Prognostic impact of the number and Temporality of heart failure Hospitalisations: Analysis of a National healthcare database

Highlights

• •Prior HF hospitalisation is the strongest predictor of adverse outcomes in reduced LVEF.
• •Both recency and number of hospitalisations have an additive negative impact on mortality and rehospitalisation.
• •Recent and/or frequent hospitalisations are simple, pragmatic markers to identify high-risk patients.
• •These patients require prioritisation for guideline-directed therapy, transitional care, and monitoring strategies.

Abstract

Aims

 To assess the prognostic impact of both the frequency and timing of prior heart failure (HF) hospitalisations on outcomes in patients with reduced left ventricular ejection fraction (LVEF).

Methods and results

This nationwide retrospective cohort study used the French national health insurance database to identify 730,052 adults with HF in 2017. A validated algorithm classified 226,747 as HF with reduced LVEF (<45 %), of whom 54,504 (24 %) had at least one HF-related hospitalisation >24 h within the preceding 24 months (worsening HF group). Patients were stratified by (1) time since the last HF hospitalisation (0–6, 6–24 months) and (2) number of hospitalisations (1, 2, ≥3). Mean age was 76 ± 15 years. Prior HF hospitalisation was the strongest predictor of mortality among all variables. After multivariable adjustment, prior hospitalisation was associated with increased risk of all-cause death (HR 1.61, 95 % CI 1.56–1.65), all-cause hospitalisation (HR 1.34, 95 % CI 1.32–1.37), and recurrent HF hospitalisation (HR 2.51, 95 % CI 2.43–2.59). Risks were greatest when the most recent hospitalisation occurred within 6 months and rose progressively with the number of prior events.

Conclusion

In patients with reduced LVEF, both recent and recurrent HF hospitalisations are strong predictors of mortality and rehospitalisation. These two simple markers identify highly vulnerable patients and should trigger intensified follow-up, optimisation of guideline-directed therapies, and implementation of transitional care and remote monitoring programs.

1. Introduction

Heart failure (HF) is a chronic, progressive syndrome characterised by alternating phases of clinical stability and episodes of worsening symptoms. Stable periods are defined by an absence of congestion under steady therapy, whereas decompensations involve symptom deterioration and/or congestion requiring therapeutic intensification. Despite advances in care, these worsening episodes frequently lead to hospitalisation and are a major driver of morbidity, mortality, and healthcare costs [1].

Increasing attention has focused on worsening heart failure (WHF), defined as symptom deterioration despite ongoing therapy, leading to hospitalisation, emergency department visits, or escalation of outpatient management [[2], [3], [4]]. WHF is consistently associated with impaired prognosis and accounts for a substantial proportion of the economic burden of HF [5,6].

HF hospitalisation represents the most visible and reproducible manifestation of WHF in large-scale databases. Patients requiring HF hospitalisation face a substantially higher risk of recurrent events and death compared with never-hospitalised patients [[7], [8], [9]]. Approximately half are readmitted within 3–6 months, often due to recurrent decompensation [10]. Each subsequent event tends to occur earlier and carries progressively worse outcomes [11].

However, available evidence remains fragmented, mainly derived from ancillary analyses of clinical trials or modest registries. Robust nationwide data are lacking.

This study aimed to describe the demographic and clinical characteristics, treatment, and outcomes of patients with reduced LVEF according to the number and timing of prior HF-related hospitalisations, using a nationwide French claims database.

2. Methods

2.1. Study design and data source

This was a nationwide, retrospective, observational cohort study using the French National Health Data System (Système National des Données de Santé, SNDS), which covers ∼ 99 % of the French population. The database includes sociodemographic information, reimbursement for ambulatory care, prescriptions (ATC codes), hospitalisations and diagnoses at discharge (ICD-10), mortality and costs [12]. For this study, hospitalisation data were available for 2017.

2.2. Study population

Eligible patients were adults (≥18 years) with either (1) ≥ 1 hospitalisation of ≥ 24 h with HF (ICD-10 I50) in 2017, or (2) registration for long-term disease status related to HF (full coverage beneficiaries for HF-related care).

