Validation and clinical application of implantable loop recorders for diagnosis of atrial fibrillation in horses

Authors: Sofie Troest Kjeldsen, Sarah D. Nissen, Nina C. Christensen, Simon L. Haugaard, Mélodie J. Schneider, Zenta Vinther, Stefan M. Sattler, Helena Carstensen, Christian Jøns, Charlotte Hopster‐Iversen, Rikke Buhl

Abstract

Background

Paroxysmal atrial fibrillation (pAF) occurs sporadically and can impair athletic performance. Gold standard for diagnosis is surface electrocardiography (ECG), however, this requires AF to be sustained. Implantable loop recorders (ILRs) are routinely used for AF detection in human medicine. While ILR placement has been studied in horses, its AF detection performance is unknown.

Objectives

(I) Validation of ILRs for AF detection in horses. (II) Determining pAF incidence using ILRs and estimate the positive predictive value (PPV).

Study design

(I) Experimental study; (II) Longitudinal observational study.

Methods

(I) Implantation of ILRs in 15 horses with AF and 13 horses in sinus rhythm. Holter ECGs were recorded 1, 4, 8, 12 and 16 weeks of AF. The ILR ECGs were compared with surface ECGs to assess diagnostic sensitivity and specificity. (II) Eighty horses (43 Warmbloods, 37 Standardbreds) with ILRs were monitored for 367 days [IQR 208–621].

Results

(I) ILRs detected AF on all recording days, in horses with AF, with a sensitivity of 66.1% (95% CI: 65.8–66.5) and a specificity of 99.99% (95% CI: 99.97–99.99). The sensitivity remained consistent across all time points. (II) The incidence of pAF was 6.3% (5/80). In horses with pAF, the PPV ranged from 8% to 87%. Increased body condition score (BCS > 6/9) was associated with an increased number of false positive episodes (p = 0.005).

Main limitations

(I) Horses were stabled during the ECG recordings, and AF was induced, rather than naturally occurring pAF. (II) Integrated algorithm in this ILR is optimised for AF detection in humans using remote monitors. Additionally, sensing is affected by motion artefacts.

Conclusion

The ILR reliably detected AF in resting horses, particularly in horses with normal BCS (6/9). The ILR proved useful to detect pAF and is recommended alongside Holter monitoring for diagnostic workup of horses with suspected pAF.

INTRODUCTION

Paroxysmal atrial fibrillation (pAF) is characterised by intermittent episodes of atrial fibrillation (AF) that spontaneously revert to normal sinus rhythm (SR). ^1^ , ^2^ , ^3^ During strenuous exercise, pAF can be detrimental to the performance level and may lead to significant increase in ventricular rates, increasing the risk of ventricular arrhythmias. ^3^ , ^4^ , ^5^ The sporadic arrhythmia may not manifest during the diagnostic workup of horses with poor performance, highlighting the importance of long‐term monitoring to detect sporadic arrhythmic events in horses with suspected pAF. Previously reported prevalences and incidences of pAF in racehorses have ranged from 0.03% to 4.9%, ^6^ , ^7^ but these reports may have underestimated the true occurrence, as they relied on post‐race surface electrocardiograms (ECGs). In a recent study using an implantable loop recorder (ILR) in a cohort of 12 horses with poor performance, pAF was diagnosed in 33%, including episodes detected while the horses were at rest. ^8^ In human cardiology, long‐term continuous monitoring devices including ILRs have been used for decades to detect arrhythmias, including pAF. ^9^ , ^10^ , ^11^ The Reveal LINQ (Medtronic, Denmark) shows a high AF duration sensitivity (98.4%) and specificity (99.5%) for AF detection. ^10^ However, sensing performance is reduced in individuals with higher body mass index. ^12^ ILRs have only recently been applied in equine medicine for monitoring horses with syncope, ^13^ , ^14^ detecting pAF in horses with poor performance ^8^ and monitoring AF recurrence after cardioversion. ^15^ Although the optimal placement of ILRs has already been investigated, ^16^ the sensitivity and specificity of ILRs for diagnosing AF in horses requires validation against the gold standard Holter ECG. Furthermore, ILRs' performance in horses of different breed and body condition scores (BCS) remains unclear to date. This study (I) To validate the ILRs ability to detect AF in horses, using Holter ECG monitoring as the gold standard; (II) To investigate the clinical application of ILRs for AF detection and investigate the incidence of pAF in a cohort of Warmbloods and Standardbreds of different disciplines, breeds, performance levels and BCS.

