Cost-effectiveness findings

Table 71 shows the mean cost and QALY estimates associated with each strategy for each of the nine scenarios defined above. In each case, the more costly strategy is compared with the next less costly alternative to estimate the incremental costs and incremental QALYs. Where strategies are more costly and less effective an alternative strategy, they are said to be dominated. Where a strategy is more costly and more effective than its comparator, its incremental cost-effectiveness ratio (ICER) is reported. This represents the additional cost incurred per QALY gained over the next less costly alternative.

TABLE 71

Cost-effectiveness analyses results (based on a 20-year time horizon)

The base case considers the cost-effectiveness of immediate MRI plus ultrasound for patients presenting at stroke prevention clinics within a few hours of symptom onset compared with a policy of immediate CT imaging plus ultrasound. The results of this analysis indicate that a strategy of immediate MRI and ultrasound is very unlikely to be considered cost-effective in the immediate presenting cohort. It works out to be more expensive, primarily because the cost of MRI is higher than the cost of CT imaging and, if anything, slightly less effective. The small difference in effectiveness is caused by a slightly greater proportion of patients with haemorrhagic stroke being missed with MRI in the early presenting cohort, and therefore receiving inappropriate antiplatelet treatment. This is a consequence of the lower sensitivity of MRI for haemorrhagic stroke when used within a few hours of the event. At the same time, as individuals without any visible lesion detected on imaging are modelled to receive secondary prevention for ischaemic stroke, and ultrasound is used to detect carotid stenosis in both arms of the model, there is no benefit associated with the higher sensitivity of MRI for identifying ischaemic stroke. Hence the estimated increase in cost of £77 and the marginal QALY loss of 0.0005.

In scenario 2, MRI with ultrasound is again compared with immediate CT imaging with ultrasound. However, in this scenario selective imaging is used in the MRI arm, with MRI and ultrasound being delayed by 7 days for patients who have a low ABCD score; i.e. a score of < 4. This strategy also turns out to be less cost-effective than the CT option and also less cost-effective than when immediate MRI and ultrasound is carried out for all. The incremental costs are higher (£126) and the QALY losses greater (0.0191) with this selective delayed MRI strategy as a significant proportion of patients with a low ABCD2 score still have high or moderate levels of carotid stenosis, which without timely diagnosis and surgery puts them at high risk of suffering a recurrent stroke. The effect of delaying diagnosis and surgery for these patients translates into higher mean incremental costs and QALY losses.

In strategy 3, instead of delaying both ultrasound and MRI by 7 days, the strategy tests the effect of delaying only imaging with MRI for those with a low ABCD2 score (i.e. < 4). This strategy still costs more than the CT base strategy and there is still a loss of QALYs, although to a lesser extent than with strategy 2. However, the strategy is still dominated by CT plus ultrasound for all, i.e. it is more costly and no more effective, and it is therefore less cost-effective. This is a consequence of patients with a low ABCD2 score, but no stenosis, still having some degree risk for recurrent stroke that would benefit from immediate secondary preventative treatment.

A further option for the immediate presenting cohort is considered in scenario 4. Here we simultaneously compare four potential immediate imaging strategies: (1) CT for brain imaging plus ultrasound for carotid imaging; (2) CT plus CTA for carotid imaging; (3) MRI plus ultrasound for carotid imaging; and (4) MRI plus MRA for carotid imaging. The results for this scenario indicate that CT plus ultrasound dominates all of the other strategies. These results are based on the assumption that neither CTA nor MRA is any more sensitive for picking up carotid stenosis than ultrasound,^83,84 whereas both are more costly.

Scenario 5 explores the impact of assuming secondary preventative drug treatment for ischaemic stroke is implemented only for patients with an identified ischaemic lesion on DWI in the MRI arm of the model (whereas treatment decisions remain unchanged from the base case in the CT arm). Under this scenario, the MRI strategy is less costly than it otherwise is owing to saving on the cost of treatment, but it is also significantly less effective owing to an increased incidence of recurrent stroke in the 66% of TIAs and 30% of minor strokes patients with no visible ischaemic lesion on DWI. It remains dominated by the CT strategy.

A final scenario explored for the immediate presenting cohort, is one where it is assumed that preventative drug treatment for ischaemic stroke is implemented for all patients upon presentation, but that brain and carotid imaging are delayed for the group with a low ABCD2 score in the MRI arm of the model (scenario 6). CT plus ultrasound remains the favoured strategy under this scenario.

