Predicting 1‐year heart failure hospitalization and mortality post‐discharge from the intensive cardiac care unit

Abstract

Aims

Despite the high risk of rehospitalization for heart failure (HF) and death among patients admitted to the intensive cardiac care unit (ICCU), no accurate prediction score for these outcomes exists. We aimed to develop a risk score to predict unplanned HF hospitalization and death 1‐year post‐discharge in an unselected cohort of patients admitted to the ICCU.

Methods

Based on a national, multicentre study, we included all consecutive patients admitted to the ICCUs in 39 French centres from 7 to 22 April 2021. We randomly selected a training cohort of 21 centres (n = 1008) to develop the ICCU‐HF score and a validation cohort of eight other centres (n = 463). The primary composite outcome was unplanned hospitalization for HF and cardiovascular death at 1‐year follow‐up after discharge. Using the score, patients were stratified into three risk groups to evaluate the prognostic value.

Results

Using a least absolute shrinkage and selection operator (LASSO) regression approach, we identified seven predictors: left ventricular ejection fraction, significant valvular disease grade 2+, Killip score >1, NT‐proBNP, creatinine level, previous ventricular arrhythmia and use of inotropes during hospitalization. In 1471 patients (63 ± 15 years, 70% men), 99 (6.7%) experienced the primary outcome. The ICCU‐HF score outperformed NT‐proBNP, the strongest individual predictor (area under the curve [AUC] 0.77, 95% CI [0.71–0.83] vs. AUC 0.72, 95% CI [0.66–0.79], P = 0.008), demonstrating excellent performance with an AUC of 0.83 (95% CI: 0.77–0.89) to predict outcomes in the validation cohort. Compared with the low‐risk group, the intermediate‐risk and high‐risk groups had significantly higher risks of the composite outcome (HR 4.09, 95% CI [2.23–7.50], P < 0.001 and 12.69, 95% CI [7.02–22.95], P < 0.001), proving strong risk stratification capability of the ICCU‐HF score.

Conclusions

The ICCU‐HF score showed good performance in predicting the 1‐year risk of unplanned HF hospitalization and death in a large cohort of unselected patients admitted to the ICCU, with excellent results in the validation cohort. This score effectively stratifies patients into risk groups, enhancing its utility in clinical decision‐making.

Introduction

Acute heart failure (HF) is a major cause of hospitalization in the year following discharge from the intensive cardiac care unit (ICCU)¹ and remains the main cause of hospitalization for patients aged >65 years, associated with 1-year mortality ranging from 25% to 30% and high rates for HF readmission.² It is, therefore, crucial to offer a high-performance prognostic stratification tool to identify these patients at risk of HF outcomes after discharge from ICCU, allowing for personalized management. The existing prognostic scores for predicting the risk of hospitalization for HF or death on discharge from ICCU were built using cohort with only acute coronary syndrome (ACS)^3–5 or acute HF patients.^6–8 However, ICCU patients are often afflicted by coexistence of several cardiovascular diseases.^1,9 Hence, it would be relevant to propose a global prognostic stratification tool including all patients admitted to the ICCU.

Although the post-discharge prognosis for patients hospitalized for ACS and acute HF has improved during the recent decades,^10–12 the morbidity, mortality and readmission rate for acute HF remain high.^13,14 Hence, development of an applicable post-discharge risk prediction score in a population of unselected cardiovascular patients hospitalized in the ICCU is needed. Such a score should include easy-accessible and reproducible information on comorbidities, clinical presentation, echocardiographic measurements and laboratory results collected during the hospitalization in the ICCU.

Therefore, this study aimed to investigate a large cohort of unselected consecutive patients hospitalized in the ICCU to assess the feasibility and accuracy of a prediction score—the ICCU-HF score—to predict the occurrence of cardiovascular death and unplanned HF hospitalization within 1 year of hospital discharge. Furthermore, we compared its performance to existing models and scores. Finally, we validated its performance using a validation cohort.

Methods

Study population

This study was based on the ADDICT-ICCU study; a multicentre, prospective, observational study including all consecutive patients admitted to the ICCUs in 39 French centres from 7 to 22 April 2021. All administrative regions of France were represented, and a list of total inclusion sites can be found in Table S1. The details of the ADDICT-ICCU study design have previously been described.¹⁵ The main exclusion criterion was in-hospital death during the index hospitalization. We randomly selected a training cohort of 21 centres (n = 1008) to develop the ICCU-HF score and a validation cohort of eight other centres (n = 463) for its validation.

