Prognostic value of brain natriuretic peptide vs history of heart failure hospitalization in a large real‐world population

Abstract Background In heart failure (HF) patients, both natriuretic peptides (NP) and previous HF hospitalization (pHFH) have been used to predict prognosis. Hypothesis In a large real‐world population, both NP levels and pHFH have independent and interdependent predictive value for clinical outcomes of HFH and all‐cause mortality. Methods Linked electronic health records and insurance claims data from Decision Resource Group were used to identify HF patients that had a BNP or NT‐proBNP result between January 2012 and December 2016. NT‐proBNP was converted into BNP equivalents by dividing by 4. Index event was defined as most recent NP on or after 1 January 2012. Patients with incomplete records or age < 18 years were excluded. During one‐year follow up, HFH and mortality rates stratified by index BNP levels and pHFH are reported. Results Of 64 355 patients (74 ± 12 years old, 49% female) with available values, median BNP was 259 [IQR 101‐642] pg/ml. The risk of both HFH and mortality was higher with increasing BNP levels. At each level of BNP, mortality was only slightly higher in patients with pHFH vs those without pHFH (RR 1.2 [95%CI 1.2,1.3], P < .001); however, at each BNP, HFH was markedly increased in patients with pHFH vs those without pHFH (RR 2.0 [95%CI 1.9,2.1], P < .001). Conclusion In this large real‐world heart failure population, higher BNP levels were associated with increased risk for both HFH and mortality. At any given level of BNP, pHFH added greater prognostic value for prediction of future HFH than for mortality.

ability to define disease phenotype, and define populations that are responsive to existing and novel management strategies.
Among the clinical and laboratory metrics that have been shown to predict morbidity and mortality in chronic heart failure patients (HF), two have demonstrated the most promise: previous HF hospitalizations (HFH) and natriuretic peptides (NP), particularly B-type natriuretic peptide (BNP). [1][2][3][4][5][6][7] To date, the experimental designs used to examine the utility of these metrics have had several limitations. For example, previous studies have had limited sample sizes, been constrained by exclusion of co-morbidities, had short follow-up periods, and not reflected a non-selected "real world" population. Furthermore, previous studies have examined binary cut-off values of BNP, above vs below median values, data division in tertiles or quartiles. They have not examined and compared HF with a reduced ejection fraction (HFrEF) vs HF with a preserved EF (HFpEF). While both HFH and NP levels are consistently associated with greater risk for morbidity and mortality in patients with HF, the extent to which these two parameters in combination help to stratify risk compared with either parameter alone remains unclear.
Accordingly, the purpose of this study was to define the rate of HFH and all-cause mortality that occurs over a wide range of BNP in an unselected real world population of patients with HF and determine whether the addition of a previous HFH independently altered the morbidity and mortality rates predicted by BNP in this population.

| Data source
Data were extracted from a US wide, real world, data repository from the Decision Resources Group. This data links medical encounter claims, prescription claims, and electronic medical records (EMR) to provide longitudinal patient-level data covering the majority of the US healthcare system. The data were collected from the four largest clearing houses in US and covers over 1.8 million health care providers and 300 million patients. These data consist of records for tests ordered, test results, diagnoses, comorbidities, medications, therapies, patient demographics, healthcare encounters, and death. Encounters recorded in the insurance claims included in-patient/out-patient/ emergency/urgent-care healthcare provider visits, hospital admissions, and nursing home, rehabilitation facility, or hospice encounters.

| Cohort design and eligibility
HF patients were identified from the data repository that had at least one measurement of BNP or NT-proBNP test result available between 1 January 2012 and 31 December 2016. In these patients, the first BNP or NT-proBNP result on or after 1 January 2012 was used as index event. An individual was identified as being a HF patient based on their ICD-9-CM or ICD-10-CM diagnoses for HF (Table S5) in any field in the insurance claims or EMR data on or before index event. Those with age <18 years, discontinuous insurance enrollment or missing encounter types were excluded. The status of insurance enrollment was determined using encounters with the healthcare system that resulted in an insurance claim, which included medication refill, healthcare provider visits admissions, nursing home, rehabilitation, or hospice.

