The effectiveness and cost-effectiveness of biomarkers for the prioritisation of patients awaiting coronary revascularisation: a systematic review and decision model

Article history

The research reported in this issue of the journal was commissioned by the HTA programme as project number 05/40/04. The contractual start date was in November 2006. The draft report began editorial review in August 2008 and was accepted for publication in March 2009. 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

© 2010 Queen’s Printer and Controller of HMSO. 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.

2010 Queen’s Printer and Controller of HMSO

Biomarkers

There is intense interest in the measurement and evaluation of biomarkers in order to better target clinical care for many diseases. 1,2 The hope is that biomarkers will provide new information about the patient and his or her disease condition, which will help optimise the type, amount or timing of subsequent intervention. The development, evaluation and use of biomarkers represents a major technology in health care, with growing investment from large companies, including Dade Behring, Roche Diagnostics, Abbott Diagnostics, Acon International and Beckman Instruments. Within the last decade there have been rapid increases in the number of reports of individual biomarkers and their incorporation into prognostic risk scores. 3 This interest has in part been stimulated by the high cost and long timescales involved in the development of new therapeutic drugs and devices. The concept of ‘personalised medicine’ seeks to exploit information from biomarkers in order to maximise the probability of benefit and minimise harms for a given treatment.

A biomarker has been defined as ‘a characteristic that is objectively measured and evaluated as an indicator of normal biologic processes, pathogenic processes, or pharmacologic responses to a therapeutic intervention.’4 Thus the term ‘biomarker’ encompasses a range of measures; biomarkers that are circulating – that is, assessed from a blood sample – have been the focus of most interest in prognosis research. Compared with imaging biomarkers (e.g. computed tomography or magnetic resonance imaging), circulating biomarkers have the advantages of being relatively low cost, low burden (patients expect to have blood taken) and no risk (compared with the radiation exposure of computed tomography). Clinicians are used to interpreting and acting on a single numerical value from a blood test (e.g. low haemoglobin defines anaemia), and are increasingly using numerical values derived from scores. In the setting of coronary artery disease, more than 100 measures (beyond those widely made in routine clinical practice) have been related to the risk of subsequent death or heart attack in one or more study. 1

Prognosis, outcomes and NHS quality initiatives

Clinicians are increasingly invited to scrutinise the outcomes of their care, in an effort to improve quality. Cardiac surgery in children,5 and subsequently in adults, has been the subject of high profile inquiries into the performance of individual units and clinicians. Under the Darzi review,6 from April 2010 all health-care providers working for the NHS will be legally obliged to publish ‘quality accounts’ on safety, patients’ experience and clinical outcomes, in the same way that they publish financial accounts. Indeed, the 2007 White Paper Trust, Assurance and Safety7 states that ‘recertification will be supported by information that shows how clinically effective each doctor’s treatment of his or her patients has been’ requiring ‘analysis of the outcomes of their treatment’.

Coronary artery bypass grafting and the NCEPOD Report

Coronary artery bypass grafting (CABG) remains the standard of care for patients with three-vessel or left main coronary artery disease, because the use of CABG, as compared with percutaneous coronary intervention (PCI), resulted in lower rates of the combined end point of major adverse cardiac or cerebrovascular events. 8 The 2008 National Confidential Enquiry into Patient Outcome and Death (NCEPOD) report on death after CABG9 found that ‘in the opinion of the advisors for 57/821 (7%) of cases there was a delay from referral to the first cardiothoracic review and in 33 of these patients outcome was adversely affected.’ One of the principal recommendations was to ‘use protocols for referrals. These protocols should be standardised nationally for patients who require coronary artery bypass graft surgery. The degree of urgency of referral should be emphasised within these protocols.’

Thus, cost-effective means of improving institutional performance are of considerable interest. There is an established culture of using risk prediction scores (the euroSCORE; European System for Cardiac Operative Risk Evaluation)10 for operative mortality after CABG, for the purposes of risk-adjusting comparisons between institutions and individual surgeons. This score does not use novel biomarkers and is not designed to assess event rates on the waiting list.

