Abstract
Heart failure with preserved ejection fraction(HFpEF) is underrecognized in clinical practice. Although a previously developed risk score, termed H2FPEF, can be used to estimate HFpEF probability, this score requires imaging data, which is often unavailable. Here we sought to develop a HFpEF screening model that is based exclusively upon clinical variables and that can guide the need for echocardiography and further testing. In a derivation cohort (n = 414, 249 women), a clinical model using age, body mass index (BMI) and history of atrial fibrillation (termed the HFpEF-ABA score) showed good discrimination (AUC 0.839(95% CI 0.800–0.877), p<0.0001). The performance of the model was validated in an international, multicenter, cohort (n=736, 443 women; AUC 0.813(95% CI 0.779–0.847), p<0.0001) and further validated in two additional cohorts: a cohort including patients with unexplained dyspnea (n=228, 136 women; AUC 0.840(95% CI 0.782–0.900), p<0.0001) and a cohort for which HF hospitalization was used instead of hemodynamics to establish HFpEF diagnosis (n=456, 272 women; AUC 0.929(95% CI 0.909–0.948), p<0.0001]. Model-based probabilities were also associated with increased risk of HF hospitalization and/or death among patients from the Mayo Clinic (n=790) and a US national cohort across the Veteran Affairs Health system (n=3076, 110 women). Using the HFpEF-ABA score, rapid and efficient screening for risk of undiagnosed HFpEF can be performed in patients with dyspnea using only age, BMI and history of AF.
Keywords: heart failure, exercise hemodynamics
Introduction
Diagnosis of heart failure (HF) with preserved ejection fraction (HFpEF) is challenging, contributing to underrecognition in everyday practice1,2 along with delays in appropriate referral, diagnosis and treatment.3,4 Easily applicable, validated screening models can enhance detection, which is even more important given the recent identification of effective therapies.5–8 Probabilistic scoring systems have been developed but require echocardiography,9,10 which is not available when patients initially seek medical care from most non-cardiologists for dyspnea.
The H2FPEF and HFA-PEFF scores were developed to estimate the probability of HFpEF in patients with dyspnea using a combination of echocardiographic and clinical variables.9–11 While imaging plays an essential role in the diagnostic evaluation for HF, echocardiography measures accounted for only 2 out of 9 possible total points in estimation of HFpEF probability using the H2FPEF score. This led us to hypothesize that HFpEF likelihood might be reasonably estimated at the initial point of care using only clinical data, which could then help guide echocardiography and further testing.
To test this hypothesis, we derived a model based on clinical variables, without imaging data, using HFpEF case definitions established definitively by invasive testing. We then validated this algorithm in an international, multicenter cohort of patients presenting with unexplained dyspnea across 3 continents undergoing the same invasive evaluation, a separate secondary validation cohort, and a third cohort in whom the diagnosis of HFpEF was alternatively established by hospitalization for volume overload. To provide complementary value and validate against a different HF endpoints aside from diagnosis alone, we also determined whether higher model-based probability could predict future risk of adverse HF outcomes in 2 separate cohorts.
Results
Baseline characteristics of derivation cohort
The derivation cohort included 414 patients from one center (65% HFpEF, 35% non-cardiac dyspnea, NCD). Compared to those with NCD, patients with HFpEF were older and more obese, with a greater prevalence of hypertension and diabetes (Table 1). Atrial fibrillation (AF) was overall more common in HFpEF for both paroxysmal and permanent/persistent AF. Natriuretic peptides were on average elevated in the HFpEF group compared to NCD. There were more echocardiographic abnormalities in the HFpEF group including higher estimated right ventricular systolic pressure, LA volume and E/e’ as is typical of HFpEF (Table 1).
Table 1:
Baseline characteristics of the derivation and validation cohorts
| Derivation Cohort | Primary Validation Cohort | Hospitalized HFpEF validation cohort | ||||||
|---|---|---|---|---|---|---|---|---|
| Non-cardiac dyspnea (n=147) |
Ambulatory HFpEF (n=267) |
P value | Non-cardiac dyspnea (n=173) |
Ambulatory HFpEF (n=563) | P value | Hospitalized HFpEF (n=456) |
P value# | |
| Age | 56.1 ± 14.6 | 68.0 ± 10.8 | <0.0001 | 60.3 ± 14.5 | 69.1 ± 10.5 | <0.0001 | 78.1 ± 12.5 | <0.0001 |
| Female, % | 59 | 61 | 0.6131 | 63 | 59 | 0.4245 | 60 | 0.4407 |
| BMI, kg/m2 | 28.2 ± 5.4 | 33.0 ± 7.4 | <0.0001 | 27.9 ± 5.8 | 32.4 ± 7.2 | <0.0001 | 34.9 ± 10.3 | <0.0001 |
| Diabetes, % | 12 | 28 | 0.0001 | 14 | 19 | 0.1489 | 27 | 0.0009 |
| Hypertension, % | 53 | 86 | <0.0001 | 64 | 82 | <0.0001 | 95 | <0.0001 |
| AF,% (no/parox/perm) | 95.9/3.4/0.7 | 65.6/17.2/17.2 | <0.0001 | 92/6/2 | 62/22/15 | <0.0001 | 41/20/39 | <0.0001 |
| Race, White/Black/Other% | 97/2/1 | 98/1/1 | 0.2454 | 94/3/3 | 94/4/2 | 0.7972 | 96/2/2 | 0.4805 |
| NT-proBNP | 122 [52–259] | 384 [131–1111] | <0.0001 | 83 [50–198] | 269 [99–869] | <0.0001 | 2841 [1205, 5594] | <0.0001 |
| Creatinine | 0.96 ± 0.22 | 1.13 ± 0.40 | <0.0001 | 0.99 ± 0.32 | 1.12 ± 0.62 | <0.0001 | 1.42 ± 0.80 | <0.0001 |
| LV Ejection Fraction | 63 ± 5 | 63 ± 5 | 0.7030 | 62 ± 6 | 61 ± 6 | 0.0317 | 62 ± 6 | 0.5321 |
| LVEDD, mm | 47.5 ± 4.7 | 48.3 ± 5.2 | 0.1321 | 47.1 ± 5.5 | 48.0 ± 6.6 | 0.0886 | 48.5 ± 6.2 | 0.0098 |
| LAVI, ml/m2 | 28.6 ± 9.4 | 38.0 ± 13.8 | <0.0001 | 32.1 ± 12.0 | 36.6 ± 13.7 | <0.0001 | 45.6 ± 14.3 | <0.0001 |
| Septal e’ velocity | 0.08 ± 0.03 | 0.07 ± 0.02 | <0.0001 | 0.08 ± 0.03 | 0.07 ± 0.02 | <0.0001 | 0.06 ± 0.02 | <0.0001 |
| E/e’ | 9.5 ± 3.9 | 14.0 ± 6.5 | <0.0001 | 9.3 ± 3.8 | 12.9 ± 6.5 | <0.0001 | 20.2 ± 10.6 | <0.0001 |
| RVSP, mmHg | 29.7 ± 5.3 | 37.8 ± 12.2 | <0.0001 | 29.6 ± 10.9 | 38.5 ± 15.1 | <0.0001 | 46.2 ± 13.9 | <0.0001 |
| Beta blocker, % | 29 | 55 | <0.0001 | 28 | 56* | <0.0001 | 70 | <0.0001 |
| ACEi/ARB, % | 25 | 44 | 0.0002 | 33 | 46* | 0.0278 | 50 | 0.0030 |
| Diuretic, % | 22 | 48 | <0.0001 | 20 | 61* | <0.0001 | 76 | <0.0001 |
LAVI-Left Atrial Volume Index, RVSP-Right Ventricular Systolic Pressure, LVEDD-left Ventricular End diastolic dimension,
available only in Mayo subset (n=83 NCD, n=292 HFpEF),
Comparison of hospitalized patients with HFpEF to ambulatory patients with non-cardiac dyspnea from the international ambulatory validation cohort. Two-sided P values for continuous variables are from a T test or Wilcoxon rank sum test, and for binary variables from a Chi square test with no adjustment for multiple comparisons.
Baseline characteristics of validation cohorts
The multicenter international validation cohort (primary validation cohort) included 736 patients undergoing invasive hemodynamic exercise testing across 6 centers in 3 continents (77% HFpEF, 23% NCD). Characteristics of HFpEF in the validation cohort were similar to those of the derivation cohort, with greater obesity, atrial fibrillation, natriuretic peptides and echo abnormalities compared to NCD (Table 1). The additional ambulatory validation cohort (secondary validation cohort) included 228 consecutive patients with unexplained dyspnea undergoing exercise Right Heart Catheterization (RHC) at a single center (78% HFpEF, 22% NCD). (Extended Data Table 1).
