Cardiologist follow-up and improved outcomes of heart failure: a French nationwide cohort

Abstract

Background and Aims

Outpatient cardiology follow-up is the cornerstone of heart failure (HF) management, requiring adaptation based on patient severity. However, risk stratification using administrative data is scarce, and the association between follow-up and prognosis according to patient risk has yet to be described at a population level. This study aimed to describe prognosis and management across different strata using simple criteria, including diuretic use and prior HF hospitalization (HFH).

Methods

This nationwide cohort included all French patients reported as having HF in the previous 5 years and alive on 1 January 2020. Patients were categorized into four groups: (i) HFH within the past year (HFH ≤ 1y), (ii) HFH 1–5 years ago (HFH > 1y), (iii) not hospitalized using loop diuretics (NoHFH/LD+), and (iv) not hospitalized without loop diuretics (NoHFH/LD−). Between-group associations, all-cause mortality (ACM), and cardiology follow-up were analysed using survival models.

Results

The study included 655 919 patients [80 years (70–87), 48% female]. One-year ACM risk was 15.9%, ranging from 8.0% (NoHFH/LD−) to 25.0% (HFH ≤ 1y). Mortality risk was 1.61-fold higher for NoHFH/LD+, 1.83-fold for HFH > 1y, and 2.32-fold for HFH ≤ 1y compared to NoHFH/LD− (P < .0001). During the first year of follow-up (2020), cardiology consultation rates were similar across groups, with 40% of patients lacking an annual visit. Compared to no consultation, a single cardiology visit in the previous year (2019) was associated with a 6%–9% absolute reduction in 1-year ACM during the following year (2020) across all groups. The number needed to consult (NNC) to prevent one modelled death was 11–16. Additional visits showed greater benefit with increasing HF severity, with NNC ranging from 55 (NoHFH/LD−) to 20 (HFH ≤ 1y). The optimal follow-up to minimize the number of deaths without increasing the total number of consultations was 1 annual visit for NoHFH/LD−, 2–3 visits for NoHFH/LD+ and HFH > 1y, and 4 visits for HFH ≤ 1y patients.

Conclusions

Despite having a HF diagnosis, 40% of patients do not see a cardiologist annually, regardless of disease severity. Simple stratification based on hospitalization history and diuretic use effectively predicts outcomes. Tailoring the annual number of HF consultations according to this stratification could optimize resource use and reduce avoidable modelled deaths.

Structured Graphical Abstract

Structured Graphical Abstract.

See the editorial comment for this article ‘Heart failure: the need for cardiologist-guided follow-up to improve outcomes’, by L.H. Lund, https://doi.org/10.1093/eurheartj/ehaf229.

Introduction

Heart failure (HF) is a major global public health issue, leading to over 1 million yearly hospitalizations in both Europe and the USA and affecting more than 60 million people worldwide.^1,2 Current guidelines recommend tailoring HF management to the severity of the disease.³ However, there are no straightforward criteria to assess severity, nor is there a validated algorithm for optimal follow-up using available consultation data.

Among potential criteria to identify high-risk HF patients, the use of loop diuretic medication has been associated with poor prognosis. This is observed in both patients with and without documented HF, compared to HF patients not using diuretics.⁴ Similarly, a recent HF hospitalization (HFH) is strongly associated with increased mortality risk.⁵ In a comprehensive HF patient population study, those hospitalized for HF had a three-fold increase in 1-year mortality, reaching 27%.⁶ While this risk ultimately decreases over time, it remains higher than in patients who have never been hospitalized.⁵

Establishing the prognostic value of a simple administrative data stratification in a comprehensive national HF population—without the need for patient consultations or clinical examinations to predict HFH or all-cause mortality—would be particularly valuable. This approach would enable the implementation of a standardized and automated minimum medical follow-up based solely on administrative criteria, improving population management without requiring complex medical information. Additionally, these criteria, being easily reproducible and not dependent on biological markers such as natriuretic peptides, which may be less accessible in certain countries, would facilitate broad validation of this stratification across various healthcare systems. Real-world data on the adaptation of cardiology follow-up according to disease risk factors such as diuretic use and hospitalization remain unknown, as are the associated benefits of this approach.

The present study aimed to describe the overall prognosis and management of HF patients on a national scale using reliable administrative severity criteria. Additionally, we sought to assess the optimal cardiological follow-up based on these HF severity criteria.

Methods

Population

This nationwide cohort study (referred to as French-DataHF) included HF patients identified from French health insurance databases. Except for patients living in nursing homes, all patients over 18 years of age with a long-term condition (LTC) declaration for HF by 31 December 2019, or an HF hospitalization in the previous 5 years (from 1 January 2015, to 31 December 2019), who were still alive on 1 January 2020, were identified. The LTC declaration ensures 100% healthcare coverage for the disease, thus providing reliable data. In brief, HFH were defined by a principal diagnosis code beginning with I50, I11, I13, or J81 [in accordance with the International Classification of Diseases, 10th Revision (ICD-10)]. The exact criteria used to identify the population and define HFH are detailed in Supplementary Document S1.

Characteristics of the French health insurance database

This study was conducted using the French National Health Data System [Système National des Données de Santé (SNDS)] which provides data for the entire French population with universal public health insurance (67 000 000 subjects). It contains comprehensive data on health insurance claims and hospital discharges. The SNDS is a medico-administrative tool with several linked databases. It includes demographic data as well as exhaustive data on the medical reimbursements of all individuals with universal social security coverage. The SNDS includes the National Hospital Discharge Database [Programme de médicalisation des systèmes d’information (PMSI)], gathering all data on public and private hospital stays. The PMSI provides data such as diagnoses based on the ICD-10 codes, hospital deaths, and information regarding procedures performed during the hospital stay, coded according to the French classification of medical procedures (Classification Commune des Actes Médicaux). Patients were considered to be on treatment at inclusion if there was at least one reimbursement for the evaluated therapeutic class during 2019. The same methodology was used to define treatments received during follow-up. General practitioner (GP) consultations were defined as reimbursements for a GP, internist, or geriatrician consultation. Cardiologist visits were identified based on reimbursements for cardiologist consultations, echocardiograms, cardiology day hospitalizations, and follow-up care in cardiac rehabilitation units.

Groups based on predefined healthcare administrative criteria

Risk groups were defined using simple healthcare administrative criteria, allowing for risk stratification by both medical and non-medical teams involved in patient care, without the need for additional tests. The cohort was divided into four groups based on hierarchical criteria (see Supplementary data online, Figure S1), namely (i) HFH within the past year, (ii) HFH between 1 and 5 years ago, and the presence or absence of loop diuretics. This resulted in four groups: (i) HFH within the past year (HFH ≤ 1y), (ii) HFH more than a year ago (HFH > 1y), (iii) not hospitalized within the past 5 years with loop diuretics (NoHFH/LD+), and (iv) not hospitalized within the past 5 years without loop diuretics (NoHFH/LD−).

Outcomes

The assessed primary outcomes were all-cause mortality (ACM), HFH, and the combined outcome of HFH or ACM from 1 January 2020, to 31 December 2022. The definition of HFH during follow-up was consistent with the criteria used for defining the study population, i.e. restricted to principal diagnosis codes beginning with I50, I11, I13, or J81, as per the ICD-10.

