Characterization and Prediction of Natriuretic Peptide “Nonresponse” During Heart Failure Management: Results From the ProBNP Outpatient Tailored Chronic Heart Failure (PROTECT) and the NT-proBNP–Assisted Treatment to Lessen Serial Cardiac Readmissions and Death (BATTLESCARRED) Study
Hanna K. Gaggin MD, MPH,
Cardiology Division, Massachusetts General Hospital, Boston, MA
Address for correspondence: James L. Januzzi, Jr, MD, Cardiology Division, Massachusetts General Hospital, 55 Fruit Street, Yawkey 5984, Boston, MA, 02114 E-mail: email@example.com
Many proven heart failure (HF) therapies decrease N-terminal pro B-type natriuretic peptide (NT-proBNP) values over time, yet some patients have an NT-proBNP >1000 pg/mL following treatment, which is associated with poor outcomes. A total of 151 patients with left ventricular systolic dysfunction were treated with aggressive HF therapy in the ProBNP Outpatient Tailored Chronic Heart Failure (PROTECT) study. Clinical characteristics and NT-proBNP were measured at each visit during 10 months. In this post hoc analysis, biomarker nonresponse was defined as an NT-proBNP >1000 pg/mL and its relationship with echocardiographic and clinical characteristics and outcomes were explored. A risk model predictive of nonresponse was derived and externally validated. A rising NT-proBNP over time was associated with increased cardiovascular event rates while a decreasing NT-proBNP was associated with better clinical outcomes (58.2% vs 27.6%, P=.001). A higher percentage of time in biomarker response was associated with lower event rates (P<.001). Importantly, responders showed improved left ventricular remodeling parameters (all P<.001), while nonresponders did not. A risk model for predicting nonresponse had a C statistic of 0.82 (P<.001) and predicted outcomes well. Using data from the NT-proBNP–Assisted Treatment to Lessen Serial Cardiac Readmissions and Death (BATTLESCARRED) cohort, the risk score was validated for its ability to predict nonresponse (C statistic 0.73, P<.001). Serial changes in NT-proBNP inform risk for adverse outcome and are associated with prognostically meaningful metrics. Prediction of future NT-proBNP nonresponse to HF therapy is possible.
Heart failure (HF) is a global epidemic, affecting 23 million people worldwide. While there have been significant recent advances in the medical management of chronic HF (including the use of β-blockers, angiotensin-converting enzyme inhibitors, angiotensin II receptor blockers, and aldosterone blockers), the ability to characterize, monitor, and predict a patient's response to HF therapy is poor. One patient's symptoms may improve dramatically with therapy while another patient's symptoms may not respond to the same guideline-derived medical therapy. In addition, the benefit (or lack thereof) from chronic HF therapy is usually not apparent until many months or years after the initiation of therapy, often after an adverse outcome. Thus, an objective method to monitor a patient's response and an early identification of nonresponders to HF therapy is needed for better outcomes and for an efficient use of limited resources.
Natriuretic peptides such as B-type natriuretic peptide (BNP) and N-terminal pro BNP (NT-proBNP) are biomarkers of cardiomyocyte stretch. These biomarker levels typically rise and fall in parallel with parameters such as cardiac filling pressures and are used in the diagnosis of HF.[5, 6] Elevated NT-proBNP has been validated for its ability to predict adverse outcomes in chronic HF,[7-9]and the role of serial natriuretic peptide measurement[9, 10] and the novel concept of biomarker response to HF therapy have been introduced.[11, 12] While the concept of biomarker-guided HF management has been examined in several trials with various designs,[13-18] data are limited on how exactly HF biomarker response should be defined and what biomarker response may mean relative to biomechanical and clinical outcomes. A better characterization of natriuretic peptide response to HF therapy, along with the ability to better understand those patients likely to be a “nonresponder” would be essential knowledge in further studying guided therapy of HF.
In this study our aim was to characterize the meaning of NT-proBNP nonresponse and to examine whether this biochemical nonresponse to HF therapy can be predicted prior to initiation of therapy. We did so by using the well-characterized cohort from the ProBNP Outpatient Tailored Chronic Heart Failure Therapy (PROTECT) study (www.clinicaltrials.gov, NCT #00351390).[15, 19]
The PROTECT study was a prospective, single-center, randomized controlled trial[15, 19] that compared NT-proBNP-guided HF care (in addition to guideline-compliant standard of care) vs standard of care management (without NT-proBNP guidance) in patients with HF due to left ventricular systolic dysfunction. This study complies with the Declaration of Helsinki, and the Partners Healthcare institutional board review board has approved the research protocol and informed consent has been obtained from the patients.
