Early versus delayed insertable cardiac monitor implantation after ESUS stroke and the yield of atrial fibrillation detection: A systematic review and meta-analysis

Introduction

Covert atrial fibrillation (AF) is considered one of the most important underlying mechanisms in patients with cryptogenic or embolic stroke of undetermined source (ESUS). ¹ Indeed, AF can be detected in almost 30% of cryptogenic or ESUS patients during long-term follow-up as shown by previous randomized controlled trials of prolonged cardiac monitoring like the CRYSTAL-AF(Study of Continuous Cardiac Monitoring to Assess Atrial Fibrillation After Cryptogenic Stroke), ² EMBRACE (30-Day Cardiac Event Monitor Belt for Recording Atrial Fibrillation After a Cerebral Ischemic Event), ³ and Find-AF_RANDOMISED (A Prospective, Randomized, Controlled Study to Determine the Detection of Atrial Fibrillation by Prolonged and Enhanced Holter Monitoring as Compared to Usual Care in Stroke Patients). ⁴ Furthermore, previous meta-analyses have also demonstrated the clinical utility of prolonged rhythm surveillance using implantable cardiac monitors (ICMs), showing improved AF detection rates compared with conventional follow-up. ⁵ However, despite the increasing adoption of ICMs in post-stroke care pathways, the optimal timing of device implantation remains unclear. This systematic review and meta-analysis aimed to synthesize available evidence on ICM use after cryptogenic or ESUS ischemic stroke or transient ischemic attack (TIA), with a specific focus on the timing of implantation. We assessed whether the delay between the index cerebrovascular event and ICM implantation influences AF detection rates and the time to AF diagnosis.

Methods

Results

The systematic literature search initially identified 2096 records. After removing 593 duplicates, 1563 titles and abstracts were screened, leading to the exclusion of 838 records that did not meet inclusion criteria. A total of 725 full-text articles were assessed for eligibility, of which 678 were excluded for reasons such as inadequate outcome reporting or overlapping populations. Ultimately, 47 studies were included in the final quantitative synthesis. The study selection process, including reasons for exclusion at each stage, is illustrated in the PRISMA flow diagram (Supplemental Figure 1).

Pooled proportion of atrial fibrillation detection across studies

We performed a random-effects meta-analysis to estimate the overall proportion of AF detected by ICM in patients after a cryptogenic stroke or ESUS index event. A total of 47 studies were included, enrolling 6918 patients who underwent ILR monitoring after ischemic stroke. Across these studies, 1873 patients were diagnosed with AF. The pooled AF detection rate was estimated at 27.3% (95% CI: 24.6–30.2). Between-study heterogeneity was considerable, with an I² of 80.8%, suggesting substantial clinical and methodological differences among the included studies (Figure 1). Individual study estimates varied widely, ranging from 10.3% (Toyoda et al. ⁷ ) to 70.0% (Cuadrado-Godia et al. ⁸ ) Several studies reported AF detection rates above 40%, including Melis et al.,^9,10 (45.7%), Gambino et al. (41.7%), and Kitsiou et al. ¹¹ (41.5%). Conversely, detection rates were below 15% in studies such as Toyoda et al. ⁷ (10.3%) and Yushan et al. ¹² (12.0%). We performed a leave-one-out sensitivity analysis to assess the influence of individual studies on the overall estimate (Supplemental Figure 2). The pooled AF detection proportion remained stable across iterations, and no single study had a disproportionate effect on the overall results or heterogeneity estimates, confirming the robustness of the findings. The results remained highly consistent across iterations, with predicted AF detection proportions ranging narrowly from 26.7% to 27.9%, and confidence intervals showing minimal variation. Visual inspection of the funnel plot showed no clear asymmetry, and Egger’s test did not provide statistically evidence of small-study effects (z = −0.69, p = 0.490; Supplemental Figure 3).

OPEN IN VIEWER

Figure 1.

Forest plot of atrial fibrillation (AF) detection rates across the 47 included studies. Each point represents the proportion of patients with AF detected by insertable cardiac monitor (ICM) during follow-up. Horizontal lines denote 95% confidence intervals. Studies are ordered by AF detection rate.

