Heart failure (HF) is a major health concern, with mortality ranging from 5 to 40% [1], corresponding with a fivefold increased risk of death, compared to the general population [2]. It is even estimated that HF patients have a worse life expectancy than the majority of cancer patients, with a median survival of approximately 2 to 3 years [3, 4]. More than 400,000 patients in the USA are being diagnosed with HF, annually [5]. Moreover, prevalence rates are progressively rising and are expected to increase with 46% from 2012 to 2030 [6, 7].

In addition, heart failure is the diagnosis with the highest readmission rates among all diseases [8,9,10,11], as it accounts for 1 to 2% of all hospital admissions [12, 13]. In elderly people, it is the major cause of hospitalization [8]. Most patients are hospitalized at least once a year after diagnosis (i.e., 68 to 78% of patients) [8, 14, 15], and more than one-fourth is at risk of being readmitted within 30 days after initial diagnosis [8, 15,16,17,18]. Comparatively to prevalence rates, the total number of hospitalizations is also expected to rise, by 50% in the near future [19, 20].

Hospitalization places a great burden on patients [21]. Patients may experience various limitations in their activities of daily living [22,23,24], which highly impact their quality of life and level of satisfaction [21, 25]. Moreover, aside from a reduced quality of life, patients who are hospitalized have a significantly higher risk of death than non-hospitalized patients [26, 27]. Additionally, hospitalization due to HF places a great burden on the healthcare system, as it accounts for more than half of total healthcare costs [28, 29] corresponding with more than > 15 billion dollars a year for the American healthcare system [24, 30, 31]. HF is the most costly condition in western countries and since long time hospitalization for HF even exceeds the hospitalization costs for both cancer and myocardial infarction combined [32, 33]. Accordingly, hospitalization is judged as a highly important outcome measure in (inter)national literature and registries [34, 35].

Nevertheless, despite the rising prevalence rates, it seems that up to 40% of hospitalizations could be classified as preventable [36,37,38,39,40]. Therefore, the reduction of hospitalizations is the most promising factor as target to improve patients’ reported experiences or outcomes and to reduce the costs of HF management [25, 41, 42]. The combined measure of patient outcomes and costs are the main goal in value based healthcare, a well-known and promising strategy in healthcare in order to improve patient value [43,44,45].

Multiple previous studies examined the effect of various interventions to reduce (re)hospitalization in HF, mostly in patients with an left ventricular ejection fraction (LVEF) < 40% (i.e., patients with HFrEF) [46], but contrasting findings are found within the literature regarding the effectiveness of these interventions in reducing hospital admissions [47, 48]. Moreover, there is some considerable heterogeneity in strategies and methods used in previous studies [49]. Some studies, for example, focused on remote monitoring to prevent readmissions, while others examined quality improvement of interventions or transitional care systems [36, 37, 50,51,52]. Therefore, there remains a gap in information concerning which interventions could effectively contribute to effective prevention of HF hospitalization or readmission [47, 48, 53, 54].

Hence, even though multiple interventions have been included in the guidelines for treatment of HF [46, 55], there is a compelling need of a comprehensive overview of which types of interventions prove effective specifically in reducing HF hospitalizations, especially in HFrEF patients. This umbrella review therefore aims to systematically review all published meta-analyses conducted in the past 10 years that examined the incremental effect of different interventions in addition to standard care, to reduce (re)hospitalization in HFrEF patients, in order to highlight different levels of evidence regarding their effectiveness.


The systematic review protocol of this review was registered, in accordance with the PRISMA guidelines, at the International Prospective Register of Systematic Reviews (PROSPERO) on July 6, 2020 (registration number: 247872).

Search strategy

An electronic literature search was performed in PubMed, Web of Science, PsycInfo, Cochrane Reviews, CINAHL, and Medline to identify eligible studies published in the English language from January 2010 up to the end of June 2020. Search terms were developed using MeSH terms. Key words were related to (1) interventions, (2) heart failure, (3) hospitalization, and (4) meta-analysis (Table 1).