Statistical analysis

Categorical variables are expressed as frequencies (percentages) and compared using the chi-square test. Continuous variables are expressed as medians with interquartile ranges (25–75) or means (±standard deviation). Given the large sample size, differences between groups were assessed using the absolute standardized mean difference (ASMD). Absolute standardized mean difference is interpreted as an indicator of the balance between groups, with a value below 0.1 typically considered acceptable for well-balanced groups.⁷

To account for variable follow-up durations, the assessment of healthcare utilization and treatment follow-up was calculated in event-years, by dividing the number of events by the follow-up time in years (the event-year value differs from the raw number only for patients who died during the year). To facilitate the interpretation of the results expressed as medians, the analyses are presented for the entire population and then excluding patients with a count of zero.

Survival analyses were initiated with 1 January 2020, as the baseline (time zero). Survival probabilities are illustrated using Kaplan–Meier curves, compared with log-rank tests. Cox proportional hazards regression models, adjusted for age, sex, social indicators, diabetes, chronic respiratory diseases, coronary artery disease, liver or pancreatic diseases, end-stage renal disease, cancer, atrial fibrillation, and treatments [beta-blockers, angiotensin-converting enzyme (ACE) inhibitors/angiotensin receptor blockers (ARB), digoxin, ivabradine, mineralocorticoid receptor antagonists (MRA)], were used to assess associations between severity criteria groups and long-term prognosis. To assess the added predictive value of severity groups on top of adjustment variables, the C-index of a baseline model was calculated including all the adjustment variables and the increase in C-index when adding severity groups to the model.

In addition, the cohort was characterized based on the number of cardiology consultations in 2019 (0, 1, 2–3, or 4 or more consultations, corresponding to quartiles of consultation frequency). Logistic regression was performed to identify factors associated with cardiology follow-up. Subsequently, two analyses were performed to assess the relationship between cardiology follow-up in 2019 and outcomes observed after 1 January 2020. Consultations from 2019 were used to mitigate survival bias in the analysis, as patients must be alive to attend a consultation. The first analysis included the entire population, incorporating patients with de novo HF diagnosed in 2019, who consequently had <1 year of follow-up. In this analysis, the absolute risk reduction (ARR) of ACM was assessed at 1 year based on the number of consultations in 2019 (0, 1, 2–3, or 4 or more consultations, corresponding to quartiles of the number of consultations) within each group based on predefined healthcare administrative criteria. The second analysis focused on patients with a HF diagnosis prior to 2019, excluding those with de novo HF in 2019, to ensure a minimum follow-up duration of 1 year. Additionally, the number needed to consult (NNC) to prevent one modelled death within a year (1/ΔARR) was calculated. The NNC was calculated as an exploratory measure to describe the association of cardiology consultations in 2019 with 1-year ACM risk on an absolute scale. Given the observational nature of the data, these results should be interpreted as hypothesis-generating rather than as causal inferences.

The optimal allocation of cardiology follow-up visits across the different patient severity groups (NoHFH/LD−, NoHFH/LD+, HFH > 1y, HFH ≤ 1y) to reduce the modelled number of deaths ($D_{\mathrm{Reduction}}$) was calculated while adhering to specified consultation limits (from 500 000 to 2 500 000 visits). For each scenario, the total number of cardiology consultations ($C_{\mathrm{total}}$), the reduction in modelled deaths, and the ratio of consultations per modelled death avoided ($C_{\mathrm{per}\mathrm{death}}$) were calculated using the formula provided in the Supplementary material.

Additionally, the association between cardiology follow-up visits and long-term prognosis was assessed in fully adjusted Cox models (i.e. including the aforementioned adjustment variables and severity groups). The added predictive value of cardiology follow-up visits on top of all other variables was also assessed.

Ethical statement

According to French law, ethical approval was not required for this retrospective observational study as patient anonymity was fully preserved. Analyses were conducted by the Statistical Service of the Regional Directorate of the Medical Service of the National Health Insurance Fund.

Results

Patient baseline characteristics

A total of 851 304 patients met the inclusion criteria. Of the latter, 5% (n = 46 450) were excluded due to residing in nursing homes, and 18% (n = 149 130) had died by 31 December 2019. The final cohort included 655919 patients alive on 1 January 2020, with an LTC designation for HF or an HFH within the past 5 years (48% female, median age 80 years). Patient distribution was balanced between the four groups: 20.4% in HFH ≤ 1y, 27.6% in HFH > 1y, 28.3% in NoHFH/LD+, and 23.7% in NoHFH/LD− (Table 1). Predominant comorbidities included coronary artery disease (36%), atrial fibrillation (34%), diabetes (29%), and chronic respiratory diseases (25%), with prevalence increasing with HF severity (all ASMD > 0.1). At baseline, 68.9% of patients were on beta-blockers (ASMD = 0.13), 57.6% on ACE inhibitors/ARBs (including angiotensin receptor-neprilysin inhibitors) (ASMD = 0.07), and 21.5% on MRAs (ASMD = 0.19), with NoHFH/LD− having the lowest treatment proportion.

Table 1.

