Author + information
- Received May 25, 2009
- Revision received August 31, 2009
- Accepted August 31, 2009
- Published online October 27, 2009.
- Catherine Klersy, MD, MSc⁎,⁎ (, )
- Annalisa De Silvestri, MSc⁎,
- Gabriella Gabutti, MA†,
- François Regoli, MD⁎,‡ and
- Angelo Auricchio, MD‡
- ↵⁎Reprint requests and correspondence:
Dr. Catherine Klersy, Servizio di Biometria ed Epidemiologia Clinica, Fondazione IRCCS Policlinico San Matteo, 27100 Pavia, Italy
Objectives The purpose of this study was to assess the effect of remote patient monitoring (RPM) on the outcome of chronic heart failure (HF) patients.
Background RPM via regularly scheduled structured telephone contact between patients and health care providers or electronic transfer of physiological data using remote access technology via remote external, wearable, or implantable electronic devices is a growing modality to manage patients with chronic HF.
Methods After a review of the literature published between January 2000 and October 2008 on a multidisciplinary heart failure approach by either usual care (in-person visit) or RPM, 96 full-text articles were retrieved: 20 articles reporting randomized controlled trials (RCTs) and 12 reporting cohort studies qualified for a meta-analysis.
Results Respectively, 6,258 patients and 2,354 patients were included in RCTs and cohort studies. Median follow-up duration was 6 months for RCTs and 12 months for cohort studies. Both RCTs and cohort studies showed that RPM was associated with a significantly lower number of deaths (RCTs: relative risk [RR]: 0.83, 95% confidence interval [CI]: 0.73 to 0.95, p = 0.006; cohort studies: RR: 0.53, 95% CI: 0.29 to 0.96, p < 0.001) and hospitalizations (RCTs: RR: 0.93, 95% CI: 0.87 to 0.99, p = 0.030; cohort studies: RR: 0.52, 95% CI: 0.28 to 0.96, p < 0.001). The decrease in events was greater in cohort studies than in RCTs.
Conclusions RPM confers a significant protective clinical effect in patients with chronic HF compared with usual care.
Patients with chronic heart failure (HF) frequently experience repeated hospitalizations that are not only a result of progression of underlying disease but more often due to poor adherence to drug therapy, inadequate drug therapy, changes in diet, poor self-care, and inadequate patient support. Approximately 70% of all direct and indirect costs generated by HF patients are due to hospitalization (1). Recent guidelines of both European and American scientific societies recommended a multidisciplinary care approach that coordinates care along the continuum of HF and throughout the chain of care delivery by various services within the health care systems (1,2)
The multidisciplinary HF care approach is implemented by in-person follow-up visits and is regarded as usual care of HF patients. More recently, alternative approaches have been proposed including regularly scheduled structured telephone contact between patients and health care providers and electronic transfer of physiological data using remote access technology via external, wearable, or implantable electronic devices. This latter approach allows frequent or continuous assessment of some physiological parameters related to HF exacerbation, and such technology-based monitoring is the base for early detection of HF worsening, thus permitting remote disease management (1).
Results of some randomized controlled trials (RCTs) (3–22) and several observational studies (23–34) support the hypothesis that the multidisciplinary care approach or the management strategy of structured communication with the health care provider may reduce both the incidence of hospitalizations and death and eventually related costs with respect to more traditional follow-up of patients with chronic HF. Moreover, recent systematic reviews and meta-analyses provided further evidence in favor of implementation of telemonitoring in chronic HF patients (35–38). Of note, guidelines recommend remote monitoring of symptoms (including drug adverse effects) and signs of HF (Class I recommendation, Level of Evidence: C) (1). Since the last review by Clark et al. (35), which exclusively included RCTs and all published articles until mid-2006, several observational studies and RCTs have become available (6,9,10,21).
To update earlier systematic reviews, we conducted a literature search including both RCTs and observational studies and performed a meta-analysis of the use of telemonitoring in chronic HF and its related outcomes compared with usual care of HF patients.
The National Guideline Clearinghouse, PubMed, EMBASE, Cinhail, and the Cochrane Library databases were searched throughout October 2008 using the following search criteria: 1) full-text articles in peer-reviewed journals published between January 2000 and October 2008 in which at least 2 treatment arms were evaluated (thus uncontrolled studies were excluded); 2) RCTs or observational cohort (C) studies; and 3) language of publication could be English, Spanish, German, French, or Italian (Table 1).A total of 253 abstracts were retrieved; however, 56 studies were excluded because they were simultaneously present in more than 1 database; thus, 197 abstracts were finally collected and reviewed.
