- Email: [email protected]
Stroke incidence in patients with cardiac implantable electronic devices remotely controlled with automatic alerts of atrial fibrillation. A sub-analysis of the HomeGuide study
International Journal of Cardiology 219 (2016) 251–256
Contents lists available at ScienceDirect
International Journal of Cardiology journal homepage: www.elsevier.com/locate/ijcard
Stroke incidence in patients with cardiac implantable electronic devices remotely controlled with automatic alerts of atrial fibrillation. A sub-analysis of the HomeGuide study Renato Pietro Ricci a,⁎,1, Diego Vaccari b,1, Loredana Morichelli a,1, Gabriele Zanotto c,1, Leonardo Calò d,1, Antonio D'Onofrio e,1, Antonio Curnis f,1, Ennio C.L. Pisanò g,1, René Nangah h,1, Marco Brieda i,1, Vittorio Calzolari j,1, Donato Melissano k,1, Nicola Rovai l,1, Alessio Gargaro l,1 a
San Filippo Neri Hospital, Rome, Italy Civil Hospital, Montebelluna, Italy c Mater Salutis Hospital, Legnago, Italy d Casilino Hospital, Rome, Italy e V. Monaldi Hospital, Naples, Italy f Spedali Civili, Brescia, Italy g Vito Fazzi Hospital, Lecce, Italy h Civil Hospital, Portogruaro, Italy i Santa Maria degli Angeli Hospital, Pordenone, Italy j Cà Foncello Hospital, Treviso, Italy k F. Ferrari Hospital, Casarano, Italy l Clinical Department, BIOTRONIK Italy, Vimodrone, Italy b
a r t i c l e
i n f o
Article history: Received 19 April 2016 Accepted 12 June 2016 Available online 14 June 2016 Keywords: Pacemakers Implantable cardioverter defibrillators Remote monitoring Telemedicine Atrial fibrillation Stroke
a b s t r a c t Background: Remote Monitoring (RM) of cardiac implantable electronic devices (CIEDs) is recommended in management of Atrial Fibrillation (AF), which is a recognized risk factor for thromboembolism. We tried to elucidate whether stroke incidence observed in a large, remotely monitored population was consistent with the CHA2DS2VASc risk profile. Methods: Data from 1650 patients [76% male, age 72 (63–68), CHA2DS2VASc score 3.0 (2.0–4.0)] enrolled during the HomeGuide study and monitored with a daily-transmission RM system providing automatic alerts for AF, were analysed. Of those, 25% had a pacemaker and 75% an implantable cardioverter defibrillator with or without cardiac resynchronization. Estimations of the expected thromboembolic events were based on the population CHA2DS2VASc score profile used in a computer-simulated Markov model. Results: Eight thromboembolic events were observed with a 4-year cumulative stroke rate of 0.8% (confidence interval, 0.4%–1.5%). Simulations returned from 18.7 to 17.1 expected events, depending on the AF duration assumed to trigger anticoagulation (one-sample log-rank p b 0.03). During the study period, 681 (84%) AF episodes and 129 (16%) atrial tachycardias were detected in 291 patients (18%): 93% of episodes were detected remotely in 269 patients, 66% of whom had no history of AF. Medical interventions were necessary in 305 episodes, 85% of which were detected remotely. Reaction time was 1 (0–6) days for remotely-detected episodes and 33 (14–121) days for episodes detected in clinic (p b 0.0001). Conclusions: In a large CIED population followed remotely for up to 4 years, the incidence of thromboembolic events was less than half the estimations based on the CHA2DS2VASc risk profile. © 2016 Elsevier Ireland Ltd. All rights reserved.
1. Introduction
⁎ Corresponding author at: Cardiology Department, San Filippo Neri Hospital, Via Martinotti, 20, Roma, Italy. E-mail address: [email protected] (R.P. Ricci). 1 All authors take responsibility for all aspects of the reliability and freedom from bias of the data presented and their discussed interpretation.
http://dx.doi.org/10.1016/j.ijcard.2016.06.016 0167-5273/© 2016 Elsevier Ireland Ltd. All rights reserved.
One of the most appealing implications of Remote Monitoring (RM) of Cardiac Implantable Electronic Devices (CIED) is the potential of a fine tuning of medical therapy in response to early detection and notification of atrial fibrillation (AF). This hypothesis is applicable in standard practice [1,2] and the latest consensus statement of the Heart Rhythm
252
R.P. Ricci et al. / International Journal of Cardiology 219 (2016) 251–256
Society has set a class I recommendation with level of evidence A for the use of RM to early detect and quantify AF [3]. However there is still no evidence that a prompt intervention in response to early AF detection may be effective in reducing the incidence of thromboembolic events. The HomeGuide study is a recently completed large registry collecting data from routine RM application in CIED follow-up [4,5]. The study included 1650 subjects in 75 sites testing a standard pre-defined workflow model to implement RM in outpatient clinics. We further analyzed the study data to assess whether the stroke incidence observed during the follow-up period was consistent with expectations based on the CHA2DS2VASc risk [6] profile of the selected population. In addition we analyzed (1) the incidence of symptomatic and asymptomatic AF remotely detected in patients with and without prior history of AF; (2) the time from onset/detection to medical decision; (3) the type of medical interventions. 2. Methods 2.1. Study design, organizational model, HM alert programming for AF Patients with standard indication to pacemaker, implantable defibrillator (ICD), or cardiac resynchronization therapy (CRT) implant, who received a RM system (specifically, Home Monitoring [HM], BIOTRONIK SE & Co. KG, Berlin, Germany) at post-implant discharge, were enrolled in the HomeGuide study (Clinical Trials identifier NCT01459874) [4]. HM is characterized by daily automatic transmissions from the implant to a portable patient unit forwarding data to a central server via GSM (Global System for Mobile communication). Data include both technical and diagnostic information normally retrieved in a full device interrogation in hospital. HM transmissions occur every 24 hours automatically. In addition, alerts are triggered by selectable critical events and notified with email, fax or messages. Programming options of automatic alerts depend on the implanted device type. AF alerts are delivered within 24 h to the attending site staff and are generally triggered by long sustained episodes and/or a cumulative AF daily burden higher than a programmable threshold: options range from N 0% to N 75% of 24 h. AF notifications are not available in devices without atrial sensing. In addition to RM, patients recruited in the HomeGuide study were visited in hospital annually (biyearly in case of a CRT device). Participating sites were trained to adopt an organizational model based on a cooperative interaction within a nurse-physician couple univocally assigned to individual patients [4,7]. Concerning alert programming for and medical reaction to AF episodes, investigators were advised to apply the following scheme: first episode of AF; AF burden N10% for at least five consecutive days, AF burden = 100% for two consecutive days (persistent AF) [2]. Any option of such scheme (or others, if deemed clinically necessary) could be selected according to medical evaluation based on AF history of the individual patient and on optimal antithrombotic and/or antiarrhythmic therapy. Medical reaction to AF alert was left to investigators' discretion.
atrial sensing were directly downloaded from the HM central server and used to double-check investigators' AF episode reporting. For each reported AT/AF episode, investigators were asked to annotate patientreferred symptoms and associated medical decision/action. All severe adverse events occurring during the HomeGuide study had to be reported and documented by the investigators. These include any ischemic or hemorrhagic cerebrovascular accident (stroke) or transient ischemic attack (TIA). 2.3. Methods for estimating the expected stroke incidence Due to the lack of a control study arm, direct observations on stroke incidence were not possible. However crude estimations were made possible through computer simulations [8] using baseline characteristics, duration of study participation, AF incidence and related symptoms of individual study subjects as input data. Calculations were based on the CHA2DS2VASc score profile of the enrolled population used in a two-state Markov chain model within a Monte Carlo simulation (Fig. 1). The basic assumption was that the real study population was not monitored remotely with a RM system. Therefore, as described below, the attending physician could become aware of an AF occurrence only during an in-hospital evaluation either scheduled or triggered by patient-referred symptoms. Further assumptions were selected following a “best-case” principle, in order to retrieve the least expected number of stroke events at the end of the simulations: • Individual patients without history of AF would remain in a “healthy” state until onset of AF. This was a conservative assumption, as it is equivalent to the hypothesis that the transition probability to a “stroke” state would be zero until AF onset. • Upon new AF onset during the study period, patient-referred AFrelated symptoms would determine timing of OAT introduction: immediately (i.e. after 3 days in the simulation), in case of symptomatic AF; at the next in-hospital visit, in case of asymptomatic AF. • In case of AF history at baseline, the transition probability was set according to the subject's CHA2DS2VASc score, reduced by 22% or 62% [9] if the subject was already taking aspirin or oral anticoagulation therapy (OAT), respectively.