Baseline characteristics of the study population

	Global (n = 655 919)	NoHFH/LD− (n = 155 755)	NoHFH/LD+ (n = 185 484)	HFH > 1y (n = 180 891)	HFH ≤ 1y (n = 133 789)	ASMD mean
Demographics
Age (years)	80 (70–87)	74 (64–84)	81 (71–88)	81 (71–88)	83 (74–89)	0.3045
Female	314 215 (47.9%)	67 713 (43.5%)	91 937 (49.6%)	88 427 (48.9%)	66 138 (49.4%)	0.0630
Recipients of French healthcare assistance programs	52 549 (8.0%)	14 102 (9.1%)	13 638 (7.4%)	14 486 (8.0%)	10 323 (7.7%)	0.0328
Disadvantaged socioeconomic index	0.37 (−0.67–1.33)	0.23 (−0.82–1.22)	0.37 (−0.66–1.32)	0.41 (−0.62–1.37)	0.44 (−0.58–1.40)	0.0686
At least one GP consultation in 2019	635 299 (96.9%)	148 770 (95.5%)	182 443 (98.4%)	173 622 (96.0%)	130 464 (97.5%)	0.2033
1 annual GP consultation in 2019	15 143 (2.3%)	5413 (3.5%)	3244 (1.7%)	4327 (2.4%)	2159 (1.6%)
2–3 annual GP consultations in 2019	54 322 (8.3%)	19 216 (12.3%)	13 650 (7.4%)	15 000 (8.3%)	6456 (4.8%)
≥4 annual GP consultations in 2019	565 834 (86.3%)	124 141 (79.7%)	165 549 (89.3%)	154 295 (85.3%)	121 849 (91.1%)
At least one cardiology consultation in 2019	451 530 (68.8%)	100 528 (64.5%)	130 576 (70.4%)	121 729 (67.3%)	98 697 (73.8%)	0.2339
1 annual cardiology consultation in 2019	143 107 (21.8%)	36 551 (23.5%)	40 194 (21.7%)	41 744 (23.1%)	24 618 (18.4%)
2–3 annual cardiology consultations in 2019	169 001 (25.8%)	38 207 (24.5%)	48 502 (26.1%)	50 974 (28.2%)	31 318 (23.4%)
≥4 annual cardiology consultations in 2019	139 422 (21.3%)	25 770 (16.5%)	41 880 (22.6%)	29 011 (16.0%)	42 761 (32.0%)
Medical history
Diabetes mellitus	191 184 (29.1%)	34 315 (22.0%)	53 662 (28.9%)	58 984 (32.6%)	44 223 (33.1%)	0.1375
Chronic respiratory diseases excluding cystic fibrosis	165 856 (25.3%)	25 530 (16.4%)	44 881 (24.2%)	45 417 (25.1%)	50 028 (37.4%)	0.2460
Acute coronary syndrome	24 966 (3.8%)	9011 (5.8%)	9471 (5.1%)	1599 (0.9%)	4885 (3.7%)	0.1523
Chronic coronary disease	203 936 (31.1%)	35 583 (22.8%)	52 279 (28.2%)	63 399 (35.0%)	52 675 (39.4%)	0.2055
History of stent	85 013 (13.0%)	19 939 (12.8%)	25 385 (13.7%)	22 500 (12.4%)	17 189 (12.8%)	0.0187
History of coronary artery bypass grafting	14 802 (2.3%)	3817 (2.5%)	4242 (2.3%)	4459 (2.5%)	2284 (1.7%)	0.0283
Acute coronary syndrome or chronic coronary disease or history of stent or coronary artery bypass grafting	233 117 (35.5%)	45 359 (29.1%)	63 220 (34.1%)	66 643 (36.8%)	57 895 (43.3%)	0.1580
End-stage chronic renal failure	15 557 (2.4%)	2128 (1.4%)	4257 (2.3%)	6225 (3.4%)	2947 (2.2%)	0.0697
Liver or pancreatic diseases	29 900 (4.6%)	6045 (3.9%)	8835 (4.8%)	7032 (3.9%)	7988 (6.0%)	0.0556
Cancer	129 043 (19.7%)	27 936 (17.9%)	36 449 (19.7%)	35 374 (19.6%)	29 284 (21.9%)	0.0499
Atrial fibrillation	223 862 (34.1%)	25 749 (16.5%)	61 457 (33.1%)	57 979 (32.1%)	78 677 (58.8%)	0.4739
Cardiac resynchronization therapy	8108 (1.2%)	511 (0.3%)	1615 (0.9%)	3724 (2.1%)	2258 (1.7%)	0.0943
Implantable cardiac defibrillator	26 382 (4.0%)	3292 (2.1%)	6016 (3.2%)	11 314 (6.3%)	5760 (4.3%)	0.1146
Baseline medication
Beta-blockers	451 656 (68.9%)	94 867 (60.9%)	132 003 (71.2%)	127 578 (70.5%)	97 208 (72.7%)	0.1279
Angiotensin-converting enzyme inhibitors	271 054 (41.3%)	60 907 (39.1%)	80 982 (43.7%)	70 914 (39.2%)	58 251 (43.5%)	0.0610
Angiotensin II receptor blockers	129 862 (19.8%)	28 035 (18.0%)	38 152 (20.6%)	33 088 (18.3%)	30 587 (22.9%)	0.0700
Angiotensin-converting enzyme inhibitors or angiotensin II receptor blockers	377 649 (57.6%)	85 676 (55.0%)	111 911 (60.3%)	100 186 (55.4%)	79 876 (59.7%)	0.0686
Mineralocorticoid receptor antagonists	140 791 (21.5%)	18 869 (12.1%)	45 122 (24.3%)	41 583 (23.0%)	35 217 (26.3%)	0.1884
SGLT2 inhibitors	0 (0.0%)	0 (0.0%)	0 (0.0%)	0 (0.0%)	0 (0.0%)	0.0000
Diuretics	485 176 (74.0%)	28 884 (18.5%)	185 484 (100%)	149 565 (82.7%)	121 376 (90.7%)	1.3432
Loop diuretics	445 798 (68.0%)	0 (0.0%)	185 484 (100.0%)	141 626 (78.3%)	118 688 (88.7%)	1.3639
Digoxin	37 620 (5.7%)	4265 (2.7%)	12 939 (7.0%)	10 980 (6.1%)	9436 (7.1%)	0.1069
Ivabradine	14 339 (2.2%)	2757 (1.8%)	4593 (2.5%)	3881 (2.1%)	3108 (2.3%)	0.0265

GP, general practitioner; NoHFH/LD−, not hospitalized in the past 5 years, without loop diuretics; NoHFH/LD+, not hospitalized in the past 5 years, but using loop diuretics; HFH > 1y, heart failure hospitalization 1–5 years ago; HFH ≤ 1y, heart failure hospitalization within the past year.

Healthcare utilization during the first year of follow-up

Healthcare utilization during follow-up is presented in Table 2. Ninety-five percent of the population had at least one GP consultation without significant group differences (ASMD = 0.06). The average number of GP consultations was 10.5 per year, increasing from 8.4 to 12.7 proportionally according to severity (ASMD = 0.16).

Table 2.

Healthcare utilization during the first year of follow-up (2020)

	Global	NoHFH/LD−	NoHFH/LD+	HFH > 1y	HFH ≤ 1y	ASMD mean
Clinical follow-up
At least one general practitioner consultation, %	94.6	94.4	95.9	94.5	93.3	0.0582
Number of general practitioner consultations, median (interquartile range)b	7 (4–12)	6 (4–10)	8 (5–13)	7 (4–13)	9 (5–15)	0.2197
Number of general practitioner consultations, median (interquartile range)a,b	8 (5–13)	6 (4–10)	8 (5–13)	8 (5–13)	9 (6–15)	0.2537
At least one cardiology consultation (%)	59.2	58.5	60.3	59.3	58.4	0.0222
Number of cardiology consultations, median (interquartile range)b	1 (0–2)	1 (0–2)	1 (0–2)	1 (0–2)	1 (0–3)	0.0635
Number of cardiology consultations, median (interquartile range)a,b	2 (1–3)	2 (1–3)	2 (1–4)	2 (1–3)	2 (1–4)	0.1752
Functional tests or cardiac rehabilitation
Six-minute walk test (%)	0.6	0.4	0.6	0.5	0.9	0.0359
Cardiopulmonary exercise test (%)	3.1	5.3	2.8	2.3	2.2	0.0861
Cardiopulmonary exercise test with VO2 max (%)	0.4	0.5	0.5	0.3	0.5	0.0148
Six-minute walk test or exercise stress test with VO2 max (%)	4.1	6.0	3.8	3.1	3.5	0.0735
Cardiac rehabilitation (%)	1.5	1.7	1.6	0.8	2.0	0.0492
Biological follow-up
At least one electrolyte/creatinine test (%)	86.8	82.7	88.9	87.2	87.8	0.0928
Number of electrolyte/creatinine tests, median (interquartile range)b	3 (1–6)	2 (1–4)	3 (1–7)	3 (1–7)	4 (2–10)	0.3422
Number of electrolyte/creatinine tests, median (interquartile range)a,b	4 (2–7)	2 (1–4)	4 (2–7)	4 (2–8)	5 (3–11)	0.4128
At least one natriuretic peptide test (%)	47.0	29.6	51.2	49.7	57.7	0.2978
Number of natriuretic peptide tests, median (interquartile range)b	0 (0–2)	0 (0–1)	1 (0–2)	0 (0–2)	1 (0–3)	0.3575
Number of natriuretic peptide tests, median (interquartile range)a,b	2 (1–4)	1 (1–2)	2 (1–4)	2 (1–4)	3 (1–6)	0.3901
At least one iron panel test (%)	41.5	36.8	43.4	41.9	43.8	0.0762
Number of iron panel tests, median (interquartile range)b	0 (0–1)	0 (0–1)	0 (0–1)	0 (0–1)	0 (0–2)	0.1106
Number of iron panel tests, median (interquartile range)a,b	2 (1–3)	1 (1–2)	2 (1–3)	2 (1–3)	2 (1–3)	0.2289
Heart failure medications
Beta-blocker (%)	66.1	60.7	68.0	69.2	65.5	0.0973
Beta-blocker, number of prescriptions, median (interquartile range)a,b	11 (7–14)	10 (6–13)	11 (7–14)	11 (7–14)	11 (7–15)	0.1403
ACE inhibitors or ARBs (%)	52.6	53.3	54.5	53.1	48.6	0.0596
ACE inhibitors or ARBs, number of prescriptions, median (interquartile range)a,b	10 (6–13)	8 (5–12)	10 (6–13)	10 (6–13)	10 (6–13)	0.1482
MRAs (%)	19.9	12.1	22.3	22.4	22.4	0.1371
MRAs, number of prescriptions, median (interquartile range)a,b	7 (5–11)	7 (4–11)	8 (5–11)	8 (5–11)	7 (4–11)	0.0513
SGLT2 inhibitors (%)	0.3	0.2	0.3	0.3	0.3	0.0076
SGLT2 inhibitors, number of prescriptions, median (interquartile range)a,b	3 (1–4)	3 (1–5)	3 (1–4)	3 (1–5)	2 (1–4)	0.0836
Loop diuretics (%)	64.5	10.3	87.0	75.5%	81.6	1.1323
Loop diuretics, number of prescriptions, median (interquartile range)a,b	12 (9–19)	5 (2–10)	12 (10–17)	12 (10–20)	13 (10–21)	0.6773