The hospitalized HFpEF validation cohort included 456 patients who were older, more obese, and had greater prevalence of diabetes, hypertension and atrial fibrillation compared to the international cohort of NCD confirmed by exercise RHC (Table 1). As expected, N-terminal pro b-type natriuretic peptide (NTproBNP) levels were markedly elevated compared to controls along with worse echo estimates of left sided filling pressures, left atrial volume and estimated right ventricular systolic pressure.
In an additional validation analysis for adverse HF outcomes, we utilized a national cohort from the Veteran Affairs Health System of HFpEF with type 2 diabetes who demonstrated features typical of HFpEF with a mean age of 70.6 ± 8.4 years and a high prevalence of atrial fibrillation (42%) and obesity (BMI 37.1±7.5 kg/m2). In contrast to other cohorts, and consistent with the known demographics of the VA health system, there was a male predominance in this cohort (96.4%) (Extended Data Table 2).
Derivation of the HFpEF-ABA score
The predictive value of individual clinical variables from the H2FPEF score along with other clinical variables in isolation (age, BMI, atrial fibrillation, >1 antihypertensive medication, diabetes) from the derivation cohort, primary validation cohort and hospitalized HFpEF cohorts are presented in Extended Data Table 3. Each variable considered individually was mildly discriminatory for HFpEF in logistic regression, with the exception of history of diabetes which was not significant in the primary validation cohort [AUC 0.524 (0.493–0.555), p=0.067].
Multivariable models using these clinical variables were then used to predict HFpEF (Table 2). The models combining age + BMI, AF + BMI and age + AF each independently discriminated HFpEF from NCD, with retained discriminatory performance in the validation cohorts. Combining age in years, BMI in kg/m2, and AF (present/absent) in a single model (henceforth termed the HFpEF-ABA score) resulted in incremental diagnostic performance (AUC 0.839 (0.800–0.877), p<0.0001, Figure 1), with model comparison p=0.005, p=0.037 and p<0.0001 for AUC comparisons with age + BMI, AF + BMI, or age + AF models, respectively.
Table 2:
Diagnostic performance of risk models in the derivation and validation cohorts
| Derivation AUC |
AUC Difference |
Primary Validation AUC |
AUC Difference |
Hospitalized HFpEF Validation AUC |
AUC Difference* |
|
|---|---|---|---|---|---|---|
| Age + BMI | 0.806 (0.761–0.850), p<0.0001 |
−0.033 (−0.056 to −0.010), p=0.0047, (n=414) |
0.770 (0.730–0.810), p<0.0001 |
−0.043 (−0.067 to −0.018), p=0.0006, (n=736) |
0.908 (0.884–0.932), p<0.0001 |
− 0.021 (−0.045 to −0.005), p=0.0105, (n=629) |
| Age + AF | 0.776 (0.731–0.820), p<0.0001 |
−0.063 (−0.094 to −0.032), p<0.0001, (n=414) |
0.730 (0.689–0.771), p<0.0001 |
−0.083 (−0.111 to −0.055), p<0.0001, (n=736) |
0.874 (0.845–0.902), p<0.0001 |
−0.055 (−0.035 to −0.075), p<0.0001, (n=629) |
| AF + BMI | 0.799 (0.756–0.841), p<0.0001 |
−0.040 (−0.068 to −0.013), p=0.0041, (n=414) |
0.787 (0.750–0.824), p<0.0001 |
−0.026 (−0.059 to −0.002), p=0.0372 (n=736) |
0.861 (0.832–0.889), p<0.0001 |
−0.068 (−0.087 to −0.049), p<0.0001, (n=629) |
| HFpEF-ABA Model (reference model) | 0.839 (0.800–0.877), p<0.0001 |
Reference | 0.813 (0.779–0.847), p<0.0001 |
Reference | 0.929 (0.909–0.948), p<0.0001 |
Reference |
| ABA Model + >1 HTN drug | 0.843 (0.805–0.881), p<0.0001 |
+0.005 (−0.006 to +0.015), p=0.3961, (n=414) |
0.817 (0.782–0.851), p<0.0001 |
+0.004 (−0.006 to +0.014), p=0.4657, (n=736) |
0.942 (0.923–0.961) p<0.0001 |
+0.006 (+0.001 to + 0.011), p=0.0166, (n=600) |
| ABA Model + diabetes | 0.840 (0.802–0.878), p<0.0001 |
+0.001 (−0.004 to +0.006), p=0.6298, (n=414) |
0.809 (0.775–0.844), p<0.0001 |
−0.003 (−0.008 to +0.001), p=0.1478, (n=736) |
0.929 (0.090–0.948), p<0.0001 |
0.000 (−0.003 to +0.003) p=0.9567, (n=629) |
| ABA Model + NTproBNP | 0.859 (0.825–0.894), p<0.0001 |
+0.021 (+0.006 to +0.035), p=0.0053, (n=414) |
0.844 (0.807–0.881), p<0.0001 |
+0.014 (−0.002 to +0.030), p=0.0820, (n=560) |
0.987 (0.980–0.994) p<0.0001 |
+0.052 (0.035 to 0.069), p<0.0001, (n=540) |
Results are presented as Area Under the Curve [AUC] values or AUC difference (95% confidence interval), p value. Two-sided P values are from logistic regression and the Delong test for AUC comparison with no adjustment for multiple comparisons.
Age – age in years, AF- paroxysmal or permanent atrial fibrillation, BMI – body mass index in kg/m2, HTN- Hypertension.
Includes 456 hospitalized patients with HFpEF and 173 ambulatory patients with non-cardiac dyspnea.
Figure 1 – Discrimination of HFpEF from non-cardiac dyspnea using the HFpEF-ABA score.
(A-D) Logistic regression derived receiver operating curves for discrimination of HFpEF from non-cardiac dyspnea in derivation (A) and validation cohorts (B,C,D) using HFpEF-ABA score derived probabilities.
The addition of treatment with >1 antihypertensive or history of diabetes did not add incremental value to the simpler 3 variable HFpEF-ABA model (Table 2). Addition of NTproBNP levels statistically improved discrimination, but the impact was modest, with 2.5% increase in AUC (+0.021 (+0.006 to +0.035), p=0.005, when compared to ABA model]). The logistic regression equations to calculate the probability of HFpEF using the HFpEF-ABA model and other clinical models are provided in Extended Data Table 4. Calibration of the predicted probabilities with the empirical probabilities for HFpEF was robust in the derivation sample (Hosmer-Lemeshow goodness-of-fit p=0.68 for the HFpEF-ABA model and p=0.48 for HFpEF-ABA + NTproBNP model (Extended Data Figure 1).
Validation in ambulatory international HFpEF cohort
The HFpEF-ABA model was validated in the international primary validation cohort, with an AUC of 0.813 [95% CI 0.779–0.847], p<0.0001 for the ABA model and 0.844 [95% CI 0.807–0.881], p<0.0001 for the ABA model + NTproBNP (p=0.08 for comparison of ABA + NTproBNP vs ABA alone). NT-proBNP levels were missing at random in 24% of the primary validation cohort. Clinical characteristics of those with missing NT-proBNP values were similar to those with available NT-proBNP values (Extended Data Table 5). Imputation of missing values in a sensitivity analysis were consistent with lack of improved discrimination of the ABA model + NTproBNP (AUC 0.818 [95% CI 0.785–0.851],p<0.0001) compared to the ABA model alone (AUC difference +0.005 [95% CI −0.007 to +0.018], p=0.39). Calibration of the predicted probabilities with the empirical probabilities in the primary validation cohort was again supportive of proper calibration (Hosmer-Lemeshow goodness-of-fit test p=0.10 for ABA model and p=0.18 for ABA model + NTproBNP) (Extended Data Figure 1). Similar to the derivation cohort, the addition of diabetes and use of >1 hypertensive medications did not add discriminatory value (Table 2). Findings in the secondary validation cohort (n=226) were consistent with the results above with the HFpEF-ABA model demonstrating an AUC of 0.840 (95% CI 0.782–0.900) (Extended Data Table 6) (Figure 1).
Sex stratified analyses demonstrated comparable diagnostic performance of the HFpEF-ABA score in females and males (Extended Data Table 7). We tested HFpEF-ABA model performance in the subgroup of non-white participants (n=44) in the primary validation cohort where the AUC was 0.824, p=0.002 compared to 0.813, p<0.0001 in the larger white subgroup (n=692). Although the predicted HFpEF probability lies on a continuous scale from 0–100, sensitivity, specificity and likelihood ratios at deciles of predicted HFpEF probability are presented in Table 3. Use of higher HFpEF probability diagnostic thresholds improved specificity at the cost of sensitivity, whereas use of lower HFpEF probability diagnostic thresholds improved sensitivity at the cost of specificity. Shift from pre to post-test probability for HFpEF at various HFpEF-ABA thresholds is shown in Extended Data Table 8. The HFpEF-ABA score provided similar diagnostic discrimination as the integer based H2FPEF score, but model performance was superior to the more complex HFA-PEFF score that relies on echocardiography and natriuretic peptides (Extended Data Table 9).
