Effectiveness and Cost-Effectiveness of Case Management in Advanced Heart Failure Patients Attended in Primary Care: A Systematic Review and Meta-Analysis

Aims: Nurse-led case management (CM) may improve quality of life (QoL) for advanced heart failure (HF) patients. No systematic review (SR), however, has summarized its effectiveness/cost-effectiveness. We aimed to evaluate the effect of such programs in primary care settings in advanced HF patients. We examined and summarized evidence on QoL, mortality, hospitalization, self-care, and cost-effectiveness. Methods and results: The MEDLINE, CINAHL, Embase, Clinical Trials, WHO, Registry of International Clinical Trials, and Central Cochrane were searched up to March 2022. The Consensus Health Economic Criteria instrument to assess risk-of-bias in economic evaluations, Cochrane risk-of-bias 2 for clinical trials, and an adaptation of Robins-I for quasi-experimental and cohort studies were employed. Results from nurse-led CM programs did not reduce mortality (RR 0.78, 95% CI 0.53 to 1.15; participants = 1345; studies = 6; I2 = 47%). They decreased HF hospitalizations (HR 0.79, 95% CI 0.68 to 0.91; participants = 1989; studies = 8; I2 = 0%) and all-cause ones (HR 0.73, 95% CI 0.60 to 0.89; participants = 1012; studies = 5; I2 = 36%). QoL improved in medium-term follow-up (SMD 0.18, 95% CI 0.05 to 0.32; participants = 1228; studies = 8; I2 = 28%), and self-care was not statistically significant improved (SMD 0.66, 95% CI −0.84 to 2.17; participants = 450; studies = 3; I2 = 97%). A wide variety of costs ranging from USD 4975 to EUR 27,538 was observed. The intervention was cost-effective at ≤EUR 60,000/QALY. Conclusions: Nurse-led CM reduces all-cause hospital admissions and HF hospitalizations but not all-cause mortality. QoL improved at medium-term follow-up. Such programs could be cost-effective in high-income countries.


Introduction
Heart failure (HF) occurs when blood flow is insufficient to meet tissue metabolic needs [1]. Advanced HF (stage D according to America Guidelines) is defined by the • Studies where the nurse-led CM model effect was measured. • Community interventions including those commencing in hospital.
Exclusion criteria: • Nurse-led CM interventions developed only in hospitals. • Cardiac rehabilitation programs, unless providing elements of nurse-led CM. • Community interventions from specialized HF clinics directed by cardiologists.

•
Only one educational session, without follow-up phone calls/patient interaction.

Type of Comparator/Control
Studies comparing the intervention with usual care or another nurse-led CM program within primary/community care.

Outcomes Primary Outcome
Nurse-led CM program effects on mortality in primary care settings on advanced HF patients.

Secondary Outcomes
Results regarding QoL, hospitalization, adherence to treatment, undesirable effects, costs, and cost-effectiveness.

•
QoL measured by EuroQol-5D, SF-8, SF-36, and the Kansas City Cardiomyopathy Questionnaire (KCCQ) scales, etc. • All-cause and HF mortality. • Number of HF hospitalizations or for any other cause during follow-up. • Self-care measured by the Appraisal of Self-care Agency (ASA) Scale, European Heart Failure Self-care Behavior Scale, and Self-care of Heart Failure Index.

•
Costs associated with health resources. Outcomes were measured by follow-up time when available (<6 months, 6-12 months, >12 months), age, and type of consultations (home visits/telemedicine).

Electronic Searches
Searches were performed with MEDLINE, CINAHL, Embase, Clinical Trials, WHO, Registry of International Clinical Trials, and Central Cochrane. The World Health Organization's International Clinical Trials Registry (ICTRP) platform (http://apps.who.int/ trialsearch/, accessed on 24 March 2022), and the ISRCT registry (https://www.isrctn.com/, accessed on 24 March 2022) were used. One hundred ongoing studies were identified by the USA ClinicalTrials.gov registry (https://ClinicalTrials.gov/, accessed on 24 March 2022), however, there were no partial results published and they were excluded.
The database was EndNote χ 2 software. Publications were included up to March 2022 (See Appendix A).

Other Resources
A manual inspection of the references in previous systematic reviews of HF patient nurse-led CM models was conducted. Gray literature was reviewed, and experts consulted.

Data Collection and Analysis 2.3.1. Study Selection
Initial screening of titles/abstracts was performed by a reviewer. A random sample of 20% of the retrieved references was evaluated by a second reviewer in order to guarantee the quality of the process.
Two reviewers then independently assessed eligibility of the 405 studies based on full-text reading. In case of discrepancy, there were consensus sessions. The Rayyan program [14] was employed throughout.
A PRISMA flow chart depicts the study selection ( Figure 1). For excluded studies at the full-text level, see Supplementary Material Table S1.