MATERIALS AND METHODS

Study samples

Sub study (I) Validation study

The AF group consisted of 15 horses (age 7.6 ± 2.0 years, weight 494 ± 51 kg, height 162 ± 5 cm), which had AF induced by atrial tachypacing using a neurostimulator (Itrel4, Medtronic) according to a protocol described elsewhere. ^17^ In brief, the horses underwent right atrial tachypacing with 10 Hz (600 stim/min). The pacing was continued for 2 weeks or until self‐sustained AF was achieved. The control group consisted of 13 horses in SR (age 11.6 ± 6.3 years, weight 531 ± 65 kg, height 160 ± 5 cm). Four of the control horses were later enrolled in the AF group (Figure 1). Before the experiments, all horses had 24‐h ECGs recorded with no clinically relevant abnormal arrhythmias. Blood samples for biochemical and haematological profiles were all within reference values. Finally, an echocardiographic examination was performed to evaluate cardiac structure and function. Mild valvular regurgitations were accepted in the inclusion process and was found in eight horses. None of the examinations resulted in the exclusion of any of the horses. All horses were recruited over a period of 12 months, had retired from racing and were owned by the Large Animal Teaching Hospital, University of Copenhagen with consent for use in research from the previous owners.

Sub study (II) Clinical application study

Horses were selected through convenience sampling and included according to the following Physically active and in regular training, Standardbred or Warmblood horses aged 2 years or older, moreover only horses with an ILR where the battery lasted until first interrogation were included. Initially 87 horses (45 Warmbloods and 42 Standardbreds), were included and had ILRs (33 reused, 54 new) implanted. However, seven horses were subsequently excluded from the sub study either due to unrelated deaths (n = 3) or because the ILR depleted its battery (n = 4) before the first interrogation (Figure 1). Of the 80 horses included, 24 horses (17 Standardbreds and 7 Warmbloods) had a history of poor performance, 47 horses (11 Standardbreds and 36 Warmbloods) had a normal performance level whereas the final 9 horses (Standardbreds only) had unknown performance status. Poor performance level was defined as unspecific fatigue during training or racing if they had encountered two or more episodes of suddenly reduced performance during training or racing within 1 month and had no recent history of lameness (Figure 1). All horses had unremarkable clinical examinations and resting ECGs. Further diagnostic investigations were performed on a number of horses, including exercise ECG (40/80 of horses) and echocardiography (60/80 of horses). No findings from the examinations resulted in the exclusion of horses from the sub study. The BCS of each horse was evaluated according to the Henneke scale. ^18^ The horses continued their normal training routine throughout the monitoring period. The included horses were active in the following low to moderate intensity exercise (dressage [n = 15] or driving [n = 13]); moderate to high intensity exercise (eventing [n = 14] or showjumping [n = 1]) and intensive exercise (racing [n = 37]). Two horses with pAF that previously were enrolled in other published studies were included in the PPV and incidence analysis in the current sub study as well. ^8^ , ^15^

Implantation and programming of loop recorders

In both studies, all horses had an ILR (Reveal LINQ, Medtronic) ^19^ inserted on the left side thorax, positioned between the fifth and sixth intercostal space, as previously described. ^15^ , ^16^ The ILR was programmed for the detection of ‘AF only’ with a ‘balanced sensitivity’ programme using the following nominal ectopy rejection, R‐wave sensitivity above 0.035 mV, blank after sense 150 ms and sensing threshold decay delay 150 ms. Minimum episode duration was set to include all episodes. Programming and subsequent data interrogation were performed using a programming device (CareLink™ 2090, Medtronic).

