The sponsor submitted clinical and economic evidence according to a scope issued by NICE . The economic evidence included a de novo cost-consequence analysis. The EAC critically appraised the submission and carried out additional analyses to evaluate the outcomes identified in the scope.
Clinical Effectiveness Evidence
Sponsor’s Review of Clinical Effectiveness Evidence
The sponsor performed a literature search and identified and presented a total of ten studies that they considered were relevant to the decision problem. The EAC considered five were relevant to the decision problem (i.e., within scope); these were four published studies [22–25] and one study that was awaiting publication at the time of assessment and has since been published . Five studies were not fully appraised by the EAC as they were BP measurement validation studies or measurement of BP was their sole outcome, and they were not used by the sponsor in their evaluation of the device. Of these excluded studies, three were peer reviewed and published [21, 27, 28], one was a conference abstract , and an additional, unpublished, randomized controlled trial (RCT) that was potentially highly relevant to the decision problem was in the process of patient recruitment, but no results were available .
Of the five studies that were available and relevant to the decision problem, three were in a secondary-care clinical setting [22, 24, 25] and two (one published  and one unpublished at the assessment ) were based in a home setting. The studies in a home setting, although within scope, were not used for further analysis by the EAC as this scenario was not explored further by the sponsor, and the only published study  was an uncontrolled case series with a population not within the scope of the decision problem (people with a previous history of AF). The unpublished study included patients with AF and used a non-standard comparator (ECG events monitor) . This left three studies [22, 24, 25] that were potentially useful to inform the clinical effectiveness of the WatchBP Home A as well as to support the assumptions of the economic analysis.
Descriptions of the relevant studies, and an additional study that was identified by the sponsor and used to supply data concerning the comparator (pulse palpation) , are given in Table 1. These were cross-sectional diagnostic studies, with the diagnostic accuracy of the AF detection algorithm of the WatchBP Home A compared with 12-lead ECG as their main outcome. Overall, these studies indicated that the sensitivity of the AF detection algorithm was 93–100 % [22, 24], specificity was 69–91 % [22, 24], positive predictive value (PPV) was 68 %, and negative predictive value (NPV) was 98 %  (Table 2).
The sponsor critically appraised the literature using a checklist, but did not attempt to place the studies within the context of the decision problem. In particular, the sponsor did not discuss how limitations in the external validity of the studies might impact on their generalisability to UK primary healthcare practice.
Critique of Clinical Effectiveness Evidence
The EAC considered that only three of the provided studies were relevant to the decision problem. None of these studies used the WatchBP Home A as the intervention, although the sponsor confirmed that all used an algorithm equivalent to that which is embedded in the device.
The three cross-sectional diagnostic studies [22, 24, 25] were considered by the EAC to be appropriate to assess the diagnostic accuracy of the device (that is, they were of a suitable design to assess how accurately the device identifies people with irregular pulse and possible AF, and how well the device excludes people who do not have an irregular pulse). All the studies used 12-lead ECG as the reference standard, which is widely regarded as the gold standard test for the diagnosis of AF, and, because it uses an entirely different mechanism to detect AF, it is not subject to incorporation bias . The EAC critically appraised these studies using methodology adopted from the Critical Appraisal Skills Programme , and considered that they had adequate internal validity.
The EAC considered the external validity of the selected studies and the extent to which they related to the primary-care setting of the NHS of England and Wales. It identified specific areas where the population, intervention and comparator used in the studies differed to those described in the scope. Briefly, the populations described in the studies were recruited from secondary care in the USA and Greece, lending them susceptible to referral and spectrum bias . The greater prevalence of AF in these populations would be expected to significantly impact on the PPV and NPV recorded by the device . The intervention used in the studies was not specifically the WatchBP Home A device, and the devices used different modes to the WatchBP Home A; for instance, in the number of readings required to deliver a positive ‘detection’ of AF. Finally, the comparator used in the studies, although appropriate as a gold standard to evaluate diagnostic accuracy, was not that of the scope, which was pulse palpation.