| Outcomes
All-cause mortality and HFH rate were examined across seven differ-  (Table S7 for the list of codes used). The date of death was determined using last-updated-field in the EMR and from discharge date and status from healthcare encounters. Only when these two dates were not available, we needed to approximate using the last date in prescriptions, lab tests, or encounter data.
Hospitalizations of all types were categorized as being heart failure related (HFH) or not based on presence of comprehensive list of diagnosis codes reported in Table S5. The HFHs were determined using presence of primary or admission diagnosis of HF from inpatient hospital claims, or an MS-DRG of 291, 292, or 293. In instances wherein multiple claims were generated from a single in-patient hospitalization event, the claim from and claim through dates were used for consolidation of a hospitalization event. Any hospitalization spanning the index event was deemed to be a pre-index event and not counted toward the post-index follow-up period.

| Sub-group analysis
The impact of BNP on clinical outcomes was examined in patients with HFH in previous year vs those without and patients with heart failure and a reduced ejection fraction (HFrEF) vs heart failure and a preserved ejection fraction (HFpEF). In a cohort in which BMI and HF type were known, we adjusted the BNP and examined the outcomes of HFH rate and mortality rate. This adjustment was based on the Guide-HF clinical trial, design, and rationale. BNP was adjusted by increasing it by 75 pg/ml for a HFpEF patient, and by 4.1% for every unit increase in BMI above 25 units. 8

| Patients with HFH in previous year vs those without
Hospitalizations of all types were adjudicated as being heart failure related or not. A count of the HFH in the 365 days preceding and including the index date was used to classify patients into two groups, "no previous year HFH" and "previous year HFH".

| HFrEF vs HFpEF
HFrEF was identified by the presence of ICD-9-CM code of 428.2 (systolic heart failure) or ICD-10-CM code I50.2 (systolic [congestive] heart failure) available in the healthcare encounters. HFpEF was identified by presence of ICD-9-CM code of 428.3 (diastolic heart failure) or ICD-10-CM code I50.3 (diastolic [congestive] heart failure). In cases where an echocardiographic test result was known and the ejection fraction was less than 40, the patient was characterized as HFrEF patient regardless of the diagnosis code.

| Statistics
Continuous variables are presented as mean ± sd and were compared using a Student t test unless otherwise stated. Categorical variables are presented as n (%) and were compared using chi-squared test.  Demographic data for the study cohort are presented in Table S1.

| Correlations between BNP and heart failure hospitalization rate
The risk of HFH increased as the BNP level rose ( Figure 1A. Patients with BNP ≥1500 pg/ml had highest risk of HFH (RR: 3.3; 95% CI 3.0-3.5) followed by BNP 1000 to 1249, 1250 to 1499, 750 to 999, 500 to 749, 250 to 499 and 0 to 249 groups. Compared with BNP 0 to 249 pg/ml, the risk ratios incrementally increased with each subsequent BNP level, except group 1250 to 1499, which has nonsignificant difference with group 1000 to 1249. For example, as shown in Figure 1A, the RR for HFH is 1. 8

| Correlations between BNP and all-cause mortality
The risk of all-cause death increased as the BNP level rose as shown in Figure 1B. Patients with BNP ≥1500 pg/ml had highest risk of death (RR: 3.9; 95%CI 3.7-4.2) followed by BNP 1250 to 1499, 1000 to 1249, 750 to 999, 500 to 749, 250 to 499 and 0 to 249 pg/ml groups. Compared with BNP 0 to 249 pg/ml, the hazard ratios incrementally increased with each subsequent BNP level. For example, as shown in Figure 1B,    However, these differences did not reach statistical significance in all of the BNP brackets (Table S2). By contrast, for any given range of BNP, the rate of all-cause mortality was similar in the HFpEF patients compared with the HFrEF patients ( Figure 5).

