Evidence-Based Application of Natriuretic Peptides in the Evaluation of Chronic Heart Failure With Preserved Ejection Fraction in the Ambulatory Outpatient Setting.
Jan 22, 2025DOI: 10.1161/circulationaha.124.072156Journal ArticleAHA
Written by: Yogesh N V ReddyAtsushi TadaMasaru ObokataRickey E CarterDavid M KayeM Louis HandokoMads J AndersenKavita SharmaRyan J TedfordMargaret M RedfieldBarry A Borlaug
Find these authors on Roon
YR
Dr. Yogesh Reddy
Advanced Heart Failure · Mayo Clinic College of Medicine
KS
Dr. Kavita Sharma
Advanced Heart Failure · Johns Hopkins
Introduction
Natriuretic peptides (NP) are the most frequently used biomarkers in heart failure (HF). Elevated NP levels are associated with adverse hemodynamics and poorer prognosis.1 As compared to HF with reduced ejection fraction (EF), individuals with HF with preserved EF (HFpEF) have lower NP, often in the normal range.1,2 This is related in part to greater obesity (believed to impact NP production and degradation),3–5 smaller left ventricular dimension with concentric remodeling, and frequently normal filling pressures at rest,1,6–8 which reduce wall stress, the stimulus for NP production.5,9,10 Despite these facts,1 NP testing continues to be recommended for evaluation of HFpEF based upon limited evidence.11 Such reliance may contribute to clinical underrecognition,12 and therefore undertreatment for those with HFpEF, and potentially, misdiagnosis in those with spuriously high NP in the absence of HFpEF.Prior studies evaluating NP for HF diagnosis included patients with acute dyspnea seeking urgent care,13–15 a setting where accuracy was found to be high.16–18 In contrast, most patients with HFpEF have indolent, gradually progressive dyspnea, leading to presentation in the non-acute setting. The few studies evaluating NP in the ambulatory setting have focused on HF with reduced EF (HFrEF),19,20 without adequate HFpEF representation,20–22 and with no objective reference standard beyond expert opinion.4,23 We hypothesized that when compared to gold-standard diagnosis using exercise right heart catheterization (RHC), NP testing for ambulatory HFpEF with chronic dyspnea could be optimized using separate rule-in and rule-out diagnostic thresholds, and that these would be optimally stratified by obesity and atrial fibrillation (AF) status. Derived thresholds were then validated in six independent cohorts with definitive diagnosis.
Methods
The study was approved by the Mayo, Johns Hopkins, Gumma University, De Videnskabsetiske Komitéer For Region Midtjylland, Alfred Hospital and the Amsterdam University Medical Centers institutional review boards with waiver of informed consent given the use of deidentified patient data. The data that support the findings of this study are availabel from the corresponding author upon reasonable requestData from some of these cohorts have been previously published, but not as they relate to the present analyses focused on NP.4,24,25 Briefly, the derivation cohort included the study sample from which the HFPEF 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.
2
The primary validation cohort (Validation cohort 1) included patients from an international, multicenter sample of patients with unexplained dyspnea and EF≥50% undergoing invasive hemodynamic exercise testing across 2 centers in the US (Mayo Clinic, Rochester, MN; Johns Hopkins Hospital, Baltimore, MD) and 3 international centers in The Netherlands (Amsterdam University Medical Centre, Amsterdam), Denmark (Aarhus University Hospital, Aarhus), and Australia (Alfred Hospital, Melbourne, Victoria). Only patients with N-terminal pro-brain Natriuretic Peptide (NT-proBNP) levels availabel were included in the analysis. Therefore in contrast to our prior international cohort reports,24,25 patients from Medical University of South Carolina, Charleston, SC were not included as they only had B-type Natriuretic Peptide (BNP) levels measured based upon local laboratory preferences.
Validation cohort 2 included 207 consecutive unique patients with unexplained dyspnea undergoing invasive hemodynamic exercise testing at the Mayo Clinic between 2018–2021 with NT-proBNP levels availabel who were not included in the prior cohorts.25
Validation cohort 3 included 77 consecutive patients who underwent invasive hemodynamic exercise testing for the evaluation of unexplained dyspnea at the Gunma University Hospital, Maebashi, Japan between 2020–2024 with availabel NT-proBNP levels.
