Arrhythmia and Sudden Death in Hemodialysis Patients: Protocol and Baseline Characteristics of the Monitoring in Dialysis Study

Role of Dialysis Baths

Low-potassium dialysate of 0 or 1 mEq/L (17) or <2 mEq/L (18) and lower calcium dialysate (<2.5 mEq/L) (19) have been implicated as risk factors for SCD in large database studies. Similarly, low serum magnesium levels were associated with higher all-cause mortality in Japanese dialysis patients (20), but few data exist on SCD risk and magnesium dialysate levels.

Dialytic Cycle

Although great enthusiasm exists among nephrology professionals for more frequent maintenance HD, most patients dialyze three times weekly, necessitating two gaps of 1 day and one gap of 2 days. This has long been viewed as physiologically challenging in patients with limited capacity to maintain homeostasis in the presence of metabolic and volume-related excursions from normality, where background cardiovascular disease is the norm.

In a nationally representative United States sample between 2004 and 2007, one study compared rates of death and cardiovascular admissions on the day after the 2-day interdialytic interval with rates on other days (21). As shown in Figure 4, although overall mortality, cardiovascular mortality, and cardiovascular admission rates were higher on the day after the long interval, relative event disparities were especially marked for congestive heart failure (by a factor of 1.8) and dysrhythmia (by a factor of 1.9).

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

Association of dialytic intervals cycle with mortality and cardiovascular hospitalizations. Annualized mortality (A) and cardiovascular admission rates (B) per 100 patient-years, on different days of the dialysis week. Error bars represent 95% confidence intervals. CHF, congestive heart failure; CVD, cardiovascular disease; HD, hemodialysis (session); MI, myocardial infarction. Adapted from reference 21.

Similar associations have been seen in several large studies (22–24). For example, in a recent study of newly incident HD patients in the United Kingdom between 2002 and 2006, hospital admission rates after the 2-day gap were 1.7-fold higher, whereas all-cause mortality rates were 1.22 times higher. As with almost all of the literature to date, this study was not specifically configured to compare rates of SCD. Nevertheless, out-of-hospital death rates were 1.59 times higher after the long interval than after the shorter interval and only 1.06 times higher for in-hospital death (23).

Incidence and Prevalence of AF

Although asystole or ventricular arrhythmias are the most likely types of arrhythmia to result in sudden death, atrial arrhythmias, particularly AF, may result in significant morbidity in patients with ESRD. AF is increasingly common in HD patients. In a study of 258,605 older participants, the AF incidence in incident dialysis patients was 14.8/100 person-years, and adjusted probabilities of developing AF during the first year of dialysis increased from 11.3% in 1995 to 14.3% in 2007 (44). Similarly, the prevalence of AF (diagnosed from administrative claims) was 10.7% in 2006—a three-fold increase from 1992 (45). Finally, the overall AF prevalence in a meta-analysis of 25 dialysis studies was 11.6% (46).

Dialysis as a Trigger

Most evidence that HD triggers AF is indirect. Echocardiographic intra-atrial and interatrial activation times, for example, correlate with ultrafiltration volume and shorten significantly toward the end of HD (47). Similarly, there are peridialytic changes in P-wave dispersion (standard electrocardiography), P-wave duration, and the root mean square voltage of the final 20 ms of the filtered P wave (signal-averaged electrocardiography)—parameters that measure atrial conduction delay and are associated with AF (48,49). In several studies, intradialytic changes were prominent and were attenuated at the conclusion of dialysis (50,51). In others, changes were not detected or were seen only at the conclusion of dialysis and then quickly attenuated (52). However, in one study, P-wave duration decreased and root mean square voltage of the final 20 ms of the filtered P wave decreased from the beginning to the end of dialysis, changes suggesting a reduced risk of AF after HD (53). Changes in electrolytes were associated with P-wave parameters in several studies (51,52), but associations with ultrafiltration appear more robust (50,51). On balance, this literature suggests that rapid ultrafiltration or other factors during HD acutely affect conduction in a way that could trigger AF but that effects rapidly attenuate after HD.

More recently, implanted defibrillators were used to monitor patients enrolled in the Implantable Cardioverter Defibrillator-2 Trial. AF was detected in 14 of 40 dialysis patients—11 receiving HD, 3 receiving peritoneal dialysis, and 9 without known AF (54). AF frequency was three-fold higher on dialysis days (P=0.001) but did not differ with duration of the interdialytic interval. Intradialytic AF was 13-fold more frequent than in the 7 hours before dialysis (P=0.03) and two-fold higher than in the 7 hours after dialysis (P=0.001), although results were largely driven by two patients. Lower dialysate potassium concentration and higher ultrafiltration volumes were associated with AF occurrence. More recently, a trial using an ILR in 50 HD patients found that the long interdialytic interval was the most frequent period of arrhythmia. AF was detected in 42% of patients, and new-onset AF was asymptomatic in 86% (16).

