Cutoff Value of Phase Angle by Bioelectrical Impedance Analysis at Admission as a Prognostic Factor in Patients with Acute Heart Failure

Summary

This protocol presents R codes for performing restricted cubic spline multivariable Cox regression analysis to determine the cutoff value of the phase angle for prognosis in acute heart failure as well as codes for calculating statistics that measure the discriminatory ability of each model and evaluate its goodness of fit.

Abstract

Acute heart failure is the most common cause of hospital admission in patients aged > 65 years, with congestion being the predominant symptom. It is widely known that congestion is the main cause of readmission and its association with poor outcomes. This study aimed to determine the cutoff value of the phase angle using bioelectrical impedance analysis in the first 24 h after admission for the prediction of in-hospital mortality and all-cause mortality or rehospitalization 90 day post discharge in patients with acute heart failure.

The dataset analyzed for this protocol included 219 patients who presented to the emergency department with acute heart failure, of whom 115 died during hospitalization, 90 days post discharge, or were readmitted during the same period. Through univariable analysis using Cox proportional hazards regression analysis, potential predictors were identified, and a multivariable Cox proportional hazard model was used to identify the significant prognostic factors associated with patient survival and to control for confounding variables. Analysis of the phase angle adjusted by different predictors using restricted cubic splines revealed that lower values were directly associated with the risk of 90 day readmission or death post discharge, particularly when the cutoff value of PhA was <4.0°. All developed models showed moderate discriminatory ability (C-statistic < 0.80); however, the in-hospital mortality model demonstrated the highest discrimination ability (C-statistic, 0.75).

Introduction

Acute heart failure (AHF) is the most common cause of hospital admission in patients aged > 65 years, with congestion being the predominant symptom in the majority of cases as well as the main reason for hospitalization¹. Heart failure (HF) can lead to malnutrition through different mechanisms, including malabsorption due to intestinal edema, hypercatabolism due to sustained neurohormonal activation, and constant inflammation^2,3. Therefore, congestion and malnutrition directly influence body composition.

Bioelectrical impedance analysis (BIA) is a quick, noninvasive, and safe method that provides immediate bedside and reliable measurements to estimate body composition by introducing a constant low-level alternating electrical current that measures the impedance (Z) or resistance/opposition generated by human cells and tissues. The BIA involves two parameters: resistance (R) and reactance (Xc)⁴.

Additionally, it measures the delay between the current and voltage required to penetrate the cell membrane and tissue interfaces, known as the phase angle (PhA), which has been suggested to be an indicator of cellular health, cell membrane integrity, and vitality, which are affected by fluid distribution, inflammation, and malnutrition due to tissue resistivity and electrical properties. Lower PhA values indicate apoptosis and reduced components of the cellular matrix, whereas higher values reflect greater cellularity and better cell function⁵. Different studies have consistently demonstrated the prognostic role of PhA in multiple outcomes, such as mortality, rehospitalization, length of hospital stay, and risk of poor nutritional status in different clinical settings, including patients with HF^6,7.

This study aimed to describe the steps for performing multivariable Cox regression analysis with penalized splines to determine the cutoff value of PhA by BIA in the first 24 h after admission to predict in-hospital mortality and all-cause mortality or rehospitalization following 90 day post discharge in patients with AHF.

Protocol

This study comprises four steps. The data used in this study included adult patients (≥18 years) with AHF confirmed by a previous echocardiogram according to the European Society of Cardiology (ESC) guidelines⁸ and with expected hospitalization, as judged by clinicians. This study was approved by the Ethics and Human Biomedical Research Committee of the Instituto Nacional de Ciencias Médicas y Nutrición Salvador Zubirán and was conducted according to the principles of the Declaration of Helsinki. Written informed consent was obtained from the patient or their next of kin if they were unable to provide consent.

1. Data acquisition and general information

NOTE: BIA measurements require specific patient preparation, as described in a separate protocol⁹. Once the BIA equipment has completed the measurement, the equipment screen will display the following data: impedance values (Z) at different frequencies (5, 50, 100, and 200 kHz), as well as the resistance and reactance values at 50 kHz.

1. After BIA measurements, open the software to adjust the resistance (R) and reactance (Xc) by patient's height and to calculate the phase angle (PhA) values.
   NOTE: The analysis of BIA parameters using BIVA should be performed within 24 h of patient admission.
2. Enter the results into the dataset and save it as a .csv file for analysis in RStudio.
   NOTE: The patient's data are shown as a reference in Supplemental Table S1.

2. Installing, loading packages, and data preparation

NOTE: All procedures were performed using R software using the packages survival, MASS, and pspline.