To evaluate broader validity beyond patients referred for exercise RHC, we also evaluated diagnostic performance of NT-proBNP in 3 additional validation cohorts (Validation cohorts 4–6). Validation cohort 4 included a random sample of HFpEF patients diagnosed following referral for resting RHC (without exercise RHC) at Mayo Clinic. We additionally evaluated performance in ambulatory HFpEF patients with definite prior HF hospitalization who had NT-proBNP levels measured during subsequent outpatient evaluation when volume status had normalized (Validation cohort 5).25 Since HFpEF cannot be definitively excluded by resting RHC or clinical evaluation in the outpatient setting4, we used patients with definite NCD from the primary international validation cohort for control group comparisons for both Validation cohorts 4 and 5. The final validation cohort (Validation cohort 6) included individuals who underwent non-invasive diagnosis of HFpEF by exercise echocardiography26 at the Gunma University Hospital, Maebashi, Japan between 2020–2024, where the diagnosis of HFpEF was determined by elevated rest or exercise E/e’ >15, with patients not meeting this criteria considered to have NCD.
Ambulatory patients were studied in the supine position at rest and during supine cycle ergometry exercise to exhaustion.4,24 The diagnosis of HFpEF was definitively ascertained based on objective hemodynamic evidence of HF, with elevated PCWP at rest (≥15 mmHg) or exercise (≥25 mmHg) as per current guidelines.26 Controls with non-cardiac dyspnea (NCD) were defined based on normal rest and exercise hemodynamics.
We tested the recommended “rule-out” diagnostic threshold of 125 pg/ml used for diagnosis of HF.11 We also tested recent combined age and BMI recommended cutpoints as recently proposed by expert consensus panel from the Heart Failure Association of the European Society of Cardiology.27 This algorithm recommends a single rule-out diagnostic threshold of <125 pg/ml regardless of age, and age specific rule-in of ≥125, ≥250 and ≥500 pg/ml at ages <50, 50–75 and ≥75 years respectively. An additional BMI adjustment for rule-in was proposed with reduction of cutpoints by 25%, 30% and 40% with BMI of 30–35, 35–40 and ≥40 kg/m2 respectively.
We then independently evaluated separate cutpoints that individually optimized sensitivity and specificity that could be used to rule-in and rule-out HFpEF with robust accuracy in the derivation cohort, and then tested these cutpoints in the primary multicenter validation cohort and in the 5 separate validation cohorts. Operating characteristics for NT-proBNP were also evaluated separately in subgroups of patients where diagnostic value may differ based upon prior studies, such as patients living with grade 2 or greater obesity [BMI≥35 kg/m2],5,28 and prevalent AF (defined as persistent/permanent AF at the time of RHC).29,30 Assay details are included in the online supplement.
Logistic regression was used to determine the strength of association and discriminatory performance of NT-proBNP on a continuous scale and for various cutpoints across a range of observed specificity and sensitivity. Operating characteristics were evaluated separately for patients with HFpEF with BMI≥35 kg/m2, as this threshold has been previously used to characterize the obesity phenotype of HFpEF5 and because relationships between NT-proBNP and hemodynamics differ in patients falling above or below this cutpoint.31 Diagnostic accuracy of NT-proBNP cutpoints were tested by logistic regression to predict HFpEF or NCD determined by gold standard diagnosis using invasive exercise testing. Missing data in the derivation cohort were imputed using random forest imputation as previously described (missForest package version 1.4).4 No imputation was performed for missing variables in the validation cohort with analysis restricted to those with measured NT-proBNP levels. Because binary NT-proBNP diagnostic thresholds typically sacrifice either sensitivity or specificity, we anticipated that we would require separate rule-out and rule-in thresholds to guide diagnosis with acceptable accuracy, such that patients with NT-proBNP values falling between these thresholds would remain considered as indeterminate and require further testing. Accordingly, our focus was only on one sided diagnostic accuracy with sensitivity (and negative likelihood ratio) for rule-out thresholds, and specificity (and positive likelihood ratio) for rule-in cutpoints. We then validated the sensitivity of the rule-out threshold and specificity of the rule-in threshold in the validation cohorts. To examine potential incremental value for NP testing, we then evaluated posttest probabilities of HFpEF using derived likelihood ratios based upon rule-out and rule-in NT-proBNP values over a range of pretest probabilities of HFpEF. We calculated posttest probabilities for both empirically estimated pre-test probability ranges, as well as more quantitatively determined pre-test probabilities using the validated H2FPEF score instrument.4
Results
The derivation cohort included 414 patients from one center (65% HFpEF, 35% non-cardiac dyspnea, NCD). Patients with HFpEF were older, more obese, and had greater prevalence of AF and renal dysfunction compared to NCD (Table 1). On average, NT-proBNP was elevated in the HFpEF group compared to NCD. Echocardiographic abnormalities were observed consistent with HFpEF, with higher estimated right ventricular systolic pressure, LA volume and E/e’ compared to NCD (Table 1).
