The RATPAC (Randomised Assessment of Treatment using Panel Assay of Cardiac markers) trial: a randomised controlled trial of point-of-care cardiac markers in the emergency department

Article history

The research reported in this issue of the journal was commissioned by the HTA programme as project number 06/302/19. The contractual start date was in April 2007. The draft report began editorial review in January 2010 and was accepted for publication in October 2010. As the funder, by devising a commissioning brief, the HTA programme specified the research question and study design. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the referees for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

none

Copyright statement

© Queen’s Printer and Controller of HMSO 2011. This work was produced by Goodacre et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This journal is a member of and subscribes to the principles of the Committee on Publication Ethics (COPE) (http://www.publicationethics.org/). This journal may be freely reproduced for the purposes of private research and study and may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising.Applications for commercial reproduction should be addressed to: NETSCC, Health Technology Assessment, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2011 Queen’s Printer and Controller of HMSO

Diagnostic assessment of acute chest pain

Chest pain is responsible for around 700,000 patient attendances per year in England and Wales, and around one-quarter of hospital admissions. 1 The main reason for attendance is the possibility of acute myocardial infarction (AMI). Patients with ST-elevation myocardial infarction (STEMI) can benefit from early coronary reperfusion using primary angioplasty or intravenous thrombolysis. Patients with non-ST-elevation myocardial infarction (NSTEMI), who are much more numerous than those with STEMI, benefit from hospital admission and treatment. The potential benefits of treatment for AMI have led to public awareness campaigns encouraging people with acute chest pain to call for emergency medical help and national guidelines recommending that patients with chest pain should call for an emergency ambulance to take them to a hospital emergency department (ED) rather than contact their general practitioner (GP). 2 Thus, ED attendances with chest pain are increasing.

Standard initial assessment of patients with acute chest pain consists of a clinical history and examination, 12-lead electrocardiogram (ECG) recording and a chest radiograph. On the basis of this assessment around 34% will have a clinical diagnosis of suspected acute coronary syndrome (ACS), 11% will have an ECG diagnosis of suspected acute coronary syndrome, 19% will have a clear diagnosis of benign non-cardiac chest pain (such as muscular pain) and 12% will have suspected serious non-cardiac chest pain (such as pulmonary embolus) or comorbidity (such as heart failure). The remaining 24% have no clear diagnosis but have a significant risk (5–10%) of undiagnosed AMI. 1

Acute myocardial infarction is usually diagnosed by serial cardiac biomarker testing (‘cardiac enzymes’) typically by measurement of troponin in the blood. A patient with ischaemic symptoms (such as chest pain) would be diagnosed as having AMI according to the universal definition for acute, evolving or recent AMI if a troponin level is recorded above the 99th percentile of the values for a reference control group. 3 These patients are likely to benefit from hospital admission. 4 Patients with no troponin elevation have a low risk of adverse outcome and are unlikely to benefit from hospital admission, unless they have a serious non-coronary cause for their pain (such as pulmonary embolus) or serious comorbidity (such as heart failure or arrhythmia). Rapid and accurate identification of patients with AMI is thus a hallmark of effective emergency care for chest pain.

Current recommendations suggest that patients with chest pain due to suspected, but not proven, AMI should receive diagnostic testing with a troponin sample taken 12 hours after symptom onset,5 the delay being necessary because troponin sensitivity does not reach optimal levels until this time. This approach is inconvenient and potentially costly because it requires many patients to be unnecessarily admitted to hospital until the time delay has elapsed. Most patients presenting to the ED with suspected AMI do not actually have subsequent confirmation of AMI, so their admission will ultimately prove avoidable. Economic analysis suggests that admitting patients for cardiac marker testing is not a cost-effective use of health service resources compared with early cardiac marker testing. 6 Recent studies7,8 have suggested that new high-sensitivity troponin assays may be able to rule out AMI earlier than 12 hours, although the effect of using these assays in practice has not yet been evaluated.

Evidence also suggests that these guidelines are often not followed in a busy emergency setting where acute beds are limited. Collinson et al. 9 showed that 7% of patients discharged after ED assessment for acute chest pain had elevated troponin levels at follow-up 2 days later. Goodacre et al. 10 showed that, in the standard-care arm of a randomised trial of a chest pain unit, 14% of patients with an elevated troponin level at 2-day follow-up had been sent home from the ED. A national survey of EDs11 asked the lead consultant what proportion of patients with undifferentiated chest pain would be admitted to hospital. Estimates varied from less than 20% to over 80%. Hence, it appears that the theoretical ideal of a 12-hour troponin is not realised in practice and, as a result, patients are inadvertently discharged home with undetected AMI.

The point-of-care cardiac marker panel

Rapid point-of-care testing using a panel of markers offers an alternative approach that may be more effective and cost-effective than current practice. This technology has two elements that offer putative benefits: (1) point-of-care testing to reduce the turnaround time to results being available and (2) the use of a combination of markers, including measurement of marker gradients, to optimise early sensitivity. This is based on the idea that other markers, such as myoglobin and creatine kinase MB (mass) [CK-MB (mass)], may start to rise earlier than cardiac troponin.

Point-of-care testing involves using an analyser in the ED for marker assays. The analyser needs to be operated quickly and reliably by clinical staff. It can then allow rapid provision of marker results with short turnaround times to guide decision-making. Several randomised controlled trials (RCTs) have compared point-of-care tests with the same test performed on a laboratory analyser to determine whether use of point-of-care testing reduces turnaround times and changes practice. Collinson et al. 12 showed that the use of point-of-care troponin testing on a coronary care unit led to reduced length of coronary care unit stay and overall hospital stay. Renaud et al. 13 showed that point-of-care troponin testing in an ED reduced time to anti-ischaemic therapy and physician notification of troponin results, but did not change ED length of stay or patient outcomes. Ryan et al. 14 evaluated point-of-care troponin testing in four EDs and found that the effect varied between settings, with length of stay in the ED being increased in one hospital and decreased in another. Kendall et al. 15 evaluated a variety of point-of-care tests in an ED and found that point-of-care testing reduced turnaround times and times to decisions being made, but did not influence clinical outcomes or length of stay. Overall, therefore, there is reasonable evidence that point-of-care testing reduces turnaround times and possibly also times to decisions being made, but no consistent evidence of an effect on length of stay. One crucial component may be the need to incorporate point-of-care testing within a decision-making protocol.

Combining markers in a panel aims to overcome the limitations of individual markers. Different cardiac markers have optimal sensitivity for AMI at different times after symptom onset. 16,17 Combining markers to form a panel that is positive if any one marker is positive should optimise sensitivity and ensure that fewer cases of AMI are missed. Measuring the gradient rise of markers between baseline (when the patient arrives at hospital) and a specified time later (typically 90 minutes) has been shown to improve early sensitivity18,19 and can also improve specificity if only the gradient rise (rather than the absolute value) is considered positive.

Most studies of point-of-care cardiac marker panels have evaluated the combination of CK-MB (mass), myoglobin and troponin I measured at presentation and 90 minutes later. These have shown that the panel has high sensitivity and can accurately rule out AMI by 90 minutes after presentation. 20–27 This results in earlier identification of AMI than with laboratory testing22 and expedited decision-making with turnaround times reduced by 55%. 23 Meanwhile, comparison of patient management with the panel with previous practice showed a 40% reduction in coronary care unit admissions. 24

These studies show that the point-of-care combination of CK-MB (mass), myoglobin and troponin I measured at presentation and 90 minutes has appropriate diagnostic accuracy, but they do not reliably tell us whether the panel will alter patient care, improve outcomes or reduce health service costs. Early diagnostic accuracy and reduced turnaround times will lead to changes in practice only if clinicians act upon the additional diagnostic information. The before and after study by Ng et al. 24 may be confounded by changes in coronary care referrals over time and, originating from the USA where coronary care usage is much higher than in the UK, may not be applicable to the UK NHS. To date, there have been no randomised trials comparing marker panels with routine practice.

Existing data therefore suggest that point-of-care cardiac markers can accelerate decision-making in the ED and that a panel of markers consisting of troponin, CK-MB (mass) and myoglobin, measured at baseline and 90 minutes later, can accurately identify patients with AMI. This strategy could allow rapid and accurate diagnosis in the ED or clinical decision unit, facilitating hospital admission for those with AMI and discharge home for those without. However, such a strategy needs to be evaluated in practice and compared with current management before it can be recommended for widespread adoption throughout the NHS. Evaluation allows us to determine whether clinical decision-making is changed by using the point-of-care panel, whether hospital admissions are actually reduced in practice, whether reduced hospital admissions save sufficient health service costs to compensate for the additional costs of point-of-care testing and whether patient outcomes are improved. In addition, advances in assay methodology for troponin suggest that troponin alone may be used to rule in and rule out AMI soon after presentation, and marker panels may be unnecessary. 28 This has also not been tested in a randomised controlled manner.

Research objectives

We aimed to measure the effect of using a point-of-care cardiac marker panel upon the following outcomes in patients presenting to the ED with suspected but not proven AMI:

1. the proportion of patients successfully discharged home after ED assessment

2. health utility and satisfaction with care

3. the use of coronary care beds and cardiac treatments

4. subsequent re-attendance at and/or re-admission to hospital

5. major adverse events (death, non-fatal AMI, life-threatening arrhythmia, emergency revascularisation or hospitalisation for myocardial ischaemia)

6. health and social care costs.

In addition, we planned to undertake secondary analysis of the data of the Randomised Assessment of Treatment using Panel Assay of Cardiac markers (RATPAC) trial to evaluate the TIMI (thrombolysis in myocardial infarction)29 and GRACE (Global Registry of Acute Coronary Events)30 clinical prediction scores and store blood samples taken from participants in the intervention arm to evaluate potential new or alternative markers. The results of the evaluation of TIMI and GRACE scores are presented in this report. Funding has been sought to undertake secondary blood sample analysis.

Study design

We undertook a multicentre pragmatic RCT and economic evaluation of a point-of-care cardiac marker panel in the management of patients with suspected, but not proven, AMI in six EDs in the UK.

Setting

The participating hospitals were Barnsley District General Hospital, Derriford Hospital in Plymouth, Edinburgh Royal Infirmary, Frenchay Hospital in Bristol, Leeds General Infirmary and Leicester Royal Infirmary. They were selected to provide a range of different settings that reflected the variation in current NHS practice and the variation in facilities available to manage patients with acute chest pain. Table 1 outlines the characteristics of the participating centres.

Participants

We recruited people presenting to the ED with chest pain due to suspected but not proven AMI in whom a negative point-of-care marker test could potentially rule out AMI and allow discharge home. All patients with chest pain were considered for participation but then excluded if they met any of the following criteria:

1. Diagnostic ECG changes for AMI or high-risk acute coronary syndrome (> 1 mm ST deviation or > 3 mm inverted T waves). These patients are at high risk of adverse outcome and require inpatient care even if marker tests are negative.

