Can Brain Natriuretic Peptide Be Used as an Indicator of Over-infusion in Patients with Acute Sepsis? : A Pilot Study Evaluating Correlation Between Brain Natriuretic Peptide and Global End-diastolic Volume Index

Background: Recently, for patients with sepsis, under-infusion, that hastens the progression to multiple organ failure and death, has been avoided; however, the rate of over-infusion and resulting complications, such as mortality has increased. There are few indicators of over-infusion, such as transpulmonary thermodilution; therefore, minimally invasive, simple, and quick indicators of over-infusion are needed. This pilot study aimed to determine the association between brain natriuretic peptide (BNP) and global end-diastolic volume index (GEDI). Methods: This was a post-hoc analysis of a multicenter prospective randomized controlled study, performed in the intensive care units (ICU) of 10 separate hospitals in Japan from September 2013 to March 2016. Patients with sepsis who were expected to require mechanical ventilation for more than 48 hours were included. We measured BNP and GEDI every 24 hours from ICU admission until next 72 hours. The primary outcome was the correlation between BNP and GEDI that was assessed in two ways: 1) GEDI considered a continuous variable, and 2) GEDI divided into two groups based on a cutoff of 850 ml/m 2 , which is the upper limit of normal GEDI. We used a univariable generalized linear mixed-effects model in which patients are considered as a random effect. Results: Fifty-seven patients were analyzed. GEDI remained under the upper limit of normal GEDI (850 ml/m 2 ) from ICU admission to 72 h, and there were no signicant changes in the trends of BNP and GEDI over time. BNP tended to be higher in group 1 (GEDI >850 ml/m 2 ) from ICU admission to 72 h, but there was no signicant difference in the BNP values. The regression coecient and 95%CI of BNP for GEDI were 2.5 and 1.3 to 3.7 when considered as a continuous variable, and 166 and -273 to 605 when divided into two groups based on GEDI cutoff of 850 ml/m 2 . Conclusion: There could be a weak but signicant correlation between GEDI and BNP considering GEDI as a continuous variable. Using BNP as an indicator of over-infusion may be current pregnancy; burn; drug toxicity; severe pancreatitis; acute phase of stroke; acute coronary syndrome; severe valvular disease; arrhythmia detected at admission; unstable hemodynamic status with extracorporeal membrane oxygenation (ECMO), or intra-aortic balloon pumping (IABP); unsuitable situation for monitoring central venous pressure (CVP),or TPTD; time elapsed > 24 h prior to eligibility evaluation; a “Do not resuscitate” code status; regular hemodialysis; and attending physician’s decision. of the following also lead to inaccurate TPTD this study: ischemic heart disease (including medical history); chronic heart cardiomyopathy (including medical history); chronic kidney disease (including maintenance dialysis); BNP not collected; GEDI not collected; consent withdrawal; or duplicate registration. surviving sepsis campaign guideline; SVV, volume variation; TPTD, transpulmonary thermodilution

hours were included. We measured BNP and GEDI every 24 hours from ICU admission until next 72 hours.
The primary outcome was the correlation between BNP and GEDI that was assessed in two ways: 1) GEDI considered a continuous variable, and 2) GEDI divided into two groups based on a cutoff of 850 ml/m 2 , which is the upper limit of normal GEDI. We used a univariable generalized linear mixed-effects model in which patients are considered as a random effect.
Results: Fifty-seven patients were analyzed. GEDI remained under the upper limit of normal GEDI (850 ml/m 2 ) from ICU admission to 72 h, and there were no signi cant changes in the trends of BNP and GEDI over time. BNP tended to be higher in group 1 (GEDI >850 ml/m 2 ) from ICU admission to 72 h, but there was no signi cant difference in the BNP values. The regression coe cient and 95%CI of BNP for GEDI were 2.5 and 1.3 to 3.7 when considered as a continuous variable, and 166 and -273 to 605 when divided into two groups based on GEDI cutoff of 850 ml/m 2 .
Conclusion: There could be a weak but signi cant correlation between GEDI and BNP considering GEDI as a continuous variable. Using BNP as an indicator of over-infusion may be possible; however, further studies are needed.
Trial registration: None.