Timing from index event to ICM implant and AF detection

We conducted a mixed-effects meta-regression to investigate whether the timing of ICM implantation—categorized as early versus delayed—was associated with AF detection rates, while adjusting for the duration of ILR monitoring. The grouping was based on the median time from the index stroke/TIA to ICM implantation across studies, which was 31.5 days. Studies with a median ILR implantation time below 31.5 days were classified as the early group, whereas those with a median time equal to or greater than 31.5 days were classified as the delayed group. This analysis included 45 studies with complete data as two studies^13,14 did not report the duration of the ICM monitoring. We found that the median ICM monitoring duration was longer in the delayed group (707 days, IQR: 592–974) compared with the early group (382, IQR: 364–465) (p < 0.001). The meta-regression analysis adjusted for the duration of the ICM monitoring showed that timing of ICM implantation emerged as a predictor of AF detection (β = −0.3198, p = 0.0117), indicating that delayed implantation was associated with lower log-odds of AF detection compared with early implantation, even after adjusting for monitoring duration (Figure 2). The model explained 41.95% of the between-study heterogeneity (R² = 41.95%) and reduced residual heterogeneity, with an I² of 80.84%. Based on model predictions at the mean ICM monitoring duration, the estimated percentage of AF detection was 30.0% (95% CI: 26.7–33.5) in the early group and 23.7% (95% CI: 20.6–27.2) in the delayed group (Figure 2). There were no differences in the distribution or proportions of key variables between the early and delayed ICM groups, including age, sex, and CHA₂DS₂-VASc score. See supplemental results for additional analysis. To assess the potential impact of unmeasured confounding by these variables, we calculated E-values for each. The E-value for age was 1.16, for female sex was 1.60, and for CHA₂DS₂-VASc score was 1.27. The E-value for CHA₂DS₂-VASc marginally exceeded the commonly used threshold of 1.25.

OPEN IN VIEWER

Figure 2.

Pooled proportion of AF detection by timing of ICM implantation (early vs delayed), adjusted for monitoring duration. Patients receiving ICM implantation within 31.5 days from the index event had a higher AF detection rate compared with those with delayed implantation (30.0% vs 23.7%, p = 0.0117).

To further investigate the association between the timing of ICM implantation and AF detection, we categorized the interval between the index event and ILR implantation into three groups based on tertiles of the distribution across the 45 included studies (two studies were excluded from this analysis as they did not report ICM monitoring duration). Specifically, studies with a median time from stroke to ILR implantation of less than 27.7 days were categorized as early, those between 27.7 and 62.3 days as intermediate, and those with more than 62.3 days as delayed. We observed a gradual increase in the median duration of monitoring from early to delayed implantation (387, 510, and 569 days, respectively). We then conducted a mixed-effects meta-regression including these three timing categories as dummy variables and adjusting for ILR monitoring duration. Based on model predictions at the median ICM monitoring duration, the estimated percentage of AF detection was 30.8% (95% CI: 26.8–35.2) in the early group, 28.5% (95% CI: 24.7–32.7) in the intermediate group, and 22.0% (95% CI: 18.6–25.8) in the delayed group (Figure 3). These estimates were adjusted for ICM duration using a random-effects meta-regression model with logit-transformed proportions and REML estimation. In the first model, using early implantation as the reference group, the delayed group was associated with lower log-odds of AF detection (β = −0.459, 95% CI: −0.752 to −0.166, p = 0.0022), while the intermediate group showed no difference (β = −0.111, 95% CI: −0.389 to 0.169, p = 0.438). Longer ICM monitoring duration was independently associated with increased AF detection (β = 0.0011, p < 0.0001). The model explained 45.8% of the between-study heterogeneity (R² = 45.8%), with substantial residual heterogeneity (I² = 79.5%).

OPEN IN VIEWER

Figure 3.

Pooled proportion of AF detection stratified by tertiles of ICM implantation timing, adjusted for monitoring duration. The detection rate was highest in the early group (<27.7 days, 30.8%) and progressively declined in the intermediate (27.7–62.3 days, 28.5%) and delayed (⩾62.3 days, 22.0%) groups. Early and intermediate groups both showed higher detection rates compared with the delayed group (p = 0.0022 and p = 0.0189, respectively).