2. Known coronary heart disease (CHD) presenting with prolonged (> 1 hour) or recurrent episodes of typical cardiac-type pain. These patients have unstable angina and require inpatient care for symptom control even if marker tests are negative.

3. Proven or suspected serious non-coronary pathology (e.g. pulmonary embolus) that requires inpatient care even if AMI is ruled out.

4. Comorbidity or social problems that require hospital admission even if AMI can be ruled out.

5. An obvious non-cardiac morbidity (e.g. pneumothorax or muscular pain) in a patient in whom AMI can be excluded as a possible cause without resorting to further diagnostic testing.

6. Presentation more than 12 hours after the most significant episode of pain, in which case a single troponin measurement would clearly be more appropriate than point-of-care panel assessment.

7. Previous participation in the RATPAC trial.

8. Inability to understand the trial information owing to cognitive impairment.

9. Non-English-speaking patients for whom translation facilities were not available.

For every fourth week of trial recruitment the research nurse at each hospital examined ED attendance lists to identify patients attending with chest pain and record basic demographic details and reasons for exclusion. The huge number of attendances with chest pain meant that undertaking this process throughout the whole trial would have produced an excessive workload, whereas monitoring every fourth week achieved the aim of reporting sample selection within acceptable use of resources.

Recruitment and randomisation

Research nurses and ED staff identified eligible patients, provided trial information and obtained written consent. Participants were then randomly allocated to receive either (1) diagnostic assessment using the point-of-care biochemical marker panel or (2) conventional diagnostic assessment without the panel.

The Nottingham Clinical Trials Unit (CTU) generated a simple randomisation sequence, stratified by centre, which was not revealed to any person involved in patient recruitment. Recruiting doctors and research nurses accessed a secure website provided by Nottingham CTU and entered participant details. The CTU revealed the participant’s allocated treatment group to the ED only after the participant’s details were entered, written consent was confirmed and the participant irrevocably entered into the trial.

Planned interventions

Participants were randomised to receive either:

• diagnostic assessment using the point-of-care biochemical marker panel

  or

• conventional diagnostic assessment without the panel.

The only difference between the two arms of the trial was that patients in the intervention arm received assessment with the point-of-care panel. The use of all other tests and treatments, and decision-making in the ED, was at the discretion of the attending clinician.

The point-of-care cardiac marker panel comprised CK-MB (mass), myoglobin and troponin I, measured at presentation and 90 minutes later, using the Siemens Stratus^® CS Analyser (Dade Behring, Milton Keynes, UK). Clinical staff were trained to use the test and given guidance in interpretation of the results. We provided a recommended protocol that advised a first panel test immediately after initial ED assessment and a second panel test 90 minutes later, and then advised hospital admission or discharge on the basis of point-of-care results (Appendix 1). Decisions were ultimately at the discretion of clinical staff. We did not police the use of the point-of-care protocol to ensure that it was being followed.

Other than obtaining consent, collecting data and random allocation to use of the point-of-care test, the only change to routine practice was that we asked clinical staff to take an additional sample of blood for storage (without repeating venepuncture) each time a point-of-care blood sample was required. The additional blood sample taken after point-of-care testing was transported to the hospital laboratory to be centrifuged and the serum separated and then frozen. Batches of frozen samples were then transported quarterly to St George’s Hospital for longer-term storage and future secondary analysis of new biomarker assays.

The RATPAC trial was a pragmatic trial, intended to determine whether point-of-care panel assessment should be standard practice for patients presenting to the ED with suspected AMI. It was designed to compare two pragmatic alternatives (management with and without point-of-care panel assessment) under routine conditions to determine whether use of the test changes costs or outcomes. This pragmatic design had the following implications:

1. There was no attempt to blind clinical staff, patients or carers to the allocated treatment group after randomisation.

2. The point-of-care test was provided with a recommended protocol for use but management decisions were ultimately at the discretion of the clinical staff.

3. All other diagnostic tests and the use of laboratory blood tests in the control group were at the discretion of the clinical staff. Blood samples were taken only for the purposes of clinical management. We did not take blood samples at additional time points to evaluate theoretical management strategies or to evaluate the accuracy of diagnostic assessments. The blood samples taken for storage and future analysis were taken at the same time as the point-of-care samples used for clinical care and did not require additional venepuncture.

Point-of-care assays and analyser

Cardiac troponin I, CK-MB (mass) and myoglobin were measured by point-of-care testing in the ED on the Stratus CS Analyser. The original project protocol planned to use the Biosite Triage^® analyser (Biosite, San Diego, CA, USA), as this was the most widely used analyser at the time of protocol development. However, as protocol development continued it became apparent that new-generation high-sensitivity troponin assays, as well as point-of-care analysers with high-sensitivity troponin assays, were becoming available. 31 Discussions with experts in the field revealed reservations about the analytical performance of the Biosite system, particularly the analytical sensitivity of its troponin assay. In view of this, and the need to ensure the future generalisability and the development towards more sensitive troponin assays, it was deemed prudent to use a more sensitive system in the trial. Recent publications7,8 have since confirmed the potential of high-sensitivity troponin assays to improve early assessment.

The Stratus CS Analyser was selected on the basis of having most data as an instrument suitable both for the emergency laboratory and for use as a point-of-care instrument,32–34 as well as a troponin assay with performance characteristics close to current recommendations for analytical goals. It has the advantage that it can measure troponin, myoglobin and CK-MB (mass) on the same sample. 35 In addition, it has the health and safety advantages of using a closed system that does not require the operator to open the blood tube and pipette the sample on to a test strip.

Blood is drawn directly into a Vacutainer tube (a standard no-touch aseptic blood-sampling system) using lithium heparin as an anticoagulant. The tube is then inverted twice to mix the blood and anticoagulant. It is then ready for analysis. The sample is introduced directly into the machine, the sample door closed and analysis is fully automatic.

The analytical characteristics of the assays were as follows. Cardiac troponin I: detection limit 0.02 µg/l, analytical range 0.02–50 µg/l, interassay coefficient of variation (CV) 4.3–5.1% (0.03–0.22 µg/l). The 99th centile of the assay is 0.07 µg/l. Myoglobin: detection limit 1 µg/l, analytical range 1–900 µg/l, interassay CV 1.9–12.7% (56–308 µg/l); 95% reference interval, males 21–98 µg/ml, females 19–56 µg/l, combined 20–82 µg/l. CK-MB (mass): detection limit 0.3 µg/l, analytical range 0.3–150 µg/l, interassay CV 0.15–1.27% (3.7–39.3 µg/l); 95% reference interval 0.6–3.5 µg/l.

The recommended point-of-care protocol

The point-of-care protocol is outlined in Appendix 1. The sample(s) were deemed positive and hospital admission advised if any of the following were met:

1. any CK-MB (mass) exceeded 5 µg/l

2. the CK-MB (mass) gradient exceeded 1.6 µg/l

3. the myoglobin gradient increase exceeded 25% of the baseline value

4. troponin exceeded 0.02 µg/l.

The CK-MB (mass) gradient is based on data from Fesmire et al. ,18,19 suggesting that this value allows optimisation of both sensitivity and specificity. The myoglobin gradient was used alone because absolute myoglobin levels have very poor specificity, whereas using only the gradient rise maintains the value of early sensitivity without compromising specificity.

The Stratus CS Analyser is able to detect troponin levels in the range of 0.03–0.07 µg/l even though this is below the 99th centile of normal, which is conventionally used as a diagnostic threshold. Our understanding of the significance of very low troponin levels developed during the trial and is continuing to develop now (indeed we anticipate that secondary analysis of RATPAC blood samples will provide a further contribution). At the start of the trial we decided to err on the side of caution and recommended that any detectable troponin above 0.02 µg/l should be considered positive in case it represented the start of a significant rise. However, it became apparent from emerging data that levels between 0.03 and 0.07 µg/l do not typically represent an early troponin rise and where they do this is evident upon measuring the difference between baseline and 90-minute values. We therefore amended the guidance to recommend that the tests only be considered positive on the troponin assay if:

• any troponin level exceeded 0.07 µg/l

or

• the initial troponin is below 0.03 µg/l and the second sample is above 0.02 µg/l.

The new guidance was disseminated and implemented between December 2008 and February 2009.

Standard care

The standard-care group were managed without point-of-care panel assessment according to existing guidance for management of low-risk chest pain due to suspected ACS at each of the participating hospitals. We deliberately selected a variety of hospitals that were operating a range of different strategies and had a variety of different facilities available for chest pain management. Table 2 summarises the location of care and biomarker(s) used for low-risk patients in each hospital. Patients were referred to the cardiology team if biomarkers were positive and discharged (with or without exercise treadmill testing) if negative.

Outcome measures

The primary outcome was the proportion of patients successfully discharged home after ED assessment. To be considered successfully discharged the patient had to have both (1) left the hospital or be awaiting transport home with a discharge decision having been made at 4 hours after initial presentation and (2) suffered no major adverse event (as defined below) during the following 3 months.

Secondary outcomes were:

1. Health utility measured using the European Quality of Life-5 Dimensions (EQ-5D) self-complete questionnaire at 1 and 3 months after attendance.

2. Satisfaction with care measured at 1 month after attendance using a modified Group Health Association of America questionnaire that had been used successfully in previous studies of diagnostic strategies for acute chest pain. 10,36

3. The proportion of patients admitted to the coronary care unit, receiving cardiac medications (aspirin, heparin, clopidogrel or glycoprotein IIb/IIIa inhibitors) or receiving cardiac interventions [angiography, percutaneous coronary intervention (PCI) or bypass grafting].

4. Length of initial hospital stay and total inpatient days over 3 months.

5. Re-attendance at and/or re-admission to hospital and outpatient attendances over the following 3 months.

6. Adverse events (death, non-fatal AMI, life-threatening arrhythmia, emergency revascularisation or hospitalisation for myocardial ischaemia).

7. The proportion of admitted patients ultimately diagnosed as having AMI by the universal definition. 3

We selected successful discharge home as the primary outcome because the main purpose of point-of-care cardiac marker testing in this patient group is to facilitate discharge home. This outcome is beneficial for patients, who avoid the inconvenience and risks of hospital admission, and is beneficial for the health service, which avoids unnecessary admissions and pressure upon acute and emergency services. Patients who suffered an adverse event after discharge were not classified as a successful discharge home because they may have benefited from hospital admission. We also recorded the proportion of admitted patients who were ultimately diagnosed as having AMI to provide a measure of the appropriateness of admissions.

Assessment of outcomes

Recruiting staff recorded baseline data (including the variables required to calculate the TIMI or GRACE score), the results of initial assessment (including any biochemical cardiac tests) and admission or discharge from the ED. Research nurses then used ED and hospital inpatient notes to record management decisions at initial attendance and admission, extract resource use data and identify subsequent attendances/admissions and adverse events up to 3 months. Time and date of discharge for the initial admission were recorded as precisely as possible using computer records, case notes and contact with hospital staff. The total number of inpatient days over 3 months was recorded using hospital notes. An inpatient day was defined as being an overnight stay.