The multicenter international validation cohort 1 included 560 patients undergoing invasive hemodynamic exercise testing across 5 centers on 3 continents (79% HFpEF, 21% NCD). Validation cohort 2 included 207 consecutive patients with unexplained dyspnea undergoing exercise RHC at a single center (79% HFpEF, 21% NCD). Consistent with the derivation cohort, obesity, AF, renal dysfunction and echo abnormalities were more common in HFpEF compared to NCD (Table 1), with NT-proBNP again elevated on average in both cohorts.
Validation cohort 3 included 77 Japanese patients (100% Asian race) who underwent invasive hemodynamic exercise testing (75% HFpEF, 25% NCD). (Table S1) Consistent with the other cohorts, patients with HFpEF were older and had higher NT-proBNP levels, a larger left atrial volume, and a higher E/e’ ratio than NCD, but BMI levels were lower than in other cohorts consistent with a primarily non-obese HFpEF cohort.
The fourth and fifth validation cohort included 145 and 332 HFpEF patients respectively with expected comorbidity burden (older with more AF and greater obesity) along with progressively greater average NT-proBNP elevation, renal dysfunction and echocardiographic abnormalities consistent with their mode of diagnosis through resting RHC (validation cohort 4) or diagnosed clinically through prior HF hospitalization (validation cohort 5) (Table S2).
The sixth validation cohort included 517 Japanese patients (100% Asian race) who underwent non-invasive diagnosis of HFpEF or NCD through exercise echocardiography (36% HFpEF, 64% NCD). Resting echo abnormalities, age, hypertension, diabetes, renal dysfunction and NT-proBNP were higher with HFpEF. Like the invasive Japanese cohort, BMI was lower and comparable between HFpEF and NCD consistent with a non-obese HFpEF cohort (Table S3).
In the derivation cohort, viewed continuously and using the optimal binary threshold, NT-proBNP provided marginal discrimination between HFpEF and NCD, with an AUC of 0.73 [95% CI 0.68–0.78], p<0.0001 at the optimal threshold of 272 pg/ml by Youden index, providing 60% [95% CI 54–66] specificity and 77% [95% CI 70–84] sensitivity. The 2023 ESC Consensus criteria rule-out threshold of <125 pg/ml had a sensitivity of only 77% [95% CI 72–82]. Using a recommended age-based rule-in threshold had a specificity of only 69% [95% CI 62–77], which did not improve with addition of recommended BMI-based rule-in threshold adjustments with a specificity of only 65% [95% CI 57–72%] (Table S4). Diagnostic performance of NT-proBNP on a continuous scale was similar when restricted to the subgroup of patients with grade 2 obesity only (n=109, 84% HFpEF) with an AUC of 0.76 [95% CI 0.67–0.86], p<0.0001, but in this group the optimal threshold was lower at 192 pg/ml, providing 88% [95% CI 73–100] specificity and 62% [95% CI 52–72] sensitivity. Of the 47 patients with AF, 46 (97.9%) had HFpEF, with 1 patient with unexplained dyspnea and AF found to have NCD (Figure 1). This precluded any ability for NT-proBNP thresholds to incrementally discriminate HFpEF from NCD in patients with AF, since the presence of AF itself was sufficiently diagnostic.