Because analysis of the appropriate comparator, pulse palpation, was critical to answer the decision problem, the EAC undertook a brief literature search to find relevant studies that might address this issue. As no head-to-head studies of the WatchBP Home A (or equivalent device) and pulse palpation were identified, the EAC made an indirect comparison using data derived from the SAFE (Screening for Atrial Fibrillation in the Elderly) study , which was considered to be of highest methodological quality and have greatest external validity. The SAFE study was a large (n = 14,802), multicentre, pragmatic RCT commissioned by the NHS Research and Development Health Technology Assessment (HTA) Programme. It investigated the effectiveness and cost effectiveness of an opportunistic strategy to screen for AF using pulse palpation (followed by confirmatory 12-lead ECG) with systematic screening with nurse- or GP-led ECG (followed by confirmatory 12-lead ECG) in an elderly population. In addition, there was a control arm who received neither screening intervention. The opportunistic strategy proved to be most cost effective. The diagnostic parameters reported concerning pulse palpation were a sensitivity of 87.2 %, a specificity of 81.3 %, a PPV of 30.1 %, and an NPV of 98.6 % when comparing GP- or nurse-led pulse palpation with 12-lead ECG . The results of this study, together with those from the sponsor-identified studies on the Watch BP Home A, are presented in Table 2. These data were used in subsequent economic analysis.
Sponsor’s Economic Submission
The sponsor performed a basic literature search which did not identify any economic studies relevant to the decision problem (i.e., they cited no studies which used the WatchBP Home A device). The sponsor developed a de novo economic model using a cost-consequence design , taken from the perspective of the NHS and Personal Social Services (PSS), with a 1-year timeframe. This was constructed using an executable Microsoft Excel spreadsheet. The year of valuation was 2011. In the model, a simulated cohort of patients who were symptomatic of AF underwent screening for AF using either the WatchBP Home A device (intervention) or pulse palpation (comparator). If AF was detected, it would be confirmed with 12-lead ECG and be managed appropriately with anticoagulation or antiplatelet therapy. The differences in sensitivity and specificity between the intervention and comparator meant that there were different proportions of true positives (TPs), false positives (FPs), true negatives (TNs), and false negatives (FNs) for the intervention and comparator groups. Each diagnostic classification was associated with a corresponding clinical consequence; thus, for instance, patients who were classed as TP had AF and were appropriately managed; patients who were FP did not have AF but incurred a charge for an unnecessary ECG; patients who were TN were correctly excluded from further management; and patients who were FN were wrongly excluded as not having AF, and were therefore not treated.
Costs to the healthcare system arose from money spent on ECGs, anticoagulation, adverse effects of anticoagulation (limited in the model to gastrointestinal bleeding) and management of stroke. The device itself was not considered to incur a cost, as it was reasoned an alternative BP monitor would be required, and the incremental cost of the WatchBP Home A compared with other oscillometric BP monitors was relatively small. Pulse palpation was assumed to have a time cost of 1 minute, consistent with the SAFE study . The sponsor derived estimates of diagnostic accuracy of the WatchBP Home A from the published literature [16, 22, 24, 25, 36] (Table 2). Annual and entry costs attributed to each health state were mainly derived from the NICE costing study Atrial Fibrillation , and adjusted upwards at an assumed rate of 5 % per annum, since publication, to account for inflation. These values are presented in Table 3.
In the base-case analysis, the sponsor assumed that all patients would present to their GP with symptoms of AF and be tested for pulse irregularity with the WatchBP Home A or pulse palpation. Using an incidence estimate of AF of 0.175 % derived from a NICE costing report , the sponsor calculated that 87,000 new cases of AF would present to primary care each year. The sponsor derived a prevalence rate of 4.4 % from these data and a calculation taken from the SAFE study . Overall, the sponsor found that detection of AF per se was associated with a net cost to the NHS, because the costs associated with detecting AF through ECG and subsequent anticoagulation management outweighed costs associated with preventing stroke. However, the sponsor calculated that the improved diagnostic accuracy of the WatchBP Home A would lead to an annual cost saving to the NHS of approximately £11,600,000, with the prevention of 221 strokes, when compared with pulse palpation. This cost saving resulted mainly from lower ECG costs associated with fewer FP tests.
