Echocardiography in newly diagnosed atrial fibrillation patients: a systematic review and economic evaluation

Significance for patients in terms of ill health (burden of disease)

Atrial fibrillation is a common and significant cause of cardiovascular-related morbidity and mortality. The condition is a major predictor of atrial thrombosis, peripheral embolism and stroke, especially in elderly patients. 16,27,28 Evidence for the Framingham Heart Study noted a four- to fivefold rise in the risk of stroke in patients with AF. 14 The observed increase in risk has been attributed to the presence of LV hypertrophy, which is often associated with long-standing AF. 27 The risk of stroke also increases with age. 14,20

Coexisting cardiovascular disease and AF significantly reduce quality of life in those patients with symptoms such as palpitations, light-headedness and fatigue. Furthermore, patients with AF with underlying coronary heart disease and chronic lung disease are more likely to suffer from myocardial infarction (MI) and acute respiratory failure. 29 Evidence from a longitudinal cohort study has also suggested an increased risk of dementia in patients with AF. 30

Atrial fibrillation is also associated with an increase in mortality rate. In the Framingham study,19 AF was associated with 1.5- to 1.9-fold increase in the risk of mortality, following adjustments for the underlying cardiovascular diseases in affected patients. 31 This conferred risk of death is similar in both men and women and does not vary significantly by age. 31 Data from the Renfrew/Paisley cohort with a 20-year follow-up period demonstrated an increase of 1.8- to 2.8-fold and 1.5- to 2.2-fold in cardiovascular-related death and all-cause death, respectively. 32

Significance for the NHS

In many developed countries, AF is increasingly becoming a significant public health challenge. It is a major cause of increased hospitalisation in the UK. 18,33 The number of hospital admissions for patients with AF has at least doubled in recent times. 34

A study of the health and social care-related expenditure on patients with AF in 1995 showed that an estimated £244M, which accounted for 0.62% of the total NHS budget was spent on patients with AF. 35 Of this, half of the cost covered the hospitalisation of patients, whereas 20% of the total expenditure was for the cost of drug prescriptions. An extra £46.4M was used to provide long-term nursing home care following admission. Based on projections of AF expenditure in 1995, it was estimated that direct costs of the condition would be approximately £459M in the year 2000. 35

Aim of transthoracic echocardiography

Transthoracic echocardiography is used to assist the diagnosis and management of a broad range of heart conditions. The commonest indications are for heart failure, murmur, palpitations/arrhythmias/blackouts and hypertension. 49

Transthoracic echocardiography

The standard adult transthoracic echocardiogram measures structure and function of the heart. It should reliably describe quantitative LV systolic and diastolic function and assess all valves, including minor abnormalities that may progress or need follow-up, basic prosthetic valve function, and common congenital abnormalities cardiomyopathies, as well as detecting the presence and significance of pericardial fluid. 50

The following cardiac and vascular structures are routinely evaluated as part of a complete adult echocardiographic report: left ventricle, mitral valve, left atrium, aortic valve, aorta, right ventricle, tricuspid valve, right atrium, pulmonary valve, pulmonary artery, pericardium, inferior vena cava and pulmonary veins. LV size is one of the most important components of LV function quantification. Changes in LV dimensions are frequently interpreted as indices of progression or regression of a disease state that affects the left heart. 51

A complete transthoracic study includes two-dimensional (2D) and, usually, M-mode (motion mode) echocardiography, as well as spectral and colour Doppler techniques. M-mode supplies additional information when indicated; it is obtained by selecting any of the individual sector lines from which a 2D image is constructed. It is useful for quantifying linear dimensions of the cardiac chambers and walls when the correct direction is verified under 2D imaging. Doppler modalities provide functional information on intracardiac flow haemodynamics, including measurement of systolic and diastolic blood flow velocities and volumes, assessment of the severity of valvular lesions, and location and severity of intracardiac shunts. Pulsed-wave Doppler is useful for locating and timing blood flow within the physiological range of velocities. Continuous wave Doppler can accurately measure the highest flow velocities and estimate the gradients across valves or interventricular defects. Colour flow mapping provides a composite picture of flow over a larger area and is most useful for screening valves for regurgitation and stenosis, and detecting the presence of intracardiac shunts. Colour flow M-mode is useful for timing blood flow information. 51

Transthoracic echocardiography provides comprehensive evaluation of cardiac and vascular structures and function, and can immediately affect the diagnostic and management work-up of the patient. It is accepted that 2D TTE can accurately assess cardiac chamber size, wall thickness, ventricular function, valvular anatomy, and the size of great vessels. Pulsed-wave, continuous wave and colour flow Doppler echocardiography provide measurements of blood flow velocities and assessment of intracardiac pressures and haemodynamics, and can detect and quantify stenosis, regurgitation and other abnormal flow states. 51

The diagnosis of heart conditions requires the integration of clinical, laboratory and echocardiographic data. The contribution of TTE in the diagnosis of heart conditions depends on the particular condition. It is particularly useful for the assessment and management of valve disease, providing good structural information about the valve and its supporting structures. Doppler provides good information about the severity of the lesion and whether the valve is repairable. The impact of valve lesion on the heart as a whole can also be assessed. 52 The usefulness of TTE in an intensive care setting has been reported by Stanko et al. 53 TTE resulted in a change of diagnosis in 29% of studies, and a change of management in 41% of studies.

Indications

According to the British Society of Echocardiography (BSE),50 TTE is indicated for the following conditions if certain circumstances (relating to seriousness) are fulfilled: heart murmurs, native valvular stenosis, native valvular regurgitation, prosthetic valve assessment, infective endocarditis, ischaemic heart disease, cardiomyopathy, pericardial disease, cardiac masses, pulmonary disease, neurological disease, arrhythmia/palpitations/syncope, echocardiography before cardioversion, hypertension, aortic and major arterial disease, and preoperative echocardiography for elective and semi-urgent surgery.

Various guidelines for the use of TTE for all indications have been reported in recent years. Details of the clinical indications for echocardiography are provided by the BSE. 50 The Bedfordshire and Hertfordshire Cardiac Network49 describes the effective use of TTE in adults for indications including heart murmur and palpitations, and the National Imaging Board52 provides guidance relating to a number of cardiac imaging modalities, including echocardiography.

Technical difficulties

Some patients give poor images and the information derived using TTE from these patients can therefore be limited. Furthermore, the accuracy of TTE depends on the experience of the person reporting the images. 52 The risks associated with TTE are extremely low.

Setting and equipment required

Transthoracic echocardiography is a non-invasive imaging technique performed with the use of an ultrasound machine. It provides real-time images, is portable and of low cost. 51 It is usually performed in cardiology clinics and is less used in primary or non-specialist secondary care, and may be undertaken by a cardiologist, BSE-accredited echocardiographer or general practitioner (GP) with special interests. 1

Criteria for use

Patients currently meeting the criteria for recommended TTE in the NICE guidelines1 for AF comprise:

• younger patients for whom a baseline echocardiogram is important for long-term management

• patients for whom cardioversion (electrical or pharmacological) is being considered

• patients in whom there is a high risk or a suspicion of underlying structural/functional heart disease (such as heart failure or heart murmur) that influences their subsequent management

• patients in need of clinical risk stratification for antithrombotic therapy, where clinical evidence is needed of LV dysfunction or valve disease.

Guidelines from NICE1 state that TTE is not recommended for patients with AF for whom the need to initiate anticoagulation therapy has already been decided on clinical criteria.

Inclusion

Exclusion

Inclusion

Quantity and quality of research available

The literature search yielded 15,824 article citations when duplicates had been removed. Figure 1 shows study selection, in a modified version of the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) flow diagram. 61 Citations presenting purely economic analyses were not included in this chapter. References excluded at the full paper screening stage (n = 38), with reason for exclusion, are presented in Appendix 8.

FIGURE 1.

Study selection for diagnostic review.

There were 44 diagnostic accuracy studies62–105 and five prognostic studies106–110 accepted into the review.

A summary of included diagnostic accuracy studies is presented in Table 1 and a summary of included prognostic studies is presented in Table 2.

