H2FPEF Score for heart failure with preserved ejection fraction

Clinical background

The diagnosis of heart failure with preserved ejection fraction (HFpEF) is particularly difficult in euvolaemic patients with exertional dyspnoea but without clear signs of congestion on examination, chest radiography or biomarkers. The reference standard is invasive haemodynamic exercise testing with right heart catheterisation, in which heart failure is defined as a resting pulmonary capillary wedge pressure (PCWP) ≥15 mmHg or an exercise PCWP ≥25 mmHg [1]. Because catheterisation is resource-intensive and cannot be offered to every patient with unexplained dyspnoea, a non-invasive tool is needed that can estimate the probability of HFpEF and thereby guide the decision on further investigation. The H₂FPEF score was developed to fill this gap and to replace earlier expert consensus algorithms that lacked empirical validation [1].

Calculating the H₂FPEF score

The score is a weighted additive model with six variables, drawn from the history, the medication list and standard echocardiography:

$$\text{H}$$2\text{FPEF} = 2 \cdot H{\text{BMI}} + 1 \cdot H_{\text{med}} + 3 \cdot F + 1 \cdot P + 1 \cdot E + 1 \cdot F_{\text{fill}}

where:

• $H_{\text{BMI}} = 1$ if body mass index >30 kg/m² (Heavy), otherwise 0
• $H_{\text{med}} = 1$ if ≥2 antihypertensive drugs (Hypertensive), otherwise 0
• $F = 1$ if atrial fibrillation (paroxysmal or persistent), otherwise 0
• $P = 1$ if echocardiographically estimated pulmonary artery systolic pressure >35 mmHg (Pulmonary hypertension), otherwise 0
• $E = 1$ if age >60 years (Elder), otherwise 0
• $F_{\text{fill}} = 1$ if the echocardiographic E/e' ratio >9 (Filling pressure), otherwise 0

The score ranges from 0 to 9. Atrial fibrillation carries the greatest weight (3 points), obesity gives 2 points, and the remaining variables give 1 point each. The weighting is based on the beta coefficients of the multivariable logistic regression in the derivation cohort [1].

The derivation cohort consisted of 414 consecutive patients referred to the Mayo Clinic, Rochester, for invasive exercise haemodynamics because of unexplained dyspnoea between 2006 and 2016 [1]. Of these, 267 had HFpEF and 147 served as controls with normal haemodynamic values at rest and during exercise, giving an HFpEF prevalence of 64%. Exclusion criteria included ejection fraction <50%, significant valvular disease, pulmonary arterial hypertension, constrictive pericarditis, primary cardiomyopathy and cardiac transplant recipients [1]. A separate test cohort of 100 consecutive patients (61 with HFpEF) was used for internal validation [1].

Interpretation in practice

The score translates into three probability bands with differing clinical action:

The probability rises stepwise with the score: the odds of HFpEF double for each point (OR 1.98, 95% CI 1.73–2.30) [1]. NT-proBNP adds no further discriminatory information beyond the score and should not be added as a seventh step [1].

Validation and performance

In the derivation cohort an AUC of 0.841 (95% CI 0.802–0.881) was achieved, with performance maintained in the independent test cohort (AUC 0.886) [1]. Bootstrap validation with 1000 replicates gave an optimism-corrected AUC of 0.838, indicating little overfitting [1]. Calibration was good, with close agreement between model-predicted and observed HFpEF prevalence at each score value (Hosmer–Lemeshow p >0.1) [1]. The score significantly outperformed both the 2007 and the 2016 ESC consensus algorithms (AUC difference +0.169 and +0.173 respectively, both p <0.0001) [1].

A multicentre validation published in 2022 included 736 patients from six centres in the USA, the Netherlands, Denmark and Australia, with invasive exercise haemodynamics as the reference standard [2]. HFpEF prevalence was 76%. H₂FPEF achieved an AUC of 0.845 (95% CI 0.810–0.875) and was significantly superior to the HFA-PEFF algorithm (AUC 0.710, difference −0.134, p <0.001) [2]. The false negative rate for low scores was 25% for H₂FPEF compared with 55% for HFA-PEFF [2].

