Observational cohort study of patients with HF referred to the HF unit at our hospital for the first time between January 1, 2011 and December 31, 2017. Data were obtained from the SiMon clinical record management system, a purpose-built application designed by the hospital's computer department that includes a tabulated database containing prospectively collected data. Patients provided informed consent permitting us to use their clinical data for research purposes.

The study setting was the HF unit at the Cardiology Department of Complejo Hospitalario Universitario A Coruña in Galicia, Spain. This unit is the referral unit for HF patients in the health care district of A Coruña, which has approximately 550 000 inhabitants. It also receives patients with refractory HF from other hospitals in the autonomous community of Galicia (≈2 700 000 inhabitants) for assessment of suitability for advanced therapies (heart transplant and ventricular assist devices). The unit was awarded double SEC-Excellence certification—as a specialized and an advanced HF unit—in 2017. Our hospital was also awarded SEC certification for excellence in ventricular assist device implantation in 2018.

The core HF unit team is formed by 5 cardiologists and 2 nurses. It offers the full spectrum of services for HF patients, including health education, treatment optimization, outpatient telemonitoring, ward care for patients admitted for decompensation, assessment of suitability for advanced therapies, and clinical follow-up after heart transplant or ventricular assist device implantation. To cover this array of services, the HF unit forms part of a broad multidisciplinary team comprising other cardiology units (imaging, hemodynamics, electrophysiology, intermediate care, family heart disease, congenital heart disease) and the hospital's heart surgery and intensive medicine departments. It also works closely with other departments and units (internal medicine, emergency medicine, home hospitalization, physiotherapy, social work, and mental health) and primary care centers.

The main outcome variable was all-cause mortality, which is the main survival indicator used in the SEC-Excellence program.5 The secondary outcome variable was the cumulative incidence of hospitalization for HF. We also studied changes in LVEF and New York Heart Association (NYHA) functional class from the start to the end of active clinical follow-up by the HF unit.

For the primary and secondary outcome variables, patients were followed up from baseline (first visit to the HF unit) to death or heart transplant, regardless of follow-up status. Data for patients who did not die or receive a heart transplant were censored on December 31, 2018. Survival data for patients discharged by the HF unit were obtained from the electronic medical record database held by the Galician Health System.

Predicted mortality at 1 and 3 years was calculated for all patients using the MAGGIC (Meta-Analysis Global Group in Chronic Heart Failure) risk score.6 This model predicts the probability of death using 13 clinical predictor variables derived from individual patient data for 39 372 patients from 30 studies, including 6 clinical trials. The patients represent a broad spectrum of HF patients, aiding the generalizability of the model.

The MAGGIC risk score has been externally validated in various populations of HF patients7–9 and has been found to have adequate calibration and discrimination and superior performance in this respect to other risk prediction models.10 The prognostic value of the MAGGIC risk score adds to that of natriuretic peptide levels.8,9

Quantitative variables are presented as mean±SD or median (interquartile range [IQR]) depending on the normality of distribution. Categorical variables are reported as proportions. The Kaplan-Meier method was used to calculate the cumulative probability of death and hospitalization due to HF throughout follow-up; it was also used to build the corresponding survival curves.

Predicted 1- and 3-year mortality was calculated using the MAGGIC risk score. The discrimination power of the model in our population was analyzed using the c-statistic.

The difference between observed and predicted mortality was expressed as the observed/predicted ratio (O/P). The chi-square test was used to measure the statistical significance of this difference for the whole cohort, the quartiles of predicted risk, and several clinically relevant subgroups, defined by year of inclusion, age, sex, healthcare district of origin, LVEF, previous hospitalization for HF, and the presence of coronary artery disease.

All the analyses were performed in SPPS 20. Statistical significance was set at P<.5.

Between January 1, 2011 and December 31, 2017, 1280 patients with HF were seen for the first time at our HF unit. Of these, 648 (53.4%) had been hospitalized for HF at least once; 890 (69.5%) belonged to the healthcare district covered by our hospital and 390 (30.5%) were from other areas of Galicia. The clinical characteristics of the population are summarized in table 1.

The patients were treated and monitored by the HF unit for a median of 621±863 days (range, 190-1053 days). In total, 427 patients (33.4%) were still under active follow-up on December 31, 2018. Of the 853 patients no longer under follow-up (66.6%), 598 (46.7%) had been discharged by the unit, 170 (13.3%) had died, and 85 (6.6%) had been referred for advanced therapy with a heart transplant or ventricular assist device.

Prescription rates for the different classes of drugs at the first and last visits and during follow-up are shown in figure 1. Overall prescription rates for the cohort were 93% for beta-blockers, 67% for angiotensin-converting enzyme (ACE) inhibitors, 22% for angiotensin II receptor blockers (ARBs), 73% for mineralocorticoid receptor antagonists, and 16% for sacubitril-valsartan. The respective rates for the 940 patients with an LVEF <40% were 98%, 71%, 22%, 84%, and 22%. The proportion of patients with an LVEF <40% on sacubitril-valsartan still being monitored by the unit on December 31, 2018 was 43.2% (n=338).

Figure 1.

Prescription rates by drug class at baseline, during follow-up, and at the end of follow-up for patients seen at the heart failure unit. A: Whole cohort. B: Patients with LVEF <40%. ACE, angiotensin-converting enzyme; ARBs, angiotensin II receptor blockers; LVEF, left ventricular ejection fraction; MRAs, mineralocorticoid receptor antagonists.

(0.18MB).