NoHFH/LD−, not hospitalized in the past 5 years, without loop diuretics; NoHFH/LD+, not hospitalized in the past 5 years, but using loop diuretics; HFH > 1y, HF hospitalization 1–5 years ago; HFH ≤ 1y, HF hospitalization within the past year; ACE, angiotensin-converting enzyme; ARB, angiotensin II receptor blocker; MRA, mineralocorticoid receptor antagonist.

^aExcluding patients with a count of zero.

^bCount-per-year, calculated by dividing the number by the follow-up time in years (the count-per-year differs from the raw count only for patients who died during the year)

Although cardiology follow-up differed between groups in 2019 (Table 1), with a higher proportion of patients having at least one consultation in the order HFH ≤ 1y > NoHFH/LD+> HFH > 1y > NoHFH/LD− (ASMD = 0.23), this disparity appears to diminish in 2020 (Table 2). Overall, 59% of patients had a cardiology follow-up during the first year (2020). The proportion and median number of consultations (2 per patient) were similar across groups (ASMD: 0.02 and 0.06, respectively). Among those with at least one consultation, follow-up slightly increased with severity, ranging from 2 (1–3) visits per year in NoHFH/LD− to 2 (1–4) in HFH ≤ 1y (ASMD = 0.18).

Biomarker and functional testing, including the 6-min walk test, cardiopulmonary exercise test, and cardiac rehabilitation, were consistent across groups (ASMD < 0.1), except for natriuretic peptide testing. Respectively, 29.6%, 51.2%, 49.7%, and 57.7% of NoHFH/LD−, NoHFH/LD+, HFH > 1y, and HFH ≤ 1y had at least one natriuretic peptide test during follow-up (ASMD = 0.30). Among those tested, the number of tests increased with severity, from 2.4 to 5.4 tests per year (ASMD = 0.21).

Heart failure medication during the first year of follow-up

Beta-blockers, renin–angiotensin system (RAS) inhibitors, MRAs, or sodium-glucose co-transporter 2 inhibitors (SGLT2i) were used by 66.1%, 52.6%, 19.9%, and 0.3% of the cohort, respectively, without significant group differences (ASMD < 0.1), except for MRAs, which were less prescribed in NoHFH/LD− (12% vs. 22% in other groups, ASMD = 0.14). Except for SGLT2i (not prescribed in 2019), the proportion of patients on beta-blockers, RAS inhibitors, and MRAs decreased slightly during the first year of follow-up, mainly in the HFH ≤ 1y group.

Prognosis of HF based on the simple French-DataHF risk stratification

The risk of HFH and ACM is presented in Table 3 and Figure 1. In this national cohort of HF patients, the ACM rate was 15.9% (104 438 patients) at 1 year and 26.7% (174 821 patients) at 2 years, with significant variability across severity groups. Specifically, 1-year mortality rates were 8.0%, 14.8%, 17.2%, and 25.0% (ASMD = 0.25), and 2-year mortality rates were 13.4%, 25.6%, 29.7%, and 39.3% (ASMD = 0.32) in the NoHFH/LD−, NoHFH/LD+, HFH > 1y, and HFH ≤ 1y groups, respectively.

Table 3.

Outcomes of interest during follow-up

	Global	NoHFH/LD−	NoHFH/LD+	HFH > 1y	HFH ≤ 1y	ASMD mean
All-cause mortality
All-cause mortality at 1 year	104 438 (15.9%)	12 490 (8.0%)	27 406 (14.8%)	31 105 (17.2%)	33 437 (25.0%)	0.2465
All-cause mortality at 2 years	174 821 (26.7%)	20 922 (13.4%)	47 570 (25.6%)	53 796 (29.7%)	52 533 (39.3%)	0.3194
Heart failure hospitalization
Heart failure hospitalization at 1 year	67 713 (10.3%)	3960 (2.5%)	15 779 (8.5%)	20 833 (11.5%)	27 141 (20.3%)	0.3139
Heart failure hospitalization at 2 years	106 638 (16.3%)	7199 (4.6%)	26 518 (14.3%)	34 456 (19.0%)	38 465 (28.8%)	0.3652
All-cause mortality or heart failure hospitalization
All-cause mortality or heart failure hospitalization at 1 year	148 847 (22.7%)	15 403 (9.9%)	38 064 (20.5%)	45 049 (24.9%)	50 331 (37.6%)	0.3596
All-cause mortality or heart failure hospitalization at 2 years	233 187 (35.6%)	25 793 (16.6%)	62 890 (33.9%)	72 885 (40.3%)	71 619 (53.5%)	0.4329

NoHFH/LD−, not hospitalized in the past 5 years, without loop diuretics; NoHFH/LD+, not hospitalized in the past 5 years, but using loop diuretics; HFH > 1y, HF hospitalization 1–5 years ago; HFH ≤ 1y, HF hospitalization within the past year.

Figure 1.

Cumulative incidence curves of (A) all-cause mortality, (B) first heart failure hospitalization, and (C) first heart failure hospitalization or all-cause mortality

Rehospitalization rates and the combined outcome of ACM and HFH varied from 2.5% to 20.3% (ASMD = 0.31) for HFH and from 9.9% to 37.6% (ASMD = 0.36) for ACM or HFH at 1 year, and from 4.6% to 28.8% (ASMD = 0.37) for HFH and from 16.6% to 53.5% (ASMD = 0.43) for ACM or HFH at 2 years according to the same hierarchical order among the groups.

As illustrated in Figure 1, classification by severity level provided effective stratification of the risk of HFH or ACM.