Table 3:
Sensitivity and specificity of HFpEF-ABA model based probabilities in Derivation and Primary Validation Cohort
| Derivation cohort | Primary Validation cohort | |||||||
|---|---|---|---|---|---|---|---|---|
| Predicted HFpEF probability | Sensitivity | Specificity | LR− | LR+ | Sensitivity | Specificity | LR− | LR+ |
| 10 | 99 | 12 | 0.031 | 1.135 | 100 | 9 | 0.000 | 1.088 |
| 20 | 98 | 22 | 0.086 | 1.254 | 99 | 29 | 0.029 | 1.220 |
| 30 | 96 | 35 | 0.119 | 1.468 | 98 | 26 | 0.091 | 1.310 |
| 40 | 94 | 48 | 0.132 | 1.811 | 95 | 35 | 0.156 | 1.460 |
| 50 | 86 | 60 | 0.238 | 2.137 | 88 | 47 | 0.251 | 1.675 |
| 60 | 77 | 71 | 0.320 | 2.700 | 79 | 64 | 0.330 | 2.171 |
| 70 | 69 | 77 | 0.396 | 3.086 | 69 | 75 | 0.418 | 2.758 |
| 80 | 57 | 90 | 0.472 | 6.017 | 55 | 85 | 0.527 | 3.798 |
| 90 | 38 | 98 | 0.639 | 18.352 | 38 | 96 | 0.648 | 10.806 |
LR− indicates negative likelihood ratio, LR+ indicates positive likelihood ratio.
Validation in HFpEF cohort with prior HF hospitalization
When applied to patients with HFpEF defined by prior HF hospitalization, the ABA model retained high discrimination [AUC 0.929 (0.909–0.948), p<0.0001] with incremental diagnostic value compared to age +BMI or age + AF models alone (Table 2).
Association of ABA model probability with outcome
Higher HFpEF-ABA model probability of HFpEF was associated with greater risk of adverse HF events (HF hospitalization, death or the composite of HF hospitalization or death) in both the derivation, validation and combined cohorts of ambulatory patients from the Mayo Clinic (Table 4, Figure 2). The combined ABA + NTproBNP model was also associated with increased risk of events. The prognostic ability for adverse HF outcomes of both models was independently confirmed in the VA cohort (Table 4).
Table 4:
Risk of clinical events by HFpEF-ABA score probability
| Derivation Mayo Cohort (n=414) |
Primary Validation Mayo Cohort (n=376) |
Combined Mayo Cohort (n=790) |
National VA Cohort (N=3,076) |
|
|---|---|---|---|---|
| HFpEF-ABA model | ||||
| HF hospitalization | 1.21 (1.05–1.43), p=0.0070 | 1.25 (1.02–1.62), p=0.0308 |
1.23 (1.09–1.40), p=0.0004 |
1.12 (1.06–1.18) p<0.0001 |
| Death | 1.20 (1.08–1.35), p=0.0004 | 1.24 (0.97–1.67), p=0.0854 |
1.20 (1.09–1.33), p=0.0001 |
1.07 (1.02–1.12) p=0.003 |
| HF hospitalization or death | 1.23 (1.12–1.37), p<0.0001 | 1.24 (1.05–1.51), p=0.0109 | 1.22 (1.12–1.34), p<0.0001 | 1.10 (1.05–1.13) p<0.0001 |
| HFpEF-ABA Model + NTproBNP * | ||||
| HF hospitalization | 1.32 (1.13–1.57), p=0.0002 | 1.33 (1.07–1.74), p=0.0069 | 1.32 (1.17–1.52), p<0.0001 | 1.12 (1.04–1.21) p=0.002 |
| Death | 1.35 (1.20–1.55), p<0.0001 | 1.29 (1.002–1.79), p=0.0480 | 1.34 (1.20–1.51), p<0.0001 | 1.10 (1.02–1.17) p=0.013 |
| HF hospitalization or death | 1.36 (1.23–1.53), p<0.0001 | 1.28 (1.08–1.58), p=0.0031 | 1.34 (1.22–1.48), p<0.0001 | 1.11 (1.05–1.17) p<0.0001 |
Results are presented as hazard ratio (95% confidence interval) per 10% increase in model-based probability. Two-sided P values are from a Cox proportional model with no adjustment for multiple comparisons.
The HFpEF-ABA model was applied across the 4 cohorts among patients in whom NTproBNP levels were available. This included 414 patients in the Derivation Mayo Cohort, 333 patients in the Primary Validation Mayo Cohort, 745 patients in the Combined Mayo Cohort and 1114 patients in the National VA Cohort.
Figure 2 – Risk of HF hospitalization and death by HFpEF-ABA score.
Kaplan Meier curves showing the risk of the composite of HF hospitalization or death (A) and its individual components (B,C) with HFpEF-ABA score predicted HFpEF probability >75%. Hazard ratios were calculated by Cox proportional model.
Discussion
We derived and then validated a novel screening tool for HFpEF among patients with dyspnea using 3 universally available clinical metrics: age, body mass index and history of atrial fibrillation. The HFpEF-ABA score demonstrated reasonably strong discrimination and calibration across ambulatory patients with dyspnea in the derivation cohort, a second international multicenter validation cohort and a third independent ambulatory cohort. The HFpEF-ABA score also identified patients with HFpEF and propensity for clinical volume overload (those experiencing prior HF hospitalization) and was able to predict future risk of HF hospitalization when tested in multiple independent ambulatory cohorts.
While the probabilities provided by the HFpEF-ABA score should not replace clinical diagnosis, the present data show that the HFpEF-ABA score may be applied in clinical and research settings to quickly and efficiently screen for patients who have a greater probability of underlying HFpEF who should be referred for echocardiography. This strategy may be especially valuable in primary care and non-cardiology clinics where imaging is unavailable or rarely performed, or other settings where the diagnosis of HFpEF is not often considered. Further prospective validation of its screening utility is needed, and the score should only be applied in patients complaining of exertional dyspnea, in whom the score was derived and validated. Provision of the HFpEF-ABA score may assist in more efficient downstream use of echocardiography and cardiology referral to confirm a suspected diagnosis of HFpEF, when applied among patients with dyspnea. The HFpEF-ABA score could be an important tool to evaluate the probability of HFpEF in larger observational or trial cohorts where imaging data or natriuretic peptide testing was not performed or is unavailable.
The need for novel screening approaches to HFpEF is emphasized by current estimates that up to one-third of dyspneic patients in the community may have unrecognized HFpEF contributing to their symptoms.1,2,12 Unexplained dyspnea is a common problem in clinical practice, reported by a quarter of individuals over 65 year of age in the community,13 up to two thirds of patients with atrial fibrillation,14 and one-third of younger obese individuals.15 Symptomatic individuals with these risk factors are at elevated risk for HFpEF, but most are initially seen in non-cardiology clinics where symptoms may be incorrectly attributed to lung disease, aging, deconditioning, atrial fibrillation or obesity itself, without performing echocardiography or cardiology referral to consider HFpEF. There is therefore significant potential to improve identification of HFpEF with a goal of initiating proven therapies that can improve symptoms and outcomes.5–8
The H2FPEF score, which employs a computationally simpler algorithm that can be performed the bedside, is recommended by ACC/AHA guidelines to enhance diagnosis of HFpEF,16 but requires downstream diagnostic information from echocardiography, which is often not available at the point of screening. An analogous score, the HFA-PEFF score additionally requires natriuretic peptide testing in addition to echocardiography. Although the HFpEF-ABA score is mathematically more complex, the need for only simple demographic information to obtain HFpEF probability, provides a more universally applicable prediction algorithm that is not immediately dependent on echocardiographic information or natriuretic peptide testing. In our international validation cohort, 15% of individuals lacked data to compute a H2FPEF score, 21% lacked data to compute the HFA-PEFF score and 24% lacked NT-proBNP levels. In contrast, there was no patient who did not have the basic demographic information required for determining HFpEF probability across all cohorts using the HFpEF-ABA score with diagnostic performance comparable to the H2FPEF score and superior to the HFA-PEFF score. Absent echocardiographic data and NT-proBNP levels are expected to be even more common in real-world clinical screening of dyspneic patients. This makes the HFpEF-ABA score more applicable than the H2FPEF score or HFA-PEFF score for population-based or automated electronic medical record screening to identify patients with symptomatic dyspnea where clinical HFpEF may be present but not yet considered, and this screening approach requires testing in prospective clinical trials. Pending further study, ABA model derived probabilities are likely best utilized in a Bayesian approach to guide additional testing with patients through shared decision making. With intermediate or high model probabilities, patients may value echocardiography or further testing to evaluate the cause of dyspnea, whereas there may be less diagnostic yield in those with low model based probabilities. The specific thresholds of estimated HFpEF probability at which further testing is considered can be determined based on individual treatment goals and preferences.