Background
Until recently, sepsis has been one of the most serious diseases, with a mortality rate up to 50% (1) . Since the cause of death in sepsis is the lack of oxygen delivery, and multiple organ failure due to the progression of sepsis, early uid resuscitation has been considered important to avoid the progression to multiple organ failure and death (2) (3) . With the widespread use of the Surviving Sepsis Campaign Guideline (SSCG), a treatment bundle for sepsis that recognized proper uid resuscitation, an improvement in sepsis-related mortality of approximately 30% was reported (4) (5) . Consequently, uid resuscitation became established because of robust evidence. However, in an attempt to prevent underinfusion, the rate of over-infusion has increased and there are growing concerns on the complications resulting from over-infusion. Currently, complications such as mortality due to over-infusion are being increasingly reported, and it is important to control not only the under-infusion but also the over-infusion of uids (6) (7) .
The usefulness of indicators of infusion, such as pulmonary artery wedge pressure (PAWP), stroke volume variation (SVV), lactic acid, and urine output, has been examined in patients with sepsis; however, these have been mainly useful for monitoring under-infusion, and cannot be effective in monitoring overinfusion (8) (9) (10) . Conversely, transpulmonary thermodilution (TPTD) is a medical device that can be used to monitor over-infusion (11) . In particular, accumulating evidence suggests that the global end-diastolic volume index (GEDI), which is measured using TPTD and re ects the volume of blood in the end-diastolic heart cavity, can be used as a monitoring index for under-infusion in patients with sepsis (12) (13) . However, TPTD may not be always be available due to the complexity of the procedure and the need for specialized devices.
Brain natriuretic peptide (BNP) is a ventricle-derived hormone that is elevated by ventricular load and is widely used in the diagnosis of heart failure (14) . In addition BNP can be measured in blood by a minimally invasive procedure. Few studies have reported the dynamics of BNP in diseases other than heart failure.
Although the association of BNP with prognosis and prediction of infusion responsiveness has been studied in patients with sepsis, it is not known whether it can be used as an indicator of over-infusion in patients with sepsis (15)(16)(17) (18) . Therefore, in this study, we hypothesized that BNP, could be used as an indicator of over-uidization in patients with sepsis instead of GEDI. The objective of the current study is to evaluate an association between GEDI and BNP, what will ultimately lead to the understanding whether BNP can be a surrogate for over-infusion assessment.

Design, patients
This study was a post-hoc analysis of a multicenter randomized controlled trial which was already published in Acute Medical Surgery on November 22, 2019 (19) . Patients admitted to the intensive care units (ICU) at 10 centers in Japan between October 2013 and March 2016 were included. The main study was approved by the clinical research ethics review committee of the central institution, St. Marianna University School of Medicine Hospital. This paper adhered to the standards laid down in the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement (20) .
Inclusion criteria of the main study were patients who met 1) the diagnosis of sepsis (meeting the diagnostic criteria for infection and systemic in ammatory response syndrome (SIRS)) and 2) the expected need for mechanical ventilation for more than 48 hours (21) . Patients having any of the following conditions that could lead to inaccurate TPTD readings were excluded in the main study: age < 18 years; current pregnancy; burn; drug toxicity; severe pancreatitis; acute phase of stroke; acute coronary syndrome; severe valvular disease; arrhythmia detected at admission; unstable hemodynamic status with extracorporeal membrane oxygenation (ECMO), or intra-aortic balloon pumping (IABP); unsuitable situation for monitoring central venous pressure (CVP),or TPTD; time elapsed > 24 h prior to eligibility evaluation; a "Do not resuscitate" code status; regular hemodialysis; and attending physician's decision.
In addition, patients with any of the following conditions that could also lead to inaccurate TPTD readings were excluded in this study: ischemic heart disease (including medical history); chronic heart failure; cardiomyopathy (including medical history); chronic kidney disease (including maintenance dialysis); BNP not collected; GEDI not collected; consent withdrawal; or duplicate registration.

Data collection
The extracted items were age, sex, comorbidities, Simpli ed Acute Physiology Score II (SAPS II), worst systolic blood pressure, maximum noradrenaline dose and total volume of infusion (within 24 h of ICU admission), BNP, GEDI, lactic acid level, Continuous Central Venous Oxygen Saturation (ScvO 2 ) and sequential organ failure assessment (SOFA) score (0, 24, 48, and 72 h), and renal replacement therapy (RRT) during ICU admission. We terminated the follow-up after 72 h of data collection or earlier in case of ICU discharge, transfer, discharge, or death. Information on the extraction factors was not masked because the information was collected by the physicians-in-charge at each center, and furthermore, the outcome assessors were not blinded. Missing data were not completed and patients having missing data were excluded.

