Introduction

Implantable cardioverter-defibrillator (ICD) therapy improves survival of patients at high risk of sudden cardiac death.¹ Current guidelines recommend that patients with ICDs should be evaluated every 3 to 6 months to assess device function.² This can impose a considerable burden on patients, physicians, and health systems if patients are evaluated in the office. Remote patient monitoring (RPM) has been promoted as a strategy to improve efficiency and reduce the burden associated with device follow-up by replacing at least some in-office device follow-up visits with remote monitoring transmissions.^3–6 Moreover, RPM enables frequent assessment of device-related parameters (eg, lead impedance and battery status). This may allow early detection of device and lead malfunction.^7–11 Thus, the routine use of RPM enhances device safety and resource use and may improve clinical outcomes.^12–17

Clinical Perspective on p 2383

Currently, all ICDs are RPM-capable, and professional societies endorse the routine use of RPM in clinical practice.^18,19 The extent to which RPM has been incorporated into routine clinical practice and the factors associated with failure to use RPM have not been identified. Successful transmission of RPM data by the patient to the healthcare provider relies on 2 main factors: (1) enrollment of the patient into the specific RPM system and (2) subsequent activation of RPM by the enrolled patient to allow transmission of device information. Barriers to widespread adoption of RPM may occur at both the RPM enrollment and RPM activation/transmission stages. Enrollment into RPM may depend largely on the local practice environment of the implanting physician and institution. However, RPM activation and transmission by the patient may be more dependent on patient factors. Accordingly, we sought to determine the degree of RPM use and to understand the patient, physician, and institutional determinants of RPM use in patients with newly implanted ICDs.

Methods

This study represents a collaborative effort between Boston Scientific Corporation, the American College of Cardiology Foundation and the Yale/New Haven Hospital Center for Outcomes Research and Evaluation. Use of the ALTITUDE database was approved by Boston Scientific Corporation. Use of the American College of Cardiology Foundation National Cardiovascular Data Registry ICD Registry was approved by the ICD Registry Research and Publications Committee. Data set linkage and analysis was approved by the Yale University School of Medicine Human Investigation Committee.

Data Sources

A limited data set without patient identifiers was derived from the Boston Scientific Corporation ALTITUDE database, which includes the patients’ data related to the ICD implantation (date of implant, device model number, age at implant, sex, zip code, survival days from implantation), the RPM data (enrollment date/time, device data transmission date/time), and the implanting hospital (Medicare provider number, facility name and address). A similarly limited data set was derived from the ICD Registry. The registry captures ≈90% of all implants being performed in the United States, regardless of device indication or patient age in their reporting periods, and contains detailed information about patient and provider characteristics.²⁰ Participating hospitals met the data quality standards established by the National Cardiovascular Data Registry of 95% completeness of specific core data elements and participation in a site-auditing program (annual on-site chart review and data abstraction for at least 5% of participating sites). The US 2000 Census of Population and Housing and the Area Resource File 2008 Release were used to identify regional characteristics such as socioeconomic status, education, and population density.

Cohort Derivation

The study cohort was derived by linking the limited data set derived from the ICD Registry with the limited data set derived from the Boston Scientific ALTITUDE data set with the use of indirect methods. The characteristics of the patients who did not match were compared with those who matched to assess for any bias introduced by the matching process. The following fields were used for linking the 2 data sets: the hospital Medicare provider number, patient age, sex, and implantation date. The algorithm used to derive the cohort is documented in Figure 1. Before the merge, the following exclusion criteria were applied: age >89 years, implant date not between 2006 and 2010, nonwireless devices, previous ICD implantation to eliminate RPM use bias introduced by previously implanted devices, unknown age, sex, or invalid hospital ID, patients in hospitals with a missing Medicare provider number or in hospitals sharing the same number, and duplicate admissions. Patients were excluded if they could not be linked between the ALTITUDE data set and the ICD Registry data set. After the match, the following exclusion criteria were applied to the linked cohort: patients who had received a cardiac transplant or surgical epicardial lead, patients implanted at hospitals not reporting all their ICDs, patients with an unknown vital status, patients who died during the hospital stay, and patients with erroneous enrollment/activation data (eg, RPM activation date before RPM enrollment date).