The clinical need for improved HFpEF identification is emphasized by the introduction of effective therapies. For example, the sodium glucose cotransporter-2 (SGLT2) inhibitors dapagliflozin and empagliflozin improve quality of life, hemodynamics and reduce risk of hospitalization for HF.5,6,17 It is particularly notable that in cardiovascular outcome trials of SGLT2 inhibitors in patients with type 2 diabetes and chronic kidney disease, there was a clear reduction in risk of HF hospitalization, even as most patients had a preserved EF and did not have recognized HF at baseline.18–21 In the DECLARE-TIMI 58 trial (>15000 patients, most without clinical diagnosis of HF at randomization), the SGLT2 inhibitor dapagliflozin reduced risk of HF hospitalization by 23% over only 3–4 years.18 A subset of these patients in DECLARE-TIMI 58 likely developed de novo HFpEF for the first time during the follow-up period of the trial, but it is perhaps more likely that many of these patients in fact had occult, unrecognized HFpEF at the time of enrollment, especially considering the proportionately consistent reductions in HF hospitalization risk observed in DECLARE-TIMI 58 and cardiovascular outcome trials in HFpEF such as EMPEROR-Preserved and DELIVER.5,6 In addition to SGLT2 inhibitors, the glucagon like peptide-1 receptor agonist semaglutide has also recently been shown to reduce symptom severity and improve exercise function in patients with the obesity phenotype of HFpEF,7,8 and reduce the risk of cardiovascular events and HF hospitalization in patients with obesity/overweight and vascular disease.22 The predictive value of ABA model derived probabilities for future HF hospitalization events may help guide optimization of therapies in otherwise stable outpatients to prevent incident HF hospitalization, an approach that merits prospective study
Elevated HFpEF-ABA model-based probabilities may help guide caregivers to consider earlier imaging or referral for additional evaluation with cardiology or hemodynamic stress testing. Use of the HFpEF-ABA model probabilities might conceivably be useful to guide preventive therapies by defining patients with preclinical HFpEF.23 Development of HFpEF generally involves a long period of exposure to risk factors such as excess body fat, hypertension, and systemic inflammation, that ultimately results in sufficient accumulation of myocardial and extra-myocardial deficits that culminate in frank HFpEF.24,25 Prior to this watershed moment in its natural history, there is a stage of preclinical HFpEF, where preventive interventions may be optimally applied. For example, even among patients where HFpEF has been definitively excluded, individual with higher HFpEF probabilities based upon the H2FPEF score display hemodynamic and functional impairments that are typical of, but less severe than, clinically overt HFpEF.23 Interventions to prevent frank HFpEF and HF hospitalization, such as SGLT2 inhibitors or weight loss26,27 may therefore be more optimally delivered in patients with higher HFpEF-ABA score, and this targeted preventive approach requires testing in clinical trials,23 a key unmet need in the field.28
Our study results should be viewed in the context of certain limitations. Although the validation cohort was multicenter and spanned patients across 3 continents, there was no representation from Asia, South America or Africa and the study population was predominantly of white race. This limitation is particularly relevant given the lower BMI observed among individuals of Asian ancestry and although performance in the smaller non-white subset was comparable, further validation in other race/ethnicities is needed. Only patients presenting with symptomatic dyspnea were included, so the results may not be applicable to populations without dyspnea, or where dyspnea status is unknown. Although the study cohort is selected by virtue of referral for invasive hemodynamic exercise, this was essential to definitively ascertain case/control status, which is not possible without gold standard testing. The ability of the model to identify more “traditionally defined” HFpEF based on overt congestion (as in the hospitalized HFpEF validation cohort) mitigates concern of selection bias. The ability of NTproBNP levels to provide incremental diagnostic value to the clinical HFpEF model was of borderline statistical significance in both ambulatory validation cohorts (both p=0.08) which may have been the result of insufficient statistical power since NT-proBNP levels were not available in all patients in the validation cohort. However, incremental discrimination adding NT-proBNP to the model was modest (2.5% increase in AUC in the derivation cohort), and regression equations both with and without NTproBNP are provided, so either can be used based upon local preferences and availability of laboratory results. Some patients in the community with dyspnea will have low EF or other specific etiologies that were excluded from this study, and the probabilities provided by the HFpEF-ABA score are conditional probabilities that require performance of a subsequent echocardiogram to exclude HF with reduced EF, as well as alternative causes of clinical HF in the setting of normal EF (pulmonary arterial hypertension, valve or pericardial disease). Importantly, the HFpEF-ABA score is not meant to supplant clinical evaluation and imaging, but rather to serve as a screening tool to facilitate appropriate referral and consideration of HFpEF as a possible explanation for dyspnea. The loss of diagnostic information from reduction in input variables in the HFpEF-ABA score compared to the H2FPEF score necessitated inclusion of the variables on a continuous scale to optimize model performance. Therefore, as opposed to the simpler integer based H2FPEF score that can be applied to an individual patient at the bedside to guide clinical practice, the HFpEF-ABA score utilizes a complex logistic regression equation to generate probabilities that cannot be applied without an online calculator (Supplemental material) or other automated method. The HFpEF-ABA score may therefore be more appropriate for screening in the electronic medical record or large clinical/research cohorts of patients with dyspnea, although the clinical utility and diagnostic performance for community screening of such patients requires additional study.
In ambulatory patients with dyspnea, the HFpEF-ABA score accurately estimates the probability of underlying HFpEF and future risk of adverse HF outcomes based only on age, body mass index and history of atrial fibrillation. This screening approach can help guide the use of echocardiography and further evaluation for heart failure in symptomatic patients regardless of care setting, to enhance identification of patients with dyspnea due to undiagnosed HFpEF.
Methods
Patient cohorts
Data from these cohorts evaluated have been previously published, but not as they relate to the present analyses.9,11,29 Briefly, the derivation cohort included the study sample from which the H2FPEF score was originally derived including patients with EF≥50% and unexplained dyspnea between 2006 to 2016 at the Mayo Clinic who were referred for invasive exercise hemodynamic testing.
The primary validation cohort included patients from an international, multicenter sample of patients with unexplained dyspnea and EF≥50% undergoing invasive hemodynamic exercise testing across 3 centers in the US (Mayo Clinic, Rochester, MN; Johns Hopkins Hospital, Baltimore, MD; Medical University of South Carolina, Charleston, SC) and 3 international centers, in The Netherlands (Amsterdam University Medical Centre, Amsterdam), Denmark (Aarhus University Hospital, Aarhus), and Australia (Alfred Hospital, Melbourne, Victoria). Patients from the original derivation cohort from the Mayo Clinic were not included in the validation cohorts. An additional validation cohort included patients with unexplained dyspnea from the Mayo Clinic between 2018–2021.
In a supplementary validation analysis, we tested the diagnostic utility of the model to identify patients with HFpEF established by prior hospitalization for volume overload from the Mayo Clinic, Rochester, MN. In these patients, the diagnosis of HFpEF was unequivocally established clinically by hospitalization for congestion requiring intravenous diuretic administration for the first 48 hours of admission (details below).
Outcome assessment
To support the validity of the derived model for prediction of HFpEF, we evaluated the prognostic utility of score predicted probabilities to predict future risk of HF hospitalization and/or death. This was performed in the ambulatory Mayo Clinic derivation and validation cohorts where outcomes were available,30 and separately in an external sample of ambulatory patients with type 2 diabetes across the US Veteran Affairs (VA) Health Department, including patient data from more than 170 hospitals and 1025 outpatient facilities. The latter analysis included patient-level data accessed from VA databases, including the Corporate Data Warehouse, Vital Status File, and the Natural Language Processing left ventricular ejection fraction tables, using the VA Informatics and Computing Infrastructure. This administrative data includes clinical information regarding inpatient and outpatient visits, pharmacy prescription history and laboratory results. Vital status files are linked to the Social Security Index, Beneficiary Identification Records Subsystem and Center for Medicare and Medicaid Services. This cross-linking of data enables researchers to accurately identify relevant long-term events for each patient.
