5.1 The sponsor submitted a de novo analysis of the costs and consequences of using WatchBP Home A, compared with pulse palpation by a nurse or GP, to detect irregular pulse in people with suspected or existing hypertension or those being screened for hypertension. The model was based in primary care and used estimates of diagnostic accuracy from published evidence. Full details of all cost evidence and modelling considered by the Committee are available in the assessment report overview.
5.2 The main consequences included in the model were: referral for confirmation of atrial fibrillation with consultant-led 12-lead ECG, use of anticoagulant drugs and aspirin, adverse effects of anticoagulants and aspirin, and number of strokes prevented. The time horizon of the model was 1 year.
5.3 The key assumptions used in the model were:
New cases of atrial fibrillation would occur at a fixed number (87,000) each year based on a published annual incidence rate.
Each of these people would be screened for atrial fibrillation.
Atrial fibrillation was related to an annual 4% risk of having a stroke.
56% of patients diagnosed with atrial fibrillation would be prescribed anticoagulants and 32% would be prescribed aspirin.
2.4% of all patients prescribed anticoagulants would have major bleeds and 15.8% would have minor bleeds.
For every 5.7 patients detected as having an irregular pulse, only 1 would be confirmed as having atrial fibrillation by ECG.
5.4 The diagnostic accuracy of WatchBP Home A was estimated from Wiesel et al. (2009) (96.8% sensitivity and 88.8% specificity) and Stergiou et al. (2009) (100% sensitivity and 89% specificity). The diagnostic accuracy of pulse palpation was estimated from Morgan et al. (2002) (91% sensitivity and 74% specificity) and Hobbs et al. (2005) (87.2% sensitivity and 81.3% specificity). The clinical variable estimates used in the model were derived from the costing report from the NICE clinical guideline on atrial fibrillation. These were: the probability of starting anticoagulant or antiplatelet drugs, absolute risk reduction of having a stroke if on anticoagulant or antiplatelet therapy, and the probability of a minor bleed or a major bleed.
5.5 The cost of an ECG was estimated to be £36.03, taken from the costing report from the NICE clinical guideline on atrial fibrillation, adjusted for inflation (5%). The cost of pulse palpation was estimated to be £2.32, derived from Hobbs et al. (2005) and adjusted for inflation. The capital cost of the WatchBP Home A device was not included in the analysis because the sponsor assumed that GPs would need to purchase a blood pressure monitor for routine practice.
5.6 The base-case analysis showed that in primary care the WatchBP Home A device would lead to an annual saving of £9,165,000 for the NHS in England and Wales by displacing pulse palpation, based on the assumption that all people with atrial fibrillation were symptomatic and that the cost of stroke was £9906 per person. The WatchBP Home A device was associated with the prevention of 221 strokes, freeing NHS resources equivalent to £2,289,000.
5.7 A deterministic sensitivity analysis was carried out by the sponsor testing the proportion of people with atrial fibrillation who were symptomatic (100%, 65% or 50%) and varying the cost of stroke (£44,000 rather than £9906). The sensitivity analysis demonstrated that increasing the proportion of asymptomatic patients would lead to a reduced saving when using the WatchBP Home A device. If 50% of patients were assumed to be asymptomatic, using the device would incur costs. Conversely, if the more expensive estimate of stroke costs was used, using WatchBP Home A would generate further savings.
5.8 The External Assessment Centre carried out a multivariate sensitivity analysis to examine the impact of changing the following parameters:
prevalence of atrial fibrillation in the study population
sensitivity and specificity of pulse palpation and the WatchBP Home A device
including and excluding the cost of pulse palpation.
5.9 The sensitivity analyses showed that excluding the cost of time associated with pulse palpation did not make pulse palpation less costly than the WatchBP Home A in any single scenario. In addition, the worst-case sensitivity and specificity values increased costs in all cases; similarly, the best-case sensitivity and specificity values decreased costs in all cases compared with the base-case sensitivity and specificity. Total cost was highly sensitive to the prevalence of atrial fibrillation, with costs for populations with a higher prevalence higher than for those with a lower prevalence, both for the WatchBP Home A device and for pulse palpation.
5.10 The sponsor's de novo cost analysis in primary care was limited to a 1-year time horizon and did not include device costs. In its initial review of the cost analysis, the Committee requested further modelling of the longer-term impact of the device on the treatment pathway. The External Assessment Centre therefore carried out additional modelling to establish the longer-term cost impact of using WatchBP Home A in primary care for the opportunistic detection of atrial fibrillation during routine blood pressure measurement. This model also aimed to address uncertainties surrounding the costs of stroke present in the sponsor's original analysis.
5.11 In the External Assessment Centre's additional model, the comparator was defined as blood pressure measurement using a manual or automated sphygmomanometer, with manual pulse palpation to detect atrial fibrillation. A decision tree was used to model the costs associated with the diagnostic accuracy of the device and pulse palpation compared with 12-lead ECG, in terms of true or false positives or negatives. This was accompanied by a state transition model to simulate 10-year cost consequences of managing identified atrial fibrillation (true positives) in a simulated cohort of patients. Outcomes were also modelled for those with atrial fibrillation not detected by either diagnostic method (false negatives).