Outcome
The primary outcomes was an association between the values of the BNP and GEDI.

Statistical analysis
Continuous variables are expressed as medians and interquartile ranges (IQR) and analyzed using the Mann-Whitney U test; categorical data are expressed as absolute numbers and percentages (%). The correlation between BNP and GEDI was assessed in two ways: 1) GEDI considered a continuous variable, and 2) GEDI divided into two groups based on a cutoff of 850 ml/m 2 , which is the upper limit of normal GEDI since we intended to consider an index of over-infusion. Regarding the association between BNP and GEDI and considering the fact that this study used repeated measures data, we used a generalized linear mixed-effects model in which patients are treated as variable effect. In this study, covariates affecting the outcome were considered to be excluded according to the exclusion criteria of the study, and therefore, the outcomes are analyzed via univariate analysis using EZR (Version 1.38), R (3.5.2.tar.gz) and SAS (Version 9.4) (22) (23) . We used p < 0.05 to detect statistically signi cant differences.

Results
A total of 372 patients were enrolled and 164 patients were analyzed in the main study. Regarding the reasons for exclusion, 53 patients had arrhythmias at the initial phase, 32 patients did not matched physician's treatment policy, 26 patients failed to provide informed consent, 9 patients had di culty in CVP or TPTD monitoring, while the other reasons are mentioned in Fig. 1. A total of 164 patients were thus eligible for this study of which 57 patients were included in the analysis. Regarding the reasons for exclusion in this study, 9, 4, and 3 patients had a history of ischemic heart disease, chronic heart failure, and cardiomyopathy, respectively. In addition, 9 patients had a history of chronic kidney disease, while the other reasons are mentioned in Fig. 1. There were 181 BNP measurements and 162 GEDI measurements, and 157 pairs were matched. Finally, data of 57 unique patients were analyzed.  Figure 2 shows the trends of BNP and GEDI during 72 h from the initiation of resuscitation for sepsis. GEDI remained under the upper limit of normal GEDI (850 ml/m 2 ) from ICU admission over the next 72 h, and there were no apparent changes in the trends of either BNP or GEDI over time (The statistical test was not performed).
GEDI was divided into two parts based on the cutoff of 850 ml/m 2 , and the changes in BNP according to the time are described in Table 2. Although BNP tended to be higher in group 2 (GEDI > 850 ml/m 2 ) from ICU admission to 72 h, this difference was not statistically signi cant(p ≧ 0.05). (Table 2) Using a univariable generalized linear mixed-effects model considering the patients as random effects, the regression coe cients and 95% CIs of BNP for GEDI were 2.5 and 1.3 to 3.7 (p < .001), when considered a continuous variable, and 166 and − 273 to 605 (p = 0.46), when divided into two 2 groups based on GEDI, with 850 ml/m 2 as the cutoff. (Table 3