Case Definitions and Ascertainment
Across all centers among ambulatory patients with unexplained dyspnea and no overt congestion at rest, patients were studied in the supine position at rest and during supine cycle ergometry exercise to exhaustion.9,11 The diagnosis of HFpEF was definitively determined based on an elevated PCWP either at rest (≥15 mm Hg) or exercise (≥25 mm Hg) as per current guidelines.10 Controls with NCD were defined based on normal rest and exercise hemodynamics. Patients with a current or prior reduced EF, cardiomyopathies, high output heart failure, pulmonary arterial hypertension, severe valvular heart disease or constrictive pericarditis were excluded. History of AF was defined by any known paroxysmal or persistent/permanent AF. Atrial flutter was considered interchangeable with AF given known comparable clinical significance for anticoagulation and risk of underlying LA myopathy.31,32 Sex of participants was determined based on self-report. For comparative purposes, we calculated the H2FPEF score9 and HFA-PEFF score based on clinical and echocardiographic features.10,11
For the sensitivity analysis using patients with HFpEF hospitalized for volume overload, the initial study sample was created by using International Classification Disease (ICD) codes for HF in combination with filtering according to the terms decompensated or acute within the clinical notes from Mayo Clinic, Rochester, MN between 2010–2015.29 Patients were included only at the time of first HF hospitalization and were required to have an echocardiography result within 1 year of admission that demonstrated a left ventricular ejection fraction ≥50%. All charts were manually reviewed by a board-certified cardiologist (YNR) to confirm acute HFpEF as the primary diagnosis of admission, and the same exclusion criteria as above were employed for the hospitalized HFpEF cohort.
Patients with HFpEF within the VA health system were identified using a validated algorithm for identification of HFpEF within the national VA electronic health records.33 This required each patient to fulfil each of the following criteria: 1) An ICD diagnosis code of heart failure, 2) Either elevated natriuretic peptides or chronic use of loop diuretics, and 3) Ejection fraction >50%. Diseases that can mimic HFpEF including valvular heart disease, hypertrophic cardiomyopathy, peripartum cardiomyopathy, amyloidosis, sarcoidosis, and myocarditis were excluded. Patients with any EF<50% prior to the first ICD diagnosis code for HF were excluded.
Statistical analysis
Logistic regression with receiver operating characteristic (ROC) curves were used to determine the strength of association and discriminatory performance of the continuous variable models created based on the original H2FPEF score components using the area under the curve (AUC). Candidate variables were selected a priori based on known independent associations of clinical variables with HFpEF including age, BMI, AF, NT-proBNP, diabetes and >1 anti-hypertensive medication.9 Diagnostic accuracy of the constructed models using clinical variables were tested by logistic regression to predict HFpEF or NCD determined by gold standard diagnosis using invasive exercise testing. We utilized the clinical variables on a continuous scale when possible to maximize diagnostic information given the anticipated loss in diagnostic performance from the decrease in the number of input variables compared to the original H2FPEF score. Only clinical variables were included in the prediction model, without echocardiographic data, to enable application of developed models to assess cohorts where echocardiography was not yet available. A separate model with addition of serum NTproBNP levels was created to assess its incremental value prior to echocardiographic data to diagnose HFpEF. Missing data in the derivation cohort were imputed using random forest imputation (missForest package version 1.4).34 No imputation was performed for missing variables in the validation cohort to better reflect real world feasibility and validity of calculating HFpEF probability using the clinical score. Model performance was compared using the DeLong test. Calibration of the predicted probabilities with the empirical probabilities for HFpEF was assessed with the Hosmer-Lemeshow goodness-of-fit test. In the validation analysis using clinically diagnosed HFpEF after heart failure hospitalization, we tested model discrimination from those with confirmed non-cardiac dyspnea in the international cohort where HFpEF had been conclusively excluded by exercise right heart catheterization. Cox proportional Hazard ratios were calculated for risk of HF hospitalization and/or death using HFpEF probability as the predictor. The study was approved by the Mayo IRB (18–002030, 21–000738) with waiver of informed consent given use of de-identified patient data. All tests were 2-sided, with a value of P<0.05 considered significant. Analyses were performed with JMP, version 14.1.0 (JMP Statistical Discovery LLC), R 3.4.1 (R Foundation for Statistical Computing, Vienna, Austria) and BlueSky Statistics, version 7.40 (BlueSky Statistics LLC).
Extended Data
Extended Data Figure 1:
Calibration plots in Derivation and Validation Cohort predicted probabilities of HFpEF by the HFpEF-ABA score are grouped by deciles and plotted against the actual prevalence of HFpEF in each decile in ambulatory derivation (A) and validation cohort (B).
Extended Data Table 1:
Baseline characteristics of Ambulatory Secondary Validation Cohort
| Non-cardiac dyspnea (n=49) |
Ambulatory HFpEF (n=179) | p value | |
|---|---|---|---|
| Age | 59.8 ± 14.4 | 67.7 ± 12.0 | 0.0001 |
| Female, % | 74 | 56 | 0.0232 |
| BMI, kg/m2 | 27.7 ± 4.8 | 33.6 ± 8.0 | <0.0001 |
| Diabetes, % | 4 | 25 | 0.0003 |
| Hypertension, % | 49 | 76 | 0.0004 |
| Atrial fibrillation, % (no/parox/perm) | 92/8/0 | 68/21/11 | 0.0003 |
| Race, White/Black/Other% | 94/0/6 | 99/1/0 | 0.0031 |
| NT-proBNP | 60 [42–159] | 216 [84–751] | <0.0001 |
| Creatinine | 0.94 ± 0.22 | 1.12 ± 0.39 | 0.0017 |
| LV Ejection Fraction | 62 ± 4 | 62 ± 5 | 0.4346 |
| LVEDD, mm | 47.4 ± 4.2 | 49.6 ± 5.1 | 0.0133 |
| LAVI, ml/m2 | 26.7 ± 9.8 | 36.4 ± 18.1 | 0.0020 |
| Septal e’ velocity | 0.08 ± 0.02 | 0.07 ± 0.02 | 0.1048 |
| E/e’ | 9.8 ± 4.8 | 12.9 ± 6.6 | 0.0066 |
| RVSP, mmHg | 29.2 ± 6.5 | 40.5 ± 17.3 | 0.0007 |
| Beta blocker, % | 16 | 44 | 0.0003 |
| ACEi/ARB, % | 23 | 34 | 0.1293 |
| Diuretic, % | 20 | 50 | 0.0001 |
Two-sided P values for continuous variables are from a T test or Wilcoxon rank sum test, and for binary variables from a Chi square test with no adjustment for multiple comparisons.
Extended Data Table 2:
Baseline characteristics of VA ambulatory HFpEF cohort with type 2 diabetes
| Ambulatory HFpEF (n=3078) | |
|---|---|
| Age | 70.6 ± 8.4 |
| Female, % | 3.6 |
| BMI, kg/m2 | 37.1 ± 7.5 |
| Diabetes, % | 100 |
| Hypertension, % | 55.9 |
| Atrial fibrillation, % | 41.6 |
| NT-proBNP | 384 [131–1111] |
| Creatinine | 1.5 ± 0.9 |
| LV Ejection Fraction | 55 ± 5 |
Extended Data Table 3:
Diagnostic performance of Single Variable based Prediction Algorithm in Derivation and Validation Cohorts
| Derivation AUCa |
Primary Validation AUC |
Hospitalized HFpEF Validation AUC |
|
|---|---|---|---|
| Age | 0.743 (0.692–0.795), p<0.0001, (n=414) | 0.681 (0.634–0.728), p<0.0001, (n=736) | 0.831 (0.799–0.864), p<0.0001, (n=629)** |
| BMI | 0.703 (0.652–0.755), p<0.0001, (n=414) | 0.696 (0.651–0.741), p<0.0001, (n=736) | 0.720 (0.678–0.762), p<0.0001, (n=629) |
| AF | 0.652 (0.619–0.685), p<0.0001, (n=414) | 0.648 (0.619–0.676), p<0.0001, (n=736) | 0.753 (0.723–0.784), p<0.0001, (n=629) |
| >1 HTN medication | 0.678 (0.630–0.725), p<0.0001, (n=414) | 0.647 (0.606–0.688), p<0.0001, (n=736) | 0.752 (0.710–0.794), p<0.0001, (n=600) |
| Diabetes | 0.579 (0.541–0.617), p<0.0001, (n=414) | 0.524 (0.493–0.555), p=0.0675, (n=736) | 0.563 (0.529–0.596), p=0.0001 (n=629) |
Results are presented as Area Under the Curve [AUC] values (95% confidence interval), p value. Two-sided P values are from logistic regression with no adjustment for multiple comparisons.
Age – age in years, AF- paroxysmal or permanent atrial fibrillation, BMI – body mass index in kg/m2, HTN- Hypertension
Data previously published in reference9
Includes 456 hospitalized HFpEF and 173 non-cardiac dyspnea
Extended Data Table 4.