Of the 44 included studies,62–105 there were 17 studies62,69,72,74,75,77,79,80,82,86,90,91,95,97,100–102 that included AF patients in the population. Of these, for two studies79,91 all participants had AF. Although two studies63,71 stated that there were no patients with AF in the population, in other studies it was not reported.

For all categories of pathologies sought, studies of diagnostic accuracy were identified. AF population studies were available for the categories of structural defect, ischaemia/thrombosis, pulmonary disease and valvular heart disease.

Methods of TTE represented were 2D, M-mode, pulsed and continuous wave Doppler, and colour Doppler. All studies had a high percentage of usable, good images from TTE.

All five prognostic accuracy studies included106–110 had a population of non-valvular/non-rheumatic AF. Of the categories of pathologies sought, only heart failure and valvular heart disease were represented. Three of the studies were prospective studies106,107,110 with follow-up ranging from mean 1.3 to 9.6 years. Two of the studies108,109 were retrospective. The TTE methods represented were 2D, M-mode and colour Doppler.

Quality of included studies

Quality assessment forms are in Appendix 5. According to the level of hierarchy proposed by Merlin et al. , 57 studies ranged from level 2 (higher quality) to level 3c (lower quality). Twelve studies were of level 2,64–67,76,80,81,83,88,91,94,102 a study of test accuracy with an independent, blinded comparison with a reference standard among consecutive patients. Six of the studies were level 3a,68,73,74,84,99,103 i.e. they differed from level 2 only in being among non-consecutive patients. Twenty-two of the studies were level 3b,62,69,71,72,77–79,82,85–87,89,90,92,93,95,96,98,100,101,104,105 comparisons with a reference standard that did not meet criteria for higher levels of evidence. There were four diagnostic case–control studies, level 3c. 63,70,75,97

Considering only diagnostic accuracy studies with AF populations, there were three level 2 studies,80,91,102 one level 3a study,74 11 level 3b studies62,69,72,77,79,82,86,90,95,100,101 and two level 3c studies. 75,97 As all AF population studies were included, and non-AF population studies selected according to hierarchy of evidence, this explains the higher proportion of level 2 and 3a studies with non-AF populations.

For the prognostic studies, one study was level 2,107 a prospective cohort study; two studies were level 3b;106,110 and two studies were level 3c,108,109 retrospective cohort studies.

Selected items from QUADAS were also addressed (see Appendix 5). We did not ask about representativeness of patients in the study for participants receiving the test in practice, as this review is concerned with screening patients with AF, and so an AF population, although relevant to this review, would not necessarily reflect quality of the diagnostic studies. All included diagnostic studies were of high quality in terms of all patients receiving TTE and a reference standard, and the reference standard being administered whatever the TTE results, and the reference standard being independent of TTE. More than half of the studies were blinded.

Some studies selected participants on the basis of having usable TTE images, and some excluded indeterminate images from the analysis of sensitivity or specificity, whereas six studies explicitly included either poorer images in analysis78,92,96–98,105 or provided separate analyses by the inclusion or exclusion of poor-image-quality TTE. 67

Diagnostic accuracy results

Eight studies64,78,86,92,96–98,105 reported diagnostic accuracy of TTE in structural defects (Table 3). TTE was presumed the gold standard for one study of ventricular septal defect. 64 Sensitivity ranged from 0.25 for atrial septal hypertrophy86 to 1 for ostium primum atrial septal defect96 or ventricular septal rupture. 98 Two studies86,97 reported specificity, which ranged from 0.9 for rupture of chordae tendineae97 to 0.909 for atrial septal hypertrophy. 86 Six78,92,96–98,105 of the eight studies used catheterisation, surgery or autopsy as the comparator diagnostic test, whereas one used clinical cardiac examination,64 and one used TOE86 (see Table 3).

Nine studies62,77,79,82,90,91,95,100,102 reported diagnostic accuracy of TTE in ischaemic heart disease (Table 4). Sensitivity ranged from 0 for right atrial appendage (RAA)79 or left atrial appendage (LAA)102 thrombus to 0.955 for thrombosis of ventricle. 100 Specificity ranged from 0.857 for thrombosis of ventricle100 to 1 for LA79 or right atrial (RA)79 thrombus. Five of the studies62,77,90,95,100 used surgery or angiography, three used TOE82,91,102 and one used CT79 as comparators (see Table 4).

Two studies66,80 reported diagnostic accuracy of TTE in pulmonary disease (Table 5). Sensitivity ranged from 0.523 for PE66 to 1 for pulmonary hypertension. 80 Specificity ranged from 0.6 to 1 for pulmonary hypertension. 80 The study of PE66 used perfusion lung scan with radiography or pulmonary angiography as a comparator, whereas the study of pulmonary hypertension80 used catheterisation (see Table 5).

Three studies67,76,94 reported diagnostic accuracy of TTE in endocarditis (Table 6). Sensitivity ranged from 0.4494 to 0.871;67 specificity ranged from 0.61567 to 0.98. 94 Two of the studies used TOE as a comparator,67,76 whereas the other study94 used information obtained from clinical follow-up (see Table 6).

Twelve studies64,65,68,69,72–75,88,101,103,104 reported diagnostic accuracy of TTE in valvular heart disease (Table 7). TTE was presumed gold standard for four studies of MR,64,68 aortic stenosis,64,68 mitral valve prolapse (MVP),64 valvular heart disease,64,88 aortic regurgitation (AR)64,68 and tricuspid regurgitation. 73 Sensitivity ranged from 0.222 for mitral stenosis leaflet calcification103 to 1 for mitral stenosis104 or mitral regurgitation75 or severe AR. 69 Specificity ranged from 0.655 for mitral stenosis to 1 for AR or MR. Six of the studies used catheterisation/aortography or radiography/cinefluorography as comparators,69,74,75,101,103,104 four of the studies used clinical examination,64,65,68,88 one used TOE68 and one used CT73 (see Table 7).

One study85 reported the accuracy of TTE in differentiating between ischaemic and non-ischaemic cardiomyopathy (ICM and non-ICM) (Table 8). This study reported a sensitivity of 0.77 and specificity of 0.77 for differentiating between ICM and non-ICM with a comparator of angiography.

Five studies63,71,81,87,99 reported accuracy of TTE in the diagnosis of heart failure (Table 9). TTE was presumed the gold standard for two studies of CHF81 and LV dysfunction. 99 Sensitivity ranged from 0.737 for CHF63 to 0.93 for LV hypertrophy. 87 Specificity ranged from 0.75 for CHF63 to 1 for LV dysfunction. 71 Two of the studies used clinical diagnosis as comparators,81,99 two studies63,71 used radiography or catheterisation, and one used autopsy results87 (see Table 9).

Four studies70,83,84,89 reported diagnostic accuracy of TTE in aortic dissection (Table 10). Sensitivity ranged from 0.59383 to 0.953. 89 Specificity ranged from 0.50889 to 0.977. 89 Three studies83,84,89 included surgery, autopsy or angiography as comparators; the other study used aortography70 (see Table 10).

One study93 reported the diagnostic accuracy of TTE for intracardiac masses (Table 11). This study reported sensitivity 0.882 and specificity 0.953, with a comparator of surgery. 93

Prognostic study results

Five studies106–110 reported prognosis based on TTE-diagnosed pathologies in AF populations.

The pathologies were left atrial diameter (LAD); mitral annular calcification; MVP global, moderate to severe or reduced LV systolic dysfunction; any or severe MR; and valvular abnormality (Table 12).