Stable angina

In women and men, coronary heart disease (CHD) is the most common cause of premature death in the UK and most Western countries,11 and is predicted to remain so at least until the next decade. Coronary disease commonly presents as angina pectoris, which is characterised by chest pain or discomfort typically on exertion and relieved by rest, in association with atherosclerotic narrowing of the coronary arteries (assessed at angiography). Between 1991 and 2003, while the incidence of heart attack declined rapidly (about 8% per year), the prevalence of angina pectoris diagnosed by a doctor based on five waves of Health Survey for England data showed no evidence of decline,12 suggesting that the relative importance of chronic symptomatic coronary disease may be increasing. The prevalence of angina in the UK is about 3–5% for women and men,13 suggesting that approximately 1.3 million people have symptoms (www.heartstats.org/datapage). The economic burden of angina is high, estimated at 1.3% of the NHS budget in the UK14 and costing $75 billion in 2000 in the USA. 15

Current use of circulating biomarkers in coronary disease

Several types of blood measurement are widely used in the management of coronary disease. These include markers of myocardial necrosis – evolving from aspartate transaminase in the 1950s, creatinine kinase (CK) in the 1960s, CK-MB in the 1970s, and troponins in the 1980s – which are used primarily as diagnostic tests with high negative and positive predictive value. This is one of the clearest examples in clinical medicine where marker measurement and urgent clinical decision-making are closely related. However, a range of other blood markers are routinely ‘taken’ among patients with coronary disease, but their use, if any, in clinical decision-making is less clear. For example, a measure of kidney function, the serum creatinine, has been estimated among people with suspected coronary disease for decades, but only in the last decade has its potential prognostic value been considered.

Proposed use of circulating biomarkers in stable angina

Circulating biomarkers have been recommended as potentially useful measures in the management of patients with stable angina. For example, the Centers for Disease Control/American Heart Association statement for health-care professionals recommended that, among patients with stable coronary disease, one biomarker, C-reactive protein (CRP), ‘may be useful as an independent prognostic marker’. 18 The costs and other characteristics of selected biomarkers are shown in Table 1. The European Society of Cardiology angina guidelines recommend (class IIb, strength of evidence B) measurement of CRP, lipoprotein a [Lp(a)], apoA-I (apolipoprotein A-I), apoB (apolipoprotein B), homocysteine, N-terminal brain natriuretic peptide (NT-proBNP), fibrinogen and interleukin 6 (IL-6). 19 Although there are no surveys on the variability between clinicians and centres in which biomarkers are measured, anecdotal evidence suggests that in NHS practice in 2008 most, if not all, of these eight biomarkers are not routinely evaluated among patients with angina. The biological mechanisms by which these markers may influence prognosis are varied, and span molecules with differing function, including markers of inflammation, lipids, hormones and vitamins. For the purposes of the evaluation of biomarkers, their biological functions are of secondary importance; the question is the extent to which they predict risk, not how.

Biomarkers and specific clinical decisions

Existing biomarker measurement recommendations, remarkably, do not specify which clinical decisions might be influenced in the light of the biomarker information. It is implicitly assumed that more information might lead to better clinical decision-making in general. Specifically, there are no professional body or government recommendations for the measurement of biomarkers in the invasive management of angina pectoris.

Revascularisation for angina

The goals of treatment for angina are to reduce mortality, lower the risk of major non-fatal events (heart attack and stroke) and to improve symptoms and quality of life. In both women and men, the diagnosis of angina is associated with markedly increased death rates from coronary disease compared with the general population: five-fold excesses among patients aged 45–55 years, and three-fold excesses among patients aged 65–75 years. 16 Coronary angiography – one of the most widely performed procedures in clinical medicine (annual numbers estimated at around 1 million in the USA, and 100,000 in the UK) – is the invasive X-ray used to diagnose coronary artery disease; without this test revascularisation cannot be considered. Among patients with angiographic luminal narrowings, coronary revascularisation is effective at relieving symptoms and improving quality of life, compared with medical management. Revascularisation with PCI with balloon angioplasty, with or without stenting, was initially proposed for patients with single- or double-vessel disease. CABG is a major surgical procedure carried out under general anaesthetic, in which the narrowings in the coronary arteries are bridged using vessels from the patient – leg veins (saphenous) or an artery from the inside of the chest wall (internal mammary). CABG is a higher cost procedure, which is associated with improvements in survival20 (unlike PCI), and tends to be associated with longer waiting times.

NHS waiting time initiatives for coronary artery bypass grafting

Coronary artery bypass grafting is carried out after an interval of days or weeks from the date the decision for surgery is made. In the 2003 report from the National Adult Cardiac Database there were about 25,000 CABG procedures carried out annually between 1997 and 2003,21 with no evidence of a decline in this number of procedures. There have been dramatic falls in the waiting time – defined as starting from the date of angiography to CABG – since the median waiting times of 214 days in 1994/5. 22 The implementation of the National Service Framework for CHD in 2000 led to declines in waiting time, and since March 2005 no NHS patient has waited longer than 3 months for CABG. 7 The most recent figures (August 2008) from the Department of Health suggest that about half the patients waiting for CABG have been waiting for between 1 and 3 months, and about half for up to 1 month. Previous policy was based on waiting from the time of angiography – which represents only one segment of patients’ waiting experience. The most recent policy focuses on the whole wait, from time of initial referral to receipt of definitive treatment, in this case CABG, with a ceiling of 18 weeks.