As a last assumption, AT episodes were excluded from the simulation, thus assuming that AT episodes would be not associated to any risk increase. The actual study period of each patient was simulated with 500 iterations each, to obtain estimates of the least number of stroke events, expected in that population, based on CHA2DS2VASc risk profile. Simulations were repeated several times. At first, all investigator-reported AF episodes were included, regardless of episode duration. In a second set of simulations, patients with single-chamber devices (without atrial sensing and diagnostic functions) were excluded and only device-detected AF episodes were included. In this second set of simulations, we assumed that patients would start OAT only with daily AF burden of 1%, 25% and 100% of 24-h.
2.2. Endpoints of the present analysis 2.4. Statistical analysis The primary objective of the main study was to evaluate the ratio of all major cardiovascular events remotely detected with HM over the number of all events that actually occurred during follow-up, including those which are not thought to be detected by a CIED. The endpoints for the present analysis were Atrial Tachycardia (AT)/AF episodes and strokes. ATs were defined as arrhythmic atrial episodes with regular atrial cycle and rate b 200 bpm; AF as any episode with irregular cycle or rate N200 bpm. AT/AF endpoints included investigator-reported episodes detected during regular or alert-triggered HM sessions, during inhospital device interrogation, other visits. AT/AF-related data (including episode duration and 24-hour burden) from implanted devices with
Continuous and categorical variables were reported as median (interquartile range) and percentages, respectively. Subgroup comparisons were performed using the Mann–Whitney statistics (for continuous variables), chi-squared or Fisher's exact tests as appropriate (for binary or categorical variables). Stroke/TIA cumulative rate in the study population (along with 95% confidence interval) was estimated with the product-limit method and the relative Kaplan–Meier curve was generated. The one-sample log-rank test was used to compare the stroke hazard rate observed in the study population with the expected estimations based on the CHA2DS2VASc score distribution as output of the Monte
R.P. Ricci et al. / International Journal of Cardiology 219 (2016) 251–256
253
Fig. 1. Markov chain model used for Monte Carlo simulations. In the simulated model, “H” stands for “healthy” state and “S” for “stroke” state. P is the transition probability. AF = atrial fibrillation; OAT = oral anticoagulation therapy.
Carlo simulations. All comparisons were considered statistically significant at or below the p = 0.05 level. Stata 12 (StatCorp, Texas, US) software was used for all statistical analysis. Custom-made Visual Basic routines were edited for Monte Carlo simulations. 3. Results 3.1. Study population Between 2008 and 2012, 1650 patients participated in the HomeGuide study for a median follow-up of 18 (10–31) months. Table 1 summarizes the main characteristics of the whole population and the subgroups of subjects with and without AF history at enrollment. Pacemakers, ICDs and CRT devices were roughly distributed in a 1:2:1 ratio. Fifty five percent of ICDs were single-chamber devices, but 117 of them had atrial sensing capabilities [10,11]. OAT was administered in 28% of the general population, antiplatelet in 45%: 359 of the 751 patients in the latter group (48%) had prior myocardial infarction. Twenty one percent of patients had history of AF at enrollment. This group more frequently had an indication for pacemaker implant, had a higher CHA2DS2VASc score, and was more likely to be on antiplatelet, OAT and antiarrhythmic therapy. Indeed aspirin was still recommended in lowrisk patients at the time of the study conduct. Of note, 40% of patients with history of AF were not treated with OAT, and 7% did not follow any antithrombotic prophylaxis. There was no statistically significant difference in terms of structural heart disease (except for valvular diseases), NYHA Class, ejection fraction, ventricular arrhythmias between the subgroups of patients with and without AF history at enrollment. 3.2. Incidence of stroke events During the HomeGuide study, 8 cerebrovascular accidents were observed: 2 ischemic strokes, 2 TIAs, 3 fatal and 1 non-fatal hemorrhagic strokes. All events occurred in patients on antiplatelet or anticoagulation therapy at baseline. The product-limit estimate of the 4-year cumulative stroke/TIA rate was 0.8% (confidence interval, 0.4%–1.5%). Several Monte Carlo simulations were run in order to assess if the number of observed cerebrovascular accidents was consistent with what could be expected as least estimations based on the CHA2DS2VASc
Table 1 Patient characteristics.
Age (years) Male (n, %) Device type (n, %) Pacemaker Single-chamber Dual-chamber ICD Single-chambera Dual-chamber CRT-D or -P Heart disease (HD) None Cardiomyopathy Ischemic HD Valvular HD Channelopathies Congenital HD Others NYHA class I II III IV LVEF (%) Ventricular tachyarrhythmias AF Paroxysmal Persistent Permanent CHAD2DS2VASc score Baseline therapy Antiplatelets Oral anticoagulants Antiarrhythmics
All patients 1650
With history of AF 356 (21%)
Without history of AF 1294 (79%)
P
72 (63–68) 1261 (76%)
74 (68–78) 274 (77%)
71 (62–77) 987 (76%)
b0.0001 0.78
421 (25%) 4 417 803 (49%) 444 359 426 (26%)
106 (30%) 3 119 160 (45%) 111 49 90 (25%)
315 (24%) 1 298 643 (50%) 333 310 336 (26%)
0.04
0.79
183 (11%) 851 (53%) 689 (43%) 107 (7%) 27 (2%) 9 (1%) 5 (0%)
32 (9%) 173 (49%) 152 (43%) 44 (12%) 2 (1%) 1 (0%) 0 (0%)
151 (12%) 678 (52%) 537 (41%) 63 (5%) 25 (2%) 8 (1%) 5 (0%)
0.15 0.21 0.70 b0.001 0.07 0.44 –
309 (19%) 751 (46%) 553 (33%) 37 (2%) 30.0 (25.0–42.0) 447 (27%)
67 (19%) 150 (42%) 127 (36%) 12 (3%) 31.0 (26.0–45.0) 92 (26%)
242 (19%) 601 (46%) 426 (33%) 25 (2%) 30.0 (25.0–41.5)
0.95 0.14 0.33 0.10 0.39
355 (27%)
0.57
356 (22%)
356 (100%) 155 (44%) 71 (20%) 130 (37%) 4.0 (3.0–5.0)
0 (0%)
–
3.0 (2.0–4.0)
b0.001
118 (33%) 215 (60%) 113 (32%)
623 (48%) 252 (19%) 282 (22%)
b0.001 b0.001 b0.001
3.0 (2.0–4.0) 741 (45%) 467 (28%) 395 (24%)
0.11
AF = atrial fibrillation; CRT-D (-P) = cardiac resynchronization therapy with defibrillation (pacing) function; ICD = implantable defibrillator; HD = heart disease; LVEF = left ventricular ejection fraction. a Including 117 single-chamber ICDs with atrial sensing capability.