Research nurses checked patient status (dead or alive) at 1 and 3 months, using hospital information systems. Deceased patients were assumed to have a score of zero on EQ-5D and were excluded from other patient-based assessments. Participants who were not recorded as dead were mailed a questionnaire at 1 and 3 months from the University of Sheffield to identify adverse events and hospital attendances, health and social care resource use, and measure EQ-5D and satisfaction with care (satisfaction at 1 month only). Our previous study suggested a 70–80% response rate to this questionnaire. 10,37

A single reviewer (SG) blinded to treatment group classified all ED re-attendances, subsequent hospital admissions and outpatient reviews as either potentially chest pain related (including non-cardiac conditions that could have initially presented as chest pain) or clearly non-chest pain related (such as limb injuries).

The initial working diagnosis and final diagnosis were recorded and categorised by the research nurse, based upon the diagnosis recorded in the notes by the most senior clinician at the end of initial ED assessment and at the end of hospital admission, respectively. Patients were classified as having AMI on the basis of the presence of a rise in their troponin level above the diagnostic threshold of the relevant assay and absence of a final diagnosis of an alternative condition (such as sepsis or pulmonary embolism) that could have produced a troponin elevation. Patients with a troponin rise consistent with AMI and final diagnosis of ACS or ‘other AMI’ were classified as having AMI. Patients with no troponin rise and a final diagnosis that was not ACS or ‘other AMI’ were classified as not having AMI. A single reviewer blinded to treatment group reviewed the initial and next-day ECGs of patients with a final diagnosis of ACS and no troponin rise and categorised these patients as having AMI if an ECG showed ST elevation and coronary reperfusion was performed, otherwise they were categorised as having no AMI. Two independent reviewers blinded to treatment group reviewed case details of all patients with a troponin rise and a final diagnosis other than ACS or ‘other AMI’, and all patients with a troponin rise that was inconsistent with AMI (e.g. if a positive troponin was shortly followed by a negative troponin result). Each decided whether AMI was the most likely diagnosis. Disagreements were resolved by discussion and patients classified as having AMI or no AMI.

Proposed sample size

We planned to recruit 3130 participants to the trial. A previous randomised trial in this patient group suggested that around 50% of the control group would be successfully discharged. 10 With 1565 evaluable subjects in each arm of the trial we expected to have 80% power to detect a 5% absolute difference in the primary outcome (50% vs 55%) at the two-sided significance level of 5%. The same sample size provided 80% power to detect a 2% absolute difference (2% vs 4%) in major adverse events (death, non-fatal AMI, life-threatening arrhythmia, emergency revascularisation or hospitalisation for myocardial ischaemia), again at the two-sided 5% level of significance.

We estimated that we would require six hospitals to recruit for 12 months each to achieve the sample size of 3130, assuming that we recruited 70% of those eligible. Previous studies of this specific patient group undertaken by our team had shown that recruitment of 550 suitable patients per year is attainable at a typical hospital. 10,37–39

Previous studies had also shown a response rate of 70–80% for postal questionnaires,10,37 thus providing an effective sample size of at least 1000 in each of the two groups to evaluate health utility, satisfaction with care and health service resource use.

Statistical analysis

We planned to analyse the primary outcome through logistic regression, fitting concurrently with intervention group the effect of centre and appropriate baseline measures (including age, gender and past history of CHD), to present adjusted odds ratios (ORs) along with their corresponding 95% confidence intervals (CIs). A similar analysis was used for adverse events. Primary analysis was on an intention-to-treat basis. Secondary analysis excluded those who were not managed according to their allocated strategy.

We undertook a descriptive assessment to explore whether use of biochemical cardiac markers or admission rates changed over time in either the intervention or control group, either as a result of staff ‘learning curves’ in the intervention group or as a result of contamination of the control group.

For each patient, GRACE and TIMI scores were calculated. Two approaches were undertaken for calculating each score: one using the first available troponin sample and the second using the highest troponin level taken during the initial hospital stay. Some patients did not have all of the data items that were required to permit calculation of the score so two analyses were undertaken: one including only cases with complete data, the other including all cases with imputation to complete the data set. Imputation involved assuming that missing variables would be negative or, where appropriate, the mean value for the RATPAC population (e.g. continuous variables used in the GRACE score). We evaluated the predictive value of GRACE and TIMI by calculating the proportion with an adverse outcome at 1 and 3 months in each quintile of GRACE score and each TIMI category. We then calculated the area under the receiver-operator characteristic curve (c-statistic) for each score.

Economic evaluation

Economic evaluation was undertaken alongside the trial using recommended practice. 40 In addition, a cost-effectiveness model was developed to duplicate the trial results (as a way of validation) and extrapolate the results to longer follow-up periods. The NHS perspective was undertaken and other methods were in line with National Institute for Health and Clinical Excellence (NICE) Technology Appraisal Guidelines. 41

Resource use data were collected for all patients covering the length of time in the ED, the use of diagnostic tests, admissions, re-admissions, outpatient reviews and cardiac procedures. Cost and outcome data were collected using patient notes and self-completed questionnaires as described previously. A small microcosting study of 30–40 patients was carried out at each site, gathering data on staff times relating to the care of patients. ED cost per minute was based on a study previously undertaken by the investigators,10 and amended using the microcosting data from this study. Panel costs were based on purchase price, and the remaining costs were valued using national unit costs. 42,43 Total NHS costs up to 3 months after initial attendance were calculated. Quality-adjusted life-years (QALYs) were calculated by the trapezium rule using the EQ-5D tariff values at all follow-up points.

Economic analysis

Both cost and QALY analysis compared bootstrap estimates of the mean cost per patient of the two groups. Cost-effectiveness analysis estimated the incremental cost per QALY of using point-of-care cardiac marker testing compared with management without point-of-care panel assessment. Results were plotted on the cost-effectiveness plane and then transformed into cost-effectiveness acceptability curves with their associated frontier. 44 A sensitivity analysis was undertaken to include production losses as reported by the patient.

We anticipated that some of the resource use and QALY data would be incomplete (missing). Thus, in order to maximise the information collected from the trial, we imputed missing values using multiple imputation within Stata (StataCorp LP, College Station, TX, USA). 45 The idea of multiple imputation draws from the fact that missing values from incomplete data are unknown and the technique of multiple imputation imputes more than one likely value for the missing data; hence, providing an unbiased representation of uncertainty. 46 Thus, an additional set of results is available from the imputed cost and QALY data.

Decision-analytic model

We constructed a decision-analytic model to describe the care observed in the trial, and likely care pathways subsequent to it. This allowed us to systematically investigate the impact of subsequent costs, quality of life and survival. These values were initially based on population norms, but then replaced with literature review estimates where appropriate. Finally, they were replaced by RATPAC trial estimates where required. The decision-analytic model was probabilistic, but with conventional sensitivity analysis used to assess the impact of structural uncertainties. 47

The main purpose of the decision-analytic model was to explore the potential impact of changes in the major adverse event rate upon cost-effectiveness. We anticipated that the trial would have adequate power to detect economically important differences in resource use and EQ-5D, but would only have limited power to detect differences in major adverse events. It was therefore possible that an intervention could be apparently cost-effective (in terms of reducing costs and increasing health utility), while being associated with a statistically non-significant increase in adverse events. The decision-analytic model was intended to explore whether uncertainty around the effect of the intervention upon the major adverse event rate could influence the potential cost-effectiveness of the intervention.

The decision-analytic model used trial data to estimate costs and QALYs up to 3 months. Beyond this, we used lifetime cost and QALY estimates from a previous economic analysis of a similar population. 48 The mean lifetime cost of care for a patient with CHD was estimated to be £10,079 [standard error (SE) = £2200] and mean QALYs accrued were estimated to be 6.829 (SE = 0.34), while estimates of £0 and 20 QALYs [standard deviation (SD) = 5] were used for patients without CHD. Trial data were used to estimate the rates of death and non-fatal AMI by 3 months after management with the point-of-care panel and standard care. We then assumed that those who had died by 3 months would accrue no further costs or QALYs, while those who survived with non-fatal AMI would accrue costs and QALYs associated with CHD. Probabilistic sensitivity analysis was then used to estimate the mean lifetime costs and QALYs of patients in the two arms of the trial.

The decision-analytic model was developed and implemented in an Excel (Microsoft Corporation, Redmond, WA, USA) spreadsheet, which included a Visual Basic macro to implement the replication of outcomes from the model. The results presented were based on 10,000 outcome simulations. The pathways through the model described (1) the choice of diagnostic tool, using either point-of-care tests or standard-care approaches; (2) the disposal from the ED, either to admission or to discharge; (3) possible interventions or care once a positive diagnosis of ACS (and therefore admission) was made, including coronary artery bypass graft (CABG), PCI, thrombolysis, hospital stay with no intervention or thrombolysis treatment, and discharge without intervention in the case of a negative updated diagnosis, or death from ACS while in hospital; and (4) possible ACS events after discharge without diagnosis from the ED, which resulted in subsequent admission and treatment according to the treatments outlined in (3). Costs were accrued as simulated patients passed through the various stages in the model, and benefits were accrued at the end of each passage. Once discharged, either from ED or after admission, costs were assumed to be unaffected by the choice of diagnostic technique, and therefore costs of follow-up care or intervention were combined across both arms.

Ethical arrangements

All participants were asked to provide written, informed consent. Although participants were recruited in an emergency setting and there was only a limited amount of time available for considering trial information, the nature of the selected group (in particular the exclusion of people clearly requiring hospital treatment) ensured that eligible patients would not be incapacitated by their medical condition. We did not therefore make provision for recruitment of incapacitated patients by personal or professional legal representatives.

Ethical approval was granted by Leeds East Research Ethics Committee and review provided by the local research ethics committee (LREC) at each participating centre. The trial was conducted in accordance with Medical Research Council (MRC) Guidelines for Good Clinical Practice in Clinical Trials. 49 The University of Sheffield was the sponsor for the trial. The Trial Steering Committee (TSC) consisted of the Chief Investigator (SG), one of the co-applicants (PC), an independent chair (MF), two independent members (SH, JK) and a consumer representative (EH). We also invited a representative of the funder to join the committee but this was declined. The Data Monitoring and Ethics Committee (DMEC) consisted of an independent statistician (HT), emergency physician (WT) and cardiologist (JG), who were asked to review trial data at regular intervals and implement stopping rules in accordance with MRC guidance. The Trial Management Group (TMG) consisted of the Chief Investigator, co-applicants, Principal Investigator at each site, project manager, statistician, health economist and research nurses.