International comparisons in waiting times

Internationally and across different systems of health-care provision, waiting times for CABG have been the subject of targets set by politicians and by professional bodies. 23 Waiting times continue to vary within and between countries, with published comparisons between the USA, Sweden and Netherlands24 and other countries. 25 Recent Canadian guidelines state: ‘The target for bypass surgery in those with high-risk anatomy is 14 days; for all others, the target is six weeks … there is an ongoing need to continually reassess current risk stratification methods to limit adverse events in patients on waiting lists and assist clinicians in triaging patients for invasive therapies.’26

Events on waiting list

People with stable coronary disease are at increased risk of death or heart attack,16 compared with the general population. Being on a waiting list for CABG per se probably has no measurable impact on these event rates. 27 Patients awaiting CABG experience continued symptoms, and some, but not all, studies suggest longer waits are associated with more anxiety and disutility. 28 There is no evidence of any benefit in deferring surgery among patients with stable coronary disease without acute myocardial infarction history. Among patients in whom there is a history of recent acute myocardial infarction, the possible increased risk of early surgery may be balanced against the potential for improved remodelling, improved quality of life and decreased hospital stay costs. 29

Need for prioritising waiting lists

Irrespective of whether target waiting times for CABG are 14 days, 6 weeks or 3 months from the date of angiography, clinicians (and the administrative systems in which they work) are faced with deciding whether an individual patient merits listing for surgery sooner. That is, does a strategy of ordering the waiting interval, according to formal scores, improve clinical outcomes and, if so, is this strategy cost-effective? However the cost-effectiveness of any strategy may be hypothesised to be lower in countries with lower median waiting times.

Usual practice

Clinicians informally prioritise waiting lists. Without recourse to formal scores, published evidence suggests that time to invasive management of coronary disease is not random but, on average, is ordered at least according to urgent, semi-urgent and non-urgent categories. 30,31 But the rules for deciding which combination of simple clinical information would place a patient in one group or another are not explicit. Although enough information is routinely collected that would allow calculation of a formal urgency or risk score (including information on some circulating biomarkers), these formal scores are seldom derived in NHS practice.

Different formal prioritisation strategies

The dominant technology, which has been proposed as a means of improving on such implicit means of prioritising waiting lists, has been the use of ‘urgency’ or ‘acuity’ scores. These scores have been developed, and to some extent implemented,31 in Canada32 and New Zealand. 33 These urgency scores apply weightings to clinical covariates based on anatomical disease severity and symptom severity, both of which are predictors of mortality. 34 The principle is that higher risk patients should undergo an operation sooner. Biomarkers are not included in these urgency scores. Scores that predict the long-term risk of events among people with stable coronary disease have been developed,17 but are not widely used and were not developed among patients awaiting CABG.

Framework for evaluation

Conventionally, the effectiveness of different health-care technologies is rigorously evaluated in randomised controlled trials, in order to address confounding. There are no randomised trials comparing different prioritisation strategies for CABG in stable coronary disease, and these are unlikely ever to be performed. Thus, observational data with decision modelling has been demonstrated as a robust, evidence-based method of evaluation that can inform policy-making and clinical decision-making. Thus, observational studies are likely to be the main basis for estimating the effects of biomarkers in the context of prioritising patients on waiting lists. It is increasingly recognised that, to inform decisions about the effectiveness and cost-effectiveness of new technologies and health-care programmes, decision-analytic models provide a valuable framework. 35 These methods are now central to the National Institute for Health and Clinical Excellence’s (NICE’s) technology appraisal programme. 36 Decision analysis is a framework for supporting decisions rather than a source of data as provided by randomised trials and observational methods. To inform decisions, these methods facilitate the synthesis of available evidence and explicit assumptions and judgements about, for example, the duration of treatment effects. Importantly, decision analysis provides a means of quantifying the uncertainty in existing evidence and hence prioritising future research.

Scientific uncertainties addressed in this monograph

The following is not known:

• The quality of individual studies reporting biomarkers in the prognosis of stable coronary disease, and the potential for biasing meta-analytic estimates of the effects of biomarkers.

• The strength of effect (relative risks) and precision of these estimates [95% confidence intervals (CIs)] of five routinely assessed and eight novel biomarkers in the prediction of CHD, cardiovascular disease (CVD) and ACM events among people with stable coronary disease.

• The most appropriate structure and input parameters for a decision-analytic model to evaluate alternative prioritisation strategies.

• Are circulating biomarkers cost-effective at prioritising the clinical acuity (urgency) of patients awaiting CABG?

• What is the incremental cost-effectiveness ratio (ICER) of adding one novel biomarker, one routinely assessed biomarker or both to a risk score to prioritise patients awaiting CABG?

