Improvements of predictive power of B-type natriuretic peptide on admission by mathematically estimating its discharge levels in hospitalised patients with acute heart failure

Backgrounds Earlier studies showed that in patients with heart failure (HF), circulating levels of B-type natriuretic peptide (BNP) at hospital discharge (BNPdis) are more predictive of prognosis than BNP levels on admission (BNPad). However, the mechanism underlying that difference has not been fully elucidated. We examined the association between confounding factors during hospitalisation and BNPdis in patients with HF. Methods We identified patients admitted to our hospital for HF (BNPad ≥100 pg/mL). Estimated left ventricular end-diastolic pressure (eLVEDP) was calculated using echocardiographic data. To identify the factors associated with the relation between BNPad and BNPdis, we performed a stepwise regression analysis of retrospective data. To validate that analysis, we performed a prospective study. Results Through stepwise regression of the patient data (n=688, New York Heart Association 3–4, 88%), we found age, blood urea nitrogen and eLVEDP to be significantly (p<0.05) associated with BNPdis. Through multivariate analysis after accounting for these factors, we created a formula for predicting BNP levels at discharge (predicted-BNPdis) from BNPad and other parameters measured at admission (p<0.05). By statistically adjusting for these factors, the prognostic power of BNPad was significantly improved (p<0.001). The prospective study also confirmed the strong correlation between predicted-BNPdis and BNPdis (n=104, r=0.625, p<0.05). Conclusion This study showed that statistically accounting for confounding factors affecting BNP levels improves the predictive power of BNP levels measured at the time of hospital admission, suggesting that these confounding factors are associated with lowering predictive power of BNP on admission. Trial registration number UMIN 000034409, 00035428.


INTRODUCTION
B-type natriuretic peptide (BNP) is widely used as a predictive biomarker in patients with heart failure (HF). 1 However, earlier studies showed that in patients with acute HF, the predictiveness of BNP levels measured at hospital admission (BNP ad ) for clinical outcomes is inferior to BNP levels measured at discharge (BNP dis ). 2 3 Although the reason for the insufficient predictive power of BNP ad compared with BNP dis in these patients has not been fully elucidated, several confounding factors are well known to influence circulating BNP levels (eg, left ventricular end-diastolic pressure (LVEDP)

Key questions
What is already known about this subject?
► Predictive power of B-type natriuretic peptide (BNP) levels at hospital admission (BNP ad ) is inferior to that of BNP levels at discharge (BNP dis ) in patients with acute heart failure (HF).However, the mechanism underlying these differences has not been fully elucidated.
What does this study add?
► Through multivariate analysis after accounting for known confounding factors related to circulating BNP levels, we created a formula for predicting BNP levels at discharge (predicted-BNP dis ) from BNP ad and other parameters measured at admission.► This study showed that statistically accounting for confounding factors affecting BNP levels improves the predictive power of BNP levels measured at the time of hospital admission.

How might this impact on clinical practice?
► This predicted-BNP dis has superior predictive power for clinical outcomes to BNP ad .This may contribute to the risk stratification of acute HF and primary intensive care management in acute vulnerable phases of HF. on September 14, 2023 by guest.Protected by copyright.
5][6][7] However, how these parameters influence the lower predictability of BNP ad than that of BNP dis remains uncertain.Given the importance of clinical risk stratification for hospitalised patients with heterogeneous clinical syndromes, this study aimed to identify the confounding factors affecting BNP ad that are associated with BNP dis .We also tested the hypothesis: whether statistical adjustments of these confounding factors are related to predictive power of BNP ad improvements in patients with acute HF.

Study design
This was a cross-sectional study of patients with HF admitted to the National Cerebral and Cardiovascular Center of Japan.

Retrospective study
Included in the retrospective study were 688 patients hospitalised for HF between January 2013 and March 2016 (BNP on admission: ≥100 pg/mL).We excluded patients who did not undergo a blood test and echocardiography, who underwent implantation of a left ventricular (LV) assist device (n=3), or who died in the hospital (n=24) during the corresponding hospitalisation.We also excluded patients who underwent mitral valve surgery (n=85) due to its influence on transmitral flow and septal mitral annular velocity.
Diagnosis of HF was based on the Framingham criteria. 8hether or not a HF episode met the Framingham criteria was determined by each attending physician and an investigator (H Takahama) via medical record review.We excluded patients who did not meet the criteria, as judged by the investigator and the attending physician for each patient.According to the guidelines of the Japanese Circulation Society, 9 we defined the cut-off value of plasma BNP level for diagnosis of HF as 100 pg/mL.

Prospective study
We prospectively collected data from 104 patients between January and June in 2019 based on the same criteria used for the retrospective study.