Study protocols

Sub study (I) Protocol for validation study

A modified base‐apex 3‐lead Holter ECG ^20^ (Televet100, Engel Engineering Service GmbH) was used to record 24‐h ECGs and served as the gold standard. To ensure data alignment, the ILR was interrogated just prior to the initiation of the 24‐h ECG, and data‐interrogation was precisely timed with the ending of the 24‐h ECG recording. The start and stop times of the ECG recordings were verified to synchronise with the ILR data. The control group had 24‐h ECGs recorded 2 and 6 weeks after the ILR implantation (Figure 2A). After AF induction, the AF group had 24‐h ECGs recorded after 1, 4, 8, 12 and 16 weeks. After recording, all ECGs were manually analysed (Televet 100 software, version 6.2.0, Engel Engineering Service GmbH), and minutes with noise artefacts persisting for more than 1 min on all three leads were excluded from the analysis. Atrial fibrillation was defined as irregularly irregular RR‐intervals, f‐waves and the absence of P‐waves. If any part of the recording was excluded from the surface ECG (due to noise artefacts or lead displacement), the exact same time period was also excluded from the ILR data even if AF was recorded by the ILR (Figure 2B). All ECG recordings from the ILR were visually inspected on the programmer and checked for artefacts and proper sensing. All episodes were analysed by a single observer (SK), with the involvement of a second observer (SDN or HC) in cases of uncertainty. An episode list containing time of AF‐onset and AF duration was extracted from the ILR. Each minute of 24‐h ECG recording/AF‐recording was categorised as True negative, false positive, false negative, or true positive (Figure 2B). The overall sensitivity and specificity were calculated as illustrated, with each minute of ECG recording contributing evenly to the calculation (Figure 2B). The AF horses in this study also participated in another study investigating AF. The enrolment of horses was consecutive, and consequently, some horses could only be examined at fewer timepoint resulting in unequal contributions of ECGs. To standardise data contributions across all horses, an individual mean sensitivity from each horse was calculated. During the 24‐h recording, the horses were confined to their stall where they were able to move freely around.

Sub study (II) Clinical application study

Captured event ECGs from the ILR were visually inspected and categorised into two True positive (AF), or false positive (atrial or ventricular premature depolarisations, motion artefacts, sinus arrhythmia and/or second‐degree atrioventricular blocks). Since the ILR only records ECG when AF is suspected, it will not record any false negative or true negative episodes, and therefore only the PPV can be estimated.

Data analysis

Statistical analysis was performed using GraphPad Prism (GraphPad Prism version 10.1.0) and R (version 4.0.2) software with the level of significance set at p < 0.05. All data were checked for normality using Shapiro–Wilk test and visual inspection of quantile‐quantile plots. Parametric data are presented with mean and standard deviation, while nonparametric data are presented with median and interquartile range or range when indicated. Sub study (I) A mixed‐effects analysis of repeated data was used to compare the sensitivity between the five time points. Sub study (II) Fisher's exact test was used to evaluate the relationship of AF and breed. Mann–Whitney test was used to compare the number of false positives between breeds and then compare BCS between breeds, separately. A linear regression analysis was used to test the correlation between BCS and false positive episodes. Positive predictive values (PPVs) were determined by assessing the proportion of correctly identified true positive episodes among the total number of true positives and false positives.

RESULTS

None of the horses in either of the studies exhibited any adverse reaction to the ILR.

Sub study (I) Validation study

At each recording day, the ILR successfully detected AF in all 15 horses with induced AF, confirmed by the 24‐h ECG. All horses in the AF group maintained AF during all recordings, verified from the Holter ECG. During nine recordings from six horses, the horse was tachypaced to sustain AF (Table 1). When examining minute‐to‐minute comparisons between Holter ECG recording and ILR sensing, the ILR demonstrated an overall sensitivity of 66.1% (95% CI: 65.8–66.5) and the specificity was 99.99% (95% CI: 99.97–99.99). These values were calculated based on the overall duration within each true 30 970 min, true 43 323 min, false 22 186 min, and false 3 min. The mean duration of the 24‐h ECGs was 23.1 h (±1.4 h) and the mean duration of readable ECG, without artefacts was 22.0 h (±1.8 h). The control group did not present with AF at any time, exerting neither true positive nor false negative minutes. The ILR incorrectly classified 3 min of sinus arrhythmia as AF, which were then classified as false positives. The mean sensitivity from each horse was used to calculate the individual mean sensitivity which was 63.2% (95% CI: 46–80) (Table 1). The sensitivity varied between the horses and there was no difference in sensitivity when comparing different durations of AF when performing a mixed‐effect analysis. Six of the horses failed to maintain AF on nine occasions during the study period, and were therefore tachypaced during the 24‐h ECG recordings. Overall, nine time points from six horses were excluded due to errors unrelated to either the ECG or ILR (Table 1). Incorrect episode categorisation by the ILR was mainly due to sensing of artefacts or detection of irregularity because of long pauses (above 2000 ms) with small amplitude P‐waves. Undersensing was rarely observed, and only few missensings were noted, none of which persisted for more than 1 min, therefore all episodes from the ILR were included in the analysis.