Since LVEF data was not directly available in the SNDS in 2017, patients with reduced LVEF (<45 %) were identified using a machine-learning algorithm. The 45 % threshold was chosen to dichotomise LVEF into reduced versus non-reduced, rather than 40 and 50 % thresholds, given variability in measurement and ongoing controversies regarding the mildly reduced EF (41–49 %) category. It should also be noted that this threshold of 45 % was used in clinical trials focusing more specifically on WHF [13]. In brief, the algorithm was developed using a machine learning approach that incorporated age and gender, as well as medico-administrative code-based proxies for comorbidities, ATC codes for drug delivery, ICD-10 codes for hospitalisations and any other type of reimbursed care. The algorithm was trained using the FRESH cohort, which included LVEF data and was linked to the SNDS [14,15]. In addition to cross-validation, the algorithm was externally validated using a second, independent cohort of heart failure (HF) patients with left ventricular ejection fraction (LVEF) data, which was also linked to SNDS [16]. Area under the curve, accuracy, sensitivity, specificity, positive and negative predictive values of this algorithm were 0.81, 0.74, 0.78, 0.70, 0.69 and 0.77 respectively.

The study cohort included patients with reduced LVEF alive on 1 January 2018 (index date).

Definition of worsening HF and subgroup classification

WHF was defined as ≥1 HF-related hospitalisation (≥24 h, ICD-10 I50 as primary diagnosis) within 24 months before the index date. Patients were stratified according to:

• •Timing of last hospitalisation: 0–6 months, 6–24 months.
• •Number of hospitalisations: 1, 2, or ≥ 3 in the preceding 24 months.

Outcomes

During 2018–2019, three outcomes were assessed:

• All-cause mortality,

• All-cause hospitalisation (>24 h),

• HF-related hospitalisation (>24 h, ICD-10 I50 as primary diagnosis).

2.3. Statistical analysis

Continuous variables are reported as mean ± standard deviation (SD), categorical variables as numbers and percentages. There was no missing data in SNDS, particularly regarding age, gender, treatment delivery and outcome. However, we cannot rule out the possibility of a few coding errors or missing data for some of the comorbidities studied (atrial fibrillation, hyperlipidaemia, COPD, hypertension or stroke). Comparisons used Student’s t-test, Mann–Whitney or Kolmogorov–Smirnov tests for skewed data, and χ² tests for categorical variables.

Cox proportional hazard models were used to estimate hazard ratios (HRs) and 95 % confidence intervals (CIs) for all available variables and for each of the three studied outcomes at 1 year. Variables with p < 0.10 in univariate analyses were included in multivariable models (backward stepwise selection). Proportional hazards assumptions were tested using Schoenfeld residuals. Kaplan–Meier curves were generated and compared with log-rank tests.

2.4. Ethics

The study was approved by the Comité d’Expertise pour les Recherches, les Études et les Évaluations dans le domaine de la Santé (TPS 347113bis, April 25, 2019). Data protection approval was obtained from the French data protection authority (CNIL, DR-2019–341, November 19, 2019).

3. Results

3.1. Patient disposition

Among 730,052 patients identified with HF in 2017, 276,935 (37.9 %) were classified as having reduced LVEF (<45 %). Of these, 226,747 were alive on the index date. Within this group, 54,504 (24.0 %) had one or more HF-related hospitalisation in the preceding 24 months (WHF group) and were categorised as follows:

• •By timing of last hospitalisation: 25,436 (46.7 %) within 0–6 months, 29,068 (53.3 %) within 6–24 months.
• •By number of hospitalisations: 35,206 (64.6 %) had one, 11,386 (20.9 %) had two, 7,912 (14.5 %) had three or more.

3.2. Baseline characteristics

The characteristics of the study population with HF and reduced LVEF are presented in Table 1. Most patients were elderly, male and had at least one comorbidity. Overall, patients had been dispensed one or more HF drugs during the 3-month period before the index date, most commonly loop diuretics (53.9 %), beta-blockers (51.0 %), angiotensin-converting enzyme inhibitors or angiotensin receptors blockers (ACEi or ARB; 58.9 %) and mineraloreceptor antagonists (MRA) in 23.7 %. No ARNi or SGLT2 inhibitors were dispensed, as they were not reimbursed during the study period.

Table 1.

Characteristics of the study population with heart failure (HF) and reduced left ventricular ejection fraction (<45 %), classified into worsening and non-worsening HF groups. COPD: Chronic obstructive pulmonary disease; ACEi: Angiotensin-converting enzyme inhibitors; MRA: Mineralocorticoid receptor antagonists; ARB: Angiotensin receptor blockers.