Aim

To determine the clinical effectiveness and cost-effectiveness of a range of formal strategies based on conventional clinical information and novel circulating biomarkers (singly or in combination) for prioritising patients with stable angina awaiting CABG.

Objectives

1. To estimate the prognostic value of circulating biomarkers in predicting events among patients with stable coronary disease. The prognostic value was determined using systematic review and meta-analytic approaches in order to estimate summary relative risks of effects on prognosis for biomarkers for which measurement in angina patients in NHS practice is widespread [e.g. estimated glomerular filtration rate (eGFR)], and is recommended but not routine (e.g. highly sensitive CRP).

2. To explore sources of uncertainty in the estimates of effect of the biomarkers which may influence the clinical effectiveness and cost-effectiveness estimates, specifically to assess the precision of estimates, the quality of individual studies, publication bias and other sources of heterogeneity.

3. To develop a decision-analytic model to compare alternative approaches to prioritisation in terms of cost-effectiveness using lifetime costs and quality adjusted life-years (QALYs). The strategies of interest in relation to circulating biomarkers were:

   1. – single routine biomarker (e.g. eGFR)

   2. – single novel biomarker (e.g. CRP)

   3. – combination of routine and novel biomarkers (e.g. eGFR + CRP),

4. added to a risk equation, as compared with other alternative approaches which are relevant comparators. These alternatives include both current practice and the more formal use of alternative methods or prioritisation based on conventional clinical information and urgency scores (e.g. New Zealand and Ontario algorithms).

5. To estimate the contemporary rates of events (death, non-fatal myocardial infarction and non-fatal stroke) among patients with chronic stable angina on a waiting list for CABG, and the extent to which conventional clinical information predicts these event rates, and thereby offer a means of prioritisation.

6. To estimate the costs of gathering prognostic information, the costs of alternative management interventions, the long-term outcomes (fatal, non-fatal and health-related quality of life), and the efficacy of revascularisation with respect to patients’ baseline risks and over time.

7. To populate the decision-analytic model using robust estimates of the full range of relevant inputs required to estimate mean lifetime costs, QALYs and overall cost-effectiveness of the alternative strategies in the context of a representative cohort of patients on a waiting list for CABG.

8. To undertake sensitivity analysis to examine the robustness of the results of the decision-analytic model to alternative input values and assumptions in relation to potentially important drivers of cost-effectiveness.

Overview of decision problem

This section summarises the key elements of the decision problem considered.

• Patient population Patients with stable coronary artery disease who have been placed on the waiting list for CABG. It is assumed that all patients will undergo CABG within 3 months of being placed on the waiting list but, within this period, prioritisation between patients is possible. Alternative scenarios representing shorter target waiting times for CABG of 15 and 40 days are also considered.

• Technology of interest Standard clinical information together with circulating biomarkers singly and in combination as a basis of prioritisation. This is represented in the form of a risk score combining information relating to clinical parameters and circulating biomarkers.

• Comparators Alternative forms of prioritisation: no formal prioritisation (routine clinical practice), urgency scores (New Zealand and Ontario algorithms), a formal risk score based on standard clinical information (without biomarkers).

• Basis of evaluation To establish which approach to prioritisation is the most cost-effective. Cost-effectiveness is determined based on a comparison of the expected (mean) estimates of costs and QALYs for the alternative strategies considered. The alternative strategies are then compared by estimating the differential costs and outcomes between successively more expensive (or more effective) strategies, expressed in terms of an ICER representing the incremental cost per additional QALY gained. The ICER can then be compared with external thresholds used to establish whether or not this represents potential value for money to the NHS. The threshold applied here is in the region of £20,000–30,000 per additional QALY, based on decisions made by NICE.

Overview of analytical approach

In order to determine the cost-effectiveness of alternative strategies (including the use of biomarkers) for prioritising patients on the waiting list for CABG, several analytical steps are required. These are outlined schematically in Figure 1. These steps are broken down into four inter-related elements, comprising:

1. Defining the baseline characteristics of the representative cohort This represents variation in baseline characteristics and risk factors among patients on a waiting list for CABG. In the base-case analysis, all patients in the cohort are assumed to have the procedure within 3 months, but the order in which they undergo the procedure is then determined by the alternative methods of prioritisation under investigation. A different ordering of patients may result in different costs and health outcomes, as these are determined by the risk of cardiovascular events while awaiting CABG, the risk of the procedure itself and the risk of cardiovascular events after CABG. These costs and health outcomes will vary according to the baseline characteristics and risk factors of the representative cohort. It should be noted that we modelled a waiting list based on a fixed cohort of patients. The reason we did not model a more complex, dynamic situation (in which patients enter and leave the cohort over time) is because once patients are given a date for their operation it is rarely appropriate to change this.