Sub study (II) Clinical application study

Paroxysmal AF was detected in 6.3% (5/80) of the horses, four Standardbreds and one Warmblood (Table 2). The horses were continuously monitored for a median duration of 367 days [208–621] with data being interrogated at a median of three times [range 1–6]. In total, 1065 ECG were stored as episodes on the ILR. Whereas 1640 episodes were stored as text files due to storage limitations without the use of remote monitoring. Among the recorded episodes with ECG data, 985 (92%) were false positives of which 686 (70%) were motion artefacts attributed to exercise estimated from higher heart rates (Figure 4B). The ILR detected 80 ECG episodes of pAF with a median episode duration of 8 min [4–20 min], where 58 of 80 episodes were ≥6 min (Table S1). All episodes of pAF were detected during periods when the horses were resting, exhibiting a low heart rate. The overall PPV was 8% (95% CI: 4%–17%). Individual PPVs for horses with pAF 8% (95% CI: 0.4%–33%), 50% (95% CI: 3%–97%), 55% (95% CI: 38%–72%), 67% (95% CI: 54%–78%) and 87% (95% CI: 70%–95%), respectively (Table S1). Warmblood horses had significantly more false positive episodes (p < 0.001, Mann–Whitney test) and significantly higher BCS (p < 0.001, Mann–Whitney test) than Standardbreds (Figure 4C and Table 2). Increasing BCS was correlated with increased number of false positive episodes (estimate = 8.73, p = 0.005) encountering breed as a variable in the linear regression. There was no difference in the incidence of pAF between breeds when performing Fisher's exact test. False positives resulted from misclassification due to non‐rhythm related factors (motion artefacts causing undersensing or oversensing) or rhythm‐related factors such as accelerated ventricular rhythm with irregular coupling, premature complexes either ventricular or atrial with poor sensing of the P‐wave, and sinus arrhythmia with poor sensing of the P‐wave (Figures 3 and 4B).

DISCUSSION

This study is the first to validate ILRs, a long‐term ECG monitoring device, as a supplementary tool for diagnosing pAF in horses. The ILRs successfully detected AF in 100% of the horses with experimentally induced AF on all recording days. With a sensitivity of 66.1%, the ILR was unable to capture every minute of AF due to its limited storage capacity. However, the very high specificity of 99.9% demonstrated that false positive detections were exceedingly rare in stabled horses. Clinically, the ILR detected pAF with an incidence rate of 6.3%. The ILR detected fewer false positive episodes among horses with a lower BCS, which may explain why Warmblood had more false positive episodes. Therefore, the ILR may prove valuable for improving the detection of pAF in racehorses.

Sensitivity and specificity

The sensitivity in this study differs from the reported 98.4% sensitivity in humans. ^10^ , ^11^ This discrepancy can be attributed to several First, the internal storage of the ILR is limited without use of remote monitors, and may lead to missed AF episodes; second, the ILR placement in horses is near the left triceps muscle can potentially introduce interference with the ILR whereas human ILRs are positioned parasternal on the thorax, ^21^ a position less susceptible to motion‐related disruptions; finally, the implantation method plays a significant role, as the creation of a large subcutaneous pocket may result in signal loss. Despite efforts to standardise the implantation technique, minor variations may have occurred.

To enhance detection sensitivity, the ILRs' AF detection threshold can be reprogrammed from ‘balanced’ to ‘more sensitive’, but this alteration carries the risk of increasing false positive detections, thereby reducing the specificity. ^11^ In a study by Buhl et al., the sensitivity threshold was programmed to ‘less sensitive’ to minimise false positive detections, but still resulted in several false positive episodes, potentially due to motion artefacts, as the horses were engaged in racing activities during the monitoring period. ^8^

In our study, the specificity was very high, with only one false positive episode detected in the control group, which is comparable to the specificity reported in human studies. ^10^ , ^11^ This finding can be attributed to the controlled and resting conditions to which the horses were stabled during the examinations, thus minimising the risk of motion artefacts and, consequently, false positive episodes. Similar outcomes were reported in a study validating the ILR positioning, where horses were monitored under similar controlled conditions, and no false positive episodes were observed. ^16^ The precision of the specificity might have been enhanced by monitoring all included horses five times instead of two times, however, practical considerations limited this possibility.