Using the NoHFH/LD− group as reference, the NoHFH/LD+ [adjusted hazard ratio (aHR) 1.57, 95% confidence interval (CI) 1.55–1.60], HFH > 1y [aHR 1.82 (1.80–1.85)], and HFH ≤ 1y [aHR 2.24 (2.20–2.27)] groups were associated with an increased risk of mortality (all P < .0001) (Table 4). Similarly, the risk of HFH increased progressively according to severity strata, with HRs of 2.54, 3.46, and 4.80. The risk of the combined outcome of ACM or HFH also increased, with HRs of 1.74, 2.11, and 2.70 in the NoHFH/LD+, HFH > 1y, and HFH ≤ 1y groups, respectively.

Table 4.

Adjusted hazard ratios for all-cause mortality, first hospitalization for heart failure, and combined outcome of all-cause mortality or first hospitalization for heart failure

	Model 1 including HFH/LD categorization		Model 2 including HFH/LD categorization and number of cardiology consultations
	HR (95% CI)	P-value	HR (95% CI)	P-value
All-cause mortality
1 cardiology consultation vs. 0a			0.76 (0.75–0.77)	<.0001
2–3 cardiology consultations vs. 0a			0.69 (0.68–0.69)	<.0001
≥4 cardiology consultations vs. 0a			0.62 (0.62–0.63)	<.0001
NoHFH/LD + vs. NoHFH/LD−	1.57 (1.55–1.60)	<.0001	1.61 (1.59–1.64)	<.0001
HFH > 1y vs. NoHFH/LD−	1.82 (1.80–1.85)	<.0001	1.83 (1.81–1.86)	<.0001
HFH ≤ 1y vs. NoHFH/LD−	2.24 (2.20–2.27)	<.0001	2.32 (2.29–2.36)	<.0001
	C-index of Model 1	0.7129 (0.7120–0.7139)	C-index of model 2	0.7186 (0.7176–0.7196)
	Delta C-index on top of Model 0	0.0099 (0.0096–0.0103) P < .0001	Delta C-index on top of Model 1	0.0057 (0.0054–0.0059) P < .0001
First HF hospitalization
1 cardiology consultation vs. 0a			1.01 (0.99–1.03)	.19
2–3 cardiology consultations vs. 0a			1.04 (1.03–1.06)	<.0001
≥4 cardiology consultations vs. 0a			1.02 (1.01–1.04)	.006
NoHFH/LD + vs. NoHFH/LD−	2.54 (2.48–2.59)	<.0001	2.53 (2.48–2.59)	<.0001
HFH > 1y vs. NoHFH/LD−	3.46 (3.39–3.54)	<.0001	3.46 (3.38–3.54)	<.0001
HFH ≤ 1y vs. NoHFH/LD−	4.80 (4.69–4.91)	<.0001	4.79 (4.68–4.90)	<.0001
	C-index of Model 1	0.7074 (0.7061–0.7087)	C-index of model 2	0.7074 (0.7061–0.7087)
	Delta C-index on top of Model 0	0.0376 (0.0367 to 0.0385) P < .0001	Delta C-index on top of Model 1	0.0000 (0.0000 to 0.0001) P = .012
All-cause mortality or first HF hospitalization
1 cardiology consultation vs. 0a			0.82 (0.81–0.82)	<.0001
2–3 cardiology consultations vs. 0a			0.77 (0.77–0.78)	<.0001
≥4 cardiology consultations vs. 0a			0.73 (0.72–0.74)	<.0001
NoHFH/LD + vs. NoHFH/LD−	1.74 (1.72–1.76)	<.0001	1.77 (1.75–1.79)	<.0001
HFH > 1y vs. NoHFH/LD−	2.11 (2.08–2.14)	<.0001	2.12 (2.09–2.14)	<.0001
HFH ≤ 1y vs. NoHFH/LD−	2.70 (2.67–2.74)	<.0001	2.77 (2.73–2.81)	<.0001
	C-index of Model 1	0.6981 (0.6971–0.6990)	C-index of model 2	0.7009 (0.7000–0.7018)
	Delta C-index on top of Model 0	0.0172 (0.0168 to 0.0176) P < .0001	Delta C-index on top of Model 1	0.0028 (0.0027 to 0.0030) P < .0001

Adjusted for age, sex, social indicators, diabetes, chronic respiratory diseases (excluding cystic fibrosis), coronary artery disease, liver or pancreatic diseases, chronic renal failure, cancer, atrial fibrillation, and the use of beta-blockers, angiotensin-converting enzyme inhibitors, angiotensin II receptor blockers, digoxin, ivabradine, and mineralocorticoid receptor antagonists.

The C-index for Model 0, which included only adjustment variables, was 0.7030 (95% CI: 0.7020–0.7040) for all-cause mortality, 0.6698 (95% CI: 0.6684–0.6712) for first hospitalization for heart failure (HFH), and 0.6809 (95% CI: 0.6799–0.6818) for the combined outcome of all-cause mortality or first HFH.

^aCardiology consultations from 2019 were analysed, with 1 January 2020, serving as the baseline (time zero) for the survival analysis.

The C-index for the baseline model (Model 0, which included only adjustment variables) was 0.7030 (95% CI: 0.7020–0.7040), 0.6698 (95% CI: 0.6684–0.6712), and 0.6809 (95% CI: 0.6799–0.6818) for ACM, first HFH, and the combined outcome of ACM or first HFH, respectively. The added value of the incorporated French-DataHF risk stratification improved discrimination on top of the baseline model, mainly for HFH [ΔC-index of Model 1 compared to Model 0: 0.0099 (95% CI 0.0096–0.0103), P < .0001 for ACM, 0.0376 (95% CI 0.0367–0.0385), P < .0001 for first HFH, and 0.0172 (95% CI 0.0168–0.0176), P < .0001 for the combined outcome].

Association of cardiological follow-up with 1-year mortality

In 2019, a total of 1 687 750 cardiology consultations were conducted, averaging 2.57 consultations per patient. Baseline characteristics of the study population according to the annual number of cardiology visits and parameters associated with cardiology follow-up are presented in Supplementary data online, Tables S1 and S2. Briefly, access to a cardiologist was only modestly influenced by severity strata, including hospitalization [odds ratio (OR) ranging from 1.02 to 1.09]. Using patients under 60 years of age as reference group, access to a cardiologist significantly decreased only after the age of 90 [OR 0.58 (95% CI 0.56–0.59), P < .0001]. Women also had more limited access [OR 0.84 (95% CI 0.83–0.86), P < .0001]. The association of comorbidities with cardiology visits varied: cancer and end-stage chronic renal failure were associated with increased cardiology consultations (OR 1.12 and 1.68, respectively), whereas diabetes and respiratory diseases were associated with slightly reduced access (OR 0.90 and 0.97, respectively).

The prescription of guideline-directed medical therapy, including beta-blockers, RAS inhibitors, and MRAs, increased significantly with the number of cardiology consultations.

Using the no cardiology consultation group as reference, having cardiology consultations in 2019 (with 1 January 2020, as the baseline for survival analyses) was associated with a significant reduction in ACM [aHR 0.76 (95% CI 0.75–0.77), 0.69 (95% CI 0.68–0.69), and 0.62 (95% CI 0.62–0.63) for 1, 2–3, and ≥4 consultations, respectively; all P < .0001] and in the combined outcome of ACM or first HFH [aHR: 0.82 (95% CI 0.81–0.82), 0.77 (95% CI 0.77–0.78), and 0.73 (95% CI 0.72–0.74) for 1, 2–3, and ≥4 consultations, respectively; all P < .0001]. The risk of first HFH was marginally increased, with an aHR ranging from 1.01 to 1.04, regardless of the number of cardiology consultations. Including the number of cardiology consultations on top of other covariates significantly improved the model’s performance, as reflected by a significant ΔC-index [0.0057 (95% CI 0.0054–0.0059), P < .0001 for ACM and 0.0028 (95% CI 0.0027–0.0030), P < .0001 for the combined outcome].