2. Establishing the clinical effectiveness of the alternative strategies for prioritisation Systematic review and appropriate evidence synthesis approaches are required to generate measures of clinical effectiveness (in terms of relative risks) for the alternative strategies considered, reflecting the effects on prognosis for biomarkers. Ultimately, what is of interest is the extent to which utilising such information actually changes the order in which patients are prioritised for CABG and the subsequent costs and outcomes of such an approach.

3. Developing and populating a decision-analytic model in order to evaluate the lifetime costs and health outcomes of the alternative strategies The decision-analytic model provides an explicit analytical framework to combine data on a range of effectiveness, resource use and value parameters necessary to provide guidance on optimal reimbursement decisions. The decision-analytic model is used to structure the decision problem to identify the relevant parameters, and the amount and quality of available evidence can then be reflected in the inputs assigned to these parameters. The model is structured around the patient population of interest, characterising the potential events that may occur within both the short-term (e.g. events on a waiting list and procedural risk) and the longer term (e.g. subsequent prognosis after CABG) in terms of overall quality-adjusted life expectancy and costs and also reflecting the specific effect of the alternative strategies on these elements. This process inevitably involves methods for synthesising evidence for a range of parameters beyond simply the effectiveness of an intervention, including generating baseline event rates (e.g. to represent current practice), quality of life estimates and costs.

4. Estimation of the mean lifetime cost and health outcomes of the different prioritisation strategies in order to determine overall cost-effectiveness Propagation of the full range of input parameters into the decision-analytic model enables the expected lifetime costs and health outcomes for patients with different characteristics and risk factors to be established, conditional on the assigned day of CABG. For example, a particular patient in the representative cohort could be assigned CABG at day 10 with one prioritisation strategy and day 20 with an alternative prioritisation strategy. In this case it is necessary to estimate the costs and health outcomes associated with undergoing CABG at day 10 and day 20 for this particular patient. Overall cost-effectiveness of the prioritisation strategies can then determined by averaging the costs and health outcomes across patients in the representative cohort for each prioritisation strategy and comparing the subsequent ICER estimates with external thresholds representing value for money to the NHS. 36

FIGURE 1.

Overview of the analytical approach. CABG, coronary artery bypass graft surgery; CAD, coronary artery disease; CRP, C-reactive protein; eGFR, estimated glomerular filtration rate; QALY, quality-adjusted life-year.

Major data inputs

Several data sources and analytical approaches are required in order to carry out the analyses outlined above. The two major elements required to populate the decision model are: (1) systematic review and synthesis of existing evidence related to the clinical effectiveness of biomarkers themselves and (2) other evidence necessary to populate the decision-analytic model and the range of alternative strategies including the risk of cardiovascular events (on the waiting list, procedural and after CABG) as well as the effect on costs and health-related quality of life. The methods and results of the systematic review and synthesis of existing evidence related to clinical effectiveness of biomarkers are described in detail in Chapters 3 and 4. The results from this review provide one of the major inputs required for the overall decision model. The methods, range of data sources and results of the decision model are reported in Chapters 5 and 6. The key data source for a number of separate elements of the decision model, including the characteristics and risk factors of the representative cohort, was the Swedish Coronary Angiography and Angioplasty Registry (SCAAR). 37 This registry includes consecutive patients without exclusion criteria in all 30 centres in Sweden, and covers a total of 201,000 angiographies. Furthermore, the decision on further management after angiography is also available in SCAAR, making it possible to identify a representative cohort of patients with a decision to perform CABG as well as longer term prognosis. The SCAAR registry, alongside other published sources used to populate the decision model, is presented in detail in Chapter 5.

Inclusion criteria

An eligible publication had to meet five criteria:

• prospective study design

• patients with stable coronary disease

• one (or more) eligible biomarker

• eligible outcome

• relative risk and 95% CI reported.

Appendix 3 shows the proforma used to assess articles for eligibility.

Exclusion criteria

We did not exclude any studies based on methodological standards, sample size, duration of follow-up, publication year or language of publication.

Search strategies

We searched MEDLINE (PubMed) and EMBASE databases between 1966 and 30 November 2008 using a strategy developed with an expert librarian (who has a Postgraduate Diploma in Information and Library Science and 10 years experience as a medical librarian) based on terms for coronary disease (Cochrane Library of Systematic Reviews and Protocols), prognostic studies39 and biomarker. The final search combined these three searches with the connector word ‘AND’. Details of the search terms are shown in Appendix 2.

Reviewing titles and abstracts for eligibility

Three reviewers (NF, JD and KM) reviewed article titles and abstracts for eligibility and obtained full text articles where eligibility was definite or unclear. Multiple publications from one study data set were eligible where they reported results from two or more different biomarkers.