Prognosis was investigated by studies for the types of valvular heart disease, mitral annular calcification, MVP, MR and valvular abnormality. Mitral annular calcification was non-significantly associated with thromboembolism by age-adjusted analysis RR of 0.6 (95% CI 0.2 to 1.5; p > 0.2). 110 MVP had a non-significant association with risk of stroke unadjusted RR of 0.29 (p = 0.22). 106 For MR, grade 1 MR (compared with no MR) odds ratio (OR) of 2.689 (95% CI 1.039 to 7.189; p = 0.0434) was significantly associated with history of thromboembolic events. 108 Severe MR had a non-significant association with risk of stroke (relative to none or mild MR) unadjusted RR of 1.7 (p = 0.59),106 was found to be protective against stroke with hazard ratio (HR) of stroke for increase in MR from mild to severe groups 0.45 (95% CI 0.20 to 0.97) by multivariate analysis (multivariate analysis includes MR, LAD, sex and age),109 and was non-significantly associated with thromboembolism by age-adjusted analysis RR 0.4 (95% CI 0.1 to 3.0; p > 0.2). 110 For MR, the retrospective studies108,109 found significant associations with prognosis, whereas the prospective studies106,110 had non-significant results. Any detected valvular abnormalities had a reported HR for mortality: diabetic participants 2.05 (95% CI 1.10 to 3.82; p = 0.0229), non-diabetic participants HR 1.88 (95% CI 1.30 to 2.70; p = 0.0007). 107 For this study, diabetics and non-diabetic groups differed in that diabetic participants were older and had higher comorbidity, and more of them received oral anticoagulation; there was also a relatively small number of diabetics. 107

Left atrial diameter had a reported non-significant association with risk of stroke unadjusted RR of 1.02/mm (95% CI 0.99 to 1.06; p = 0.10106) and was reported to have HR of stroke for every 10-mm increment in LA size of 1.06 (95% CI 0.75 to 1.49) by multivariate analysis (multivariate analysis includes MR, LAD, sex, age). 109 LAD (corrected for body surface area) as a continuous variable by univariate analysis was significantly associated with thromboembolism (p = 0.01110), and had reported HR for mortality, diabetic participants 1.01 (95% CI 0.97 to 1.05; p = 0.6445), HR non-diabetic participants 1.06 (95% CI 1.03 to 1.08; p < 0.0001). 107 Moderate to severe LV dysfunction was associated with a significantly higher risk of stroke relative to normal LV function or mild dysfunction,106 and global LV dysfunction was significantly associated with risk of thromboembolism. 110 Reduced LV function had reported HR for mortality, diabetic participants 1.52 (95% CI 0.85 to 2.70; p = 0.1598), HR non-diabetic participants 2.28 (95% CI 1.58 to 3.29; p < 0.0001). 107

Prevalence results

One prevalence study119 was identified that sought to identify prevalence of atrial septal defect, as presented in Table 14. This study, by Lip et al. ,119 found a prevalence of 0.9% for atrial septal defect. This study looked at a cross-section of patient records in UK primary care.

Fifteen studies28,111–114,116–125 investigated the prevalence of pathologies within the category ‘ischaemia/thrombosis’, as shown in Table 15. One study, that by Lip et al. 1997,119 found a 28.8% prevalence of ischaemic heart disease.

The six studies112,116,117,120,124,125 reporting prevalence of LA thrombus gave differing prevalences, ranging from 3%117 to 18%. 116 Both of these studies116,117 used TOE to diagnose thrombi. These six studies differed in terms of sample size and population, with Maltagliati et al. 120 including atrial flutter, Shen et al. 124 restricting the population to patients with subtherapeutic INR, and Kleeman et al. 117 using a population admitted for cardioversion.

Six studies28,114,120–123 investigated prevalence of LAA thrombi; the lowest reported prevalence was for patients undergoing catheter ablations 1.6%123 and the highest was 40%,122 although this study had a small sample size (n = 30). Two studies28,120 looked at RAA thrombus, reporting prevalences of 0.5%120 and 6.7%. 28 Both of these studies used TOE to diagnose thrombi.

Four studies115,118,119,122 investigated the prevalence of valvular pathologies (Table 16). Two of these studies115,119 reported prevalence of valvular heart disease as 13.4%115 to 26.1%,119 and combining rheumatic and non-rheumatic valvular heart disease from Levy et al. 118 would give 18.8% prevalence. Mitral valve disease had a reported prevalence of 10.4%,115 and Santiago et al. 122 reported prevalence of 30% for MR.

Three studies115,118,119 investigated the prevalence of pathologies within the category cardiomyopathy, as presented in Table 17. Dang et al. 115 and Lip et al. 119 reported prevalences of 4.5%115 to 5.4%. 119 Levy et al. 118 reported a prevalence of 4.5% for hypertrophic cardiomyopathy and of 9.2% for dilated cardiomyopathy.

Two studies115,118 investigated the prevalence of heart failure, as shown in Table 18. Dang et al. 115 reported a prevalence of 31.1% and Levy et al. 118 reported the prevalence of CHF to be 29.8%.

Introduction

The specific focus of the model is the clinical decision whether or not to prescribe an OAC to a patient. Three types of OAC are considered: warfarin, dabigatran and rivaroxaban. The decision involves balancing competing clinical risks, as these drugs reduce the risk of stroke, but an adverse effect increases the risk of major bleeding events, which, in some cases, can lead to clinical outcomes as, or more, severe than the strokes that the treatment aims to prevent. In patients with an underlying low risk of stroke, the added risks of treatment in terms of bleeding events can outweigh the additional benefit caused by the reduced risk of stroke, and so it is neither clinically effective nor cost-effective to prescribe anticoagulants in this patient group. Within the context of this model, the added clinical benefit of performing TTE in a patient is a direct result of the increased appropriateness of the decision whether or not to prescribe anticoagulants, measured in terms of estimated QALYs.

Uncertainty about appropriate anticoagulants

CHADS₂ decision rule and choice of oral anticoagulants

Although dabigatran and rivaroxaban are generally considered to be very similar OACs in terms of costs, clinical effectiveness, and risk of major bleeding events, the indications provided in the NICE guidance for each OAC differ slightly.

The guidance for rivaroxaban states:

• Rivaroxaban is recommended as an option for the prevention of stroke and systemic embolism within its licensed indication, that is, people with non-valvular AF with one or more risk factors, such as:

  • CHF

  • hypertension

  • ≥ 75 years of age

  • DM

  • prior stroke or transient ischaemic attack. 47

It is noted that this guidance is equivalent to stating that rivaroxaban is recommended for patients with AF with a CHADS₂ score of ≥ 1.

The guidance for dabigatran states that:

• Dabigatran etexilate is recommended as an option for the prevention of stroke and systemic embolism within its licensed indication (i.e. in people with non-valvular AF with one or more of the following risk factors):previous stroke, transient ischaemic attack or systemic embolism

  • LV ejection fraction of < 40%

  • symptomatic heart failure of New York Heart Association (NYHA) class 2 or above

  • age ≥ 75 years

  • age ≥ 65 years, with one of the following: DM, coronary artery disease or hypertension. 143

It is noted that an implication of the guidance is that dabigatran is indicated in patients who have a CHADS₂ score of ≥ 1 if they are aged ≥ 65 years. In people aged < 65 years, the correspondence between CHADS₂ score and OAC decision is not clear cut. For simplicity, and because the mathematical model developed does not model the progression of each individual disease state that is incorporated in the CHADS₂ score, the mathematical model will be run in patients aged ≥ 65 years only when considering dabigatran as the OAC of choice.

Based on the above NICE recommendations, recent ESC guidance47 and broad recognition that warfarin carries a higher risk of major bleeding events for most patient groups than either dabigatran or rivaroxaban,6 different CHADS₂ thresholds were used for each OAC, as shown in Table 21 below.

Comparator strategy: CHADS₂ plus transthoracic echocardiography

In our comparator strategy, ‘CHADS₂  + TTE’, the decision to coagulate can also be made as a result of TTE identifying a structural feature of LA abnormality that predisposes an individual to a high risk of stroke. 144 For the purposes of this report we define LA abnormality as a patient having one or more of the following conditions:

• LAA thrombi

• dense spontaneous echo contrast

• LAA low-flow velocities.

These conditions have been chosen as they are the ones used in a recent publication by Provedencia et al. 13 which provides key data for populating the model.

This means the comparator offers two alternative routes by which a decision to recommend OACs can be made:

• a CHADS₂ score at or above the threshold required to recommend OACs

• a TTE indicating the presence of LA abnormality.