The 1-year absolute risk of all-cause mortality for patients without a cardiologist consultation in the previous year was 13.0%, 21.3%, 24.8%, and 34.3% in the NoHFH/LD−, NoHFH/LD+, HFH > 1y, and HFH ≤ 1y groups, respectively (Figure 2). The ARR of 1-year mortality from having one cardiologist consultation (compared to none) was generally consistent across the four severity groups, with ARR ranging from 6.3% in the NoHFH/LD− group to 9.2% in the HFH > 1y group, corresponding to an NNC of 11 to 16 patients to reduce one modelled death within 1 year. The number of patients requiring additional cardiology consultations was inversely proportional to severity, with an NNC of 20 (for 2–3 consultations) and 26 (for ≥4 consultations) in the HFH ≤ 1y group, compared to 55 (for 2–3 consultations) and 112 (for ≥4 consultations) in the NoHFH/LD− group. In the sensitivity analysis focusing on the 462 092 patients (70.4%) with HF diagnosed before 2019 and still alive on 1 January 2020, results were fairly identical and are presented in Supplementary data online, Figure S2.

Figure 2.

Association between cardiology consultations in 2019 and 1-year absolute risk of all-cause mortality after 1 January 2020. This figure illustrates the 1-year all-cause mortality risk (%) in heart failure (HF) patients across different clinical groups based on their history of HF hospitalization and loop diuretics use (NoHFH/LD−, NoHFH/LD+, HFH > 1y, and HFHs ≤ 1y). The mortality risk is shown for patients receiving 0, 1, 2–3, or ≥4 cardiologist visits. The coloured bars represent the percentage risk of 1-year mortality for each group, while the arrows indicate the number of patients requiring an incremental number of cardiology consultations to reduce one modelled death within 1 year. Example: In the ‘NoHFH/LD−’ group, the risk of 1-year all-cause mortality is 13.0% for those with no annual cardiologist visit. However, for patients who had one cardiologist visit, the mortality risk drops to 6.7%, meaning that 15.9 patients would need to be seen once to reduce one modelled death within 1 year. To reduce an additional modelled death within 1 year, 55.4 patients would need 2 or 3 cardiologist visits (one or two additional visits compared to the previous step). NoHFH/LD−, not hospitalized in the past 5 years, without loop diuretics, NoHFH/LD+, not hospitalized in the past 5 years, but using loop diuretics, HFH > 1y, HF hospitalization 1–5 years ago, HFH ≤ 1y, HF hospitalization within the past year

Based on the 1-year absolute risk of death, the number of modelled deaths without cardiology consultations was estimated at 150 507 patients. The modelled number of deaths expected with one consultation per patient was 101,807, closely aligning with the observed number of 1-year deaths (104 438) in the population with a median of one consultation per group (Table 5). According to the model, the optimal follow-up strategy to minimize 1-year deaths, while maintaining the same total number of cardiology consultations as in 2019 (1 687 750), yielded 1 consultation for the NoHFH/LD− group, 2–3 consultations for the NoHFH/LD+ and HFH >1y groups, and 4 consultations for the HFH ≤1y group. This configuration was projected to reduce the number of modelled deaths to 80 193—a decrease of 70 314 modelled deaths compared with no cardiology consultations—with a total of 1 606 849 consultations, averaging 22.9 consultations per reduced modelled death. Table 5 outlines the optimal combinations of follow-up visits in detail.

Table 5.

Optimal annual cardiology visits combinations by severity group for minimizing modelled deaths within consultation limits in France

Modelled consultation limit	NoHFH/LD− visits per year	NoHFH/LD+ visits per year	HFH > 1y visits per year	HFH ≤ 1y visits per year	Modelled deaths	Reduction in modelled Deaths	Total consultations	Consultations per modelled death avoided
0	0	0	0	0	150 507	0	0
250 000	0	0	0	1	140 874	9633	133 789	13.9
500 000	0	0	1	1	124 232	26 275	314 680	12.0
750 000	1	1	1	1	101 807	48 700	655 919	13.5
1 000 000	1	1	1	2–3	95 117	55 390	856 603	15.5
1 250 000	1	1	1	4	89 899	60 607	1 057 286	17.4
1 500 000	1	1	2–3	4	85 015	65 492	1 328 623	20.3
1 750 000	1	2–3	2–3	4	80 193	70 314	1 606 849	22.9
2 000 000	2–3	2–3	2–3	4	77 389	73 118	1 840 481	25.2
2 250 000	1	4	4	4	73 765	76 742	2 156 411	28.1
2 500 000	2–3	4	4	4	70 961	79 545	2 390 044	30.0
	Observed NoHFH/LD− visits per year	Observed NoHFH/LD + visits per year	Observed HFH > 1y visits per year	Observed HFH ≤ 1y visits per year	Observed deaths		Observed total consultations
	1 (0–2)	1 (0–2)	1 (0–2)	1 (0–3)	104 438		1 687 750

NoHFH/LD−, not hospitalized in the past 5 years, without loop diuretics; NoHFH/LD+, not hospitalized in the past 5 years, but using loop diuretics; HFH > 1y, HF hospitalization 1–5 years ago; HFH ≤ 1y, HF hospitalization within the past year.

The optimal allocation of cardiology follow-up visits across different patient severity groups (NoHFH/LD−, NoHFH/LD+, HFH > 1y, HFH ≤ 1y) was calculated to minimize the number of modelled deaths ($D_{\mathrm{Reduction}}$) while adhering to specified consultation limits (from 500 000 to 2 500 000 visits). For each scenario, the total number of cardiology consultations ($C_{\mathrm{total}}$), the number of modelled deaths, and the ratio of consultations per modelled death avoided were calculated ($C_{\mathrm{per}\mathrm{death}}$).

$D_{\mathrm{Reduction}}=\sum _{i=1}^4(Ni\times (M_{i,\mathrm{baseline}}-M_{i,\mathrm{follow}\text{-}\mathrm{up}}))$ .

i represents each patient group (NoHFH/LD−, NoHFH/LD+, HFH > 1y, HFH ≤ 1y).

Ni is the total number of patients in group i.

$M_{i,\mathrm{baseline}}$ is the baseline mortality rate for group i without any follow-up visits.

$M_{i,\mathrm{follow}\text{-}\mathrm{up}}$ is the adjusted mortality rate for group iii after receiving the specified number of follow-up visits.

$M_{i,\mathrm{baseline}}-M_{i,\mathrm{follow}\text{-}\mathrm{up}}$ represents the reduction in mortality rate due to the follow-up visits.

$C_{\mathrm{total}}=\sum _{i=1}^4(V_i\times N_i)$ .

$V_i$ is the number of follow-up visits assigned to group i.

$N_i$ is the number of patients in group i

$C_{\mathrm{per}\mathrm{death}}={\displaystyle \frac{C_{\mathrm{total}}}{D_{\mathrm{Reduction}}}}$ .