Table 1 Search strategy for each database

Ample differences existed in the classification of categories of interventions depicted in the existing literature. For example, previous reviews classified interventions in either educational interventions, pharmacological interventions, telemonitoring (TM), structured telephone support (STS), nurse home visits, nurse care management, and disease management clinics [41]; or discharge planning protocols, comprehensive geriatric assessments, discharge support arrangements, and educational interventions [56]; or case management interventions, clinical interventions, and multidisciplinary interventions [53]; or predischarge interventions, postdischarge interventions, and interventions bridging the transition [57]. A list of 4 categories of interventions was derived following a scoping review that combine the most common interventions aimed at reducing hospital (re)admissions, cardiac rehabilitation, care pathways, medication, and invasive treatment. Both general terms linked to the concept of interventions (e.g., programs, inventions, therapy) and terms for specific examples of (categories of) interventions were included in the search strategy.

Eligibility criteria

Search results of all databases were combined, and duplicates were removed. Titles and abstracts were screened against the following inclusion criteria: (1) a meta-analysis was conducted, on (2) randomized controlled trials (RCTs), (3) that examined the effectiveness of (3.a) cardiac rehabilitation, or (3.b) care pathways, or (3.c) medication, or (3.d) invasive therapy, (4) in patients with an established diagnosis of chronic heart failure, (5) with an LVEF < 40, (6) with a primary or secondary objective to evaluate the effect on reduction of (7) HF-related hospitalization or readmissions, (8) as compared to usual care, (9) conducted in the past 10 years, (10) followed patients for at least three months, and (11) were reported in English. Meta-analyses that included both RCTs and observational or cohort studies were not excluded. Yet only the included RCTs (and corresponding meta-analyzed effect sized) were extracted and used for our analyses. Only meta-analyses that reported at least one meta-analyzed effect estimate for HF-related admissions were included. In order to assure objective assessment, the title and abstract screening were independently conducted by two researchers (FH, TG). In case of disagreement between reviewers, points of disagreement were discussed in order to reach consensus. For full-text screening, inter-rate reliability was calculated using Cohen’s kappa.

Studies were excluded when the patient population was not primarily diagnosed with heart failure (e.g., patients with diabetes and comorbid heart failure). Additionally, if studies examined HF patients in combination with other patient groups yet did not report data on the individual patient groups, the study was excluded, as we would otherwise be unable to make a distinction between the differences in patient groups. Furthermore, studies that only reported data on a combined endpoint (e.g., mortality in conjunction with HF-hospitalization) and meta-analyses that examined risk stratification, prognostic factors, or lifestyle advice in patients were excluded. Moreover, meta-analyses were also excluded when examining a specific subgroup of HF patients (e.g., patients with and LVAD) or when examining a broader category of patients that could possibly include HF patients (e.g., “older patients” in general).

Quality assessment

The “A MeaSurement Tool to Assess systematic Reviews 2” (AMSTAR 2) was used to assess the methodological quality of included meta-analyses [58]. AMSTAR 2 consists of 16 items, of which 10 items were retained from the original AMSTAR tool. Response options for the items were “yes,” “partial yes,” and “no,” with “yes” responses denoting a positive result. The overall score of this tool was converted to high quality, moderate quality, low quality, and critically low quality. High quality was achieved when studies possessed no or one non-critical weakness; moderate quality was achieved when studies had more than one non-critical weakness; low quality was achieved when studies had one critical flaw, with or without a non-critical weakness; and critically low quality was achieved when studies exhibited more than one critical flaw with or without non-critical weaknesses. Critical domains are depicted in Table 2 [58]. In order to assure objective assessment, the quality assessment was independently conducted by two researchers (GS, TG). In case of disagreement between reviewers, points of disagreement were discussed in order to reach consensus (RT).

Table 2 Critical domains of the AMSTAR 2

Data extraction

A standardized extraction form was used to extract data from the included studies. Sociodemographic data (e.g., age, sex), number of participants, left ventricular ejection fraction, type of intervention and control, follow-up period, effect size, and conclusion were extracted from either the individual RCT or the meta-analysis in which the RCT was included. Only the most recent meta-analysis was included when multiple articles were written by the same authors on the same dataset. Comparisons were made between the different categories of interventions in terms of effectiveness in reducing HF-related (re)hospitalization. Interventions were classified as having a significant effect on HF-related (re)hospitalization (as compared to usual care) based on their own reported RR statistics, findings, and conclusions.

Data synthesis

Interventions were first classified into the four predefined categories (i.e., cardiac rehabilitation, care pathways, medication, and invasive therapy) and subsequently divided into more detailed classes of interventions (e.g., TM and STS) to examine the exact effect of all unique interventions.