| Effect of age on BNP and correlations between BNP and HFH and all-cause mortality
Overall BNP levels increased with the increasing age brackets examined ( Figure S1). Additionally, at every age, the risk of HFH and allcause mortality increased as a function of increasing BNP brackets ( Figure S5 and S6).

| Outcomes for BNP adjusted for LVEF, BMI, and renal function
For any given level of heart failure severity, BNP was altered in the presence of Afib, LVEF and other factors. Risk ratios for HFH and mortality were assessed using BNP levels that were adjusted for BMI and LVEF in a subgroup of 3590 patients in whom both were known.  (Table S3) when compared to unadjusted BNP analyses.
In the current study, outcomes were adjusted for age, gender, comorbidities, and previous year HFH. Among

| Outcomes according to BNP vs NT-proBNP
The index NP tests were a combination of BNP and NT-proBNP, such that 64.4% of index test results were from BNP, and the remaining 35.6% were from NT-proBNP. The relationship of NP levels and outcomes remained similar irrespective of the test type (BNP vs NT-proBNP) used in the analysis.

| DISCUSSION
In heart failure patients, both brain natriuretic peptide and previous HF hospitalization predict prognosis. [1][2][3][4][5][6][7] However, this association has not been reported over a wide range of BNP levels and analyzed in patients with and without prior HFH for both HFH and all-cause mortality in a large, real-world population. Data from the current study support several clinically relevant findings. Over a very wide range of BNP, subdivided into seven ranges, the risk of HFH increased nearly 26% for every doubling of the BNP value. These differences persisted over time, with risk in 6 to 12 months 20% higher. The same findings were true for all-cause mortality. The risk of all-cause mortality increased nearly 31% for every doubling of the BNP value. When a history of a previous HF hospitalization was added to the risk model, there was an important added value in predicting future HFH, but less added value in predicting all-cause mortality. These finding were consistent across HF phenotypes defined by EF and age. The risk of HFH and the risk of all-cause mortality increased in both HFrEF and HFpEF patients and at every age bracket as a function of increasing BNP. It is F I G U R E 4 Heart failure hospitalizations (HFH) stratified by heart failure phenotype and B-type natriuretic peptide (BNP) levels using a Nelson-Aalen analysis. Panel A, HFH at each BNP range in patients with heart failure and a reduced ejection fraction (HFrEF). Panel B, HFH at each BNP range in patients with heart failure and a preserved ejection fraction (HFpEF) important to note that, these findings were obtained in a large, completely "unselected" (real-world) population with no exclusions for co-morbidities or any other factors that would routinely be done in RCTs.
This study represents the largest cohort report showing the association of natriuretic peptides (NP) and heart failure hospitalizations (HFH) as predictors of outcomes in HF patients. Very specifically, this study showed that higher BNP levels were associated with increased risk for both HFH and mortality. The unique observation from this study is at any given level of BNP, previous HF hospitalization added greater prognostic value for prediction of future HFH than for mortality. Similarly, regardless of whether the patient had HFH in previous year, the knowledge of BNP added greater prognostic value about future HF hospitalization and, most importantly, about mortality.

| Practical applications
The data and analysis provided by this study should help to facilitate assessment and management of health care resources for defined populations, play a pivotal role in planning sample size calculations for randomized clinical trials, enhance our ability to define disease phenotype, and define populations that are responsive to existing and novel management strategies.
For example, one of the most difficult decisions made in the design of clinical trials relates to the oppositional effects of using BNP to enhance specificity and increase event rates vs resultant effects on rates of recruitment. Within eligibility criteria, the higher the required BNP and the additional requirement of prior HFH usually results in a significant increase in the number of patients that will be "screened out" as a potential participant and will result in a slower recruitment rate for a study. A proper balance between estimated event rates and recruitment rate may be achieved by using the data in the current study. If the primary endpoint is all-cause mortality, the eligibility criteria may focus more heavily on BNP values because the requirement of prior HFH does not appear to predict mortality to as significant a degree. However, if HFH rate is the primary endpoint, adding the requirement of prior HFH will lower the needed BNP level for eligibility. In either circumstance, precise numbers may be useful from the data provided in this study. In addition, the current data suggest that BNP and prior HFH data are practical predictors of events in both HFrEF and HFpEF and at all age groups. These data also point to the facts that the rate of HFH is lower at any given BNP in patients with HFpEF and with increasing age. Uniquely, the rates of all-cause mortality at any given BNP are comparable for HFpEF vs HFrEF.