Data management

Trial data were collected on the case report form and follow-up form and then entered by the research nurses into an online database provided on a secure central server by the Sheffield CTU. The system had a full electronic audit trail. Quality control procedures were applied to validate the trial data. The project manager undertook quarterly data monitoring visits to each site, at which a random sample of data forms were checked for errors and validated against source documents. Any errors, protocol deviations or violations were reviewed with the research nurse and Principal Investigator and documented in the site files and master file. Central monitoring involved flagging discrepant or questionable data. Error reports were generated where data clarification was required. Monthly outputs were generated for the TMG and TSC summarising baseline characteristics of patients recruited, follow-up rates, data completion and adverse events by study site, but not study group. Outcome data summaries were provided for closed meetings of the DMEC. All activities were performed in accordance with Sheffield CTU standard operating procedures.

Trial progress

The project started on 1 April 2007. We planned to recruit staff and gain ethical and local governance approvals in months 1–6, recruit patients in months 7–18 and complete follow-up, data analysis, writing-up and dissemination in months 19–24. Actual trial progress was slower than anticipated: staff recruitment, ethics and regulatory approvals took up to 12 months to complete, patient recruitment was slower than expected, and the trial was terminated before the target of 3130 patients was recruited. The Gantt chart in Figure 1 outlines the planned and actual trial progress.

FIGURE 1.

Gantt chart of trial timetable and progress.

Table 3 shows the timing of processes at the six participating centres.

The reasons for the set-up phase being longer than expected are as follows:

1. The process of drawing up contracts between the sponsor (the University of Sheffield) and the six participating hospitals took longer than expected.

2. Local ethics and regulatory approvals could not be processed at some sites until contracting had been completed.

3. The time taken to provide local ethics and, in particular, governance approvals varied between hospitals.

4. In addition to trial training, some of the participating hospitals required all staff involved in the trial, including recruiting doctors, to have received formal training in Good Clinical Practice (GCP), even though the RATPAC trial was not a trial of an investigational medicinal product.

Patient recruitment was slower than anticipated and varied between 300 and 400 patients per centre per year of recruitment, compared with the expected 550 per centre per year. Details of patient recruitment are reported below (see Results) but it appeared that failure to achieve the expected recruitment rate was due to failure to recruit the expected proportion of eligible patients rather than any lack of eligible patients. Overall 35% (604/1719 – see Table 4) of eligible patients were recruited instead of the 70% anticipated. The following factors are likely to have accounted for this difference:

1. The requirement for GCP training resulted in doctors being unwilling to assist with trial recruitment. This varied between centres, depending upon the local regulatory requirements. In one centre the need for all doctors to receive formal GCP training resulted in most patients being recruited by research nurses. This centre recruited the lowest proportion of eligible patients (19%).

2. Service commitments, particularly the target of discharging 98% of patients from the ED within 4 hours of arrival, discouraged ED staff from recruiting patients during busy times.

3. The recall of point-of-care testpaks (described below) may have reduced the confidence of ED staff in the technology and thus their willingness to recruit patients to the trial.

Changes to protocol and other unanticipated events

A number of unanticipated events occurred during the trial, two of which resulted in changes to the protocol. Most of the unanticipated events related to the intervention being evaluated. Point-of-care cardiac markers are a developing technology and clinicians are learning more about the technology as it is implemented. Even mature technologies can be subject to unanticipated problems, especially if used by a wide range of staff in a variety of settings.

The unanticipated events were:

1. change of the point-of-care analyser

2. recall of point-of-care testpaks

3. amendment of the point-of-care protocol

4. early termination of trial recruitment.

The first and third of these resulted in changes to the protocol. In addition, we became aware that we had not included life-threatening arrhythmia as a major adverse event in the trial protocol, although it was included as such in the case report form. We therefore amended the protocol to include life-threatening arrhythmia as a major adverse event.

Change of the point-of-care analyser

This occurred during protocol development and before the trial started. Details and the reasons for this are outlined above (see Point-of-care assays and analyser).

Recall of point-of-care testpaks

Siemens informed the RATPAC research team on 4 July 2008 of the need to recall point-of-care assays for troponin I because of a faulty batch that could have produced low false-positive levels. All six sites were affected, resulting in a suspension of recruitment at all sites for up to 1 week between 4 and 14 July 2008 until new assays were delivered. A meeting was held with Siemens and the biochemical expert on the Research Team (PC) to address concerns. It was concluded that quality assurance was tight, particularly with high-risk tests, so this would be a rare rather than a common occurrence. It was also discussed in length at a TMG meeting in July 2008. It was agreed that patients with a potentially false-positive point-of-care troponin result would have received appropriate follow-up based on their 12-hour troponin level and clinical condition, and therefore no additional intervention was required.

Amendment of the point-of-care protocol

Details and the reasons for this are outlined above (see The recommended point-of-care protocol).

Early termination of trial recruitment

Owing to the set-up delays and slower than anticipated recruitment, a request for additional funding was submitted to allow completion of the trial. The DMEC undertook a futility analysis (Appendix 2). This showed that the conditional power for the primary outcome recalculated using data to May 2009 was > 99.9% and the conditional power against major cardiac events was < 10%. In the light of this analysis, the trial funder declined the request for additional funding, based on clear efficacy (primary outcome) and futility (major cardiac events), and, consequently, trial recruitment was terminated on 2 June 2009, with a total of 2263 patients recruited. The findings reported here represent those as accrued at the point at which this decision to halt the trial was taken.

Screening, recruitment, randomisation and follow-up

Patients were recruited between 30 January 2008 and 2 June 2009. Table 4 summarises the process of screening and recruitment at each centre. All hospitals stopped recruitment on 2 June 2009 but the staggered start meant that some recruited for more days than others. Overall, 2263 patients were recruited over a total of 2658 hospital-days (0.9 patients per day). The total at each site ranged from 327 to 469, and the recruitment rate ranged from 0.7 to 1.1 per day.

Research nurses screened 9109 chest pain attendances on a total of 667 hospital-days during the trial (13.7 patients per hospital per day). Overall, 1719/9109 (19%) were eligible for recruitment (2.6 per hospital per day), of whom 604/1719 (35%) were recruited. The proportion of eligible patients recruited varied across the sites from 19% to 75%.

The proportions excluded for each exclusion criterion were as follows: 14% had ECG changes, 15% had known CHD with prolonged or recurrent pain, 8% had suspected serious non-CHD pathology, 5% had comorbidities or social problems, 28% had obvious non-cardiac pain, 5% presented > 12 hours since their worst pain, < 1% had previously participated in the trial, 2% were unable to understand the trial information owing to cognitive impairment or being non-English speaking, 2% had other exclusion criteria, 3% had an unknown reason for exclusion and < 1% declined consent. These proportions varied across the sites. We were unable to verify whether this was due to differences in coding and classification, or differences in population characteristics.

Figure 2 shows the flow of patients through the trial after recruitment and Table 5 shows withdrawals and losses to follow-up. We intended to follow up patients who were recruited in error, provided they did not withdraw their consent, so that all consented patients could be analysed as randomised. However, six patients (three in each group) were not followed up after being recruited in error for specific reasons necessitating their exclusion (prisoners, who were defined as ‘vulnerable groups’, n = 3; clinician error, n = 2; and recruitment when the trial was suspended, n = 1). In the point-of-care group seven patients withdrew or were lost to follow-up before any outcomes were recorded, and a further 18 withdrew or were lost to follow-up by the end of the trial. These numbers were 13 and 10, respectively, in the standard-care group. Questionnaire response rates were slightly higher in the point-of-care group: 75% at 30 days and 69% at 90 days, as opposed to 71% and 66%, respectively, for the standard-care group.

FIGURE 2.

CONSORT chart showing flow of patients after recruitment.

Research nurses checked whether patients had been appropriately recruited and the results are shown in Table 6. In addition to the six patients who were randomised in error and not followed up (reported above), a total of 23 patients (1%) were inappropriately recruited: 12 in the point-of-care group and 11 in the standard-care group. These cases were followed up and analysed as planned on an intention-to-treat basis.

Protocol violations

Some 140 patients (12.4%) in the point-of-care group did not receive testing as recommended in the protocol, i.e. two panels consisting of CK-MB (mass), myoglobin and troponin (or one panel if the first troponin was positive). Of these, 34 did not receive any point-of-care testing at all, typically because of machine failure or inability to use the machine, while the remainder had fewer tests than recommended in the protocol. Two patients (0.2%) in the standard-care group received point-of-care testing.

Baseline characteristics, emergency department findings and diagnosis

Table 7 shows the characteristics and demographics of the recruited patients. The characteristics were as anticipated, with a mean age of 54.5 years, more men than women, and 12% with a past history of CHD.

Table 8 shows the presenting characteristics of the patients. The median duration between the onset of worst pain and arrival at hospital was just over 2 hours, excluding the 1% who suffered their worst pain after hospital arrival. There were no marked differences between the study groups in the nature or duration of pain, or the associated features.

Table 9 shows the initial ECG and examination findings. A small proportion of patients had ST-segment deviations or T-wave inversions, although these were not deemed to be characteristic of myocardial ischaemia by the research nurse or recruiting doctor.

Table 10 shows the events in the ED, working diagnosis after initial assessment, the next-day ECG findings and final diagnosis. The proportions in each category differed between the final and initial diagnosis in both groups. More patients had a non-specific or other final diagnosis, and fewer had a final diagnosis of angina or ACS.

Blood testing

Table 11 shows the number and proportion of the point-of-care group who received each test, the number and proportion with a positive test, and the mean (median) time from worst symptoms to sampling. Most positive cases arose from the first troponin sample, although about one-quarter arose from the second sample.

Table 12 shows the number and proportion of patients in each study group receiving each laboratory blood test. Although most patients in the point-of-care group had negative point-of-care tests, a substantial proportion went on to have laboratory testing with troponin I or T. In the standard-care group around 90% received laboratory testing with troponin I or T. Other cardiac biomarkers were rarely used.

Primary efficacy

Table 13 shows the results for the primary outcome, i.e. successful discharge home. This was defined as having left the hospital or as a decision to discharge made 4 hours after initial presentation, and no adverse event over the following 3 months. Some 509 patients were defined as discharged at initial presentation: 453 had left the hospital by 4 hours and 56 had a decision to discharge. However, five of these patients had a major adverse event over the following 3 months, so 504 were defined as being successfully discharged home.

The primary efficacy analyses were adjusted, as planned, for hospital, age, gender and known CHD. Patients in the point-of-care group were significantly more likely to be successfully discharged home (OR = 3.81, 95% CI 3.01 to 4.82, p < 0.001).

Variation between sites

Table 14 shows the results for each individual hospital. Full details are supplied in Appendix 3. There was marked variation in the primary outcome across the six hospitals. Point-of-care panel assessment was associated with substantial increases in successful discharge rates at Barnsley and Edinburgh, modest increases at Derriford and Frenchay, and no increase at Leeds or Leicester.