Primary analysis: meta-analyses

To synthesize the data, a best-evidence synthesis was used as primary analysis, in which meta-analyses were classified based on level of internal and external validity [59]. The levels of evidence regarding the significance or non-significance of a relationship between the intervention and HF-related hospitalization among studies were ranked according to the following statements: (1) strong evidence: consistent findings (> 75% of the studies reported consistent findings) in multiple high quality studies; (2) moderate evidence: consistent findings (> 75% of the studies reported consistent findings) in one high-quality study and two or more moderate quality studies or in three or more weak quality studies; (3) limited evidence: generally consistent findings (> 75% of the studies reported consistent findings) in a high quality study or in two or fewer moderate quality studies; (4) no evidence: no studies could be found; (5) conflicting evidence: conflicting findings.

Secondary analysis: extracted RCTs

It was expected that multiple meta-analyses would report identical RCTs, as it was previously found that the amount of redundancy and duplication among reviews is substantial [60, 61]. Therefore, the corrected covered area (CCA) was calculated, which is a measure of duplicates in meta-analyses divided by the frequency of duplicates, reduced by the number of original publications \((\text{Corrected}\;\mathrm{covered}\;\mathrm{area}=\frac{N-r}{r\;c-r})\) [62]. A CCA of 0–5% is considered as slight overlap, while 6–10%, 11–15%, > 15% are respectively regarded as moderate, high, and very high overlap. In order to prevent bias as a result of duplicated data, a secondary analysis was conducted to control for the effects of overlap. All unique RCTs were extracted from the meta-analyses. Individual risk ratios (RRs) and 95% CIs for each intervention were calculated using Review Manager V.5.4. or extracted from the meta-analyses. The I2-statistic was used to present the heterogeneity of intervention effect. When the I2-statistic was statistically significant, a random-effects model was used in analyses. The RR-statistics found in our own meta-analyses were compared to the reported effects in the published meta-analyses.


Search results

After removal of duplicate meta-analyses, 639 titles and abstracts were screened (see Fig. 1). A total of 202 full-text articles were assessed for eligibility, of which 44 were included in our analyses. Cohen’s kappa for full-text screening was 0.76, indicating substantial agreement [63]. Median year of publication of all included meta-analyses was 2018. The 44 included meta-analyses encompassed 348 RCTs of which 186 were unique RCTs regarding interventions to prevent HF hospitalization (Table 3). Of these 186 unique RCTs, 44 were classified as invasive therapy, 14 as cardiac rehabilitation, 60 as medication, and 67 as care pathways (Table 4). The CCA for cardiac revalidation was \(\frac{(19-14)}{\left(\left(14\times 2\right)-14\right)}= \frac{5}{14}=36\%\), the CCA for invasive therapy was \(\frac{(86-45)}{\left(\left(45\times 15\right)-45\right)}= \frac{41}{630}=7\%\), the CCA for medication was \(\frac{(100-60)}{\left(\left(64 \times 14\right)-60\right)}= \frac{40}{836}=5\%\), and the CCA for care pathways was \(\frac{(138-67)}{\left(\left(67 \times 15\right)- 67\right)}= \frac{73}{896}=8\%\). This indicates a moderate to very high overlap in included RCTs [62].

Fig. 1
figure 1

Flow diagram of study inclusion. RCT: randomized controlled trial

Table 3 Overlap between different meta-analyses in included RCTs
Table 4 Baseline characteristics of RCTs

Quality assessment

Overall, risk of bias was classified as relatively low (Table 5). Of the 44 meta-analyses, 11 scored critically low, 15 low, 1 moderate, and 17 high. Almost all meta-analyses registered their protocol before commencement of the review (item 2) and used appropriate meta-analytical methods (item 11). Reviews were mostly downgraded based on the lack of an adequate investigation of publication bias (item 15).