In effect, this means that some individuals who would not have received OAC with CHADS₂ alone will receive OAC following CHADS₂ + TTE due to detection of LA abnormality. It is noted that in no instance patients who would have received OAC using CHADS₂ alone would have treatment withheld. As such, it has been assumed that TTE will only provide information that can alter patient management when the CHADS₂ score does not indicate treatment with OAC. In the remaining patients we have not formally assessed the cost-effectiveness of TTE, noting that costs would be incurred for no assumed gain.

CHA₂ DS₂-VASc score [Congestive heart failure, hypertension, age ≥ 75 years (doubled), diabetes, stroke (doubled), vascular disease, age 65–74 years, and sex category (female)]

An alternative variation of the CHADS₂ instrument exists, which uses additional information such as gender to make the decision. It was decided not to produce an additional 14 scenarios using CHA₂ DS₂-VASc rather than CHADS₂ as the baseline strategy for a number of reasons. These include the already large number of scenarios considered; the fact the recent NICE guidance47 on rivaroxaban and dabigatran relate more clearly to CHADS₂ than CHA₂ DS₂-VASc scores, and the fact that CHADS₂ is the more established of the two instruments.

Introduction

The short-term diagnostic section of the model begins by simulating a series of male and female cohorts aged either 50 or 65 years old with newly diagnosed AF but with none of the following conditions: CHF, hypertension, DM, prior stroke or TIA, or vascular disease. This patient group has been selected as they would have a CHADS₂ score of 0 and thus would not be currently recommended treatment with an OAC. Additionally, otherwise identical cohorts of individuals with a CHADS₂ score of 1 are considered in the case of warfarin.

A summary of the sources of data used to populate the model is provided in Appendix 10.

For this patient group the age at death (assuming no AF- or AF treatment-related mortality) was simulated. The diagnosis, or not, of LA abnormality following TTE was additionally simulated.

Risk of all-cause mortality

The risks of all-cause mortality were estimated separately for males and females using gender-specific UK life table data. 145 These were converted into probabilities of males and females dying in each forthcoming year assuming initial ages of 50 and 65 years. These produced a range of distributions of death given. For simplicity, it was assumed that all remaining patients would die within their 101st year.

The assumed sensitivity and specificity of transthoracic echocardiography of diagnosing left atrial abnormality

Table 2 in the paper by Providencia et al. 13 provides sufficient information to calculate an estimate of sensitivity and specificity of TTE in diagnosing LA abnormality. These data were used to derive Tables 23 and 24. Patients were assessed using both TTE and TOE, and for TTE were assigned an echocardiographic risk score depending on how many structural features that are constituents of LA abnormality were identified through TTE. It was assumed that TOE was the gold standard and identified all patients with LA abnormality.

This allows calculation of sensitivity and specificity for this patient group:

• sensitivity = TP/(TP + FN) = 87/(87 + 5) = 0.946 (to three decimal places)

• specificity = TN/(TN + FP) = 83/(83 + 159) = 0.343 (to three decimal places).

The four cells in Table 25 also allow uncertainty to be estimated for the joint distribution of sensitivity and specificity. One thousand draws from a Dirichlet distribution using the cell counts as parameter values were used to jointly calculate sensitivity and specificity. The resulting joint distribution of sensitivity and specificity estimates is shown in Figure 4 as a contour plot. A contour plot represents variation in density by joining together points on the surface with equal heights, allowing three-dimensional information to be presented in a monochrome graph. This method of presenting the data was preferred to a scatterplot, as the relative density would have been relatively difficult to ascertain in a scatterplot containing 1000 points. In Figure 4 the peak density of the plot is, unsurprisingly, close to the point estimates of 0.95 for sensitivity and 0.34 for specificity.

FIGURE 4.

Joint distribution of sensitivity and specificity based on 1000 draws from a Dirichlet distribution.

It was assumed that the derived distributions of sensitivity and specificity were applicable to all patients and were thus assumed applicable to patients who had a CHADS₂ score of 0.

Data reported in table 2 of Provedencia et al. 13 indicate that of 24 patients with a CHADS₂ score of 0, two had a LA abnormality. Given the small number of data available (the number in the LA abnormality group being < 5) an uninformative prior of 0.5 was added to each paired data set, culminating in an expected 2.5 out of 25 patients expected to have a LA abnormality among those with a CHADS₂ score of 0. These values were used to populate beta distributions to allow for uncertainty in the true proportion of patients with LA abnormality within the CHADS₂  = 0 score.

For patients with a CHADS₂ score of 0 there were 17 patients out of 79 with a LA abnormality.

CHADS₂-related stroke risk

We assumed that higher CHADS₂ scores were associated with a higher risk of stroke. Our estimates were based on unadjusted stroke risk estimates presented in Friberg et al. 146 These estimates are presented in Table 25 (95% CIs were estimated from beta distributions).

Patients with LAA were assumed to have a risk of stroke independent of CHADS₂ score. The risk was set as 8.0% (95% CI 7.26 to 8.31) per annum, as reported in Connelly et al. 147 For simplicity, the risk of stroke was assumed to apply throughout the lifetime of the patient.

Estimating uncertainty in CHADS₂-related stroke risk using CHADS₂ score

Introduction

Simulating values

In order to ensure no estimated risks were less than zero, we assumed that the above estimates followed a log-normal distribution, producing 10,000 simulated values for the stroke risk associated with each score, as shown in Figure 5.

FIGURE 5.

The estimated distribution of annual stroke risk by CHADS₂ score.

Effect of dabigatran on stroke risk

The effect of warfarin in preventing strokes was taken from a 2006 meta-analysis. 148 The effects of dabigatran and rivaroxaban compared with placebo (i.e. no treatment) were estimated using an indirect comparison approach. Estimates for the annual risk ratio of a stroke for patients given 150 mg of dabigatran twice daily compared with warfarin were taken from a paper based on the Randomized Evaluation of Long-Term Anticoagulation Therapy (RE-LY) study. 147 Data on the effect of rivaroxaban were taken from Patel et al. 149

Overall, 1000 simulated values were sampled from each distribution, with derived values for dabigatran compared with placebo estimated by multiplying the RRs of the sampled warfarin compared with placebo value and the dabigatran compared with warfarin value, to produce a distribution of estimates of the RR of dabigatran compared with placebo (Figure 6). An identical methodology was used to produce a distribution of the RR of rivaroxaban compared with placebo.

FIGURE 6.

The assumed distributions of each OAC vs. placebo.

Table 26 shows summary statistics from these two papers, as well as for the simulated distribution produced by combining the two, whereas the density functions of the three distributions are shown below (Table 27).

The estimated risk of bleeding associated with dabigatran

The risk of major bleeding events in patients receiving dabigatran was based on results published using data from the RE-LY trial. 150 This reported that major bleeding events occurred at a rate of 2.12% per year in patients given dabigatran and who were < 75 years of age, and at a rate of 5.10% in patients aged ≥ 75 years. A simulation approach, based on the binomial distribution and incorporating information about the different sample sizes of these two age groups within the trial, was used to represent uncertainty around these estimates for use within the probabilistic sensitivity analysis (PSA). Table 27 presents credible intervals (CrIs) for these two age groups, and Figure 7 shows these results graphically. CrIs were calculated using a presumed sample size: the paper reports 10,855 participants of < 75 years and 7258 participants aged ≥ 75 years; an equal probability of assignment between the three trial arms was assumed.

FIGURE 7.

Distribution of simulated estimates for the annual risk of bleed on dabigatran.

The estimated risk of bleeding associated with warfarin

Results from the RE-LY study150 were used to estimate the annual risk of bleed associated with warfarin, as shown in Table 28. The presumed sample sizes used for estimating the risk of bleeding in patients treated with dabigatran were also used in estimating the risk of bleeding associated with warfarin.

The estimated risk of bleeding associated with rivaroxaban

The annual risk of bleeding given rivaroxaban was estimated indirectly by combining estimates of the risk of bleed given warfarin compared with placebo148 with estimates of the risk of bleed given rivaroxaban compared with warfarin. 149 The central estimates and CrIs are shown in Table 29. The presumed sample sizes used for estimating the risk of bleeding in patients treated with dabigatran were also used in estimating the risk of bleeding associated with warfarin.