Association between timing of ICM implantation and AF detection delay

To explore whether the timing of ICM implantation influences the time to AF detection, we conducted a meta-regression with time from ICM implant to AF detection as the outcome, and time from index event to ICM implant as a continuous predictor, adjusting for the duration of ICM monitoring. The model showed a statistically positive association between later ICM implantation and longer time to AF detection (β = 0.32; 95% CI: 0.14–0.51; p = 0.0007), even after adjusting for ICM monitoring duration. This suggests that each day of delay in ICM implantation was associated, on average, with an additional 0.32 days until AF diagnosis. The amount of heterogeneity explained by the model was moderate (R² = 23.1%), while residual heterogeneity remained high (I² = 100%) (Figure 4).

OPEN IN VIEWER

Figure 4.

Bubble plot of meta-regression assessing the association between timing of ICM implantation and time to AF detection. Each bubble represents a study, with size proportional to its weight in the meta-regression. Later ICM implantation was associated with longer time to AF diagnosis (β = 0.32; 95% CI: 0.14–0.51; p = 0.0007), even after adjustment for ICM monitoring duration.

Discussion

Our systematic review and meta-analysis, which included 47 studies and 6918 patients with ESUS or cryptogenic TIA/stroke, demonstrates that the timing of ICM implantation influences the diagnostic yield for AF detection. Indeed, we showed that earlier implantation of ICMs following the index cerebrovascular event was associated with higher AF detection rates and shorter times to diagnosis. These findings might support the notion that delays in initiating prolonged cardiac monitoring may reduce the clinical yield of ICMs and potentially defer opportunities for preventive anticoagulation.

One of the major elements of novelty of our analysis is that when stratifying studies based on the timing of ICM implantation—defined as early (<31.5 days from index event) vs delayed (⩾31.5 days)—we observed a higher pooled AF detection rate in the early group. Building on this, our stratified meta-regression using tertiles of implantation timing further reinforced the importance of early ICM placement. When categorizing studies into early (<14 days), intermediate (14–34 days), and delayed (⩾35 days) implantation groups and adjusting for ICM monitoring duration, we found that delayed implantation was consistently associated with lower AF detection rates even when compared to the intermediate group. In interpreting these findings, it is also important to consider the potential influence of baseline cardiovascular risk profiles. Although initially developed for patients with atrial fibrillation, the CHA₂DS₂-VASc score has been increasingly recognized as a marker of atrial cardiopathy and an independent predictor of new-onset AF in cryptogenic stroke or ESUS populations undergoing prolonged monitoring. Its inclusion as a potential confounder in our analysis reflects this broader predictive role beyond established AF. Moreover, while less invasive and less costly modalities such as prolonged Holter or external loop monitoring remain important for initial screening, further evidence consistently show that implantable loop recorders provide the highest diagnostic yield for delayed or paroxysmal AF.^2,15 Therefore, our findings should be interpreted within the context of patients in whom ILR implantation is clinically indicated, suggesting that earlier initiation may maximize detection and shorten diagnostic delay.

Our findings are aligned with emerging pathophysiological and clinical insights suggesting that AF may be particularly active or detectable in the acute phase following a cerebrovascular event. ¹⁶ The concept of “atrial fibrillation detected after stroke” (AFDAS) has gained traction as a distinct clinical entity characterized by a temporally heightened vulnerability to arrhythmia in the days to weeks after stroke or TIA.^17,18 In this context, a recent meta-analysis by Sposato et al. provided critical supporting evidence. ¹⁹ Their work included more than 7000 patients from both randomized trials and observational studies, 28% of them cryptogenic cases, undergoing serial ECGs, continuous inpatient ECG monitoring, continuous inpatient cardiac telemetry, in-hospital Holter monitoring, mobile cardiac outpatient telemetry, external loop recording, and implantable loop recording after an index event. The authors demonstrated that the odds of detecting AFDAS are over fivefold higher when continuous cardiac monitoring is initiated at a mean of 27 days after stroke than a mean of 75 days. Moreover, previous studies suggested that the early initiation of cardiac monitoring is crucial for improving atrial fibrillation diagnosis after stroke and TIA, with more than 70% of cases being diagnosed within the first 3 days after admission. ²⁰ It is unknown whether the higher AF diagnostic yield immediately after a cerebrovascular event might be explained by the pathophysiology of AFDAS. However, deferring monitoring might lead to miss transient or paroxysmal episodes of AFDAS that occur predominantly in the immediate post-stroke period. ¹⁶ In our analysis, the higher detection rate of AF observed with early monitoring may also reflect the underlying stroke pathophysiology. When AF is causally related to the index ischemic event, it is often intermittently active and thus more likely to be captured shortly after stroke. This interpretation underscores that the timing of monitoring and the pre-test probability of cardioembolic stroke are intrinsically linked, highlighting the importance of patient selection and timing in optimizing AF detection after ESUS or cryptogenic stroke.