Parameter	Study population (n = 226,747)	Previous HF hospitalization (n = 54,504)	No previous HF hospitalization (n = 172,243)	p-value‡
Females, %	21.2	23.9	20.3	<0.001
Age (years)	75.6 ± 14.5	75.8 ± 13.2	75.5 ± 14.9	0.34
Ischemic heart disease, %	54.0	59.2	52.4	<0.001
Comorbidities, %
Charlson Comorbidity Index	1.9 ± 2.7	3.2 ± 2.8	1.4 ± 2.4	<0.001
Atrial fibrillation	32.9	62.0	23.7	<0.001
Diabetes	22.3	29.1	20.1	<0.001
COPD	14.8	24.3	11.8	<0.001
Peripheral arterial disease	9.5	15.3	7.7	<0.001
Sleep apnea	8.4	13.6	6.8	<0.001
Stroke	16.0	8.2	5.4	<0.001
Pharmacological treatment, %
Loop Diuretics	53.9	81.6	45.1	<0.001
Beta-blockers	51.0	66.9	46.0	<0.001
ACEi/ARB	58.9	57.3	59.4	<0.001
MRA	23.7	36.9	19.5	<0.001
Digoxin	4.9	5.2	4.8	<0.001
Ivabradine	2.7	3.4	2.5	<0.001

^‡Comparison of previous heart failure hospitalization vs no previous heart failure hospitalization.

As compared with HF patients without previous HF-related hospitalisation during the last 2 years, patients with worsening HF had similar age but differed by significantly higher rates of comorbidities, nearly two-fold more, as well as delivery of HF drugs (Table 1).

Within the WHF group, both the recency and number of prior hospitalisations were associated with a greater burden of comorbidities and a higher rate of HF drug delivery, except for ACEi/ARB, compared to patients without a prior HF hospitalisation (Supplementary Tables 1 and 2).

3.3. Outcomes

The Fig. 1 shows Forest plots of multivariable Cox proportional hazard models for one-year all-cause death and hospitalisation. Ageing, most comorbidities, diuretics use were associated with adverse events. On the other hand, ACEi/ARB, betablockers, MRA and cardiac resynchronization were associated with the best outcome. WHF was independently associated with each type of outcome and was even the strongest predictor of death. The Fig. 2 and Supplementary Fig. 2A show outcomes according to previous or not previous hospitalisation. After multivariable adjustment, WHF patients had:

• •60 % higher risk of all-cause mortality (HR 1.61, 95 % CI 1.56–1.65),
• •35 % higher risk of all-cause hospitalisation (HR 1.34, 95 % CI 1.32–1.37),
• •2.5-fold higher risk of recurrent HF hospitalisation (HR 2.51, 95 % CI 2.43–2.59).

Fig. 1.

Multivariable Cox proportional hazard models for A/ one-year all-cause death, B/ one-year all-cause hospitalisation. Worsening: worsening heart failure (previous HF hospitalization), AF: atrial fibrillation, IHD: ischemic heart disease, COPD: chronic pulmonary disease, PAD: peripheral artery disease, VHD: valvular heart disease, MRA: mineraloreceptor antagonist, CRT: cardiac resynchronization therapy.

Fig. 2.

Cumulative incidence of A/ all-cause death, B/ heart failure-related hospitalisations over two years after the index date according to previous HF hospitalization and no previous HF hospitalization.

Risks increased progressively with both recency and frequency:

• •Timing (Fig. 3, Table 2): Mortality risk was highest with last hospitalisation within 6 months (HR 3.36) versus 6–24 months (HR 1.68). Similarly, all-cause hospitalisation was highest with last hospitalisation within 6 months (HR 2.40), versus 6–24 months (HR 1.50). Recurrent HF hospitalisation was also highest with last hospitalisation within 6 months (HR 6.04 versus 3.14 in the other subgroup).
• •Number of prior HF hospitalisations (Fig. 4, Table 2): Mortality rose stepwise from HR 1.82 (1 hospitalisation) to HR 3.84 (≥3 hospitalisations). For HF rehospitalisation, risk reached HR 9.99 with ≥ 3 prior events.

Fig. 3.

Cumulative incidence of time to A/ all-cause death, B/ heart failure-related hospitalisations over one year after the index date according to the timing of last heart failure hospitalisation prior to the index date.

Table 2.

