Approximately one-third of all ischemic strokes (IS) are characterized as cryptogenic strokes (CS), due to the lack of a possible cause for the event or incomplete diagnostic work-up [1-3]. Atrial fibrillation (AF), either paroxysmal or chronic, represents a major risk factor for stroke and systemic embolism, and is associated with a 5-fold increase in IS risk [4,5]. Although paroxysmal AF appears to be implicated in at least 30% of patients with CS and in approximately 25% of patients with unselected IS [6,7], current guidelines on secondary stroke prevention do not provide firm directions on AF screening after IS.

The American Heart Association/American Stroke Association (AHA/ASA) recommendations on secondary stroke prevention suggest that prolonged rhythm monitoring for approximately 30 days is reasonable for AF screening within 6 months after CS (Class IIa; Level of Evidence C) [8], while the recent AHA/ASA guidelines on the early management of IS patients indicate that the clinical benefit of prolonged cardiac monitoring to detect AF remains uncertain (Class of Recommendation: IIb, Level of Evidence: B) [9]. However, clinical trials [6,10] suggest that implantable cardiac monitors (ICMs) substantially increase AF detection in IS patients, due to prolonged monitoring duration.

In the present systematic review and meta-analysis, we sought to investigate the association of ICM duration with the level of AF detection in IS patients. We also assessed whether IS subtype, patient characteristics, and elapsed time between IS onset and CM implantation may affect the probability of AF detection.

The PRISMA flowchart summarizing the literature search process is shown in Supplementary Figure 1. MEDLINE and SCOPUS literature searches retrieved 375 and 417 results respectively, while comprehensive searches of key journals and conference proceedings identified 20 additional studies. Of all potentially eligible studies, 18 study protocols were excluded (Supplementary Table 1) due to overlapping data (n=4), the use of monitoring devices other than ICM (n=11), or unavailable information on ICM duration (n=3). Our literature search highlighted 28 studies for inclusion, comprising 4,531 patients (mean age 65 years, 52% male) [6,21-48]. Protocols and patient characteristics of included studies are briefly summarized in Table 1 and Supplementary Table 2, respectively. Most studies were conducted in the USA (n=9) and Germany (n=8). The most common subgroup studied was cryptogenic IS/TIA (n=17), followed by ESUS (n=9). The mean/median elapsed time from IS/TIA onset to cardiac monitoring implantation ranged from 3 to 174 days, while the mean/median ICM duration ranged from 180 to 1,080 days (Table 1). Included studies were generally found to have a low risk of bias (Supplementary Table 3), except in cases not clearly stating consecutive enrollment of patients [23,26,27,39,44-47], exclusion of AF with electrocardiogram or short-term non-invasive Holter monitoring prior to ICM impantation [21,33], or no adjudication of ICM recordings by experienced cardiologists [26-29,33,41,46,47].

In the overall analysis of all included studies, the cumulative AF detection rate in patients with ICM was 26% (95% confidence interval [CI], 22% to 30%), with significant heterogeneity among studies (I²=83%, P for Cochran Q <0.001) (Supplementary Figure 2). No evidence for publication bias was identified by funnel plot inspection (Supplementary Figure 3) or by the Egger’s statistical test (P=0.525). In univariate meta-regression analyses of all included studies (Table 2) the proportion of AF detection by ICM was positively associated with the duration of monitoring (coefficient=0.009; 95% CI, 0.005 to 0.013; P<0.001) (Figure 1A) and mean patient age (P=0.018) (Supplementary Figure 4A). No associations were detected with other patient characteristics (Supplementary Figures 5A and 9A), including sex (P=0.100) (Supplementary Figure 5A), hypertension (P=0.215) (Supplementary Figure 6A), diabetes mellitus (P=0.140) (Supplementary Figure 7A), mean patient CHA₂DS₂-VASc score (P=0.232) (Supplementary Figure 8A), or elapsed time from IS/TIA onset to cardiac monitor implantation (P=0.363) (Supplementary Figure 9A). In multivariate analyses, both monitoring duration (coefficient=0.015; 95% CI, 0.005 to 0.024; P=0.003) and mean patient age (coefficient=0.009; 95% CI, 0.003 to 0.015; P=0.004) were independently associated with the proportion of AF detection (Table 2).