Echocardiography
Through medical chart review, we retrospectively reviewed the echocardiography data collected during the hospitalisation.LV dimensions were measured according to the American Society of Echocardiography guidelines. 10LV ejection fraction (EF) was measured using the modified Simpson method or the semiquantitative twodimensional visual estimate method, as described previously. 11Transmitral inflow was measured with pulsedwave Doppler using standard methods as described previously. 12The septal mitral annular early diastolic velocity (e′) was determined with spectral tissue Doppler imaging.LVEDP was calculated as 11.96+0.596× early diastolic transmitral flow velocity (E)/e′, as previously reported. 13asurement of plasma BNP concentration All biochemical analyses were performed as routine clinical examinations.BNP were measured by human brain natriuretic peptide kit (TOSOH corporation, Tokyo, Japan).

Clinical outcomes
After the admission date, we investigated through medical chart review or a letter all causes of death and rehospitalisation for HF.Combined clinical events were defined as all-cause death or rehospitalisation for HF.

Ethics
The study was designed to be carried out without obtaining individual informed consent according to the 'opt-out' principle.Instead, we publicised a summary of the study protocol with the contact information for our office on the institution website, which provided patients with the ability to refuse enrolment to the study.This study protocol was also registered in the Japanese University Hospital Medical Information Network Clinical Trials Registration.

Statistical analyses
Results are expressed as the median and IQR.Fisher's exact test or the χ 2 test was used to compare categorical variables, as appropriate.With regard to baseline patient characteristics, Wilcoxon's rank-sum test was used for comparison of continuous variables between two groups.HRs with 95% CIs and probability (p) values determined using the likelihood ratio test are presented.The area under the receiver operating characteristics (ROC) curve (AUC) and C-statistics were also calculated.AUCs were compared using an algorithm developed by DeLong et al. 14 Pairwise comparisons of the areas under multiple ROC curves were made using the roccomp command in Stata.Multivariate analysis/regression was used to test multiple covariates.All tests were two tailed, and values of p<0.05 were considered significant.All statistical analyses were performed using JMP V.9 statistical analysis software (SAS Institute Japan, Inc, Tokyo, Japan) and Stata V.15 (Stata Corporation LLC, College Station, Texas, USA).

Retrospective data analysis
Using the inclusion and exclusion criteria described in the Methods section, we identified 688 patients with HF from our database.The patient characteristics on admission were as follows (table 1): New York Heart Association class III and IV on admission, 43% and 45%, respectively; median LVEF: 35% (IQR: 24%-55%); median LV end-diastolic diameter (LVEDD), 56 mm (IQR: 48-64 mm); and median plasma BNP levels, 671 pg/mL (IQR: 370-1170 pg/mL).Median hospitalisation length was 19.5 days (IQR: 14-27 days).The patients treated without beta-blockers were often observed in those with valvular regurgitation and with HF with preserved EF.Based on earlier studies, we selected the following clinical Heart failure and cardiomyopathies parameters known to influence circulating BNP levels for analysis: age, sex, 4 LVEDP, 5 15 blood pressure, 16 heart rate, 7 body mass index (BMI), 6 LVEF, 17 18 end-diastolic volume, 17 LVEDD, 18 cholesterol levels, 19 anaemia, 20 renal function, 7 20 atrial fibrillation 21 and diabetes mellitus. 22he results of univariate analyses of the association between BNP dis and the selected parameters are shown in table 2. Age, BMI, systolic blood pressure, diastolic blood pressure, LVEDD, LVEF, estimated LVEDP (eLVEDP), estimated glomerular filtration, blood urea nitrogen (BUN), haemoglobin, hematocrit, total cholesterol and low-density lipoprotein cholesterol on admission were all associated with BNP dis after adjusting for BNP ad .Stepwise analysis identified the following parameters as significantly associated with BNP dis after accounting for BNP ad levels: age, systolic blood pressure, LVEF, eLVEDP and BUN levels (table 3).Subsequent multivariate analysis revealed the parameter estimates of the above factors to be: 3.81 (age), −1.09 (systolic blood pressure), −0.87 6.72 (eLVEDP), 3.48 (BUN) and 0.21 (BNP ad ).Using those data, we developed the following formula to predict BNP dis from BNP ad : the 'predicted-BNP dis '=3.81 × age -1.09 × systolic blood pressure -0.87×LVEF + 6.72 × eLVEDP + 3.48 × BUN + 0.21 × BNP ad -184.5.Next, the analysis of association non-pharmacological intervention on the relation between the parameters used for estimation of the predicted-BNP dis and BNP dis were performed because the therapeutic intervention might influence the discharge levels of BNP.As shown in online supplemental table 1, no statistical significance was found.In addition, there was significant differences in both BNP ad and BNP dis in patients with initial admission and rehospitalisation for HF.As shown in online supplemental table 2, both BNP levels were higher in patients with readmission than those in initial admission.The correlation of the predicted-BNP dis with BNP dis was similar between the patients with initial admission and readmission (p<0.05).

Prospective data analysis
The prospective study revealed similar relationships between BNP ad or predicted-BNP dis and BNP dis (figure 2A,B).

