Abstract
Background
Implantable loop recorders (ILRs) are increasingly placed for arrhythmia detection. However, historically, ≈75% of ILR alerts are false positives, requiring significant time and effort for adjudication. The LINQII and LUX‐Dx are remotely reprogrammable ILRs with dual‐stage algorithms using artificial intelligence to reduce false positives, but their utility in routine clinical practice has not been studied.
Methods and Results
We identified patients with the LINQII and LUX‐Dx who were monitored by the Veterans Affairs National Cardiac Device Surveillance Program between March and June 2022. ILR programming was customized on the basis of implant indication. All alerts and every 90‐day scheduled transmissions were manually reviewed. ILRs were remotely reprogrammed, as appropriate, after false‐positive alerts or 2 consecutive same‐type alerts, unless there was ongoing clinical need for that alert. Outcomes were total number of transmissions and false positives. We performed medical record review to determine if patients experienced any adverse clinical events, including hospitalization and mortality. Among 117 LINQII patients, there were 239 total alerts, 43 (18.0%) of which were false positives. Among 105 LUX‐Dx patients, there were 300 total alerts, 115 (38.3%) of which were false positives. LINQIIs were reprogrammed 22 times, resulting in a decrease in median alerts/day from 0.13 to 0.03. LUX‐Dx ILRs were reprogrammed 52 times, resulting in a decrease from 0.15 to 0.01 median alerts/day. There were no adverse clinical events that could have been identified by superior or earlier arrhythmia detection.
Conclusions
ILRs with artificial intelligence algorithms and remote reprogramming ability are associated with reduced alert burden because of higher true‐positive rates than prior ILRs, without missing potentially consequential arrhythmias.
Keywords: alerts, artificial intelligence, implantable loop recorders, remote monitoring, reprogramming
Subject Categories: Arrhythmias
Nonstandard Abbreviations and Acronyms
- ILR
implantable loop recorder
- VANCDSP
Veterans Affairs National Cardiac Device Surveillance Program
- VHA
Veterans Health Administration
Clinical Perspective.
What Is New?
New remotely reprogrammable implantable loop recorders with artificial intelligence algorithms have been developed, with the goal of reducing false‐positive alerts.
Our study shows that customizing alert criteria and remotely reprogramming implantable loop recorders that use artificial intelligence algorithms can reduce the number of false‐positive alerts and burden for clinicians, without missing clinically important events.
What Are the Clinical Implications?
Customizing alert criteria based on clinical indication can reduce implantable loop recorder alert burden.
Remotely reprogrammable implantable loop recorders with artificial intelligence–based algorithms may be more useful in the management of patients who have these devices implanted for multiple indications.
Implantable loop recorders (ILRs) are increasingly placed to detect arrhythmias for multiple indications, 1 but their monitoring has traditionally led to many alerts that require significant, burdensome clinical time commitment for adjudication. In a study of 26 713 patients with cardiovascular implantable electronic devices, half of alerts were from ILRs, although patients with ILRs comprised less than one‐fifth of the study population. 2 A well‐established problem is false‐positive alerts. In a 4‐week study of patients with ILRs for atrial fibrillation (AF) surveillance, cryptogenic stroke, and syncope who were programmed with nominal settings, the incidence rates of false‐positive transmissions were 46%, 86%, and 71%, respectively. 3 In another study, 74.2% of atrial tachycardia/AF and 76.8% of pause alerts were false positives. 4 Patient‐triggered transmissions are another common reason for false positives, including every single of 881 transmissions in a single‐center study, questioning the utility of these transmissions. 5 Cardiac device clinic personnel time for evaluating 1 ILR transmission averaged 15 minutes, and electrophysiologist time averaged 1.5 minutes. 6 Therefore, false positives present tremendous burden of time and resources. The estimated yearly staff time needed to manage a patient with an ILR was 7.7 to 9.3 hours, compared with just 1.6 to 2.4 hours for a patient with a pacemaker or implantable cardioverter‐defibrillator. 7
To address this false‐positive alert burden of ILRs, manufacturers have recently developed and implemented dual‐stage algorithms that detect arrhythmias through algorithms developed using artificial intelligence (AI) to adjudicate and remove false‐positive detections. For example, in June 2020, Boston Scientific received US Food and Drug Administration clearance for its LUX‐Dx (Boston Scientific, MN) ILR in which the AF and tachycardia algorithms were developed using machine learning, and in February 2022, Medtronic launched its AccuRhythm AI feature for LINQII (Medtronic, MN) ILRs focused on AF and pause alerts that was developed using deep learning. AI‐based algorithms have been shown to increase the positive predictive value of ILR‐detected AF alerts. 8 The implications of use of these algorithms in routine clinical practice, however, have not been studied. First, the number of alerts and their clinical actionability has not been quantified. Second, it is unknown if the greater filtering ability of AI algorithms causes clinically actionable arrhythmias to be missed, and therefore patients experience potentially avoidable adverse outcomes, such as stroke or syncope.
In addition, the remote programming ability of these ILRs allows iterative adjustments 9 to reduce alert burden without requiring in‐person visits. If these AI‐based algorithms and remote reprogramming ability are successful, then resulting alerts would more often be reliable and clinically actionable, reducing clinician time and costs from manually reviewing electrograms of false‐positive alerts without negatively affecting clinical outcomes. One recent study showed that 24% of LUX‐Dx ILRs were reprogrammed, with 82% of reprogramming occurring remotely. 10 Because there are no published evaluations of the clinical effects of new AI‐based algorithms and remote reprogramming ability for ILRs in routine clinical practice, we studied 2 ILRs with these capabilities, the LINQII and the LUX‐Dx, paired with a customized programming strategy and close follow‐up with reprogramming as indicated.