Consequences of AF

Associations with stroke are similar in the ESRD and general populations (44–46,55,56). In a recent study, for example, stroke incidence was 5.2/100 patient-years in patients with AF compared with 1.9/100 patient-years in dialysis patients without AF (46). AF is also strongly associated with mortality due to fatal stroke and other causes, the rate of which is roughly doubled (26.9 versus 13.4/100 patient-years) when AF is present (44–46). In addition, AF is an independent risk factor for mesenteric ischemia, a frequently fatal event (57), and amputation (58); although this has not been studied in the setting of ESRD, AF may lead to tachyarrhythmia-induced cardiomyopathy when sustained (59). Frequent AF could thus be an important cause of an extended array of morbidity and mortality, including fatal stroke, heart failure, and deterioration in LV ejection fraction in HD patients.

Relevance of Subclinical AF

Although associations of subclinical AF with clinical outcomes have not been studied in ESRD, multiple studies suggest that clinically silent AF is an important stroke risk factor in the general population. In one randomized pacemaker trial, the pacemaker monitoring function was used to record rapid atrial events (rate ≥220 beats/min [BPM] lasting ≥5 minutes) presumed to represent AF (60). Overall, 51.3% of patients had events at a median onset of 100 days. Atrial events were independently associated with stroke (hazard ratio, 2.8; P=0.001). Other studies have demonstrated similar associations between subclinical AF and subsequent overt AF or stroke (61–63). More recently, the Cryptogenic Stroke and Underlying AF (CRYSTAL AF) (64) study randomly assigned 441 stroke patients without AF who were undergoing 24-hour monitoring to an ILR or conventional follow-up. AF lasting >30 seconds was detected in 30% of intervention versus 3% of control patients at 3 years.

These data suggest that subclinical AF is an important contributor to stroke while illustrating the limits of clinical observation and standard diagnostics for advancing scientific discovery. In CRYSTAL AF, newly developed implantable technologies enabled long-term, continuous monitoring, thereby facilitating discovery of a high frequency of “silent” AF and a paradigm shift in the understanding of cryptogenic stroke. Improving understanding of SCD and its association with dialysis is an obvious extension. More broadly, CRYSTAL AF highlights a new paradigm in which the expanding capabilities and decreasing size of wearable or implantable technologies will facilitate capture of previously unmeasurable parameters and collection of more granular physiologic data, thereby improving understanding of dialysis physiology.

Overview

The primary objective of the MiD study (NCT01779856) was to estimate the proportion of HD patients experiencing clinically significant arrhythmias during a 6-month primary observation period using the Reveal ILR system. Secondary goals were to broadly characterize the occurrence of arrhythmia in HD-dependent ESRD and quantify associations with electrolytes, HD procedure, and volume parameters.

Study Population

Eligibility criteria were designed to maximize generalizability and patient safety (Table 1). All research complied with the Declaration of Helsinki and was approved by applicable institutional review boards.

Study Procedures

Echocardiography (if not performed within the preceding 6 months), 12-lead electrocardiography, medical history, and medications were obtained at baseline, and ILRs were implanted within 14 days. For the first 6 months after implantation, patients transmitted Reveal data immediately before all dialysis sessions and after each session associated with a study blood draw. Postmortem transmission was encouraged but was not specifically required by the protocol. Symptoms, dialysis prescription, ultrafiltration volumes, and peri- and postdialysis vital signs were collected at every dialysis session during the first 6 months. Blood was drawn both before and after dialysis: twice weekly for 4 weeks and once weekly thereafter through 6 months (Table 2 and 3). After 6 months, transmissions occurred at least weekly but data collection was otherwise limited to adverse events.

Study coordinators reviewed transmissions after the dialysis session to identify prespecified, potentially dangerous arrhythmias, which mandated investigator review: VT ≥180 BPM for >15 seconds, asystole >5 seconds, waking (6 a.m.–10 p.m.) heart rate ≤40 BPM for ≥6 seconds, AF for >24 hours or for >12 hours over consecutive days, or symptomatic arrhythmia.