The flowchart of the study is depicted in Figure 1. The methodology of baseline characteristics collection including echocardiographic measurements is detailed in Method S1. The admission diagnosis was adjudicated by two independent experts at the end of the hospitalization based on current guidelines (Method S2). The treatment of each patient was at the discretion of the treating physicians following the current European Society of Cardiology guidelines.¹⁶ The study conforms with the principles outlined in the Declaration of Helsinki, was approved by the Committee for the Protection of Human Subjects, Ile de France-7 (APHP190870), and was registered at ClinicalTrials.gov (NCT05063097). All patients provided written informed consent for participation.

[IMAGE OMITTED. SEE PDF]

Outcomes

The follow-up consisted of a clinical visit as part of usual care or by direct contact with the patient and the referring cardiologist. Data collection ended in June 2022. The primary composite outcome was defined as cardiovascular death or hospitalization for acute HF at 1-year follow-up. According to the European guidelines,¹⁶ hospitalization for acute HF was defined as symptoms and/or signs of HF with evidence of systolic and/or diastolic left ventricular and/or right ventricular dysfunction by echocardiography and elevated levels of natriuretic peptide (brain natriuretic peptide >35 pg/mL and/or N-terminal pro-brain natriuretic peptide [NT-proBNP] > 125 pg/mL). Cardiovascular mortality is defined as sudden cardiac death with documented fatal arrhythmias, or any death immediately preceded by acute MI, acute or exacerbation of heart failure or stroke.¹⁷ Notably, all events from hospital discharge reports were analysed by two independent adjudication experts blinded to additional patient information.

Statistical analysis

Continuous variables were expressed as mean ± standard deviation for normally distributed data or median with interquartile range (IQR) for non-normally distributed data. Categorical variables were presented as counts and percentages. Group comparisons were performed with Student's t-test, Mann–Whitney test, or Pearson's chi-squared test, depending on the statistical distribution of the variables. The total ADDICT-ICCU cohort was divided into a training cohort for developing the risk score and a validation cohort for validating the score. The allocation of patients into cohorts was randomly performed, stratified by inclusion site and done under the condition of equal event rates in the two cohorts. To handle missing data, imputation was performed using the k-nearest neighbour algorithm. To build the prediction score, a least absolute shrinkage and selection operator (LASSO) Cox regression was used to identify the most relevant variables associated with the primary outcome. The predictive power of each of the selected variables for the primary outcome was analysed using univariable and multivariable Cox proportional hazards method.

The conditional inference tree method was used to determine the best cut-off to transform each continuous variable selected by the LASSO algorithm into a binary variable with the best predictive value for the occurrence of HF outcomes. A significance level of 0.05 was applied for splits, with a minimum of 100 participants in each terminal node. The ICCU-HF score was built using the beta coefficients from the final multivariable Cox regression model. These coefficients were rounded to the nearest half-point to generate a practical scoring increment. This approach was designed to balance precision with clinical applicability. The area under the curve (AUC) of the receiver operating characteristics (ROC) curves were used to assess the performance of the ICCU-HF score, 95% confidence intervals (CI) were computed by stratified bootstrapping, and to compare AUCs, DeLong's test was used.¹⁸ Patients in the training cohort were classified into three groups based on the risk score and time to event was assessed by cumulative event curves generated by Kaplan–Meier estimates and compared with the log-rank test. A two-tailed P-value < 0.05 was considered statistically significant. All analyses were conducted using R software, version 4.2.2 (R Foundation for Statistical Computing).

Results

Study population

A flowchart of study patients is depicted in Figure 1. Between April 7 and 22, 1904 patients were admitted to the ICCUs at the 39 participating inclusion sites. After exclusion, 1471 patients were included and split into a training cohort of 1008 patients (68.5%) from 21 centres and a validation cohort of 463 patients (31.5%) from eight other centres. Baseline characteristics including clinical admission information of the included patients are depicted in Table 1. The mean age of the population was 63 ± 15 years, 70% were male, and comorbidities were highly prevalent with 53% of patients with hypertension, 38% with dyslipidaemia and 21% with diabetes. Regarding known cardiovascular diseases, 32% had a history of percutaneous coronary intervention (PCI), 6% had a history of HF hospitalization, and 7% had a history of ventricular arrhythmia. The most common admission diagnoses in the ICCU were ACS (52%), acute HF (14%) and arrhythmia (6%). The median duration of hospitalization was 5 (IQR: 3–7) days.