254
R.P. Ricci et al. / International Journal of Cardiology 219 (2016) 251–256
score of the study subjects. The results are shown in Fig. 2 and further detailed in Table 2. Simulations returned a significantly higher number of expected events, as compared to the real occurrences: it ranged from 18.7 events (confidence interval, 18.4–19.1) when we included all patients and all AF episodes (regardless of their duration in terms of daily AF burden), to 17.1 (16.7–17.4), when we excluded pure single-chamber devices and limited OAT introduction only upon atrial episodes causing a 24-hour AF burden = 100%. Interestingly, simulation outputs were quite similar independently of the AF burden thresholds chosen to introduce OAT: a slight difference of only 0.4 expected stroke/TIA events was obtained, on average, between 10% and 100% AF burden. Comparisons with the real occurrences during the followup period were statistically significant at the one-sample log-rang test both including all bleeding and ischemic strokes (8 events) or only ischemic strokes or TIAs (4 events). 3.3. Incidence of symptomatic and asymptomatic AF During the study period, 810 atrial episodes were detected in 291 patients (18%): 681 (84%) were classified as AF, 129 (16%) as AT (Table 3). The majority of AT/AF episodes (93%) were detected remotely during HM sessions. Of note, 193 of the 291 patients with AT/AF episodes (66%) did not have history of AF at enrollment. Patients did not report any AF-related symptoms in 72% of episodes. Asymptomatic episodes were mostly detected remotely by HM (96%) and frequently recurred in patients without history of AF (142/213 episodes, 67%). Medical reactions were required in 305 of 810 episodes, 85% of which were detected remotely. For episodes detected remotely, medical decision was taken 1 (0–6) days after episode onset, as compared to episodes detected in-hospital: 33 (14–121) days and 40 (17–133) including only N24-hour AF episodes (p b 0.0001, for both comparisons). Table 4 shows the interventions in detail. Change in pharmacological
Fig. 2. Kaplan–Meier curve of stroke/TIA free rate. Dotted line shows the expected level of the Kaplan–Meier survival curve. Calculations were performed with the Monte Carlo method based on the CHA2DS2VASc risk profile of the study population. TIA = Transient ischemic attack.
therapy was the most frequent intervention: in 111 cases (61%) the ongoing therapy was optimized; in particular OAT, betablockers, and amiodarone were introduced in 39, 43 and 34 cases (21%, 24%, and 19%), respectively. A vast majority of interventions on medical therapy (159, 87%) were taken in response to episodes detected remotely with HM. 4. Discussion 4.1. Main results The main results of our analysis may be summarized as follows: 1) in the HomeGuide study population systematically followed up with a RM system characterized by daily automatic transmissions the incidence of thromboembolic events was less than half of estimation based on the CHA2DS2VASc risk profile of the population; 2) 94% of AT/AF episodes were detected remotely, 72% were asymptomatic and mostly occurred in patients with no history of AF (66%); 3) 85% of medical interventions for AF were triggered by RM notifications, which ultimately drove therapy and device optimization, cardioversions, ablations or hospitalizations. 4.2. Clinical implications Our results do not prove the efficacy of RM as a tool to prevent thromboembolic events, as computer simulated models cannot surrogate results from randomized prospective trials, no matter how realistic and accurate the assumptions are. However simulations can provide indications justifying further evaluations. As a matter of fact, there were significantly less ischemic or bleeding stroke events during the study period than it could be expected based on the CHA2DS2VASc score risk of individual patients. The CHA2DS2VASc score is a well-known validated and recommended risk stratification index [12,13], as it is non-inferior to, and possibly better than, previous scores such as CHADS2 in identifying patients who develop stroke and thromboembolism, and it is superior in identifying ‘truly low-risk’ patients with AF [14,15]. Estimations of the expected number of thromboembolic events in the study population were performed using a simple 2-state Markov chain model including the individual patients' CHA2DS2VASc scores for the transition probabilities, possibly reduced by the administration of antithrombotic therapy. The least number of assumptions was introduced in the simulated model in order to keep it sufficiently accurate but not critically dependent on arbitrary constraints. The basic assumption was that the real population was not monitored remotely and physician's evaluation of an AF episode could only take place if patients reported symptoms or during the next actually performed follow-up visit. It should be emphasized that all other necessary assumptions, despite generating crude approximations, were always based on a ‘best-case’ scenario, selecting conditions associated with the lowest, even if unrealistic, event probabilities. For example, by setting the Markov transition probability to 0 in the absence of AF episodes, we implicitly assumed that thromboembolic strokes may only be caused by AF. This assumption may be unrealistic but it tended to underestimate the expected event rate in the scenario of no RM. Similarly, it was assumed that patient-reported symptoms would always trigger an immediate introduction of OAT within 3 days, and that AT episodes were not associated with an increased risk of stroke. These assumptions aimed at retrieving a minimal rather than realistic estimation of stroke incidence. Nevertheless, the expected stroke number returned by the Monte Carlo simulations was always higher than that actually observed, regardless of the AF burden threshold assumed necessary to trigger OAT. AF burden threshold that can justify the introduction of OAT is still a matter of debate, as it has not been precisely identified in the range of few minutes up to several hours [16]. Monte Carlo simulations were therefore repeated with different assumptions on AF burden thresholds
R.P. Ricci et al. / International Journal of Cardiology 219 (2016) 251–256
255
Table 2 Expected number of embolic events. All observed ischemic and bleeding strokes (8)
Observed ischemic strokes/TIA(4)
Assumed AF episode duration triggering OAT
Expected
Tolerance
Chi squared
P
Chi squared
P
All investigator-reported AF Daily AF burden ≥1% Daily AF burden ≥25% Daily AF burden ≥100%
18.72 17.70 17.49 17.07
0.35 0.35 0.33 0.34
6.14 5.32 5.15 4.82
0.013 0.021 0.023 0.028
3.40 10.60 10.40 10.01
0.0007 0.0011 0.0013 0.0016
Results of the Monte Carlo simulations based on the Markov chain model are shown in Fig. 1. The expected number of strokes in the study population and the respective tolerance (semilength of the confidence interval) are reported and compared with the one-sample logrank test to the number of all ischemic and bleeding events (8) and ischemic stroke/TIAs (4) actually observed. Simulations were repeated for different assumptions on the AF incidence triggering antithrombotic therapy in the simulated model: first all investigator-reported AF was included; then only AF episodes generating a daily arrhythmic burden ≥1%, ≥25%, ≥100%. AF = atrial fibrillation; TIA = Transient ischemic attack.
Table 3 Classification of atrial tachyarrhythmia episodes. AT/AF episodes
All episodesa
Remotely detected during HM sessionsb
Detected during in-hospital visits/otherb
No. of episodes AF AT No. patientsc Without history of AF Asymptomatic episodes No. of patients Without history of AF Actionable episodes
810 681 (84%) 129 (16%) 291 193 583 (72%) 213 142 305 (38%)
753 (93%) 640 (94%) 113 (88%) 269 179 (66%) 559 (96%) 200 136 258 (85%)
57 (7%) 41 (6%) 16 (12%) 50 34 (68%) 24 (4%) 22 14 47 (15%)
AF = atrial fibrillation; AT = atrial tachycardia; HM = Home Monitoring. a Percentages were calculated with respect to the first row. b Percentages were calculated with respect to the first column. c In 28 patients different AT/AF episodes were detected both remotely and in-hospital.
to be exceeded in one single day in order to let OAT be introduced. After a first simulation including all the AF episodes reported by the investigators regardless of their durations, three more simulations were performed based on device-reported daily AF burden thresholds of 1%, 25%, and 100%, corresponding to AF durations of 14 min, 6 h, and 24 h, respectively. It is worth noting that simulations returned very similar
estimations regardless of AF burden thresholds: from the 18.7 expected events when all the patients and all the investigator-reported AF episodes were included, to 17.1 expected events when only devices with atrial sensing were included and only episodes with 100% AF burden were assumed necessary to trigger OAT. The small difference of less than 2 expected stroke events obtained in a 1650 population in an 18-
Table 4 Medical interventions to AT/AF episodes. Action
All episodesa
Episodes remotely detected during HM sessionsb
Episodes detected during in-hospital visits/otherb
Therapy optimization Suspension/change dosage Introduction of Betablockers Antithrombotic therapy Amiodarone Diuretics Class IC antiarrhythmics Sotalol Digitals Ca-antagonists ACE-inhibitors Ag II receptor blockers Other drugs Encouraging patient to improve adherence to therapy Device reprogramming Cardioversion DC shock Pharmacological RF ablation Hospitalization Others
182 111 (61%)
159 (87%) 106 (95%)
47 (13%) 5 (5%)
43 (24%) 39 (21%) 34 (17%) 13 (7%) 6 (3%) 5 (3%) 5 (3%) 4 (2%) 3 (2%) 3 (2%) 40 (22%) 20 (11%) 65 (36%)
40 (93%) 34 (87%) 32 (94%) 11 (85%) 6 (4%) 5 (3%) 5 (3%) 4 (2%) 3 (2%) 3 (2%) 38 (24%) 19 (95%) 54 (83%)
3 (7%) 5 (13%) 2 (6%) 2 (15%) 0 (0%) 0 (0%) 0 (0%) 0 (0%) 0 (0%) 0 (0%) 2 (3%) 1 (5%) 11 (17%)
33 (18%) 6 (3%) 11 (6%) 24 (13%) 14 (8%)
25 (76%) 4 (66%) 8 (73%) 15 (62%) 13 (93%)
8 (24%) 2 (33%) 3 (27%) 9 (38%) 1 (7%)
More medical interventions were required in some cases. AF = atrial fibrillation; AT = atrial tachycardia; DC = direct current; HM = Home Monitoring; RF = radiofrequency. a Percentages were calculated with respect to the first row. b Percentages were calculated with respect to the first column.