Recognizing that no single threshold of NT-proBNP could provide clinically acceptable specificity and sensitivity, we next evaluated various thresholds to determine the best rule-out and rule-in NT-proBNP thresholds for HFpEF that would provide sufficient confidence based upon sensitivity and specificity (Figure 2). The currently recommended rule-out threshold of <125 pg/ml had sensitivities of 77% in the overall cohort (Table S5, Figure S1), 82% in those with BMI<35 kg/m2, and 67% in those with grade 2 or greater obesity (Table 2, Figure 2). The sensitivities for possible alternate rule-out thresholds are presented in Table S1 and stratified by BMI of 35 kg/m2 in Table 2, where an NT-proBNP <50 pg/ml demonstrated 93% sensitivity overall; 97% in those with BMI<35 kg/m2, decreasing to 86% among those with grade 2 obesity.
Conversely, an NT-proBNP threshold ≥ 500 pg/ml provided 86% specificity overall, and 85% among those with BMI<35 kg/m2, with lower thresholds providing high false positive rates (Table S4, Table 2, Figure 2, Figure S1). However, among patients with grade 2 or greater obesity, lower thresholds for positive diagnosis provided acceptable specificity, with a threshold of ≥220 pg/ml associated with 88% specificity (Figure 2, Table 2).
NT-proBNP results in subgroups with sinus rhythm, BMI<30, BMI≥30 and BMI 30–35 kg/m2 were consistent with the overall study population results (Table S6–S9) supporting differential NT-proBNP rule-in thresholds only with grade 2 obesity.
A threshold of <50 pg/ml had a sensitivity of 91% among those with BMI<35 kg/m2 in the international validation cohort 1, 86% in validation cohort 2 and 93% in validation cohort 3, supporting use of NT-proBNP <50 pg/ml as a rule-out threshold with BMI<35 kg/m2.(Table 3, S10,S11) As in the derivation cohort, sensitivity at <50 pg/ml was lower in patients with grade 2 obesity (80%) in the international validation cohort 1 (Table 3), and in validation cohort 2 (84%) (Table S11). Using a lower diagnostic threshold of <35 pg/ml to rule-out HFpEF with grade 2 obesity resulted in modestly improved sensitivity for HFpEF in the derivation (94%) cohort suggesting this even lower threshold (<35 pg/ml) could serve as the rule-out diagnostic threshold. However, sensitivity remained imperfect even at this very low threshold in the international validation cohort 1 (88%) (Table 3), and validation cohort 2 (84%) (Table S10).
The specificity for HFpEF of a diagnostic threshold ≥500 pg/ml in patients with BMI<35 kg/m2 was 87%, 90% and 89% in the validation cohorts 1,2 and 3, supporting ≥500 pg/ml as a rule-in threshold in non-obese persons. This diagnostic threshold was unnecessarily high among those with grade 2 obesity in validation cohorts 1 and 2, where a threshold of ≥220 pg/ml retained excellent specificity (93% and 100%, Table 3, Table S10), supporting the use of ≥220 pg/ml as a rule-in threshold in grade 2 obesity.
Like the derivation cohort, 79/81 (97.5%),17/17 (100%) and 7/7 (100%) of patients with unexplained dyspnea and AF in three validation cohorts were found to have HFpEF (Figure 1), so the presence of prevalent AF by itself was sufficiently discriminatory for HFpEF without needing NT-proBNP levels, or AF-specific NT-proBNP diagnostic thresholds. With application of current recommended AF-stratified NT-proBNP thresholds for HFpEF rule-in (>660 pg/ml),26 91%, 78%, 94% and 86% of AF HFpEF patients would have been correctly identified as having HFpEF across the derivation and 3 validation cohorts. This resulted in a pooled 15.1% (23/152) of patients with proven HFpEF and AF not being correctly diagnosed as HFpEF when relying on NT-proBNP, as compared to 2.0% (3/152) being misdiagnosed simply using AF as the marker of HFpEF status (p=0.0001).
