for this project, it was deemed unfeasible to adequately present all the data in this report (even as appendices). The results section of the prognostic utility review in this report contains, where necessary, details of the studies, including the populations studied, test characteristics and quality-related features, and data for key outcomes are presented in illustrative forest plots.