METHODS
Data Sources
The first data source was the Veterans Affairs National Cardiac Device Surveillance Program (VANCDSP), 11 which monitors some patients with ILRs for device clinics around the country, including interpreting all alerts and routine, scheduled transmissions. The second data source was the Veterans Health Administration (VHA) electronic health record for clinical information, including clinical decision‐making and patient follow‐up about alerts and routine transmissions from ILRs. The third data source was information provided by Medtronic (summary‐level data) and Boston Scientific (patient‐level data) about indications as well as transmissions and alerts for devices that were not monitored by the VANCDSP. The University of California, San Francisco, institutional review board granted a waiver of informed consent and approval for analyses of these data. Because of the sensitive nature of the data collected for this study, requests to access the data set from qualified researchers trained in human subject confidentiality protocols may be sent to Dr Dhruva at the San Francisco Veterans Affairs Medical Center.
Patient Selection
We included patients with the LINQII and LUX‐Dx who were monitored by the VANCDSP, with data abstracted after a 3‐month remote monitoring period. We also included 2 concurrent control groups of patients followed in VHA device clinics: (1) patients with the LINQII and LUX‐Dx who were not monitored by the VANCDSP (the goal of this control group was to determine the effect of both customized initial programming and close follow‐up with remote reprogramming as indicated) and (2) patients with LINQI (Medtronic), a prior generation ILR, who were also not monitored by VANCDSP (the goal of this control group was to additionally determine the effect of the new AI‐based algorithms). These control patients' ILRs were monitored contemporaneously.
We only included patients who sent at least 1 transmission during the study period. Monitoring for LINQII patients began on March 7, 2022 (5 clinical sites) and proceeded to include patients on March 9 (2 sites), March 28 (1 site), and April 21 (1 site), with the last transmissions for this evaluation captured on June 15, 2022. Monitoring for all LUX‐Dx patients began on March 23, 2022, with the last transmissions for this evaluation captured on June 27, 2022. Patients' medical records were reviewed during the monitoring period to look for possible clinical sequelae not accounted for by ILR transmissions. All patient medical records were also reviewed during the monitoring period and 1 month afterwards to identify any clinical actions that occurred in response to ILR transmissions.
ILR Programming
Patients with both the LINQII and LUX‐Dx were monitored according to specific settings (Table S1 for final settings at end of monitoring period), which were informed by a survey of practicing cardiac electrophysiologists within the Department of Veterans Affairs. These settings included multiple modifications compared with nominal manufacturer settings and were customized to the patient's indication for ILR placement. On May 11, 2022, multiple settings were updated. First, alerts for AF were initiated for all patients implanted for both syncope and ventricular tachycardia among patients with both ILRs, as had been initially intended. Second, AF management and AF ablation alerts were remotely reprogrammed from a 1‐ to 6‐hour burden in the LUX‐Dx (LINQII patients had been set to 6 hours) to ensure congruity for all patients with ILRs. Third, patient‐initiated transmissions for all patients with the LUX‐Dx were disabled (subsequently, patient‐initiated transmissions could be sent only as symptom with device‐detected events), as had been initially intended; before this time, 3 patients had sent patient‐initiated transmissions, which were immediately disabled. LINQII does not allow patient‐initiated transmissions unless associated with a detected episode. Finally, for all patients, the heart rate criterion for tachycardia alerts was changed from 170 beats to (230 beats–age) at 5 seconds for the LUX‐Dx to ensure greater congruity using age‐based criteria for all patients with ILRs.
Nominal LINQII and LUX‐Dx programming includes routine transmissions every 30 days; those patients monitored by the VANCDSP instead were programmed to send routine transmissions every 90 days. Thus, summary reports were generated only every 90 days; VANCDSP used 90 days, instead of 30 days, to better align with routine monitoring for other cardiac implantable electronic devices 12 and to reduce the overall burden of transmissions, given the expectation that alerts would detect actionable findings and there would be minimal, if any, clinical benefit from routine transmissions that were not prompted by alerts. Inadvertently, not all devices were switched to the 90‐day schedule at the start of the March/April 2022 monitoring periods.
ILR Monitoring Process and Patient‐Specific Reprogramming
All patients were centrally monitored by trained nurse and technician reviewers with expertise in interpretation of ILR interrogation and remote monitoring, with transmissions reviewed on all weekdays, excluding federal holidays. We prespecified a plan to adjust settings based on repeated true‐positive episodes or false‐positive findings (Tables S3 and S4). Prespecified clinically relevant findings identified by the reviewers resulted in direct email notification to local clinicians at VHA cardiology device clinics around the United States; these local clinicians were responsible for taking the appropriate clinical actions, as needed, for patient care. For repeated true‐positive notifications, clinicians were asked if they had taken clinical action and needed the ongoing notification; if ongoing notification was not required, then the specific alert was either deactivated or the alert criteria made more stringent through remote reprogramming. Remote reprogramming changes require up to 36 hours to be synced and implemented within a given ILR. Local VHA clinicians also had access to all ILR data, including alerts and routine transmissions, for their own independent review.
Covariates
Demographic factors (age, sex, and race) as well as both cardiovascular risk factors and established cardiovascular disease–related conditions were obtained from the VHA electronic health record. Location was determined on the basis of the VHA clinic associated with the patient's ILR care.