We included 151 outpatients with New York Heart Association (NYHA) class II through IV HF, left ventricular ejection fraction (LVEF) ≤40%, and recent acute HF decompensation.
In both study arms, patients had scheduled clinic visits every 3 months at a minimum and any additional visits as needed to achieve maximally tolerated, guideline-recommended target doses of medications for a total of 908 visits. Following any treatment adjustment in either arm, patients were seen at 2 weeks (±4 week intervals) until goals were met or the patient required hospitalization. In the NT-proBNP–guided care arm, an additional goal to standard of care was to lower and sustain NT-proBNP concentration ≤1000 pg/mL and additional visits and medication adjustments were made for that purpose.
At each clinic visit, the Minnesota Living With Heart Failure Questionnaire was filled out to assess quality of life, and a blood sample for standard laboratory testing and NT-proBNP measurement (Elecsys proBNP; Roche Diagnostics, Indianapolis, IN) were both obtained. An estimated glomerular filtration rate was calculated using the Modification of Diet in Renal Disease formula. Echocardiography was performed at study enrollment and at completion of study procedures. The main echocardiographic parameters measured were LVEF, left ventricular end-systolic volume index (LVESVi), left ventricular end-diastolic volume index (LVEDVi), an estimation of left ventricular filling pressures or E/E′, and right ventricular systolic pressure.
“Response” of NT-proBNP to therapy intervention was defined as achievement of a value of ≤1000 pg/mL. For the purpose of this analysis, we used the primary endpoint of the PROTECT study, which was total cardiovascular events, a composite outcome defined as worsening HF (new or worsening symptoms/signs of HF requiring unplanned intensification of decongestive therapy), hospitalization for acutely decompensated HF (including treatment with intravenous diuretic in the emergency department setting without hospitalization), clinically significant ventricular arrhythmia, acute coronary syndrome, cerebral ischemia, and cardiac death.
Differences in categoric variables between responders and nonresponders were assessed using χ test, while for continuous variables, the Student t test, Mann-Whitney U test, or Kruskal-Wallis were employed as appropriate. Continuous variables were expressed as means±standard deviation or medians (25th–75th percentile), the latter employed in the context of non-normality. Repeated measures were analyzed using Wilcoxon tests.
The phenomenon of NT-proBNP nonresponse was assessed in numerous ways. Firstly, we evaluated the event rates of patients as a function of absolute achieved NT-proBNP at baseline and at the end of the study. Next, we examined outcomes as a function of dichotomous categorization (≤1000 vs >1000 pg/mL) at both baseline and final measurement. Then, we examined associations between NT-proBNP “slope” (defined as direction of change from baseline to final; a “positive” slope would define an aggregate rise in NT-proBNP values, while a “negative” slope an aggregate fall) with outcomes. Lastly, in order to consider the fact that certain patients achieved an NT-proBNP ≤1000 pg/mL for longer periods than others in the study, we examined the association between percentage of “time in response” and outcomes. To do so, we integrated NT-proBNP values with a time element adjusted for each interoffice measurement, with a higher time in response indicating a greater duration spent ≤1000 pg/mL.
Baseline clinical, biochemical, and echocardiographic variables predictive of nonresponse were sought using univariable screening, with a retention P=.10. Candidate variables identified in univariable screening were then entered into a multivariable model, using forward stepping logistic regression to identify independent predictors. Verification of goodness of fit was confirmed with the Hosmer-Lemeshow test. All candidate variables were started out of the logistic model and each was entered in order of largest to smallest χ statistics only, whereas the maximum likelihood estimate of the corresponding regression parameter was significantly different from zero at P<.10. An independent variable was removed from the model only when its corresponding regression parameter was not significantly different from zero at P>.10. First-order testing for interactions was performed to evaluate for interaction(s) between candidate independent variables. Odds ratios for nonresponse were generated and expressed with 95% confidence intervals.
After identification of significant independent predictors of nonresponse, β coefficients for each variable were rounded to the nearest whole integer to generate a relative weight for each variable. An integeric nonresponder risk score, composed of the sum of applicable weighted variables, was calculated for each individual. The observed prevalence of nonresponders at each score level was first compared with the predicted prevalence for each score parameter from the logistic model. Receiving operating characteristic curve analysis was used to determine the score value with the optimal sensitivity and specificity for the antecedent prediction of subsequent nonresponse.