• Download figure
• Download PowerPoint

Outcomes

Patients were categorized into 3 groups: (1) patients who were not enrolled in the Boston Scientific RPM LATITUDE system within 180 days of implant (RPM-not enrolled), (2) patients who were enrolled in the Boston Scientific RPM LATITUDE system but did not transmit information within 180 days of implant (RPM-inactive), and (3) patients who were enrolled in the Boston Scientific RPM LATITUDE system and actively transmitted within 180 days of implant (RPM-active). Although RPM enrollment and activation typically occur within a few weeks after the device implant, the 180-day timeframe was chosen as a clinically relevant period that captures the majority of RPM enrollments and activations. Furthermore, RPM enrollment and activation occurring >180 days following device implantation are likely related interim changes such as deterioration in health status and would not accurately reflect the patient, physician, and institutional factors associated with a strategy of routine, early RPM use. Sensitivity analysis was performed varying the definition of RPM enrollment and activation to 90 days.

Statistical Analysis

Patient, physician, and hospital characteristics were compared between RPM-enrolled and RPM-not enrolled patients, and also between RPM-inactive and RPM-active patients. After excluding institutions with very few device implantations (<25 devices), the hospitals were divided into quartiles based on the percentage of patients at each hospital that were enrolled in RPM. Baseline patient characteristics were compared based on the quartiles of hospital enrollment. The χ² test or analysis of variance was used for categorical variables and continuous variables, respectively. To identify the candidate variables most strongly associated with RPM enrollment or RPM activation, the study population was randomly assigned to derivation (70%) and validation (30%) cohorts

A model was developed to predict the RPM enrollment in the derivation cohort by using multivariable logistic regression analysis with stepwise selection (entry P value=0.1 and retention P value=0.05). For categorical variables with a low missing rate, we imputed missing values by using the most common category of each variable. For categorical variables with a missing rate >5%, a category was added to indicate the missingness. For continuous variables, missing values were imputed by using the median value of the entire cohort of nonmissing patients, and a dummy variable to indicate the missingness was created. A bootstrap approach to variable selection was used to assess the stability of the variables selected into the model. This consisted of bootstrapping the derivation cohort with 1000 iterations and reselecting variables within each iteration. Variables selected in >70% of the replications were included into the final model. Model discrimination in the derivation and validation cohorts was evaluated by the C statistic. For model validation, the coefficients of the model from derivation cohort were applied to the validation cohort assessing the predicted versus observed RPM enrollment rate within deciles of the predicted RPM enrollment rate. The selected variables were then incorporated into a hierarchical logistic regression model to account for the clustering of the patients within hospitals, and a hospital- specific median odds ratio was calculated to characterize the variation between hospitals in the propensity to enroll patients in RPM.²¹ The median odds ratio reports hospital level variation in RPM enrollment as an odds ratio and is defined as the median value of the odds ratios between a hospital at highest risk and that at lowest risk when chosen at random. The median odds ratio is statistically independent of prevalence and is thus a means to assess the extent in which the probability of individual RPM enrollment is determined by hospital.

Finally, the determinants of RPM activation among the RPM-enrolled patients were identified by using the approach outlined above. We conducted a secondary analysis excluding patients who died within 180 days. All analyses were performed on SAS Version 9.3.

Results

Derivation of Study Cohorts

The derivation of the study cohorts is shown in Figure 1. The overall match rate of the limited ALTITUDE data set to the limited ICD Registry data set was 90.9%. There were no clinically significant differences in patient characteristics between the matched cohort and those who did not match (tables in the online-only Data Supplement). The characteristics of ICD Registry patients who received an eligible Boston Scientific device and matched with the corresponding ALTITUDE data set were compared with those who did not match. This article focuses on the study cohort that includes deaths within 180 days following device implant. The 180-day cutoff for defining enrollment and activation captured >90% of all RPM-enrolled patients and >75% of all RPM-active patients. The distribution of the days from device implant to RPM enrollment and from device implant to RPM activation is shown in Figure 2.

• Download figure
• Download PowerPoint

Patient Characteristics

As shown in Figure 1, 39 158 patients from 883 hospitals were linked between the ALTITUDE database and ICD Registry and met study inclusion criteria. Of these, 24 113 patients (62%) were enrolled within 180 days of device implant. Among RPM-enrolled patients, 18 289 (76%) patients activated their device within 180 days (RPM-active) and 5824 (24%) did not activate their device (RPM-inactive). Thus, the effective rate of RPM use (both enrolled and activated) was 47%, in large part, owing to the lack of patient enrollment in the RPM program. Table 1 shows the characteristics of the different subgroups based on RPM enrollment and RPM activation status.