Discussion
In this study, the BNP and GEDI in patients with sepsis undergoing invasive ventilation were mildly but signi cantly correlated when GEDI was considered a continuous variable.
GEDI is considered to be a measure of the systemic uid volume, because GEDI is calculated by dividing the global end-diastolic volume (GEDV) by the body surface area, which indicates the intrathoracic and pulmonary vascular blood volume (11) . In contrast, BNP, a hormone synthesized mainly by the ventricles, has been used as a biochemical marker that sensitively re ects the degree of ventricular load, because its secretion is increased by ventricular load, causing a rise in its blood concentration (24) . BNP is a superior auxiliary diagnostic method for heart failure, because plasma BNP concentrations correlate well with the hemodynamics, and BNP re ects the left ventricular load precisely (25) . In this study, GEDI was elevated due to the increased circulating plasma volume in response to the acute phase of sepsis infusion load. Similarly, BNP was also elevated in response to the left ventricle progression in response to the infusion load. These facts could indicate that BNP may be elevated in the acute phase of sepsis as well similar to the pathogenesis of heart failure, re ecting the excess uid volume in the acute phase of sepsis.
However, the correlation coe cient between BNP and GEDI was 2.5; the value was small and the correlation was mild, which may render it di cult for use in clinical practice. There could be three possible reasons for the small correlation coe cient: rst, as shown in Fig. 2, the median values of GEDI in this study over the three days of ICU stay remained within the upper limit of normal (850 ml/m 2 ). Therefore, it could be possible that BNP was not secreted and was thus not su ciently elevated due to the insu cient increase in blood volume in the left ventricular cavity, and the lack of expansion of the left ventricular cavity associated with the infusion load. The second reason, as noted in the early goaldirected therapy (EGDT), the acute phase of sepsis requiring large amount of volume infusion is generally considered to be approximately 6 hours from the diagnosis of sepsis (26) . In this study, the inclusion of data up to 72 hours when volume infusions are no longer needed and GEDI and BNP were no longer elevated, might have resulted in a small correlation coe cient and mild correlation. The third reason: BNP rises within 1 hour of stimulation and has a biological half-life of 20 minutes in the body (17) . Since the GEDI and BNP measurements in this study were performed every 6 hours, it is possible that even though the left ventricular cavity was su ciently enlarged by volume infusion to cause an increase in BNP, the very short half-life of the BNP caused a decrease at the time of measurement, which might have resulted in a small and mild correlation coe cient.
Considering these reasons, it may be possible to obtain larger regression coe cients and stronger correlations by shortening the measurement period of BNP to within 6 hours from starting the treatment and shortening the measurement interval to every hour or so.
Since the present study has revealed a possible association between BNP and GEDI, this study could guide further studies evaluating the usefulness of BNP as an indicator of over-infusion, using the upper limit of GEDI as a cutoff value as the index of over-infusion. Although this study is a pilot study, if further studies show that BNP can be an indicator of over-infusion, a minimally invasive, simple, and quick assessment of over-infusion would be possible. This nding could lead to avoidance of over-fusion, which might lead to a reduction in the adverse events, including increased mortality, and thus might have tremendous clinical bene t.
However, there are several limitations in this study. First, as noted in the EGDT, the acute phase of sepsis in which we need large volume of infusion is generally considered to be approximately 6 hours from the diagnosis of sepsis (26) . Because the GEDI and BNP assessments in this study were conducted at 0, 24, 48, and 72 hours after ICU admission, when volume infusions were no longer needed, and GEDI and BNP were no longer elevated, this might have contributed to a small correlation coe cient and mild correlation. Second, the median SAPS II score was 53 in this study, which is not suggestive of the most critically ill patients. In addition, the median infusion volume within 24 hours from ICU admission was 4492 ml and the median GEDI time series were all within the upper limit of normal for GEDI (850 ml/m 2 ). These facts suggest that the patients in this study were not severely ill requiring large amount of infusion; thus, the increase in GEDI and BNP was small, and neither GEDI nor BNP was elevated, and thus, no correlation was found. Third, the half-life of BNP is 20 minutes, and even if BNP was elevated, we could not record an increase in BNP by the 6-hourly measurements. Fourth, due to insu cient number of analyzed patients, we might not be able to detect signi cant differences when using a generalized linear mixed-effects model with GEDI of 850 ml/m 2 as the cutoff for the two groups. Fifth, the results of this study may be biased because the patients with severe anemia or acute kidney injury, in whom errors in BNP and GEDI measurements can occur, were not excluded.
Nevertheless, this pilot study suggests that GEDI and BNP may be correlated when GEDI is considered a continuous variable. However, further studies are warranted to increase the number of patients and obtain more frequent BNP measurements within the rst six-hours.

Conclusions
In conclusion, there could be a weak but signi cant correlation between GEDI and BNP considering GEDI as a continuous variable. It may be possible to use BNP as an indicator of over-infusion; however, that needs further studies.

Availability of data and materials
The data and materials of this study are available from the corresponding author on reasonable request.
Ethics approval and consent to participate Not obtained since this was a post-hoc analysis of a published study.

Consent for publication
Not applicable. Abbreviations: BNP, brain natriuretic peptide; GEDI, global end diastolic volume. The correlations were assessed in two ways: 1) GEDI was considered a continuous variable, and 2) GEDI was divided into two groups based on a cutoff of 850 ml/m 2 , which is the upper limit of normal GEDI, as we want to assess an index of over-infusion. We used a generalized linear mixed effects model in which patients are treated as a variable effect. Abbreviations: BNP, brain natriuretic peptide; CI, con dence interval; GEDI, global end diastolic volume. Figure 1 Flow of patients screening and enrollment. Abbreviations: BNP, brain natriuretic peptide; CVP, central venous pressure; DNAR, do not resuscitate; ECMO, extracorporeal membrane oxygenation; GEDI, global end-diastolic volume; TPTD, transpulmonary thermodilution.

Figure 2
Trends of BNP and GEDI over time. Abbreviations: BNP, brain natriuretic peptide; GEDI, global enddiastolic volume.