In our analysis, each day of delay in ICM implantation from the index event was associated with an average increase of 0.33 days in the time to AF detection, even after adjustment for monitoring duration. This result should not be interpreted in absolute temporal terms, as the pooled estimate reflects heterogeneous definitions of “early” versus “delayed” implantation across studies. Rather, it reinforces the broader concept that earlier initiation of cardiac monitoring—typically within the first weeks after stroke—is associated with a higher likelihood of detecting AF during the period of increased atrial vulnerability following the index event. This result might suggest that earlier implantation of the device may not only improve overall AF detection rates but also accelerate the time to diagnosis. As anticoagulation in stroke patients with AF is the most effective strategy for secondary stroke prevention, the present findings have potentially important clinical implications. ²¹ Earlier identification of AF may facilitate more timely initiation of anticoagulant therapy, thereby minimizing the period during which patients remain at risk of recurrent cardioembolic events. In contrast, delayed ICM implantation may hinder the detection of brief, early post-stroke arrhythmic episodes—many of which are asymptomatic—ultimately limiting opportunities for effective intervention. These results reinforce the importance of prompt cardiac rhythm monitoring as a key component of individualized secondary prevention strategies in patients with ischemic stroke of undetermined source.

As expected, longer ICM monitoring duration was independently associated with higher AF detection rates, consistent with the established advantage of prolonged rhythm surveillance in this population. This further supports the overall utility of ICMs in post-stroke care pathways and highlights the diagnostic yield associated with sustained monitoring over time.

A major challenge emerging from our analysis is the lack of a uniform definition for AF episode duration across studies. The threshold used to define AF varied substantially, ranging from >30 s ²² to ⩾3 min. ²³ This inconsistency likely contributes to the wide variability in reported detection rates and limits the comparability of findings between cohorts. Establishing a standardized, clinically meaningful duration threshold for AF episodes is, therefore, essential to improve consistency in future research, facilitate meta-analytic synthesis, and ensure that AF detection after stroke is interpreted within a uniform diagnostic framework.

From a clinical standpoint, recent data from the ALONE-AF trial ²⁴ further emphasize the importance of precise rhythm monitoring. In that study, continuation of NOAC therapy after apparently successful AF ablation did not confer additional benefit, underscoring the uncertainty surrounding the prognostic meaning of brief or intermittent AF episodes. When discontinuation of oral anticoagulant therapy is considered, incorporating regular rhythm monitoring to detect AF recurrence is crucial, given the poor correlation between post-ablation AF burden and symptoms. In the trial setting, continuous invasive monitoring remains the gold standard for post-ablation surveillance, although prolonged intermittent monitoring with ambulatory ECG devices represents a practical alternative. Although this context differs from post-stroke AF detection, it reinforces a common principle across clinical scenarios—namely, that consistent, high-resolution rhythm monitoring is key to informing individualized anticoagulation strategies.

