Zio XT for detecting cardiac arrhythmias

The clinical evidence comprises 17 published studies, which include 169,063 people referred to ambulatory monitoring, and 13 abstracts:

• 1 UK-based randomised controlled trial (Kaura et al. 2019)

• 3 prospective within-subject comparative studies (Barrett et al. 2014, Eysenck et al. 2019, Rosenberg et al. 2013)

• 6 prospective non-comparative studies (Rho et al. 2018, Heckbert et al. 2018, Reed et al. 2018, Schreiber et al. 2014, Steinhubl et al. 2018, Turakhia et al. 2015)

• 7 retrospective non-comparative studies (Eisenberg et al. 2014, Go et al. 2018, Schultz et al. 2019, Solomon et al. 2016, Tung et al. 2015, Turakhia et al. 2013, Wineinger et al. 2019)

• 13 abstracts (Agarwal et al. 2015, Chandratheva et al. 2017, Ghosh et al. 2018, Hall et al. 2019, Keibel et al. 2015, Malhotra et al. 2018, Miller et al. 2014, Norby et al. 2018, Salazar et al. 2011, Sattar et al. 2012, Su et al. 2014, Turakhia et al. 2012, Ullal et al. 2013).
  
  The external assessment centre (EAC) noted that some of the non-comparative studies may have overlapping populations because the data are retrospective. For full details of the clinical evidence, see section 3 of the assessment report, which is in the supporting documents for this guidance.

Three of the 4 comparative studies compared 14‑day Zio XT with a 24‑hour Holter monitor (Barrett et al. 2014, Kaura et al. 2019, Rosenberg et al. 2013) and 1 compared it with an external loop recorder (the Novacor R. Test; Eysenck et al. 2019). The size of the studies varied, with a total of 357 people, including those with a recent stroke or transient ischaemic attack, people with pacemakers or diagnosed atrial fibrillation, and people with suspected arrhythmia. The EAC considered the multicentre UK randomised controlled trial to be the highest-quality study (Kaura et al. 2019). The EAC considered the other 3 comparative studies to be of adequate quality. The EAC did not do a meta-analysis because it considered the evidence to be too heterogeneous in terms of populations, methodology, comparators, and outcomes reported.

The randomised controlled trial compared the diagnostic yield of 14‑day Zio XT with 24‑hour Holter monitoring in 116 people with stroke or transient ischaemic attack. People randomised to the Zio XT monitoring arm also had 24‑hour Holter monitoring. There was a high withdrawal rate from both arms of the trial because 20% of the randomised patients refused to use the 24‑hour Holter monitor. This may have biased results. According to the trial authors, the study was adequately powered for the primary outcome. An independent power analysis done by the EAC found that the randomised controlled trial was likely to be underpowered because of the high withdrawal rate. The trial is underpowered for the secondary outcomes, which included anticoagulation use and mortality.

Three studies comparing arrhythmia detection rates for Zio XT with 24‑hour Holter monitoring showed an increased diagnostic yield with Zio XT over total wear time. Results from Eysenck et al. (2019) suggested that Zio XT may be more accurate in detecting the presence or absence of atrial fibrillation than the Novacor R. Test (an external event loop monitor, described as current standard practice) but less accurate than pacemaker data (described as gold standard).

Evidence from comparative studies suggested that most patients were happy to wear the Zio XT biosensor, with median wear time ranging from 10.8 days (Rosenberg et al. 2013) to 12.8 days (Eysenck et al. 2019) out of a scheduled 14 days. In Eysenck et al. (2019), the Zio XT biosensor was worn for longer than 3 of the other continuous cardiac monitors evaluated. In Barrett et al. (2014), 93.7% of patients found the biosensor comfortable to wear compared with 51.7% for the Holter monitor. A survey of patients from a UK cardiology clinic (Hall et al. 2019) found that Zio XT was statistically significantly preferred to Holter monitoring in terms of shape, comfort, practicality and returning method.