256
R.P. Ricci et al. / International Journal of Cardiology 219 (2016) 251–256
month median follow-up is clearly a consequence of the very asymmetrical and skewed distribution of AF and probably explains the current uncertainty in identifying the proper AF burden threshold. The relation between AF and thromboembolism is more complex than initially suspected. It is well known that AF is not the only mechanism potentially involved in pathogenesis of thromboembolism. In addition, two recent sub-analyses from the ASSERT [17] and the IMPACT [18] studies have demonstrated a substantial temporal disconnection between AF onset and stroke events, in contrast with the idea of a simple cause–effect model. Such findings clearly reflect the current incomplete understanding of the clinical relevance of (non-)sustained AF and mechanisms underlying thromboembolism. Regardless of whether AF-stroke relation is direct or mediated by other mechanisms, the association between AF and increased stroke risk is well established. Therefore, the theoretical basis of early AF detection with RM to trigger prompt medical reaction (e.g. anticoagulation therapy) is worth investigating further. The recently published IMPACT study was a first attempt toward this objective but it failed to demonstrate any benefit from an ‘on/off’ OAT strategy driven by continuous remote monitoring of AF episodes as compared with a conventional approach [18]. However, the study suffered from several limitations both in design and compliance, preventing straightforward conclusions. Patients randomized to the active study arm discontinued OAT after 30 days (CHADS2 1 or 2) or 90 days (CHADS2 3 or 4) without AF recurrences. More than 90% of patients with stroke at the time of study end either were not under OAT or had suboptimal INR values. In our series we observed a lower than expected rate of thromboembolic events. We have no data to support the hypothesis that this was the result of an optimized therapy management driven by RM findings without any pre-specified antithrombotic therapy plan. Nevertheless, such a low stroke incidence was observed in an unselected population with remotely controlled CIEDs, in which 94% of AF were detected and notified by the RM system, and 85% of interventions for AF (including 21% of OAT introduction) were triggered by a RM notification of AF with a median reaction time of 1 day. These figures might reasonably play a role in the explanation of the low stroke incidence reported in our population, even if we cannot exactly estimate to what extent. 5. Conclusions We observed a lower than expected incidence of thromboembolic events in an unselected CIED population remotely controlled with HM and in which most AF episodes were detected and managed upon early RM notification. Monte Carlo computer simulations based on the population CHA2DS2VASc risk profile showed that the stroke incidence observed in the population was less than half of the expected rate, irrespective of the AF burden threshold assumed to trigger OAT. Conflicts of interest RPR: minor consultancy fees by Medtronic NR and AG: employees of Biotronik Italy DG, LM, GZ, LC, ADO, AC, ECLP, RN, MB, VC and DM: nothing to disclose
Acknowledgments The authors would like thank Xenia Antoniou for her precious advice in text revision. References [1] F. Zoppo, D. Facchin, G. Molon, et al., Improving atrial fibrillation detection in patients with implantable cardiac devices by means of a remote monitoring and management application, Pacing Clin. Electrophysiol. 37 (12) (Dec 2014) 1610–1618. [2] R.P. Ricci, L. Morichelli, M. Santini, Remote control of implanted devices through Home Monitoring technology improves detection and clinical management of atrial fibrillation, Europace 11 (1) (Jan 2009) 54–61. [3] D. Slotwiner, N. Varma, J.G. Akar, et al., HRS Expert Consensus Statement on remote interrogation and monitoring for cardiovascular implantable electronic devices, Heart Rhythm. 12 (7) (Jul 2015) e69–100. [4] R.P. Ricci, L. Morichelli, A. D'Onofrio, et al., Effectiveness of remote monitoring of CIEDs in detection and treatment of clinical and device-related cardiovascular events in daily practice: the HomeGuide Registry, Europace 15 (7) (Jul 2013) 970–977. [5] R.P. Ricci, L. Morichelli, A. D'Onofrio, et al., Manpower and outpatient clinic workload for remote monitoring of patients with cardiac implantable electronic devices: data from the HomeGuide Registry, J. Cardiovasc. Electrophysiol. 25 (11) (Nov 2014) 1216–1223. [6] G.Y. Lip, R. Nieuwlaat, R. Pisters, D.A. Lane, H.J. Crijns, Refining clinical risk stratification for predicting stroke and thromboembolism in atrial fibrillation using a novel risk factor-based approach: the euro heart survey on atrial fibrillation, Chest 137 (2) (Feb 2010) 263–272. [7] R.P. Ricci, L. Morichelli, M. Santini, Home monitoring remote control of pacemaker and implantable cardioverter defibrillator patients in clinical practice: impact on medical management and health-care resource utilization, Europace 10 (2) (Feb 2008) 164–170. [8] R.P. Ricci, L. Morichelli, A. Gargaro, M.T. Laudadio, M. Santini, Home monitoring in patients with implantable cardiac devices: is there a potential reduction of stroke risk? Results from a computer model tested through Monte Carlo simulations, J. Cardiovasc. Electrophysiol. 20 (11) (Nov 2009) 1244–1251. [9] G.Y. Lip, R.G. Hart, D.S. Conway, Antithrombotic therapy for atrial fibrillation, BMJ 325 (7371) (Nov 2 2002) 1022–1025. [10] C. Sticherling, M. Zabel, S. Spencker, et al., Comparison of a novel, single-lead atrial sensing system with a dual-chamber implantable cardioverter-defibrillator system in patients without antibradycardia pacing indications: results of a randomized study, Circ. Arrhythm. Electrophysiol. 4 (1) (Feb 2011) 56–63. [11] F. Stazi, M. Mampieri, M. Cardinale, M.T. Laudadio, A. Gargaro, G.B. Del Giudice, Implant and long-term evaluation of atrial signal amplification in a single-lead ICD, Pacing Clin. Electrophysiol. 35 (9) (Sep 2012) 1119–1125. [12] G.Y. Lip, Stroke in atrial fibrillation: epidemiology and thromboprophylaxis, J. Thromb. Haemost. 9 (Suppl. 1) (Jul 2011) 344–351. [13] A.J. Camm, G.Y. Lip, R. De Caterina, et al., ESC Committee for Practice Guidelines (CPG). 2012 focused update of the ESC Guidelines for the management of atrial fibrillation: an update of the 2010 ESC Guidelines for the management of atrial fibrillation. Developed with the special contribution of the European Heart Rhythm Association, Eur. Heart J. 33 (21) (Nov 2012) 2719–2747. [14] T.B. Larsen, G.Y. Lip, F. Skjøth, K.M. Due, K. Overvad, R.L. Hvilsted, Added predictive ability of the CHA2DS2VASc risk score for stroke and death in patients with atrial fibrillation: the prospective Danish Diet, Cancer, and Health cohort study, Circ. Cardiovasc. Qual. Outcomes 5 (3) (May 2012) 335–342. [15] G. Boriani, G.L. Botto, L. Padeletti, et al., Italian AT-500 Registry Investigators. Improving stroke risk stratification using the CHADS2 and CHA2DS2-VASc risk scores in patients with paroxysmal atrial fibrillation by continuous arrhythmia burden monitoring, Stroke 42 (6) (Jun 2011) 1768–1770. [16] T.V. Glotzer, P.D. Ziegler, Silent atrial fibrillation as a stroke risk factor and anticoagulation indication, Can. J. Cardiol. 29 (Suppl. 7) (Jul 2013) S14–S23. [17] M. Brambatti, S.J. Connolly, M.R. Gold, et al., ASSERT Investigators. Temporal relationship between subclinical atrial fibrillation and embolic events, Circulation 129 (21) (May 27 2014) 2094–2099. [18] D.T. Martin, M.M. Bersohn, A.L. Waldo, et al., IMPACT Investigators. Randomized trial of atrial arrhythmia monitoring to guide anticoagulation in patients with implanted defibrillator and cardiac resynchronization devices, Eur. Heart J. 36 (26) (Jul 7 2015) 1660–1668.