Strategies 7–9 assess the cost-effectiveness of MRI strategies compared with immediate CT plus ultrasound in cohorts of patients presenting 7 days or 21 days after the initial event. Considering immediate MRI plus ultrasound compared with immediate CT plus ultrasound in the cohort presenting at 7 days (scenario 7), the MRI strategy turns out to be more costly but also slightly more effective than CT, with an ICER of £39,509 per QALY. The increase in effect observed for MRI under this scenario results from its superior sensitivity for haemorrhagic stroke (the sensitivity of CT for haemorrhages declines over time), whereas the sensitivity of MRI for ischaemic stroke also declines over time, this has no impact on outcomes as treatment for ischaemic stroke is still modelled to be implemented in this scenario unless an alternative cause of symptoms is identified. Similar results are obtained when the same strategies are compared in a cohort of patients presenting 21 days after the initial event. However, the QALY gains associated with the MRI strategy are higher in this case as a result of a further decline in the sensitivity of CT for haemorrhagic stroke and, as such, the ICER for MRI plus ultrasound is improved (£27,751 per QALY gained).

A final scenario explored for the cohort of patients presenting 7 days after the initial event was a repeat of scenario 3: delayed MRI for those with a low ABCD2 score but immediate ultrasound for all compared with immediate CT plus immediate ultrasound for all (scenario 9). Under this scenario the QALYs are again higher with MRI but less so than when immediate MRI is implemented for patients presenting at 7 days. This is, again, due to some patients with a low ABCD2 score and no carotid stenosis still being exposed to a risk of ischaemic stroke that would benefit from appropriate treatment. In this instance, however, the point estimate for the ICER is £159,325, which is well above the usually acceptable range of £20,000–30,000 per QALY gained.

The QALY differences shown in Table 71 are primarily driven by small changes in the relative incidence of recurrent stroke between the arms of the models. To illustrate this, Table 72 presents the modelled cumulative incidence of recurrent stroke in the MRI and CT arms of the model for scenarios 1, 2, 3, 5, 7 and 8 over the 20-year time horizon. In each instance, the strategy associated with the QALY gains is also associated with a lower cumulative recurrence of stroke. In the base case of immediate CT and ultrasound compared with immediate MRI and ultrasound (in the immediate presenting cohort), for example, the difference in recurrent stroke incidence between the strategies is very small, and would equate to an increase of one more haemorrhagic stroke for every 12,500 patients imaged using the MRI strategy (1/0.00007976).

TABLE 72

Modelled cumulative incidence of stroke recurrence over a 20-year time horizon, expressed as the proportion of the whole cohort (including those with a mimic in whom recurrent stroke is not explicitly modelled)

Uncertainty around cost of imaging

First of all we varied the cost of MRI to examine the effect of changes in this parameter on the results. Immediate MRI and ultrasound remains dominated by immediate CT plus ultrasound unless we assume that the cost of MRI is less than that of CT (e.g. £90). Although the MRI strategy is less costly than the CT strategy at this price (holding the CT cost constant), it also remains marginally less effective. However, CT remains relatively cost-effective in comparison with MRI, with an ICER of approximately £20,000 per QALY gained.

Uncertainty around the sensitivity of computed tomography in detecting visible haemorrhagic lesions

Lowering the sensitivity of CT for identifying haemorrhages in the immediate presenting cohort, the MRI plus ultrasound option provides a greater number of QALYs compared with CT plus ultrasound. The declining sensitivity of CT for haemorrhage over time is responsible for the improved cost-effectiveness of MRI compared with CT in later presenting cohorts (see Table 71, scenarios 7 and 8). Here we demonstrate how the cost-effectiveness of MRI improves in relation to CT if the sensitivity of CT for haemorrhage drops in a more immediate presenting cohort (i.e. within a few hours of the initial event).

Uncertainty around the increased risk of recurrence of haemorrhagic stroke when incorrect treatment with antiplatelet therapy is given

With a reduced risk to 1.05 (the lower 95% CI on the risk of recurrent haemorrhage in patients with haemorrhagic stroke treated with antiplatelet agents), the option of immediate MRI and ultrasound is still dominated; however, there is a lower loss in QALYs than in the base case of immediate CT and ultrasound. The opposite effect occurs with an increase in the hazard rate; a higher QALY loss is observed.