To externally validate the PROTECT nonresponder risk score, a patient cohort from another prospective study of NT-proBNP-guided HF therapy, the NT-proBNP–Assisted Treatment to Lessen Serial Cardiac Readmissions and Death (BATTLESCARRED) study was used to calculate the risk scores and the observed proportion of nonresponders in this group was determined.
In all statistical analyses, a 2-tailed P<.05 indicated statistical significance. All analyses were performed with SAS (version 9.2; Cary, NC) or PASW (version 17 and 18; Chicago, IL) software.
Table 1 details baseline patient characteristics for the trial by responder category. Nonresponder patients were more likely to be older (66.8±13.7 years vs 58.7±13.0 years, P<.001), were more likely to have NYHA functional class III or IV symptoms (65.1% vs 38.6%, P=.003), had more prevalent atrial fibrillation (47.7% vs 30.8%, P=.05), had more congestion on physical examination, and significantly higher baseline NT-proBNP concentrations (2498 [1770–5237] vs 1269 [448–2541] pg/mL, P<.001). Nonresponders were also more likely to have worse renal function (estimated glomerular filtration rate 56.0±17.6 mL/min/1.73 m vs 67.0±23.1 mL/min/1.73 m, P=.001).
Table 1. Baseline Characteristics of the Study Cohort as a Function of NT-proBNP Response
Abbreviations: COPD, chronic obstructive pulmonary disease; eGFR, estimated glomerular filtration rate; HF, heart failure; LVEF, left ventricular ejection fraction; NT-proBNP, amino-terminal pro-B type natriuretic peptide; NYHA, New York Heart Association symptom severity. Continuous variables are expressed as means±standard deviation or median with 25th and 75th percentile as appropriate.
Elderly (≥75 y)
NT-proBNP study arm
Body mass index, kg/m2
Heart rate, beats per min
Systolic blood pressure, mm Hg
Jugular venous distension
Lower extremity edema
Baseline laboratory results
eGFR, mL/min/1.73 m2
NT-proBNP Changes Over Time
Individual patients' NT-proBNP values over time were graphed for nonresponders and responders (Figure 1).
NT-proBNP Nonresponse and Echocardiographic Changes
Nonresponders showed little if any improvement over time in most echocardiographic measures when compared with responders (Figure 2). Although nonresponders showed some improvement in median LVEF (from 27% to 32%, P=.04) the magnitude of improvement was less than that seen in NT-proBNP responders (29% to 37%, P<.001). Among nonresponders there was no significant improvement in median LVESVi (from 56 mL/m to 56 mL/m, P=.44) or LVEDVi (79 to 81 mL/m2, P=.82), while the responders had significant improvement in the same remodeling indices (LVESVi from 58 mL/m2 to 40 mL/m2, P<.001; LVEDVi 82 mL/m2 to 64 mL/m2, P<.001).
Consistent with above differences in remodeling indices as a function of NT-proBNP response, there was no significant improvement in diastolic filling indices among nonresponders (E/E′ from 13[9–17] to 12[10–17], P=.32) while right ventricular systolic pressure (RVSP) was only marginally lowered (from 46±12 mmHg to 42±11 mmHg, P=.05). In contrast, there was a significant improvement in responders, with E/E′ decreasing from 10[7–13] to 9[6–12] (P=.040) and right ventricular systolic pressure decreasing from 41±10 mm Hg to 34±8 mm Hg, (P<.001).
NT-proBNP Nonresponse and Clinical Outcomes
As expected, patients who began and ended the study with an NT-proBNP ≤1000 pg/mL had the fewest cardiovascular events (.50±1.23 event per patient) while those who had an NT-proBNP level >1000 pg/mL throughout the study had the highest event rates (1.57±1.9 events per patient, P=.003; Figure S1). Patients who began with an NT-proBNP ≤1000 pg/mL but ended above this value had a numerically higher event rate (.71±1.11 events per patient) compared with those who started off higher but ended the study with a level ≤1000 pg/mL (.46±1.04 events per patient), but this was not statistically significant.
As expected, an increasing NT-proBNP concentration (ie, a “positive” slope) over time was associated with increased occurrence of cardiovascular events compared with a stable or a decreasing NT-proBNP value over time (58.2% vs 27.6%, P=.001).
Further, using all visits, we considered the therapeutic time in response for each patient using time-integrated values for NT-proBNP ≤1000 pg/mL. With a greater percentage of time in response over the follow-up period, we found a robust decrease in the number of cardiovascular events (Figure 3, P<.001).