| Comparisons with RCT or epidemiology based data
The rates of HFH and all-cause mortality obtained in the current study are comparable to both the data bases published in RCTs and epidemiology studies. Examples of data from several recent clinical trials that have incorporated BNP and NT-proBNP in the inclusion criteria, are shown in Table S4. [10][11][12][13][14][15][16][17]

| Limitations
Several limitations to the current data and analysis are noteworthy.
The data from this study cannot be used to predict response to therapies, particularly novel therapies. There is insufficient data to compare the relative value of BNP vs NT-proBNP. For any given level of heart failure severity, BNP is altered by the presence of Afib, BMI, and other factors. In a limited dataset, we have addressed the impact of LVEF and BMI via a BMI adjusted BNP analysis (Table S3), but are unable to assess the impact of atrial fibrillation. These co-morbid factors were not examined within the context of this study or placed into a multifactorial analysis to adjust or normalize for the HFH or mortality rates.
These analyses are beyond the scope of this study.
The influence of the number of prior HF hospitalizations on subsequent cardiovascular events in patients with reduced or preserved EF is important and has been studied in a number of clinical trials including CHARM, I-Preserve and others 18 Belle et al. demonstrated that history of an acute heart failure event is a powerful predictor of adverse cardiovascular outcomes, and the time between the last HFH and enrollment was a powerful predictor of subsequent event rates.
However, given the focus and methodological limitations of the current analysis, we are not able to perform this analysis. In the current study design, we were able to adjust for age, gender, co-morbidities, and previous year HFH as discrete variables and did not account for the number of events and time between the last HF hospitalization and the NP value. We acknowledge that this is a worthwhile analysis that will await further study.
While both BNP and NT-proBNP are widely used to aid diagnosis, assess the effect of therapy, and predict outcomes in heart failure, there is no clear consensus to guide the conversion between them. 19 However, there are examples of conversion factors chosen for recent randomized clinical trials that were used to choose and justify a conversion factor of 4 times BNP, NT-proBNP as used in the current analysis. For example, the eligibility criteria in PARADIGM-HF included elevated natriuretic peptides: BNP ≥150 pg/ml or NT-proBNP ≥600 pg/ml (for patients with HF hospitalization within 12 months, BNP ≥100 pg/ml or NT-proBNP ≥400 pg/ml), that is, a factor of 4. A similar conversion factor was chosen for the ongoing trial Guide-HF (Lindenfeld et al. 22 ). Because the conversion factor likely varies with disease states, with and without atrial fibrillation, with increasing age and decreasing renal function Rorth et al. 19 concluded that "there is no single, simple, conversion ratio of NTproBNP to BNP and factors, such as, atrial fibrillation, age, and renal function need to be taken into account." There are alternative conversion formulas proposed by Kasahara et al. 20 and Yeo et al. 21 that incorporate age, body mass index, hemoglobin, renal function, sex, and atrial fibrillation; however, these variables were not available in the context of the current study. The use of ICD codes to designate patients as having HFrEF vs HFpEF has significant limitations because it is not obligatorily linked to an objective measure of ejection fraction. It does; however, take advantage of the fact that the clinical provider who knows the patient the best is the one that can impact the choice the ICD code. For these reasons, our analysis reported in the manuscript used a conservative approached wherein we classified a patient as HFrEF, if either they had a diagnosis of systolic HF (ICD code) OR their EF was <40.
Where, a patient was classified as HFpEF when their claims carry a diagnosis of diastolic HF and their EF was not <40 if it was known.

| CONCLUSION
In a large real-world heart failure population, higher BNP levels were prognostic of both future HFH and mortality whereas at any given level of BNP, previous HFH added prognostic value to a greater degree for prediction of future HFH than for mortality.