Table 74 presents the results of one-way sensitivity analysis on several key variables for the comparison of immediate MRI plus ultrasound compared with immediate CT plus ultrasound in the later presenting (day 7) cohort of patients. The results indicate that the cost-effectiveness of the MRI strategy is quite sensitive to the cost of obtaining the MRI scan relative to the cost of CT, and also the relative increased risk of recurrent haemorrhagic stroke in patients with a minor haemorrhagic stroke treated with antiplatelet medication. When the cost of MRI is reduced or the RR associated with inappropriate antiplatelet medication is increased, the cost-effectiveness of MRI improves in the cohort of patients presenting at 7 days. The relative cost-effectiveness of alternative strategies is much less sensitive to the costs of recurrent stroke owing to only very small differences in the overall cumulative incidence of stroke between strategies. Cost-effectiveness also improves slightly when running the analysis over a period of 30 years.

TABLE 74

Sensitivity analysis surrounding key parameters that influence the cost-effectiveness of immediate MRI plus ultrasound vs. immediate CT plus ultrasound in a cohort of patients presenting at 7 days after the initial event

Statement of principal findings

Magnetic resonance imaging strategies are generally found to be more expensive and no more effective than CT. The exceptions to this are for those strategies that involve later presenting cohorts. Here, MRI becomes more cost-effective, and its cost-effectiveness improves as its sensitivity for detecting haemorrhages improves over time, although its ICER falls only within, and does not fall below, conventionally accepted cost-effectiveness thresholds ranges (£20,000–30,000 per QALY) in the cohort presenting 21 day after the initial event. The cost-effectiveness of MRI in late-presenting cohorts (7 and 21 days) was found to be sensitive to fairly small changes in the cost of MRI relative to the cost of CT, and also the assumed relative increased risk of recurrent haemorrhagic stroke associated with inappropriate antiplatelet treatment of those patients who have suffered a minor haemorrhage. It should also be noted here that, although we present findings for the cost-effectiveness of MRI compared with CT at 7 days and 21 days, this was primarily due to available evidence suggesting that CT has high sensitivity for haemorrhagic stroke up to 7 days, and significantly lower sensitivity from day 7 onwards. However, if the sensitivity of CT for haemorrhage drops off more quickly, MRI may become the more effective strategy earlier than 7 days after the initial event.

Strengths and limitations of the economic model

A strength of the model is that it offers an explicit framework for the synthesis of both evidence and expert opinion. It uses structures developed in two previous HTA cost-effectiveness analyses that combine assessment of brain imaging (cross-sectional diagnostic decision tree approach) with risks of stroke (variable time-dependent approach), the former developed to assess the cost-effectiveness of brain imaging in acute stroke²⁶ and the latter to assess the cost-effectiveness of carotid imaging in stroke prevention.⁵ From this combined approach, gaps in the evidence base are revealed. There is good evidence from high-quality large RCTs on the effectiveness of different forms of medical and surgical management on the probability of stroke recurrence and by extrapolation on health-related quality of life. The main gaps appear to be in relation to the relative sensitivity and specificity of MRI and CT in TIA and minor stroke, and the impact these imaging modalities have on subsequent treatment and outcomes; the available evidence base is weighted towards sensitivity and specificity in moderate to severe stroke patients, and so, as discussed in earlier chapters, some assumptions have been made with respect to how these apply to TIA and minor stroke.