Table 5 AMSTAR 2 scores

Study characteristics

A total of 425,220 patients were included in the 44 meta-analyses and 186 RCTs (Table 4). RCTs included between 16 and 10,917 patients. The mean age of patients ranged from 33 to 96 years. Mean LVEF varied between 17 and 40%. Percentage of male patients ranged from 25 to 100%. Follow-up period ranged widely from 30 days to 10 years. Studies that tried to prevent hospital admissions with cardiac rehabilitation focused on either exercise only or multicomponent cardiac rehabilitation. Care pathways could be divided into either TM, STS, and self-management promotion programs or multidisciplinary clinics. Invasive therapy encompassed catheter ablation (CA), cardiac resynchronization therapy (CRT), mitral valve repair, or stem cell therapy. Medication subtypes were angiotensin-converting enzyme inhibitors (ACE), angiotensin II receptor blockers (ARBs), mineralocorticoid receptor antagonists (MRAs), beta-blockers, statins, anticoagulation, and a miscellaneous subcategory.

Effect of interventions

Primary analysis: meta-analyses

Meta-analytic results of the 44 included meta-analyses are demonstrated in Table 6 and Fig. 2. According to our best-evidence synthesis, strong evidence suggests that CA, CR, and TM could prevent heart failure hospitalization. Furthermore, moderate evidence was found for the effectiveness of RAAS inhibitors, and CRT in reducing HF-related hospitalizations, while only limited evidence suggests the beneficial effects of beta-blockers, statins, mitral valve therapy, and multidisciplinary clinics or self-management promotion programs. There is conflicting evidence regarding the effect of cell therapy on HF hospitalization, and no evidence was found that anticoagulation should reduce HF-related hospitalizations.

Table 6 Effectiveness of interventions
Fig. 2
figure 2

Effects of different interventions on HF-related hospitalization in meta-analyzed and single-study results. ACE, angiotensin-converting enzyme inhibitors; ARB, angiotensin II receptor blockers; MRA, mineralocorticoid receptor antagonists; CR, cardiac rehabilitation; CRT, cardiac resynchronization therapy; CA, catheter ablation; TM, telemonitoring; STS, structured telephone support

Secondary analysis: extracted RCTs

In order to prevent bias as a result of duplicated data, all unique RCTs (N = 186) were extracted in a secondary analysis from the meta-analyses and compared to the results from our primary analysis.

Cardiac rehabilitation

A total of 14 studies examined the effects of cardiac rehabilitation. Of these individual studies, 1 reported a significant effect. When examined in a meta-analysis, a significant positive effect of cardiac rehabilitation was found (RR: 0.66, 95% CI: 0.44 | 0.97) (Fig. 3). This is in accordance with the general findings reported by the studied meta-analyses. Upon visual inspection, the funnel plots suggest no publication bias (Fig. 7).

Fig. 3
figure 3

Forest plot of RR for HF-related hospitalization between cardiac rehabilitation and control. Random effects model

Invasive therapy

There were 5 studies examining the effect of CA. Of these studies, 2 studies reported a significant effect. A positive effect of CA on HF-related hospitalization was found in our meta-analyses (RR: 0.57, 95% CI: 0.46 | 0.72) (Fig. 4). This is consistent with the general findings reported by the studied meta-analyses.

Fig. 4
figure 4figure 4

(AD) Forest plots of RR for HF-related hospitalization between (A) catheter ablation, (B) cardiac resynchronization therapy, (C) mitral valve therapy, and (D) stem cell therapy, and control. Fixed effects model

A total of 23 studies examined CRT to prevent HF-related hospitalization. Of these, 8 studies found a positive effect. Our meta-analysis suggested a positive effect of CRT (RR: 0.85, 95% CI: 0.78 | 0.92). This is in line with the general findings reported by the studied meta-analyses.

Of the 4 studies that examined mitral valve repair, 3 reported an effective reduction in HF-related hospitalization. Our meta-analyses suggested a positive effect (RR: 0.74, 95% CI: 0.64 | 0.86), which is in agreement with the general findings reported by the studied meta-analyses.

Stem cell therapy was in 0 of the 13 studies related to reduced HF-related hospitalization, which is in line with our meta-analyzed result (RR: 0.71, 95% CI: 0.45 | 1.14) and the conflicting evidence suggested by the studied meta-analyses.

The funnel plots indicate no, or only minimal publication bias (Fig. 7).