Outcomes
The primary outcome was the total number of remote transmissions (alert transmissions and scheduled downloads) during the monitoring period. The secondary ILR‐focused outcomes included the reason for alert transmissions and the true‐/false‐positive rates of all alerts. We also examined the number of times that the ILR was remotely reprogrammed and the prereprogramming and postreprogramming alert transmission rates. The secondary clinical outcomes were clinical actions performed by local cardiology device clinicians as a result of remote ILR transmissions (routine follow‐up that did not stem from ILR transmissions was not included).
An additional secondary outcome was any of the following clinical outcomes: death (including cause of death), stroke, syncope, pacemaker placement, implantable cardioverter‐defibrillator placement, and arrhythmia‐related emergency department visit or hospitalization (counted as hospitalization if occurred after an emergency department visit). These clinical outcomes were adjudicated by 2 cardiologists (M.H.R. and S.S.D.) to determine if they were caused by an arrhythmia that could have been identified sooner through a more sensitive ILR alert.
Among patients with ILRs not monitored by VANCDSP, we examined the total number of alerts, stratified by reason for ILR placement.
Statistical Analysis
We used descriptive statistics to summarize results. We used a 1‐way ANOVA test to compare the association of device type with or without VANCDSP monitoring with the mean number of alert transmissions. P<0.05 was considered statistically significant.
RESULTS
ILRs Monitored and Patient Characteristics
Overall, there were 354 patients with the LINQII, of whom 117 (33%) across 9 different clinical sites were monitored by the VANCDSP for a median (interquartile range [IQR]) of 98 (71–100) days (Table 1). There were 444 patients with the LUX‐Dx, of whom 105 (24%) across 19 different sites were monitored by the VANCDSP for a median (IQR) of 96 (89–96) days.
Table 1.
Characteristics of Patients With ILRs
| Characteristic | LINQII with VANCDSP monitoring | LINQII without VANCDSP monitoring | LUX‐Dx with VANCDSP monitoring | LUX‐Dx without VANCDSP monitoring | LINQI without VANCDSP monitoring |
|---|---|---|---|---|---|
| Patients, N | 117 | 237 | 105 | 339 | 2608 |
| Age, median (IQR), y | 69 (60–75) | 69 (62–74) | |||
| Sex, N (%)* | |||||
| Male | 108 (92.3) | 97 (92.4) | |||
| Female | 9 (7.7) | 8 (7.6) | |||
| Race, N (%)* | |||||
| American Indian or Alaska Native | 2 (1.7) | 2 (1.9) | |||
| Asian | 2 (1.7) | 1 (1) | |||
| Black | 24 (20.5) | 25 (23.8) | |||
| Hawaiian or Pacific Islander | 1 (0.9) | 0 (0) | |||
| White | 86 (73.5) | 71 (67.6) | |||
| Region, N (%) | |||||
| Northeast | 18 (15.4) | 9 (8.6) | |||
| Midwest | 43 (36.8) | 3 (2.9) | |||
| South | 28 (23.9) | 63 (60.0) | |||
| West | 28 (23.9) | 30 (28.6) | |||
| Comorbidities, N (%) | |||||
| Hypertension | 105 (89.7) | 89 (84.8) | |||
| Dyslipidemia | 106 (90.6) | 88 (83.8) | |||
| Diabetes | 48 (41.0) | 47 (44.8) | |||
| Coronary artery disease | 45 (38.5) | 34 (32.4) | |||
| Prior percutaneous coronary intervention | 18 (15.4) | 17 (16.2) | |||
| Prior coronary artery bypass graft surgery | 12 (10.3) | 12 (11.4) | |||
| Prior myocardial infarction | 13 (11.1) | 8 (7.6) | |||
| Atrial fibrillation | 61 (52.1) | 46 (43.8) | |||
| Peripheral vascular disease | 8 (6.8) | 15 (14.3) | |||
| Heart failure | 29 (24.8) | 21 (20.0) | |||
| Stroke | 40 (34.2) | 41 (39.0) | |||
| Prior ablation | 27 (23.1) | 15 (14.3) | |||
| Previous syncope | 34 (29.1) | 35 (33.3) | |||
| Indications for ILR placement, N (%) | |||||
| Cryptogenic stroke | 35 (29.9) | 75 (31.6) | 24 (22.9) | 68 (20.1) | 677 (26.0) |
| Syncope | 36 (30.8) | 80 (33.8) | 40 (38.1) | 149 (44.0) | 1013 (38.8) |
| Palpitations | 10 (8.5) | 17 (7.2) | 4 (3.8) | 30 (8.8) | 219 (8.4) |
| Atrial fibrillation management | 16 (13.7) | 22 (9.3) | 18 (17.1) | 34 (10.0) | 338 (13.0) |
| Suspected atrial fibrillation | 8 (6.8) | 28 (11.8) | 9 (8.6) | 37 (10.9) | 192 (7.4) |
| Post–atrial fibrillation ablation | 7 (6.0) | 4 (1.7) | 4 (3.8) | 7 (2.1) | 67 (2.6) |
| Ventricular tachycardia | 4 (3.4) | 9 (3.8) | 3 (2.9) | 6 (1.8) | 54 (2.1) |
| Other | 1 (0.9) | 2 (0.8) | 3 (2.9) | 8 (2.4) | 48 (1.8) |
| ILRs in place before start of evaluation period, N | 81 | 78 | |||
| ILR placed before start of evaluation period, median (IQR), d | 83 (42–115) | 176 (93–315) | |||
| ILRs placed after start of evaluation period, N | 36 | 27 | |||
| ILR placed after start period, median (IQR), d | 33 (17–55) | 40 (16–59) | |||
| Monitoring during study period, median (IQR), d | 98 (71–100) | 100 | 96 (89–96) | 96 (96–96) | 100 |
ILR indicates implantable loop recorder; IQR, interquartile range; and VANCDSP, Veterans Affairs National Cardiac Device Surveillance Program.