In the sensitivity univariate meta-regression analyses of published studies (Supplementary Table 4) duration of ICM (coefficient=0.007; 95% CI, 0.001 to 0.014; P=0.049) (Figure 1B), history of hypertension (coefficient=0.005; 95% CI, 0.001 to 0.010; P=0.029) (Supplementary Figure 6B), and diabetes mellitus (coefficient=0.013; 95% CI, 0.001 to 0.024; P=0.033) (Supplementary Figure 7B) were positively associated with higher rates of AF detection, while no association was detected with other patient characteristics (Supplementary Figures 4B, 5B, 8B, and 9B). However, in multivariate analyses only monitoring duration (coefficient=0.009; 95% CI, 0.003 to 0.015; P=0.006) and mean patient age (coefficient=0.037; 95% CI, 0.013 to 0.062; P=0.007) were independently associated with the proportion of AF detection (Supplementary Table 4).

In subgroup analyses of all included studies (Table 3), there were significant differences (P for subgroup differences <0.001) in the rates of AF detection in subgroups stratified by ICM duration (<6 months: 5% [95% CI, 3% to 6%]; ≥6 and ≤12 months: 21% [95% CI, 16% to 25%]; >12 and ≤24 months: 26% [95% CI, 22% to 31%]; and >24 months: 34% [95% CI, 29% to 39%]) (Supplementary Figure 10). No differences were found in the subgroup analyses of all included studies stratified by study type (P=0.178) (Supplementary Figure 11), IS/TIA subgroup (P=0.093) (Supplementary Figure 12), the time threshold used for AF definition (P=0.234) (Supplementary Figure 13), the elapsed time from IS/TIA onset to cardiac monitor implantation (P >0.999) (Supplementary Figure 14) or the device used (P=0.174) (Supplementary Figure 15). Similarly, in the subgroup analysis of published studies (Supplementary Table 5) there were significant differences (P for subgroup differences <0.001) in AF detection rates in subgroups stratified by ICM duration (<6 months: 5% [95% CI, 3% to 6%]; ≥6 and ≤12 months: 22% [95% CI, 16% to 28%]; and >12 and ≤24 months: 22% [95% CI, 14% to 29%]) (Supplementary Figure 16). There were no differences in subgroup analyses of published studies stratified by IS/TIA subtype (P=0.429) (Supplementary Figure 17), the time threshold used for AF definition (P=0.149) (Supplementary Figure 18), the elapsed time from IS/TIA onset to cardiac monitor implantation (P=0.864) (Supplementary Figure 19), or the device used (P=0.174) (Supplementary Figure 20). In the aforementioned subgroup analyses, considerable heterogeneity was found within all subgroups (I²>70%), except for the subgroup of studies reporting the 6 min interval as a threshold for AF detection (I²=8%) and the subgroup of studies reporting more than 24 months of ICM duration (I²=24%).

Finally, in a post hoc analysis of available studies, we found that among IS/TIA patients with AF detected during ICM, a total of 87% (95% CI, 78% to 96%) experienced asymptomatic AF episodes (Supplementary Figure 21), with no evidence of heterogeneity between studies (I²=29%, P for Cochran Q=0.21).

Our meta-analysis showed that AF detection in patients with history of IS/TIA is positively associated only with the duration of ICM and patient age. We failed to find any other independent association between AF detection rates and IS/TIA subtype, device type, other patient characteristics, or elapsed time between IS/TIA onset and cardiac monitor implantation. Approximately nine out of 10 patients, with positive ICM for AF, experienced asymptomatic AF episodes during the monitoring period.

Our findings agree with a previously published systematic review and meta-analysis reporting improved AF detection with ICM, compared to wearable devices, in CS patients (23.3% [95% CI, 13.83% to 32.29%] vs. 13.6% [95% CI, 7.91% to 19.32%]; P for subgroup differences <0.05) [49]. However, compared to the previous meta-analysis, we included a significantly higher number of studies and patients (seven studies with 774 patients vs. 28 studies with 4,531 patients). Moreover, we assessed the potential modifying effect of stroke subtype, baseline characteristics, and time interval between ischemic event and implantation.