The sponsor also performed a limited univariate and threshold sensitivity analysis based on a scenario where only a proportion of patients were symptomatic of AF. They found that if 50 % of patients were asymptomatic then the WatchBP Home A was not cost saving. This was because time and ECG costs associated with pulse palpation were absent in 50 % of patients simulated. The sponsor also performed a sensitivity analysis on the cost of stroke, by including societal costs. It was found that if an annual societal cost of stroke of £44,000 was assumed , the WatchBP Home A would result in annual savings of approximately £64,600,000 compared with £45,400,000 for pulse palpation. Threshold sensitivity analysis showed that an assumed annual cost of stroke of approximately £16,000 would result in savings for the WatchBP Home A.
Critique of Economic Evidence
The EAC identified several weaknesses in the sponsor’s economic model, primarily relating to the fact that it did not match the scope of the decision problem, particularly in terms of model structure, the modelled population, the intervention and the comparator, and parameter inputs.
The EAC considered that the structure of the economic model was incorrect in that it derived important parameters in a reverse manner; i.e., it used the incidence data derived from different epidemiological studies together with diagnostic parameters taken from an RCT to inform about population prevalence [16, 36]. From this, the number of people who would need to be screened to achieve this incidence was calculated. The EAC considered it would have been better to select and estimate the population consistent with the scope (i.e., people having a clinical measurement of BP), estimate the likely prevalence of AF in this population, and calculate forwards from this point, using sensitivity analyses as appropriate. An additional limitation of the sponsor’s model was that it only had a time horizon of 1 year, which is insufficient to estimate the long-term implications of managing stroke risk (and likely to underestimate the benefits of the device).
The EAC considered that the population included in the model, people who are symptomatic of AF, was not the population described in the scope, and it should have been people who were having their BP measured (i.e., people asymptomatic of AF). The EAC considered this was a fundamental error in the sponsor’s analysis, and noted that the sponsor’s own sensitivity analysis indicated the WatchBP Home A was not cost saving in asymptomatic populations. Additionally, an important area of research described in the scope which the sponsor omitted from the model was the use of the WatchBP Home A for the diagnosis or monitoring of hypertension in a home setting (with the possible incidental detection of AF).
The EAC was asked to provide an analytic model of the device in a home setting. The EAC found that the WatchBP Home A was always cost-incurring in this setting, due to the lack of a suitable comparator and the fact that the costs of anticoagulation management outweighed those of preventing stroke.
The EAC did not fully agree with the sponsor’s choice of study to inform the diagnostic accuracy of the WatchBP Home A and pulse palpation; these issues are presented in Table 2. Part of the claimed cost saving of the WatchBP Home A was that the device saved time compared with manual pulse palpation, but the EAC did not accept this was likely due to the repeat measurements required by the device’s instructions for use. Additionally, the EAC noted that, according to the NICE processes and methods guides, cost estimates should be taken from the perspective of the NHS and PSS and not society as a whole (i.e., indirect costs should not be counted) , so the sponsor’s use of societal costs for stroke in their sensitivity analysis fell outside the scope of the MTEP’s processes and methods. Other issues with parameters and costs are presented in Table 3.
Overall, the EAC considered that there were significant uncertainties about the model structure, inputs and parameter. The Committee (MTAC) considered that the issues raised by the EAC were valid and therefore asked for further modelling to be carried out. The EAC developed a cost-consequence decision-analytic model, from the perspective of NHS and personal social care costs, to simulate incidental detection of asymptomatic AF in a primary-care population. The number of critical events (fatal and non-fatal strokes, gastro-intestinal bleeds) and accumulated costs for the WatchBP Home A were estimated and compared with manual pulse palpation for two age cohorts (65–74 and 75–84 years). The year of valuation was 2011 and the discount rate was 3.5 %. From this de novo model, the EAC estimated that the WatchBP Home A was associated with a per use saving of £2.98 for patients aged 65–74 years and £4.26 for people aged 75–84 years (potential savings in younger age groups could not be meaningfully quantified). Given the frequency that BP measurement is undertaken, this would represent significant savings for the NHS in England and Wales. Additionally, the device was associated with the prevention of 53–117 nonfatal and 28–65 fatal strokes over a 10-year time period, per 100,000 people measured, due to incidental detection of AF and subsequent prevention of stroke.