ACE inhibitors (5/18 studies; RR: 0.64, 95% CI: 0.49 | 0.85), MRAs (4/9 studies; RR: 0.77, 95% CI: 0.71 | 0.83), ARBs (4/5 studies; RR: 0.77, 95% CI: 0.72 | 0.84), beta-blockers (8/16 studies; RR: 0.78, 95% CI: 0.74 | 0.83), and statins (2/9 studies; RR: 0.51, 95% CI: 0.36 | 0.72) showed a significant effect of reduced hospitalizations in our meta-analyses (Fig. 5). This is in line with the general findings reported by the studied meta-analyses.

Fig. 5
figure 5figure 5

(AF) Forest plots of RR for HF-related hospitalization between (A) angiotensin-converting enzyme inhibitors, (B) angiotensin II receptor blockers, (C) mineralocorticoid receptor antagonists, (D) beta-blockers, (E) statins, and (F) anticoagulation, and control. Fixed effects model

Anticoagulation (RR: 0.99, 95% CI: 0.91 | 1.08) was in none of the studies (0/3) able to reduce HF-related hospitalizations. This absence of an effect was also reported by the studied meta-analyses.

The asymmetry in the medication funnel plots suggests some publication bias towards significant effectiveness of medication in reducing HF-related hospitalizations (Fig. 7).

Care pathways

Multidisciplinary clinics or self-management promotion programs (10/23 studies; RR: 0.79, 95% CI: 0.73 | 0.85) and TM (12/33 studies; RR: 0.86, 95% CI: 0.81 | 0.92) were related to less HF-related hospitalizations (Fig. 6). This is in agreement with findings reported by the studied meta-analyses. STS (1/11 studies; RR: 0.85, 95% CI: 0.85 | 1.04) was not related to reductions in HF-related hospitalizations. This is in contrast to findings from the meta-analyses. Visual inspection of the funnel plots did not suggest publication bias (Fig. 7).

Fig. 6
figure 6

(AC) Forest plot of RR for HF-related hospitalization between (A) multidisciplinary clinics or self-management promotion programs, (B) structured telephone support, and (C) telemonitoring, and control. Fixed effects model

Fig. 7
figure 7figure 7figure 7

(AD) Funnel plots of the effects of (A) cardiac rehabilitation, (B) telemonitoring, (C) medication, and (D) invasive therapy


Heart failure is a major health concern, with the highest readmission rates among all diseases [8,9,10,11]. Yet, up to 40% of hospitalizations could be classified as preventable [36,37,38,39,40]. This umbrella review therefore aimed to systematically review all published meta-analyses conducted in the past 10 years that examined the incremental benefit of interventions in addition to standard care, in reducing HF-related (re)hospitalization, in order to provide a comprehensive overview of different levels of evidence with regard to the different interventions that aim to reduce HF-related (re)hospitalization.

Even though previous studies did examine the effectiveness of interventions in treatment for heart failure in general, this umbrella review highlights different levels of evidence regarding the effectiveness of several interventions in reducing HF-related hospitalization. All different categories of interventions (i.e., cardiac rehabilitation, invasive treatment, medication, and care pathways) entail interventions that prove able to statistically significantly reduce HF-related hospitalizations. Strong or at least moderate evidence was found for the beneficial effects of CA, CRT, ACE inhibitors, MRAs, ARBs, CR, TM, and STS. Limited evidence was found for the ability of beta-blockers,, statins, mitral valve repair, and multidisciplinary clinics or self-management promotion programs to reduce hospitalization rates. Conflicting or no evidence was found for the effects of anticoagulation and stem cell therapy.

The findings of this umbrella review were generally supported by the American Heart Association and European Society of Cardiology heart failure guidelines [46, 64]. Yet, evidence for effectiveness was still lacking for several interventions in these guidelines. A couple of interventions proposed in the guidelines had low levels of evidence, as they were only supported by a single randomized clinical trial. Although these guidelines do not solely focus on the prevention of (re)hospitalization, this umbrella review now provides additional evidence for the effectiveness of ARBs (e.g., Valsartan) and telemonitoring as effective in the prevention of (re)hospitalization in heart failure. Therefore, the results of this review may be used in addition in clinical practice, as well as by policymakers, as a guideline in deciding what treatment option might help prevent hospitalization in at risk heart failure patients.