The diagnostic accuracy of Zio XT compared with standard care was not clearly defined in any study. Barrett et al. (2014) and Rosenberg et al. (2013) did some analysis comparing a Holter monitor and Zio XT over the same 24‑hour period with different results. Rosenberg et al. (2013) reported that there was statistically significant agreement between Zio XT and the Holter monitor recordings over the same 24‑hour period. However, Barrett et al. (2014) reported the Holter monitor detected 11 arrhythmia events that were not detected by Zio XT over a simultaneous 24‑hour monitoring period. The authors stated that 2 were caused by Zio XT algorithm misclassification, which was then corrected, and 7 were errors made by the company's report reviewer. A technical study by Hannun et al. (2019) reported good diagnostic performance for the deep neural network used as part of Zio XT compared with a committee of cardiologists. There is no evidence to show that an increased diagnostic yield with Zio XT improves clinical outcomes. The EAC considered that, without more information about diagnostic accuracy, it is not clear if the changes to treatment reported in the studies were an appropriate response to the patient's condition. During consultation, the company submitted unpublished technical data to support diagnostic accuracy claims. This is detailed in section 3.8.

Five studies were done in the UK and the EAC considered the evidence is generalisable to the NHS: 2 comparative studies (Kaura et al. 2019, Eysenck et al. 2018), 1 prospective non-comparative study (Reed et al. 2018), and 2 conference abstracts (Chandratheva et al. 2017, Ghosh et al. 2018). The 2 remaining comparative studies were in the US.

During consultation, the company submitted technical data reporting the performance of Zio XT's algorithm (ZEUS) at detecting 11 arrhythmia types and sinus rhythm. Using a reference database of known clinical rhythms (over 150 examples of each), the ZEUS algorithm showed a sensitivity of over 90% and a specificity of over 95% for detecting atrial tachycardia (including atrial fibrillation and supraventricular tachycardia) from a total of over 1,400 episodes. Post-market analysis using clinical records processed by Zio XT showed a sensitivity of over 90% for detecting atrial tachycardia from a total of over 50 million episodes. Newcastle EAC reviewed the technical data provided by the company and concluded that Zio XT shows excellent agreement with cardiographic technician assessment of identified arrhythmia episodes. However, it highlighted that the analyses were done per episode and not on a per-patient basis. But it stated that, in principle, good per-episode performance should translate to good per-patient performance.

During consultation, KiTec EAC reviewed the artificial intelligence in Zio XT's ZEUS software. It noted that the algorithm to detect arrythmia uses a fixed deep neural network (a computational model made up of multiple processing layers) and that the training of the system is adequate. The EAC noted it had a large training dataset and that uncommon rhythms were well represented within this dataset. The software also used an external database for validation. The EAC also reviewed the internal quality assurance process and noted that before providing a report to clinicians, the algorithm outputs are checked by the company's cardiographic technicians (certified by The Society for Cardiological Science and Technology and Cardiac Credentialing International). The percentage of electrocardiogram (ECG) traces that are quality assured by cardiographic technicians was not reported but the agreement between algorithms and cardiographic technicians was high (over 99% overall episodic sensitivity in post-market analysis).

During consultation, 1 NHS trust and 1 sustainability and transformation partnership provided feedback and results from using Zio XT. The EAC reviewed the evidence submitted but did not consider it added anything to the existing evidence base. It concluded that clear conclusions about efficacy could not be drawn from the results because there was not enough information on the patient populations.

Five published studies reported the economic impact of the technology:

• a UK budget impact analysis (Kaura et al. 2019)

• a UK study reporting technology costs using data from the REMAP‑AF trial (Eysenck et al. 2019)

• a prospective matched cohort study reporting healthcare resource use (Steinhubl et al. 2018)

• 2 conference abstracts (Ghosh et al. 2018, Chandratheva et al. 2017).
  
  Two studies reported that the technology was cost saving. Two reported it was not cost saving compared with other devices including Holter monitoring. Studies consistently reported that Zio XT is the most efficient in terms of avoiding delays between clinic and diagnosis confirmation.