The sensitivity of HFpEF rule out at <50 pg/ml was maintained in validation cohorts 4,5 and 6 [99%, 98% and 94% respectively] (Table S12–S14). Additionally, specificity of ≥500 pg/ml for rule-in with BMI<35 kg/m2 in validation cohorts 4,5 and 6 was 87%, 87% and 81%. Among the 2 cohorts with adequate individuals with BMI>35 kg/m2 for analysis (validation cohort 4 and 5), specificity for ≥220 pg/ml for rule-in was 93% and 93% respectively). (Tables S12, S13)
Application of derived likelihood ratios to individuals who met the validated rule-in and rule-out thresholds resulted in the most meaningful changes in HFpEF probability (up or down) among those with intermediate pre-test probability at baseline (Figures 3 and 4). Among those with high pre-test probability, high NT-proBNP levels were diagnostically uninformative, and having ‘rule-out’ NT-proBNP levels still resulted in an intermediate post-test probability of HFpEF where further workup would be required (Tables S15–S17, Figure S3). The opposite was observed for individuals at lower pre-test probability. The increase in pre-test to post-test probability was generally greater for patients with grade 2 obesity, even as the rule-in diagnostic threshold of NT-proBNP was lower for positive diagnosis (220 vs 500 pg/ml, Figure 4).
In the combined derivation and validation cohorts, 490 of 927 (53%) patients with BMI<35 kg/m2 and 130 of 331 (39%) of patients with BMI≥35 kg/m2 had an NT-proBNP value that was within the “indeterminate” range (50–500 pg/ml in patients with BMI<35 kg/m2 or 50–220 pg/ml patients with BMI≥35 kg/m2). Further, as reflected by a H2FPEF score of 2–5, 53% of patients with BMI<35 kg/m2 and 64% patients with BMI≥35 kg/m2 had an intermediate probability of HFpEF. Overall, 22% of patients with intermediate probability H2FPEF scores of 2–5 had BMI stratified rule-in or rule-out diagnostic thresholds, with the other 78% having NT-proBNP levels falling in the indeterminate range where exercise testing is required to reach the diagnosis. Among those with intermediate pretest probability of HFpEF based upon H2FPEF score, 34% of those with BMI stratified rule-in achieved a high posttest probability>80%; while 31% of those with rule out threshold (<50 pg/ml) achieved a low posttest probability<20%. In contrast, application of existing NT-proBNP rule out<125 pg/ml to these intermediate pretest probability patients (57% HFpEF,43% NCD, n=648) would result in 42% of patients with true HFpEF having HFpEF incorrectly excluded; while existing age and BMI rule-in thresholds would result in 36% of patients with true NCD being misdiagnosed with HFpEF.
We next evaluated clinical characteristics associated with discordance between NT-proBNP classification and objectively measured evidence of congestive left heart failure by invasive hemodynamics. Compared to patients with HFpEF and high NT-proBNP, those with lower NT-proBNP were found to be younger, with more obesity, less AF and better renal function, like prior studies (Table S18). Compared to patients with NCD and normal NT-proBNP, those with NCD and high NT-proBNP were found to have multiple non-HF contributors to NT-proBNP elevation, including being older, leaner, and having more renal dysfunction (Table S19). Importantly, although patients with NCD and high NT-proBNP had structural changes consistent with HFpEF including larger left atria and higher resting E/e’, all displayed normal PCWP with absence of left heart failure when measured directly at rest and during exercise.
Although Mayo Clinic is a tertiary referral center, performance of proposed rule-in and rule-out thresholds were consistent among the subset of HFpEF patients who were from the local, predominantly rural population (66%), those with more advanced HFpEF defined by elevated PCWP at rest, those without imputed NT-proBNP values (76%), and in subgroup analyses stratified by sex, antihypertensive, or diuretic use (Table S20). There was some attenuation of specificity in older patients, and those with chronic kidney disease (CKD) (Table S20, S21). Accordingly, thresholds providing 95% specificity in those >60 years or with eGFR<60 ml/min/1.73m2 were higher at 1141 and 1529 pg/ml respectively (Table S22). However, these 2 subgroups with older age and CKD with unexplained dyspnea had a very high prevalence of HFpEF (83 and 89%), such that the modest decline in specificity was still associated with high positive predictive value (Table S23). Given the high pre-test probability (prevalence) >80%, a positive test result was associated with a small non-clinically meaningful shift from pre to posttest probability (Table S16, S17, Fig S2), such that age and CKD stratified NT-proBNP rule-in thresholds were not clinically useful in diagnosis of ambulatory HFpEF.