Cost of dabigatran

This is assumed to cost £920.43 per year, assuming two 150-mg tablets daily at a cost of £2.52 per day. 142 This cost is fixed within all runs of the PSA.

Cost of rivaroxaban

The cost of rivaroxaban was assumed to be £767 per year, based on 20 mg per day. 151 This price was fixed in all runs.

Cost of warfarin

The annual cost of warfarin includes both drug costs and monitoring costs. The dosage received depends on the results of monitoring, although the costs of different dosages of drugs differ only marginally in comparison with the costs of monitoring. The annual monitoring costs were assumed to be £241 per annum, in line with assumptions made by the appraisal committee in the review of dabigatran. 142 The annual cost of the drug was taken from the British National Formulary (BNF) website, and suggested prices varied from 3.1 to 4.8 pence per tablet, depending on dose, equivalent to between £11.22 and £17.87 per year. 152

Including monitoring costs, this suggests a range for average total costs of between £252.22 and £258.87 per year. The average of this range (£255.54) was used as the central estimate. In the PSA the total costs were assumed to be drawn from a uniform distribution ranging from £252.22 to £258.87.

Cost of transthoracic echocardiography

TTE has been estimated to cost £66 using HRG code RA60Z Simple Echocardiogram. 153 A second, more expensive estimate of £425 was listed for HRG code EA45Z Complex Echocardiogram (including Congenital, Transoesophageal and Fetal Echocardiography), which was deemed not appropriate for TTE. Consideration was given to the use of alternative values for the cost of TTE, such as £100 to allow for variation in cost. However, on viewing the initial results it was seen that a small change in the cost of TTE would not materially alter the cost per QALY. This was due to the main component of incremental costs being the costs associated with prescribing OACs minus any savings in the reduced numbers of stroke plus any additional costs associated with bleeding episodes.

Introduction

The long-term part of the model uses an individual-level DES approach to simulate health trajectories experienced by a large series of patients who experience competing risks of major health events.

Within a DES, an individual begins in one of a range of discrete states. They remain in this state until the ‘next event’ occurs. The next event the individual experiences, and the time they remain in their current state, are both determined by the competing risks of possible events that may occur next given the individual's current state. In this DES, the individual begins aged 60 years. They are assumed to be newly diagnosed patients with AF, and not to have previously taken OACs, and thus not to have a risk of experiencing a major bleeding event.

Extended example

Within the DES, the patient who enters the model is assigned a life expectancy using data from national life tables. 145 This produces a time to event against which other competing risks are compared. For example, for a patient aged 60 years and assigned a life expectancy of 75 years, this baseline next event is assigned a value of 15 years (75–60 years).

This value of 15 years is compared against the risk of alternative next events. The two alternative next events in this model are:

• risk of stroke

• risk of major bleeds due to an OAC.

Only if a patient is receiving an OAC do they experience a risk of major bleeds due to that OAC, and so this event will not occur in patients who are not receiving OACs. As previously discussed, taking an OAC reduces the risk of stroke, so patients not treated with an OAC have no risk of bleed but a higher risk of stroke.

In the DES, higher risks are represented by, on average, shorter times to competing next events. For example, if an event has a 20% risk of occurring per year then it has an expected time to occurrence of 5 years (1/0.2). An event with a 50% risk of occurring per year, however, has an expected time to occurrence of just 2 years (1/0.5). As events do not all occur at the expected time to occurrence, and have a range of times to occurrence around this expected time, simulated values for each of these times of events are sampled from exponential distributions parameterised by the expected time to occurrence.

Within the DES, the ‘next event’ an individual experiences is the event out of a series of candidate events with the shortest simulated time to occurrence. As a hypothetical example, consider Table 30 for a 60-year-old patient on OACs.

Out of the three candidate next events, the event with shortest time to occurrence is that of a bleed. This means that the next event the individual experiences is a bleed, and that this event occurs 12 years from the patient's current age. In the model, the candidate's profile is updated to increase their age by 12 years and their most recent event to ‘bleed’.

Assuming the bleed is non-fatal (discussed later), this updating of the patient's profile has knock-on effects for the subsequent next events too. As the individual is now 12 years older, the time to occurrence of the baseline candidate ‘death due to other causes’ has to be reduced by 12 years, from 15 years to 3 years. Having experienced a major side effect from the OAC, it is assumed the individual will no longer be prescribed the drug, as later detailed, so the risk of major bleeds is set to zero. However, in no longer being prescribed the OAC, the risk of stroke is increased. Assuming, for example, the annual risk of stroke increases as a result of this from 5% per year to 12.5% per year, the table of candidate next events that the individual (now 72 years old) could experience is as follows (Table 31).

As ‘death from other causes’ is the next event candidate with the shortest time to occurrence, it is this next event for this individual, and occurs 3 years after the previous event, at the age of 75 years.

Over the course of the 15 years from the age of 60–75 years for which this simulated individual lives, it is assumed that resources are consumed and QALYs accrued. These patterns of resource use and utility depend on the events experienced and the order they are experienced in, which is partly determined both by the patient's underlying risk of stroke and the decision made about whether or not to prescribe OACs, on the basis either of the CHADS₂-alone diagnostic strategy or the CHADS₂  + TTE diagnostic strategy.

The differences between the costs and utilities following these two diagnostic strategies are considered to result from the addition of TTE to the diagnostic package. As this is just one of a range of ways that information from TTE could improve clinical management of the patient, the estimates provided are thus partial and conservative estimates of the cost-effectiveness of TTE for this patient group.

Dynamic features of the model

Dynamic features of the model include:

• Updating the CHADS₂ score when a patient reaches the age of 75 years. This is because being aged ≥ 75 years is a risk factor within CHADS₂ , and increases the CHADS₂ score by one point. This means the annual risk of stroke increases at the age of 75 years.

• Updating the CHADS₂ score by two points when a patient has a first stroke thus resulting in an increased risk of subsequent strokes.

• If a patient suffers a major bleeding event after taking OACs, they stop being prescribed the OACs, leading to the risk of bleeds reducing to zero, but the risk of stroke increasing.

• If a patient experiences a stroke and is not already taking an OAC, they are prescribed OACs, reducing the risk of stroke but increasing the risk of bleeds assuming that the patient had not previously had a bleed event.

• The risk of a major bleeding event when taking dabigatran (150 mg twice daily) was also assumed to change at the age of 75 years.

• The life expectancy given a Glasgow Outcome Scale state 2 (GOS2) was reduced to a maximum of 3.4 years.

Determining the category of stroke

This section describes the methods used to estimate the probabilities of different mutually exclusive states following a stroke, and the costs and utilities associated with each of these states.

The outcome following a stroke is estimated using a two-stage process, using data from Rivero-Arias et al. 154 This paper154 reported that, of 1283 patients who had a stroke within the Oxford Vascular Study (OXVASC) cohort, 24.8% (319/1283) were dead within 24 months. Of those who survived, the degree of disability following the stroke was graded according to the modified Rankin Scale (mRS) 24 months after the event in 425 patients. For simplicity, this 24-month state is assumed to be the patient's permanent state until another event occurs, and the patients for whom mRS outcomes were reported were assumed to be representative of those for whom the data were not collected. The mRS has six discrete non-dead states, categorised 0–5, as shown in Table 32.

By convention, mRS states 0–2 are categorised as ‘independent’ states, and states 3–5 as ‘dependent’ states. Of those with mRS states recorded at 24 months, 74.1% of those living after a stroke were in an independent state, and 25.9% were in a dependent state, as indicated in Figure 8.

FIGURE 8.

Distribution of stroke outcomes at 24 months (survivors at 24 months only).

Uncertainty in the proportion of patients who survive a stroke was represented using a binomial distribution. As it is required for the accurate calculation of utility multipliers associated with dependent and independent states, the proportion of patients in each of the six mRS outcome states was used to parameterise a Dirichlet distribution in order to represent uncertainty in the distribution of non-dead outcome states following a stroke. These values were then converted back into estimated proportions of those alive in dependent and independent states following stroke. Results from the two-state (dead/alive) Binomial simulation and the six-state (mRS 0–6) Dirichlet simulation were used to estimate uncertainty in the proportions of patient outcomes following stroke in the three mutually exclusive categories: ‘dead’, ‘dependent stroke’, and ‘independent stroke’. The estimated proportions, together with 95% CrIs, are shown in Table 33 and graphically in Figure 9.