Contents lists available at ScienceDirect
International Journal of Cardiology journal homepage: www.elsevier.com/locate/ijcard
Stroke incidence in patients with cardiac implantable electronic devices remotely controlled with automatic alerts of atrial fibrillation. A sub-analysis of the HomeGuide study Renato Pietro Ricci a,⁎,1, Diego Vaccari b,1, Loredana Morichelli a,1, Gabriele Zanotto c,1, Leonardo Calò d,1, Antonio D'Onofrio e,1, Antonio Curnis f,1, Ennio C.L. Pisanò g,1, René Nangah h,1, Marco Brieda i,1, Vittorio Calzolari j,1, Donato Melissano k,1, Nicola Rovai l,1, Alessio Gargaro l,1 a
San Filippo Neri Hospital, Rome, Italy Civil Hospital, Montebelluna, Italy c Mater Salutis Hospital, Legnago, Italy d Casilino Hospital, Rome, Italy e V. Monaldi Hospital, Naples, Italy f Spedali Civili, Brescia, Italy g Vito Fazzi Hospital, Lecce, Italy h Civil Hospital, Portogruaro, Italy i Santa Maria degli Angeli Hospital, Pordenone, Italy j Cà Foncello Hospital, Treviso, Italy k F. Ferrari Hospital, Casarano, Italy l Clinical Department, BIOTRONIK Italy, Vimodrone, Italy b
a r t i c l e
i n f o
Article history: Received 19 April 2016 Accepted 12 June 2016 Available online 14 June 2016 Keywords: Pacemakers Implantable cardioverter defibrillators Remote monitoring Telemedicine Atrial fibrillation Stroke
a b s t r a c t Background: Remote Monitoring (RM) of cardiac implantable electronic devices (CIEDs) is recommended in management of Atrial Fibrillation (AF), which is a recognized risk factor for thromboembolism. We tried to elucidate whether stroke incidence observed in a large, remotely monitored population was consistent with the CHA2DS2VASc risk profile. Methods: Data from 1650 patients [76% male, age 72 (63–68), CHA2DS2VASc score 3.0 (2.0–4.0)] enrolled during the HomeGuide study and monitored with a daily-transmission RM system providing automatic alerts for AF, were analysed. Of those, 25% had a pacemaker and 75% an implantable cardioverter defibrillator with or without cardiac resynchronization. Estimations of the expected thromboembolic events were based on the population CHA2DS2VASc score profile used in a computer-simulated Markov model. Results: Eight thromboembolic events were observed with a 4-year cumulative stroke rate of 0.8% (confidence interval, 0.4%–1.5%). Simulations returned from 18.7 to 17.1 expected events, depending on the AF duration assumed to trigger anticoagulation (one-sample log-rank p b 0.03). During the study period, 681 (84%) AF episodes and 129 (16%) atrial tachycardias were detected in 291 patients (18%): 93% of episodes were detected remotely in 269 patients, 66% of whom had no history of AF. Medical interventions were necessary in 305 episodes, 85% of which were detected remotely. Reaction time was 1 (0–6) days for remotely-detected episodes and 33 (14–121) days for episodes detected in clinic (p b 0.0001). Conclusions: In a large CIED population followed remotely for up to 4 years, the incidence of thromboembolic events was less than half the estimations based on the CHA2DS2VASc risk profile. © 2016 Elsevier Ireland Ltd. All rights reserved.
1. Introduction
⁎ Corresponding author at: Cardiology Department, San Filippo Neri Hospital, Via Martinotti, 20, Roma, Italy. E-mail address: [email protected] (R.P. Ricci). 1 All authors take responsibility for all aspects of the reliability and freedom from bias of the data presented and their discussed interpretation.
http://dx.doi.org/10.1016/j.ijcard.2016.06.016 0167-5273/© 2016 Elsevier Ireland Ltd. All rights reserved.
One of the most appealing implications of Remote Monitoring (RM) of Cardiac Implantable Electronic Devices (CIED) is the potential of a fine tuning of medical therapy in response to early detection and notification of atrial fibrillation (AF). This hypothesis is applicable in standard practice [1,2] and the latest consensus statement of the Heart Rhythm
252
R.P. Ricci et al. / International Journal of Cardiology 219 (2016) 251–256
Society has set a class I recommendation with level of evidence A for the use of RM to early detect and quantify AF [3]. However there is still no evidence that a prompt intervention in response to early AF detection may be effective in reducing the incidence of thromboembolic events. The HomeGuide study is a recently completed large registry collecting data from routine RM application in CIED follow-up [4,5]. The study included 1650 subjects in 75 sites testing a standard pre-defined workflow model to implement RM in outpatient clinics. We further analyzed the study data to assess whether the stroke incidence observed during the follow-up period was consistent with expectations based on the CHA2DS2VASc risk [6] profile of the selected population. In addition we analyzed (1) the incidence of symptomatic and asymptomatic AF remotely detected in patients with and without prior history of AF; (2) the time from onset/detection to medical decision; (3) the type of medical interventions. 2. Methods 2.1. Study design, organizational model, HM alert programming for AF Patients with standard indication to pacemaker, implantable defibrillator (ICD), or cardiac resynchronization therapy (CRT) implant, who received a RM system (specifically, Home Monitoring [HM], BIOTRONIK SE & Co. KG, Berlin, Germany) at post-implant discharge, were enrolled in the HomeGuide study (Clinical Trials identifier NCT01459874) [4]. HM is characterized by daily automatic transmissions from the implant to a portable patient unit forwarding data to a central server via GSM (Global System for Mobile communication). Data include both technical and diagnostic information normally retrieved in a full device interrogation in hospital. HM transmissions occur every 24 hours automatically. In addition, alerts are triggered by selectable critical events and notified with email, fax or messages. Programming options of automatic alerts depend on the implanted device type. AF alerts are delivered within 24 h to the attending site staff and are generally triggered by long sustained episodes and/or a cumulative AF daily burden higher than a programmable threshold: options range from N 0% to N 75% of 24 h. AF notifications are not available in devices without atrial sensing. In addition to RM, patients recruited in the HomeGuide study were visited in hospital annually (biyearly in case of a CRT device). Participating sites were trained to adopt an organizational model based on a cooperative interaction within a nurse-physician couple univocally assigned to individual patients [4,7]. Concerning alert programming for and medical reaction to AF episodes, investigators were advised to apply the following scheme: first episode of AF; AF burden N10% for at least five consecutive days, AF burden = 100% for two consecutive days (persistent AF) [2]. Any option of such scheme (or others, if deemed clinically necessary) could be selected according to medical evaluation based on AF history of the individual patient and on optimal antithrombotic and/or antiarrhythmic therapy. Medical reaction to AF alert was left to investigators' discretion.
atrial sensing were directly downloaded from the HM central server and used to double-check investigators' AF episode reporting. For each reported AT/AF episode, investigators were asked to annotate patientreferred symptoms and associated medical decision/action. All severe adverse events occurring during the HomeGuide study had to be reported and documented by the investigators. These include any ischemic or hemorrhagic cerebrovascular accident (stroke) or transient ischemic attack (TIA). 2.3. Methods for estimating the expected stroke incidence Due to the lack of a control study arm, direct observations on stroke incidence were not possible. However crude estimations were made possible through computer simulations [8] using baseline characteristics, duration of study participation, AF incidence and related symptoms of individual study subjects as input data. Calculations were based on the CHA2DS2VASc score profile of the enrolled population used in a two-state Markov chain model within a Monte Carlo simulation (Fig. 1). The basic assumption was that the real study population was not monitored remotely with a RM system. Therefore, as described below, the attending physician could become aware of an AF occurrence only during an in-hospital evaluation either scheduled or triggered by patient-referred symptoms. Further assumptions were selected following a “best-case” principle, in order to retrieve the least expected number of stroke events at the end of the simulations: • Individual patients without history of AF would remain in a “healthy” state until onset of AF. This was a conservative assumption, as it is equivalent to the hypothesis that the transition probability to a “stroke” state would be zero until AF onset. • Upon new AF onset during the study period, patient-referred AFrelated symptoms would determine timing of OAT introduction: immediately (i.e. after 3 days in the simulation), in case of symptomatic AF; at the next in-hospital visit, in case of asymptomatic AF. • In case of AF history at baseline, the transition probability was set according to the subject's CHA2DS2VASc score, reduced by 22% or 62% [9] if the subject was already taking aspirin or oral anticoagulation therapy (OAT), respectively.