For each article, we collected the following information: type of study (multicenter or single center), total number of patients included in the trial, number of arms/periods, mean duration of follow-up, age, sex, New York Heart Association functional class, and left ventricular ejection fraction of included patients and for each arm, person-years of follow-up, and the modality of care. Three different approaches of care were identified: 1) a usual care approach, which referred to in-person visits at the doctor's office, at a multidisciplinary outpatient clinic, or at emergency department without additional phone calls to and from the patient; 2) a telephone monitoring approach including regularly scheduled structured telephone contact between patients and health care providers (with or without home visits) and reporting of symptoms and/or physiological data; and 3) a technology-assisted monitoring approach relying on information communication technology, with transfer of physiological data collected via remote (at the patient's home) external monitors or via cardiovascular implantable electronic devices. Finally, the latter 2 approaches were collectively considered and identified as remote patient monitoring (RPM).
For each of these approaches, great attention was paid to retrieval of information about clinical and cardiovascular parameters monitored such as symptoms, body weight, blood pressure, electrocardiogram, heart rate, arrhythmias, shock device, heart rate variability, activity log, oxygen saturation, and right ventricular pressures.
The following outcomes were considered: death from any cause, first hospitalization for any cause and first hospitalization for HF, and a combined end point of first hospitalization or death from any cause. Only a few articles reported the cause-specific mortality, and, thus, this was not included.
Two authors (C.K. and A.D.S.) reviewed all abstracts and selected articles to ensure that they met the inclusion criteria. Each of them separately extracted the information from the articles, and whenever a discrepancy was noted, it was reconciled by consensus. The quality of the study was rated based on adherence to the CONSORT and STROBE statements and graded on a 0 to 10 visual analog scale.
The primary end point of the study was the comparison of the cumulative incidence of events (number of patients with events/total number of patients per arm) between the usual care approach and RPM strategies (telephone and technology-assisted monitoring approaches) for each of the outcomes considered (death from any cause, first hospitalization for any cause, and hospitalization for HF), as well as for the combined outcome of first hospitalization or death from any cause. Death and hospitalization from any cause were assessed separately for RCTs and cohort (between) studies, whereas hospitalization for HF could be assessed for RCTs only. A secondary analysis of the primary end point comparing incidence rates of events (number of events/total person time per arm) gave similar results (data not shown).
To assess the stability of our conclusions, we performed a series of sensitivity analyses: 1) comparison of the cumulative incidence of events with the usual care approach with that of the telephone monitoring and the technology-assisted monitoring approach; 2) comparison of the cumulative incidence of events with the usual care approach with that of the RPM approach, according to duration of follow-up (≤6 and >6 months) and study quality (<8 and ≥8 on a visual analog scale).
The relative risk (RR) and 95% confidence interval (CI) for each outcome in each study were calculated. Study RRs were then pooled according to the Mantel-Haenszel fixed-effects method. To better account for differences among studies, we also fitted DerSimonian and Laird random-effects models. Statistical heterogeneity was evaluated by the Cochran Q test and measured by the I2statistic. When the I2statistic was >20%, we considered the random-effects RR to be preferable. The presence of severe publication and other biases, for each outcome, was excluded by funnel plots (Online Appendix).
Stata version 10.1 (Stata Corp., College Station, Texas) was used for computation.
Identification of articles
Sixty-three full-text articles were retrieved from 197 abstracts; an additional 33 full-text articles were identified from references in the articles retrieved; thus, a total of 96 full-text articles were collected. However, 64 of them were excluded for 1 or more of the following reasons: duplicate publication, review or meta-analysis, lack of pertinence, study protocol, insufficient information provided, or lack of a usual care arm (Fig. 1).They are listed in the Online Appendixtogether with the reason for exclusion. Thus, 20 RCTs (3–22) and 12 cohort studies (23–34) were available. Six cohort studies had a between-arm and 6 had a within-arm (before-after) design.
As shown in Table 2,RCTs were evenly distributed over the 9 years of literature reviewed, and 4 of them were published in 2008. Results of cohort studies started to be published only in the year 2001. A limited proportion of studies (45% of RCTs and 17% of cohort studies) had been conducted at multiple centers.
A total of 6,258 patients were enrolled in RCTs and 2,354 patients in cohort studies. Women were well represented (36% and 40% of the study population for RCTs and cohort studies, respectively). The median age over studies was 70 years in RCTs and 66 years in cohort studies. Median left ventricular ejection fraction was 35% in RCTs and 40% in cohort studies. Fifty-four percent of patients in RCTs were of New York Heart Association functional class III to IV compared with 83% of patients in cohort studies. However, left ventricular ejection fraction and New York Heart Association functional class were only very partially retrievable and thus are to be interpreted with caution.