Office Visits and Therapy Changes
There were no differences in the number of office encounters between nonresponders and responders (5.0 vs 5.0 visits, P=.19). Consistent with the goal of optimal medical therapy in HF, the care delivered in both subgroups of the PROTECT study was highly compliant with HF practice guidelines, with high use of β-blockers, angiotensin-converting enzyme inhibitors, angiotensin receptor blockers, and aldosterone inhibitors (Table S1). Differences in medication use tended to favor the responder category, although none of these differences were significant. An example of medication titrations and NT-proBNP concentrations in two HF patients (1 nonresponder and 1 responder) are shown in Figure S2.
Nonresponse, Adverse Events, and Quality of Life
There were no statistically significant differences in the rates of treatment-related adverse events between nonresponders and responders (Table S2). Compared with responders, who had a significant improvement in quality of life from baseline to the end of study (from 27.5 [13.8–51.0] to 15.5 [2.0–39.3], P<.001), the change in Minnesota Living With Heart Failure score in the nonresponders was not significant (from 30.0 [14.0–43.0] to 21.0 [8.0–50.0], P=.12). In addition, responders trended toward a greater magnitude of improvement in quality-of-life measures during the study (−8.0 [−21.3 to −1.0] vs −5.0 [−15.0 to 8.0], P=.06).
Predicting Nonresponse: the PROTECT Nonresponder Risk Score
Of the original 18 candidate variables examined in multivariable logistic regression analysis for the prediction of future nonresponse, 4 independent variables remained statistically significant. These variables are detailed in Table 2 and lead to a possible final score of 0 to 7 points. The full model containing these predictors was robust and statistically significant (χ [5, N=142] 50.01, P<.001), indicating that it was well able to identify future nonresponders. The model as a whole explained between 29.7% (Cox and Snell r) and 40.2% (Nagelkerke r) of the variance in nonresponder status and correctly identified 75.4% of cases.
Table 2. Independent Clinical and Biochemical Predictors of Nonrespondersa
Abbreviations: NT-proBNP, amino-terminal pro-B type natriuretic peptide; NYHA, New York Heart Association. aExpressed with their respective odds ratio, 95% confidence interval (CI), β coefficient, and the integeric score derived from each.
Heart rate <60 beats per min
NYHA class III or IV
History of atrial fibrillation
Examining patients as a function of their score, the overall mean PROTECT nonresponder risk score was 2.75 points. The mean score among nonresponders was significantly higher than those of responders (3.34 vs 1.75, respectively; P<.001). Receiving operating characteristic curve analysis had an area under the curve (AUC) of 0.82 for predicting nonresponse (Figure 4A P<.001), and a PROTECT nonresponder risk score of ≥3 points was found optimal for maximizing overall accuracy with a sensitivity of 79%, specificity of 68%, positive predictive value of 79%, and negative predictive value of 68%. This value was thus selected as the threshold for identification of nonresponders. As shown in Figure 4B, a rising percentage of nonresponders were seen with a higher score.
Among patients from the BATTLESCARRED study, 223 had data that defined nonresponse status at 1 year and comprised our validation cohort. Receiving operating characteristic curve analysis of the PROTECT nonresponder risk score in the BATTLESCARRED cohort produced AUC of 0.73 (Figure 4C; 95% confidence intervals, 0.65–0.81; P<.001) with a sensitivity of 67%, specificity of 67%, positive predictive value of 81%, and negative predictive value 50%. The risk score distribution was similar to that of the derivation cohort, with higher scores associated with higher rates of nonresponse (Figure 4D).
The PROTECT Nonresponder Risk Score and Clinical Outcomes
As would be expected based on its ability to predict higher post-treatment NT-proBNP values, higher scores were associated with higher event rates (1.44 vs 0.44 events per person, P<.001; Figure 5), and cumulative hazard (Figure S3, P<.001).
In this study, we have defined NT-proBNP nonresponse as a value of >1000 pg/mL and have characterized biomarker response to HF therapy in several ways, showing it to be strongly prognostic of improved clinical outcomes. On the other hand, failure to achieve NT-proBNP values ≤1000 pg/mL was strongly associated with adverse outcome and with less favorable changes in prognostically powerful echocardiographic parameters. Lastly, we derived and validated a concise risk score that predicts nonresponse to HF therapy using simple baseline characteristics.
The relationship between biochemical response to HF therapy and clinical outcomes was consistently demonstrated whether taken as a single achieved value, a trend over time, or most importantly when considering serial NT-proBNP measurement, as a novel concept of percentage of time spent in response. While therapeutic time in response is well-accepted with regard to predicting benefits of drug therapy with agents such as warfarin sodium, to our knowledge this is the first link between NT-proBNP measurements and the concept of percentage of time in response as a function of outcomes in HF.