As a last assumption, AT episodes were excluded from the simulation, thus assuming that AT episodes would be not associated to any risk increase. The actual study period of each patient was simulated with 500 iterations each, to obtain estimates of the least number of stroke events, expected in that population, based on CHA2DS2VASc risk profile. Simulations were repeated several times. At first, all investigator-reported AF episodes were included, regardless of episode duration. In a second set of simulations, patients with single-chamber devices (without atrial sensing and diagnostic functions) were excluded and only device-detected AF episodes were included. In this second set of simulations, we assumed that patients would start OAT only with daily AF burden of 1%, 25% and 100% of 24-h.
2.2. Endpoints of the present analysis 2.4. Statistical analysis The primary objective of the main study was to evaluate the ratio of all major cardiovascular events remotely detected with HM over the number of all events that actually occurred during follow-up, including those which are not thought to be detected by a CIED. The endpoints for the present analysis were Atrial Tachycardia (AT)/AF episodes and strokes. ATs were defined as arrhythmic atrial episodes with regular atrial cycle and rate b 200 bpm; AF as any episode with irregular cycle or rate N200 bpm. AT/AF endpoints included investigator-reported episodes detected during regular or alert-triggered HM sessions, during inhospital device interrogation, other visits. AT/AF-related data (including episode duration and 24-hour burden) from implanted devices with
Continuous and categorical variables were reported as median (interquartile range) and percentages, respectively. Subgroup comparisons were performed using the Mann–Whitney statistics (for continuous variables), chi-squared or Fisher's exact tests as appropriate (for binary or categorical variables). Stroke/TIA cumulative rate in the study population (along with 95% confidence interval) was estimated with the product-limit method and the relative Kaplan–Meier curve was generated. The one-sample log-rank test was used to compare the stroke hazard rate observed in the study population with the expected estimations based on the CHA2DS2VASc score distribution as output of the Monte
R.P. Ricci et al. / International Journal of Cardiology 219 (2016) 251–256
253
Fig. 1. Markov chain model used for Monte Carlo simulations. In the simulated model, “H” stands for “healthy” state and “S” for “stroke” state. P is the transition probability. AF = atrial fibrillation; OAT = oral anticoagulation therapy.
Carlo simulations. All comparisons were considered statistically significant at or below the p = 0.05 level. Stata 12 (StatCorp, Texas, US) software was used for all statistical analysis. Custom-made Visual Basic routines were edited for Monte Carlo simulations. 3. Results 3.1. Study population Between 2008 and 2012, 1650 patients participated in the HomeGuide study for a median follow-up of 18 (10–31) months. Table 1 summarizes the main characteristics of the whole population and the subgroups of subjects with and without AF history at enrollment. Pacemakers, ICDs and CRT devices were roughly distributed in a 1:2:1 ratio. Fifty five percent of ICDs were single-chamber devices, but 117 of them had atrial sensing capabilities [10,11]. OAT was administered in 28% of the general population, antiplatelet in 45%: 359 of the 751 patients in the latter group (48%) had prior myocardial infarction. Twenty one percent of patients had history of AF at enrollment. This group more frequently had an indication for pacemaker implant, had a higher CHA2DS2VASc score, and was more likely to be on antiplatelet, OAT and antiarrhythmic therapy. Indeed aspirin was still recommended in lowrisk patients at the time of the study conduct. Of note, 40% of patients with history of AF were not treated with OAT, and 7% did not follow any antithrombotic prophylaxis. There was no statistically significant difference in terms of structural heart disease (except for valvular diseases), NYHA Class, ejection fraction, ventricular arrhythmias between the subgroups of patients with and without AF history at enrollment. 3.2. Incidence of stroke events During the HomeGuide study, 8 cerebrovascular accidents were observed: 2 ischemic strokes, 2 TIAs, 3 fatal and 1 non-fatal hemorrhagic strokes. All events occurred in patients on antiplatelet or anticoagulation therapy at baseline. The product-limit estimate of the 4-year cumulative stroke/TIA rate was 0.8% (confidence interval, 0.4%–1.5%). Several Monte Carlo simulations were run in order to assess if the number of observed cerebrovascular accidents was consistent with what could be expected as least estimations based on the CHA2DS2VASc
Table 1 Patient characteristics.
Age (years) Male (n, %) Device type (n, %) Pacemaker Single-chamber Dual-chamber ICD Single-chambera Dual-chamber CRT-D or -P Heart disease (HD) None Cardiomyopathy Ischemic HD Valvular HD Channelopathies Congenital HD Others NYHA class I II III IV LVEF (%) Ventricular tachyarrhythmias AF Paroxysmal Persistent Permanent CHAD2DS2VASc score Baseline therapy Antiplatelets Oral anticoagulants Antiarrhythmics
All patients 1650
With history of AF 356 (21%)
Without history of AF 1294 (79%)
P
72 (63–68) 1261 (76%)
74 (68–78) 274 (77%)
71 (62–77) 987 (76%)
b0.0001 0.78
421 (25%) 4 417 803 (49%) 444 359 426 (26%)
106 (30%) 3 119 160 (45%) 111 49 90 (25%)
315 (24%) 1 298 643 (50%) 333 310 336 (26%)
0.04
0.79
183 (11%) 851 (53%) 689 (43%) 107 (7%) 27 (2%) 9 (1%) 5 (0%)
32 (9%) 173 (49%) 152 (43%) 44 (12%) 2 (1%) 1 (0%) 0 (0%)
151 (12%) 678 (52%) 537 (41%) 63 (5%) 25 (2%) 8 (1%) 5 (0%)
0.15 0.21 0.70 b0.001 0.07 0.44 –
309 (19%) 751 (46%) 553 (33%) 37 (2%) 30.0 (25.0–42.0) 447 (27%)
67 (19%) 150 (42%) 127 (36%) 12 (3%) 31.0 (26.0–45.0) 92 (26%)
242 (19%) 601 (46%) 426 (33%) 25 (2%) 30.0 (25.0–41.5)
0.95 0.14 0.33 0.10 0.39
355 (27%)
0.57
356 (22%)
356 (100%) 155 (44%) 71 (20%) 130 (37%) 4.0 (3.0–5.0)
0 (0%)
–
3.0 (2.0–4.0)
b0.001
118 (33%) 215 (60%) 113 (32%)
623 (48%) 252 (19%) 282 (22%)
b0.001 b0.001 b0.001
3.0 (2.0–4.0) 741 (45%) 467 (28%) 395 (24%)
0.11
AF = atrial fibrillation; CRT-D (-P) = cardiac resynchronization therapy with defibrillation (pacing) function; ICD = implantable defibrillator; HD = heart disease; LVEF = left ventricular ejection fraction. a Including 117 single-chamber ICDs with atrial sensing capability.
254
R.P. Ricci et al. / International Journal of Cardiology 219 (2016) 251–256
score of the study subjects. The results are shown in Fig. 2 and further detailed in Table 2. Simulations returned a significantly higher number of expected events, as compared to the real occurrences: it ranged from 18.7 events (confidence interval, 18.4–19.1) when we included all patients and all AF episodes (regardless of their duration in terms of daily AF burden), to 17.1 (16.7–17.4), when we excluded pure single-chamber devices and limited OAT introduction only upon atrial episodes causing a 24-hour AF burden = 100%. Interestingly, simulation outputs were quite similar independently of the AF burden thresholds chosen to introduce OAT: a slight difference of only 0.4 expected stroke/TIA events was obtained, on average, between 10% and 100% AF burden. Comparisons with the real occurrences during the followup period were statistically significant at the one-sample log-rang test both including all bleeding and ischemic strokes (8 events) or only ischemic strokes or TIAs (4 events). 3.3. Incidence of symptomatic and asymptomatic AF During the study period, 810 atrial episodes were detected in 291 patients (18%): 681 (84%) were classified as AF, 129 (16%) as AT (Table 3). The majority of AT/AF episodes (93%) were detected remotely during HM sessions. Of note, 193 of the 291 patients with AT/AF episodes (66%) did not have history of AF at enrollment. Patients did not report any AF-related symptoms in 72% of episodes. Asymptomatic episodes were mostly detected remotely by HM (96%) and frequently recurred in patients without history of AF (142/213 episodes, 67%). Medical reactions were required in 305 of 810 episodes, 85% of which were detected remotely. For episodes detected remotely, medical decision was taken 1 (0–6) days after episode onset, as compared to episodes detected in-hospital: 33 (14–121) days and 40 (17–133) including only N24-hour AF episodes (p b 0.0001, for both comparisons). Table 4 shows the interventions in detail. Change in pharmacological
Fig. 2. Kaplan–Meier curve of stroke/TIA free rate. Dotted line shows the expected level of the Kaplan–Meier survival curve. Calculations were performed with the Monte Carlo method based on the CHA2DS2VASc risk profile of the study population. TIA = Transient ischemic attack.