As categorized within Veterans Health Administration electronic health record.
The median (IQR) age of patients was 69 (60–75) years for the LINQII group and 69 (62–74) years for the LUX‐Dx group monitored by the VANCDSP. A total of 205 (92.3%) patients were categorized as men and 17 (7.7%) as women in the electronic health record. The most common comorbidities were hypertension (194 [87%]), dyslipidemia (194 [87%]), and AF (107 [48%]). The most common ILR indications for patients monitored by the VANCDSP were syncope (76 [34%]), cryptogenic stroke (59 [27%]), and AF management (34 [15%]). Indications were similar for LINQII and LUX‐Dx patients monitored by the VANCDSP compared with control patients.
Alert Transmission Volume
The LINQII group with VANCDSP monitoring sent 331 total transmissions: 183 alert transmissions and 148 scheduled transmissions (Table 2). Within the 183 alert transmissions, patients had 239 total alerts. A total 43 (36.8%) of the 117 patients in this group had at least 1 alert transmission over the median 98‐day period. Among these patients, the mean (SD) was 4.3 (4.3) alert transmissions per patient, with the highest for patients post‐AF ablation (mean, 8.6; SD, 5.6; Figure [A]).
Table 2.
ILR Transmissions and Alerts
| Variable | LINQII with VANCDSP monitoring (n=117) | LINQII without VANCDSP monitoring (n=237) | LUX‐Dx with VANCDSP monitoring (n=105) | LUX‐Dx without VANCDSP monitoring (n=339) | LINQI without VANCDSP monitoring (n=2608) |
|---|---|---|---|---|---|
| Total alert transmissions, N | 183 | 598 | 242 | 1759 | 9838 |
| Total alerts within alert transmissions, N | 239 | 300 | |||
| Total scheduled downloads, N | 148 | 233 | 89 | 669 | 2606 |
| Patient‐initiated transmissions, N | 0 | 5 | 3 | 83 | 5023 |
| Patients with ≥1 alert transmission, N (%) | |||||
| Cryptogenic stroke | 7 (20.0) | 11 (14.7) | 13 (54.2) | 25 (36.8) | 238 (35.2) |
| Syncope | 17 (47.2) | 41 (51.3) | 26 (65.0) | 103 (69.1) | 404 (39.9) |
| Palpitations | 4 (40.0) | 9 (52.9) | 2 (50.0) | 22 (73.3) | 113 (51.6) |
| Atrial fibrillation management | 6 (37.5) | 10 (45.5) | 6 (33.3) | 14 (41.2) | 186 (55.0) |
| Suspected atrial fibrillation | 3 (37.5) | 11 (39.3) | 8 (88.9) | 18 (48.6) | 99 (51.6) |
| Post–atrial fibrillation ablation | 5 (71.4) | 3 (75.0) | 3 (75.0) | 4 (57.1) | 38 (56.7) |
| Ventricular tachycardia | 1 (25.0) | 5 (55.6) | 1 (33.3) | 1 (16.7) | 28 (51.9) |
| Other | 0 (0) | 0 (0) | 1 (33.3) | 4 (50.0) | 20 (41.7) |
| Total/overall | 43 (36.8) | 90 (38) | 60 (57.1) | 191 (56.3) | 1126 (43.2) |
| Alert transmissions per patient among those with ≥1 alert, mean (SD) | |||||
| Cryptogenic stroke | 2.6 (1.3) | 4.1 (4.2) | 4.3 (3.3) | 8.2 (17.5) | 7.9 (13.1) |
| Syncope | 3.9 (4.7) | 4.2 (7.4) | 4.7 (6.7) | 7.2 (11.9) | 7.0 (12.6) |
| Palpitations | 3.0 (3.4) | 9.1 (12.5) | 5.5 (2.1) | 12.5 (23.9) | 9.7 (15.3) |
| Atrial fibrillation management | 4.3 (3.2) | 6.6 (5.9) | 4.0 (3.6) | 6.2 (6.8) | 12.7 (19.8) |
| Suspected atrial fibrillation | 5.3 (5.8) | 6.8 (8.3) | 2.0 (0.9) | 13.3 (20.7) | 9.7 (17.4) |
| Post–atrial fibrillation ablation | 8.6 (5.6) | 23.7 (29.2) | 2.3 (1.2) | 19.3 (26.1) | 11.1 (16.9) |
| Ventricular tachycardia | 1.0 (N/A) | 17.6 (28.4) | 2.0 (N/A) | 1 (N/A) | 9.4 (18.3) |
| Other | N/A | N/A | 1.0 (N/A) | 33.5 (41.4) | 0.1 (0.6) |
| Total/overall | 4.3 (4.3) | 6.6 (12.2) | 4.0 (4.9) | 9.2 (16.7) | 8.8 (15.2) |
| Type of alerts, N (%) | |||||
| True atrial fibrillation | 120 (86.3) | 79 (59.8) | |||
| False atrial fibrillation | 19 (13.7) | 53 (40.2) | |||
| True tachycardia | 63 (75.0) | 79 (63.7) | |||
| False tachycardia | 21 (25.0) | 45 (36.3) | |||
| True bradycardia | 4 (100) | 10 (100) | |||
| False bradycardia | 0 (0) | 0 (0) | |||
| True pauses | 9 (75.0) | 17 (50.0) | |||
| False pauses | 3 (25.0) | 17 (50.0) | |||
| Total alerts that were true positives, N (%) | 196 (82.0) | 185 (61.7) | |||
| Total alerts that were false positives, N (%) | 43 (18.0) | 115 (38.3) | |||
| Total patients with at least 1 true positive, N (% of patients with ≥1 alert) | 43 (100) | 44 (73.3) | |||
| Total patients with at least 1 false positive, N (% of patients with ≥1 alert) | 16 (37.2) | 35 (58.3) | |||
ILR indicates implantable loop recorder; N/A, not available because not calculable; and VANCDSP, Veterans Affairs National Cardiac Device Surveillance Program.