The company created 3 de novo cost analyses comparing the 14‑day Zio XT with blended strategies, based on a 24‑hour Holter monitor or a cardiac event recorder, in different care pathways:

• The cardiology model (presented as a base case) considered people with symptomatic palpitations or syncope and assessed the costs associated with the diagnostic process only.

• The stroke model (presented as a base case) considered people who have had a stroke or transient ischaemic attack and assessed the costs associated with the diagnostic process only.

• The downstream stroke model was presented as a scenario analysis and extrapolated the economic consequences of the extra risk of recurrent stroke because of delayed or missed diagnosis of atrial fibrillation.
  
  All models had a time horizon of 1 year. Overall, the company's models showed that using Zio XT saves between £55 and £85 per patient because of reductions in repeat testing, referrals or cardiology outpatient review, and events in stroke populations. For full details of the cost evidence, see section 4 of the EAC's assessment report in the supporting documents for this guidance.

The EAC revised the base-case (cardiology and stroke) models to address some potential limitations:

• the proportion of patients having repeat Holter tests after 24‑hour Holter monitoring was changed to 27%

• NHS reference costs were used for Holter monitoring rather than Patient Level Information and Costing System (PLICS) data

• the cost of an outpatient appointment before discharge was included for all tests.
  
  The EAC revised the downstream stroke model to:

• include the cost of anticoagulants (and their side effects)

• lower the estimated stroke risk

• include repeated diagnostic test costs.
  
  The EAC considered the downstream stroke cost model the most informative. After these revisions, the EAC concluded that Zio XT is unlikely to be cost saving at the original price when compared with current practice. Zio XT became cost incurring by:

• £0.82 per patient per year in the cardiology model

• £70.81 per patient per year in the stroke model

• £20.83 per patient per year in the downstream stroke model.

The EAC did a scenario analysis to explore the impact of repeat monitoring after a negative test. Zio XT was cost incurring when all monitoring was repeated after a negative test. When monitoring with a 24‑hour Holter or a 7‑day cardiac event recorder was repeated after a negative first test, but Zio XT was not repeated, the technology was cost saving. The EAC also explored the impact of excluding follow-up outpatient appointments after monitoring for some or all tests. Full descriptions of the 5 scenarios explored and results for each of these scenarios can be found in section 2 of the addendum to the EAC's assessment report in the supporting documents for this guidance.

During consultation, the company revised the cost for the technology to £265 per person. The EAC recalculated the results from the cost modelling with the lower price in the 2 most relevant scenarios, which included the following assumptions:

• no appointment needed after a negative result, and no repeated test (scenario 3)

• no appointment needed after any negative result, whether or not the test is repeated (scenario 5).
  
  The results showed Zio XT is likely to be cost saving or cost neutral compared with standard care in the cardiology model (£59.80 less for scenario 3; £3.47 less for scenario 5 per patient per year). For the stroke model, the results were still cost incurring (£14.93 more for scenario 3; £79.47 more for scenario 5 per patient per year) but the downstream model results for the same population showed the results moved towards Zio XT being cost saving (£72.55 less for scenario 3; £33.79 more for scenario 5 per patient per year). Full results for all 5 scenarios explored can be found in section 1 of the additional economic analyses in the supporting documents for this guidance.

Additional sensitivity analysis was done to examine the effect of a potential change in the use of implantable cardiac monitors. The experts did not all agree that use of implantable cardiac monitors is likely to increase in the future. In the stroke and cardiology models, a small percentage of patients (2%) who have a negative test will go on to have implantable cardiac monitoring. Increasing the probability of testing with implantable cardiac monitors to 10% did not substantially change the results of the Zio XT cost modelling. For full details, see the additional sensitivity analysis report and additional clinical expert advice in the supporting documents for this guidance.

The EAC also did a threshold analysis to assess the number of repeat tests per person in the current care arm needed to equalise the cost of Zio XT and current care. The average number of repeat tests after a negative result with either 24‑hour Holter or cardiac event recorders was assumed to be 1.389 per person in the base case. Threshold analysis results ranged from 1.30 to 1.78 repeat tests per person, depending on the model and scenario evaluated. For full details, see the additional sensitivity analysis report in the supporting documents for this guidance.