Median follow-up duration was 6 months, and approximately 25% of the follow-up was 3 months or less in RCTs. In contrast, median follow-up duration was 12 months in cohort studies; only 1 study had a follow-up of 3 months or less. One-half of the RCTs had a high-quality scoring (≥8) compared with only 17% of cohort studies.
RPM approach (telephone and technology-assisted monitoring)
Among the RCTs, 2 studies (3,5) compared 3 strategies (usual care, telephone monitoring, and technology-assisted monitoring), whereas the remaining 18 studies (4,6–22) compared the usual care approach with the telephone-monitoring approach (11 studies) or technology-assisted monitoring approach (7 studies). In contrast, in all cohort studies, the usual care approach was always compared with the technology-assisted monitoring approach. Detailed methodology of care in the analyzed studies is given in Table 3.Of note, in 9 of 20 RCTs and in 9 of 12 cohort studies, the modality for performing usual care was not provided.
In RCTs, symptoms and body weight were always recorded. In a similar manner, symptoms and body weight were monitored in nearly all cohort studies (11 of 12 studies and 10 of 12 studies, respectively). Blood pressure, patient physical activity, and heart rate were monitored in a limited number of RCTs (5 of 20 studies); in contrast, blood pressure and heart rate were monitored twice as frequently in cohort studies than in RCTs (7 of 12 studies and 8 of 12 studies, respectively). All other cardiovascular parameters were monitored anecdotally (1 to 2 studies each), as detailed in Figure 2.
Incidence rates of events
Table 4shows the study outcomes for both RCTs and cohort (between) studies in terms of cumulative incidence and incidence rates per 100 person-years, together with their 95% CIs for each of the relevant outcomes and for each care approach. Table 4also reports the total number of patients, the total number of patients with events, and the person-years of observation.
In the cohort studies with a before-after design, the incidence rate of hospitalization was calculated over 4 studies and equaled 86.4 (95% CI: 75.7 to 98.3) in the study period with the usual care approach and 21.1 per 100 person-years (95% CI: 16.0 to 27.4) in the study period with the RPM approach. The incidence rate of hospitalization for HF was calculated over 2 studies and equaled 100% (95% CI: 98% to 100%) in the study period with the usual care approach and 28.4 per 100 person-years (95% CI: 21.1 to 37.3) in the study period with the RPM approach. Overall, the incidence of death, hospitalization for all causes, hospitalization for HF, or combined death and hospitalization was lower with the RPM compared with the usual care approach.
Outcomes in RCTs
RPM was associated with a significantly lower number of deaths (RR: 0.83, 95% CI: 0.73 to 0.95, p = 0.006) compared with usual care (Fig. 3A). No heterogeneity between studies was shown (p = 0.82). A similar, yet less important, protective effect was found comparing RPM with usual care when hospitalization was considered (RR: 0.93, 95% CI: 0.87 to 0.99, p = 0.030); the higher heterogeneity (18.4%, p = 0.27) resulted in a loss of statistical significance when considering the random-effects RR (Fig. 3B). The strongest protective effect of RPM compared with usual care was found when hospitalization for HF was analyzed (RR: 0.71, 95% CI: 0.64 to 0.80, p < 0.001) with little heterogeneity between studies (Fig. 3C). The combined end point of death and first hospitalization (available in a few studies) gave comparable results (RR: 0.86, 95% CI: 0.79 to 0.94, p < 0.001). Some heterogeneity was present (28%, p = 0.22), although without loss of significance in the random-effects model (Fig. 3D).
In a sensitivity analysis, the telephone or technology-assisted monitoring approach provided an equally large benefit compared with usual care for almost every outcome. Of note, the protective effect on death of the technology-assisted approach was slightly greater than that provided by the structured telephone approach (Table 5). Moreover, when RRs were computed separately according to duration of follow-up (short/long) and to quality of study (low/high), the association between RPM and death or RPM and hospitalization for HF was maintained (Table 5).
Outcomes in cohort studies
Figure 4shows the effect of RPM on death and hospitalization. RPM was associated with a significantly lower number of deaths (random-effects RR: 0.53, 95% CI: 0.29 to 0.96, p < 0.001) and hospitalizations (random-effects RR: 0.52, 95% CI: 0.28 to 0.96, p < 0.001) compared with usual care. High heterogeneity between studies was shown (I2= 59% for death and I2= 82% for hospitalization) (Figs. 4A and 4B).