Figure 1. Mean alert transmissions per patients with at least 1 alert among patients with Medtronic LINQ (A) and Boston Scientific LUX‐Dx (B) implantable loop recorders.
Error bars represent SEs. AF indicates atrial fibrillation; and VANCDSP, Veterans Affairs National Cardiac Device Surveillance Program.
The LUX‐Dx group sent 343 total transmissions: 242 alert transmissions, 89 scheduled transmissions, 3 patient‐initiated transmissions, and 9 transmissions for symptoms without events (Table 2). Within the 242 alert transmissions, patients had 300 total alerts. A total 60 (57.1%) of the 105 patients had at least 1 alert transmission over the median 96‐day period. The mean (SD) was 4.0 (4.9) alert transmissions per patient among those who had an alert, with the highest for patients with an ILR placed for palpitations (mean, 5.5; SD, 2.1; Figure [B]).
In comparison, concurrent control patients with the LINQII who were not monitored by the VANCDSP had a mean of 6.6 (SD, 12.2) alert transmissions per patient with at least 1 alert during the same time period. The highest mean for LINQII for those who were not monitored by the VANCDSP was those with a post‐AF ablation indication (mean, 23.7; SD, 29.2). Concurrent control patients with the LINQI had a mean of 8.8 (SD, 15.2) alert transmissions per patient with at least 1 alert.
The LUX‐Dx without VANCDSP monitoring group had a mean of 9.2 (SD, 16.7) alert transmissions per patient with at least 1 alert during the same time period. Excluding 4 patients with other reasons for ILR placement, the highest mean for LUX‐Dx without VANCDSP monitoring was for patients with a post‐AF ablation indication (19.3; SD, 26.1).
One‐way ANOVA indicated there was a statistically significant difference in the mean number of alert transmissions across groups (P=0.0231).
False‐Positive Alerts
The rate of false‐positive alerts was 43 of 239 (18.0%) for the LINQII group monitored by the VANCDSP, including 13.7% of AF, 25.0% of tachycardia, 0% of bradycardia, and 25.0% of pause alerts (Table 2). For the LUX‐Dx group monitored by the VANCDSP, 115 of 300 (38.3%) alerts were false positives, including 40.2% of AF, 36.3% of tachycardia, 0% of bradycardia, and 50.0% of pause alerts. Of the 43 patients with the LINQII who sent at least 1 alert transmission, only 16 (37.2%) had ≥1 false‐positive transmissions. Similarly, of the 60 patients with the LUX‐Dx who sent at least alert 1 transmission, only 35 (58.3%) had ≥1 false‐positive transmissions.
Remote Reprogramming
Remote reprogramming was performed 22 times for 18 patients with LINQII monitored by the VANCDSP, most commonly for AF (18 times) (Table 3). Reprogramming was performed 52 times for 37 patients with LUX‐Dx monitored by the VANCDSP, most commonly for AF (34 times) and tachycardia (8 times). All patients with the LINQII whose ILR was reprogrammed multiple times had the reprogramming performed for the same reason. Of the 37 patients with the LUX‐Dx whose ILR was reprogrammed multiple times, 4 (10.8%) patients were reprogrammed for >1 reason. The most common alert adjustments were increasing AF episode or burden threshold durations after 2 notifications in the same range or turning off AF alert for chronic AF as well as increasing tachycardia alert durations after 2 notifications for supraventricular tachycardia and AF with rapid ventricular response. Multiple reprogramming in a single patient was performed with progressively more stringent criteria (eg, for AF detection until false positives attributable to premature atrial contractions decreased). For the LINQII, the median (IQR) time before and after reprogramming was 23 (17–60) and 37 (28–57) days, respectively. For the LUX‐Dx, the median (IQR) time before and after reprogramming was 23 (15–50) and 56 (25–74) days, respectively. For patients whose LINQII was reprogrammed, the median (IQR) number of alert transmissions per day decreased from 0.13 (0.09–0.24) before to 0.03 (0–0.06) after the first reprogramming. For LUX‐Dx patients, the median (IQR) alert transmissions per day decreased from 0.15 (0.05–0.36) to 0.01 (0–0.06).
Table 3.
ILR Reprogramming
| Variable | LINQII with VANCDSP monitoring | LUX‐Dx with VANCDSP monitoring |
|---|---|---|
| Times ILR remotely reprogrammed, N | 22 | 52 |
| Patients with ILR remotely reprogrammed, N | 18 | 37 |
| Reasons for remote reprogramming, N (%) of patients | ||
| Atrial fibrillation | 18 (14) | 34 (26) |
| Tachycardia | 4 (4) | 8 (6) |
| Bradycardia | 0 (0) | 2 (2) |
| Pause | 0 (0) | 6 (5) |
| Symptoms | 0 (0) | 2 (2) |
| Among patients who had remote reprogramming | ||
| Alert transmissions/d before first reprogramming, median (IQR) | 0.13 (0.09–0.24) | 0.15 (0.05–0.36) |
| Alert transmissions/d after first reprogramming, median (IQR) | 0.03 (0–0.06) | 0.01 (0–0.06) |
ILR indicates implantable loop recorder; IQR, interquartile range; and VANCDSP, Veterans Affairs National Cardiac Device Surveillance Program.