Determinants of Enrollment in RPM

Table 2 shows patient characteristics based on quartiles of hospital RPM enrollment rates. Clinically meaningful differences were observed in race/ethnicity, insurance status, and geographic region. A total of 21 variables were significantly associated with RPM enrollment in the final logistic regression model (Figure 3). The area under the receiver operating characteristic curve was 0.67. For RPM enrollment, the C statistics in the derivation and validation cohorts were 0.669 and 0.668. Several baseline characteristics including race, ethnicity, health insurance status, and geographic factors were associated with RPM enrollment. Specifically, patients with nonwhite ethnicity, those with Medicaid or no medical insurance, and residents of the mid- and south-Atlantic and Pacific regions were less likely to be enrolled in RPM. Residents of rural areas were more likely to be enrolled in RPM than those living in more population-dense regions.

• Download figure
• Download PowerPoint

Disease-related patient characteristics were also related to RPM enrollment. Patients who were electively admitted for ICD implantation and those receiving cardiac resynchronization devices were more likely to be enrolled. The presence of comorbidities such as lung disease, renal dysfunction, hyponatremia, or atrial fibrillation/flutter was associated with lower RPM enrollment.

Physician characteristics associated with increased likelihood of RPM enrollment included electrophysiology board certification and Boston Scientific Corporation manufacturer implant volume. However, the single most important determinant of RPM enrollment was related to the providing institution in which the ICD was implanted. The hospital-specific median odds ratio for RPM enrollment was 3.43, signifying a strong institutional effect in which a randomly selected patient receiving an ICD at 1 hospital would have a 3.5 times higher odds of being enrolled in RPM than an identical patient at a different randomly selected hospital in the sample. Furthermore, hospitals varied widely in the proportions of eligible patients who were enrolled in the RPM program (Figure 4).

• Download figure
• Download PowerPoint

Determinants of Activation and Transmission of RPM

A total of 17 variables were identified as correlates of RPM activation (Figure 5). The area under the receiver operating characteristic curve was 0.62. For RPM activation, the C statistics in the derivation and validation cohorts were 0.624 and 0.617. Patient-specific factors played a more prominent role in determining RPM activation in comparison with the model for RPM enrollment. Factors such as race, health insurance, and reason for admission influenced RPM activation in a manner similar to RPM enrollment. In addition, an age-related increase in the likelihood of RPM activation was observed. The presence of comorbidities including diabetes mellitus, renal disease, hemodialysis, and hypo- or hypernatremia, and severe impairment in left ventricular function, as well, were associated with a lower likelihood of RPM activation. Patients experiencing procedure-related adverse events during the index hospital stay were also less likely to activate RPM. A distance-dependent increase in the likelihood of RPM activation was observed up to 100 miles from the implanting facility. This effect diminished with distances >100 miles. Geographic variation also played a role, with patients living in the Pacific region being significantly less likely to activate RPM than patients living in other census regions.

• Download figure
• Download PowerPoint

Physician characteristics did not play a very significant role in affecting RPM activation. Similarly, the hospital median odds ratio for RPM activation was 1.69, demonstrating a smaller institutional effect in comparison with RPM enrollment. In sensitivity analyses, results of the models for RPM enrollment and RPM activation were not significantly changed when we excluded patients who died within 180 days or changed the time period for both enrollment and activation to 90 days.

Discussion

This is the first study to examine the patterns of patient enrollment in RPM programs, and reasons for the failure of enrolled patients to actively transmit remote data, as well. Our study demonstrates that, despite its universal availability, RPM technology is substantially underused in this patient population. Less than half of patients who receive an RPM-capable device use this technology within 180 days of implant. Lack of enrollment of eligible patients into RPM systems is the major cause of underuse, with more than one-third of all patients receiving RPM-capable devices not being enrolled within 180 days of device implant. Furthermore, the use of RPM is not uniform among different institutions. Whether creating institutional systems that incorporate allied health professional and ancillary support to facilitate patient enrollment as a routine part of postimplant care would increase RPM use needs to be examined.