therapy was the most frequent intervention: in 111 cases (61%) the ongoing therapy was optimized; in particular OAT, betablockers, and amiodarone were introduced in 39, 43 and 34 cases (21%, 24%, and 19%), respectively. A vast majority of interventions on medical therapy (159, 87%) were taken in response to episodes detected remotely with HM. 4. Discussion 4.1. Main results The main results of our analysis may be summarized as follows: 1) in the HomeGuide study population systematically followed up with a RM system characterized by daily automatic transmissions the incidence of thromboembolic events was less than half of estimation based on the CHA2DS2VASc risk profile of the population; 2) 94% of AT/AF episodes were detected remotely, 72% were asymptomatic and mostly occurred in patients with no history of AF (66%); 3) 85% of medical interventions for AF were triggered by RM notifications, which ultimately drove therapy and device optimization, cardioversions, ablations or hospitalizations. 4.2. Clinical implications Our results do not prove the efficacy of RM as a tool to prevent thromboembolic events, as computer simulated models cannot surrogate results from randomized prospective trials, no matter how realistic and accurate the assumptions are. However simulations can provide indications justifying further evaluations. As a matter of fact, there were significantly less ischemic or bleeding stroke events during the study period than it could be expected based on the CHA2DS2VASc score risk of individual patients. The CHA2DS2VASc score is a well-known validated and recommended risk stratification index [12,13], as it is non-inferior to, and possibly better than, previous scores such as CHADS2 in identifying patients who develop stroke and thromboembolism, and it is superior in identifying ‘truly low-risk’ patients with AF [14,15]. Estimations of the expected number of thromboembolic events in the study population were performed using a simple 2-state Markov chain model including the individual patients' CHA2DS2VASc scores for the transition probabilities, possibly reduced by the administration of antithrombotic therapy. The least number of assumptions was introduced in the simulated model in order to keep it sufficiently accurate but not critically dependent on arbitrary constraints. The basic assumption was that the real population was not monitored remotely and physician's evaluation of an AF episode could only take place if patients reported symptoms or during the next actually performed follow-up visit. It should be emphasized that all other necessary assumptions, despite generating crude approximations, were always based on a ‘best-case’ scenario, selecting conditions associated with the lowest, even if unrealistic, event probabilities. For example, by setting the Markov transition probability to 0 in the absence of AF episodes, we implicitly assumed that thromboembolic strokes may only be caused by AF. This assumption may be unrealistic but it tended to underestimate the expected event rate in the scenario of no RM. Similarly, it was assumed that patient-reported symptoms would always trigger an immediate introduction of OAT within 3 days, and that AT episodes were not associated with an increased risk of stroke. These assumptions aimed at retrieving a minimal rather than realistic estimation of stroke incidence. Nevertheless, the expected stroke number returned by the Monte Carlo simulations was always higher than that actually observed, regardless of the AF burden threshold assumed necessary to trigger OAT. AF burden threshold that can justify the introduction of OAT is still a matter of debate, as it has not been precisely identified in the range of few minutes up to several hours [16]. Monte Carlo simulations were therefore repeated with different assumptions on AF burden thresholds
R.P. Ricci et al. / International Journal of Cardiology 219 (2016) 251–256
255
Table 2 Expected number of embolic events. All observed ischemic and bleeding strokes (8)
Observed ischemic strokes/TIA(4)
Assumed AF episode duration triggering OAT
Expected
Tolerance
Chi squared
P
Chi squared
P
All investigator-reported AF Daily AF burden ≥1% Daily AF burden ≥25% Daily AF burden ≥100%
18.72 17.70 17.49 17.07
0.35 0.35 0.33 0.34
6.14 5.32 5.15 4.82
0.013 0.021 0.023 0.028
3.40 10.60 10.40 10.01
0.0007 0.0011 0.0013 0.0016
Results of the Monte Carlo simulations based on the Markov chain model are shown in Fig. 1. The expected number of strokes in the study population and the respective tolerance (semilength of the confidence interval) are reported and compared with the one-sample logrank test to the number of all ischemic and bleeding events (8) and ischemic stroke/TIAs (4) actually observed. Simulations were repeated for different assumptions on the AF incidence triggering antithrombotic therapy in the simulated model: first all investigator-reported AF was included; then only AF episodes generating a daily arrhythmic burden ≥1%, ≥25%, ≥100%. AF = atrial fibrillation; TIA = Transient ischemic attack.
Table 3 Classification of atrial tachyarrhythmia episodes. AT/AF episodes
All episodesa
Remotely detected during HM sessionsb
Detected during in-hospital visits/otherb
No. of episodes AF AT No. patientsc Without history of AF Asymptomatic episodes No. of patients Without history of AF Actionable episodes
810 681 (84%) 129 (16%) 291 193 583 (72%) 213 142 305 (38%)
753 (93%) 640 (94%) 113 (88%) 269 179 (66%) 559 (96%) 200 136 258 (85%)
57 (7%) 41 (6%) 16 (12%) 50 34 (68%) 24 (4%) 22 14 47 (15%)
AF = atrial fibrillation; AT = atrial tachycardia; HM = Home Monitoring. a Percentages were calculated with respect to the first row. b Percentages were calculated with respect to the first column. c In 28 patients different AT/AF episodes were detected both remotely and in-hospital.
to be exceeded in one single day in order to let OAT be introduced. After a first simulation including all the AF episodes reported by the investigators regardless of their durations, three more simulations were performed based on device-reported daily AF burden thresholds of 1%, 25%, and 100%, corresponding to AF durations of 14 min, 6 h, and 24 h, respectively. It is worth noting that simulations returned very similar
estimations regardless of AF burden thresholds: from the 18.7 expected events when all the patients and all the investigator-reported AF episodes were included, to 17.1 expected events when only devices with atrial sensing were included and only episodes with 100% AF burden were assumed necessary to trigger OAT. The small difference of less than 2 expected stroke events obtained in a 1650 population in an 18-
Table 4 Medical interventions to AT/AF episodes. Action
All episodesa
Episodes remotely detected during HM sessionsb
Episodes detected during in-hospital visits/otherb
Therapy optimization Suspension/change dosage Introduction of Betablockers Antithrombotic therapy Amiodarone Diuretics Class IC antiarrhythmics Sotalol Digitals Ca-antagonists ACE-inhibitors Ag II receptor blockers Other drugs Encouraging patient to improve adherence to therapy Device reprogramming Cardioversion DC shock Pharmacological RF ablation Hospitalization Others
182 111 (61%)
159 (87%) 106 (95%)
47 (13%) 5 (5%)
43 (24%) 39 (21%) 34 (17%) 13 (7%) 6 (3%) 5 (3%) 5 (3%) 4 (2%) 3 (2%) 3 (2%) 40 (22%) 20 (11%) 65 (36%)
40 (93%) 34 (87%) 32 (94%) 11 (85%) 6 (4%) 5 (3%) 5 (3%) 4 (2%) 3 (2%) 3 (2%) 38 (24%) 19 (95%) 54 (83%)
3 (7%) 5 (13%) 2 (6%) 2 (15%) 0 (0%) 0 (0%) 0 (0%) 0 (0%) 0 (0%) 0 (0%) 2 (3%) 1 (5%) 11 (17%)
33 (18%) 6 (3%) 11 (6%) 24 (13%) 14 (8%)
25 (76%) 4 (66%) 8 (73%) 15 (62%) 13 (93%)
8 (24%) 2 (33%) 3 (27%) 9 (38%) 1 (7%)
More medical interventions were required in some cases. AF = atrial fibrillation; AT = atrial tachycardia; DC = direct current; HM = Home Monitoring; RF = radiofrequency. a Percentages were calculated with respect to the first row. b Percentages were calculated with respect to the first column.