Clinical Actions From ILR Transmissions and Clinical Outcomes
Patients with both the LINQII and LUX‐Dx monitored by the VANCDSP had similar clinical sequela and outcomes (Table 4). Clinical actions that followed from transmissions for the LINQII and LUX‐Dx groups were notable for the following: 25 and 29 in‐person clinic visits, 11 and 26 telephone calls, and 7 and 5 anticoagulation medication changes (generally initiation of oral anticoagulants), respectively.
Table 4.
Clinical Actions and Clinical Sequelae for ILRs During 3‐Month Monitoring Period
| Variable | LINQII with VANCDSP monitoring | LUX‐Dx with VANCDSP monitoring |
|---|---|---|
| Transmissions leading to local VHA clinic notification, N | 107 | 113 |
| Actions followed from ILR transmissions as documented in EHR, N* | ||
| In‐person clinic visit | 25 | 29 |
| Telephone call | 11 | 26 |
| Video visit | 2 | 0 |
| ECG performed | 5 | 4 |
| Pacemaker implanted | 1 | 1 |
| ICD implanted | 0 | 0 |
| Oral anticoagulant change | 7 | 5 |
| Medication change other than oral anticoagulant | 14 | 7 |
| Catheter ablation | 1 | 1 |
| ILR explant | 1 | 2 |
| Cardioversion | 2 | 0 |
| Clinical outcomes between ILR transmissions | ||
| Death (and cause of death) | 0 | 1 (Suicide) |
| Syncope | 0 | 4 |
| Pacemaker placement | 0 | 0 |
| ICD placement | 1 | 0 |
| Stroke | 1 | 0 |
| Arrhythmia‐related emergency department visit | 2 | 0 |
| Arrhythmia‐related hospitalization | 0 | 1 |
EHR indicates electronic health record; ICD, implantable cardioverter‐defibrillator; ILR, implantable loop recorder; VANCDSP, Veterans Affairs National Cardiac Device Surveillance Program; and VHA, Veterans Health Administration.
*List is not exclusive, as some patients may have had multiple actions.
For the LINQII group monitored by the VANCDSP, clinical outcomes during the monitoring period not resulting from ILR transmissions were as follows: 2 arrhythmia‐related emergency department visits, 1 each of pacemaker and implantable cardioverter‐defibrillator placement, and 1 stroke. For the LUX‐Dx group monitored by the VANCDSP, there was 1 death, 4 episodes of syncope, and 1 arrythmia‐related hospitalization. After medical record review of all of these clinical sequelae by 2 cardiologists, it was determined that changes in ILR programming or transmission filtering would not have identified any diagnoses that could have prevented these outcomes. For example, the LINQII group's implantable cardioverter‐defibrillator placement was for primary prevention, and no arrythmia led to this placement. The 2 arrhythmia‐related emergency department visits were attributable to known AF. The episodes of syncope in patients with the LUX‐Dx were volume dependent or determined to be caused by medication, and not by arrhythmia.
DISCUSSION
We found that ILRs with dual‐stage algorithms that use AI decrease the well‐known problem of high alert burden, particularly with the aid of remote reprogramming. Compared with prior studies demonstrating false‐positive rates of ILR transmissions are often as high as 75%, 3 , 4 we found overall reductions in false‐positive alerts to 18.0% and 38.3% for 2 ILRs. Customized programming and remote reprogramming also reduced the median number of alerts per day, without the need for an in‐person clinic visit, as had been intended with the study design. There were no adverse clinical sequelae identified that could have been prevented through better arrhythmia detection. Given the significant burden of time that clinical personnel and electrophysiologists must contribute to managing ILR alerts, 6 , 7 this study indicates that ILRs with AI‐based algorithms and remote reprogramming ability have better utility in clinical practice.
In addition to remote reprogramming by changing alert thresholds and removing alerts once they were known, which reduced the number of transmissions, we also set the scheduled alerts to once every 90 days (instead of the nominal every 30 days). Although remote interrogation of ILRs can be reimbursed every 30 days, our goal was to align with the every 90‐day remote transmission schedule more common for pacemakers and implantable cardioverters‐defibrillators, 12 paired with the expectation that any significant abnormalities would be identified through alerts; this reduced the number of scheduled downloads by two‐thirds. In addition, our criteria for alerts, informed by a survey of cardiac electrophysiologists practicing in the Department of Veterans Affairs, were also more restrictive (ie, less likely to trigger an alert) compared with nominal ILR settings. Even if scheduled transmissions are reviewed every 30 days, they are likely to have a significantly smaller burden of data for review.