During months 0–6, the study sponsor reviewed patient-marked events and transmissions with potential arrhythmias. A core laboratory adjudicated those possibly consistent with the primary endpoint. Long-term follow-up continued for a maximum of 12 months. The study concluded when the last study participant had completed 6 months of follow-up. ILR removal at study termination was optional.

Endpoints

The primary study objective was to estimate the proportion of HD patients experiencing clinically significant cardiac arrhythmias (CSAs) over 6 months. CSAs are arrhythmias considered most likely to be associated with syncope and cardiac arrest or to cause symptoms of hypoperfusion. They were based on standard definitions (69–71), recommendations of an advisory panel, and the detection capabilities of the Reveal device and included the following:

• VT≥115 BPM lasting ≥30 seconds (the rate limit was subsequently changed to ≥130 BPM with a protocol amendment).

• Bradycardia with heart rate ≤40 BPM for ≥6 seconds.

• Asystole for ≥3 seconds.

• Patient-marked (symptomatic) events where electrocardiographic review showed an arrhythmia considered clinically relevant in the judgment of the site cardiologist.

Secondary objectives were designed to assess device safety, characterize cardiac rhythm and associations with dialysis or clinical events, and assess the ability of the Reveal device to detect short-term changes in electrocardiographic morphology and their association with treatment parameters.

The following secondary objectives were specified: (1) quantifying device and procedure-related adverse events; (2) recording health-related events, specifically death, cardiovascular events, and health care utilization; (3) analyzing the association of arrhythmic events with health-related events and HD treatment parameters (this objective included CSA, other arrhythmias [e.g., AF], and parameters such as heart rate variability and heart rate trend); (4) quantifying atrial arrhythmia burden and analyzing associations with HD treatment parameters; and (5) assessing association of captured electrocardiographic morphology and pre- and postdialysis serum electrolyte levels.

Statistical Analyses

The planned sample size was 125 implanted patients. With 10% attrition this would have enabled estimation of the proportion of patients experiencing CSAs with a 95% confidence interval (95% CI) half-width of less than 0.1 for any proportion. Preliminary data from the first 50 enrolled patients revealed CSA rates of ≥70%, which allowed estimation of 95% CI half-widths ≤0.15 without additional enrollment. Thus, recruitment of the full sample size was not necessary to reliably estimate CSA event rates or to determine whether there was a clinically relevant incidence of CSA in the HD population. Given the financial and logistic challenges of recruitment and limited potential for qualitative effect from additional enrollment, the study was capped at 66 implanted patients. This final sample size of 66 enables the estimation of the proportion of patients experiencing CSA with a 95% CI half-width of <0.13 for any proportion.

The primary CSA proportion will be calculated from the cohort completing 6-month follow-up with 95% CI bounds estimated by the Clopper–Pearson “exact” method using the MiD-p modification (72). The analysis will be repeated in two populations: (1) patients with complete follow-up, plus those with incomplete data who experienced CSAs, and (2) all implanted patients, assuming that those with incomplete follow-up and no CSAs while under observation had no events during unobserved follow-up. This will be supplemented with Kaplan–Meier survivor plots of CSA-free survival. A negative binomial model, appropriate for recurrent events and allowing for variable follow-up (73), will be used to estimate mean and 95% CI of CSA per patient-year.

Special Considerations

Use of an invasive monitoring device in a clinical study is an unusual feature of MiD and raises unique issues by exposing patients to risk in an otherwise observational study. The ethical issues are not unique to this design. Radiographic endpoints are frequently used in studies, for example, and also require radiation exposure without clear benefit; phase 1 drug studies or “first-in-human” device studies also expose people to risk without clear benefit.

Given the need to understand SCD and arrhythmia cause in the dialysis population, the minimally invasive nature of ILR devices, and the low rate of infections when used in other populations, the use of Reveal XT and LINQ in MiD was felt to be ethically acceptable. In general, we advocate assessing “interventional-observational” research designs, with an ethical framework that balances the importance of the knowledge gained against potential risks. High-risk, non–minimally invasive devices would rarely be acceptable without a potential for individual benefit. Conversely, as devices become smaller and less invasive, they may enable observational studies of nonserious conditions. Use of noninvasive tools should generally be considered as an alternative, and investigators are obligated to carefully review individual risk before enrolling patients in an observational study requiring use of an implantable monitor.

A particularly thorny issue is that monitoring data can alter observed outcomes and bias estimated event rates. Detected arrhythmias in MiD, for example, could mandate changes in clinical care (e.g., pacemaker implantation, change in dialysis prescription) that modify the risk of subsequent arrhythmia, thereby reducing CSAs and resulting in underestimation of the true population CSA rate.