FIGURE 9.

The estimated distribution of patients 24 months after a stroke.

The effect of a stroke on a patient's utility

The utilities associated with independent and dependent states following strokes were estimated from the same data set used to determine the outcome following stroke. 154

Utility multipliers following a stroke were estimated from data that presented EuroQOL Five Dimension (EQ-5D) mapped utility estimates for the utility associated with each of the six mRS scores. As the mildest of these categories (mRS 0) is a full recovery, this is assumed to represent baseline patient utility. Multipliers for mRS 1–5 were thus calculated by dividing utility estimates of these worse states by the utility estimates of mRS 0. Uncertainty in both nominators and denominators were estimated using a simulation approach, with 10,000 random draws from EQ-5D estimates of each of the states mRS 1–5 divided by 10,000 random draws from the EQ-5D estimates for state mRS 0.

In order to derive estimates of the utility multiplier associated with both dependent and independent strokes, the proportions of each of the constituent mRS states within the dependent and independent stroke categories need to be estimated. Uncertainty in our knowledge of these proportions thus also needs to be represented. This is done as follows:

1. sample from a Dirichlet distribution with all six mRS states (as detailed in Introduction, above)

2. divide the six states into the independent stroke category (mRS 0–2) and dependent stroke category (mRS 3–5)

3. calculate the relative proportion of mRS states 0–2 within the independent stroke category, and relative proportion of mRS states 3–5 within the dependent stroke category

4. use weight utility multiplier estimates of mRS states 0, 1 and 2 in proportion to these states' relative prevalence within the independent stroke category, and weight utility multiplier estimates of mRS states 3, 4 and 5 in proportion to these states' relative prevalence within the dependent stroke category.

As our interest is in the mean utility multiplier for dependent and independent stroke multipliers, the mean values of 10,000 bootstraps of the distributions produced were then calculated in order to estimate both the means and uncertainty around the means. The mean utility multipliers produced are shown in Table 34.

For simplicity, it was assumed that patients who had a fatal stroke accrued no further QALYs. This is a limitation as not all patients would have died instantly; however, data that could be used to accurately populate this parameter were not identified.

Comparison with previous utility multiplier estimates

Our estimated utility multipliers are very similar to those presented by Dorman et al. 155 for independent strokes, but somewhat higher than those reported in that paper for dependent strokes. This is largely due to the distribution of mRS states within the ‘independent stroke’ and ‘dependent stroke’ categories, which for both categories of stroke are weighted towards less severe mRS states (as shown in Figure 8). In the case of dependent strokes (mRS 3–5), for example, only around 4% were the worst category mRS 5, which has an estimated EQ-5D score of around 0, and around 75% were in the least severe category mRS 3, which has an estimated EQ-5D score of > 0.5. The discrepancy may reflect improvements in the prognosis following strokes in the decade that separates the studies used.

The assumed costs following stroke

Costs following categories of events were subdivided into one-off costs, such as the cost of admission to an emergency department, and ongoing costs, such as rehabilitation costs, which are assumed to continue indefinitely.

Determining the category of major clinical bleed

Not all major clinical bleeds are the same in their consequences. The immediate outcome following a bleed is divided into three categories:

• death from major bleed

• non-fatal gastrointestinal (GI) haemorrhage

• non-fatal intracranial haemorrhage (ICH).

There is a wide variation in the effects of an ICH. To represent this variation, outcomes following non-fatal ICH are further subdivided into four distinct states according to the GOS (described later):

• GOS 2 vegetative state

• GOS 3 severely disabled

• GOS 4 moderately disabled

• GOS 5 good recovery.

The probabilities of discrete states following a bleed were calculated using a two-stage approach as described below.

Initial stage

In the model, three possible outcomes are assumed to result from a major bleeding event:

1. death from bleed

2. non-fatal GI haemorrhage

3. non-fatal ICH.

The proportions of these three events are estimated from table 79 of a 2009 HTA monograph,158 which is derived from a 2003 meta-analysis. 159 Uncertainty about the relative distribution of these three outcomes was represented using a Dirichlet distribution. These results are shown in Table 38.

If a non-fatal ICH occurs, the effect of this bleed on patient outcome is simulated using data that maps outcomes on to the GOS, which categorises a patient's state after a traumatic brain injury. 160 Uncertainty about the relative distribution of these three outcomes was represented using a Dirichlet distribution. These results are shown in Table 39 and use data in Holmes et al. 161

A GOS state of 2 is associated with a severely reduced life expectancy. This was taken into account in the model by replicating an assumption in Holmes et al. ,161 reducing the life expectancy to 3.4 years where it was otherwise expected to be greater.

Utility multiplier following a major clinical bleed

The assumed costs following a major clinical bleed

Costs following categories of events were subdivided into one-off costs (such as the cost of admission to an emergency department) and ongoing costs (such as nursing), which are assumed to continue indefinitely. Mean costs were calculated by adding together distributions from component costs associated with the event then using a bootstrapping approach to identify the mean and uncertainty around the mean of the distribution.

Fifty-year-old males, initial CHADS₂ score of 0, informing the decision whether to treat patients with warfarin

Summary results for this treatment option and patient group are shown in Table 45 and Figure 10, below. They indicate that the majority of the PSA scatterplot is in the north-west quadrant, suggesting that the addition of TTE is ruled out by simple dominance in this patient group when using warfarin as the OAC. This is confirmed in the table of results, indicating that the mean cost of conducting full TTE screening is greater than not screening, whereas the mean QALY score is lower. Summary statistics from the patient experience simulation indicate that although the number of deaths from stroke is predicted to reduce as a result of using TTE in this way, the number of deaths due to major bleeding events is predicted to increase.

FIGURE 10.

Probabilistic sensitivity analysis scatterplots, CEAFs and mean ICERs in the W_50_0_M comparison. (a) Scatterplot of difference in costs (£) against difference in QALYs. (b) CEAF (If there is no solid line in the figure, this indicates that the screening option using TTE is not optimal over all willingness-to-pay thresholds ranging from £0/QALY to £50,000/QALY.) (c) Mean costs, QALYs and ICERs.

The CEAF indicates that not screening (dashed line) remains the optimal choice compared with screening (solid line) at willingness-to-pay thresholds varying from £0/QALY to £50,000/QALY. In this case, the addition of TTE is not predicted to be the optional choice at willingness-to-pay thresholds of either £20,000/QALY or £30,000/QALY

Fifty-year-old females, initial CHADS₂ score of 0, treated with warfarin

Summary results for this treatment option and patient group are shown in Table 46 and Figure 11, below. They also indicate that the majority of the PSA scatterplot is in the north-west quadrant, suggesting that the addition of TTE is ruled out by simple dominance in this patient group when using warfarin as the OAC. This is confirmed in the table of results, indicating that the mean cost of conducting full TTE screening is greater than not screening, whereas the mean QALY score is lower. Summary statistics from the patient experience simulation indicate that although the number of deaths from stroke is predicted to reduce as a result of using TTE in this way, the number of deaths due to major bleeding events is predicted to increase.

FIGURE 11.

Probabilistic sensitivity analysis scatterplots, CEAFs and mean ICERs in the W_50_0_F comparison. (a) Scatterplot of difference in costs (£) against difference in QALYs. (b) CEAF (If there is no solid line in the figure, this indicates that the screening option using TTE is not optimal over all willingness-to-pay thresholds ranging from £0/QALY to £50,000 per QALY.) (c) Mean costs, QALYs and ICERs.

The CEAF indicates that not screening (dashed line) remains the optimal choice compared with screening (solid line) at willingness-to-pay thresholds varying from £0/QALY to £50,000/QALY. In this case, the addition of TTE is not predicted to be the optional choice at willingness-to-pay thresholds of either £20,000/QALY or £30,000/QALY.