Importantly, the relationship between biochemical response and clinical outcomes is further supported by mechanistic evidence of favorable ventricular remodeling. Patients who achieved NT-proBNP ≤1000 pg/mL had a robust improvement in remodeling parameters, as well as diastology and right ventricular pressures, while biomarker nonresponse was associated with weaker or nonexistent improvement of the same parameters despite aggressive HF therapy. Taken together, our results may elucidate why previous guided therapy studies not achieving adequate natriuretic peptide suppression were negative for their primary endpoint,[12, 16-18] set the stage for future larger trials of biomarker-guided HF therapy, and provide guidance for clinicians currently utilizing BNP or NT-proBNP in serial fashion while monitoring patients during treatment for chronic HF.
When faced with a nonresponder to HF therapy, a prompt exploration should be undertaken to determine an underlying cause of nonresponse such as lifestyle or medication noncompliance or a subclinical HF decompensation. Importantly, the optimal application of available guideline-supported medications that decrease NT-proBNP values (as well as improve HF mortality) such as β-blockers, angiotensin-converting enzyme inhibitors, angiotensin receptor blockers, aldosterone inhibitors, and cardiac resynchronization therapy, should be reviewed with the goal to achieve guideline-derived medical therapy as per consensus practice guidelines. When all attempts to optimize medical care have been exhausted and natriuretic peptides remain elevated, our data indicate a high likelihood for adverse ventricular remodeling, poor quality of life, and unfavorable prognostic outlook. It is fair to envision the next step for such a patient to be considered for referral for advanced HF management strategies, such as mechanical support or listing for heart transplantation. Based on serial NT-proBNP values plotted over time for all patients in the study, response to aggressive HF therapy appears to materialize early, within about 3 to 4 months of therapy.
We derived and validated a risk model for predicting NT-proBNP nonresponse to HF management at the very first office visit. This risk model, based on simple baseline characteristics routinely available at any outpatient clinics–NT-proBNP, heart rate, NYHA functional class, and a history of atrial fibrillation performed well on the derivation cohort as well as on the validation cohort from the BATTLESCARRED study. The robust validation results were shown despite the fact that the cohort from Christchurch, New Zealand, was heterogeneous in baseline demographics, guided therapy, and outcomes from the PROTECT study cohort. The overall result of the BATTLESCARRED study was negative, although benefit was shown in patients younger than 75 years in a subgroup analysis. Interestingly, age was not included in the final risk model. Once NT-proBNP was entered into the model, age was no longer predictive of nonresponse and may reflect the combination of information embedded in NT-proBNP. A higher score on the risk model is probably reflective of a more advanced nature of HF in nonresponders as evidenced by the presence of atrial fibrillation, higher NYHA class, and higher NT-proBNP values. Nonresponders with lower heart rate at baseline tended to be patients who appeared to have underlying conduction disease, as they were not taking high-dose β-blockers, also preventing aggressive uptitration of β-blockers during the trial. Thus, knowledge of the potential for future nonresponse might assist clinicians in deciding on earlier referral for more invasive or palliative strategies.
Limitations of our post hoc study include the fact that it was based on a small cohort from a single tertiary care academic medical center in Boston. However, the smaller size allowed for extensive characterization of the cohort with more than 900 office visits and NT-proBNP concentrations tracked across all visits. Additionally, while the PROTECT risk score for predicting nonresponse was accurate and had excellent operating characteristics, our validation was retrospective rather than prospective. Lastly, this was not a prespecified analysis, thus our dataset was not specifically powered for the examination of nonresponse.
Despite our limitations, our results on the novel concept of nonresponse to HF therapy were consistent and robust across various modalities and show promise for improving the care and resource allocation in HF. Large prospective studies are needed to further explore this topic.
Funding: This work was supported in part by Roche Diagnostics Corporation (Indianapolis, IN). The sponsor had no role in design, data collection, analysis, manuscript preparation, interpretation, or decision to submit the manuscript for publication. Drs Rehman, Mohammed, Bhardwaj, and Gaggin are supported by the Dennis and Marilyn Barry Cardiology Fellowship (Boston, MA); Dr Gaggin is supported by the Ruth and James Clark Fund for Cardiac Research Innovation (Boston, MA); and Dr Januzzi is supported in part by the Roman W. DeSanctis Clinical Scholar Endowment (Boston, MA).
Relationship with Industry: None of the authors have any disclosures other than Dr Januzzi, who reports receiving compensation in the form of grant support (Roche Diagnostics, Siemens, Critical Diagnostics, and BRAHMS, GmBH).