Effectiveness of reported interventions was measured in terms of a reduced risk for heart failure related hospitalizations. However, it would be naïve to suggest that this equals the clinical, genuine effect of treatment. Non-effectiveness of treatment could also be related to non-adherence or non-acceptance of the intervention by the patient, since it is estimated that non-adherence ranges between 30 and 50% in patients with chronic illnesses [65]. And non-adherence not only holds for medication, yet also for cardiac rehabilitation [66, 67] and telemonitoring [68, 69]. It has been shown that worsening of HF is often related to non-adherence of patients [70] and is in fact associated with 10% of hospitalizations [65, 71] and a 10% increased risk of readmission [72]. The other way around, reductions in non-adherence are found to result in less hospital admissions [73].

Differences in non-adherence to different forms of interventions were also found. For example, patients are found to be more adherent to ACE-inhibitors (77.8%) as compared to beta-blockers (69.8%) [74]. These differences could be explained by cognitions of patients regarding the efficacy of the intervention and the usability of the intervention [75]. Moreover, low health literacy or simply a lack of knowledge about the syndrome could also contribute to non-adherence [76,77,78]. In clinical practice, one should therefore educate patients about the importance of disease management with medication, invasive therapy, cardiac rehabilitation, and care pathways [65, 79].

Moreover, when implementing interventions in practice, one should not only focus on effectiveness, yet also incorporate, for example, the costs of the intervention. Especially, since HF is the most costly condition in western countries, with at least twice the costs of the estimated consumption of healthcare in the general population in a year [32, 33, 80], mainly due to HF-related hospitalization [28, 29]. Research has shown that, in terms of cost-effectiveness, medication treatment with beta-blockers, ARBs, or ACE inhibitors could be preferred over more cost expensive therapies as device therapy with CRT [81, 82]. More specifically, with regard to specific forms of medication, ivabradine seems a cost-effective treatment option, while this does not hold for valsartan [82]. In addition, general HF treatment combined with telemonitoring has been found to be between 27 and 52% more cost-efficient than usual care alone [83, 84]. Furthermore, telemonitoring seems not only cost-efficient; but nowadays, with the pandemic consequences of COVID-19 it seems more desired than ever [85]. The pandemic served as a catalyst, as both healthcare professionals as patients wanted optimal care in a time of reduced ambulatory outpatient clinics, with being compliant to social distancing [84]. Our review shows, in addition, that, even though the terms are interchangeably used to both describe some form of “remote care,” telemonitoring and structured telephone support have different levels of effectiveness with regard to prevention of heart failure related (re)hospitalizations, which should be accounted for in clinical practice.

In this umbrella review, we only aimed to provide an overview of effective treatment options for prevention of heart failure (re)hospitalization. Consequently, no conclusions could be drawn regarding the hierarchy of effectiveness based upon this review. In future research, it should be examined what factors contribute to effectiveness of interventions, as our study only showed that particular interventions could reduce heart failure hospitalizations, but not why per se. Research should focus on the effective mechanisms of care pathway programs, for example, or on determinants of successful implementations of interventions for heart failure.

The aim of our review was to assess interventions which are currently used in clinical practice and examined in large populations. Our results are based upon meta-analyses performed within the past 10 years. Yet, most recent innovative treatment options are probably underrepresented. For example, no study examined the effects of SGLT-II inhibitors, while the European Society of Cardiology stated that SGLT-II inhibitors could be preferred in heart failure patients [86]. Future studies should examine whether the use of SGLT-II inhibitors could show effective in reducing hospitalization. Moreover, as the aim of our review was to assess interventions which are currently used in clinical practice and examined in large populations, we expected to find multiple meta-analyses examining the same interventions. A large amount of overlap in RCTs in included meta-analyses was found. This stresses the importance of registration of protocols and knowing whether the intended research subject has a significantly different research objective than existing, or outdated reviews [62].

To conclude, this umbrella review highlights different levels of evidence regarding the effectiveness of several interventions in reducing HF-related hospitalization in HFrEF patients. It provides an overview of all, known, meta-analyses conducted in the past 10 years that examined interventions to prevent heart failure related hospitalizations. All different categories of interventions entail interventions that prove able to statistically significantly reduce HF-related hospitalizations. Most evidence was found for the beneficial effects of angiotensin-converting enzyme inhibitors (ACE), mineralocorticoid receptor antagonists (MRAs), angiotensin II receptor blockers (ARBs), cardiac rehabilitation, and telemonitoring. The results of this review may be used in clinical practice, as well as by policymakers, to guide treatment for heart failure patients at risk of hospitalization.