Discussion

Analyses of this national cohort of HF patients, identified either by a recent HFH or a LTC declaration for HF, showed that: (i) stratifying the population based on the simple French-DataHF risk stratification including recent HFH and the use of loop diuretics enabled categorizing the cohort into four similarly-sized groups but with markedly different prognoses; (ii) the risk of adverse events varied significantly among the four groups, with 1-year combined rates of HFH or ACM ranging from 10% in non-hospitalized patients without diuretics to 38% in patients with a recent HFH within the past year; (iii) despite these differences in risk, follow-up care—including medications, GP consultations, and cardiology visits—was relatively consistent across the four groups; and (iv) the benefit of cardiology follow-up was dependent on the severity group and the number of annual consultations. Tailoring HF follow-up care based on the French-DataHF risk stratification would enable optimal resource utilization, maximizing the efficiency of cardiology consultations, and achieving the best ratio of consultations to reductions in mortality (Structured Graphical Abstract).

The prevalence of HF is influenced by both the region and age of the studied cohort, estimated to affect 3.4% of the adult population.^8,9 Its incidence, especially among younger individuals, is on the rise in developed countries, and HF-related mortality appears to be increasing.^10,11 Using our inclusion criteria, we identified a cohort exceeding 650 000 adult patients living with HF, corresponding to almost 1% of the global French population. These results highlight the major public health challenge posed by HF, especially due to its clinical prognosis and the associated healthcare costs. For instance, HF-related costs represent a significant healthcare expenditure, escalating from $3000 per year in the first year to $15 000 annually at 5 years in the UK, and an estimated $25 000 annually in the USA, with $16 000 attributable to HFH alone.^8,12 Hence, providing adequate medical care as a function of disease severity is crucial not only to reduce the human and societal burden of HF but also to provide the best health care for these patients.

Despite the exclusion of patients residing in nursing homes—due to their advanced prognosis and the unavailability of data on their healthcare utilization—ACM associated with HF in this unselected population was notably higher than previously reported, particularly among outpatient groups. The European Society of Cardiology Heart Failure Long-Term Registry (ESC-HF-LT), a prospective, observational registry spanning 21 European or Mediterranean countries and including over 12 000 patients, reported a 1-year mortality rate of 23.6% for acute HF and 6.4% for chronic HF.¹³ In our study, the 1-year mortality rate was 8% for outpatients not on diuretics and exceeded 15% in other groups, highlighting the high baseline risk in HF patients, even those who are considered stabilized. These results are slightly higher than those reported by Friday et al.,⁴ also taking advantage of a large, unselected set of patients, but in a younger cohort.

The ESC/HFA guidelines recognize that general follow-up care for chronic HF is under-studied and that the specific role of the cardiologist in the management of stable patients remains to be clearly defined.³ These guidelines recommend follow-up at least every 6 months, with more frequent visits advised for patients who have recently been hospitalized or are undergoing medication titration.³ In France, the French National Health Authority (HAS) empirically recommends at least one consultation for NYHA Class I HF patients, and 4 to 12 annual consultations for the most symptomatic patients.¹⁴

Our study shows that, within this national cohort, the frequency of cardiology consultations was not significantly different across the four severity groups. Notably, 40% of patients did not see a cardiologist within the first year of follow-up, even among those recently hospitalized. When considering 2019, the proportion of patients who did not see a cardiologist was 31%, mainly due to a high proportion of cardiology follow-up in the HFH ≤ 1y group (74%). In the next year, among the 60% of patients who saw a cardiologist at least once annually, the number of consultations slightly increased according to patient severity. Taken together, these data indicate that cardiology follow-up is conducted without consideration of patients’ baseline risk except for the very initial follow-up after an HFH. Furthermore, female patients, elderly individuals, and those with a higher disadvantaged socioeconomic index were less likely to receive cardiology consultations, highlighting persistent inequities in access to specialized care.

Although the use of beta-blockers, RAS inhibitors, and MRAs for HF indications was limited to HF with reduced ejection fraction (HFrEF) in the 2016 ESC/HFA guidelines, the overall prescription rate remains low, indicating significant room for improvement, particularly for MRAs, which were prescribed in only 20% of patients.¹⁵ The proportion of patients treated with beta-blockers (∼70%) and RAS inhibitors (∼60%) was homogeneous across the four groups at baseline. However, ambulatory patients without diuretics received MRA treatment half as often compared to other groups. Of note, the prescription of these three essential drug classes was positively correlated with the number of cardiology consultations in the prior year. During follow-up, there was a slight decline in the proportion of patients treated with these three drug classes, predominantly due to reduced prescriptions in the group of patients hospitalized within the past year. These results indicate a prescription rate lower than that reported in the STRONG-HF study, where over 90% of patients received beta-blockers or RAS inhibitors, but with a much lower discontinuation rate compared to certain international registries, where cessation of these drug classes ranged from 30% to 40% at 1 year.^16,17 SGLT2i prescriptions (label extension for use in HFrEF in France since March 2021 for dapagliflozin and January 2022 for empagliflozin) remained very marginal (0.3%) and consistent across all groups. Although biomarkers play a prognostic role, routine biomarker follow-up to guide therapy is not recommended.³ Nevertheless, one in two patients had at least one annual natriuretic peptide test, with significantly fewer tests among ambulatory patients without diuretics (30% vs. 50%), with the number of tests increasing with severity. In contrast to natriuretic peptides, renal function, iron status, and cardiac functional tests were prescribed uniformly across groups. Given that renal function deterioration or exercise intolerance is part of the definition of advanced HF, one would expect increased use among high-risk patients.¹⁸

To our knowledge, this is the first study in which the benefits associated with cardiological follow-up have been described in a comprehensive national population based on risk profile and the number of consultations. A clear dose–response relationship was observed between the frequency of cardiology consultations and improved survival outcomes, ranging from a 24% reduction in mortality with one annual consultation to a 38% reduction with four or more consultations in fully adjusted models. Notably, this benefit was achieved without a clinically significant increase in HF hospitalizations, highlighting the critical role of specialized follow-up care. This effect translates into one annual cardiology visit being associated with one fewer death for every 11–16 patients seen, irrespective of HF severity. However, the clinical benefits of more frequent consultations (2 or more) appeared to increase with HF severity. Based on our modelling analysis, the number of consultations for each group can be effectively optimized to maximize survival by tailoring the frequency of visits to the patient's HF severity, while maintaining the same overall number of consultations. This finding holds particular importance for public health strategies. Cardiology consultations emerge as a key factor associated with survival when adjusting for available clinical data, further exceeding the expected benefits of pharmacological therapy titration. The strong statistical association between cardiology consultations and survival reflects the integration of patients into a comprehensive care network. This includes enhanced adherence to non-pharmacological strategies and educational initiatives, alongside implementation of evidence-based pharmacological therapies. It also highlights factors such as overall health status, socioeconomic conditions, and other determinants that influence care, collectively underscoring the magnitude of this effect.

Impact for clinical practice

These real-world epidemiological data highlight both the severe prognosis associated with HF, even among outpatients, with a 1-year mortality rate of 8%–15%, and reveal suboptimal therapeutic management, as evidenced by the low prescription rate of MRAs. These findings underscore the need for actions directed at both cardiologists and GPs to emphasize the importance of guideline-directed medical therapy in HF management.