Data Extraction/Management
Data included author/s, publication year, design, sample and intervention characteristics, and outcomes. Funding information was from economic studies. Data extraction was performed in duplicate.

Risk-of-Bias Assessment
The Cochrane risk-of-bias 2 tool (RoB2) [15] was employed for clinical trials and the Risk-of-Bias in Non-randomized Studies of Interventions (Robins-I) [16] for cohort studies.
The Robins-I was adapted for quasi-experimental studies (Supplementary Material Table  S2). Disease decline over time was assumed as a confounder and penalized if the participants were overstable/decompensated during follow-up. In the risk of bias due to intervention classification, intervention/inclusion criteria were either clearly predefined or not.
The Consensus Health Economic Criteria (CHEC) list was employed for economic evaluations [17].
All studies were peer-reviewed, and in the case of any discrepancies, consensus was reached. In studies that measured more than one outcome, these were assessed separately in the outcome domains of the tools.

Intervention Characteristics
Interventions were classified as basic or intensive and determined according to staff availability and issue management during follow-up. Table 1 depicts the intervention and  characteristics. Subgroup analyses were carried out by age ranges (>85 years, 65-85 years, <65 years) and time to follow-up (<6 months, 6-12 months, and >12 months). Population characteristics and outcome measurement instruments were evaluated. RCT: randomized control trial. HF: heart failure. NYHA: New York Heart Association. * This classification was based on the number of contacts made with the patients, staff availability, and the extent to which they addressed the issues during follow-up. The main component of each intervention was also described: a -clinical consultations, b -home visits, c -remote vital sign monitoring, d -videophone, e -messaging, f -scheduled telephone calls, g -telephone availability of staff (unscheduled). ∞ There is no exact information about the "no program" group but it seems to be similar to the uncompleted follow-up. π Case management vs. control not reported. β Population with advanced chronic diseases. Advanced HF disease not reported.

Data Synthesis and Registry
The analysis was an intention-to-treat approach, and all participants were included to reduce the potential selection bias.
Outcome data were evaluated at <6 months, 6-12 months, and >12 months follow-up when available.
Mortality/hospitalization were meta-analyzed using Review Manager (RevMan, version 5.3.5., Cochrane Collaboration, Oxford, UK) and STATA software (v.14.0, STATA Corp, College Station, TX, USA). Pooled relative risk ratios (RRs) and standard mean differences (SMDs) for binary and continuous outcomes were evaluated with the random effect model approach. When means and standard deviations (SDs), or changes of means and SDs from baseline were not reported, they were calculated using standard errors (SE), confidence intervals (CI), or the correlation coefficient.
For all meta-analyses with at least 10 included studies, the publication bias was assessed by a visual inspection of Begg's funnel plot and statistically, using Egger's test for small study effects (funnel plot asymmetry).
Systematic Reviews of Economic Evaluations guides were followed to analyzed costs and the cost-effectiveness of primary studies [42].
The study was registered in the International Prospective Register of Systematic Reviews (PROSPERO) and published with ID number CRD42020160810.

Selection of Primary Studies
From 5944 records from four databases, 2129 studies remained. We then reviewed 405 full-text articles: 14 were selected for synthesis. We also reviewed all primary studies obtained from 55 systematic reviews identified in the title/abstract screening and selected 16. In total, 30 studies were included in the evidence synthesis. Of these 30, 25 described the benefits/risks of a nurse-led CM model, and 5 were economic evaluations (Figure 1).

Evidence of Effects (Benefits and Risks)
The 25 included studies were published between 1997 and 2016 ( Table 1). The majority (17) were from the US and European countries. Most were randomized controlled trials, except for five quasi-experimental studies and one prospective cohort.
Populations were mainly men >60 years. The identification was performed primarily at the hospital or community level. In 22 of the 25 included studies, the comparator was usual care. Most studies had a follow-up of more than six months, with a maximum of one year.
Twelve intensive and thirteen basic programs compared their effect with usual care. The intensity classification was based on the number of contacts made with the patients, staff availability, and to what extent they addressed the issues in the follow-up visits. Telemedicine and home-visit interventions were mainly intensive programs whereas others (clinical consultations, phone calls) were basic.

Cost-Effectiveness Studies
Five economic evaluations were identified. Three were cost-effectiveness studies, and two cost-benefit analyses. All had been performed in high-income countries, four in European ones and another in the US (Table 2).