Table 1 Baseline characteristics of the study population (n = 1471)

Characteristics	Overall population (n = 1471)	Training cohort (n = 1008)	Validation cohort (n = 463)	P-value
Age, years	63 ± 15	63 ± 14	63 ± 16	0.9
Men	1026 (70%)	701 (70%)	325 (70%)	0.8
BMI, kg/m2	27 ± 6	27 ± 5	27 ± 6	0.015
Duration of hospitalization, days	5 (3–7)	5 (3–7)	5 (3–7)	0.7
CV risk factors
Diabetes	312 (21%)	228 (23%)	84 (18%)	0.051
Hypertension	773 (53%)	550 (55%)	223 (48%)	0.022
Dyslipidaemia	566 (38%)	404 (40%)	162 (35%)	0.062
Family history of CAD	247 (17%)	159 (16%)	88 (19%)	0.12
Current or former smoking	372 (25%)	257 (25%)	115 (25%)	0.8
Alcohol, more than once per week	789 (55%)	529 (53%)	260 (57%)	0.2
History of CVD
Previous ACS	224 (15%)	167 (17%)	57 (12%)	0.035
Previous PCI	476 (32%)	351 (35%)	125 (27%)	0.003
Previous CABG	50 (3%)	36 (4%)	14 (3%)	0.6
Previous HF hospitalization	85 (6%)	56 (6%)	29 (6%)	0.6
Previous ventricular arrhythmia	95 (7%)	67 (7%)	28 (6%)	0.7
Previous stroke	3 (0.2%)	2 (0.2%)	1 (0.2%)	>0.9
Known CKD	150 (10%)	100 (10%)	50 (11%)	0.6
Atrial fibrillation	170 (12%)	118 (12%)	52 (11%)	0.8
Other comorbidities
Cancer	57 (4%)	38 (4%)	19 (4%)	0.8
HIV	13 (0.9%)	11 (1.1%)	2 (0.4%)	0.4
COPD	62 (4%)	39 (4%)	23 (5%)	0.3
Medications
Beta blockers	425 (29%)	288 (29%)	137 (30%)	0.7
ACEi	294 (20%)	200 (20%)	94 (20%)	0.8
ARB	242 (16%)	175 (17%)	67 (14%)	0.2
MRA	77 (5%)	61 (6%)	16 (4%)	0.038
ARNi	36 (2%)	24 (2%)	12 (3%)	0.8
SGLT2i	11 (0.7%)	10 (1.0%)	1 (0.2%)	0.2
Diuretics	347 (24%)	237 (24%)	110 (24%)	>0.9
Statins	489 (33%)	336 (33%)	153 (33%)	>0.9
Anti-arrhythmic drugs	97 (7%)	64 (6%)	33 (7%)	0.6
Admission diagnosis
ACS	758 (52%)	535 (53%)	223 (48%)	0.080
Acute HF	202 (14%)	135 (14%)	67 (14%)	0.6
Arrhythmias	93 (6%)	63 (6%)	30 (7%)	0.9
Other	418 (28%)	275 (27%)	143 (31%)	0.2
Signs of HF at admission
No HF sign	1157 (79%)	797 (79%)	360 (78%)	0.6
Left ventricular failure	167 (11%)	108 (11%)	59 (13%)	0.3
Right ventricular failure	31 (2%)	18 (2%)	13 (3%)	0.2
Biventricular failure	92 (6%)	69 (7%)	23 (5%)	0.2
Cardiogenic shock	24 (2%)	16 (2%)	8 (2%)	0.8
Clinical data at admission
Systolic blood pressure, mmHg	136 ± 27	137 ± 27	134 ± 26	0.081
Heart rate, bpm	82 ± 24	83 ± 24	82 ± 23	0.9
Arterial oxygen saturation, %	97 ± 5	97 ± 5	97 ± 5	0.7
Killip score >1	248 (17%)	158 (16%)	90 (19%)	0.073
Illicit drug detected	157 (11%)	99 (10%)	58 (13%)	0.12
Echocardiography
LVEF, %	55 (45–60)	55 (45–60)	55 (45–60)	0.5
VTI, cm	19 ± 5	19 ± 5	19 ± 6	0.3
Dilated LV, LVEDD >55 mm	101 (7%)	63 (7%)	38 (9%)	0.2
E/e′ ratio	8 (6–10)	8 (6–11)	8 (6–10)	0.3
TAPSE, mm	21 ± 5	21 ± 5	21 ± 4	0.3
Presence of valvular disease grade 2+	212 (15%)	141 (14%)	71 (16%)	0.5
Laboratory results
NT-proBNP, pg/mL	700 (172–2500)	696 (181–2507)	700 (166–2383)	0.5
hsTroponin peak, ng/L	250 (40–3763)	304 (42–5618)	175 (33–1875)	<0.001
Creatinine, μmol/L	80 (67–99)	81 (68–99)	79 (66–97)	0.12
Haemoglobin, g/dL	14 (12–15)	14 (13–15)	14 (12–15)	0.2
Treatment during hospitalization
Use of i.v. diuretics	261 (18%)	183 (18%)	78 (17%)	0.5
Coronary angiography	968 (66%)	659 (65%)	309 (67%)	0.6
PCI	476 (32%)	351 (35%)	125 (27%)	0.003
Use of inotropes	27 (1.8%)	21 (2.1%)	6 (1.3%)	0.3
Use of MCS	5 (0.3%)	1 (0.1%)	4 (0.9%)	0.036
Use of mechanical ventilation	8 (0.5%)	4 (0.4%)	4 (0.9%)	0.3