Regression equations for clinical variable models
| Regression equation for log odds of HFpEF* | |
|---|---|
| Age | −3.997742 + 0.073396 (age) |
| BMI | −3.168224 + 0.124006 (BMI) |
| Age + BMI | −8.334640 + 0.079543 (age) + 0.128005 (BMI) |
| Age + AF | −3.345276 + 0.058542 (age) + 1.914686 (AF history if present) |
| AF + BMI | −3.963373 + 0.136347 (BMI) + 2.704942 (AF history if present) |
| ABA Model | −7.788751 + 0.062564 (age) + 0.135149 (BMI) + 2.040806 (AF history if present) |
| ABA + >1 HTN drug | −7.436121 + 0.057758 (age) + 1.957936 (AF history if present) + 0.121762 (BMI) + 0.666309 (>1 HTN drug if present) |
| ABA + diabetes | −7.619470 + 0.060999 (age) + 2.047362 (AF history if present) + 0.130498 (BMI) + 0.361005 (diabetes if present) |
| ABA + NTproBNP | −7.999067 + 0.052368 (age) + 0.150643 (BMI) + 1.505152 (AF history if present) + 0.001010 (NT-proBNP) |
Age expressed in units of years, BMI in kg/m2, NTproBNP in pg/ml, and AF, >1 HTN drug, and diabetes are coded as present (=1) or absent (=0).
Probability of HFpEF is then derived from log odds as P= 1 / (1 + ê(−log odds))
Extended Data Table 5:
Clinical characteristics of those with missing NT-proBNP values in the primary validation cohort compared to those with available NT-proBNP levels.
| NT-proBNP available (n=560) |
Missing NT-proBNP (n=176) |
P value | |
|---|---|---|---|
| Age | 67.4 ± 12.2 | 66.0 ± 12.1 | 0.1867 |
| Female, % | 59 | 65 | 0.1075 |
| BMI, kg/m2 | 31.5 ± 7.2 | 30.9 ± 6.9 | 0.2733 |
| Diabetes, % | 19 | 15 | 0.1844 |
| >1 HTN drug, % | 61 | 56 | 0.2369 |
| Atrial fibrillation, % | 30 | 32 | 0.7146 |
| Creatinine | 1.08 ± 0.38 | 1.14 ± 0.97 | 0.2229 |
| LV Ejection Fraction | 60.9 ± 6.0 | 61.6 ± 6.3 | 0.2185 |
| LVEDD, mm | 48.0 ± 6.5 | 47.0 ± 5.9 | 0.0628 |
| LAVI, ml/m2 | 35.1 ± 13.6 | 36.6 ± 12.8 | 0.2155 |
| Septal e’ velocity | 0.07 ± 0.02 | 0.07 ± 0.02 | 0.2706 |
| E/e’ | 12.1 ± 6.3 | 11.7 ± 5.7 | 0.4367 |
Two-sided P values for continuous variables are from a T test or Wilcoxon rank sum test, and for binary variables from a Chi square test with no adjustment for multiple comparisons.
Extended Table 6:
Diagnostic performance of Clinical Models in Ambulatory Secondary Validation Cohort
| Validation AUC |
AUC Difference |
|
|---|---|---|
| Age + BMI | 0.787 (0.714–0.859), p<0.0001, (n=228) |
−0.053 (−0.088 to −0.018), p=0.0030, (n=228) |
| Age + AF | 0.736 (0.664–0.809), p<0.0001, (n=228) |
−0.103 (−0.158 to −0.048), p=0.0002, (n=228) |
| AF + BMI | 0.810 (0.751–0.869), p<0.0001, (n=228) |
−0.030 (−0.073 to +0.013), p=0.1731 (n=228) |
| HFpEF-ABA Model (reference model) | 0.840 (0.782–0.900), p<0.0001, (n=228) |
Reference |
| ABA Model + >1 HTN drug | 0.842 (0.783–0.900), p<0.0001 (n=228) |
+0.002 (−0.015 to +0.019), p=0.8147, (n=228) |
| ABA Model + diabetes | 0.845 (0.788–0.901), p<0.0001 (n=228) |
+0.005 (−0.0001 to +0.011), p=0.0550, (n=228) |
| ABA Model + NTproBNP | 0.867 (0.813–0.922), p<0.0001 (n=207) |
+0.017 (−0.002 to −0.037), p=0.0785 (n=207) |
Results are presented as Area Under the Curve [AUC] values or AUC difference (95% confidence interval), p value. Two-sided P values are from logistic regression and the Delong test for AUC comparison with no adjustment for multiple comparisons.
Age – age in years, AF- paroxysmal or permanent atrial fibrillation, BMI – body mass index in kg/m2, HTN- Hypertension
Extended Data Table 7:
Diagnostic performance of HFpEF-ABA Score in Derivation and Validation Cohorts Stratified by Sex
| Derivation AUC (95% CI) |
Sex interaction p value | International Ambulatory Validation AUC (95% CI) |
Sex interaction p value | Second Ambulatory Validation AUC (95% CI) |
Sex interaction p value |
Hospitalized HFpEF Validation AUC (95% CI) |
Sex interaction p value |
|
|---|---|---|---|---|---|---|---|---|
| Female | 0.865 (0.820–0.911), p<0.0001, (n=249) |
0.4973 | 0.836 (0.795–0.878), p<0.001, (n=443) |
0.3160 | 0.846 (776–0.915), p<0.0001, (n=136) |
0.5409 | 0.938 (916–0.961), p<0.0001, (n=381) |
0.8947 |
| Male | 0.812 (0.746–0.878), p<0.0001, (n=165) |
0.774 (0.714.−0.833), p<0.001, (n=293) |
0.814 (0.703–0.925), p=0.0003, (n=92) |
0.918 (0.883–0.953), p<0.0001, (n=248) |
Two-sided P values are from logistic regression with no adjustment for multiple comparisons.
Extended Data Table 8:
Change from pre to post test probability of HFpEF at various rule-out or rule-in HFpEF-ABA thresholds
| Empiric pretest probability of HFpEF (%) |
HFpEF ABA probability (%) |
Post-test probability of HFpEF (%) |
|---|---|---|
| 20.0 | <10 | 0.8 |
| <20 | 2.1 | |
| <30 | 2.9 | |
| >70 | 43.6 | |
| >80 | 60.1 | |
| >90 | 82.1 | |
| 50.0 | <10 | 3.0 |
| <20 | 7.9 | |
| <30 | 10.6 | |
| >70 | 75.5 | |
| >80 | 85.8 | |
| >90 | 94.8 | |
| 80.0 | <10 | 10.9 |
| <20 | 25.6 | |
| <30 | 32.2 | |
| >70 | 92.5 | |
| >80 | 96.0 | |
| >90 | 98.7 |
Extended Data Table 9:
Diagnostic performance of HFpEF-ABA Score in Derivation and Validation cohorts compared to H2FPEF and HFA-PEFF scores
| Derivation AUC (95% CI) |
AUC difference (95% CI) |
Primary Ambulatory Validation AUC (95% CI) |
AUC difference (95% CI) |
Second Ambulatory Validation AUC (95% CI) |
AUC difference (95% CI) |
|
|---|---|---|---|---|---|---|
| HFpEF-ABA Score | 0.839 (0.801–0.877), p<0.0001, (n=414) |
Reference | 0.813 (0.779–0.847), p<0.0001, (n=736) |
Reference | 0.840 (0.782–0.900), p<0.0001, (n=228) |
Reference |
| H2FPEF Score | 0.841 (0.802–0.880), p<0.0001, (n=414) |
+0.002 (−0.027 to +0.023), p=0.8531, (n=414) |
0.845 (0.813–0.878), p<0.0001, (n=627) |
+0.011 (−0.012 to +0.035), p=0.3460, (n=627) |
0.862 (0.788–0.937), p<0.0001. (n=163) |
+0.009 (−0.045 to +0.063), p=0.7389, (n=163) |
| HFA-PEFF Score | 0.764 (0.717–0.811), p<0.0001, (n=414) |
−0.075 (−0.122 to −0.028), p=0.0016, (n=414) |
0.712 (0.664–0.760), p<0.0001, (n=594) |
−0.100 (−0.152 to −0.047), p=0.0002, (n=594) |
0.765 (0.687–0.843), p<0.0001, (n=194) |
+0.088 (+0.005 to +0.172), p=0.0381, (n=194) |
Two-sided P values are from logistic regression and the Delong test for AUC comparison with no adjustment for multiple comparisons
Supplementary Material
Acknowledgements
This work was supported by National Institute of Health grants - R01 HL128526 (B.A.B.), R01 HL162828 (B.A.B.), U01 HL160226 (B.A.B.) and K23HL164901 (Y.N.R.) from the National Institutes of Health; and W81XWH2210245 (B.A.B.) from the US Department of Defense.
Footnotes
Competing Interests Statement
BAB receives research support from the National Institutes of Health (NIH) and the United States Department of Defense, as well as research grant funding from AstraZeneca, Axon, GlaxoSmithKline, Medtronic, Mesoblast, Novo Nordisk, and Tenax Therapeutics. Dr. Borlaug has served as a consultant for Actelion, Amgen, Aria, BD, Boehringer Ingelheim, Cytokinetics, Edwards Lifesciences, Eli Lilly, Janssen, Merck, and Novo Nordisk. BAB and SJA are named inventors (US Patent no. 10,307,179) for the tools and approach for a minimally invasive pericardial modification procedure to treat heart failure.