Randomized Control Trials
The RoB2 Cochrane [15] tool was used to evaluate the risk of bias (Supplementary  Material Table S3). Most studies presented issues with random sequence generation; however, baseline group characteristics did not suggest a randomization concern.
To evaluate the risk of bias of the reported results, all study original protocols were examined to compare the planned statistical analysis with the final result. In nine studies, the protocol was missing and thus referred to as a lack of information with some concerns. Most studies with protocol (8 out of 10) were assessed as a low risk of bias. Hospitalization/mortality was considered a low risk of bias. QoL and self-care outcomes presented some concerns as the interventions were not blinded, and the questionnaires were generally self-reported (Supplementary Material Table S3).

Nonrandomized Trials
The risk of bias in the quasi-experimental and cohort studies was assessed with the ROBINS-I tool (16) (Supplementary Material Table S4). Only one cohort study was identified (Schellinger 2011 [35]). Studies were classified as having a high risk of bias when patients were too stable/decompensated during follow-up. If no control group was present, the progression of advanced HF was considered to play a role in the intervention effectiveness.
Since nonrandomized control trials had lower levels of evidence than randomized ones, they were not included in the meta-analysis, although descriptively reported.

Economic Evaluations
The CHEC tool [17] was used to assess the economic evaluations ( Table 2). All studies had a clear research question, with a well-defined population. The economic evaluations were considered as social ones, since they included costs related to patient care beyond hospital admissions. The quality of such studies was therefore downgraded.
Intervention cost-effectiveness was taken to be >6 months although three of the five studies had a shorter follow-up time. Only one study declared its source of funding (Sahlen et al. [47]).
Overall, three economic evaluations had a moderate/low risk of bias, and two were high risk.
The follow-up was 12 months in three studies and 6 months in the others. To avoid one death, 32.15 patients were required (Figure 2).
The forest plot did not suggest a marked heterogeneity, nor did the subgroup analysis by length of follow-up indicate differences amongst subgroups (p = 0.34). There were, however, some differences in the type of CM (p = 0.07). Telemedicine was more effective than home visits (RR 0.47, 95% CI 0.27 to 0.83; participants = 2686 in two studies I 2 0%, low risk of bias) with 6 and 12 months of follow-up (Goldberg 2003 [28] and Lynga 2012 [13], respectively) (Supplementary Material Figure S1).

Mortality for Heart Failure
None of the studies reported deaths due to HF.

Hospitalizations for Heart Failure
Eight studies described HF hospitalizations and results showed CM as effective in avoiding them (HR 0.79, 95% CI 0.68 to 0.91; participants = 1989; studies = 8; I 2 = 0%, low risk of bias).
Five studies lacked full information, and two were not randomized controlled trials, and thus excluded. Nevertheless, the results showed that CM was beneficial for advanced HF (Supplementary Material Table S5).

All-Cause Hospitalizations
Five studies reported hospitalizations for all causes and demonstrated CM as protective for this outcome (HR 0.73, 95% CI 0.60 to 0.89; participants = 1012; studies = 5; I 2 = 36%, low risk of bias).
The subgroup analysis by age, time to follow-up, and CM type did not suggest any differences among groups (p = NA, p = 0.05 and p = 0.40, respectively) (Supplementary Material Figure S1).
Seven studies were excluded since four were not randomized controlled trials and three lacked information. Nevertheless, results indicated the benefits of nurse-led CM (Supplementary Material Table S5).
Subgroup differences in follow-up showed that the beneficial effect started at 6 months but was lost at 12 months (p = 0.02). In addition, testing for subgroup differences in the type of nurse-led CM suggested an improvement in home visits rather than telemedicine or other means (p = 0.02) (Supplementary Material Figure S1).
3.5. Costs and Cost-Effectiveness of Nurse-Led CM 3.5.1. Cost of the Intervention Except for one study (Sahlen et al. [47]), all reported that investment in a new intervention was greater than in usual care. Cost varied according to intensity, year of implementation, and country. Related costs were mainly linked to healthcare professionals and telemedicine devices in those studies proposing remote data transfer ( Table 2).

Cost-Effectiveness (Cost per QALY)
Three studies reported results in the incremental cost-effectiveness ratio (ICER) per QALY of the intervention compared with usual care. They also presented the lowest risk of bias and the largest time horizon (Postmus et al., ). They reported that intensive interventions, compared with basic ones/usual care, obtained larger benefits in terms of QALY and LY in NYHA III/IV patients. Regarding the ICER per QALY, ICERs of EUR 59,289 and EUR 14,027, respectively, were observed when comparing intensive versus usual care. Figures were below EUR 60,000/QALY.

Cost-Benefit Studies
Studies reported savings due to fewer hospital admissions. The net benefit was mainly determined by the price of the intervention.