1. Download R and RStudio.
   1. Install R software (version can change over time) and RStudio and open it.Click on the File tab | New File option | select the R script. RStudio will display the script file in the upper-left corner, positioned above the Console tab.
   2. Install the packages by executing the following commands in the script file:
      install.packages("survival")
      install.packages("MASS")
      install.packages("pspline")
   3. Load the packages by running the following commands
      library(survival)
      library(MASS)
      ​library(pspline)
   4. Import the dataset by executing the following command. #Dataset will be the name of the data. Here the file Acute heart failure data is the example.
      Dataset <-read.csv("…/Acute heart failure data.csv")

3. Model construction

1. Determine the Cox proportional hazards regression model adjusted for potential confounding variables and calculate the 95% confidence intervals (95% CI) through the following commands.
   Model_1 <- coxph (Surv (Survival_days, Outcome) ~ factor (Diabetes_mellitus) + Systolic_blood_pressure + Serum_sodium + Phase_angle, method="breslow")
   exp(confint(Model_1))
2. Calculate Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), C statistic, and 95% CI by running the following commands.
   #Akaike Information Criterion
   AIC(Model_1)
   #Bayesian Information Criterion
   BIC(Model_1)
   C-statistic
   summary(Model_1)$concordance[1]
   #Lower interval of C-statistic
   summary(Model_1)$concordance[1]-(1.96*summary(Model_1)$concordance[2])
   #Upper interval of C-statistic
   summary(Model_1)$concordance[1]+(1.96*summary(Model_1)$concordance[2])

4. Graphical representation

1. Create the survival object using the Surv() function.
   surv_1 <- Surv(Survival_days, Outcome)
2. Fit the Cox proportional hazards regression model developed on step 3.1. and predict the values for fitted spline by executing the following commands.
   fit_model_1 <- coxph(surv_1 ~ pspline(Phase_angle, df=2)+factor(Diabetes_mellitus)+ Systolic_blood_pressure + Serum_sodium)
   predicted_1 <- predict(fit_model_1, type="terms", se.fit=TRUE ,terms=1)
3. Plot the fitted model and the corresponding lines on the graph by executing the following commands. Identify the cut-off point where the HR adjusted (spline regression) is greater than 1, which suggests that a lower PhA values predict an increased risk of in-hospital mortality and 90-day readmission or death.
   plot(Phase_angle, exp(predicted_1$fit), type="n", xlab="Phase angle", ylab="Adjusted HR (95% IC)", xlim=c(1.2,12), ylim= c(0,2.5), main="All clinical outcomes")
   lines(sm.spline(Phase_angle , exp(predicted_1$fit)) , col = "#b23500" , lty = 1, lwd=2 )
   lines(sm.spline(Phase_angle , exp(predicted_1$fit + 1.96 * predicted_1$se)) , col = "#b23500" , lty = 2,lwd=1.2)
   lines(sm.spline(Phase_angle , exp(predicted_1$fit - 1.96 * predicted_1$se)) , col = "#b23500" , lty = 2, lwd=1.2)
   abline(h=1.0,col="#050505", lty= 2, lwd= 0.00001 )
   text(4.27,1.12, "4.0°")
   abline(v=4.0,col="black", lty= 2, lwd= 0.00001)

Results

This study included 219 patients with acute heart failure (AHF), both de novo and acute decompensated heart failure. The median age of the patients was 71 (p25th 60 - p75th 81) years, and 63% were female. Diabetes mellitus and hypertension were the most common comorbidities (41.6% and 69.9% of patients, respectively). During a median follow-up of 79 days, 115 events (52.5%) occurred, of which 32.2% died in the hospital, and 67.8% were readmitted or died within 90 days after discharge.

The patient characteristics at admission are presented in Table 1. Regarding the comparison of BIA parameters, the values of PhA, hemoglobin, and serum sodium were significantly lower among patients in whom an event was observed. In contrast, total body water (TBW), extracellular water (ECW), and brain natriuretic peptide (BNP) levels were significantly lower in patients in whom a non-event was observed. Additionally, patients who died in-hospital showed significantly lower systolic blood pressure (SBP), diastolic blood pressure (DBP), PhA, and higher BNP levels than survivors

Table 1: The clinical characteristics of the patients studied upon admission. The characteristics were classified according to the events. Values are expressed as medians (p25-p75) or frequencies (percentages). *<10% missing values. Abbreviations: BMI = body mass index; HFrEF = Heart failure with reduced ejection fraction; HFmrEF = Heart failure with mildly reduced ejection fraction; HFpEF = Heart failure with preserved ejection fraction; SBP = Systolic blood pressure; DBP = Diastolic blood pressure; OSA = Obstructive sleep apnea; COPD = Chronic obstructive pulmonary disease; TBW = Total body water; ECW = Extracellular water; PhA = Phase angle; BUN = Blood urine nitrogen; BNP = Brain natriuretic peptide. Please click here to download this Table.