A systematic review and meta-analysis from 2026 pooled ten studies and found a pooled sensitivity of 0.76 (95% CI 0.56–0.87) and a pooled specificity of 0.72 (95% CI 0.59–0.82) for H₂FPEF, with AUC values from 0.74 to 0.886 [3]. Heterogeneity was substantial (I² = 84% for sensitivity). A key finding was that the prevalence of atrial fibrillation in the study population explained 89.3% of the variance in sensitivity: in populations with an atrial fibrillation prevalence above 50% sensitivity exceeded 99%, whereas in populations with a lower atrial fibrillation prevalence it fell to 58–84% [3]. This reflects the fact that atrial fibrillation is the most heavily weighted variable in the score. Studies using invasive haemodynamics as the reference standard showed a higher AUC (0.85) than studies using a clinical diagnosis as the reference (0.76) [3].

In a Japanese prospective cohort study of 356 stable outpatients with cardiovascular risk factors (mean age 73.2 years, mean score 3.1), the H₂FPEF score predicted future heart failure events (hospitalisation or cardiovascular death) over a mean follow-up of 517 days, with an AUC of 0.77–0.78 [4]. This shows that the score also has prognostic value, even though it was developed primarily as a diagnostic tool.

Limitations

The score applies only to patients with preserved ejection fraction (≥50%) and unexplained exertional dyspnoea. It should not be used in patients with HFrEF, significant valvular disease, pulmonary arterial hypertension, constrictive pericarditis or primary cardiomyopathy, since these conditions were excluded from the derivation cohort [1].

All variables are dichotomised, which entails a loss of information. A patient with a BMI of 31 receives the same 2 points as one with a BMI of 45, and a patient with an E/e' ratio of 9.1 receives the same 1 point as one with an E/e' of 20. The continuous version of the model performed marginally better (AUC difference +0.022, p = 0.03) but is less clinically tractable [1].

The echocardiographic variables (PASP and E/e') are operator-dependent and may vary between examiners and machines. PASP is estimated indirectly from the tricuspid regurgitation velocity and can be difficult to measure in patients without appreciable regurgitation. The E/e' ratio has limited reliability in atrial fibrillation and in resistant hypertension with left ventricular hypertrophy.

The considerable heterogeneity in sensitivity between studies, driven by atrial fibrillation prevalence, means that the score performs best in populations in which atrial fibrillation is common [3]. In populations with a low atrial fibrillation prevalence, sensitivity falls appreciably, which increases the risk of false negative results.

Intermediate scores (2–5) constitute the greatest problem area: here the score can neither confirm nor exclude HFpEF with sufficient certainty, and further objective testing is indicated [1]. Treating an intermediate score as though it were diagnostic is the commonest error in clinical use.

References

1. Reddy YNV, Carter RE, Obokata M, Redfield MM, Borlaug BA. A Simple, Evidence-Based Approach to Help Guide Diagnosis of Heart Failure With Preserved Ejection Fraction. Circulation. 2018;138(9):861–870. PMID: 29792299
2. Reddy YNV, Kaye DM, Handoko ML, et al. Diagnosis of Heart Failure With Preserved Ejection Fraction Among Patients With Unexplained Dyspnea. JAMA Cardiology. 2022;7(9):891–899. PMID: 35830183
3. Estes-Schmalzl KJ, Wolden M, Lefebvre KM. Diagnostic Accuracy of H₂FPEF and HFA-PEFF Algorithms for Heart Failure with Preserved Ejection Fraction (HFpEF): A Systematic Review and Meta-Analysis. Journal of Clinical Practice and Research. 2026;48(1):9–18. PMID: 41908354
4. Suzuki S, Kaikita K, Yamamoto E, et al. H₂FPEF Score for Predicting Future Heart Failure in Stable Outpatients With Cardiovascular Risk Factors. ESC Heart Failure. 2020;7(1):65–74. PMID: 31967406