The use of RPM is advocated by professional societies.^18,19 In clinical studies, RPM use has been demonstrated to reduce resource use, and it may improve safety and patient outcomes.^{3–6,12–17} However, whether these findings apply to the real-world clinical experience is unknown. Our findings suggest that physician and hospital factors play an important and potentially modifiable role in determining who is enrolled into an RPM system. Patient enrollment varied significantly across hospitals, with 117 (13.3%) of hospitals not enrolling any patients, and only 93 (10.5%) of hospitals enrolling all their eligible patients. Although differences in patient characteristics relating to race/ethnicity, insurance status, and geographic region exist between institutions with low versus high RPM enrollment, these were accounted for in the hierarchical model. Furthermore, the hospital-specific median odds ratio for RPM enrollment that accounts for patient clustering was consistent with a strong institutional effect that may reflect differences in the extent to which clinicians and institutions have adopted this technology. The logistics of RPM enrollment, in general, depend on the physician/institution, but this process may be affected by external factors such as local industry representative availability.

We observed significant geographic variation in RPM use. Enrollment into RPM systems was lower along coastal regions of the United States in comparison with the central United States and in urban areas in comparison with suburban regions. Similarly, patients who lived closer to the implanting center were in general less likely to be enrolled than patients who lived further away. Although this finding may be related to clinicians making pragmatic assessments about who would benefit most from RPM, it may simply reflect patient-related factors such as a desire to have in-person visits with the healthcare provider as long as it is not very inconvenient. Of note, our study indicates that enrollment in and activation of RPM programs may be underused among certain populations, including patients with no health insurance, and racial and ethnic minorities, as well. These observations are consistent with previous reports demonstrating racial and socioeconomic differences in the use of medical devices and indicate that, despite receiving an ICD, there are differences in the use of remote follow-up in these populations.^22,23 However, the reasons underlying these differences are not known. One potentially important factor is access to a telephone landline, a requirement for the RPM devices studied in this analysis. The availability of a telephone landline varies based on age, race/ethnicity, geography, and socioeconomic status.^24–26 Further study will be required to fully explore the complex relationships between these factors, RPM use, and landline availability.

Finally, patients with comorbidities, severe left ventricular dysfunction, and in-hospital procedure-related complications and those not admitted specifically for device implantation were less likely to use RPM. Thus, sicker patients who may be more likely to benefit from RPM were less likely to use this technology, which is an example of a risk-treatment paradox. The underlying reasons for this paradox are unknown, but it is possible that sicker patients prefer or are deemed by clinicians to need in-office encounters. Furthermore, they may be more frequently hospitalized, or may spend a considerable amount of time in intermediate-care facilities, and thus are not enrolled as frequently in RPM.

Limitations

This study examined the RPM system of a single device manufacturer. Although this may limit extension of the findings to other RPM systems, the logistics of RPM enrollment and activation are fairly similar among the major RPM systems that rely mainly on landline use in the United States. This report uses Area Resource File data that reflect the proportion of households with a landline in a particular zip code as opposed to whether an individual patient had a landline or not. Moreover, the area resource file and census data were obtained before the performance of this study and may not exactly reflect current conditions.

Although this study examined individual physician-related determinants of RPM use, it could not distinguish between group physician practice and institutional factors. The institutional variation observed in this study is related to differences in local group practice that is largely determined by physician groups at implanting institutions. Furthermore, the degree to which reimbursement plays a role in the adoption of RPM is unknown and warrants further examination. Specifically, the Centers for Medicare and Medicaid Services released new Current Procedural Terminology codes for RPM reimbursement in January of 2009 that may have affected the use of RPM during the course of the study.

The models provide relatively modest discrimination, and thus other factors are involved in RPM enrollment and activation. This report does not contain information about changes in clinical status following ICD implantation that might have influenced RPM enrollment or activation. As mentioned, limiting the RPM enrollment and activation period to 180 days postimplant diminishes the likelihood of occurrence of such events. The 180-day timeline is a clinically relevant period that minimizes change in patient clinical status as a cause of RPM enrollment or activation. The intent of this study was to examine the degree and the determinants of RPM enrollment and initial activation. Examination of compliance and continuing use of RPM following initial activation is beyond the scope of this study. Furthermore, the extent to which increased RPM use leads to enhanced patient outcomes was not examined in this study and needs to be further evaluated.

Conclusions

Despite its availability, the use of RPM in patients with ICDs is low, owing in large part to the failure to enroll in RPM programs. The local practice pattern is the most important determinant of RPM enrollment. Differences exist between determinants of RPM enrollment and those of RPM activation leading to transmission. RPM enrollment depends on provider and institutional factors, whereas RPM activation depends more on patient-related factors. Because many of the determinants of RPM use are modifiable, healthcare providers, institutions, and payers may consider strategies that promote RPM use.

Acknowledgments

We thank Bonnie Garmisa for her invaluable help in organizing this project and preparing the manuscript.

Footnotes