256
R.P. Ricci et al. / International Journal of Cardiology 219 (2016) 251–256
month median follow-up is clearly a consequence of the very asymmetrical and skewed distribution of AF and probably explains the current uncertainty in identifying the proper AF burden threshold. The relation between AF and thromboembolism is more complex than initially suspected. It is well known that AF is not the only mechanism potentially involved in pathogenesis of thromboembolism. In addition, two recent sub-analyses from the ASSERT [17] and the IMPACT [18] studies have demonstrated a substantial temporal disconnection between AF onset and stroke events, in contrast with the idea of a simple cause–effect model. Such findings clearly reflect the current incomplete understanding of the clinical relevance of (non-)sustained AF and mechanisms underlying thromboembolism. Regardless of whether AF-stroke relation is direct or mediated by other mechanisms, the association between AF and increased stroke risk is well established. Therefore, the theoretical basis of early AF detection with RM to trigger prompt medical reaction (e.g. anticoagulation therapy) is worth investigating further. The recently published IMPACT study was a first attempt toward this objective but it failed to demonstrate any benefit from an ‘on/off’ OAT strategy driven by continuous remote monitoring of AF episodes as compared with a conventional approach [18]. However, the study suffered from several limitations both in design and compliance, preventing straightforward conclusions. Patients randomized to the active study arm discontinued OAT after 30 days (CHADS2 1 or 2) or 90 days (CHADS2 3 or 4) without AF recurrences. More than 90% of patients with stroke at the time of study end either were not under OAT or had suboptimal INR values. In our series we observed a lower than expected rate of thromboembolic events. We have no data to support the hypothesis that this was the result of an optimized therapy management driven by RM findings without any pre-specified antithrombotic therapy plan. Nevertheless, such a low stroke incidence was observed in an unselected population with remotely controlled CIEDs, in which 94% of AF were detected and notified by the RM system, and 85% of interventions for AF (including 21% of OAT introduction) were triggered by a RM notification of AF with a median reaction time of 1 day. These figures might reasonably play a role in the explanation of the low stroke incidence reported in our population, even if we cannot exactly estimate to what extent. 5. Conclusions We observed a lower than expected incidence of thromboembolic events in an unselected CIED population remotely controlled with HM and in which most AF episodes were detected and managed upon early RM notification. Monte Carlo computer simulations based on the population CHA2DS2VASc risk profile showed that the stroke incidence observed in the population was less than half of the expected rate, irrespective of the AF burden threshold assumed to trigger OAT. Conflicts of interest RPR: minor consultancy fees by Medtronic NR and AG: employees of Biotronik Italy DG, LM, GZ, LC, ADO, AC, ECLP, RN, MB, VC and DM: nothing to disclose
Acknowledgments The authors would like thank Xenia Antoniou for her precious advice in text revision. References [1] F. Zoppo, D. Facchin, G. Molon, et al., Improving atrial fibrillation detection in patients with implantable cardiac devices by means of a remote monitoring and management application, Pacing Clin. Electrophysiol. 37 (12) (Dec 2014) 1610–1618. [2] R.P. Ricci, L. Morichelli, M. Santini, Remote control of implanted devices through Home Monitoring technology improves detection and clinical management of atrial fibrillation, Europace 11 (1) (Jan 2009) 54–61. [3] D. Slotwiner, N. Varma, J.G. Akar, et al., HRS Expert Consensus Statement on remote interrogation and monitoring for cardiovascular implantable electronic devices, Heart Rhythm. 12 (7) (Jul 2015) e69–100. [4] R.P. Ricci, L. Morichelli, A. D'Onofrio, et al., Effectiveness of remote monitoring of CIEDs in detection and treatment of clinical and device-related cardiovascular events in daily practice: the HomeGuide Registry, Europace 15 (7) (Jul 2013) 970–977. [5] R.P. Ricci, L. Morichelli, A. D'Onofrio, et al., Manpower and outpatient clinic workload for remote monitoring of patients with cardiac implantable electronic devices: data from the HomeGuide Registry, J. Cardiovasc. Electrophysiol. 25 (11) (Nov 2014) 1216–1223. [6] G.Y. Lip, R. Nieuwlaat, R. Pisters, D.A. Lane, H.J. Crijns, Refining clinical risk stratification for predicting stroke and thromboembolism in atrial fibrillation using a novel risk factor-based approach: the euro heart survey on atrial fibrillation, Chest 137 (2) (Feb 2010) 263–272. [7] R.P. Ricci, L. Morichelli, M. Santini, Home monitoring remote control of pacemaker and implantable cardioverter defibrillator patients in clinical practice: impact on medical management and health-care resource utilization, Europace 10 (2) (Feb 2008) 164–170. [8] R.P. Ricci, L. Morichelli, A. Gargaro, M.T. Laudadio, M. Santini, Home monitoring in patients with implantable cardiac devices: is there a potential reduction of stroke risk? Results from a computer model tested through Monte Carlo simulations, J. Cardiovasc. Electrophysiol. 20 (11) (Nov 2009) 1244–1251. [9] G.Y. Lip, R.G. Hart, D.S. Conway, Antithrombotic therapy for atrial fibrillation, BMJ 325 (7371) (Nov 2 2002) 1022–1025. [10] C. Sticherling, M. Zabel, S. Spencker, et al., Comparison of a novel, single-lead atrial sensing system with a dual-chamber implantable cardioverter-defibrillator system in patients without antibradycardia pacing indications: results of a randomized study, Circ. Arrhythm. Electrophysiol. 4 (1) (Feb 2011) 56–63. [11] F. Stazi, M. Mampieri, M. Cardinale, M.T. Laudadio, A. Gargaro, G.B. Del Giudice, Implant and long-term evaluation of atrial signal amplification in a single-lead ICD, Pacing Clin. Electrophysiol. 35 (9) (Sep 2012) 1119–1125. [12] G.Y. Lip, Stroke in atrial fibrillation: epidemiology and thromboprophylaxis, J. Thromb. Haemost. 9 (Suppl. 1) (Jul 2011) 344–351. [13] A.J. Camm, G.Y. Lip, R. De Caterina, et al., ESC Committee for Practice Guidelines (CPG). 2012 focused update of the ESC Guidelines for the management of atrial fibrillation: an update of the 2010 ESC Guidelines for the management of atrial fibrillation. Developed with the special contribution of the European Heart Rhythm Association, Eur. Heart J. 33 (21) (Nov 2012) 2719–2747. [14] T.B. Larsen, G.Y. Lip, F. Skjøth, K.M. Due, K. Overvad, R.L. Hvilsted, Added predictive ability of the CHA2DS2VASc risk score for stroke and death in patients with atrial fibrillation: the prospective Danish Diet, Cancer, and Health cohort study, Circ. Cardiovasc. Qual. Outcomes 5 (3) (May 2012) 335–342. [15] G. Boriani, G.L. Botto, L. Padeletti, et al., Italian AT-500 Registry Investigators. Improving stroke risk stratification using the CHADS2 and CHA2DS2-VASc risk scores in patients with paroxysmal atrial fibrillation by continuous arrhythmia burden monitoring, Stroke 42 (6) (Jun 2011) 1768–1770. [16] T.V. Glotzer, P.D. Ziegler, Silent atrial fibrillation as a stroke risk factor and anticoagulation indication, Can. J. Cardiol. 29 (Suppl. 7) (Jul 2013) S14–S23. [17] M. Brambatti, S.J. Connolly, M.R. Gold, et al., ASSERT Investigators. Temporal relationship between subclinical atrial fibrillation and embolic events, Circulation 129 (21) (May 27 2014) 2094–2099. [18] D.T. Martin, M.M. Bersohn, A.L. Waldo, et al., IMPACT Investigators. Randomized trial of atrial arrhythmia monitoring to guide anticoagulation in patients with implanted defibrillator and cardiac resynchronization devices, Eur. Heart J. 36 (26) (Jul 7 2015) 1660–1668.