Data extraction

Two reviewers (NF and JD) independently abstracted data from eligible articles using a pre-defined coding protocol (see Appendix 4). Non-English articles were translated. Individual item disagreement between the two reviewers was resolved by consensus or, rarely, adjudication by a third reviewer (HH). The main details extracted were the year of publication; the number of patients at baseline that were included in the analysis, their mean age and the percentage of women; the baseline coronary morbidity (proportion with symptoms of angina, angiographic disease or previous myocardial infarction); mean [standard deviation (SD)] levels of biomarker at baseline [or median (interquartile range)]; type of assay; follow-up duration; the number and type (CHD, CVD, or ACM) of outcome events; the crude annual risk of these events calculated; whether or not the multivariate adjustment models included terms for age, sex, smoking status, obesity (nearly always body mass index), diabetes and one or more lipid variable [from total cholesterol, LDL cholesterol, high density lipoprotein (HDL) cholesterol, triglycerides]; and the comparisons (grouped or continuous).

Statistical analysis

Conversion of literature estimates of relative risk to a common cut-point

For eGFR, fasting glucose and haemoglobin, a single value is used to define chronic kidney disease, diabetes and anaemia respectively. However, reports used different single cut-points, and two or more cut-points. Therefore, in order to obtain an estimate for the presence versus absence of these diseases, we illustrate the method with chronic kidney disease, as defined by an eGFR lower than 60, versus its absence (eGFR ≥ 60). eGFR differs further from the situation with CRP, because eGFR is normally distributed and the relation between eGFR and CHD risk is not linear, as demonstrated in a previous meta-analysis in healthy population studies. 55

Where the relative risk was reported on a continuous scale, we used the mean and SD to estimate the proportion of the sample lying above and below the cut-point of 60 ml/min. Then we calculated the difference in means, expressed on the standard normal deviate scale, between these two groups and multiplied this by the reported eGFR SD to obtain the scaling factor.

Where the relative risk was reported in categories with the reference group as greater than or equal to 60 ml/min, we combined the two (or three) risk groups by weighting the relative risk estimates by the inverse of their variance. When the reference group was greater than or equal to 90 or to 75, we used that as reference group.

Meta-analysis

For each study, the relative risk estimate and its corresponding standard error were transformed to their natural logarithms to stabilise the variance and normalise the distributions. Summary relative risk estimates and their 95% CIs were estimated from a random effects model that considers both within- and between-study variation. 56 Statistical heterogeneity among studies was evaluated using the Cochran’s Q and I² statistics. 57 Small study bias, consistent with publication bias was assessed with funnel plot (i.e. a plot of study results against precision), by Begg’s adjusted rank correlation test, and by Egger’s regression asymmetry test. 58 To explore other potential sources of study heterogeneity, such as age, sex, annual risk rate, CRP levels, sample size, degree of covariates adjustment, duration of follow-up, study start year, events number and type of adjustment, we employed a meta-regression model that included these variables as covariates in categorical forms. We also performed meta-regression for continuous covariates, but due to the fact that these were often aggregated individual-level covariates, the results were interpreted with caution because of possible ecological bias. All analyses were conducted using stata, version 8.0 (StataCorp, College Station, TX, USA). All statistical tests were two-sided.

Results of searches

We reviewed a total of 14,723 unique titles and abstracts for eligibility and included 390 reports (see Figure 2 and Appendix 2). We translated studies from French, German, Italian, Portuguese, Spanish, Russian, Farsi, Japanese, Mandarin and Czech. A list of full text articles reviewed and rejected is given in the appendices.

FIGURE 2.

Flow chart of search results and selection of eligible studies.

We identified no previous meta-analyses, or systematic reviews, of any circulating biomarker in the prognosis of stable coronary disease. We identified no eligible studies in the ‘ideal’ population – among patients on the waiting list for CABG – in relation to novel circulating biomarkers.

Quality of individual studies

We included 390 reports of biomarker effects in our review. The number of events per study, and the demographic and clinical characteristics of patients were similar across biomarkers. The quality of individual study reports was similar across different biomarkers, and is summarised for the CRP studies in Table 3. Given that one data set commonly reported relative risks for the effects of more than one biomarker, and given that the quality of studies did not vary substantially between biomarkers, we show the 109 studies that reported results for CRP, the biomarker with the most eligible studies. No (0%) studies reported a clear statement of pre-specified research question.

C-reactive protein systematic review

We identified a total of 109 reports among patients with stable coronary disease where CRP was related to the risk of subsequent events (Table 4). Current angina symptoms were present in median 70.5 (range 9–100%) among the 45 studies providing data. Previous myocardial infarction was present in median 39 (range 8–100%) of patients in the 65 studies providing data. The mean age of patients across studies was a median of 62 years, and only one study had a mean age above 70 years. The median proportion of women in studies was 24.1%, and only three studies reported separate estimates among women. Eight thousand three hundred and sixty-nine outcome events were reported, with a median number of events per study of 51 (range 4–825).