5.12 Patients identified as having atrial fibrillation were entered into the model in 2 cohorts according to age (65 or 75 years). Cohorts were classified according to stroke risk, using a clinical risk prediction scoring system of 0 to 6 (CHADS2) based on cumulative risk factors (for example congestive heart failure, hypertension, age over 75 years, diabetes mellitus and previous stroke). CHADS2 scores were used to stratify patients into low, moderate or high risk in order to determine the allocation of aspirin or anticoagulant therapy within the cohorts, in line with clinical guidelines. The false positive cohort was classified according to age and CHADS2 score but received no management. The clinical outcomes considered in the model were the prevention of fatal and non-fatal stroke and gastrointestinal bleeding caused by prophylactic drug therapy to reduce the risk of stroke. Costs were applied to each clinical consequence and state in the model, and accumulated to form an overall cost.
5.13 Key assumptions in the model were:
A confirmatory 12-lead ECG with 100% sensitivity and specificity would be carried out immediately after the detection of suspected atrial fibrillation.
Patients had 1 blood pressure measurement during the 10-year period of the model.
Patients received the same drug treatment throughout the duration of the model.
The model considered gastrointestinal bleeding as a one-off event with no change in management and no difference in cost or clinical outcome dependent on cause.
No treatments other than aspirin and anticoagulant therapy were considered.
Stroke was considered in 3 clinical states: the first year after the stroke, subsequent years after the stroke, and fatality, with subsequent risk adjusted accordingly. The possibility of recurrent non-fatal strokes was not included in the model, nor was the increased likelihood of stroke occurring over the 10-year period.
5.14 The diagnostic accuracy of WatchBP Home A was estimated from Wiesel et al. (2009) (96.8% sensitivity and 88.8% specificity). The diagnostic accuracy of pulse palpation was estimated from Hobbs et al. (2005) (87.2% sensitivity and 81.3% specificity). The cohorts were allocated CHADS2 scores based on a study by Gage et al. (2001), adjusted according to starting age. Other transitional variables such as stroke incidence, proportion of fatal strokes, proportion experiencing major gastrointestinal bleeding, proportion dying from other causes and relative risk reduction of stroke were taken from national statistics or published evidence.
5.15 Annual treatment costs and costs for major gastrointestinal bleeding were taken from the costing report from the NICE clinical guideline on atrial fibrillation, adjusted for inflation (5%). The estimates were £489 per patient for anticoagulant therapy, a negligible cost for aspirin and £2008 for a major bleed. The costs of non-fatal stroke were considered in 2 phases and were taken from a health technology assessment study by Hemingway et al. (2010) and estimated to be £12,565 for the first year and £3315 for subsequent years. The cost of fatal stroke was estimated to be £3036, taken from dabigatran etexilate for the prevention of stroke and systemic embolism in atrial fibrillation, NICE technology appraisal 249. The cost of a 12-lead ECG was estimated to be £31, taken from Department of Health 2010/11 reference costs. The time taken for the clinical measurements was considered to be the same for either method and therefore no costs for this were included in the model. The capital outlay cost of immediate replacement of existing monitors with WatchBP Home A was not included in the model but is considered further in section 5.23.
5.16 The External Assessment Centre concluded that, given the use over the lifespan of the device (5 years), the difference in costs per use between WatchBP Home A and a standard home blood pressure monitor would be negligible.
5.17 The diagnostic cost per patient of using WatchBP Home A was calculated to be slightly less than manual pulse palpation: £2.16 in the younger cohort (65–74 years) and £1.94 in the older cohort (75–84 years).
5.18 The model established that treating patients diagnosed with atrial fibrillation by any means was likely to be cost saving in most patient groups. Treating those in the younger cohort would incur a cost of £509 over a 10-year period for those with a CHADS2 score of 2, although this would become cost saving for those with a higher risk of stroke, with a potential saving of £3520 per patient for those with a CHADS2 score of 5. Similarly, in the older cohort, treating those with a CHADS2 score of 1 would incur a cost of £352 per patient, but for those with a score of 6, the cost saving would be £4401 per patient.
5.19 A base-case analysis found that the overall cost saving per patient from using WatchBP Home A in place of a standard monitor and manual pulse palpation was £2.98 for the younger cohort and £4.26 for the older cohort. Limited sensitivity analyses were carried out, varying the costs of stroke and anticoagulation. Results from this indicated that using WatchBP Home A would remain cost saving despite changes in treatment costs. Raising the cost of stroke increased the potential cost saving of using the device.
5.20 The results revealed that WatchBP Home A was likely to be clinically effective, with 53–117 fatal strokes prevented, and 28–65 non-fatal strokes prevented per 100,000 patients, depending on age. This benefit is likely to be offset against an increase in gastrointestinal bleeds resulting from adverse effects of prophylactic drug therapy to reduce the risk of stroke, of 34–68 per 100,000 patients, also depending on age.
