3 Results
3.1 Baseline Characteristics
There were 109 patients in each group. No statistical difference was detected in demographic data between patients in the two groups (Table 1).
3.2 Primary Outcome
After follow-up of 344.0 ± 60.5 days in BT Group, 70/109 patients (64.2%) were free from AF recurrence (70.8% in paroxysmal AF and 54.5% in persistent AF). While in TF Group, the AF free rate was 86/109 patients (78.9%, P=0.0163<0.05) after follow-up of 346.8 ± 59.9 days (88.2% in paroxysmal AF and 57.6% in persistent AF). The BigThumb detected more AF recurrence in paroxysmal AF after ablation (P=0.0099<0.05) but not in persistent AF (P=0.7910>0.05).
Adherence on anticoagulation in patients with CHA2DS2-VASc score >1 was significantly higher in BT Group (25/49, 51.0%) than TF Group (16/63, 25.4%, P=0.0052<0.05). There were 11/109 (10.1%) and 2/109(1.8%) patients receiving a second ablation in BT Group and TF Group respectively (P=0.0101<0.05).
The Kaplan‐Meier curves for recurrence stratified by follow-up strategies at time of enrollment are shown in Figure 2. Log rank tests showed a significantly earlier documentation of AF recurrence in BT Group than TF Group (P < 0.05). In a multivariate Cox model, the likelihood of recurrence detection was greater in patients in BT Group with larger left atrium (Table 2).
3.3 Feasibility of BigThumb in follow-up after ablation
A total of 26133 bECGs were recorded, among which 3299 (12.6%) were confirmed as AF by cardiologists’ manual review, 3860 (14.8%) were diagnosed as AF by the automated AF detection algorithm and 3457 (13.2%) by the AI algorithm. The sensitivity of the AI algorithm was significantly higher than the automated AF detection algorithm [(94.4%, 95% CI, 93.5% - 95.1%) VS (90.7%, 95% CI, 89.7% - 91.7%)]. The specificity was also higher in the AI algorithm [(98.5%, 95% CI, 98.3% - 98.6%) VS (96.2%, 95% CI, 95.9% - 96.4%)]. There were 2514/3299 (76.2%) AF bECGs marked with symptoms by participants.
3.4 Compliance of BigThumb Monitoring
The monitoring frequency was 0.53 ± 0.02/day in all the participants. They were used more frequently in the first three-month follow-up than after (1.19 ±0.03/day VS 0.36 ± 0.02/day, P<0.05) as shown in Figure 3. The monitoring was most frequently collected in the daytime than the nighttime. Hence, the episodes of AF were more frequently detected during daytime (Figure 4).