Cox proportional hazard of Worsening Heart Failure after multivariable adjustment and according to recency and frequency of prior HF hospitalisations. Patient without prior heart failure related-hospitalisation as reference.

	Mortality	All-cause hospitalisation	HF-hospitalisation
last HF hospitalisation < 6 months	HR 3.36 [95 % CI 3.23–3.50]	HR 2.40 [95 % CI 2.34–2.46]	HR 6.04 (95 % CI 5.76–6.35]
last HF hospitalisation 6–24 months	HR 1.68 [95 % CI 1.62–1.74]	HR 1.50 [95 % CI 1.47–1.54]	HR 3.14 (95 % CI 2.99–3.29]
1 prior HF hospitalisation	HR 1.82 [95 % CI 1.76–1.89]	HR 1.60 [95 % CI 1.57–1.64]	HR 3.08 [95 % CI 2.95–3.22]
2 prior HF hospitalisations	HR 2.68 [95 % CI 2.53–2.84]	HR 2.14 [95 % CI 2.07–2.21]	HR 5.61 [95 % CI 5.19–6.06]
≥3 prior HF hospitalisations	HR 3.84 [95 % CI 3.57–4.12]	HR 3.02 [95 % CI 2.89–3.16]	HR 9.99 [95 % CI 9.05–11.02]

Fig. 4.

Cumulative incidence of A/ all-cause death, B/ heart failure-related hospitalisations according to the number of HF-related hospitalisation prior to the index date.

Fig. 5 shows the gradual increase in the rate of death or HF-related hospitalisation according to the length of time since the last HF-related hospitalisation and the number of HF hospitalisations prior to the index date.

Fig. 5.

Cumulative incidence of death or heart failure-related hospitalisation at one year according to both length of time since the last HF-related hospitalisation and the number of HF-related hospitalisation prior to the index date.

4. Discussion

This nationwide study highlights the prognostic significance of prior HF hospitalisation in patients with reduced LVEF. Three major findings emerge:

• Prior HF hospitalisation is the strongest predictor of death and rehospitalisation.

• Both recency and number of hospitalisations independently worsen prognosis in an additive manner.

• These markers are simple, easily identifiable, and highly relevant for risk stratification.

Our findings confirm and extend prior evidence. Previous studies have suggested a vulnerable post-discharge period and a cumulative adverse impact of repeated decompensations [[17], [18], [19], [20]]. By leveraging nationwide data, our study demonstrates the generalisability of these associations at the population level.

Real-world cohort studies complement clinical trials by providing insights into patient characteristics and treatment practices in unselected populations, more representative in terms of age, comorbidities, and routine care [17]. In our study, 24 % of patients with reduced LVEF had a history of HF-related hospitalization within 24 months before the index date, defining the group with possible WHF. Mean age and sex distribution were similar across groups, suggesting these factors were not major contributors to worsening. In contrast, patients with previous HF hospitalizations had a substantially higher comorbidity burden, particularly atrial fibrillation, hyperlipidemia, diabetes, and COPD—highlighting comorbidities as key risk factors for adverse outcomes. Such comorbidities may also limit treatment optimization or adherence to healthy behaviors, notably physical activity.

Rates of evidence-based HF drug delivery were low overall, especially in patients without prior hospitalization, underscoring the persistent gap between guidelines and practice. While some registries report high prescription rates in HFrEF, administrative datasets comparable to ours show similarly modest fill rates [18]. This likely reflects both therapeutic inertia in patients perceived as low risk and poor treatment adherence, since our data capture dispensed, not prescribed, medications. Importantly, HF drug delivery was associated with improved outcomes in our Cox models, emphasizing the prognostic value of optimized therapy. Early post-hospitalization optimization has proven beneficial [19], though possibly less effective in the most severe patients [20]. Consistent with this, our cohort showed higher rates of HF drug delivery in those with prior hospitalizations. Despite optimised therapy following hospitalisation for HF, the prognosis for these patients is worse than for patients with less optimised therapy but without a prior HF hospitalisation. This highlights the strong prognostic significance of recent and repeated HF hospitalisations.