Univariate Cox regression analysis showed that BMI, TBW, ECW, PhA, serum sodium, and erythrocyte levels were significantly associated with adverse events, whereas PhA, Xc, SBP, and DBP were significantly associated with in-hospital mortality

Table 2: Prognostic models for events, in-hospital mortality, and 90-day readmission or death based on multivariable Cox regression analysis. Abbreviations: SBP = systolic blood pressure; HFpEF = Heart failure with preserved ejection fraction; HFmrEF = Heart failure with mildly reduced ejection fraction; HFrEF = Heart failure with reduced ejection fraction; AIC = Akaike information criterion; BIC = Bayesian information criterion. Please click here to download this Table.

Two models were developed using Cox regression to determine predictive capacity in combination with different variables. After multivariate analyses, PhA remained an independent prognostic factor for events and for in-hospital mortality (adjusted Hazard Ratio [aHR] 0.76 [95% CI 0.65-0.90], p = 0.002 for model 1; aHR 0.79 [95% CI 0.67-0.93], p = 0.006, for model 2; and aHR 0.57 [95% CI 0.42-0.76], p ≤0.001, for in-hospital mortality) (Table 2). Likewise, when PhA was analyzed and adjusted for different predictors using restricted cubic splines, lower values were directly associated with the risk of 90 day readmission or death post discharge, particularly when PhA was <4.0°. All developed models showed moderate discrimination ability (C-statistic <0.80); however, the in-hospital mortality model demonstrated the highest discriminatory ability (C-statistic 0.75).

Figure 1: Adjusted hazard ratio of phase angle using penalized splines model. Abbreviations: HR = hazard ratio; CI = confidence interval. Please click here to view a larger version of this figure.

Supplemental Table S1: Example patient data. Please click here to view a larger version of this File.

Discussion

This study demonstrated that patients with AHF admitted to the ED with lower PhA values in the first 24 h after admission had a higher risk of poor prognosis, such as death or rehospitalization in the following 90 days post discharge, which increased when the PhA values were below 4.0°, independent of serum sodium, SBP, presence of diabetes mellitus, or sex. Similarly, lower PhA values predicted in-hospital mortality after adjusting for sex, SBP, serum potassium levels, and HF type. Currently, there is evidence demonstrating an association between low PhA values and poor outcomes in different diseases⁷. Additionally, some studies have determined different cutoff values ranging from 4.2° to 5.0°^10,11,12 to detect high-risk patients with poor prognosis, both within chronic HF and AHF.

According to previous studies, the rate of in-hospital mortality mostly ranges between 4% and 8%; however, in this study, it was higher (16.83%)^13,14,15. The readmission rate reported in this study was consistent with those reported by Muhammad et al. (34.7% vs 34.6%, respectively)¹⁵. According to the vulnerable phase, which lasts up to 6 months post discharge¹⁶, it is widely known the association with an increase in readmission and mortality remains up to 30% and 10%, respectively.

Accordingly, a rapid and reliable assessment of fluid congestion is essential for clinical decision-making and management of AHF, as it determines its prognosis¹³. The advantage of using PhA is that it is an easy predictor at the bedside, which can be useful at admission to identify patients at a high risk of poor outcomes. In combination with BIVA¹⁷, it is recommended to provide additional information related to hydration status, membrane integrity, permeability, and therefore, cell function^4,6.

There are some limitations to consider when using the Cox proportional hazards model for prediction, which assumes that the effects of covariates remain constant over time (the proportional hazards assumption). Violation of this assumption may lead to biased or unreliable estimations¹⁸, as well as informative censoring that can lead to biased estimations¹⁹. In Cox analysis, it assumes that censoring is non-informative; therefore, researchers must be aware of the type of censoring in their data.

Moreover, several factors may vary throughout the follow-up period and were not accounted for in this protocol. These include a history of coronary artery disease (CAD), previous revascularization, and in-hospital treatments, which were not registered despite their potential association with the outcomes.

Materials

Name	Company	Catalog Number	Comments
BIVA tolerance R-Xc software			Linked to the BIA equipment
Computer system	Microsoft	Windows 10	Windows 10 or higher
Quadscan 4000  equipment	BodyStat	BS-4000	Impedance measuring range: 20 - 1300 Ω ohms Test Current: 620 μA Frequency: 5, 50, 100, 200 kHz Accuracy: Impedance 5 kHz: +/- 2 Ω   Impedance 50 kHz: +/- 2 Ω   Impedance 100 kHz: +/- 3 Ω   Impedance 200 kHz: +/- 3 Ω   Resistance 50 kHz: +/- 2 Ω   Reactance 50 kHz: +/- 1 Ω   Phase Angle 50 kHz: +/- 0.2° Calibration: A resistor is supplied for independent verification from time to time. The impedance value should read between 496 and 503 Ω.
R software	None	Not Applicable	Version 3.6.2 or higher
RStudio	None	Not Applicable	Version 2022.07.2+576