C-reactive protein meta-analysis

The 109 study reports came from 77 unique studies. The pooled relative risk from the random effects model of top versus bottom third of CRP based on 77 unique studies was 1.96 (95% CI 1.76 to 2.17). There was marked heterogeneity, with an I-squared value of 79.1%. Evidence of small study bias is seen with smaller studies showing more extreme (positive) results. The funnel plot (not shown) was asymmetrical and the Egger test was significant (p < 0.001). This overall effect was weaker among those with more adjustment variables and earlier studies, but study characteristics did not account for the substantial heterogeneity between studies. Effects did not differ according to morbidity at baseline or among studies which reported CHD, CVD or ACM outcome events (data not shown).

Estimated glomerular filtration rate systematic review and meta-analysis

The systematic review of eGFR is shown in Table 5 and serum creatinine in Table 6. Given that eGFR is a more accurate reflection of renal function, taking account of age and sex, we did not meta-analyse the serum creatinine results. For routinely measured biomarkers, relative risks were 2.00 (95% CI 1.65 to 2.42) for eGFR below versus above 60 ml/min (based on 12 studies, 31,839 patients, 1639 outcome events).

Summary of systematic review for five routinely measured and eight novel biomarkers

We included 390 reports of biomarker effects in our review (Table 7), and Appendix 1 contains the results of the systematic reviews of each of the biomarkers. The number of events per study, and age, sex and baseline morbidity characteristics were similar across all the biomarkers. The quality issues identified for the CRP studies were likewise found for the other biomarkers (data not shown). Routinely assessed biomarkers contributed fewer studies than did novel biomarkers. Thus, CRP had 109 reports in the systematic review and 100 in the meta-analysis, by contrast with haemoglobin (15 and 4 respectively).

The estimated summary relative risks were 1.74 for fasting glucose higher than 7 mmol/l, 2.92 for haemoglobin lower than 13 g/dl, and 1.30 and 1.33 for total and LDL cholesterol (top versus bottom tertile) respectively.

For novel circulating biomarkers, relative risks comparing the top and bottom third were: 1.96 (95% CI 1.76 to 2.17) for CRP (based on 77 studies, 56,496 patients, 5798 outcome events) and, based on a smaller literature, 2.93 for BNP, 2.06 for homocysteine, 1.63 for IL-6, 1.59 for fibrinogen, 1.39 for ApoB, 1.24 for Lp(a) and 0.81 for ApoA-I. The quality of individual study reports was variable with many studies lacking clear description of population selection, and variable adjustment for simple clinical information – age, sex, smoking, diabetes, obesity and lipids.

Introduction

As outlined in Chapter 2, several analytical steps are required in order to determine the cost-effectiveness of alternative strategies for prioritising patients on the waiting list for CABG. While the results of the systematic review provide the most appropriate basis for estimating the clinical effectiveness of single and combination biomarkers (both routine and novel) in terms of their prognostic value in predicting events among patients with stable coronary disease, this addresses only one element of the overall decision problem. These results do not directly consider the effect of employing biomarkers in terms of their effect on final health outcomes expressed in generic terms (e.g. QALYs gained), subsequent health-care resource utilisation and costs; neither do these results provide a comparison against a range of alternative approaches to prioritisation which may be considered relevant comparators. Hence, in order to address the overall decision problem outlined in Chapter 2, a number of additional steps are subsequently required to determine the cost-effectiveness of alternative prioritisation strategies.

The additional steps comprise the methods and analytical approaches of the decision-analytic model itself, as well as the additional approaches required to integrate the results from the systematic review of circulating biomarkers within this framework. This chapter provides details of the methods, analytical approaches and sources of input data into the decision-analytic model. The chapter also outlines the approaches required to incorporate the results from Chapter 4 within the broader evaluation of cost-effectiveness. The results of the separate analyses, alongside the final estimates of cost-effectiveness, are reported in detail in Chapter 6.

Cost-effectiveness analysis

The cost-effectiveness analysis was undertaken from a UK health service perspective and costs were expressed in UK pounds sterling (GBP) at 2006/7 prices. A lifetime time horizon was employed and health outcomes were estimated in terms of QALYs. Costs and QALYs were discounted by 3.5% per annum. 191

Prioritisation strategies

There are several ways of formally prioritising patients waiting for revascularisation, of which circulating biomarkers represent one potential approach. It is not possible to establish the cost-effectiveness of using biomarkers without an explicit comparison against other formal approaches to prioritisation that are considered relevant and feasible options which could also be implemented in the NHS. Similarly, the use of any formal approach to prioritising waiting lists for CABG also needs to be evaluated in the context of current NHS practice. That is, any additional costs that may be imposed on the NHS because of the use of novel biomarkers (or any formal approach to prioritising waiting lists) need to be considered in relation both to the additional gains in health outcomes that may subsequently be achieved as well as to other ways in which these resources might be productively used elsewhere within the NHS. It is only through such an explicit comparison that the cost-effectiveness of alternative prioritisation strategies can subsequently be determined.