The training cohort had a significantly higher prevalence of hypertension (53% vs. 48%, P = 0.022), previous PCI (35% vs. 27%, P = 0.003) and use of mineralocorticoid receptor antagonists (6% vs. 4%, P = 0.038), as well as higher peak high-sensitivity troponin concentration (304, IQR: 42–5618 ng/L vs. 175, IQR: 33–1875 ng/L, P < 0.001) compared with the validation cohort. Otherwise, the cohorts had homogeneous characteristics.

Prognostic factors of HF outcomes 1 year after ICCU discharge

In the training cohort (n = 1008 patients), 71 patients reached the primary outcomes (7.0%) at 1 year, including 34 hospitalizations for acute HF (3.4%) and 37 cardiovascular deaths (3.7%). A total of 58 demographic, clinical, echocardiographic and laboratory candidate variables were incorporated as input for the LASSO feature selection method. The LASSO model selected seven variables for further analysis: left ventricular ejection fraction (LVEF), presence of significant valvular disease grade 2+ (according to ESC guidelines, see Method S3), Killip score >1 at admission, NT-proBNP, creatinine, previous documented ventricular arrhythmia and use of inotropes during hospitalization. In univariable analysis, all of these seven variables were associated with the occurrence of HF outcomes (all P < 0.001; Table 2). An overview of univariable analyses of all 58 candidate variables can be found in Table S4.

Table 2 Univariable and multivariable Cox regressions for HF outcomes at 1 year in the training cohort (n = 1008)

	Univariable	Multivariablea
Characteristic	N	Events	HR	95% CI	P-value	HR	95% CI	P-value
Previous documented ventricular arrhythmia	1008	71	3.80	2.12, 6.81	<0.001	3.34	1.84, 6.08	<0.001
Killip score >1 at admission	1008	71	3.32	2.05, 5.39	<0.001	2.07	1.22, 3.50	0.007
NT-proBNP, pg/mLb	1008	71	1.00	1.00, 1.00	<0.001	1.00	1.00, 1.00	<0.001
Creatinine, μmol/Lb	1008	71	1.00	1.00, 1.01	<0.001	1.00	1.00, 1.01	0.023
LVEF, %b	1008	71	0.96	0.95, 0.97	<0.001	0.98	0.97, 1.00	0.051
Significant valvular disease grade 2+	1008	71	3.36	2.05, 5.49	<0.001	1.62	0.94, 2.80	0.083
Use of inotropes during hospitalization	1008	71	7.22	3.46, 15.1	<0.001	3.42	1.56, 7.50	0.002

In multivariable analysis, previously documented ventricular arrhythmia, Killip score >1 at admission, NT-proBNP, creatinine and use of inotropes during hospitalization remained independently associated with the occurrence of HF outcomes (Table 2).

Performance of the ICCU-HF score

Using conditional inference trees, the following optimal cut-offs were determined: <45% for LVEF, ≥125 μmol/L for creatinine and ≥175 and ≥2200 pg/mL for NT-proBNP. Based on the beta coefficients of the Cox regression model with discretized variables (Table S5), the proposed ICCU-HF score was created (Table 3). The ICCU-HF score can be calculated as the weighted sum of the seven variables. This score was applied to the training cohort (AUC 0.77, 95% CI [0.71–0.83]) (Figure 2A). Compared with the strongest individual predictor, NT-proBNP (AUC 0.72, 95% CI [0.66–0.79]), the ICCU-HF score exhibited superior performance (P = 0.008). The ICCU-HF score was then used to categorize patients into three risk groups: a low-risk group with an ICCU-HF score of <3, an intermediate-risk group with an ICCU score of 3–6, and a high-risk group with an ICCU-HF score of >6. The cumulative event curves showed a clear division of the three groups (Figure 3A), and with the low-risk group as reference, the intermediate-risk group (HR 4.09, 95% CI [2.23–7.50], P < 0.001) and high-risk group (HR 12.69, 95% CI [7.02–22.95], P < 0.001) demonstrated significantly increased risk of HF hospitalization or cardiovascular death.