Our study has several strengths. All patients included across the 47 studies (n = 6918) met established definitions of cryptogenic stroke or ESUS. This homogeneous population enhances the clinical relevance and interpretability of our findings. By synthesizing data from these studies, we were able to explore both absolute AF detection rates and time to diagnosis using multiple analytic approaches, including stratified and continuous meta-regressions adjusted for ICM monitoring duration. Our findings were robust across sensitivity analyses and supported by consistent patterns in both binary and continuous exposure models. Moreover, we applied established statistical methods for proportion meta-analysis, heterogeneity exploration, and effect modification, lending methodological rigor to our estimates. Despite these strengths, several limitations should be acknowledged. First, the included studies were mostly observational and heterogeneous in design, population characteristics, and reporting standards. Although we employed rigorous data extraction procedures and conducted multiple sensitivity analyses, residual confounding and potential publication bias cannot be entirely excluded. Second, in a subset of studies, key timing variables (e.g., stroke-to-ICM interval or ICM-to-AF detection) were reported as medians rather than means. While we applied validated transformation methods (e.g., Wan et al.) to approximate means from medians, such conversions may introduce estimation error, particularly in the context of skewed distributions. Third, data on anticoagulation initiation timing or uptake following AF diagnosis were inconsistently available, limiting our ability to evaluate the downstream clinical implications of early detection. Studies differed in how they defined and reported AF episodes—particularly with respect to minimal duration thresholds required for AF classification—which may have influenced AF detection rates and introduced heterogeneity across studies. An important consideration is that patients who underwent delayed ICM implantation may represent a selected subgroup of the initially eligible population—specifically, those who had not already experienced AF detection or recurrent stroke by the time of implantation. This introduces the possibility of attrition bias, whereby individuals at higher baseline risk are systematically excluded from later cohorts due to early clinical events. As a result, the lower AF detection rate observed in the delayed group may not solely reflect the effect of timing but could be partially driven by this depletion of higher-risk individuals over time. At the same time, such delays in ICM implantation likely reflect real-world clinical practice, as the majority of included studies were observational and conducted across diverse international settings. While this introduces some methodological heterogeneity, it also enhances the external validity of our findings and helps contextualize the observed variability in ICM timing. Together, these factors underscore the importance of early monitoring—not only to maximize diagnostic yield but also to avoid missing critical windows of AF vulnerability in the acute post-stroke phase. In addition, a broader selection bias may exist at the population level, as ICM implantation was generally offered to older patients with higher CHA₂DS₂-VASc scores or features suggestive of atrial cardiopathy. Although a few cohorts reported recurrent stroke, heterogeneity in outcome definitions and limited data on anticoagulation after AF detection precluded estimation of outcome-based metrics such as the number of patients who would need ICM implantation to prevent one recurrent stroke. Nevertheless, the higher detection rate observed with early monitoring may not solely reflect a temporal biological effect. It may also result, at least in part, from selection and attrition mechanisms, as patients with stroke-related AF are more likely to be identified shortly after the index event, while those at lower risk remain arrhythmia-free over time. This interpretation is consistent with the possibility of selection bias discussed above and supports the view that both biological and methodological factors may jointly contribute to the observed difference between early and delayed implantation. An additional limitation is the variability in sample size across the included studies, ranging from 22 to 500 patients. This imbalance may have contributed to the heterogeneity observed in the pooled estimates. However, rigorous data extraction procedures, the use of random-effects models, and multiple sensitivity analyses were applied to mitigate this issue, and the consistency of findings across methods supports the robustness of our conclusions. Most included cohorts had a mean age below 75 years, reflecting the demographic profile of ESUS populations typically enrolled in cardiac monitoring studies. Consequently, the generalizability of our findings to very elderly or frail patients may be limited, as age-related differences in arrhythmia burden, comorbidities, and competing risks could influence both AF detection and clinical outcomes. Finally, our meta-regression analyses relied on study-level aggregate data rather than individual patient data (IPD), which restricts the precision of effect estimates and limits the capacity to assess interaction effects or confounding at the patient level. Future IPD meta-analyses may help overcome these limitations and further clarify the temporal dynamics between monitoring initiation and clinical outcomes.

In conclusion, this study underscores that earlier ICM implantation following cryptogenic or ESUS acute ischemic stroke is associated with both higher detection rates and shorter time to diagnosis of AF. These findings highlight the importance of prioritizing early ICM implantation as an integral component of secondary stroke prevention strategies. Future studies are warranted to validate these associations and explore whether earlier AF detection translates into reduced stroke recurrence and improved patient outcomes.

Supplemental Material

Supplemental Material

Supplemental Material

Supplemental Material

Supplemental Material