This meta-analysis showed that RPM significantly reduced the risk of death, hospitalization for any cause, and hospitalization for HF compared with usual care in RCTs. The reduction in risk of death and hospitalization was even more pronounced when meta-analyzing cohort studies.
Our analysis confirms, extends, and updates previous systematic reviews (35–38). To the best of our knowledge, this represents the largest number of meta-analyzed patients. A previously published analysis (35) included 14 RCTs (all published before May 2006) and totaled 4,369 patients; in contrast, our analysis included 20 RCTs (all published before October 2008) and reported on 6,133 patients, which represents 1,764 patients or 40.3% more patients than the most recent meta-analysis of RCTs by Clark et al. (35). Since the publication of that meta-analysis, 5 additional RCTs were published, including 4 in 2008 (6,9,10,21); moreover, nearly all cohort studies included in our meta-analysis were published in the past 5 years. That clearly indicates a growing interest in RPM in the cardiology community; a similarly great interest in telemedicine is also found at the highest European institutional level, which recently urged health care systems across Europe to embrace the technology. There are, however, technological and organizational challenges for health care systems that are expected to increase in view of the aging population in Western countries and the increase in the prevalence of HF in the general population worldwide. Which parameters to monitor, how to monitor them most efficiently, and how to organize the response of the health care professionals to data obtained from monitoring to optimize patient care are all questions that need to be answered.
In contrast to most previous reviews and meta-analyses, our study included both RCTs and cohort studies. RCTs are very accurate in their design but may not reflect real life well enough, which, in contrast, is probably better represented in cohort studies; conversely, cohort studies often do not sufficiently control for confounding factors. The fact that the use of RPM in cohort studies led to a reduction of both mortality and hospitalization, which was of similar magnitude or even greater than that observed in RCTs, may be considered a confirmation of the value of the technology as such. This observation also emphasizes the importance of including in meta-analyses both RTCs and cohort studies. There are additional differences between this meta-analysis and most of the previous meta-analyses; indeed, our meta-analysis and that of Clark et al. (35) considered only those studies in which RPM was compared with usual care. These criteria are more stringent than those of other meta-analyses (36–38), which included different multidisciplinary approaches. The beneficial effect of RPM on mortality and hospitalization observed in our meta-analysis was consistent with that reported by Clark et al. (35) (Table 6),who, however, focused their analysis on RCTs. Some difficulty may exist in comparing the results of our meta-analysis with those of a previous review of cohort studies by Gonseth et al. (36). They included cohort studies that were published before 2004 and summarized the data of comprehensive disease management programs and very little, if any, of technology-assisted RPM. Telemonitoring technology has substantially changed over the past years and has moved from structured phone contact to more automatic external, wearable, or implantable devices (21,31,39). In addition, modern technology relies more and more on central servers, sophisticated algorithms for automatic review of transmitted data and alerts, expert systems for direct interaction with either the patient or health care provider, and dedicated online health care providers and in-hospital remote device and disease management units (39,40).
Although well-designed cohort studies may be as accurate as RCTs (41,42), they should be considered with caution because they often do not sufficiently control for confounding factors, thus resulting in high estimates of risk. As a matter of fact, the estimates of the pooled risk reduction of death and hospitalization in the cohort studies in our meta-analysis were unadjusted for confounding, which resulted in an almost 50% risk reduction compared with at best 20% in the RCTs (Table 6). Moreover, only a few cohort studies satisfied our inclusion criteria, leading to wide CIs for the estimates of RR. An additional caveat exists regarding the interpretation of observational studies included in this meta-analysis because enrollment in the RPM arm was voluntary; this might bias the results toward a high adherence to the monitoring program given the high level of motivation of these patients and resulting in a lower incidence of events. Moreover, clinical, social, and/or demographic factors may be considered additional confounders and should be adjusted for in any analysis. All these reasons emphasize the difficulty in stating a difference in effect size between RCTs and cohort studies. It is important to note that same drawbacks apply to a previous meta-analysis and explain the similar large differences observed between estimates of risk reduction in RCTs and cohort studies (36). Future cohort studies should be designed to assess the role of RPM in large populations having as end points death and hospitalization (any cause and HF) while accounting for the underlying patients' characteristics.