Table 3 Description of the ICCU-HF score

Variables	Points
LVEF <45%	1
Presence of significant valvular disease grade 2+	1.5
Killip score >1 at admission	1.5
NT-proBNP >175 pg/mL	1.5
NT-proBNP >2200 pg/mL	2.5
Creatinine >125 μmol/L	2.5
Previous documented ventricular arrhythmia	2.5
Use of inotropes during hospitalization	3

[IMAGE OMITTED. SEE PDF]

[IMAGE OMITTED. SEE PDF]

Performance of the ICCU-HF score in the validation cohort

In the validation cohort (n = 463 patients), 28 patients exhibited HF outcomes (6.0%) during the follow-up of 1 year, including 17 hospitalizations for acute HF (3.7%) and 11 cardiovascular deaths (2.8%). The performance of the ICCU-HF score using the validation cohort was excellent (AUC 0.83, 95% CI [0.77–0.89]) (Figure 2B). Moreover, the ICCU-HF score showed a better performance than the ACUTE HF score⁸ (AUC 0.83, 95% CI [0.77–0.89] vs. 0.69, 95% CI [0.59–0.79], P = 0.006). Similar to the training cohort, the validation cohort was categorized into a low-, intermediate- and high-risk group based on the ICCU-HF score. The cumulative event curves showed that this risk stratification was robust in the validation cohort, but not able to significantly distinguish the intermediate- and high-risk group (P = 0.25) (Figure 3B).

Subgroup analyses

As a sensitivity analysis to assess whether the ICCU-HF score performed inconsistently in selected subgroups, we tested the ICCU-HF score in patients hospitalized for ACS and patients hospitalized for acute HF separately. In these two subgroups, the ICCU-HF score exhibited a good performance with AUC values of 0.73 (95% CI [0.65–0.82]) in ACS patients and 0.70 (95% CI [0.62–0.78]) in acute HF patients (Figure S1).

Since age was not selected by the LASSO model, we performed a sensitivity analysis in two age subgroups, defined by the median age of 65 years. The score demonstrated good performance in both the subgroup <65 years (AUC 0.78, 95% CI [0.67–0.89]) and the subgroups ≥65 years (AUC 0.74, 95% CI [0.68–0.80]).

Discussion

In a large multicentric cohort of consecutive patients hospitalized in the ICCU, we constructed the ICCU-HF score using seven traditional variables collected during the hospitalization for predicting the risk of 1-year HF outcomes: LVEF, significant valvular disease grade 2+, Killip score >1 at admission, NT-proBNP, creatinine, previous documented ventricular arrhythmia and use of inotropes during hospitalization. Notably, the ICCU-HF score showed a good performance in both the training and validation cohorts with better accuracy than existing models or scores. To our knowledge, this is the first large-scale investigation of prediction scores for prognostic risk evaluation of HF outcomes for all consecutive patients hospitalized in the ICCU.

Description of the ICCU-HF score

The ICCU-HF score consists of elements from the medical history (previous ventricular arrhythmia), clinical presentation (Killip score), echocardiographic measurements (LVEF and valvulopathy), laboratory results (NT-proBNP and creatinine) and treatment intensity (use of inotropes) of the patient. Thus, it covers a variety of features to characterize the patients.

The echocardiographic measurements, LVEF and the presence of significant valvular disease have previously been included in scores predicting outcomes related to HF.⁸ However, the predictive value of LVEF in a population of unselected cardiovascular patients has not been studied. In patients hospitalized for HF, reduced LVEF has been associated with a higher mortality.¹⁹ Likewise, no previous study has evaluated the presence of valvular disease in this population, although mitral and aortic valve diseases have been associated with excess mortality in both community cohorts²⁰ and cardiovascular patient cohorts.²¹ Moreover, due to the growing age of the population, the incidence of valvular heart diseases is increasing and will gradually become a future economic burden of the healthcare system²² and possibly a more substantial risk factor in cardiovascular patients.

The Killip classification dates more than half a century back and has been used as a predictive tool for in-hospital and post-discharge mortality in several patient populations.^23,24 The Killip classification provides a four-piece severity score indicating worsening prognosis with increasing scores: Class I, no evidence of HF; Class II, findings of mild to moderate HF (third heart sound, rales less than halfway up lung fields or elevated jugular venous pressure); Class III, pulmonary oedema; and Class IV, cardiogenic shock.²⁵ In the present study, we showed the predictive value of a Killip score >1 in a population of patients admitted to the ICCU with various admission diagnoses. A high Killip score suggests signs of congestion at admission, and as shown, 17% of the patients presented as such. An indicator of more severe decompensation, a state of cardiogenic shock, is the need for inotropic treatment during hospitalization. Only 1.8% of the patients received inotropic treatment, and we identified this feature as a strong predictor of post-discharge HF rehospitalization and death, consonant with previous findings.²⁶

Used as a screening tool for patients with suspected HF, and as an established prognostic marker, NT-proBNP is a routine measurement in HF patients. However, NT-proBNP provides prognostic value in various patient populations.^27,28 We identified NT-proBNP as the best predictor of the composite endpoint of unplanned HF hospitalization and cardiovascular death. Singly, NT-proBNP exhibited an AUC of 0.72 in the training cohort, which was inferior to the performance of the combined ICCU-HF score (P = 0.008). Similar to LVEF, the Killip score and the use of inotropes, we expect the variable selection of NT-proBNP to be associated with the HF hospitalization component of the composite endpoint. However, in our sensitivity analyses, both NT-proBNP and the ICCU-HF score showed inferior performance in a subset of patients admitted with acute HF compared with the training or validation cohort.