Discussion
This systematic review summarized the quality of evidence regarding the effectiveness/ cost-effectiveness of nurse-led CM programs in advanced HF populations. We included 30 studies, 25 reported effectiveness (19 randomized controlled trials, 5 quasi-experimental, and 1 cohort), and 5 economic evaluations. Only meta-analyzed studies with a low risk of bias, or with the lowest risk of bias available, were analyzed. Nonrandomized trials or studies lacking data were excluded. The latter were presented as narrative results and showed the same direction of effectiveness.
Our results were nonsignificant to indicate that nurse-led CM intervention reduced all-cause mortality. Interventions with telemedicine were the most effective. No study reported mortality due to HF.
Regarding HF hospitalizations, we found eight low-risk-of-bias randomized controlled trials. Five additional studies were narratively summarized. Nurse-led CM, telemedicine, and home visits were effective in preventing HF and all-cause hospitalizations.
Eight studies reported an improvement in QoL at 6 months which did not extend at 12 months. Three studies suggested that patients in the program had better self-care, although this was not statistically significant and the intervention costs among studies ranged from USD 4975 (2003-year value) to EUR 27,538 (2015-year value).
The most recent similar review (Takeda) explored different CM interventions for allstage HF patients. Whilst our results concurred, they were not statistically significant, in contrast to other authors [11,48,49]; the fact that our population was at the final stage of the disease may have played a role.
In a similar manner to Bashi et al. [50], we found that patients with lower mortality were those who received telemedicine. Such results are, however, controversial as Flodgren et al. described no differences with usual care [51]. Reasons for this may include sociocultural differences, and in this sense, further research is required.
We observed that the nurse-led CM interventions reduced the risk of hospitalizations. This is a relevant finding since hospitalizations for advanced HF are common [52] and avoiding hospitalization can also reduce mortality [53].
In agreement with Rice et al. [54], QoL improved with the intervention. We found, however, that in our population this was at 6 months after the intervention and only lasted up to 12 months. Nevertheless, due to the advanced stage of the disease, we believe any gain, or even maintenance, in the QoL of patients with advanced HF to be relevant.
QoL did not improve in the telemedicine group, concurring with Bauce et al. [55]. Personal contact with healthcare professionals can produce a certain emotional proximity which may have a positive impact on QoL and should be further evaluated.
Nurse-led CM interventions could also improve self-care [56]. Nevertheless, our findings were not statistically significant, and there was considerable heterogeneity among studies. The intervention effect was lost with time as patients lost motivation. Factors favoring long-term self-care should be further explored as they have an impact on the reduction of hospital admissions [57].
Nurse-led CM could be cost-effective, a finding that concurs with Rice et al. [54], probably due to the savings from fewer hospitalizations. In terms of QALYs, Fergenbaum et al. concluded that a home-based intervention improved the QALY by 0.11 and reduced costs [58]. In our review, all the studies that reported QALYs described improvements above the figure described by Fergenbaum except for the Postmus study.
Advanced HF patients require more resources to improve their QALY thus increasing incremental cost. Nurse-led CM was not found to be particularly cost-effective, nevertheless, a threshold of EUR 60,000/QALY may be considered affordable for high-income countries.
Further studies should consider differentiating advanced HF from the general HF population, since this subgroup has different needs.

•
The limitations of this systematic review are mostly derived from those of the primary included studies. We found seven, five, and four studies corresponding to all-cause hospitalizations, HF hospitalizations, and QoL, respectively, with concerns of a high risk of bias leading to their exclusion from the pooled analysis. We did, however, narratively summarize these data and found similar results in most cases. • Nurse-led CM interventions may have varying characteristics according to their settings which could result in heterogeneity. For clarification, we created a descriptive table with all the characteristics of each intervention.

•
The CM overall effect can be affected over time. We observed a short-term beneficial effect that was depleted on the medium/long term. We therefore carried out the meta-analysis with different follow-up time groups to analyze this factor.

Conclusions
Nurse-led CM can reduce all-cause hospital admissions and HF hospitalizations but not all-cause mortality. QoL improved in medium-term follow-up, and better selfcare/survival was reported, although it was not statistically significant. The intervention could be cost-effective for less than EUR 60,000/QALY. More intensive nurse-led case management studies are needed to determine the cost-effectiveness of the program.
Supplementary Materials: The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/ijerph192113823/s1, Figure S1: Meta-analysis of subgroups according follow-up time, type of nurse-led case management delivered and age Table S1: Excluded articles at full text level; Table S2: ROBINS-I adaptation; Table S3: Assessment of Risk of bias (RoB) in Randomized controlled trial studies; Table S4: Assessment of Risk of bias with adaptation ROBINS-I tool in non-Randomized controlled trial studies; Table S5: Descriptive tables of studies with incomplete outcome data.