Episode categorisation

When testing ILRs in a clinical setting that involved training and racing, the current study uncovered a high prevalence of false positive episodes. From the estimation of heart rate, these false positives were primarily attributed to motion artefacts that occurred during exercise, leading to over‐ or undersensing by the ILR. Similar observations have been reported in previous studies ^8^ , ^15^ , ^16^ indicating the challenge of maintaining device sensitivity during physical activity. Despite this challenge, most riding and racing horses spent several hours stabled or walking in the field/paddock, which will allow proper sensing. The integration of a P‐wave algorithm has improved AF detection in humans. ^10^ Nevertheless, when the P‐wave amplitude is very low or when baseline noise is present, the ILR has difficulties in differentiating premature atrial complexes or sinus arrhythmia from AF. Moreover, previous studies have reported a sudden loss of R‐wave amplitude as the cause of false positives in approximately one third of human patients. ^21^ , ^22^ , ^23^ To overcome the issue of false positives in human patients, an artificial intelligence‐based algorithm using a deep neural network has shown significant improvement in specificity. ^24^

In a clinical setting, the number of false positives were significantly higher in horses with higher BCS, implying a better sensing in more lean horses, for example, Standardbreds. Lower BCS may increase the R‐wave amplitude and improve the AF detection algorithm. This phenomenon aligns with findings in human studies. ^12^ The overall PPV in sub study (II) was low, however, it should be noted that the PPV is influenced by the event rate, meaning that the PPV will be high in horses with more episodes of pAF, while horses without AF will have a PPV of 0% as only false positives will be detected. This study included horses with varying performance levels and found a low incidence of pAF, and thus a low PPV. Similarly, human studies have also reported performance limitations in cohorts with a low incidence of AF. ^22^ , ^23^

The main limitation of the ILR is the storage capacity, which allows automatic saving up to 14 episodes with ECG data. If the ILR records numerous short episodes, instead of fewer, longer episodes, the storage threshold is quickly reached, and the oldest episodes are only stored as text files. In the second part of this study, 1640 episodes were stored as text files and could not be analysed. More frequent data extraction could have reduced this number or, alternatively, ILRs could have been reprogrammed to store only episodes of longer duration, for example, >6 min to avoid storage of several short episodes. However, the latter approach could increase the risk of missing pAF episodes of short duration. A more advanced solution to overcome the problem of storage capacity could involve implementing remote monitors, which daily transfer data from the ILR to an online server. While the system has become an integrated part for monitoring human patients, its compatibility with the Reveal LINQ in horses still remains undetermined. Moreover, the ILR has a limited battery status of 3 years, which should be considered, especially if the device is reused.

Incidence of pAF detected using continuous monitoring

This study revealed a higher incidence of pAF than previously reported in studies using non‐continuous monitoring methods. ^6^ , ^7^ , ^25^ A previous study using ILRs reported an even higher incidence, but the study only monitored a smaller cohort of Standardbreds suspected of having AF. ^8^ In the current study, we monitored horses with and without poor performance and did not find Standardbreds to have a significantly higher incidence of pAF, despite their intense exercise level.