Discussion
In this study, we provide the first evidence-based evaluation of NT-proBNP decision thresholds to evaluate ambulatory patients with HFpEF and chronic exertional dyspnea without overt congestion, using gold standard diagnostic criteria through direct measurement of left heart filling pressures during exercise, in a large international, multicenter cohort. This knowledge gap has been acknowledged by guidelines.11 In contrast to prior literature focused on acute decompensated HF16–18 or chronic HF defined imprecisely by investigator opinion,20–22 we found that currently recommended NT-proBNP thresholds have only modest accuracy among ambulatory patients with chronic dyspnea when applied in a binary fashion using a single diagnostic threshold. Importantly, optimal rule-in and rule-out cutpoints were critically influenced by adiposity, being lower among those individuals with grade 2 or greater obesity. Utilization of a lower rule-out threshold of 50 pg/ml (and potentially even 35 pg/ml with BMI≥35 kg/m2) provided lower false negative rates. Corresponding optimal rule-in NT-proBNP thresholds were identified as 500 pg/ml and 220 pg/ml in patients with BMI <35 kg/m2 and ≥35 kg/m2, with findings again replicated in the 6 different validation cohorts with consistency of results regardless of care setting or HFpEF definition used. For patients in AF with dyspnea, there was no incremental value to NP testing, since the presence of AF was by itself a highly predictive biomarker for HFpEF. As expected from Bayes theorem, use of NT-proBNP rule-in and rule-out findings were most informative among patients with intermediate pretest probability assessed using the H2FPEF score. Finally, we observed a notable group of patients with NCD with increased NP levels despite normal rest and exercise PCWP, characterized by greater renal dysfunction, higher age, and lower BMI. Consistent with this finding, there was a modest decrease in specificity of NP rule-in criteria with older age and CKD, but the high prevalence (pre-test probability) of HFpEF in these symptomatic subgroups resulted in small non-meaningful pre to posttest probability shifts with NP rule-in. Therefore, patients with older age and CKD had high post-test probability for HFpEF regardless of whether they achieved NP ‘rule-in’ levels or not, and this may not justify more complex age and renal function sub-stratified NP cutpoints beyond BMI stratification for ambulatory HFpEF diagnosis. These data therefore suggest that NT-proBNP testing may be most optimally applied to chronically dyspneic individuals in sinus rhythm, with diagnostic value maximized by using separate rule-in and rule-out stratified by BMI to minimize false negative and false positive diagnoses, allowing for those with intermediate probability to undergo more definitive evaluation with hemodynamic exercise testing.
NP testing has consistently demonstrated excellent value as a diagnostic biomarker for patients with HFrEF, and most prior validation studies have focused on this population.19,20 These studies were conducted when there was general under-recognition of the existence of HFpEF, before the obesity related phenotype of HFpEF was reported,5 and subsequently validated with treatment trials.32–34 Patients with HFpEF were underrepresented in earlier NP studies, and often analyzed together with HFrEF (where NP is highly accurate diagnostically).20–22 This may overestimate diagnostic performance estimates for HFpEF. Ambulatory patients with HFpEF are also poorly represented in prior studies, without adequate inclusion of individuals with obesity. Even more important, the reference standard criteria for these studies was expert consensus20–22, which is poorly discriminatory in chronic HFpEF where the diagnosis may be challenging to ascertain without objective hemodynamics (particularly with obesity),4,24 and adjudication was potentially biased by knowledge of NP levels. In the absence of more rigorous studies using ambulatory patients with HFpEF with robust case ascertainment, expert consensus and guideline documents15,26,27 have extrapolated HFrEF diagnostic NT-proBNP thresholds to ambulatory HFpEF without appropriate validation, while also acknowledging the lower NP observed in HFpEF compared with HFrEF.1,3–10,35 The current study demonstrates that for every 100 patients with suspected ambulatory HFpEF and intermediate pretest probability (and HFpEF prevalence similar to our cohorts of ~57%), application of existing NT-proBNP diagnostic thresholds would result in 24 patients with HFpEF being incorrectly labelled as having NCD, while 15 patients with NCD would be incorrectly diagnosed with HFpEF. This 39% diagnostic error rate emphasizes the limitations of current NT-proBNP based diagnostic thresholds for evaluation of ambulatory HFpEF.