TABLE 14 - Primary outcome by study centre

PoC, point of care; RD, risk difference; SC, standard care.

Adjusted for age, gender and known CHD. Overall result is also adjusted for centre. The p-value corresponds to adjusted OR.

Absolute risk difference.

	PoC [n (%)]	SC [n (%)]	ORa (95% CI)	RDb (95% CI)	p-value
Overall	358/1125 (32)	146/1118 (13)	3.81 (3.01 to 4.82)	18.7 (15.4 to 22.1)	< 0.001
Barnsley	110/162 (68)	43/164 (26)	6.97 (4.18 to 11.63)	41.3 (31.4 to 51.1)	< 0.001
Derriford	43/164 (26)	21/164 (13)	2.48 (1.37 to 4.49)	13.4 (5.0 to 21.9)	0.003
Edinburgh	104/228 (46)	16/224 (7)	11.07 (6.23 to 19.66)	38.4 (31.1 to 45.6)	< 0.001
Frenchay	50/233 (21)	9/231 (4)	7.03 (3.35 to 14.75)	17.4 (11.6 to 23.2)	< 0.001
Leeds	1/173 (1)	8/171 (5)	0.12 (0.01 to 1.03)	–4.0 (–7.2 to –0.7)	0.054
Leicester	50/165 (30)	49/164 (30)	1.11 (0.66 to 1.84)	0.4 (–9.5 to 10.3)	0.699

Figure 3 shows how the proportion of patients in hospital varies with time from arrival and Figure 4 shows this for each individual hospital. Overall, point-of-care panel assessment is associated with fewer patients being in hospital between 4 and 24 hours after arrival. However, there was again variation between the hospitals. Point-of-care panel assessment was associated with markedly fewer patients being in hospital up to 24 hours at Barnsley and Edinburgh. At Derriford the difference in proportion in hospital was apparent only between 4 and 8 hours. At Frenchay the difference was marked up to 12 hours, but after 12 hours the proportion of patients in hospital was greater in the point-of-care group. At Leeds the difference between the groups did not emerge until 6 hours after attendance, but between 6 and 24 hours point-of-care panel assessment was associated with markedly fewer patients being in hospital. Finally, at Leicester there was no difference in the proportions in hospital up to 12 hours and only slightly fewer patients in hospital in the point-of-care group from 12 to 36 hours.

FIGURE 3.

Duration from arrival to discharge from hospital (all centres).

FIGURE 4.

Duration from arrival to discharge from hospital (individual centres).

These differences probably reflect differences in standard care practice or the facilities for patients with chest pain at the hospitals. Standard care at Derriford was based on a troponin level measured 6 hours after arrival in hospital. This explains why the effect of point-of-care panel assessment at Derriford was limited to the 4- to 8-hour window. At Leeds all patients with chest pain are admitted to a clinical decision unit where diagnosis and management are undertaken without the pressure of the 4-hour target. This probably explains why the effect of point-of-care panel assessment was delayed at Leeds until 6 hours after arrival. At Leicester the point-of-care tests did not seem to alter decision-making with regards to hospital admission.

Secondary efficacy

Length and location of hospital stay

The mean length of the initial hospital stay was 29.6 hours in the point-of-care group and 31.8 hours in the standard-care group [mean difference = 2.1 hours, 95% CI –3.7 to 8.0 hours, p = 0.462 (t-test)]. The median length of initial hospital stay was markedly shorter for both groups, reflecting the positively skewed nature of length-of-stay data: 8.8 hours for the point-of-care group and 14.2 hours for the standard-care group (p < 0.001, equality-of-medians test). The two groups had significantly different distributions of length of stay (p < 0.001, Mann–Whitney U-test). Point-of-care panel assessment was therefore associated with reduced probability of hospital admission and shorter median length of stay, but there was no significant difference in mean length of stay.

Table 17 shows the total number of days spent in hospital over the 3-month follow-up and the number of days spent in coronary care or intensive care, including the initial hospital admission. We did not record fractions of days spent in hospital for this analysis and recorded an inpatient day only if the patient stayed overnight. Hence, a proportion of patients were recorded as having no inpatient days if their initial hospital visit did not result in an overnight stay and they were not admitted on a subsequent occasion. More patients in the point-of-care group had no inpatient days recorded during follow-up (54% vs 40%, p < 0.001). This reflected the difference in the proportion of patients in Figure 5 who were no longer in hospital 12 hours after attendance, as patients who are discharged before 12 hours and are not re-admitted are likely to be recorded as having no inpatient days over follow-up.

TABLE 17 - Inpatient care at 3 months

PoC, point of care; SC, standard care.

	PoC (n = 1125)	SC (n = 1118)	Total (n = 2243)	Percentage difference (95% CI)	p-value
Days in hospital at any location
n (%) with at least one day	518 (46.0)	670 (59.9)	1188 (53.0)	–13.9 (–18.0 to –9.8)	< 0.001
Mean days per patient	1.8	1.7	1.7		0.815
Total days (all patients)	1961	1908	3869
Days in coronary care
n (%) with at least one day	50 (4.4)	31 (2.8)	81 (3.6)	1.7 (0.1 to 3.2)	0.041
Mean days per patient	0.17	0.09	0.13		0.033
Total days (all patients)	190	105	295
Days in intensive care
n (%) with at least one day	8 (< 1)	3 (< 1)	11 (< 1)	0.4 (–0.2 to 1.0)	0.225
Mean days per patient	0.08	0.01	0.04		0.259
Total days (all patients)	76	14	90

FIGURE 5.

Histograms of number of inpatient days.

Despite this difference in the proportion with no inpatient days, there was no difference in the mean days per patient or the total inpatient days in the two groups. Figure 5 compares histograms for total days spent in hospital in the two groups to show why this was. The difference between the two groups in the proportions of patients having no inpatient stay was reflected in the difference in proportions having one inpatient day. So point-of-care panel assessment saved one inpatient day in about 14%. However, a few patients in both groups had a very high number of inpatient days (over 30). There were a few more of these in the point-of-care group, and enough to contribute disproportionately to the mean and total inpatient days for the group. Hence, point-of-care panel assessment reduced the proportion of patients completely avoiding admission but did not reduce overall inpatient days. Point-of-care panel assessment was also associated with a higher use of coronary care over follow-up (4% vs 3%, p = 0.041), although only a small minority of patients received any coronary care or intensive care.

Economic analysis

Tables 32 and 33 show the unit costs used in the main analysis and the microcosting, respectively.

Table 34 shows the unit costs used to estimate the cost per patient of providing point-of-care panel assessment. The total cost per patient of point-of-care panel assessment was estimated as the sum of a number of components: the cost of the machine and of periodic maintenance, the cost of calibration and quality control during testing, and the annual testing rate for the machine. Information was provided by Siemens Healthcare Diagnostics. Machine and maintenance costs were annualised to the index year (2009). The cost applied to the analyses reported here is that for an annual rate of 1500 full panels (£21.33 per panel). It is worth noting that the relationship between cost and annual rate of testing is known with certainty only for the rates reported in Table 34.

Data were collected from 246 patients for the microcosting study. The average age of this sample was 54 years, 52% were male (slightly lower than the overall percentage: 58% – Table 7) and 12% had a history of CHD. The results are shown in Table 35. The point-of-care panel assessment added £38.13 per patient managed in this arm of the trial. Staff costs were £11.55 higher in the point-of-care group, reflecting the increased level of staff involvement required to deliver the intervention. Overall, the microcosting study showed that point-of-care panel assessment added £53.16 to the costs of ED management.

Table 36 shows the number of patients in each group receiving other interventions during their initial assessment, the number of interventions received and the cost per patient of providing these interventions. The small differences noted in the use of aspirin and clopidogrel between the two groups did not produce a significant difference in the cost per patient of medications. The standard-care group received more laboratory blood tests, leading to an excess cost of £9.42 per patient compared with the point-of-care group. The standard-care group also received more ECGs than the point-of-care group, although this resulted in only a small (£1.95) additional cost per patient. More patients in the point-of-care group received angiography or PCI, but the differences in mean cost per patient attributable to these two interventions were not statistically significant. A cost for ED stay for discharged patients was estimated on a pro rata basis using the ward cost for a day (£276.21) for patients who were discharged from the ED without any subsequent hospital admission. This cost was higher by £5.83 for patients discharged from the point-of-care group, who spent about a half-hour longer in the ED.

Table 37 shows resource use after the initial assessment. Items of resource use were costed only if they were chest pain related and could be matched with an appropriate unit cost. This explains why the numbers in this table for some items are slightly lower than corresponding data in the main outcomes analysis (Tables 16–21). Patients in the point-of-care group received more coronary care and intensive care, while patients in the standard-care group received more general inpatient care. None of these resulted in significant differences in mean cost, although the differences in the point estimates for costs relating to coronary care and intensive care were marked, with point-of-care panel assessment being associated with an additional £70.77 and £62.58 per patient, respectively. Patients in the point-of-care group received more outpatient follow-up. Standard care is associated with more nurse home visits and social work visits, but with relatively small differences in mean cost per patient.

Table 38 shows resource use and costs for community health service support use when missing values for each item are imputed by the mean cost for the corresponding treatment arm. Note that this approach usually underestimates the variability in the data so that p-values are slightly underestimated.

Table 39 shows the mean cost per patient by treatment group at each site and across all sites. Inferential analyses were conducted on imputed data; totals based on complete case means are included for comparison. Across all sites the mean cost per patient was higher in the point-of-care group by £211.22 in the imputed analysis (p = 0.056). Mean differences at individual sites ranged from £214.49 less in the point-of-care group at Leeds to £646.57 more in the point-of-care group at Edinburgh, with only the difference at Edinburgh being statistically significant (p = 0.025). An ANOVA test across centres (based on permuted test summed across the imputed data sets) yielded a p-value of 0.0803.

Tables 40 and 41 show the mean EQ-5D scores at 30 and 90 days using complete cases only (i.e. only those with data at both time points) and with imputation of missing values. There was no significant difference at either time point. Imputation did not markedly change the mean scores in either group at either time point.

Table 42 shows the total QALYs accrued (the difference) over 3 months at each site for both treatment groups. Data are reported assuming that EQ-5D was zero at baseline, although means for any baseline score between 0 and 1 can be estimated by adding a constant k/24, where k is the baseline EQ-5D score of interest. As 3 months is approximately one-quarter of a year, the maximum possible number of QALYs accrued is 0.25 (assuming EQ-5D was 1 at baseline). In practice, there is very little difference between treatment arms evident (a difference of 0.00273 is equivalent to 1 day), and there were no significant differences in QALYs accrued at any site or across all sites.