Because RPM exclusively collects symptom and physiological data related to HF, a greater effect on death or hospitalization from HF rather than on death or hospitalization from any cause was expected. In keeping with the stated hypothesis, the protective effect of RPM on hospitalization for HF was greater than on hospitalizations for any cause. The reduction in hospitalization for HF (−7%) was similar to that found in the Clark et al. (35) meta-analysis (−5%) and was statistically significant in our meta-analysis due to the larger sample size. Unfortunately, only hospitalizations for HF could be meta-analyzed here, given the scarce information on cause of death in the selected articles. However, RPM also showed a significant protective effect on death from any cause in our review; usually elderly patients would have comorbidities, and some of these may have been exacerbated by their cardiac condition. Thus, better follow-up and care for the latter might have also increased their overall well-being and survival.
Particular care was taken to evaluate the robustness of our results in a series of sensitivity analyses. We did not observe major differences in risk reduction when analyzing separately follow-up of ≤6 and >6 months and high- or low-quality of the study. Similarly, risk reduction was consistent for all outcomes when evaluating separately the telephone and technology-assisted approach, despite these 2 approaches being inherently different. This is in line with the results of the only 2 available articles that directly compared these 2 modalities (3,5). Similar results were reported by Gonseth et al. (36) and Roccaforte et al. (38), whereas others have reported that regularly scheduled structured telephone contact, referral to a family physician (36–38), and monitoring of symptoms alone (35) were less effective in reducing risk than more comprehensive monitoring approaches (Table 6).
The heterogeneity of RCTs was small in this review and lower than that reported by Roccaforte et al. (38) (40% to 50%), but comparable to that reported by Clark et al. (35). One possible explanation for this discrepancy may reside in the fact that Clark et al. (35) considered older studies, a more general intervention pattern, and possibly the fact that the usual care arm included a large variety of situations. Much greater heterogeneity was observed for cohort studies (50%), although this was less than the heterogeneity reported by Gonseth et al. (36) (60% to 85%); this might be due to both the lack of control of confounders and the selection bias in the enrollment process of either arms.
The results of this meta-analysis support the benefit of RPM on mortality and hospitalization rates. This benefit was present in both RCTs and cohort studies. This analysis provides further support for the recent recommendation by European and American scientific societies. Mid- and long-term cost-effectiveness of remote patient monitoring, however, remains to be evaluated.
For a table showing the study excluded and reason for exclusion as well as funnel plots on cohort and RCT studies, please see the online version of this article.
Dr. Klersy is a consultant to Boston Scientific and Medtronic. Dr. De Silvestri is a consultant to Boston Scientific. Dr. Gabutti is a consultant to Boston Scientific. Dr. Auricchio is a consultant to Philips, Medtronic, Biotronik GmbH, and Sorin and has received speaker fees from GE Healthcare, Philips, Medtronic, Biotronik GmbH, and Sorin.
- Abbreviations and Acronyms
- confidence interval
- heart failure
- randomized controlled trial
- remote patient monitoring
- relative risk
- Received May 25, 2009.
- Revision received August 31, 2009.
- Accepted August 31, 2009.
- American College of Cardiology Foundation
- Dickstein K.,
- Cohen-Solal A.,
- Filippatos G.,
- et al.
- Hunt S.A.,
- Abraham W.T.,
- Chin M.H.,
- et al.
- Cleland J.G.,
- Louis A.A.,
- Rigby A.S.,
- Janssens U.,
- Balk A.H.,
- TEN-HMS Investigators
- Galbreath A.D.,
- Krasuski R.A.,
- Smith B.,
- et al.
- Blue L.,
- Lang E.,
- McMurray J.J.,
- et al.
- GESICA Investigators
- Kasper E.K.,
- Gerstenblith G.,
- Hefter G.,
- et al.
- Krumholz H.M.,
- Amatruda J.,
- Smith G.L.,
- et al.
- Bourge R.C.,
- Abraham W.T.,
- Adamson P.B.,
- et al.
- Scalvini S.,
- Zanelli E.,
- Paletta L.,
- et al.
- Scalvini S.,
- Zanelli E.,
- Volterrani M.,
- et al.
- Adamson P.B.,
- Magalski A.,
- Braunschweig F.,
- et al.
- Scalvini S.,
- Capomolla S.,
- Zanelli E.,
- et al.
- Clark R.A.,
- Inglis S.C.,
- McAlister F.A.,
- Cleland J.G.,
- Stewart S.
- Gonseth J.,
- Guallar-Castillon P.,
- Banegas J.R.,
- Rodriguez-Artalejo F.
- McAlister F.A.,
- Stewart S.,
- Ferrua S.,
- McMurray J.J.