YNR receives research support from the National Institutes of Health (NIH), Sleep Number, Bayer, Merck and United pharmaceuticals.
MLH reported receiving grants from The Dutch Heart Foundation and educational, speaker, and consultancy fees from Novartis, Boehringer Ingelheim, AstraZeneca, Vifor Pharma, Bayer, Merck Sharp & Dohme, Abbott, Daiichi Sankyo, and Quin outside the submitted work.
RJT reports no direct conflicts of interest related to this manuscript. He is co-chair of the PH due to left heart disease task force for 7th World Symposium on Pulmonary Hypertension. He reports general disclosures to include consulting relationships with Abbott, Acorai, Aria CV Inc., Acceleron/Merck, Alleviant, CareDx, Cytokinetics, Edwards LifeSciences, Gradient, Lexicon Pharmaceuticals, Medtronic, and United Therapeutics. RJT serves on steering committee for Merck, Edwards, and Abbott as well as a research advisory board for Abiomed. He also does hemodynamic core lab work for Merck.
MJA reports no direct conflicts of interest related to this manuscript. He reports a consulting relationship with Johnson & Johnson.
FHV reports no direct conflicts of interest related to this manuscript. He reports a consulting relationship with Abbott Laboratories, Abiomed, AstraZeneca, Bayer, Boehringer Ingelheim, Bristol-Myers Squibb Belgium, Daiichi-Sankyo, Menarini Benelux, MSD, Novartis Pharma, Novo Nordisk Pharma, Pfizer, Roche Diagnostics, & Qompium.
The remaining authors declare no competing interests.
Code availability
The code utilized for derivation and validation of the HFpEF-ABA score is available at 10.6084/m9.figshare.26020810.
Data availability
The data from the derivation and validation cohorts necessary to derive and validate the HFpEF-ABA score is publicly available at 10.6084/m9.figshare.26020810 with restrictions, as follows. Data from the national Veteran Affairs (VA) Health System can only be accessed through a remote desktop connection within the VA network, and raw data cannot be transferred outside of the remote desktop environment. If other investigators are interested in performing additional analyses, requests can be made to the corresponding author, B.A.B, and analyses could be performed in collaboration with the Mayo Clinic.
References
- 1.Selvaraj S, Myhre PL, Vaduganathan M, Claggett BL, Matsushita K, Kitzman DW, Borlaug BA, Shah AM, Solomon SD. Application of Diagnostic Algorithms for Heart Failure With Preserved Ejection Fraction to the Community. JACC Heart Fail. 2020;8:640–653. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Borlaug BA, Sharma K, Shah SJ, Ho JE. Heart Failure With Preserved Ejection Fraction: JACC Scientific Statement. J Am Coll Cardiol 2023;81:1810–1834. [DOI] [PubMed] [Google Scholar]
- 3.Verwerft J, Soens L, Wynants J, Meysman M, Jogani S, Plein D, Stroobants S, Herbots L, Verbrugge FH. Heart failure with preserved ejection fraction: relevance of a dedicated dyspnoea clinic. Eur Heart J 2023;44:1544–1556. [DOI] [PubMed] [Google Scholar]
- 4.Saito Y, Obokata M, Harada T, Kagami K, Wada N, Okumura Y, Ishii H. Prognostic benefit of early diagnosis with exercise stress testing in heart failure with preserved ejection fraction. Eur J Prev Cardiol 2023;30:902–911. [DOI] [PubMed] [Google Scholar]
- 5.Anker SD, Butler J, Filippatos G, Ferreira JP, Bocchi E, Böhm M, Brunner-La Rocca H-P, Choi D-J, Chopra V, Chuquiure-Valenzuela E, Giannetti N, Gomez-Mesa JE, Janssens S, Januzzi JL, Gonzalez-Juanatey JR, Merkely B, Nicholls SJ, Perrone SV, Piña IL, Ponikowski P, Senni M, Sim D, Spinar J, Squire I, Taddei S, Tsutsui H, Verma S, Vinereanu D, Zhang J, Carson P, Lam CSP, Marx N, Zeller C, Sattar N, Jamal W, Schnaidt S, Schnee JM, Brueckmann M, Pocock SJ, Zannad F, Packer M, EMPEROR-Preserved Trial Investigators. Empagliflozin in Heart Failure with a Preserved Ejection Fraction. N Engl J Med 2021;385:1451–1461. [DOI] [PubMed] [Google Scholar]
- 6.Solomon SD, McMurray JJV, Claggett B, de Boer RA, DeMets D, Hernandez AF, Inzucchi SE, Kosiborod MN, Lam CSP, Martinez F, Shah SJ, Desai AS, Jhund PS, Belohlavek J, Chiang C-E, Borleffs CJW, Comin-Colet J, Dobreanu D, Drozdz J, Fang JC, Alcocer-Gamba MA, Al Habeeb W, Han Y, Cabrera Honorio JW, Janssens SP, Katova T, Kitakaze M, Merkely B, O’Meara E, Saraiva JFK, Tereshchenko SN, Thierer J, Vaduganathan M, Vardeny O, Verma S, Pham VN, Wilderäng U, Zaozerska N, Bachus E, Lindholm D, Petersson M, Langkilde AM, DELIVER Trial Committees and Investigators. Dapagliflozin in Heart Failure with Mildly Reduced or Preserved Ejection Fraction. N Engl J Med 2022;387:1089–1098. [DOI] [PubMed] [Google Scholar]
- 7.Borlaug BA, Kitzman DW, Davies MJ, Rasmussen S, Barros E, Butler J, Einfeldt MN, Hovingh GK, Møller DV, Petrie MC, Shah SJ, Verma S, Abhayaratna W, Ahmed FZ, Chopra V, Ezekowitz J, Fu M, Ito H, Lelonek M, Melenovsky V, Núñez J, Perna E, Schou M, Senni M, van der Meer P, Von Lewinski D, Wolf D, Kosiborod MN. Semaglutide in HFpEF across obesity class and by body weight reduction: a prespecified analysis of the STEP-HFpEF trial. Nat Med 2023; [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Kosiborod MN, Abildstrøm SZ, Borlaug BA, Butler J, Rasmussen S, Davies M, Hovingh GK, Kitzman DW, Lindegaard ML, Møller DV, Shah SJ, Treppendahl MB, Verma S, Abhayaratna W, Ahmed FZ, Chopra V, Ezekowitz J, Fu M, Ito H, Lelonek M, Melenovsky V, Merkely B, Núñez J, Perna E, Schou M, Senni M, Sharma K, Van der Meer P, von Lewinski D, Wolf D, Petrie MC, STEP-HFpEF Trial Committees and Investigators. Semaglutide in Patients with Heart Failure with Preserved Ejection Fraction and Obesity. N Engl J Med 2023; [Google Scholar]
- 9.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;138:861–870. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Pieske B, Tschöpe C, de Boer RA, Fraser AG, Anker SD, Donal E, Edelmann F, Fu M, Guazzi M, Lam CSP, Lancellotti P, Melenovsky V, Morris DA, Nagel E, Pieske-Kraigher E, Ponikowski P, Solomon SD, Vasan RS, Rutten FH, Voors AA, Ruschitzka F, Paulus WJ, Seferovic P, Filippatos G. How to diagnose heart failure with preserved ejection fraction: the HFA-PEFF diagnostic algorithm: a consensus recommendation from the Heart Failure Association (HFA) of the European Society of Cardiology (ESC). Eur Heart J 2019;40:3297–3317. [DOI] [PubMed] [Google Scholar]
- 11.Reddy YNV, Kaye DM, Handoko ML, van de Bovenkamp AA, Tedford RJ, Keck C, Andersen MJ, Sharma K, Trivedi RK, Carter RE, Obokata M, Verbrugge FH, Redfield MM, Borlaug BA. Diagnosis of Heart Failure With Preserved Ejection Fraction Among Patients With Unexplained Dyspnea. JAMA Cardiol 2022;7:891–899. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Obokata M, Olson TP, Reddy YNV, Melenovsky V, Kane GC, Borlaug BA. Haemodynamics, dyspnoea, and pulmonary reserve in heart failure with preserved ejection fraction. Eur Heart J 2018;39:2810–2821. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Ramalho SHR, Santos M, Claggett B, Matsushita K, Kitzman DW, Loehr L, Solomon SD, Skali H, Shah AM. Association of Undifferentiated Dyspnea in Late Life With Cardiovascular and Noncardiovascular Dysfunction: A Cross-sectional Analysis From the ARIC Study. JAMA Netw Open. 2019;2:e195321. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.van der Velden RMJ, Hermans ANL, Pluymaekers NAHA, Gawalko M, Elliott A, Hendriks JM, Franssen FME, Slats AM, van Empel VPM, Van Gelder IC, Thijssen DHJ, Eijsvogels TMH, Leue C, Crijns HJGM, Linz D, Simons SO. Dyspnea in patients with atrial fibrillation: Mechanisms, assessment and an interdisciplinary and integrated care approach. Int J Cardiol Heart Vasc 2022;42:101086. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Sin DD, Jones RL, Man SFP. Obesity is a risk factor for dyspnea but not for airflow obstruction. Arch Intern Med 2002;162:1477–1481. [DOI] [PubMed] [Google Scholar]