We identified previous ventricular arrhythmia as an uncommon (6.5%) but strong predictor of HF hospitalization and cardiovascular death. There are several aetiologies of ventricular arrhythmias, some being reversible, but despite improvements in revascularization, treatment with anti-arrhythmic drugs and the introduction of implantable cardioverter devices, ventricular arrhythmias remain a significant risk factor in cardiovascular disease.²⁹

The absence of age as a predictive variable in the ICCU-HF score is noteworthy given its ubiquity in similar prognostic tools. This may reflect the robustness of the selected variables, which could capture the risk attributable to age, or it may suggest a unique profile of the ICCU population where traditional age-related risks are less discriminative. Further, the ICCU-HF score showed consistent performance in age subgroups below and above the median age of the study population (Figure S2).

The potential of the ICCU-HF score

To our knowledge, the ICCU-HF score is the first post-discharge risk prediction score for a population of consecutive, unselected cardiovascular patients admitted to the ICCU at multiple centres. These features distinguish it from previous risk prediction scores mainly focusing on certain subpopulations of cardiovascular patients. The ADDICT-ICCU study was performed in 39 centres across France including all administrative regions of France. All patients admitted to the ICCUs were included regardless of admission diagnosis. This unrestricted multicentre design ensured coverage of all socioeconomic groups, and, hence, the study cohort should be representative of the general patient population admitted to the ICCU. We developed the score in a randomly selected training cohort from the ADDICT-ICCU study population and performed validation in a discrete validation cohort. The score includes easy-accessible, standardized information on comorbidities, clinical presentation, echocardiographic measurements and laboratory results that provide an extensive characterization of the patients. Thus, we consider the ICCU-HF score to be applicable in other cohorts of unselected patients admitted at the ICCU, and we will encourage further external validation of the score.

Limitations

This present study has some limitations. First, a few variables were excluded prior to the variable selection due to a low prevalence (<1% for dichotomous variables) or a high degree of missingness (>100 missing values, corresponding to >10% missing data). Hence, potential candidate variables may have been omitted. These variable exclusion criteria were specified prior to any data analysis. Second, we determined unplanned HF hospitalization and cardiovascular death as the composite outcome. We considered these to be the most severe and clinically relevant post-discharge outcomes for patients admitted to the ICCU. The inclusion of HF hospitalization as an endpoint might explain some of the variables selected in the prediction score, even though 52% of the ADDICT-ICCU patients were admitted due to ACS. However, in the sensitivity analyses, the score seemed to perform better in the unselected patient cohorts compared with subpopulations of acute HF or ACS patients. Third, the short enrolment period of 2 weeks may have introduced selection bias, as it does not capture known seasonal variations in heart failure admission rates and outcomes.^30,31 This limitation could affect the generalizability of our findings. Fourth, the 1-year event rates in the training and validation cohorts were 7% and 6%, respectively, revealing the ICCU population being at moderate risk for HF hospitalization or cardiovascular death. Although higher event rates would improve the strength of the predictions, we consider these event rates to be sufficient in the present analyses. Fifth, when applied to the validation cohort the ICCU-HF score could not significantly differentiate the 1-year risk of HF hospitalization or cardiovascular death between the intermediate- and high-risk groups. Although the event curves diverge, this is probably due to a low number of events in the groups. Sixth, the ICCU-HF score was compared against the ACUTE HF score, a tool developed to predict mortality in patients hospitalized for acute HF, which is not directly aligned with the present study's cohort of unselected ICCU admissions. Given the absence of risk prediction models derived from unselected ICCU patient cohorts, we chose to benchmark the ICCU-HF score against the ACUTE HF score, recognizing that it represents the closest available, though not entirely comparable, reference.

Conclusions

In a large multicentric cohort of consecutive patients hospitalized in the ICCU, the ICCU-HF score constituted by seven traditional variables collected during the hospitalization was an accurate method for predicting 1-year cardiovascular death and hospitalization for acute HF and proved superior to any metrics alone including NT-proBNP. The ICCU-HF score exhibited a higher prognostic value to predict HF outcomes than traditional models or existing scores including the ACUTE HF score. The performance of the ICCU-HF score was excellent using the validation cohort.