We identified four general approaches to the prioritisation of patients on a waiting list for CABG. These approaches comprised: (1) no formal prioritisation (routine clinical practice); (2) urgency scores; (3) risk score without the use of biomarkers; and (4) risk score with biomarkers. Within several of the general approaches there also exists a number of alternative approaches that could be considered (e.g. different approaches to evaluating urgency scores, different single and combination biomarkers comprising both novel and routine biomarkers, etc.). Each of these general approaches (and variants therein) represents potentially relevant and separate strategies that should be considered as part of an overall evaluation of cost-effectiveness.

While it remains desirable to evaluate all plausible strategies in the context of the decision problem, this also needs to be balanced against the analytical feasibility and data requirements required to provide robust inputs into such a comprehensive evaluation that would then provide a robust basis for informing NHS policy. This issue is particularly pertinent to the evaluation of alternative strategies that could be considered within the general approach of using risk score approaches with biomarkers. The five routine and eight novel biomarkers considered in Chapter 4 could feasibly be used either singly or in combination with any one or more of the remaining biomarkers. Consequently, there exists a large number of potentially relevant strategies – even with the eight novel biomarkers there are over 40,000 possible combinations that could be considered, and when combined with the five routine biomarkers this increases to 16 factorial. Ultimately, any attempt to comprehensively evaluate all of these strategies is likely to be a relatively futile exercise, the problems of which will also be compounded by the lack of robust clinical data on the majority of these strategies and the potential difficulties that may subsequently be encountered in terms of linking this evidence to the broader decision model itself.

Hence, rather than attempting to be comprehensive in terms of the strategies considered, with particular reference to the biomarker strategies, the decision model evaluates a more restrictive range of strategies. The strategies that were finally selected were chosen on the basis that these were considered to be particularly important questions relevant to existing NHS decision-making in terms of routine biomarkers that are already widespread (e.g. eGFR) and those whose use remains variable (e.g. CRP). A total of seven separate strategies, within the four general approaches, were thus evaluated in detail as part of the decision model. These separate strategies are summarised below.

Strategy 1: No formal prioritisation (routine clinical practice) This strategy reflects how prioritisation is currently undertaken in routine clinical practice. Routine clinical practice will inevitably reflect the variability that exists in different centres in terms of the approaches employed to prioritising patients on a waiting list. This variability will reflect differences in the formal and informal approaches to ordering waiting lists that are currently being applied. This provides an appropriate baseline with which to assess the alternative prioritisation strategies that are based on a more formal and systematic approach to prioritisation.

Strategies 2–3: Urgency scores Within this general approach, formal prioritisation is guided by the use of urgency scores. Explicit urgency scores, where those with the most disabling symptoms and worst prognosis should be prioritised first, have been proposed as a formal approach to the prioritisation of waiting lists. The two particular algorithms considered here are based on the Ontario31 and New Zealand168 scoring systems. These algorithms are subsequently evaluated as two separate strategies within the decision model.

Strategy 4: Risk score without biomarkers An alternative to the use of urgency scores based primarily on symptom measures is to consider formal prioritisation approaches guided entirely by the predicted risk of cardiovascular events. Such an approach could be implemented by employing a risk equation based on routinely measured and/or observable risk factors (excluding information based on biomarkers) which can be demonstrated to be predictive of the potential risk of experiencing a cardiovascular event. The use of such an approach would mean that individuals within a cohort of subjects on a waiting list could be stratified according to their individual risk score, with patients predicted to be at a higher absolute risk of experiencing a cardiovascular event on the waiting list subsequently prioritised above individuals predicted to be at a lower absolute risk.

Strategies 5–7: Risk score with biomarkers This general approach is similar to that outlined for Strategy 4. However, in addition to the routinely available information considered within Strategy 4, the risk score is refined by including additional prognostic information provided by the biomarkers themselves. A total of three separate strategies are considered based on a risk prediction equation that also incorporates the additional prognostic information generated by biomarkers. The three strategies considered were: (1) adding a single, routinely available biomarker to the risk prediction equation (eGFR); (2) adding a single, novel biomarker (CRP); and (3) adding a combination of biomarkers (both CRP and eGFR). Hence, the strategies reflect the use of a single routine or a novel biomarker as well as employing a combination of biomarkers.