Recommended approach for clinical use

For optimal clinical utilisation, ILRs should be individually programmed for each patient, with the possibility of reprogramming to enhance AF episode detection. Initially, a good battery status should be ensured before insertion. Key considerations when programming ILRs include data extraction frequency, sensitivity thresholds, and P‐wave detection aggressiveness. It is essential to adjust the ILR settings to balance data collection and sensitivity, as more frequent data extraction allows for lower detection thresholds but may increase the incidence of false positives. Conversely, less sensitive thresholds could lead to missing pAF episodes. If an ILR detects many false positive episodes, reprogramming to increase the sensing threshold is recommended. The clinical goal is to identify horses with pAF, enabling veterinarians to identify the potential cause of poor performance. It has been experimentally demonstrated how AF severely affects heart rate and performance in horses. ^5^ However, the clinical significance of the AF burden in horses is currently a topic of ongoing debate. ^8^ , ^15^ While episodes lasting 5 min or more are deemed clinically significant in humans because of an increased stroke risk, ^26^ research investigating hypercoagulation in horses with AF has not proven this association. ^27^ Nevertheless, even short instances of pAF may hamper the performance level if occurring during racing. Further research is needed to understand the possible progression of pAF to persistent AF in horses, which may be dependent on the pathophysiology underlying the triggering mechanisms. It is important to note, that while ILRs can provide continuous monitoring for pAF over months or even years, they cannot replace the Holter ECG, which remains the gold standard for diagnosing arrhythmias in horses. Nevertheless, the use of ILRs can significantly enhance the chances of detecting this sporadic arrhythmia in horses.

Limitations

In sub study (I), all ECGs were obtained from stabled horses, which does not fully replicate clinical conditions where AF may occur during exercise. Moreover, the horses in this study had induced persistent AF, which resembles persistent AF more closely than pAF. This study exclusively assessed the settings of the Reveal LINQ for AF detection. The device offers additional functions and settings for detection of other arrhythmias, but these aspects were beyond the scope of this investigation.

CONCLUSION

In conclusion, this study found that ILRs with a dedicated AF algorithm effectively detects the presence and absence of AF in stabled horses with experimentally induced AF. In clinical conditions, the ILR demonstrated fewer false positive detections in horses with a normal BCS. In combination with surface ECG monitoring, ILRs are recommended for evaluating horses with poor performance suspected of having pAF. This technology has the potential to improve pAF diagnosis, a potential cause of poor performance in horses, and, more importantly, may help identify the cause of poor performance. Future validation should focus on remote monitoring with ILRs and explore the impacts of other ILR programming settings to enhance AF detection. Finally, future validation of ILRs in horses with naturally occurring pAF is needed.

FUNDING INFORMATION

Sarah D. Nissen was funded by the Independent Research Fund Denmark (Grant: 1032‐00053B). This study was generously funded by the Hans Nielsen Foundation.

CONFLICT OF INTEREST STATEMENT

Medtronic has generously sponsored the implantable loop recorders for the purpose of this study.

AUTHOR CONTRIBUTIONS

Sofie Troest Kjeldsen: Methodology; validation; visualization; writing – review and editing; project administration; software; data curation; conceptualization; investigation; funding acquisition; resources; formal analysis; writing – original draft. Sarah D. Nissen: Conceptualization; investigation; funding acquisition; methodology; validation; writing – review and editing; project administration; supervision; resources; data curation. Nina C. Christensen: Conceptualization; investigation; methodology; validation; software; project administration; formal analysis; visualization; data curation. Simon L. Haugaard: Investigation; writing – review and editing; methodology; software; project administration; supervision; data curation. Mélodie J. Schneider: Supervision; project administration; software; methodology; writing – review and editing; investigation. Zenta Vinther: Supervision; project administration; writing – review and editing; investigation; methodology. Stefan M. Sattler: Investigation; conceptualization; methodology; supervision; writing – review and editing; resources. Helena Carstensen: Investigation; writing – review and editing; supervision; methodology; validation. Christian Jøns: Conceptualization; investigation; validation; supervision; writing – review and editing. Charlotte Hopster‐Iversen: Writing – review and editing; methodology; conceptualization; investigation; validation; supervision; data curation; funding acquisition; project administration. Rikke Buhl: Supervision; writing – review and editing; methodology; conceptualization; investigation; funding acquisition; validation; project administration; resources.

DATA INTEGRITY STATEMENT

I, Sofie Troest Kjeldsen, confirm that all data are original and take full responsibility for the integrity and data analysis.

ETHICAL ANIMAL RESEARCH

All experiments were approved by the Danish Animal Inspectorate (licence no.: Sub study 2020‐15‐0201‐00425; sub study 2021‐15‐0201‐00823) and by the local ethics committee.

INFORMED CONSENT

Informed written consent was obtained from all horse owners before inclusion in both sub studies.

PEER REVIEW

The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer-review/10.1111/evj.14112.

Supporting information