There were 27 studies relevant to the decision problem in the scope:
5 randomised controlled trials (RCTs)
7 diagnostic accuracy studies
1 case-control study
13 single-arm observational studies
1 case report.
Of the 27 included studies, 16 studies were peer reviewed, including 4 UK studies (Bray et al. 2021, Dimarco et al. 2018, Reed et al. 2021, Reed et al. 2019), one of which is an RCT (Reed et al. 2019). The included studies covered 6 population groups:
people with palpitations
people with a history of atrial fibrillation (AF), who have had treatment (ablation, cardioversion, or medical therapy) to restore sinus rhythm and used KardiaMobile to identify recurrence
people with diagnosed AF to assess AF burden
people with transient AF after surgery or hospitalisation whose heart rhythms reverted back to sinus rhythm before discharge and used KardiaMobile to identify recurrence
people after stroke or transient ischaemic attack who were monitored using KardiaMobile
mixed population including people with known or suspected AF.
All published evidence is on the single-lead KardiaMobile device. For full details of the clinical evidence, see section 4 of the assessment report in the supporting documentation.
Three RCTs including 1 UK trial (Goldenthal et al. 2019, Koh et al. 2021, Reed et al. 2019) found that significantly more people in the KardiaMobile monitored group had AF detected compared with those who had standard care, which included 24‑hour Holter monitoring. This was supported by the results from an observational study (Yan et al. 2020).
Four peer reviewed studies (Hermans et al. 2021, Lowres et al. 2016, Selder et al. 2019, William et al. 2018) reported on the diagnostic accuracy of AF detection using the KardiaMobile algorithm compared with clinical interpretation of the KardiaMobile ECG as the reference standard. Its sensitivity ranged between 92% and 99% per recorded ECG, with specificity between 92% and 98%. However, the external assessment centre (EAC) highlighted that diagnostic accuracy was reported on a per ECG recording and not a per person basis. Also, these 4 studies had 4 different patient populations with a pre-test probability of AF between 4.8% and 35.6%. The EAC also noted that KardiaMobile is not intended to be used to confirm the presence of AF as a standalone test but to help detect AF. All interpretations should be reviewed by healthcare professionals for clinical decision making. It is expected that false positives and negatives are likely to be captured by the clinical reviews.
Reed et al. (2019) showed that people using KardiaMobile had their symptomatic cardiac arrhythmia detected significantly earlier than those having standard care (9.9 days compared with 48.0 days, p=0.0004). This finding was supported by 1 observational study (Yan et al. 2020) which also reported that KardiaMobile significantly reduced the time to AF detection when compared with standard care. There was no direct published evidence to show that using KardiaMobile improves clinical outcomes (such as reduction in stroke) after a diagnosis of AF.
The evidence from 12 studies and a patient survey reported that KardiaMobile was easier to use compared with other ECG monitors such as Holter monitors. People felt that KardiaMobile would be useful in self-monitoring at home and improving their ability to access the care they needed. Two RCTs (Caceres et al. 2020, Guhl et al. 2020) showed that people who used KardiaMobile had a significant improvement in AF-specific quality-of-life scores compared with people in the control groups. The EAC noted that both trials used additional interventions, and the effect of KardiaMobile alone on quality of life may be difficult to interpret.
Evidence reported that there were a proportion of ECG traces that did not fit the current KardiaMobile algorithm classifications, ranging from 9.6% to 27.6%. These outputs are presented as unclassified. However, the company stated that software updates are improving the classification algorithm, and the number of unclassified outputs is reducing. Also, around 0.6% to 1.9% of KardiaMobile outputs were unreadable. This often happens when an ECG trace has interference and cannot be interpreted by the Kardia app; however, a proportion of these can be interpreted by a clinician.
Three published studies reported the economic impact of KardiaMobile:
a cost-effectiveness analysis done alongside a UK RCT compared the cost per symptomatic rhythm diagnosis using KardiaMobile in addition to standard care with standard care alone (Reed et al. 2019)
a UK budget impact analysis (York Health Economics Consortium et al. 2018)
a US single-arm study estimated the cost saving using data from a patient survey (Praus et al. 2021).
All studies reported that KardiaMobile was cost saving. Two studies reported that the main driver for the saving was a reduction in healthcare appointments.
The company developed a de novo model comparing KardiaMobile with Holter monitoring and the Zio patch. The model included people aged 64 and over with known or suspected AF who were referred for ambulatory ECG monitoring in a secondary care setting. The model assessed the costs associated with diagnosing and managing AF. Overall, the company's base case showed that using KardiaMobile could save between £320 and £380 per person over 5 years because of the cost of the technology, reductions in repeat testing, referrals to secondary care and stroke events.
For full details of the cost evidence, see section 9 of the assessment report in the supporting documentation.
The EAC was unable to validate the company's original model, and highlighted limitations and errors in some of the parameters and assumptions used. The complexity of the model meant that inconsistencies could not be investigated and corrected. There was a lack of robust evidence to support the need for such complex time dependencies in the diagnostic phase of the model, and this approach required several additional assumptions. The EAC considered that the diagnosis phase could have been modelled more simply. Overall, the EAC considered the model to be overly complex, not transparent and not verifiable. It also did not agree with underlying structural assumptions, parameter choice or their implementation in the model.
The company submitted an updated cost model, which was modified and simplified to address the EAC's comments. Furthermore, various scenario analyses and probabilistic sensitivity analyses were done to explore uncertainties of the cost impact. The model was informed by 6 comparative studies including 1 UK study (Reid et al. 2019) and the results showed that using KardiaMobile resulted in cost saving in 2 studies of people with palpitations, and in 3 studies of people who were monitored for recurrent AF but was cost incurring in a study for monitoring AF. The EAC was, however, unable to validate the company model without further clarifications from the company (see Appendix B - EAC commentary), limiting the certainty in the results presented.
The EAC initially did a simple cost calculator to explore the expected costs of using KardiaMobile to detect and treat AF over a 1‑year time horizon. It then went on to develop a new cost model that better captured the clinical pathway of using KardiaMobile for detection and ongoing monitoring of AF, compared with Holter monitoring in the NHS. The additional analyses included people presenting with undiagnosed palpitations and people who need to monitor AF recurrence after treatment. The base case results of the additional analyses showed a saving of £13.22 per person over 2 years when KardiaMobile was used for detecting AF in people presenting with undiagnosed palpitations but an additional £85.91 cost per patient over 10 years when KardiaMobile was used for monitoring AF recurrence detection in a population at low risk of having a stroke (CHA2DS2‑VAScC score of 1).