The present data demonstrate that current consensus recommended cutpoints for NP testing15 work poorly in an ambulatory population to diagnose/exclude HFpEF, with unacceptably high false positive and false negative results. However, by utilizing separate rule-out and rule-in diagnostic thresholds that optimize sensitivity and specificity, diagnostic information from NP can be maximized with acknowledgement of an intermediate zone of uncertainty where additional exercise testing is needed (50 to 500 pg/ml with BMI<35 and 35–50 to 220 pg/ml with BMI≥35). These rule-in and rule-out diagnostic thresholds appear most diagnostically informative in individuals with intermediate pre-test probability for HFpEF, emphasizing the Bayesian approach required to utilize NP in evaluation of ambulatory HFpEF, identifying a role for incorporating NP testing into estimation of pretest probability using current scoring instruments. Indeed, 22% of patients with intermediate probability H2FPEF scores achieved BMI stratified NP rule-in or rule-out thresholds, demonstrating modest incremental diagnostic utility with a simple blood test, where additional testing may not be necessary. Although more complicated modeling can also be utilized to quantify HFpEF probability by incorporating clinical characteristics and absolute NT-proBNP values without using diagnostic thresholds36, such approaches are more computationally complex and cannot be applied without an online calculator, limiting widespread clinical use. As a more practical approach for individual patient decision making, the simplified BMI-stratified NP thresholds derived herein can be more easily applied at the bedside. Importantly, application of more complicated models will still frequently result in patients with intermediate probabilities of underlying HFpEF, where additional testing is still required.
Obesity related HFpEF has become the dominant phenotype in the Western world, with the magnitude of obesity strongly related to abnormal hemodynamics and quality of life.5,7,28,37–39 The direct causality of adipose tissue to the HFpEF disease process has now been proven by randomized trials demonstrating reduction in worsening HF and improvement in exercise performance, quality of life and NT-proBNP levels with therapeutic weight loss.32–34 Patients with HFpEF and obesity also derive greater symptomatic benefit from sodium glucose co-transporter 2 inhibitors.40 Multiple studies have indicated that congestion and HF is clinically underrecognized in obese individuals.1,28,31,41 Some have suggested lower diagnostic NP cutpoints in obesity (recommendations ranging from 25–50% lower)42,43, but prior to the present analysis, there has been no objective data to quantify the veracity of these obesity-specific cutpoints. When a recent consensus recommendation for age-adjusted BMI criteria for NT-proBNP based rule-in was applied27, specificity was poor at only 65%, emphasizing the need for evidence-based reference criteria. Differential diagnostic performance of NP was apparent only with more severe obesity (BMI>35 kg/m2) and use of the current validated obesity stratified NT-proBNP thresholds will help enhance diagnosis and thus treatment with agents shown to improve functional status, quality of life,32 and reduce risk of HF hospitalization.44–47 This will also lead to greater representation of patients living with obesity in clinical trials, most of which to date have utilized NT-proBNP based inclusion criteria that generally biases towards exclusion of such patients, reducing generalizability of trial results. This is important since patients living with obesity and lower NT-proBNP levels also derive clinical benefit from treatment.48 This was recently demonstrated in the SUMMIT trial, where the GIP/GLP-1 receptor agonist tirzepatide reduced risk of cardiovascular death or worsening HF by 38% compared with placebo.34 Notably, the median NT-proBNP in SUMMIT was only 175 pg/ml, meaning that most patients would not have been eligible using NT-proBNP entry criteria from other trials. While sodium glucose co-transporter 2 inhibitor outcome trials in HFpEF required elevated NT-proBNP levels, other studies have shown that these agents also improve hemodynamics40 and prevent HF hospitalization49 among those with lower NT-proBNP levels, many of whom have obesity.