Sixty-five-year-old males, initial CHADS₂ score of 0, treated with warfarin

Summary results for this treatment option and patient group are shown in Table 47 and Figure 12. The PSA scatterplot suggests that although the use of TTE is clearly associated with increased costs, it is not associated with substantially increased predicted QALYs. This is confirmed by the table in Figure 12c, which shows the TTE strategy to cost almost £1000 more than the no-TTE strategy, but to result in almost no increase in QALYs. Summary statistics from the patient experience simulation indicate that although the number of deaths from stroke is predicted to reduce as a result of using TTE in this way, the number of deaths due to major bleeding events is predicted to increase.

FIGURE 12.

Probabilistic sensitivity analysis scatterplots, CEAFs and mean ICERs in the W_65_0_M comparison. (a) Scatterplot of difference in costs (£) against difference in QALYs. (b) CEAF (If there is no solid line in the figure, this indicates that the screening option using TTE is not optimal over all willingness-to-pay thresholds ranging from £0/QALY to £50,000 per QALY.) (c) Mean costs, QALYs and ICERs.

The CEAF indicates that not screening (dashed line) remains the optimal choice compared with screening (solid line) at willingness-to-pay thresholds varying from £0/QALY to £50,000/QALY. In this case, the addition of TTE is not predicted to be the optional choice at willingness-to-pay thresholds of either £20,000/QALY or £30,000/QALY.

Sixty-five-year-old females, initial CHADS₂ score of 0, treated with warfarin

Summary results for this treatment option and patient group are shown in Table 48 and Figure 13. The PSA scatter indicates that slightly more of the estimates were in the north-east than the north-west quadrant. The table in part (c) of Figure 13 shows that the differences in mean costs between strategies are around £1000, but the differences in mean QALYs are less than one-tenth of a QALY. The mean ICER is around £40,000/QALY, and the CEAF indicates that the TTE strategy (solid line) is unlikely to be the optimal decision at standard NICE willingness-to-pay thresholds of £20,000/QALY and £30,000/QALY.

FIGURE 13.

Probabilistic sensitivity analysis scatterplots, CEAFs and mean ICERs in the W_65_0_F comparison. (a) Scatterplot of difference in costs (£) against difference in QALYs. (b) CEAF. (c) Mean costs, QALYs and ICERs.

Fifty-year-old males, initial CHADS₂ score of 1, treated with warfarin

In this patient group and treatment option (Table 49) the mathematical model shows that the majority of the estimates from the PSA (as shown in Figure 14a) are in the north-east quadrant, indicating that the TTE strategy is both more costly and more effective than the no-TTE strategy. There is a difference in mean QALYs between the two strategies of approximately half a QALY (the table shown in Figure 14c) and difference in mean costs of approximately £3000. The mean ICER is estimated to be slightly over £6000/QALY, and the CEAF indicates that, compared with the no-TTE strategy (dashed line), the TTE strategy (solid line) appears optimal at both the £20,000/QALY and £30,000/QALY willingness-to-pay thresholds; at both thresholds the probability that the TTE strategy is more cost-effective is estimated to be > 99%.

FIGURE 14.

Probabilistic sensitivity analysis scatterplots, CEAFs and mean ICERs in the W_50_1_M comparison. (a) Scatterplot of difference in costs (£) against difference in QALYs. (b) CEAF. (c) Mean costs, QALYs and ICERs.

Fifty-year-old females, initial CHADS₂ score of 1, treated with warfarin

In this patient group and treatment option (Table 50) the mathematical model shows that the majority of the estimates from the PSA (as shown in Figure 15a) are in the north-east quadrant, indicating that the TTE strategy is both more costly and more effective than the no-TTE strategy. There is a difference in mean QALYS between the two strategies of approximately half a QALY (the table in Figure 15c) and a difference in mean costs of approximately £3000 . The mean ICER is estimated to be slightly over £7000/QALY, and the CEAF indicates that, compared with the no-TTE strategy (dashed line), the TTE strategy (solid line) appears optimal at both the £20,000/QALY and £30,000/QALY willingness-to-pay thresholds; at both thresholds the probability that the TTE strategy is more cost-effective is estimated to be > 99%.

FIGURE 15.

Probabilistic sensitivity analysis scatterplots, CEAFs and mean ICERs in the W_50_1_F comparison. (a) Scatterplot of difference in costs (£) against difference in QALYs. (b) CEAF. (c) Mean costs, QALYs and ICERs.

Sixty-five-year-old males, initial CHADS₂ score of 1, treated with warfarin

In this patient group and treatment option (Table 51) the mathematical model shows that the majority of the estimates from the PSA (as shown in Figure 16a) are in the north-east quadrant, indicating that the TTE strategy is both more costly and more effective than the no-TTE strategy. There is a difference in mean QALYs between the two strategies (the table in Figure 16c) of approximately one-fifth of a QALY and a difference in mean costs of approximately £2500. The mean ICER is estimated to be around £11,000/QALY, and the CEAF indicates that, compared with the no-TTE strategy (dashed line), the TTE strategy (solid line) appears optimal at both the £20,000/QALY (93% probability most cost-effective) and £30,000/QALY (98.5% probability most cost-effective) willingness-to-pay thresholds.

FIGURE 16.

Probabilistic sensitivity analysis scatterplots, CEAFs and mean ICERs in the W_65_1_M comparison. (a) Scatterplot of difference in costs (£) against difference in QALYs. (b) CEAF. (c) Mean costs, QALYs and ICERs.

Sixty-five-year-old females, initial CHADS₂ score of 1, treated with warfarin

In this patient group and treatment option (Table 52) the mathematical model the majority of the estimates from the PSA (as shown in Figure 17a) are in the north-east quadrant, indicating that the TTE strategy is both more costly and more effective than the no-TTE strategy. There is a difference in mean QALYs between the two strategies of approximately one-fifth of a QALY (the table in Figure 17c) and a difference in mean costs of approximately £4000. The mean ICER is estimated to be slightly < £15,000/QALY, and the CEAF indicates that, compared with the no-TTE strategy (dashed line), the TTE strategy (solid line) appears optimal at both the £20,000/QALY (73% probability most cost-effective) and £30,000/QALY (92% probability most cost-effective) willingness-to-pay thresholds.

FIGURE 17.

Probabilistic sensitivity analysis scatterplots, CEAFs and mean ICERs in the W_65_1_F comparison. (a) Scatterplot of difference in costs (£) against difference in QALYs. (b) CEAF. (c) Mean costs, QALYs and ICERs.

Fifty-year-old males, initial CHADS₂ score of 0, treated with rivaroxaban

In this patient group and treatment option (Table 53) the mathematical model slightly more of the estimates from the PSA are in the north-west than north-east quadrant, implying that the TTE strategy is likely to be ruled out by simple dominance compared with the no-TTE strategy. This is confirmed by the mean values, which indicate that the TTE strategy is around £2000 more expensive and slightly less effective than the no-TTE strategy. The CEAF (Figure 18) indicates that the TTE strategy (solid line) does not appear to be the optimal strategy at all willingness-to-pay thresholds from £0/QALY to £50,000/QALY compared with the no-TTE strategy (dashed line).

FIGURE 18.

Probabilistic sensitivity analysis scatterplots, CEAFs and mean ICERs in the R_50_0_M comparison. (a) Scatterplot of difference in costs (£) against difference in QALYs. (b) CEAF (If there is no solid line in the figure, this indicates that the screening option using TTE is not optimal over all willingness-to-pay thresholds ranging from £0/QALY to £50,000 per QALY.) (c) Mean costs, QALYs and ICERs.

Fifty-year-old females, initial CHADS₂ score of 0, treated with rivaroxaban

In this patient group and treatment option (Table 54) the mathematical model slightly more of the estimates from the PSA are in the north-west than north-east quadrant, implying that the TTE strategy is likely to be ruled out by simple dominance compared with the no-TTE strategy. This is confirmed by the mean values, which indicate that the TTE strategy is around £2500 more expensive and slightly less effective than the no-TTE strategy. The CEAF (Figure 19) indicates that the TTE strategy (solid line) does not appear to be the optimal strategy at all willingness-to-pay thresholds from £0/QALY to £50,000/QALY compared with the no-TTE strategy (dashed line).