DISCUSSION
Circulating BNP levels are affected by a number of confounding factors during the acute phase of HF and are associated with fluctuations in measured BNP levels during the initial few days after hospital admission or intensive treatments. 23 24These fluctuations are mainly due to changes in ventricular preload and the systemic fluid overload state.Our retrospective study confirmed known confounding factors affecting BNP levels and showed that these factors statistically were associated with   Heart failure and cardiomyopathies lowering the predictive power of BNP ad levels for patient outcomes.This study also showed that after statistically accounting for these confounding factors, the predictive power of the resultant BNP value (predicted-BNP dis ) did not significantly differ from that of BNP dis .Moreover, the finding of the correlation of predicted-BNP dis with BNP dis was validated by obtaining similar results in a prospective study.
Several earlier studies reported on the factors influencing circulating BNP levels.For example, Iwanaga et al 5 measured LVEDP using a LV catheter system and clearly demonstrated that LV wall stress strongly correlated with plasma BNP levels.In addition, factors such as increased cardiac preload likely due to excess body fluid, which may stretch ventricular cardiomyocytes, sharply increases circulating BNP levels in patients with HF.By the time BNP dis levels are measured, however, the patient has reached an appropriate volume state through removal of the excess body fluid.Consequently, BNP dis levels may more closely reflect myocardial quality per se or a 'true ventricular trait'.This may explain why BNP ad levels measured during the acute phase of HF have less predictive power than BNP dis levels.In the present study, we observed that after accounting for several factors, including LVEDP and LVEF, as well as blood pressure and renal function, the predicted BNP levels, which we termed predicted-BNP dis , strongly correlated with BNP dis levels and were equally predictive of patient outcome.Thus, by determining the impact of factors responsible for the difference in predictive power between BNP ad and BNP dis , we were able to shed light on the relationship between these factors and BNP dis .These findings may further our understanding of BNP levels, which are influenced by various factors in patients with acute HF.

Limitation
The present study has several limitations.First, this was a single-centre investigation with a limited number of patients.Nevertheless, we were able to confirm a formula that predicts BNP levels at discharge from clinical parameters at the time of admission.Second, several patients did not undergo a blood test for BNP and E/e′; we excluded these patients from our analysis.Next, it is widely known that the therapeutic intervention might influence the discharge levels of BNP.These therapeutic interventions might also influence the relationship between the clinical parameters on admission with the BNP dis , although no statistical association were found in online supplemental table 1.Furthermore, this study enrolled the patients between 2013 and 2016, and at that time, sacubtrilvarsartan was not approved in Japan, which is known to influence circulating BNP levels.Further investigation will be necessary to confirm the effects of sacubtrilvarsartan on the predicted BNP levels at discharge.The association of the predicted-BNP dis with BNP dis is statistically significant, but the degree of correlation was modest; we could not exclude the possibility that the other unknown factor or therapeutic effects, which are not included for the estimation of BNP dis in this study, might be also associated with the regulation of BNP.Taken together, this study was not designed to investigate the effects of the prospectively controlled pharmacological or non-pharamacological intervention on BNP dis .Further prospective study will be necessary to address these problems.
In addition, there was no discharge criteria for the research in this study, and in general, discharges were determined by the attending physician.Although the discharge levels are determined by clinical findings including BNP levels in stable phases of HF, the variation of HF severity at discharge exists among the attending physicians, which might create the further variation of the discharge levels of BNP.

CONCLUSION
We have shown the confounding factors affecting measured BNP levels and demonstrated BNP prediction at discharge in hospitalised patients with HF.This predicted BNP values have superior predictive power for clinical outcomes to raw BNP values on admission.This may contribute to the risk stratification of acute HF and primary intensive care management in acute vulnerable phases of HF.

Figure 1
Figure1Association between BNP levels at hospital discharge and levels on admission or predicted BNP levels at discharge (predicted-BNP dis ) and their prognostic power.(A) Correlation between circulating BNP levels at hospital discharge (BNP dis ) and BNP on admission (BNP ad ).(B) Correlation between BNP dis and predicted BNP levels at discharge (predicted-BNP dis ).(C) Area under the receiver operating characteristics curve (AUC) analysis of the occurrence of the combined clinical events.AUC for the predicted BNP levels at discharge (predicted-BNP dis : blue) was superior to BNP levels on admission (BNP ad : red) (p<0.001).(D) There was no significant difference in the AUC for the predicted-BNP dis (blue) and BNP levels at discharge (BNP dis : red).BNP ad , B-type natriuretic peptide on hospital admission; BNP dis , B-type natriuretic peptide at discharge.

Table 1
Baseline patient characteristics

Table 2
Association of discharge levels of BNP with clinical parameters on admission (univariate analysis) Abbreviations are shown in table1.

Table 3
Stepwise regression analysis for association of BNP dis with variables on admission Stepwise regression analysis for association of BNP dis with variables on admission after accounting for BNP ad .Abbreviations are shown in table1.