Furthermore, the present results demonstrate that the French-DataHF stratification effectively assesses the risk of ACM and HFH on a population level using two simple hierarchical criteria: the timing of the last hospitalization and treatment with loop diuretics. Diuretic treatment, a hospitalization between 1 and 5 years, and a hospitalization within the past year increased the risk of death by 60%, 80%, and 124%, respectively, after adjusting for traditional risk factors. The risk of hospitalization increased by 2.5-, 3.5-, and 4.8-fold, respectively. This stratification allows for effective medium-term risk prognostication even in the absence of other clinical and biological parameters, providing valuable guidance for cardiologists in defining the risk of outcomes and optimal follow-up frequency.

The major advantage of this classification lies in its administrative criteria, enabling the automated implementation of minimum cardiology follow-up tailored to disease severity without requiring direct medical input. With 40% of HF patients missing annual cardiology follow-ups regardless of disease severity, it is crucial from a public health perspective to ensure that all HF patients receive at least one cardiologist visit per year. This approach not only maximizes survival benefits with a limited number of consultations but also addresses the challenges posed by a shortage of cardiologists and the increasing prevalence of HF. Moreover, offering more frequent consultations—beyond one per year—provides substantial benefits, particularly for patients recently hospitalized, and to a lesser extent for those with earlier hospitalizations. This is of critical importance in light of our results, highlighting that the number of cardiology consultations is actually similar regardless of the patient profile and associated risk. This essential follow-up could be integrated into health insurance tracking systems or healthcare organizations’ reimbursement data platforms, ensuring adherence to appointments and the automatic rescheduling of missed consultations (e.g. as part of a Disease Management Program with mandatory consultations). Additionally, implementing structured referral pathways and strengthening multidisciplinary collaboration between GPs and cardiologists could help ensure that all HF patients receive appropriate specialist care. Finally, leveraging telemedicine services and patient education programs can further expand access to cardiology care, enhancing treatment outcomes, and improving patient management.

These strategies for rationalizing consultations will be essential to optimize the overall health of the population within a constrained healthcare system. Similar to the KDIGO guidelines, which recommend a specific number of annual nephrologist consultations based on glomerular filtration rate (GFR) and albuminuria levels, our optimal analysis identified that a follow-up scheme of 1, 2–3, and 4 annual consultations for the NoHFH/LD−, NoHFH/LD+, HFH > 1y, and HFH ≤ 1y groups, respectively, was the most effective strategy to minimize the modelled 1-year deaths (>69 000) within the available consultation limit (<1 700 000).¹⁹

With a rate of 92 cardiologists per million inhabitants, France has a cardiologist density comparable to the European average of 101 per million inhabitants, according to the ESC Atlas of Cardiology.²⁰ At the extremes, optimization data could be crucial for countries such as Ireland, the UK, and Germany, which have fewer than 46 cardiologists per million inhabitants and must adapt and rationalize follow-up care accordingly. Our modelling approach allows for the optimization of patient management by tailoring the latter to local epidemiology and available resources.

In the future, it will be essential to test our optimum simulation based on observational data, which assesses the benefits of incremental follow-up tailored to HF severity, in randomized clinical trials. While our modelling provides valuable insights, randomized evidence is needed to confirm these projections. A trial (likely a cluster trial) would allow determining whether adjusting the frequency of cardiology consultations based on patient risk leads to better outcomes. For pragmatic reasons, a key starting point for this initiative would be to focus on providing one annual cardiology consultation to the 40% of HF patients who currently do not receive one. This group represents a significant opportunity to improve survival outcomes with only a modest increase in the total number of consultations. Healthcare systems could formally assess the impact of this annual consultation through large-scale cluster trials, allowing for an evaluation of the broader population-level benefits of this approach.

Limitations

Our study has some limitations. Patients were identified based on an HFH or an LTC declaration for HF and were still alive on 1 January 2020. Patients who died before this date were excluded from the final analysis, leading to an underestimation of morbidity and mortality. Additionally, some HF patients who had not been recently hospitalized for HF and did not submit the LTC paperwork, as well as those with an LTC declaration for another illness who did not file additional paperwork specifically for HF because they were already receiving full reimbursement, were not included in the cohort. As this cohort is based on reimbursement data, only data on outpatient therapeutics and healthcare utilization were available. Moreover, certain clinically relevant data that could refine prognosis—such as ejection fraction, which helps characterize HF phenotype, as well as prognostic biomarker values such as glomerular filtration rate and natriuretic peptides—were unavailable. Specifically, the implementation of left ventricular ejection fraction coding is recent and currently limited to hospitalized patients, resulting in incomplete data coverage, which precluded its use in this analysis. In addition, although associated with prognosis, data on the doses of loop diuretics or HF guideline-directed medical therapy were not available.²¹ Importantly, despite adjustments for available variables in multivariable models, this study remains observational, and causality cannot be established, as is inherent to all observational studies. The definition of HFH relies almost exclusively on principal diagnoses, except in cases of pulmonary infections or chronic bronchopulmonary diseases with an associated HF diagnosis, due to the frequent overlap of these conditions. This strategy was chosen to avoid over-categorization of HF hospitalizations. Nevertheless, the approach used herein allows for estimating HF-related risk in a population using medical-administrative data commonly available during consultations, which is a strength of this real-world study. The NNC values presented in this study are derived from observational data and are subject to confounding factors, both known and unknown. Unlike the number needed to treat (NNT), which is based on randomized trial data, the NNC estimates we presented herein cannot establish causality. However, as conducting a randomized trial on cardiology follow-up would raise ethical and practical concerns, the NNC remains a useful tool for descriptive purposes.

Conclusion

In this national cohort of HF patients, the French-DataHF risk stratification effectively identified four clinically relevant groups for practical application in daily care. These groups are distinguished based on simple administrative criteria: the timing of HFH (1-year threshold) and the use of loop diuretics (yes/no). Despite significant prognostic differences between these groups, the intensity of HF management does not appear appropriately tailored to the severity of the patients’ conditions. Cardiology follow-up is essential for all patients with HF, while risk stratification based on loop diuretic use and HFH helps identify those who would benefit from more frequent consultations.

Supplementary data

Supplementary data are available at European Heart Journal online.

Declarations

Disclosure of Interest

G.B. reports personal fees from Abbott, AstraZeneca, Boehringer Ingelheim and Novartis, outside the submitted work. F.R. reports grants from Air Liquide and Abbott, outside the submitted work; consulting fees from Abbott, Air Liquid, Bayer, and Pfizer; lectures fees from AstraZeneca, Servier, Boerhinger Ingelheim, Vifor, Bayer, Novartis, Novonordisk, Air Liquid, Abbott, and QuidelOrtho; travel fees from Novartis and Boerhinger Ingelheim; participation on a data safety monitoring board or advisory board for Carmat; and part of Boehringer Ingelheim, Vifor Pharma, and Novartis board. T.D. reports grants from Pfizer outside the submitted work, consulting fees from Pfizer, lecture fees from Pfizer, and travel fees from Pfizer. N.G. reports personal fees from AstraZeneca, Bayer, Boehringer, Novartis and Vifor outside the submitted work. All other authors have nothing to disclose.

Data Availability

The data used in this study are sourced from the French National Health Data System (SNDS) and are available to authorized individuals. Access to these data is subject to approval from relevant French authorities and is restricted to qualified researchers who meet the necessary regulatory and ethical requirements.

Funding

All authors declare no funding for this contribution.

Ethical Approval

According to French law, ethical approval was not required for this retrospective observational study as patient anonymity was fully preserved.

Pre-registered Clinical Trial Number

Not applicable.