Conflict of interest

None declared.

Funding

This work was supported by an institutional grant from the French Heart Foundation (Fondation Coeur et Recherche) (Paris, France).

References

Woolridge S, Alemayehu W, Kaul P, Fordyce CB, Lawler PR, Lemay M, et al. National trends in coronary intensive care unit admissions, resource utilization, and outcomes. Eur Heart J Acute Cardiovasc Care 2020;9:923‐930. doi:10.1177/2048872619883400

Crespo‐Leiro MG, Anker SD, Maggioni AP, Coats AJ, Filippatos G, Ruschitzka F, et al. European Society of Cardiology Heart Failure Long‐Term Registry (ESC‐HF‐LT): 1‐year follow‐up outcomes and differences across regions. Eur J Heart Fail 2016;18:613‐625. doi:10.1002/ejhf.566

Eagle KA, Lim MJ, Dabbous OH, Pieper KS, Goldberg RJ, van de Werf F, et al. A validated prediction model for all forms of acute coronary syndrome estimating the risk of 6‐month postdischarge death in an international registry. JAMA 2004;291:2727‐2733. doi:10.1001/jama.291.22.2727

de Araújo GP, Ferreira J, Aguiar C, Seabra‐Gomes R. TIMI, PURSUIT, and GRACE risk scores: sustained prognostic value and interaction with revascularization in NSTE‐ACS. Eur Heart J 2005;26:865‐872. doi:10.1093/eurheartj/ehi187

Pocock SJ, Huo Y, Van de Werf F, et al. Predicting two‐year mortality from discharge after acute coronary syndrome: an internationally‐based risk score. Eur Heart J Acute Cardiovasc Care 2019;8:727‐737. doi:10.1177/2048872617719638

Lee DS, Austin PC, Rouleau JL, Liu PP, Naimark D, Tu JV. Predicting mortality among patients hospitalized for heart failure: derivation and validation of a clinical model. JAMA 2003;290:2581‐2587. doi:10.1001/jama.290.19.2581

O'Connor CM, Hasselblad V, Mehta RH, Tasissa G, Califf RM, Fiuzat M, et al. Triage after hospitalization with advanced heart failure: the ESCAPE risk model and discharge score. J Am Coll Cardiol 2010;55:872‐878. doi:10.1016/j.jacc.2009.08.083

Cameli M, Pastore MC, De Carli G, et al. ACUTE HF score, a multiparametric prognostic tool for acute heart failure: a real‐life study. Int J Cardiol 2019;296:103‐108. doi:10.1016/j.ijcard.2019.07.015

Buddeke J, Bots ML, van Dis I, Liem A, Visseren FLJ, Vaartjes I. Trends in comorbidity in patients hospitalised for cardiovascular disease. Int J Cardiol 2017;248:382‐388. doi:10.1016/j.ijcard.2017.06.106

Ibanez B, James S, Agewall S, Antunes MJ, Bucciarelli‐Ducci C, et al. 2017 ESC guidelines for the management of acute myocardial infarction in patients presenting with ST‐segment elevation: the task force for the management of acute myocardial infarction in patients presenting with ST‐segment elevation of the European Society of Cardiology (ESC). Eur Heart J 2018;39:119‐177. doi:10.1093/eurheartj/ehx393

Collet JP, Thiele H, Barbato E, Barthélémy O, Bauersachs J, Bhatt DL, et al. 2020 ESC guidelines for the management of acute coronary syndromes in patients presenting without persistent ST‐segment elevation. Eur Heart J 2021;42:1289‐1367. doi:10.1093/eurheartj/ehaa575

Johansson S, Rosengren A, Young K, Jennings E. Mortality and morbidity trends after the first year in survivors of acute myocardial infarction: a systematic review. BMC Cardiovasc Disord 2017;17:53. doi:10.1186/s12872‐017‐0482‐9

Heidenreich PA, Bozkurt B, Aguilar D, Allen LA, Byun JJ, Colvin MM, et al. 2022 AHA/ACC/HFSA guideline for the management of heart failure: a report of the American College of Cardiology/American Heart Association joint committee on clinical practice guidelines. Circulation 2022;145:e895‐e1032. doi:10.1161/cir.0000000000001063

Taylor CJ, Ordóñez‐Mena JM, Roalfe AK, Lay‐Flurrie S, Jones NR, Marshall T, et al. Trends in survival after a diagnosis of heart failure in the United Kingdom 2000‐2017: population based cohort study. BMJ 2019;364:l223. doi:10.1136/bmj.l223