The cost-effectiveness plane based on the resampled cost-effectiveness data points for the point-of-care test strategy (Figure 6, with mean values shown by grey target) shows that, owing to the negative sign of the incremental QALYs, there is a high probability that the strategy is dominated (empirical probability 0.888). Conversely, the probability of the point-of-care test strategy being cost-effective at £20,000 per QALY is (understandably) very low (0.004).

FIGURE 6.

Cost-effectiveness plane for point-of-care treatment strategy.

Cost-effectiveness results are presented in Table 43. Empirical probabilities of a policy being dominated [i.e. has higher cost and is less effective than the index (standard care) policy], and of being cost-effective (i.e. having an incremental cost-effectiveness ratio no larger than £20,000 when the policy is more effective), are also presented.

Table 43 also shows the results of two analyses used to explore whether the findings were robust to two important assumptions. First, we explored whether a cheaper point-of-care test could be cost-effective by assuming that the test performance for a single troponin test, which at £6.17 incurs the lowest feasible point-of-care test cost, is the same as that of the trial strategy. In this scenario only costs would be affected, and test costs are reduced from £38.13 to £6.17, so that the mean cost difference would reduce to £179.26. Despite this, the probability of the point-of-care strategy being cost-effective remained low. Second, we explored whether excluding intensive care costs would alter the findings. Intensive care costs are only weakly related to the intervention and one patient in the point-of-care group had spent 52 days on intensive care, incurring substantial costs and increasing the variance of cost estimates. The results show that excluding intensive care costs reduced the mean cost per patient of point-of-care but did not markedly alter the probability that this strategy would be cost-effective.

These sensitivity analyses do not vary the results of the base case (i.e. trial) cost-effectiveness analysis, which might be expected as the alternative scenarios affect only costs.

Figure 7 shows estimated 95% confidence ellipses, assuming bivariate normality, of incremental costs and QALYs, for each of the hospital sites. Note that the plots are based on smaller sample sizes than the overall result and this is reflected in the higher variability. The plots generally reflect the overall result, and highlight, for example, the greater incremental cost variability of Edinburgh (where the patient with the long intensive care stay was recruited) and the possibility that, for Derriford or Leeds, the strategy may have a reasonable probability of being cost-effective.

FIGURE 7.

Cost-effectiveness plane for point-of-care treatment strategy, by site.

Decision-analytic modelling

The parameters used in the model are shown in Appendix 4. The probabilistic analyses based on the decision-analytic model yielded a mean difference in costs (point of care minus standard care) of £169 (empirical 95% CI –£1229 to £1658) and in QALYs of –0.002 (95% CI –0.108 to 0.104), which correspond reasonably with the trial-based results, particularly in the light of the simplifications outlined above (see Methods). The estimated probability that point-of-care panel assessment is cost-effective for a willingness to pay of £20,000 per QALY under these circumstances was 0.223. The model extrapolated to lifetime survival resulted in a mean cost difference of £329 (95% CI –£1312 to £2028) and corresponding value for QALYs of –0.087 (95% CI –12.3 to 12.2) accrued over a lifetime. The probability of cost-effectiveness of point-of-care panel assessment based on lifetime survival was estimated at 0.336.

Evaluation of GRACE and TIMI scores

Table 44 shows the number and proportion of patients with and without a major adverse event up to 1 month in each quintile of the GRACE score and each category of the TIMI score. Table 45 shows the same data for major adverse events up to 3 months. The proportion with a major adverse event at either 1 month or 3 months increases progressively with increasing quintiles of GRACE score regardless of the method used to incorporate troponin or the method used to analyse the data. The relationship between TIMI score and proportion with a major adverse event is not as clear, especially at 1 month. Patients with a TIMI score of 2 had the highest rate of major adverse events by 1 month.

Table 46 reports the area under the receiver-operator characteristic curve for GRACE and TIMI scores for predicting major adverse events up to 1 month and 3 months, and Figure 8 shows the curves. As before, these are presented for both methods of incorporating the troponin measurement and both methods of analysing the data. The method of analysis made little difference but the use of any troponin measurement resulted in consistently higher discriminant value than the use of the first troponin only. Both scores tended to predict events slightly better over the longer follow-up period. Overall, GRACE had better discriminant value for predicting adverse events than TIMI, although neither score predicted events with the kind of accuracy required to be clinically useful (i.e. with an area under the receiver-operator characteristic curve greater than 0.8). The area under the receiver-operator characteristic curve for age alone as a predictor was 0.66 for major adverse events up to 1 month and 0.69 for events up to 3 months. Hence, knowing the TIMI score appears to be no better than knowing the patient’s age for predicting subsequent major adverse events in this cohort.

FIGURE 8.

Receiver-operator characteristic curves for GRACE and TIMI scores. (a) TIMI score and 30-day major adverse events (MAEs). (b) GRACE score and 30-day MAEs. (c) TIMI score and 90-day MAEs. (d) GRACE score and 90-day MAEs.

Effectiveness of point-of-care panel assessment

Point-of-care panel assessment with CK-MB (mass), myoglobin and troponin at baseline and 90 minutes increased the proportion of patients with acute undifferentiated chest pain successfully discharged home after ED assessment (defined as having left the hospital or with a decision to discharge by 4 hours after attendance and no major adverse event over the following 3 months). The difference in the proportion of patients in hospital persisted until 24 hours after attendance and resulted in a shorter median length of initial hospital stay for patients in the point-of-care group and a greater proportion of the point-of care group having no inpatient days over the 3 months of the trial. However, it did not lead to any difference in the mean length of initial hospital stay or any difference in the mean number of inpatient days over the 3-month follow-up. This appeared to be due to a small number of patients with very long hospital stays and/or large number of inpatient-days that increased the variance for these variables and which were over-represented in the point-of-care group.

These findings suggest that the benefit of point-of-care panel assessment may depend upon one’s perspective. The proportion successfully discharged and the median length of stay describe the experience of the typical patient. Patients and clinicians may perceive benefit from point-of-care panel assessment, reducing the proportion of patients admitted and the duration of hospital stay for a typical patient. Hospital bed use is best estimated from the mean length of stay and mean number of inpatient days. When these are multiplied by the number of patients they produce an estimate of the total bed-days used by the patient cohort. Point-of-care panel assessment may reduce patient turnover (i.e. the number of patients being admitted and using hospital beds), but our data suggest that it will not significantly reduce hospital bed use.

Point-of-care panel assessment was associated with increased use of coronary care and administration of clopidogrel, but reduced use of aspirin. The last two findings appear to match each other, with 5% more patients in the point-of-care group receiving clopidogrel and 5% fewer receiving aspirin. These findings suggest a potential benefit from point-of-care panel assessment but one that is difficult to extrapolate into measures that are meaningful to the patient, such as improved survival or quality of life. They do, however, incur additional costs and are considered in the economic analysis.

Point-of-care panel assessment was not associated with any significant differences in investigations, cardiac interventions, ED attendances and subsequent hospital admissions over the following 3 months, although the non-significant differences in these areas of resource use contributed to the economic analysis. Point-of-care panel assessment was associated with a borderline significant increase in chest pain-related outpatient attendances, but this was one of many hypothesis tests and could be a chance finding.

From the patient’s perspective, point-of-care panel assessment was associated with no significant differences in health utility at 1 or 3 months, or time taken off work. It was associated with higher overall satisfaction with care, with more patients in the point-of-care group assessing their care as being excellent or very good, and fewer assessing it as good, fair or poor. Point-of-care panel assessment received a more favourable rating on 2 of the 10 dimensions of satisfaction measured: the urgency of assessment and the personal interest shown in the patient and his or her medical problems. There were no significant differences between the study groups on the other dimensions.

These findings suggest a modest patient preference for point-of-care panel assessment. However, patients were not blinded and would have been aware of whether they were receiving the experimental intervention or standard care. Their preference for point-of-care panel assessment may therefore reflect a positive response to receiving the ‘new’ intervention in a trial rather than a specific preference for the intervention.

Major adverse event rates were low, with only 62 out of 2243 patients (3%) having any major adverse event over 3 months, including only eight deaths and 10 non-fatal MIs. There were no significant differences in overall major adverse event rates or the rates of any individual event. CIs for adverse event rates were wide and the study was only originally powered to detect a relatively high 2% absolute increase in major adverse event rates.

We looked in detail at the major adverse events occurring in patients who were discharged after ED assessment, as these are the patients in whom a causal link between initial diagnostic assessment and major adverse event might be strongest. For example, if a patient was discharged after ED assessment had failed to identify MI then this could increase the risk of adverse outcome. Four patients in the point-of-care arm and one in the standard-care arm suffered adverse events after ED discharge. In all but one case the adverse event occurred more than 1 month after initial assessment. This relatively long time between intervention and outcome makes it difficult to assert a causal link. Furthermore, all five patients presented relatively late after their pain and received biomarker testing after a relatively long delay from worst symptoms. It is therefore very unlikely that any of these patients had a missed diagnosis of MI at their initial presentation.

Cost-effectiveness of point-of-care panel assessment

Overall, the economic analysis showed that point-of-care panel assessment was unlikely to be cost-effective compared with standard care. The main analysis produced a 0.888 probability that standard care dominated point-of-care panel assessment (i.e. was cheaper and more effective). The cost-effectiveness analysis was planned as an analysis of the incremental cost per QALY of point-of-care panel assessment. However, the main difference between the strategies and the main driver of cost-effectiveness was cost. The difference in QALYs between the two groups was small (0.003 QALYs) and non-significant (p = 0.250). The difference in costs between the two groups was £211 and bordered on being significant (p = 0.056). Hence, it appears that point-of-care panel assessment may increase costs despite reducing hospital admissions and inpatient days.

There appear to be a number of reasons for this. First, point-of-care panel assessment is associated with an additional £53 per patient for initial ED assessment. Most of this (£38) is related to the point-of-care machine, but additional staff time to undertake the tests also adds to the costs. It is possible that using the machine for other ED patients (thus reducing the amount charged for equipment) or using a simple protocol (such as a single troponin measurement) could reduce the additional costs associated with point-of-care panel assessment. We undertook a sensitivity analysis assuming that the machine costs were only £6.17 per patient instead of £38.13, but this did not markedly change the overall result, with there still being a 0.869 probability that standard care was dominant.

Point-of-care panel assessment was also associated with markedly higher mean costs for cardiac interventions, coronary care and intensive care. These failed to compensate for a £31 per patient saving from reduced inpatient days. The cost differences were not statistically significant but still added to the overall cost difference between point-of-care panel assessment and standard care. It could be argued that intensive care admissions are unlikely to be related to the initial diagnostic process. Indeed, one patient who suffered serious complications after CABG accounted for most of the intensive care days. We therefore undertook a sensitivity analysis with intensive care costs excluded. Although the difference in costs was reduced from £211 to £154, the probability of standard care being dominant remained high at 0.876.