One potential limitation of the modelling strategy relates to the assumption regarding treatment of patients following information provided by imaging results. In the modelled base-case scenario, the sensitivity of MRI for detecting an ischaemic lesion is 0.63, whereas for CT it is 0.39. However, it is assumed that implementation of preventative treatment (for recurrent stroke) is based on clinical assessment as well as imaging, and not just the results from imaging in isolation, as is the case in clinical practice. Where either CT or MRI is unable to rule out a minor ischaemic stroke or TIA (i.e. by identifying a haemorrhage or mimic), we have assumed that implementation of secondary prevention for ischaemic stroke will prevail based on clinical assessment in patients without a visible ischaemic lesion. This is current practice. Further, we have assumed that CT and MRI, when combined with clinical assessment, are essentially equivalent in their ability to appropriately identify and rule out mimics, and that they do not result in different numbers of patients with true ischaemic stroke being misdiagnosed as mimics. The available evidence on the contribution of CT or MR to establishing a diagnosis in most of the mimics is not good but, on the other hand, there is information on the effect of MR in minor stroke on treatment decisions and on detection rates of positive findings that confirm the type of mimic in many types of mimics. There were no data on which to base other approaches, as mimics were excluded from all studies of DWI in TIA/minor stroke (see Chapter 6). It could be argued that the assumptions used may underplay the potential advantage that MRI offers through being better able to detect ischaemic lesions and therefore differentiate some true TIAs/minor strokes with a DWI lesion (one-third/two-thirds) from the two-thirds of TIAs/one-third of minor strokes without a DWI lesion and the mimics that make up 45% of clinic attendances in most of whom DWI would be negative. However, some mimics can produce a DWI-positive finding (migraine, hypoglycaemia, postictal, MS, etc.), so the impact on decision-making is not transparent. Differentiating these categories essentially boils down to clinical expertise with careful history and examination backed up by imaging. Thus, it is not clear whether that added advantage changes treatment decisions, and previous studies suggested that the differences conferred by MR DWI and T2* were small (< 10%),²⁵² no treatment differences were modelled. It remains possible that in settings where clinical assessments and further tests are less expertly applied – and particularly where there was too much reliance on ABCD2 score-based filtering – the use of CT would result in more patients with a true ischaemic stroke or TIA (e.g. those with low ABCD2 scores) being misclassified as a mimic or haemorrhage, which could end up delaying appropriate preventative treatment for these patients. On the other hand, if MRI were to be relied on in place of expert opinion, it would only help sift out the one-third of TIAs and two-thirds of minor strokes with a DWI-positive lesion, leaving the two-thirds of TIAs/one-third of minor strokes without a DWI lesion, but still with a high risk of recurrent stroke, with no protection from stroke prevention therapy. Additionally, basing treatment decision more heavily on the findings of MRI, could result in more patients with a mimic being misclassified as an ischaemic stroke, or with an ischaemic event failing to receive treatment, which could lead to unnecessary or missed ischaemic stroke treatment and delayed diagnosis and appropriate treatment for the true underlying problem. These scenarios have been modelled by using ABCD2 score to triage the rapidity of imaging assessment and also by assuming that only patients with a DWI-positive lesion would receive stroke prevention treatment, neither of which was cost-effective.

Another issue that is problematic relates to estimation of the proportion of patients already on antiplatelet therapy and other medical treatment. HR data from randomised trials will probably overestimate potential benefits, as many patients who arrive with TIA or stroke may well already be on such therapy. The model may therefore overestimate the benefits of diagnosing and treating ischaemic stroke, although such an effect comes into play only in scenarios in which it is assumed that treatment is not implemented for those patients with no visible ischaemic lesion on MRI. In fact, as the UK survey and other sources²⁹⁸ indicate that many patients are started on stroke prevention treatment before they reach the clinic, and indeed this is the recommendation in the NICE guidelines,⁹⁷ the benefit of MRI becomes even less and the relative cost even worse. Furthermore, the survey and other data indicate that MRI is often used in addition to CT, further increasing costs, burdening imaging services and reducing the incremental effectiveness of MRI.

A further potential limitation is that health-related quality of life has been aggregated into three broad categories: (1) alive and free from stroke; (2) alive and independent following a recurrent stroke; (3) and alive and dependent following a recurrent stroke. This makes the model insensitive to changes in health-related quality of life within particular categories; however, as the focus of the model relates to the contribution of imaging to diagnostic accuracy and subsequent prevention of recurrent stroke, these broad categories should be sufficient to capture the average expected impact of strategies on QALYs.

On the costs side, there is wide disparity in the available estimates for the costs of imaging. What is important for cost-effectiveness, however, is the size of the difference in costs between CT and MRI, rather than whether or not the exact values applied are too high or too low. In this case, there is a good deal of evidence available to suggest that MRI uses more health-care resources than CT; in addition to its greater acquisition price, the test takes longer to perform and it uses more consumables and materials. The important issue then is whether the cost increase is worth incurring in light of any expected health benefits or downstream cost savings. Taking the cost of a medical consultant as £146 per patient-hour,³⁹⁴ an extra 20 minutes of a consultant radiologist time costs approximately £60. It is possible then that the £70 difference is an underestimate. The effect of a larger and smaller difference, however, is tested in sensitivity analyses. In particular, the relative cost of MRI in comparison with CT is a key determinant of the cost-effectiveness of MRI in later-presenting cohorts.