Full disclosure to participants and clinicians maximizes potential benefits to participants but also maximizes bias. Sequestering the monitoring data eliminates outcome contamination, but nondisclosure of potentially harmful and treatable conditions discovered during research is ethically unacceptable.

With end-of-study batch analysis of raw monitoring data, results and interpretation of the monitoring data do not become available until after the study has concluded. This is similar to storing blood samples for batch analysis at the end of a study. Because the samples (in this case monitoring information) are not analyzed or interpreted until after the study, this approach eliminates the ethical problems inherent to withholding analyzed data; however, it imposes substantial data storage requirements and may inhibit the ability to optimally understand clinical features of events or to determine the best clinical response. Batch analysis at regular intervals offers a useful intermediate approach. Finally, it may be possible to select endpoints, monitoring time frames, or data disclosures in order to limit biasing of important endpoints while allowing clinical use of research data. In MiD, arrhythmias felt to be dangerous, such as asystole and sustained VT, were reviewed in an ongoing manner and shared with study patients and their clinicians. Conversely, disclosure of arrhythmias of uncertain importance, such as nonsustained VT, was not required.

Baseline Characteristics of the Study Population

Patients were enrolled in India (23 of 66) and the United States (43 of 66). Several characteristics of the study population differed from those of the United States dialysis population (1). Mean age was lower; only 12.1% of patients were ≥70 years of age. A higher percentage of patients were black (53.0%), Asian (34.8%), and male (69.7%). Diabetes was present in 63.6% of patients, while nearly half of the patients (48.5%) had a history of ischemic heart disease. A minority of patients had a history of arrhythmia (31.8%), and 10.6% had a history of AF (Table 4). Use of cardiovascular medications was common, with 55%, 33%, and 48% of patients using β-blockers, angiotensin-converting enzyme inhibitors, or angiotensin-receptor blockers or statins, respectively. Laboratory parameters (Table 5) suggested adequate delivery of dialysis with a mean single pool Kt/V of 1.5±0.4. Potassium (5.0±1.0 mEq/L), hemoglobin (10.6 ±1.2 g/dl), and phosphorous (5.5±2.0 mg/dl) were each mildly elevated but within the expected range for patients undergoing long-term dialysis.

Several characteristics differed significantly between United States and Indian patients. In particular, Indian patients weighed less (66.5 versus 97.6 kg), were more likely to have diabetic kidney disease (56.5% versus 34.9%), were more likely to dialyze via an arteriovenous fistula (91.3% versus 59.5%), had a shorter dialysis vintage (2.2 versus 3.5 years), and were less likely to have hypertension (60.9% versus 97.7%) than participants in the United States. Laboratory characteristics were generally similar, but bicarbonate (18.8 versus 23.8 mEq/L) and sodium (134.1 versus 138.4 mEq/L) concentrations were lower in India.

This regional variation and increased variability in laboratory and clinical characteristics present challenges and opportunities. Analysis of country-specific arrhythmia rates is important, but the wide range of characteristics, such as body mass index, hypertension, and bicarbonate concentrations, may provide a greater ability to analyze associations of these factors with arrhythmia than would have been possible had the study recruited only within the United States. Global enrollment should thus limit the effect of country-specific dialysis practices on our findings and may enhance relevance to the global HD population while requiring more cautious extrapolation to the United States population.

MiD as a Paradigm

The MiD study illustrates the potential of invasive monitoring to improve understanding of the pathophysiology of ESRD and elucidate the true nature or cause of clinical events in dialysis patients. From our experience, it seems clear that when investigators are engaged, risks are reasonably low, research questions are important, and clinical benefit is possible, dialysis patients can be engaged and their collaboration secured in observational studies using invasive monitoring procedures.

Future research on ILR technology in ESRD should include studies to determine the association between subclinical AF and the risk of stroke, long-term studies with sufficient sample size to capture terminal cardiac rhythms at the time of SCD, and interventional studies testing the effect of individualization of the dialysis prescription on arrhythmia. Devices with miniaturized impedance technology will facilitate studying associations of volume status with arrhythmia and more sophisticated investigations into links between ultrafiltration rate, BP, volume status, cardiovascular death, and intradialytic hypotension. Outside of the cardiovascular arena, implantable glucose monitors stand out as a technology that could be readily used to definitively determine the true relationship between serum glucose levels and hemoglobin A1c or clinical outcomes in ESRD (74). Similarly, one can envision using pressure or flow monitoring devices to better understand access physiology and trajectories of access maturation and failure.