The finding that there is a very low probability of point-of-care panel assessment being cost-effective is perhaps surprising given that the trial showed it was associated with a statistically significant improvement in its primary outcome measure, the proportion of patients successfully discharged home. However, it appears that earlier diagnosis may have an impact on other clinical decisions in addition to discharge home from the ED. In particular, admissions to coronary care and cardiac interventions were more frequent in the point-of-care group. One possible explanation for the increase in coronary care costs, for example, is that early recognition of troponin elevation may prompt admission to the coronary care unit, whereas an elevated 12-hour troponin after admission to a general ward is unlikely to prompt transfer to coronary care.

Variation between the participating sites

The analysis of the primary outcome and cost-effectiveness analysis showed marked differences between the participating sites. As described in the primary efficacy results section (see Chapter 3), these differences are probably due to differences in the facilities available for managing patients with chest pain and the standard care processes normally undertaken at the hospital. All hospitals, except Leicester, showed an effect from point-of-care panel assessment upon the primary outcome. The most marked effect occurred at hospitals in which chest pain patients were subject to a target of spending no more than 4 hours in the ED and standard care involved a 12-hour troponin. In these hospitals there was a significant difference in the primary outcome, which persisted until about 24 hours after attendance. At Leeds, where all patients with chest pain were cared for on a clinical decision unit that was not subject to a 4-hour target, the effect of point-of-care panel assessment was delayed until around 6 hours after attendance. At Derriford, where the standard care guidelines recommended a 6-hour troponin, the effect of point-of-care panel assessment was short-lived and not noticeable after 8 hours from attendance. The exception of Leicester is probably explained by trial factors. The Leicester Research Office stipulated that only doctors who had received full training in GCP could undertake trial recruitment. This meant that the research nurses were overwhelmingly responsible for recruiting patients and producing point-of-care test results. This probably meant that the clinical decision-makers (doctors) were not engaged in the trial and did not act on the test results.

The differences between the sites in the primary outcome are not reflected in differences in mean costs per patient. This is perhaps unsurprising as the overall analysis showed that the increase in the proportion successfully discharged with point-of-care panel assessment did not translate into cost savings. Mean costs were higher in the point-of-care arm at five hospitals and lower at one. It is possible that the overall conclusion that point-of-care panel assessment is unlikely to be cost-effective does not apply at some of the sites, such as Leeds and Derriford. However, we should be careful about drawing conclusions from individual site cost data because cost variances are large and most of the cost differences at individual sites were not statistically significant. The apparently dramatically higher costs associated with point-of-care panel assessment at Edinburgh, in particular, can be at least in part accounted for by two patients in the point-of-care arm with very high costs due to prolonged hospital stays.

However, in general terms it is possible that inpatient treatment patterns differ between hospitals and that in some settings point-of-care panel assessment in the ED does not lead to higher costs of inpatient care. If this were the case then point-of-care would have the potential to be cost-effective, as is suggested with the confidence ellipses in Figure 7. Previous work on chest pain units has tried to identify the combination of costs and admission rates that would make such initiatives cost-effective. 53 Similar studies could be worthwhile here.

Variation over time

The effect of point-of-care panel assessment could change over time as a result of clinicians becoming familiar with the technology and changing their decision-making in response to the point-of-care results. In particular, the change in the recommended decision-making threshold for point-of-care troponin could have changed the proportion of patients successfully discharged after point-of-care assessment. Conversely, standard care could change over time and start to mimic point-of-care panel assessment, thus leading to contamination of the control group and a bias towards no effect.

We analysed the primary outcome and deviations from the recommended point-of-care protocol at quarterly intervals to explore whether the effect of point-of-care panel assessment changed over time. Only two patients in the control group received the combination of tests used in the point-of-care machine (CK-MB, myoglobin and troponin) either from the machine itself or from the hospital laboratory, so we did not examine deviations from standard care over time. Deviations from the recommended point-of-care protocol remained relatively constant over time, so there was no evidence that developing experience with the technology eliminated these problems. However, the proportion successfully discharged in the point-of-care arm decreased during the middle third of the trial and increased in the last third, while the proportion successfully discharged in the standard-care arm steadily fell over the trial. The first finding may be explained by the amendment to the troponin threshold in the recommended point-of-care protocol that advised discharge instead of admission for patients with a non-rising troponin in the range 0.03–0.07 µg/l. The second finding may be due to clinical staff becoming progressively more cautious during the trial. However, we must be careful not to overinterpret these findings as they are based on relatively small numbers.

Other studies of point-of-care cardiac markers

The RATPAC trial is the first randomised trial to evaluate whether a panel of point-of-care cardiac markers designed to rule out MI in ED patients with acute chest pain can change clinical decision-making and lead to changes in patient care. Previous studies22,23 have evaluated the point-of-care panel and laboratory testing in the same cohort to compare the times taken to receive a test result and have shown that turnaround times are shorter with point-of-care testing, but few studies have compared practice in cohorts with or without point-of-care testing. Ng et al. 24 compared management with the point-of-care panel with previous practice without the test and found a 40% reduction in coronary care unit admissions. This contrasts with RATPAC, in which we found that point-of-care panel assessment was associated with increased coronary care use. There are two potential explanations for this discrepancy. First, the study by Ng et al. 24 may have been confounded by changes in practice occurring over time that would have been controlled for in RATPAC by the randomised design. Second, the study by Ng et al. 24 was undertaken in a health-care system and patient cohort with much higher coronary care use, so there was much greater potential to reduce admissions. This illustrates the difficulties in extrapolating findings relating to health service use between diverse settings.

Other studies, including randomised trials, have evaluated point-of-care testing by directly comparing use of a point-of-care troponin assay with use of a laboratory assay, rather than evaluating a point-of-care protocol as we did in RATPAC. As outlined in Chapter 1, such studies have produced mixed results, with Collinson et al. 12 finding reduced length of coronary care unit stay and overall hospital stay, Renaud et al. 13 reporting reduced time to anti-ischaemic therapy and physician notification of results, but no change in ED length of stay or patient outcomes, and Ryan et al. 14 finding that the effect of point-of-care testing varied between settings, with length of stay in the ED being increased in one hospital and decreased in another. These studies are not directly comparable with RATPAC because they compare only the process of using a point-of-care versus a laboratory assay, whereas RATPAC evaluated a point-of-care protocol involving a specific panel of tests designed to rule out MI in patients with acute chest pain. However, the inconsistencies noted between the findings of these studies and the variability between the sites involved in RATPAC suggest that the effects of point-of-care testing are inconsistent, whether one is simply examining the effect of point-of-care testing itself or point-of-care testing is being used to deliver a specific protocol.

Strengths and weaknesses of the study

As a randomised trial, RATPAC has the strength of providing an unbiased comparison between the point-of-care protocol and standard care. Allocation of treatment group was concealed until the patient was irreversibly entered into the trial and the low rate of withdrawals before the primary outcome, coupled with analysis being as randomised, meant that the benefits of randomisation were preserved. The trial was pragmatic, in that the only difference between study groups was the availability of point-of-care panel assessment, and involved six diverse hospitals, so findings from the study can be generalised across the NHS. We measured a range of important outcomes and undertook cost-effectiveness analysis using data from real patients, thus avoiding having to make assumptions about how patients would be managed. The economic analysis used standard methods recommended by NICE41 to allow comparison of the cost-effectiveness of point-of-care panel assessment with other competing demands for NHS resources.

The study also had a number of limitations that should be taken into account when interpreting the findings. We obviously could not blind patients or carers to whether they had been randomised to intervention or control. In some respects this does not matter for a pragmatic trial where we are simply interested in whether an intervention works rather than how or why it works, but it may be important in interpreting patient-reported outcomes. Patients who are randomised to the standard-care group may feel that they have had a potentially valuable intervention withheld and may respond negatively when asked for their opinion. Response rates for the patient questionnaire were, as anticipated, around 70%, so there is the potential for responder bias. It is notable in this respect that response rates were lower in the standard-care group, perhaps reflecting a reduced level of engagement in those who did not receive the study intervention.

Point-of-care testing, and troponin assays in particular, is a rapidly developing area of technology. This created inevitable difficulties in selecting the appropriate machine and assay to use, and in providing guidance for the interpretation of results. We believe that our decision to change from the Biosite troponin assay to the more sensitive Stratus CS Analyser has been vindicated by recent publications7,8 that show the value of high-sensitivity troponin assays and suggest that these will become standard. However, working with a relatively new assay created problems with interpreting low positive results and led to our having to change the recommended protocol during the trial. These issues are not limited to the RATPAC trial and the introduction of highly sensitive troponin assays into practice is likely to be complicated by difficulties deciding how to manage patients with levels close to, or just below, the diagnostic threshold. Thus, it can be argued that the difficulties encountered in the trial in terms of machine malfunction and assay interpretation merely reflect the difficulties encountered in normal practice. However, an inevitable side effect of undertaking evaluation of a developing technology may be a degree of bias against the new technology, in that clinicians may take time to develop familiarity and the technology may ultimately perform better when it is in widespread use.

The trial was terminated before the target of 3130 patients was reached so it did not have the originally planned power. For the primary outcome this was compensated by the actual control admission rate being higher than the original estimate of the baseline admission rate. The trial was originally powered to detect a doubling of the major adverse event rate from 2% to 4%. In the event, at the time it was terminated it only had conditional power of < 10% to detect such a difference. It is therefore possible that point-of-care panel assessment may achieve increased successful discharges at the expense of an increase in major adverse events that the trial was not powered to detect. However, detailed examination of the adverse events suggests that there is unlikely to be a strong association between point-of-care panel assessment and adverse events. Most of the events occurred in patients who had been admitted after ED assessment. Of the five cases that occurred after discharge, only one occurred within the time span of a typical hospital admission. In general, the low adverse event rate in this study, especially deaths and non-fatal MI, should provide some reassurance that in this selected low-risk population the risk of adverse outcome shortly after attendance is very low.

The trial was deliberately designed as a pragmatic evaluation of a specific point-of-care protocol. The intervention was therefore a composite of the point-of-care technology and the cardiac marker panel. This type of pragmatic evaluation carries the disadvantage that we cannot clearly isolate which element of the intervention is effective (or ineffective). It is thus possible that the effects recorded in this study were due to the biomarker panel, rather than point-of-care technology, and could have been achieved using laboratory assays if turnaround times were sufficiently rapid.

The trial was undertaken at six diverse NHS hospitals and all participating centres contributed substantially to recruitment. The variation in the participating hospitals is likely to reflect variation in practice across the NHS. 11 This variation means that although we may conclude that out findings apply in general across the NHS it is possible that point-of-care panel assessment may be more or less effective or cost-effective in specific hospitals. The methods used for cost-effectiveness in particular are appropriate for NHS-wide decision-making but may not reflect the costs, effects and priorities of specific individual hospitals.

