Big data have several advantages, including extremely large sample size and associated high power to detect differences. Big data are also generally thought to have good generalizability or “whole-system relevance” because of their incorporation of large-scale representative samples from the population. As per the third V (ie, velocity), big data are advantageous because they can provide real-time analysis, usually in a frequent or iterative fashion. Finally, big data often have the potential to increase the efficiency and cost-effectiveness of research. However, the operative word here is “potential.” Like any question in research, the answer is only as good as the source or data that support it. Although big data may provide a very large sample with which to answer a given question, big data techniques are only efficient if they have the information and detail required to answer the relevant clinical question. This is one of several potential limitations to big data and big data approaches.

Big data are also beset by challenges. Like any observational analysis, including prespecified ones, big data techniques can identify association but they cannot determine causality without randomization. Thus, there is great interest in deploying randomization techniques in the context of big data resources. Another relevant challenge with big data is that they are also limited by the types of data and variables available. Big data cohorts often have trade-offs that investigators need to consider. Although big data can bring large patient numbers that may be of great value, there may also be a loss of data “depth.” Lack of “depth” in big data sources can present several difficulties, including an inability to adjust for known confounders. Big data are often collected under standard operations and thus may not include important variables that have relevance for research; there may also be considerable missing data for standard or expected variables. In remote monitoring databases, structured variables such as the number of intervals to detect are readily available, but data on device indications, patient characteristics, or appropriateness of shocks are not available. Thus, missing data and overall data quality are important areas of concern in many big data analyses. Finally, it is important to note that both ethical and privacy concerns are particularly noteworthy in big data. Inadvertent disclosures because of cybersecurity lapses are an important concern for multiple stakeholders, including regulatory bodies, patients, and researchers.

Acknowledging the strengths and limitations of big data, large datasets remain an important asset in CIED research. These data represent an opportunity to investigate ways of optimizing healthcare delivery and patient outcomes.⁴ In the following sections, we will review how big data have contributed to the field of remote monitoring, particularly in the areas of healthcare utilization, atrial fibrillation, implantable cardioverter defibrillator (ICD) therapies, and cardiac resynchronization therapy (CRT).

The ALTITUDE investigators were among the first to evaluate the association between remote monitoring and all-cause mortality. In an analysis that included 69 556 patients with Boston Scientific ICD or cardiac resynchronization therapy-defibrillator (CRT-D), the use of remote monitoring was associated with a 50% lower risk of
all-cause death compared with in-person follow-up only (hazard ratio [HR] of ICD [HR_ICD] 0.56 with P < 0.0001 and HR_CRT-D 0.45 with P < 0.0001).⁵

This association was confirmed in another large cohort study that included 269 471 patients with St Jude pacemakers, ICDs, and CRT devices.⁶ In this confirmatory nationwide analysis, only 47% of whom were followed with remote monitoring, the majority of the patients in the cohort were never followed with remote monitoring. Utilization of remote monitoring was associated with improved survival and this association exhibited a “dose-dependent” relationship. When the percentage of follow-up with remote monitoring was 75% or greater, patient survival was 2-fold greater (HR 2.10, 95% confidence interval [CI] 2.04-2.16, P < 0.001) compared to patients without remote monitoring. When high-use remote monitoring was compared to infrequent remote monitoring, a beneficial association was still observed with all-cause survival (HR 1.32, 95% CI 1.27-1.36). In a subsequent analysis from the same large database (n = 106 027 patients), earlier initiation of remote monitoring was also associated with improved survival.⁷ In this analysis, patients who initiated remote monitoring early (median 4 weeks [interquartile range (IQR) 2-8]) had a lower mortality rate and improved adjusted survival (HR 1.18, 95% CI 1.13-1.22, P < 0.001) compared with patients who received delayed initiation of remote monitoring (median 24 weeks [IQR 18-34 weeks]). Lower rates of mortality in patients with early initiation of remote monitoring were observed across all device types, including pacemakers, defibrillators, and cardiac resynchronization devices.

Thus, the data from these large single-vendor databases suggest that remote monitoring is associated with lower mortality. Moreover, these data suggest that earlier initiation and consistency of remote monitoring are also associated with lower mortality. Based upon these reports and other evidence, current expert consensus documents recommend remote monitoring in addition to at least annual in-person visits (Class I, Level of Evidence A; Table 8.1).¹ Moreover, it is recommended that initiation of remote monitoring within 2 weeks of device implantation may be beneficial
(Class IIa, Level of Evidence C). Despite the advantages offered by remote monitoring, implementation of remote monitoring and patient adherence remain challenging. Data from US patients enrolled in the Medtronic CareLink remote monitoring system suggest adherence (leniently defined as ≥2 transmissions in a 14-month period) is imperfect, with overall nonadherence at 21%. Of note, patients 40 years or younger were more than twice as likely to be nonadherent (odds ratio [OR] 2.64, 95% CI 2.42-2.88).

Data on the cost-effectiveness of remote monitoring have been mixed. In the Clinical Evaluation of Remote Notification to Reduce Time to Clinical Decision (CONNECT) trial, patients undergoing remote monitoring had a shorter mean length of stay per hospitalization compared with patients followed in the office only (3.2 vs 4.3 days, P = 0.007). This reduction in length of stay was associated with approximately $2000 lower costs per hospitalization.⁸ In the Effectiveness and Cost of ICD follow-up Schedule and Telecardiology (ECOST) trial, which was conducted in Europe, the cost of hospitalization in patients randomized to remote monitoring was &U20AC;720 lower, although this difference was not statistically significant (P = 0.46).⁹ Overall, the evidence from clinical trials suggests remote monitoring is cost-effective and leads to cost savings based upon a meta-analysis of 21 randomized trials.¹⁰