The percentage of patients with an LVEF <40% on the target dose of beta-blockers increased from 18% at baseline to 29.6% at the last follow-up visit (P<.001). During the same period, there was also a significant increase in the proportion of patients with an LVEF <40% on the target dose of ACE inhibitors/ARBs/sacubitril-valsartan (11.3% vs 24.6%, P<.001). Finally, the proportion of patients with an LVEF <40% prescribed 50% or more of the target dose increased from 50.9% to 58.1% for beta-blockers (P<.001) and from 38.2% to 47.1% for ACE inhibitors/ARBs/sacubitril-valsartan (P<.001).

When seen for the first time at the HF unit, 161 patients (12.6%) had a defibrillator, 26 (2%) had a defibrillator-cardiac resynchronization therapy (CRT) device, and 4 (0.3%) had a pacemaker-CRT device. During follow-up, an additional 117 defibrillators, 91 defibrillator-CRT devices, and 10 pacemaker-CRT devices were implanted.

At the last follow-up visit, 211 (26.3%) of the 800 patients with a baseline LVEF <35% had a defibrillator, 78 (9.8%) had a defibrillator-CRT device, and 10 (1.3%) had a pacemaker-CRT device.

In addition, 6 patients (0.5%) underwent percutaneous mitral valve repair, 6 (0.5%) were fitted with a permanent ventricular assist device, and 81 (6.3%) underwent heart transplant.

The distribution of NYHA class changed significantly from baseline to the last visit (P<.001), with an increase in patients in NYHA class I (31.1% vs 11.1) and a decrease in patients in NYHA class II (42.9% vs 54.1) and III (20.9% vs 30.7%) (figure 2A). An improvement of at least 1 NYHA class was observed in 484 patients (37.8%).

Figure 2.

A: New York Heart Association functional class at first and last visit to heart failure unit. B: LVEF at first and last visit. LVEF, left ventricular ejection fraction.

(0.11MB).

There was also a significant improvement in LVEF from the first to the last visit (34.3%±13.9% vs 39.2±14.5%, P<.001). Of the 940 patients with an LVEF <40% at baseline, 239 (25.4%) had an LVEF ≥40% at the last visit. Just 25 patients (7.4%) experienced a reduction from an LVEF ≥40% to an LVEF <40% between the first and last visit. The distribution of LVEF values at baseline and the end of follow-up is shown in figure 2B.

The median survival follow-up time was 1238 days (IQR, 588-1888 days). During this period, 283 patients died (22.1%); 113 of these were no longer under follow-up at the unit, but they were nonetheless included in the survival analysis.

The estimated probability of survival according to the Kaplan-Meier method calculations was 99.7% for 30 days, 93.8% for 1 year, 83.3% for 3 years, and 74.5% for 5 years (figure 3A).

Figure 3.

A: Cumulative probability of survival over long-term follow-up according to Kaplan-Meier method. B: Cumulative probability of hospitalization for HF over long-term follow-up according to Kaplan-Meier method. HF, heart failure.

(0.13MB).

Overall, 413 patients (32.3%) were hospitalized for HF during the survival follow-up period. The cumulative probability of hospitalization due to HF was 4.2% for 30 days, 16.0% for 1 year, 28.3% for 3 years, and 36.8% for 5 years (figure 3B).

The median MAGGIC risk score was 19 [range, 13-24]. The score categories for the quartiles of predicted risk were 0-13 for the first quartile, 14-18 for the second quartile, 19-23 for the third quartile, and ≥24 for the fourth quartile. The distribution of scores is shown in figure 1 of the supplementary data.

The discrimination of the MAGGIC risk score (c-statistic) was 0.71 (95%CI, 0.65-0.77; P<.001) for 1-year mortality and 0.76 (95%CI, 0.72-0.81; P<.001) for 3-year mortality. The receiver operating characteristic curves are shown in figure 2 and figure 3 of the supplementary data.

The comparative values for observed mortality and mortality predicted using the MAGGIC risk score for the full cohort and the 4 quartiles at 1 and 3 years are shown in figure 4A and figure 4B.

Figure 4.

Comparison of observed mortality and mortality predicted using Meta-analysis Global Group in Chronic Heart Failure (MAGGIC) risk score for the whole cohort and quartiles of risk. A: 1-year mortality. B: 3-year mortality.

(0.15MB).

Observed mortality at 1 year was lower than predicted mortality in all cases, with an O/P of 0.57 for the whole cohort, 0.40 for the first quartile, 0.52 for the second quartile, 0.57 for the third quartile, and 0.69 for the fourth quartile. All the differences were statistically significant (P<.05).

Observed 3-year mortality was also significantly lower (P <.05) than predicted mortality for the whole cohort (O/P=0.60) and for the first (O/P=0.54), second (O/P=0.43), and third (O/P=0.43) quartiles. The difference for the fourth quartile was nonsignificant, with an O/P of 0.88 (P=.169).

The results comparing observed and predicted mortality at 1 and 3 years are shown by subgroup (year of inclusion, sex, age, health care district of origin, previous hospitalization for HF, presence of coronary artery disease, and LVEF) in table 2.

Observed 1-year mortality was significantly lower than predicted mortality for all subgroups except patients with an LEVF ≥40% (O/P=0.79, P=.187).

Observed mortality at 3 years was significantly lower than predicted mortality for all subgroups except patients aged 70 years or older (O/P=0.86, P=.126) and patients with an LVEF ≥40% (O/P=0.95, P=.640).

This study had several limitations. First, because it is an observational study conducted using information collected from a database, there is a risk of information bias. Second, it is a single-center study conducted in a highly specific care setting and our findings should therefore be generalized with caution.

As already discussed in detail, there are also limitations attached to the methodology used to compare observed and predicted mortality. Although the discrepancies between observed and predicted mortality in our series appear to be mostly attributable to differences in clinical treatment, we cannot be certain that our results were not influenced by differences in clinical variables that were not analyzed and that therefore could not be controlled for.