AF detection rates in patients with IS/TIA were unrelated to any patient characteristics, except for mean patient age. Although increased age, increased stroke severity, left atrial enlargement, hypertension, congestive heart failure, and valvular heart disease have been associated with increased incidence of AF detection in IS patients [50], proposed prediction scores including these parameters have limited diagnostic yield, especially at their middle grades [51]. Our meta-analysis also provides no further support to the theoretical concern regarding increased AF detection during the immediate post-IS period, due to stroke-induced sympathetic activation [52]. Finally, the results of subgroup analysis, regarding the time threshold used for AF definition, do not confirm the association of improved ICM performance with increased duration of AF episodes [53].

Another intriguing finding was that we observed no differences in AF detection rates using ICM, between CS and ESUS patients. This observation challenges the notion that paroxysmal AF is the main underlying etiopathogenic mechanism of cerebral ischemia in ESUS patients [54] and is in line with the recently reported New Approach Rivaroxaban Inhibition of Factor Xa in a Global Trial versus ASA to Prevent Embolism in Embolic Stroke of Undetermined Source (NAVIGATE ESUS) trial, where the detection rate of symptomatic AF during an approximate 1-year follow-up period was only 3% [55].

Several limitations of the present study need to be acknowledged. Firstly, it should be noted that baseline characteristics of individual patients (Supplementary Table 2) and study protocol parameters (Table 1) were unavailable in a significant proportion of included studies, and particularly in abstracts from conference proceedings. Moreover, the presence of ecological bias cannot be excluded; thus, the associations of aggregate patient characteristics may not hold true also for individual patient characteristics. Secondly, since this is a studylevel meta-analysis, we could not assess the influence of other parameters on AF detection rates that were not originally provided by included studies, e.g., the recently published hypertension, age, valvular heart disease, peripheral vascular disease, obesity, congestive heart failure, and coronary artery disease (HAVOC) score [49]. Thirdly, although meta-regression analyses did not provide evidence for any association bet=ween reported study characteristics (except for hypertension history) and AF detection rate, there is a possibility that heterogeneity in AF incidence could at least partially reflect inherent differences in the patient populations of included studies. Additionally, it should be highlighted that the lack of significant associations could be attributed to the low statistical power, especially for analyses including a low number of studies. Finally, it should be noted that in the present meta-analysis, we did not assess the number of AF episodes, false positive AF episodes, cumulative AF episode duration, or the impact of AF detection in patient management and long-term outcomes [56].

Our findings challenge current AHA/ASA guidelines [8,9], while further highlighting the indispensable role of prolonged rhythm monitoring, using ICM in the identification of a substantial portion of IS/TIA patients with occult AF. According to current recommendations, secondary CS prevention strategies are mainly based on antiplatelet therapy [6], which is known to provide inadequate protection for patients with AF. In these patients, the systemic administration of anticoagulant therapy could contribute to an 8.4% annual absolute risk reduction of stroke recurrence, compared with antiplatelet therapy [57]. Also taking into account the negative results of the recent NAVIGATE ESUS trial, showing that rivaroxaban compared to aspirin increases major bleeding without reducing ischemic events in ESUS patients [54], ICM emerges as an extremely useful diagnostic tool to identify those patients with occult AF within the heterogeneous group of ESUS or CS patients [58]. Therefore, prolonged monitoring could have a substantial impact on the secondary prevention of CS patients with underlying AF, leading to prompt anticoagulant initiation and lower stroke recurrence [59]. The Detection of Silent Atrial Fibrillation aFter Ischemic StrOke (SAFFO) study, guided by implantable loop recorder [60] and the AF detected by continuous electrocardiographic monitoring using implantable loop recorder to prevent stroke in individuals at risk (LOOP) study [61] are two ongoing, multicenter, open-label RCTs, that aim to evaluate health benefits, including reduction of recurrent ischemic events and cost-effectiveness of ICM in secondary stroke prevention. Results from these studies will further characterize the target population for ICM, the optimal threshold for AF definition and whether ICM monitoring results in lower stroke recurrence through anticoagulant initiation.

In conclusion, the results of the present meta-analysis support extended-duration ICM monitoring as a reasonable option for patients with IS or TIA, and initial negative screening for AF detection [62], that may substantially enhance detection of predominantly subclinical AF episodes.