Remotely reprogramming ILRs can take significant time to determine and execute a change, but it is still more efficient than reprogramming a patient in the clinic, which requires additional patient and clinician time as well as delays and often ongoing false‐positive transmissions until an in‐person visit. Prior research has found >11 minutes of time spent per nonactionable ILR transmission. 7 The 2023 Heart Rhythm Society Expert Consensus Statement on Practical Management of the Remote Device Clinic gives a class 1 recommendation that ILR alerts be programmed tailored to clinical indications and reprogramming be performed if there is oversensing or undersensing. 13 This recommendation is supported by our findings demonstrating both a reduction in alerts overall with tailored programming that was more restrictive and then after remote reprogramming performed by the VANCDSP; in combination, these led to fewer alerts, which translate to less clinician time. The mean number of alert transmissions per patient with at least 1 alert was reduced by more than half among patients with the LUX‐Dx monitored and reprogrammed by the VANCDSP compared with LUX‐Dx that was not monitored by the VANCDSP, and by one‐third for patients with the LINQII monitored and reprogrammed by the VANCDSP compared with the LINQII not monitored and reprogrammed by the VANCDSP and even more compared with the LINQI. Although it is possible that the LUX‐Dx and LINQII not monitored by the VANCDSP were reprogrammed, prior research finding that only one‐fourth of these are reprogrammed 10 suggests it was infrequent.
Nominal ILR programming and alert criteria have traditionally erred on the side of being conservative. 3 Despite the decreased number of alerts, in a population of 222 patients followed up for 3 months, we did not identify any adverse clinical sequelae from potentially missed arrhythmias. As ILR battery life has extended and is now up to 4.5 years, 14 reducing false positives without missing important events is even more paramount because many patients may be followed up for up to a half decade.
Despite the improvements through dual‐stage algorithms that use AI, our study also demonstrates that there are still opportunities to reduce false‐positive rates. For example, the false‐positive rate for pauses was still 25% and 50% for the LINQII and LUX‐Dx groups, respectively. Even fewer false positives would reduce clinician burden. However, changes to algorithms should be accompanied by clinical evaluation, as we have performed here through manual medical record review of >200 patients, to ensure that no adverse clinical sequelae are missed (eg, true pauses that lead to syncope or sudden death are not missed).
Our study should be considered in the context of its limitations. First, we evaluated only a 3‐month period and had a modest sample size of 222 patients. Although we manually reviewed all medical records to determine if there were any adverse cardiac events, it is possible that longer follow‐up in a larger number of patients could detect a difference in adverse outcomes. Second, in our evaluation of patients monitored by the VANCDSP, we excluded patients who did not send a transmission; we did so because these patients were not connected at some point during the evaluation period. As a result, the average transmissions per patient are higher in these groups. However, the control groups did not exclude patients who were not connected. Thus, our results underestimate the benefit in reducing alert transmissions through AI‐based algorithms, our programming parameters, and remote reprogramming. Third, we only examined new clinical actions attributable to remote transmissions over a 3‐month period. However, sometimes having no abnormalities could inform clinical decisions by indicating that a current plan is working, and no new clinical action is needed (eg, patient does not develop AF, and thus does not require oral anticoagulation). Thus, we could be underestimating ILR utility over the multiple year lifespan of most ILRs. Furthermore, an ideal test would be to review all data before use of AI‐based algorithms, customized programming, or remote reprogramming and determine what may have led to clinical action. Fourth, it is possible that the LINQII and LUX‐Dx ILRs not monitored by the VANCDSP may have been remotely reprogrammed. However, we expect that this occurred infrequently. Fifth, the 3‐month evaluation period represented the VANCDSP's initial experience with ILR monitoring and with greater experience over time, we would expect to make further refinements to ILR programming to further reduce alert burden. Finally, we examined 2 ILRs that have reprogramming ability; additional ILRs with algorithms to reduce false positives have been developed by other manufacturers and may have additional clinical utility compared with earlier‐generation ILRs. 15 , 16 , 17
In summary, ILRs with dual‐stage algorithms that deploy AI, have the ability to remotely reprogram, and with customized initial programming can reduce false positives without missing clinically important events, and thus can better inform the diagnostic utility and management among patients receiving these devices in clinical practice.
Sources of Funding
This research was supported by the Department of Veterans Affairs Health Services Research and Development (1IK2HX003357) and University of California, San Francisco, Summer Explore Funding.
Disclosures
Dr Dhruva reported research support from the Department of Veterans Affairs and Arnold Ventures. He also reported serving on the Medicare Evidence Development and Coverage Advisory Committee and Institute for Clinical and Economic Review California Technology Assessment Forum. The remaining authors have no disclosures to report.
Supporting information
Tables S1–S4
Acknowledgments
We thank Roya Lisboa, CCT, CDRMS and Kristi Donovan, BSN for reviewing implantable loop recorder transmissions and Thomas L. Rotering, MPH for statistical assistance.
This article was sent to Barry London, MD, PhD, Senior Guest Editor, for review by expert referees, editorial decision, and final disposition.
Supplemental Material is available at https://www.ahajournals.org/doi/suppl/10.1161/JAHA.123.032890
For Sources of Funding and Disclosures, see page 10.