- 16.Heidenreich PA, Bozkurt B, Aguilar D, Allen LA, Byun JJ, Colvin MM, Deswal A, Drazner MH, Dunlay SM, Evers LR, Fang JC, Fedson SE, Fonarow GC, Hayek SS, Hernandez AF, Khazanie P, Kittleson MM, Lee CS, Link MS, Milano CA, Nnacheta LC, Sandhu AT, Stevenson LW, Vardeny O, Vest AR, Yancy CW. 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. J Am Coll Cardiol 2022;79:e263–e421. [DOI] [PubMed] [Google Scholar]
- 17.Nassif ME, Windsor SL, Borlaug BA, Kitzman DW, Shah SJ, Tang F, Khariton Y, Malik AO, Khumri T, Umpierrez G, Lamba S, Sharma K, Khan SS, Chandra L, Gordon RA, Ryan JJ, Chaudhry S-P, Joseph SM, Chow CH, Kanwar MK, Pursley M, Siraj ES, Lewis GD, Clemson BS, Fong M, Kosiborod MN. The SGLT2 inhibitor dapagliflozin in heart failure with preserved ejection fraction: a multicenter randomized trial. Nat Med 2021;27:1954–1960. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Wiviott SD, Raz I, Bonaca MP, Mosenzon O, Kato ET, Cahn A, Silverman MG, Zelniker TA, Kuder JF, Murphy SA, Bhatt DL, Leiter LA, McGuire DK, Wilding JPH, Ruff CT, Gause-Nilsson IAM, Fredriksson M, Johansson PA, Langkilde A-M, Sabatine MS, DECLARE–TIMI 58 Investigators. Dapagliflozin and Cardiovascular Outcomes in Type 2 Diabetes. N Engl J Med 2019;380:347–357. [DOI] [PubMed] [Google Scholar]
- 19.Heerspink HJL, Stefánsson BV, Correa-Rotter R, Chertow GM, Greene T, Hou F-F, Mann JFE, McMurray JJV, Lindberg M, Rossing P, Sjöström CD, Toto RD, Langkilde A-M, Wheeler DC, DAPA-CKD Trial Committees and Investigators. Dapagliflozin in Patients with Chronic Kidney Disease. N Engl J Med 2020;383:1436–1446. [DOI] [PubMed] [Google Scholar]
- 20.Kato ET, Silverman MG, Mosenzon O, Zelniker TA, Cahn A, Furtado RHM, Kuder J, Murphy SA, Bhatt DL, Leiter LA, McGuire DK, Wilding JPH, Bonaca MP, Ruff CT, Desai AS, Goto S, Johansson PA, Gause-Nilsson I, Johanson P, Langkilde AM, Raz I, Sabatine MS, Wiviott SD. Effect of Dapagliflozin on Heart Failure and Mortality in Type 2 Diabetes Mellitus. Circulation. 2019;139:2528–2536. [DOI] [PubMed] [Google Scholar]
- 21.Zinman B, Wanner C, Lachin JM, Fitchett D, Bluhmki E, Hantel S, Mattheus M, Devins T, Johansen OE, Woerle HJ, Broedl UC, Inzucchi SE, EMPA-REG OUTCOME Investigators. Empagliflozin, Cardiovascular Outcomes, and Mortality in Type 2 Diabetes. N Engl J Med 2015;373:2117–2128. [DOI] [PubMed] [Google Scholar]
- 22.Lincoff AM, Brown-Frandsen K, Colhoun HM, Deanfield J, Emerson SS, Esbjerg S, Hardt-Lindberg S, Hovingh GK, Kahn SE, Kushner RF, Lingvay I, Oral TK, Michelsen MM, Plutzky J, Tornøe CW, Ryan DH, SELECT Trial Investigators. Semaglutide and Cardiovascular Outcomes in Obesity without Diabetes. N Engl J Med 2023; [Google Scholar]
- 23.Koepp KE, Reddy YNV, Obokata M, Sorimachi H, Verbrugge FH, Jain CC, Egbe AC, Redfield MM, Olson TP, Borlaug BA. Identification of Patients with Preclinical Heart Failure with preserved Ejection Fraction Using the H2FPEF Score. Nat Cardiovasc Res 2022;1:59–66. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Reddy YNV, Obokata M, Jones AD, Lewis GD, Shah SJ, Abouezzedine OF, Fudim M, Alhanti B, Stevenson LW, Redfield MM, Borlaug BA. Characterization of the Progression From Ambulatory to Hospitalized Heart Failure With Preserved Ejection Fraction. J Card Fail 2020;26:919–928. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Wohlfahrt P, Redfield MM, Lopez-Jimenez F, Melenovsky V, Kane GC, Rodeheffer RJ, Borlaug BA. Impact of general and central adiposity on ventricular-arterial aging in women and men. JACC Heart Fail 2014;2:489–499. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Kitzman DW, Brubaker P, Morgan T, Haykowsky M, Hundley G, Kraus WE, Eggebeen J, Nicklas BJ. Effect of Caloric Restriction or Aerobic Exercise Training on Peak Oxygen Consumption and Quality of Life in Obese Older Patients With Heart Failure With Preserved Ejection Fraction: A Randomized Clinical Trial. JAMA 2016;315:36–46. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Reddy YNV, Anantha-Narayanan M, Obokata M, Koepp KE, Erwin P, Carter RE, Borlaug BA. Hemodynamic Effects of Weight Loss in Obesity: A Systematic Review and Meta-Analysis. JACC Heart Fail. 2019;7:678–687. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Shah SJ, Borlaug BA, Kitzman DW, McCulloch AD, Blaxall BC, Agarwal R, Chirinos JA, Collins S, Deo RC, Gladwin MT, Granzier H, Hummel SL, Kass DA, Redfield MM, Sam F, Wang TJ, Desvigne-Nickens P, Adhikari BB. Research Priorities for Heart Failure With Preserved Ejection Fraction: National Heart, Lung, and Blood Institute Working Group Summary. Circulation. 2020;141:1001–1026. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Verbrugge FH, Reddy YNV, Sorimachi H, Omote K, Carter RE, Borlaug BA. Diagnostic scores predict morbidity and mortality in patients hospitalized for heart failure with preserved ejection fraction. Eur J Heart Fail. 2021;23:954–963. [DOI] [PubMed] [Google Scholar]
- 30.Omote K, Verbrugge FH, Sorimachi H, Omar M, Popovic D, Obokata M, Reddy YNV, Borlaug BA. Central haemodynamic abnormalities and outcome in patients with unexplained dyspnoea. Eur J Heart Fail. 2023;25:185–196. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Reddy YNV, Obokata M, Gersh BJ, Borlaug BA. High Prevalence of Occult Heart Failure With Preserved Ejection Fraction Among Patients With Atrial Fibrillation and Dyspnea. Circulation. 2018;137:534–535. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Reddy YNV, Obokata M, Verbrugge FH, Lin G, Borlaug BA. Atrial Dysfunction in Patients With Heart Failure With Preserved Ejection Fraction and Atrial Fibrillation. J Am Coll Cardiol 2020;76:1051–1064. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Patel YR, Robbins JM, Kurgansky KE, Imran T, Orkaby AR, McLean RR, Ho Y-L, Cho K, Michael Gaziano J, Djousse L, Gagnon DR, Joseph J. Development and validation of a heart failure with preserved ejection fraction cohort using electronic medical records. BMC Cardiovasc Disord 2018;18:128. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Stekhoven DJ, Bühlmann P. MissForest--non-parametric missing value imputation for mixed-type data. Bioinformatics. 2012;28:112–118. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data from the derivation and validation cohorts necessary to derive and validate the HFpEF-ABA score is publicly available at 10.6084/m9.figshare.26020810 with restrictions, as follows. Data from the national Veteran Affairs (VA) Health System can only be accessed through a remote desktop connection within the VA network, and raw data cannot be transferred outside of the remote desktop environment. If other investigators are interested in performing additional analyses, requests can be made to the corresponding author, B.A.B, and analyses could be performed in collaboration with the Mayo Clinic.