FIGURE 19.

Probabilistic sensitivity analysis scatterplots, CEAFs and mean ICERs in the R_50_0_F comparison. (a) Scatterplot of difference in costs (£) against difference in QALYs. (b) CEAF (If there is no solid line in the figure, this indicates that the screening option using TTE is not optimal over all willingness-to-pay thresholds ranging from £0/QALY to £50,000 per QALY.) (c) Mean costs, QALYs and ICERs.

Sixty-five-year-old males, initial CHADS₂ score of 0, treated with rivaroxaban

For this patient group and treatment combination (Table 55), the majority of the PSA scatterplot (Figure 20a) is either in the north-east or north-west quadrant. Slightly more of the scatterplot appears to be in the north-east quadrant, and the TTE strategy has both a greater mean cost and greater mean QALY estimate than the no-TTE strategy. The mean ICER is slightly over £30,000, and the CEAF (see Figure 20b) indicates that the TTE strategy is still not the optimal strategy compared with no TTE at the £30,000/QALY threshold; the TTE strategy is estimated to become optimal only at around £33,700/QALY.

FIGURE 20.

Probabilistic sensitivity analysis scatterplots, CEAFs and mean ICERs in the R_65_0_M comparison. (a) Scatterplot of difference in costs (£) against difference in QALYs. (b) CEAF. (c) Mean costs, QALYs and ICERs.

Sixty-five-year-old females, initial CHADS₂ score of 0, treated with rivaroxaban

The mathematical model results (Table 56) suggest that the TTE strategy is more expensive and slightly more clinically effective, on the average, than the no-TTE strategy, with the PSA scatterplot appearing slightly more predominant in the north-east than the north-west quadrant. The mean ICER estimate is slightly > £20,000, and the CEAF indicates that the TTE strategy starts to become optimal at a willingness-to-pay threshold of approximately £21,000/QALY (Figure 21). If the willingness-to-pay threshold is £30,000 then the TTE strategy is estimated to have a 55% probability of being most cost-effective.

FIGURE 21.

Probabilistic sensitivity analysis scatterplots, CEAFs and mean ICERs in the R_65_0_F comparison. (a) Scatterplot of difference in costs (£) against difference in QALYs. (b) CEAF. (c) Mean costs, QALYs and ICERs.

Sixty-five-year-old males, initial CHADS₂ score of 0, treated with dabigatran

The mathematical model results (Table 57) suggest that the TTE strategy is more expensive and slightly more clinically effective, on the average, than the no-TTE strategy, with the PSA scatterplot appearing slightly more predominant in the north-east than the north-west quadrant. The mean ICER estimate is slightly < £15,000, and the CEAF indicates that the TTE strategy starts to become optimal at a willingness-to-pay threshold of approximately £13,800/QALY (Figure 22). If the willingness-to-pay threshold is £20,000, then the TTE strategy is estimated to have a 57% probability of being most cost-effective, and with a willingness-to-pay threshold of £30,000 the TTE strategy has a 65% probability of being most cost-effective.

FIGURE 22.

Probabilistic sensitivity analysis scatterplots, CEAFs and mean ICERs in the D_65_0_M comparison. (a) Scatterplot of difference in costs (£) against difference in QALYs. (b) CEAF. (c) Mean costs, QALYs and ICERs.

Sixty-five-year-old females, initial CHADS₂ score of 0, treated with dabigatran

The mathematical model results (Table 58) suggest that the TTE strategy is more expensive and slightly more clinically effective, on the average, than the no-TTE strategy, with the PSA scatterplot appearing slightly more predominant in the north-east quadrant than in the north-west quadrant. The mean ICER estimate is slightly over £12,000, and the CEAF (Figure 23) indicates that the TTE strategy starts to become optimal at a willingness-to-pay threshold of approximately £11,800/QALY. If the willingness-to-pay threshold is £20,000 then the TTE strategy is estimated to have a 64% probability of being most cost-effective, and with a willingness-to-pay threshold of £30,000 the TTE strategy has a 72% probability of being most cost-effective.

FIGURE 23.

Probabilistic sensitivity analysis scatterplots, CEAFs and mean ICERs in the D_65_0_F comparison. (a) Scatterplot of difference in costs (£) against difference in QALYs. (b) CEAF. (c) Mean costs, QALYs and ICERs.

Sensitivity of incremental cost-effectiveness ratios to joint sensitivity and specificity of transthoracic echocardiography, and true prevalence of left atrial abnormality

In order to explore the influence of certain parameters on the cost-effectiveness estimates, the effects of changing the sensitivity and specificity estimates, holding all other values at their means, were calculated for each of the 14 comparisons. These are presented in Appendix 11.

Additionally, the influence of difference assumptions about the true proportion with LA abnormality was estimated in a similar way. These are presented in Figure 25 for people aged 50 years with a CHADS₂ score of 0, Figure 26 for people aged 65 years with a CHADS₂ score of 0, and in Figure 27 for people aged either 50 or 65 years with a CHADS₂ score of 1.

FIGURE 25.

Effect of true proportion of population with LA abnormality on estimated ICER in 50-year-olds who have a CHADS₂ score of 0.

FIGURE 26.

Effect of true proportion of population with LA abnormality on estimated ICER in 65-year-olds who have a CHADS₂ score of 0.

FIGURE 27.

Effect of true proportion of population with LA abnormality on estimated ICER in people aged either 50 or 65 years who have a CHADS₂ score of 1.

The figures indicate that, over the range of values considered here, the true prevalence of the LA abnormality could have a significant influence on the ICER in people aged 65 years with a CHADS₂ score of 0, suggesting that identifying the true value of this parameter in these patient populations may be more important than in 50-year-olds or people with a CHADS₂ score of 1. Among people with a CHADS₂ score of 1, it may be more valuable to identify the true value of this parameter in females than in males.

Search strategy

A comprehensive literature search was undertaken across five databases: MEDLINE, EMBASE, Cumulative Index to Nursing and Allied Health Literature (CINAHL), WoS, and Cochrane Database of Systematic Reviews (CDSR). Details of the full search strategy are provided in Appendix 12.

Inclusion/exclusion criteria

The inclusion criteria for this systematic review were as follows:

• Population Patients with AF.

• Intervention Transthoracic echocardiogram.

• Comparators Conventional therapy.

• Setting Interventions delivered within any geographical jurisdiction.

• Outcomes Cost per QALY.

• Study designs Studies reporting a full economic evaluation, with results expressed in terms of both costs and health outcomes.

Studies were excluded from this review if:

• the patients suffered from cardiac problems other than AF

• the intervention was not transthoracic echocardiogram

• they reported only costs or outcomes

• they were not of full economic evaluation, such as cost-minimisation analysis

• they were not published papers, such as editorials, commentaries and letters

• they were not published in the English language.

All of the potentially relevant citations were imported to Reference Manager Software, version 12 (Thomson ResearchSoft, San Francisco, CA, USA) and duplicates were removed. The titles and abstracts of the unique studies were then screened according to the predetermined inclusion criteria as outlined above. Any disagreements concerning possible inclusion of papers were resolved by discussion among the researchers of the team, or through retrieval and subsequent examination of the full study publication. Full papers of all the potentially relevant citations were retrieved for an in-depth assessment concerning study inclusion in the review.

Data extraction and evidence synthesis

Data concerning the characteristics of the population, interventions, comparators, outcomes, study location, time horizon, costs, outcomes and the perspective of the evaluations undertaken were extracted from the included study. These were then tabulated and discussed in a narrative manner.

Number of studies identified and included

In total, 1038 studies were identified through the systematic searches. Following the removal of duplicate citations, 881 unique studies were retrieved. Of these, 43 potentially relevant citations were retrieved for a more detailed inspection. Further, 26 studies were excluded and 17 studies were retrieved for double screening. After double screening, 17 studies were excluded as they failed to satisfy one or more of the inclusion criteria.