5.21 The sponsor's de novo cost analysis did not consider the cost consequences of using WatchBP Home A for home blood pressure monitoring. The External Assessment Centre was therefore asked to carry out additional analyses to estimate the costs and consequences of using the device in this setting.
5.22 The External Assessment Centre did not consider 2 identified studies carried out in a home setting (Wiesel et al., 2012 and Wiesel et al., 2007) to be suitable to provide parameters to inform the model. It therefore developed a monetised cost consequence model for patients needing home monitoring for hypertension. The population was a simulated cohort of 100,000 including some with paroxysmal atrial fibrillation and some with persistent atrial fibrillation. Each subject was randomly assigned the outcome of either atrial fibrillation detected or absence confirmed. The intervention was use of the WatchBP Home A device at home for either 4 or 7 days. The comparator was defined as an alternative home blood pressure monitoring device without an atrial fibrillation detection algorithm. The study by Wiesel et al. (2009) was used to provide a sensitivity value of 95.3% and a specificity value of 86.4% for atrial fibrillation detection from single diagnostic measurements. The number of strokes prevented was considered as a separate output.
5.23 The External Assessment Centre included costs associated with replacing home blood pressure monitors with the WatchBP Home A device, using costs taken from the NICE clinical guideline on hypertension for use of a typical home blood pressure monitoring device (£45 annually). The External Assessment Centre considered that, other than the purchase cost of WatchBP Home A (£75), all other costs would be equivalent. Additional capital outlay for purchasing the WatchBP Home A device was estimated at around £620,000 assuming 1 device was purchased per GP practice (8245 practices in England and Wales). This would increase to around £2,700,000 if 1 device was purchased per individual GP.
5.24 The External Assessment Centre was unable to ascertain the prevalence or patterns of paroxysmal atrial fibrillation from the literature, so the following prevalence estimates were used in the model:
0.5%, representing the prevalence in people aged 50–59 years (from NICE clinical guideline 36 on atrial fibrillation)
1.28%, representing the prevalence in the general population (Majeed et al., 2001)
4.4%, representing the sponsor's estimate of atrial fibrillation prevalence
7.9%, representing the prevalence in people aged over 65 years (Hobbs et al., 2005)
9%, representing the prevalence in people aged 80–89 years (from NICE clinical guideline 36 on atrial fibrillation).
Sensitivity analysis was carried out to model time spent in atrial fibrillation using varying estimates.
5.25 A base-case analysis of using WatchBP Home A for home blood pressure monitoring showed that using the device would consume NHS and personal social services resources of around £5.32 per person and prevent 22 strokes per 100,000 people screened. In all cases, use of the WatchBP Home A device incurred a cost to the NHS and personal social services. Excluding those in permanent atrial fibrillation (100%), the additional costs incurred per patient ranged from £4.44 per person (9% prevalence, 2% time in atrial fibrillation, 7 day monitoring) to £10.30 per person (9% prevalence, 50% time in atrial fibrillation, 7 day monitoring).
5.26 The sponsor's cost analysis showed that WatchBP Home A was cost saving when compared with manual pulse palpation based on the assumption that people with atrial fibrillation were symptomatic. The Committee acknowledged that there were uncertainties in the economic model presented in the sponsor's base case, mainly concerning the costs of stroke and the 1-year time horizon.
5.27 Further modelling was carried out by the External Assessment Centre to establish the longer-term cost impact of using WatchBP Home A in primary care and to provide an estimate of the cost of stroke that was as reliable as possible. From these data, the Committee accepted that the WatchBP Home A device would be cost saving compared with standard blood pressure monitors and manual pulse palpation in patients aged 65 years and older. The Committee also noted that the analyses showed that atrial fibrillation-related fatal and non-fatal strokes could be prevented.
5.28 The Committee considered advice from clinical experts, which indicated that if the device was purchased for use in primary care, it was likely to be used for blood pressure measurement across the whole GP practice population once purchased. The Committee considered 2 scenarios for purchasing WatchBP Home A for use in primary care, which were immediate wholesale replacement of existing monitors and replacement at the end of device lifespan. The Committee was advised that, as cost per use of WatchBP Home A over its 5-year lifespan would be minimal, either scenario would still lead to cost savings over a 10-year time horizon, as modelled.
5.29 The Committee acknowledged that the cost impact of using the device in a population younger than that included in the model was unknown, because there was little data on the prevalence of atrial fibrillation or associated stroke risk in this group. The Committee did consider that despite the potential for the number of false positive results to increase, incidental detection of atrial fibrillation in patients under the age of 65 years would have significant clinical benefits. Advice from clinical experts suggested that treatment for atrial fibrillation in younger people at low risk of stroke would be less costly than that for older people. For those at higher risk of stroke, for example with other risk factors contributing to higher CHADS2 scores, use of the device would be likely to generate similar cost savings to those in older people at higher risk.
5.30 The Committee considered that although the findings from the economic modelling indicated that using WatchBP Home A for home blood pressure monitoring could potentially prevent a number of strokes, use of the device in this setting would not be cost saving.