Dillinger JG, Pezel T, Fauvel C, et al. Prevalence of psychoactive drug use in patients hospitalized for acute cardiac events: rationale and design of the ADDICT‐ICCU trial, from the Emergency and Acute Cardiovascular Care Working Group and the National College of Cardiologists in Training of the French Society of Cardiology. Arch Cardiovasc Dis 2022;115:514‐520. doi:10.1016/j.acvd.2022.05.012

McDonagh TA, Metra M, Adamo M, et al. 2021 ESC guidelines for the diagnosis and treatment of acute and chronic heart failure. Eur Heart J 2021;42:3599‐3726. doi:10.1093/eurheartj/ehab368

Hicks KA, Tcheng JE, Bozkurt B, Chaitman BR, Cutlip DE, Farb A, et al. 2014 ACC/AHA key data elements and definitions for cardiovascular endpoint events in clinical trials. Circulation 2015;132:302‐361. doi:10.1161/CIR.0000000000000156

DeLong ER, DeLong DM, Clarke‐Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics 1988;44:837‐845.

Solomon SD, Anavekar N, Skali H, McMurray J, Swedberg K, Yusuf S, et al. Influence of ejection fraction on cardiovascular outcomes in a broad spectrum of heart failure patients. Circulation 2005;112:3738‐3744. doi:10.1161/CIRCULATIONAHA.105.561423

Tung M, Nah G, Tang J, Marcus G, Delling FN. Valvular disease burden in the modern era of percutaneous and surgical interventions: the UK biobank. Open Heart 2022;9:e002039. doi:10.1136/openhrt‐2022‐002039

Pai RG, Varadarajan P. Prognostic implications of mitral regurgitation in patients with severe aortic regurgitation. Circulation 2010;122:S43‐S47. doi:10.1161/CIRCULATIONAHA.109.927921

Coffey S, Roberts‐Thomson R, Brown A, Carapetis J, Chen M, Enriquez‐Sarano M, et al. Global epidemiology of valvular heart disease. Nat Rev Cardiol 2021;18:853‐864. doi:10.1038/s41569‐021‐00570‐z

Khot UN, Jia G, Moliterno DJ, Lincoff AM, Khot MB, Harrington RA, et al. Prognostic importance of physical examination for heart failure in non–ST‐elevation acute coronary syndromes. The enduring value of Killip classification. JAMA 2003;290:2174‐2181. doi:10.1001/jama.290.16.2174

Rudiger A, Harjola VP, Müller A, Mattila E, Säila P, Nieminen M, et al. Acute heart failure: clinical presentation, one‐year mortality and prognostic factors. Eur J Heart Fail 2005;7:662‐670. doi:10.1016/j.ejheart.2005.01.014

Killip T, Kimball JT. Treatment of myocardial infarction in a coronary care unit: a two year experience with 250 patients. Am J Cardiol 1967;20:457‐464. doi:10.1016/0002‐9149(67)90023‐9

Mortara A, Oliva F, Metra M, Carbonieri E, di Lenarda A, Gorini M, et al. Treatment with inotropes and related prognosis in acute heart failure: contemporary data from the Italian Network on Heart Failure (IN‐HF) outcome registry. J Heart Lung Transplant Off Publ Int Soc Heart Transplant 2014;33:1056‐1065. doi:10.1016/j.healun.2014.05.015

Paniagua R, Ventura MD, Ávila‐Díaz M, Hinojosa‐Heredia H, Méndez‐Durán A, Cueto‐Manzano A, et al. NT‐proBNP, fluid volume overload and dialysis modality are independent predictors of mortality in ESRD patients. Nephrol Dial Transplant 2010;25:551‐557. doi:10.1093/ndt/gfp395

Provan S, Angel K, Semb AG, Atar D, Kvien TK. NT‐proBNP predicts mortality in patients with rheumatoid arthritis: results from 10‐year follow‐up of the EURIDISS study. Ann Rheum Dis 2010;69:1946‐1950. doi:10.1136/ard.2009.127704

Zeppenfeld K, Tfelt‐Hansen J, de Riva M, Winkel BG, Behr ER, Blom NA, et al. 2022 ESC guidelines for the management of patients with ventricular arrhythmias and the prevention of sudden cardiac death. Eur Heart J 2022;43:3997‐4126. doi:10.1093/eurheartj/ehac262

Boulay F, Berthier F, Sisteron O, Gendreike Y, Gibelin P. Seasonal variation in chronic heart failure hospitalizations and mortality in France. Circulation 1999;100:280‐286. doi:10.1161/01.cir.100.3.280

Stewart S, McIntyre K, Capewell S, McMurray JJV. Heart failure in a cold climate. Seasonal variation in heart failure‐related morbidity and mortality. J Am Coll Cardiol 2002;39:760‐766. doi:10.1016/s0735‐1097(02)01685‐6