Atrial fibrillation is common in HFpEF and serves as a biomarker of underlying left atrial remodeling and dysfunction.4,29,30 Clinical trials commonly use higher diagnostic thresholds to establish the diagnosis of HFpEF among individuals with AF, since NT-proBNP levels are consistently higher in AF compared to sinus rhythm.30 Although this is true, we found that the mere presence of AF was sufficiently indicative of HFpEF, as 98% of patients in both the derivation and validation cohorts with AF, dyspnea, and normal EF had HFpEF when evaluated using invasive testing. This finding has important implications in that current guidelines require a higher NP level in those with dyspnea and AF, and in the present study, over 1 in 7 patients with invasively proven HFpEF and AF would have been told they did not have HF based upon NT-proBNP levels falling below the currently endorsed AF-specific diagnostic cutpoints. On a population level, application of these AF specific cutpoints for diagnosis may only add complexity with risk of misclassification, while depriving many patients with AF and HFpEF of highly effective therapies that can improve health status and reduce risk of HF hospitalization or cardiovascular death. Indeed, in the STEP-HFpEF program, efficacy of semaglutide-facilitated weight loss was significantly greater among patients with AF.50
Although the validation cohort was multicenter and included patients across 4 continents, there was no representation from South America or Africa, so the external validity in these populations requires further study. Our cohorts were also predominantly of White or Asian race, with underrepresentation of individuals of Black race. This limitation is particularly relevant given the lower NT-proBNP levels observed among those of Black race, which would suggest that rule-out and rule-in levels might be even lower in Black individuals particularly with obesity, though further study is needed. Our study results apply to single measurements of NT-proBNP with associated biological variability, and the impact of repeated measurements on diagnostic performance in HFpEF requires further study. The derivation cohort was identified retrospectively from those referred for invasive hemodynamic exercise testing due to unexplained dyspnea and had relatively high prevalence of HFpEF. This introduces bias compared to evaluation in less selected community cohorts with dyspnea in nonacademic setting, who may have lower pretest probability of HFpEF. However, this aspect of study design, was essential to accurately determine case/control status, which is not possible without gold standard testing, and the imprecision in HF diagnosis is a major limitation in the rigor of prior work evaluating the diagnostic utility of NT-proBNP in ambulatory HFpEF.20–22 Although the objective case determination is a strength of this design, the retrospective nature may introduce bias due to inconsistency in NT-proBNP measurement, which was missing at random in 24% of the derivation cohort. However, it seems unlikely that individuals in whom NT-proBNP was not tested would somehow display a more robust association between NP levels and the hemodynamic diagnosis of HFpEF compared to those where NT-proBNP was measured. Sensitivity analysis restricted to those with measured NT-proBNP were consistent with the primary results, and reproduction of the results in six independent validation cohorts adds further validity. Prospective validation in a less-selected symptomatic community cohorts, with adequate representation of obese individuals, lower HFpEF prevalence, and definitive case-control status adjudication with exercise RHC would be helpful to validate these findings. With the observed decline in specificity among older individuals, further evaluation of the utility of natriuretic peptides for rule-in among broader community cohorts of older ambulatory individuals at risk for HFpEF are needed. Although Mayo Clinic is a tertiary referral center, two-thirds of the patients in this series were local, predominantly rural patients, and findings were consistent in this subgroup as well. The external validity of our derivation cohort findings are supported by consistent observations in six different validation cohorts including patients residing across 4 continents, those with Asian race who are non-obese (Validation cohort 3 and 6), those undergoing exercise catheterization for diagnosis outside of Mayo clinic (Validation cohort 1 and 3), those diagnosed with HFpEF non-invasively (Validation cohort 5 and 6) or those diagnosed with HFpEF without exercise testing (Validation cohort 4 and 5).
Conclusion
Currently used NT-proBNP diagnostic thresholds to diagnose or exclude HFpEF provide inadequate sensitivity and specificity when applied in ambulatory patients with chronic dyspnea. Use of separate diagnostic thresholds to optimize operational characteristics allows for better diagnostic performance, particularly when utilized in context of intermediate pre-test probability. Body mass must be considered in diagnostic evaluation, as rule-out and rule-in NT-proBNP thresholds are both lower in patients with grade 2 or greater obesity. NT-proBNP may have less diagnostic utility in patients with dyspnea and AF, as AF itself is a robust biomarker of HFpEF. Patients with intermediate HFpEF probability and non-diagnostic NT-proBNP values require further evaluation with exercise hemodynamics to diagnose or exclude HFpEF.