Table 4.
Comparison of the AF detection service with three main competitors.
| Service | Apple Watch and iPhone | KardiaMobile with KardiaPro | Holter Monitor with CardioScan | |
|---|---|---|---|---|
| Performance evaluation | ||||
| Quality | PPV: 95.40% | PPV: 71% (pulse) | 8% AF yield | N/R |
| No. of patients | 82 | N/R | 50 | N/R |
| Dataset | AFDB and LTAFDB | Measurement data | Measurement data | Measurement data |
| System properties | ||||
| Signal | Heart rate | ECG | Finger ECG | ECG |
| Processing | Cloud server | Local | Cloud server | Local |
| Real-time | Yes | Yes | Yes | No |
| Diagnosis | Symbiosis between physician and DL | None | None | Feature support |
| Data storage | Unlimited | None | Snippets | Limited |
| Model update | Retraining the DL model with cloud data | None | None | None |
| Use case scenario | ||||
| Customer | Healthcare provider | Patient | Patient | Healthcare provider |
| Physical equipment | Heart rate sensor and Android phone | Apple Watch and iPhone | KardiaMobile device | Holter monitor |
| Measurement | Patient led | Patient led | Patient led | Expert led |
| Result | Diagnosis DL decision validated by a physician | Suspicion black box decision; follow-up with Holter recording for diagnosis | Suspicion black box decision; no clear follow-up | Diagnosis established by a physician with analysis support |
| Limitations | ||||
| Diagnosis | HR for diagnosis support is a new paradigm | No diagnosis; diagnosis is established through Holter recordings | No diagnosis | Inter- and intra-observer variability; labour intensive |
| Safety | Human and machine | Not critical | Not critical | Human |
| Cost | ||||
| Hardware | £ 300 | £ 1000 | £ 99 and mobile cost | £ 1885.00 |
| Service | £ 30/month | Free | £ 9.99/month | £ 50 for 10 h |