Imputation methods were used to estimate missing values on the basis of observed values. Recent methods favour the approach of generating multiple imputed data sets (multiple imputation). The main potential limitation of multiple imputation arises from the implicit assumption that the observed data are representative of those that are missing. In other words, these data must be missing at random. With regard to the imputation of staff costs using the microcosting study data, it is likely that this assumption holds, as the sample on which this study was conducted was chosen at random. In the case of the imputation of community health service costs, it may be that the assumption is more tenuous. For example, it may be that patients who are well are less likely to respond to the follow-up survey. Imputation may overestimate costs in these cases. On the other hand, severely ill patients may also have higher rates of non-response. In this case, imputation would probably underestimate costs. However, given the follow-up procedures of the RATPAC study, it is less likely that data from such patients were not recorded.

Other limitations of imputation methods include small numbers and a proportion of missing values higher than some percentage, typically set at 30%. In the case of the staff costs imputed from the microcosting data, this assumption is violated, as the missing data rate is about 90%. However, the method of imputation used in this case was one of choosing the value from the patient with complete data which matches closest that of the incomplete case, which is less prone to result in a biased estimate, although it may result in underestimation of the variability.

Implications for practice

The RATPAC trial has shown that point-of-care panel assessment with CK-MB (mass), myoglobin and troponin leads to increased successful discharge after ED assessment, but probably costs more than standard care, and is unlikely to be considered a cost-effective use of NHS resources. This means that from an NHS perspective it would not be appropriate to recommend widespread adoption of the point-of-care panel unless specific reasons can be identified as to why it might be worth paying more to reduce the proportion of patients requiring hospital admission. From the perspective of an individual hospital, the decision of whether or not to use the point-of-care panel will depend upon the specific needs of the ED and acute specialties, and whether it may be considered worthwhile to pay the potential additional costs of point-of-care panel assessment to achieve a specific service outcome, such as discharging more patients from the ED within the 4-hour time frame.

It is important to recognise that RATPAC evaluated a specific point-of-care combination for a specific purpose and that care should be taken not to extrapolate findings beyond this specific role. The RATPAC trial results suggest that a rule-out diagnostic testing protocol, which would be expected to lead to a substantial reduction in admissions, may, in practice, have only a modest effect and that a reduction in hospital admissions may not lead to health service cost savings. Although these findings may provide some useful insights when considering the wider role of point-of-care testing, they should not be extrapolated to all point-of-care tests in all settings.

Other findings from the RATPAC trial may have implications for practice and future developments in chest pain care. In particular, the very low rate of adverse events and the observation that most occur after the typical time span of hospital admission suggest that the value of hospital admission for these selected low-risk patients should be brought into question. Most of the investigations and interventions that may benefit patients can be delivered without hospital admission. If the purpose of hospital admission is simply to monitor patients thought to be at risk of adverse outcome then the data from this study suggest that admission for this purpose is unlikely to be worthwhile.

Implications for future research

The very low adverse event rates suggest that further research is required to evaluate the benefit of hospital admission and explore ways of managing acute chest pain that do not require hospital admission. RATPAC has shown that rapid diagnostic testing can make a modest impact upon hospital admissions. It may be that a substantial reduction in admissions can be achieved only through a fundamental restructuring of the way in which acute chest pain is assessed. This would involve moving from a ‘rule-out’ approach, where the aim is to identify low-risk patients who can be discharged home (with the rest being admitted), to a ‘rule-in’ approach, where the aim is to identify high-risk patients for hospital admission and treatment (with the rest being discharged home).

The RATPAC proposal recognised that this is a rapidly developing field and made provisions for future studies accordingly. We have stored blood samples from all participants in the point-of-care group and will be using these samples to test new and emerging assays. This should be coupled with economic evaluation and modelling to explore the potential impact of new early biomarkers with different levels of sensitivity and specificity. Development of biomarkers has tended to accept loss of specificity in the attempt to optimise sensitivity. However, poor specificity has potentially important economic consequences that need to be explored and better understood.

Analysis of the individual sites involved in the RATPAC trial showed marked variation in standard practice and in the effect of the intervention upon the primary outcome. This suggests that point-of-care panel assessment could be more effective or cost-effective in specific settings. Further research should be undertaken to assess in what circumstances point-of-care panel assessment could be effective and cost-effective.

Analysis of GRACE and TIMI scores

We undertook this analysis as a secondary objective of the study unrelated to the primary objectives of evaluating point-of-care panel assessment. The GRACE and TIMI scores have been developed and validated on patients with diagnosed ACS to predict adverse events, usually up to 1 month. Our analysis suggests that they have limited prognostic value in low-risk patients with undifferentiated chest pain, with TIMI performing no better than patient age as a predictor and GRACE performing only slightly better (and also being more complex to calculate). This is perhaps not surprising since both were developed in populations with diagnosed ACS and a high risk of major adverse events. Developing a new clinical risk score in the low-risk population with undifferentiated chest pain may produce an instrument with greater prognostic value. Alternatively, it could be argued that the risk of adverse outcome in this population is so low that attempts to discriminate between those at very low risk and those at moderately low risk are unlikely to be successful.

Contributions of authors

Steve Goodacre (Professor of Emergency Medicine, Health Services Research, University of Sheffield) Chief Investigator, conception and design of the study, analysis and interpretation of the data, writing the report and final approval of the version to be published.

Liz Cross (Research Associate, Health Services Research, University of Sheffield) Project management, analysis and interpretation of the data, drafting and revising the report and final approval of the version to be published.

Mike Bradburn (Senior Medical Statistician, Clinical Trials Research Unit, University of Sheffield) Trial statistician, analysis and interpretation of the data, writing the report and final approval of the version to be published.

Patrick Fitzgerald (Research Fellow, Health Economics and Decision Science, University of Sheffield) Health economics, analysis and interpretation of the data, writing the report and final approval of the version to be published.

Alasdair Gray (Consultant in Emergency Medicine, Royal Infirmary of Edinburgh) Conception and design of the study, analysis and interpretation of the data, revising the report and final approval of the version to be published.

Alistair Hall (Professor of Clinical Cardiology, University of Leeds) Conception and design of the study, analysis and interpretation of the data, revising the report and final approval of the version to be published.

Paul Collinson (Consultant in Chemical Pathology, St George’s Hospital, London) Conception and design of the study, analysis and interpretation of the data, drafting and revising the report and final approval of the version to be published.

List of contributors

Publications

Goodacre SW, Bradburn M, Cross E, Collinson PO, Gray A, Hall AS on behalf of the RATPAC research team. The RATPAC trial (Randomised Assessment of Treatment using Panel Assay of Cardiac markers): a randomised controlled trial of point-of-care cardiac markers in the emergency department. Heart 2011;97:190–6.

Bradburn M, Goodacre S, Fitzgerald P, Coats TJ, Gray AJ, Hassan TB, et al. Inter-hospital variation in the RATPAC trial (Randomised Assessment of Treatment using Panel Assay of Cardiac markers). Emerg Med J 2011. In press.

Fitzgerald P, Goodacre SW, Cross E, Dixon S. Cost-effectiveness of point-of-care biomarker assessment for suspected myocardial infarction: the RATPAC trial (Randomised Assessment of Treatment using Panel Assay of Cardiac markers). Acad Emerg Med 2011;18:488–95.

Goodacre SW, Bradburn M, Mohamed A, Gray A. Evaluation of Global Registry of Acute Cardiac Events and Thrombolysis in Myocardial Infarction scores in patients with suspected acute coronary syndrome. Am J Emerg Med 2011. In press. doi: 10.1016/j.ajem.2010.09.013.

Disclaimers

The views expressed in this publication are those of the authors and not necessarily those of the HTA programme or the Department of Health.

Step 1

• Collect an Investigation Pack from near RATPAC machine (Stratus CS Analyser).

• Contains two green top tubes, two yellow top tubes, four small clear top tubes, two rotors, two cannulae, one Lab Form.

Note: the small tubes are for the labs, please do not remove from bag.

Step 2 – sample 1

• Take point-of-care (PoC) blood samples (one yellow top and one green top) and routine bloods (normal tubes).

• Label the yellow top sample with bar code and attach corresponding bar code label to Lab Form in pack.

• Label green top sample with same bar code.

• Scan bar coded green top tube in Stratus.

• Enter patient study number (given by randomisation system) into Stratus and also add to Lab Form.

• Please process green top sample in Stratus within 20 minutes.

• Return yellow top sample and Lab Form to pack and store in fridge.

Step 3

• Collect three reagent strips from fridge: troponin I (green), myoglobin (red), creatine kinase MB (CK-MB) (blue).

• Invert green top tube gently several times (do not shake).

• Process green top tube in Stratus as per instructions.

• Results produced in 20 minutes.

• After processing remove cannula unit and tube from the Stratus and discard appropriately.

Step 4 – interpretation of sample 1 results

• Fix results slip in patient notes (important):

  • Troponin I ≥ 0.03 µg/l = positive result. Admit to hospital.

  • If no detectable troponin or < 0.03 µg/l = negative result. Finish assessing patient for 90 minutes then take sample 2.

Step 5 – sample 2

• At least 90 minutes from first sample, take second blood samples (one yellow top and one green top).

• Label as before with corresponding bar codes on tubes and Lab Form.

• Return yellow top to pack and store in fridge.

• Process green top as before in Stratus.

• Results produced in 20 minutes – keep printout with notes again.

Step 6

• Remove cannula unit and tube from the Stratus and discard appropriately.

• Add corresponding small labels (without bar code) to two small tubes for each yellow top sample.

• Send tubes in pack with Lab Form to Clinical Biochemistry Lab (within 12 hours).

Note: Ignore remaining labels.

Step 7 – interpretation of sample results

• Calculate CK-MB (mass) gradient = second CK-MB (mass) – first CK-MB (mass), for example 4.52 – 3.11 = 1.41 (negative).

• Calculate percentage increase in myoglobin (m) = 100 × [(second m – first m)/first m], for example 100 × [(40 – 25)/25] = 60% (positive).

  1. If any of the following, admit to hospital (positive cardiac markers):

     1. – troponin I ≥ 0.03 µg/l

     2. – CK-MB (mass) > 5 µg/l on both samples

     3. – CK-MB (mass) gradient on second sample 1.6 µg/l higher than first sample

     4. – myoglobin on second sample 25% higher than first sample.

  2. If cardiac markers negative, consider for discharge home.

Supplemental tables

Project title: The RATPAC trial (HTA 06/302/19) (Randomised Assessment of Treatment using Panel Assay of Cardiac markers)

A randomised controlled trial (RCT) of point-of-care cardiac markers in the emergency department.

Planned investigation

Project timetable and milestones

The project will start on 1 April 2007. Months 1–6 will involve staff recruitment and local ethics and research governance; months 7–18 will involve patient recruitment; and months 19–24 will involve completion of follow-up, data analysis, writing up and dissemination. The project will be completed by 31 March 2009. The GANTT below outlines the key milestones and shows when project staff will be employed.