References
- 1. Bisignani A, De Bonis S, Mancuso L, Ceravolo G, Bisignani G. Implantable loop recorder in clinical practice. J Arrhythm. 2019;35:25–32. doi: 10.1002/joa3.12142 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. O'Shea CJ, Middeldorp ME, Hendriks JM, Brooks AG, Lau DH, Emami M, Mishima R, Thiyagarajah A, Feigofsky S, Gopinathannair R, et al. Remote monitoring alert burden: an analysis of transmission in >26,000 patients. JACC Clin Electrophysiol. 2021;7:226–234. doi: 10.1016/j.jacep.2020.08.029 [DOI] [PubMed] [Google Scholar]
- 3. Afzal MR, Mease J, Koppert T, Okabe T, Tyler J, Houmsse M, Augostini RS, Weiss R, Hummel JD, Kalbfleisch SJ, et al. Incidence of false‐positive transmissions during remote rhythm monitoring with implantable loop recorders. Heart Rhythm. 2020;17:75–80. doi: 10.1016/j.hrthm.2019.07.015 [DOI] [PubMed] [Google Scholar]
- 4. O'Shea CJ, Middeldorp ME, Hendriks JM, Brooks AG, Harper C, Thomas G, Emami M, Thiyagarajah A, Feigofsky S, Gopinathannair R, et al. Remote monitoring of implantable loop recorders: false‐positive alert episode burden. Circ Arrhythm Electrophysiol. 2021;14:e009635. doi: 10.1161/CIRCEP.121.009635 [DOI] [PubMed] [Google Scholar]
- 5. Chorin E, Peterson C, Kogan E, Barbhaiya C, Aizer A, Holmes D, Bernstein S, Schole M, Duraiswami H, Spinelli M, et al. Comparison of the effect of atrial fibrillation detection algorithms in patients with cryptogenic stroke using implantable loop recorders. Am J Cardiol. 2020;129:25–29. doi: 10.1016/j.amjcard.2020.05.027 [DOI] [PubMed] [Google Scholar]
- 6. Afzal MR, Nadkarni A, Niemet L, Houmsse M, Devgun J, Koppert T, Ferguson K, Mease J, Okabe T, Houmsse M, et al. Resource use and economic implications of remote monitoring with subcutaneous cardiac rhythm monitors. JACC Clin Electrophysiol. 2021;7:745–754. doi: 10.1016/j.jacep.2020.10.014 [DOI] [PubMed] [Google Scholar]
- 7. Seiler A, Biundo E, Di Bacco M, Rosemas S, Nicolle E, Lanctin D, Hennion J, de Melis M, Van Heel L. Clinic time required for remote and in‐person management of patients with cardiac devices: time and motion workflow evaluation. JMIR Cardiol. 2021;5:e27720. doi: 10.2196/27720 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Mittal S, Oliveros S, Li J, Barroyer T, Henry C, Gardella C. AI filter improves positive predictive value of atrial fibrillation detection by an implantable loop recorder. JACC Clin Electrophysiol. 2021;7:965–975. doi: 10.1016/j.jacep.2020.12.006 [DOI] [PubMed] [Google Scholar]
- 9. Stolen C, Rosman J, Manyam H, Kwan B, Kelly J, Perschbacher D, Garner J, Richards M. Preliminary results from the LUX‐dx insertable cardiac monitor remote programming and performance (LUX‐dx PERFORM) study. Clin Cardiol. 2023;46:100–107. doi: 10.1002/clc.23930 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Mahajan D, Frost K, Herrmann K, McGee‐Taylor R. Real‐world characterization and use of insertable cardiac monitor remote programming. J Innov Card Rhythm Manag. 2022;13:5230–5235. doi: 10.19102/icrm.2022.13112 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Muniyappa AN, Raitt MH, Judson GL, Shen H, Tarasovsky G, Whooley MA, Dhruva SS. Factors associated with remote monitoring adherence for cardiovascular implantable electronic devices. Heart Rhythm. 2022;19:1499–1507. doi: 10.1016/j.hrthm.2022.04.025 [DOI] [PubMed] [Google Scholar]
- 12. Slotwiner D, Varma N, Akar JG, Annas G, Beardsall M, Fogel RI, Galizio NO, Glotzer TV, Leahy RA, Love CJ, et al. HRS expert consensus statement on remote interrogation and monitoring for cardiovascular implantable electronic devices. Heart Rhythm. 2015;12:e69–e100. doi: 10.1016/j.hrthm.2015.05.008 [DOI] [PubMed] [Google Scholar]
- 13. Ferrick AM, Raj SR, Denke T, Kojodjojo P, Lopez‐Cabinallas N, Abe H, Boveda S, Chew DS, Choi JI, Dagres N, et al. 2023 HRS/EHRA/APHRS/LAHRS expert consensus statement on practical management of the remote device clinic. Heart Rhythm. 2023;20:e92–e144. doi: 10.1016/j.hrthm.2023.03.1525 [DOI] [PubMed] [Google Scholar]
- 14. LINQ II insertable cardiac monitor. Medtronic. 2022. Accessed September 20, 2023. https://www.medtronic.com/us‐en/healthcare‐professionals/products/cardiac‐rhythm/cardiac‐monitors/linq‐ii.html
- 15. Piorkowski C, Busch M, Nölker G, Schmitt J, Roithinger FX, Young G, Táborský M, Herrmann G, Schmitz D. Clinical evaluation of a small implantable cardiac monitor with a long sensing vector. Pacing Clin Electrophysiol. 2019;42:1038–1046. doi: 10.1111/pace.13728 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Ip J, Jaffe B, Castellani M, Sheikh A, Castellani C, Ip R. Accuracy of arrhythmia detection in implantable cardiac monitors: a prospective randomized clinical trial comparing Reveal LINQ and Confirm Rx. Pacing Clin Electrophysiol. 2020;43:1344–1350. doi: 10.1111/pace.14076 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Quartieri F, Marina‐Breysse M, Pollastrelli A, Paini I, Lizcano C, Lillo‐Castellano JM, Grammatico A. Artificial intelligence augments detection accuracy of cardiac insertable cardiac monitors: results from a pilot prospective observational study. Cardiovasc Digit Health J. 2022;3:201–211. doi: 10.1016/j.cvdhj.2022.07.071 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Tables S1–S4