Prognosis after HF hospitalization likely varies depending on whether it reflects progressive disease or a reversible trigger. In our study, patients with previous hospitalizations were more frequently rehospitalized and had markedly higher mortality—22.0 % vs 9.4 % at 1 year and 35.9 % vs 17.3 % at 2 years (both p < 0.05). Outcomes were especially poor in those hospitalized repeatedly or within 6 months of the index date, with cumulative risks compounded when both conditions were present. These patients had similarly low rates of evidence-based therapies, indicating differences in outcomes cannot be explained by undertreatment alone. Rather, they reflect a higher baseline risk, particularly during the first 3–4 months post-hospitalization as shown by the shape of the survival curves—the “vulnerable period” [21]. Persistent divergence of survival curves highlights the residual risk in this population, consistent with other studies reporting excess mortality and rehospitalization in WHF and advanced HF, as well as after any HF hospitalization [22]. These findings reinforce guideline recommendations for early reassessment and drug titration [23], and support intensified management through close follow-up, transitional care programs, and telemonitoring. A nationwide database study recently confirmed that closer cardiology follow-up was associated with improved outcomes [24]. Improving heart failure care also necessitates attention to patient, carer and societal factors. This includes raising public awareness of early symptoms, enhancing patient and carer education to support self-management, and reallocating or augmenting funding.“

4.1. Limitations

In this study, WHF was defined exclusively by prior HF hospitalizations. Broader definitions include outpatient intensification of diuretics or unplanned consultations for IV diuretics, but such events could not be reliably identified in our dataset. However, in France in 2017, ambulatory management of decompensated HF was uncommon due to structural and reimbursement limitations. Thus, this likely had limited impact, although it may have underestimated group differences. We also acknowledge that prior hospitalization is not synonymous with WHF, yet focusing on hospitalizations ensures greater reliability. Selection bias is possible, as patients hospitalized 6–24 months before the index date survived to inclusion and may not be fully comparable. Our study focused on patients with heart failure and an LVEF of ≤ 45 %. One limitation of our study is that LVEF was not recorded directly in our claims-based database, but rather was identified using an algorithm. Coding bias may also apply, as ∼ 25 % of HF cases are not captured in administrative databases [25]. While ICD-10 I50.x codes have good specificity, they lack sensitivity [26]. Furthermore, SNDS records drug dispensing but not actual intake, and provides no clinical data such as LVEF, renal function, natriuretic peptides, or NYHA class. While this limits granularity, it also reflects the type of data realistically available for large-scale health policy decisions. Finally, our analysis predates reimbursement of ARNIs and widespread SGLT2 inhibitor use, but these advances are unlikely to alter the main conclusions.

5. Conclusion

In patients with HF and reduced LVEF, both recent and recurrent hospitalisations are strong, independent predictors of mortality and rehospitalisation. These simple markers can be readily implemented in clinical practice and health systems to identify highly vulnerable patients. Early, intensive follow-up and optimisation of therapy, together with structured transitional care and remote monitoring programs, are essential to improve outcomes.

CRediT authorship contribution statement

Damien Logeart: Conceptualization, Formal analysis, Investigation, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing. François Roubille: Investigation, Validation, Visualization, Writing – original draft, Writing – review & editing. Muriel Salvat: Investigation, Validation, Writing – review & editing. Christophe Tribouilloy: Investigation, Visualization, Writing – review & editing. Fabrice Bauer: Visualization, Writing – review & editing. Jean-Christophe Eicher: Validation, Visualization, Writing – review & editing. François Picard: Investigation, Validation, Visualization, Writing – review & editing. Jean-Jacques Von Hunolstein: Investigation, Validation, Visualization, Writing – review & editing. Jean-Noël Trochu: Investigation, Validation, Visualization, Writing – review & editing. Pascal de Groote: Investigation, Validation, Visualization, Writing – review & editing. Emmanuelle Berthelot: Validation, Visualization, Writing – review & editing. Francis Fagnani: Methodology, Validation, Visualization, Writing – review & editing. Leila Batel: Conceptualization, Funding acquisition, Supervision. Maxime Doublet: Data curation, Project administration, Resources, Software, Supervision. Thibaud Damy: Conceptualization, Supervision, Validation, Visualization, Writing – review & editing. Richard Isnard: Conceptualization, Investigation, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Doublet reports administrative support and statistical analysis were provided by Bayer AG. Logeart reports a relationship with AstraZeneca that includes: consulting or advisory and speaking and lecture fees. Logeart reports a relationship with Bayer that includes: consulting or advisory. Logeart reports a relationship with Novartis that includes: board membership, consulting or advisory, and travel reimbursement. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: