Non-Invasive Venous waveform Analysis (NIVA) correlates with pulmonary capillary wedge pressure (PCWP) and predicts 30-day admission in heart failure patients undergoing right heart catheterization

Abstract

Background:

Heart failure is the leading cause of hospitalization in the elderly and readmission is common. Clinical indicators of congestion may not precede acute congestion with enough time to prevent hospital admission for heart failure. Thus, there is a large and unmet need for accurate non-invasive assessment of congestion. Non-Invasive Venous waveform Analysis in heart failure (NIVA_HF) is a novel, non-invasive technology that monitors intravascular volume status and hemodynamic congestion. The objective of this study was to determine the correlation of NIVA_HF with pulmonary capillary wedge pressure (PCWP) and the ability of NIVA_HF to predict 30-day admission after right heart catheterization (RHC).

Methods:

The prototype NIVA_HF device was compared to PCWP in 106 subjects undergoing RHC. The NIVA_HF algorithm was developed and trained to estimate PCWP. NIVA Scores and central hemodynamic parameters [(PCWP, pulmonary artery diastolic pressure (PAD), and cardiac output (CO)] were evaluated in 84 patients undergoing outpatient RHC. Receiver Operating Characteristic (ROC) curves were used to determine whether a NIVA Score predicted 30-day hospital admission.

Results:

The NIVA Score demonstrated a positive correlation with PCWP (r=0.92, n=106, p<0.0001). NIVA Score at time of hospital discharge predicted 30-day admission with an AUC of 0.84, a NIVA Score >18 predicted admission with a sensitivity of 91% and specificity of 56%. Residual analysis suggested that no single patient demographic confounded the predictive accuracy of the NIVA Score.

Conclusions:

NIVA_HF is a non-invasive monitoring technology that is designed to provide an estimate of PCWP. A NIVA Score >18 indicated increased risk for 30-day hospital admission. This non-invasive measurement has potential for guiding decongestive therapy and prevention of hospital admission in heart failure patients.

Visual Take Home Graphic.

The NIVA_HF device is a non-invasive technology that records venous raw signal harmonics (A) of the pulse rate from the volar aspect of the wrist. These raw signals are transformed into the frequency domain (B) using a Fast Fourier Transformation (FFT) and then used to train, validate, and test a NIVA_HF algorithm to output a NIVA Score based on the harmonic patterns recorded. A Pearson correlation (C) of NIVA Scores to measured PCWP (mmHg) demonstrated a significant correlation (r=0.92, n=106, p<0.0001). A secondary analysis of 30-day admissions due to heart failure exacerbation, Receiver Operator Characteristic curve (D) demonstrated an AUC=0.84 (p<0.0001). A NIVA Score >18 had a sensitivity of 91% and specificity of 56% at predicting the need for hospitalization secondary to heart failure exacerbation in 30-days.

Lay Summary

NIVA_HF is a non-invasive technology that estimates pulmonary capillary wedge pressure (PCWP) over a wide range of values. In 106 patients having right heart catheterizations (RHC), NIVA Scores demonstrated a significant correlation to the measured PCWP (r=0.92, n=106, p<0.0001). A secondary analysis of the subset (n=84) that went home immediately after their RHC demonstrated that elevated NIVA Scores predicted 30-day admission (AUC=0.84, p<0.0001). Using a trained, validated, and tested algorithm the automated analysis of a patients’ venous harmonics can provide real time assessment of cardiac filling pressures in multiple settings (hospital, clinic, home).

INTRODUCTION

Heart failure related hospitalizations are common, costly, and impact both quality of life and prognosis.[1] Systemic and pulmonary congestion precipitate heart failure decompensation causing symptoms of edema, shortness of breath, and fatigue. Evaluation of systemic congestion relies largely on clinical examination findings of jugular venous distention, edema, hepatomegaly, rales, and ascites coupled with symptoms. Diuretic therapy is the established therapy for decongestion often in conjunction with inotropes and vasodilators.[2, 3] Physical exam and symptoms lack sensitivity for assessing response to therapy [4] and too often patients are discharged with residual congestion, increasing the risk of readmission.[5-9]

The peripheral venous system is composed of capacitance vessels which contain up to 70% of the total blood volume.[10] Due to the low pressure nature of the venous compartment, wave intensity magnitude (power) in veins is low (~5 mmHg), hence venous waveforms were not rigorously examined until adequate sensing and amplifying technologies recently became available.[11-13] Venous waveforms can be acquired with non-invasive technology using a piezoelectric sensor placed directly over the veins of the wrist. Analysis of the peripheral venous waveform has led to Non-Invasive Venous waveform Analysis which can be applied to patients with heart failure (NIVA_HF, Figure 1).[11-15] Peripheral venous waves generated by the cardiac cycle, are propagated through the venous network. The fundamental venous waveform corresponds to the pulse rate (“f₀”). Deconvolution of the raw peripheral venous signal with a fast Fourier transformation reveals the presence of harmonics of this fundamental cardiac frequency (“f_1-7”). This led to the discovery that harmonics of the fundamental cardiac frequency (f_2-7) increase with congestion.[11] A NIVA_HF algorithm was developed using a ratio of pulse rate power (f₀) to the pulse rate harmonics (f_1-7) based on pulmonary capillary wedge pressure (PCWP) (Figure 2). In a previous study, prospectively acquired peripheral venous waveforms and corresponding PCWP acquired during right heart catheterization of heart failure patients were used to derive an algorithm using a least-squares analysis.[11] NIVA technology has been used to accurately evaluate volume overload (heart failure, dialysis) and volume deficits (dehydration, over-diuresis, blood donors) suggesting that the peripheral venous waveform contains adequate information about volume status to estimate PCWP.[11, 12, 16-19] With increasing venous waveform data and experience, the peripheral venous waveform acquisition, analysis, and algorithms continue to improve.

Figure 1.

Photograph of NIVA_HF device used to obtain NIVA Scores prior to right heart catheterization.

Figure 2. Representative NIVA signals from subjects with low to high PCWP (left to right).

Raw signals are transformed from the time domain (top) to the frequency domain (bottom). The relative powers of the pulse rate (f₀, fundamental frequency) and its harmonics (f_1-7, multiples of the pulse rate) are incorporated into an algorithm to calculate a NIVA value that estimates PCWP. Abbreviations: NIVA= Non-Invasive Venous waveform Analysis; PCWP= pulmonary capillary wedge pressure; f= frequency.

Non-Invasive Venous waveform Analysis in heart failure (NIVA_HF), which detects and analyzes the peripheral venous waveform at the wrist, is a novel technology that has demonstrated accuracy in detecting circulatory volume loss and overload in both porcine models and human observational studies.[11, 12] An earlier prototype of the NIVA device and preliminary algorithm suggested that NIVA values correlated with PCWP (r=0.69, n=83, PCWP = 4-40 mmHg).[11] The primary outcome of this observational proof-of-concept study was to assess the ability of an optimized NIVA device (NIVA_HF) to analyze, train, optimize, and test a NIVA algorithm to estimate PCWP in patients with heart failure. To determine the clinical utility of NIVA_HF to assess risk of hospital admission with the subsequent goal of readmission prevention, a secondary analysis of NIVA_HF obtained at time of RHC was performed. The ability of NIVA_HF to predict a 30-day hospital admission was compared to other invasive hemodynamic parameters determined during RHC [PCWP, pulmonary artery diastolic pressure (PAD), and cardiac output (CO)].

METHODS

This observational study was approved by the University of Alabama Birmingham Institutional Review Boards through Vanderbilt University Medical Center (VUMC) Institutional Review Board secondary to institutional conflict of interest. Patients that were at least 18 years old and had a scheduled right heart catheterization (RHC) were enrolled. One hundred and six subjects with adequate PCWP tracings and NIVA signals were used for analysis, training, validation, and testing (Figure 3) of the NIVA_HF algorithm. This process is completed post evaluation of the pulmonary capillary wedge pressure and is used to ensure the best fit of the predicted NIVA Score to the volume status of the subjects. Indications for RHC included heart failure, post heart transplant, and diagnostic evaluation (Table 1).

Figure 3.

Description of patients undergoing elective right heart catheterizations and subset reviewed for 30-day admission. Abbreviations: f=frequency; RHC= right heart catheterization.

Table 1.

Demographic table of enrolled right heart catheterization patients. This table demonstrates the subject characteristics as median (IQR) as well as the representative heart failure distribution [n(%)], and total admitted [n(%)]. Receiver Operator Characteristic (ROC) curves are reported for each enrollment group demonstrating the ability of the NIVA Score to predict 30-day hospital admission post RHC. Abbreviations: RHC= right heart catheterization; BMI= body mass index; PAD= pulmonary artery diastolic pressure; PCWP= pulmonary capillary wedge pressure; NIVA= non-invasive venous waveform analysis; CI= cardiac index; EF= ejection fraction; ASA= American Society of Anesthesiology; RHC= right heart catheterization; HF= heart failure; AUC= area under the curve; Std= standard; CI= confidence interval.

	ALL RHC N=106	ADMITTED	NOT ADMITTED	P-VALUE
AGE	63 (50 -67)	63 (41 -66)	62 (50 - 67)	p = 0.72
SEX	61.3% (male)	43.7% (male)	59.7% (male)	p = 0.27
RACE	76.4% (white)	81.25% (white)	76.12% (white)	p > 0.99
BMI (kg/m2)	29.4 (26.9 – 33.6)	29.12(23.7 - 32.2)	29.5 (27.4 - 35.3)	p = 0.13
RHYTHM
sinus	101 (95%)
atrial fibrillation	2 (2%)
atrial flutter	1 (1%)
a-v paced	2 (2%)
PAD (mmHg)	17.5 (13 – 23)	28 (17 -34)	16(13 -20)	p <0.05
PCWP (mmHg)	14.5 (10.8 – 18)	22 (18 - 30)	13(11 - 16)	p <0.05
NIVA	14(11 -17)	22 (18 - 30)	13(11 - 16)	p <0.05
CI (L/min/m2)	2.4 (2.1 – 2.8)	2.4 (2.1 - 2.8)	2.5 (2.1 - 3.0)	p = 0.31
EF (%)
<45	19 (20%)	5 (31.3%)	11 (16.5%)	p = 0.17
45-55	8 (8.4%)	2 (12.5%)	3 (4.5%)
>55	68 (71.6%)	9 (56.3%)	53 (79.1%
ASA
1	0	1 (6.25%)	0 (0%)	p = 0.10
2	16 (16.2%)	1 (6.25%)	13 (19.4%)
3	82 (82.8%)	14 (87.5%)	52 (77.6%)
4	1 (1%)	0 (0%)	2 (3%)
RHC (n; %)				p = 0.55
HF DIAGNOSTIC EVALUATION	39 (36.8%)	6 (37.5%)	19 (30.6%)
ROC CURVE	AUC = 0.88	Std Error 0.10	95% CI 0.68 to 1.07	P =0.01
HF MAINTENANCE CARE EVALUATION	29 (27.4%)	5 (31.3%)	17 (27.4%)
ROC CURVE	AUC = 0.80	Std Error 0.12	95% CI 0.57 to 1.03	P =0.02
POST-TRANSPLANT GRAFT EVALUATION	38 (35.8%)	4 (25%)	26 (41.9%)
ROC CURVE	AUC = 0.77	Std Error 0.11	95% CI 0.55 to 0.98	P =0.13
TOTAL		17 (20%)	67 (80%)

Procedure:

Patients were consented prior to RHC as part of the pre-procedural assessment. The piezoelectric sensor of the NIVA_HF device (Figure 1) was secured to the volar aspect of the middle of the wrist overlying the superficial veins with a Versaflex™ wristband (PolyOne Corporation; OH, USA). NIVA signals were acquired for at least two minutes within one hour of catheterization with subjects positioned in the semi-recumbent position (head of bed elevated to 60-90 degrees) and requested to remain still while breathing normally during NIVA signal acquisition. Patients were then taken to assigned catheterization suite where, after central vein cannulation, a pulmonary artery catheter (PAC; Edwards Life Sciences Corporation, Irvine, CA, USA) was inserted with balloon tip inflated, into the pulmonary artery, and into “wedge” position. With the transducer positioned at the level of the heart, the PAC waveform was obtained at end expiration, in spontaneously breathing patients, and recorded to the electronic medical record (EMR). RHC tracings and PCWP values were uploaded to a RedCap[20] database. PCWP tracings were reviewed by a blinded cardiologist and/or cardiovascular intensivists (BA, JH, JS, JL). Cardiologists performing, and personnel reviewing RHC tracings, were blinded to all NIVA Scores. Cardiac Output (CO) was obtained after PCWP measurement using thermodilution technique.

Admission Analysis:

A subgroup of 84 of the 106 subjects who underwent outpatient RHC were investigated for 30-day hospital admission. The remaining 22 were inpatient RHCs and were not investigated for 30-day admission post RHC (RHC was not performed the same day they went home). Chart review was performed (BA, MB) to determine if hospital admission was secondary to heart failure exacerbation as documented by an admitting and/or consulting cardiologist.

Algorithm training and validation:

The NIVA_HF algorithm for the updated prototype used in this study was trained using a subset of the RHC patients with their recorded PCWP pressures in JMP Pro 13 software (SAS Institute, Cary, NC). RHC subjects were randomly assigned into training (n=64), validation (n=21), and test (n=21) groups (Figure 3).[5, 21] The 60/20/20 approach (Figure 3) is one of the most frequently used as well as one of the more conservative splits used for developing an algorithm [5, 21, 22]although some larger database-based algorithms may not retain such a large portion from the training model.[23, 24] In the 60/20/20 approach the algorithm correlation is traditionally reported as “r” value to entire data set not just the validation and/or test cohort.[21, 22]

The NIVA_HF device consists of a 22mm piezoelectric sensor that is attached via a cable to a monitoring data-logger device which used a low (~0.9 Hz) cutoff frequency created by changing the input impedance on the charge amplifier. The data was then imported to a PC computer where a custom application, written by the authors to convert time domain signal into the frequency domain, was used to determine the signal amplitude of the pulse rate as well as the amplitudes of the pulse rate harmonics by analysis of the waveform in the frequency domain (fast Fourier analysis). The changes in the relative amplitudes of the pulse rate frequency (f₀) and its harmonics (f_1-7) that occurred with changes in intravascular volume serve as the basis of NIVA waveform analysis.

A 4-fold, K-fold, dual layer neural network was used for training the algorithm to produce a “NIVA Score” that estimates PCWP. The calculated relative powers of f_0-7 obtained by the Fourier transform of the data were incorporated as variables for algorithm training. To train the algorithm to predict PCWP from the venous waveform transduced peripherally without overfitting, and to correctly power the input data, the first layer of the neural network was used to determine logistically if the subject was classified as having a “high” or “low” volume status (PCWP > 18 or ≤ 18 mmHg)[11, 25-27], followed by a layer that returned a quantitative value (NIVA Score) for the NIVA_HF device. The log-worth of venous waveform amplitudes in the frequency domain (f_0-7) and their relationship to PCWP were plotted against each other in multi-dimensional space as hyperbolic tangential functions to predict PCWP. The iterative process of training and validating the algorithm in-situ involves optimizing the Sum Square of the error between calculated NIVA Score and recorded PCWP. This process may yield results that are more correlative than real-world use and final validation. This is because of the iterative training and validation process that occur in-situ allow for fitting of the data. Realistically the validation and test correlations will be more representative of real-world use cases.

Statistical analysis:

Data analysis was performed using JMP Pro 13 software (SAS Institute, Cary, NC) and GraphPad Prism (GraphPad Software Inc., La Jolla, CA). Results for continuous variables are reported as median [Interquartile Range (IQR)]. Pearson correlation coefficients were calculated for the NIVA Score, PAD and CO compared to PCWP. For the NIVA Score a Pearson correlation coefficients were calculated not only for the entire data set, but also for each of the three subsets: training, validation, and testing. For admission prediction Receiver Operator Characteristic (ROC) curves were used to determine the accuracy of the NIVA Score to detect admission.[28] To compare if there were significant differences between the ROC curves a Z score was obtained by calculating the difference in the AUC of the two curves and then dividing this by the square root of the sum squares of the standard error of the curves. The Z score can then be used to determine the p value. Additionally, the NIVA Score was tested against a D’Agostino-Pearson normality test where it did not pass the normality test. This data was then compared using an unpaired nonparametric Mann Whitney test to determine significance, defined by p-value <0.05, between patients that were admitted within 30 days of their right heart catheterization and those that did not require an admission. Residual analyses were performed to evaluate the effects of age (years), body mass index (BMI, kg/m²), edema grade, creatinine (Cr, mg/dL) estimated glomerular filtration rate (eGFR, mL/min/1.73m²), chronic kidney disease stage (CKD), systolic pressure (SBP, mmHg) diastolic pressure (DBP, mmHg), right atrial pressure (RAP, mmHg), oxygen saturation (%), respiratory rate (RR, breaths per minute) heart rate (HR, beats per minute), systemic vascular resistance (SVR, mmHg min/L), pulmonary systolic pressure (PAS, mmHg), pulmonary artery diastolic pressure (PAD, mmHg), pulmonary artery mean pressure (PAM, mmHg), pulmonary vascular resistance (PVR, mmHg min/L), cardiac output (CO, L/min), ejection fraction (EF, %) on the predictive accuracy of NIVA Scores to PCWP.[29] Residuals were plotted as the absolute value of the standardized residual calculated from the linear regression of NIVA Scores vs. PCWP.

RESULTS

Demographic information for the algorithm analysis, training, validation, and testing cohort (n=106) are displayed in Table 1. Median age was 63 (IQR 50-67) years, 61.3% of subjects were male, and 81.3% were white. Indications for RHC included heart failure diagnostic evaluation in 39 (37%), post-transplant graft evaluation 38 (36%), or heart failure maintenance of care evaluation 29 (27%). Of those patients with heart failure who had an echocardiogram within the preceding year (n=95), ejection fraction was reduced (EF<45%) in 19 (20%) and preserved in 76 (80%).

PCWP measurements in the algorithm analysis, training, validation, and testing cohort ranged from 4 to 31 mmHg, with a median value of 14.5 (IQR10.8-18) mmHg (Table 1). Other hemodynamic indices obtained during catheterization are displayed in Table 1. There was a significant correlation between PCWP and NIVA Scores from the entire trained, validated, and tested NIVA_HF algorithm (r=0.92, n=106, p<0.0001, Figure 4A) and PAD (r = 0.64, n = 84, p>0.05, Figure 4C). There was no significant correlation between PCWP and CO (r = 0.24, n = 81 patients, p>0.05, Figure 4B). When broken down into their respective data sets, the training set NIVA Scores demonstrated a significant correlation (r=0.97, n=64, p<0.0001), the validation set NIVA Scores demonstrated a significant correlation (r=0.93, n=21, p<0.0001), and the testing set NIVA Scores demonstrated a significant correlation (r=0.78, n=21, p<0.0001).

Figure 4. Central Hemodynamic Indices and NIVA Scores in Patients Undergoing right heart catheterization at VUMC.

The correlation between NIVA Score and PCWP (A) demonstrated that NIVA Score has a high correlation with PCWP (A. r = 0.92, n = 106, p<0.0001). When broken down into their respective data sets, the training set NIVA Scores demonstrated a significant correlation (r=0.97, n=64, p<0.0001), the validation set NIVA Scores demonstrated a significant correlation (r=0.93, n=21, p<0.0001), and the testing set NIVA Scores demonstrated a significant correlation (r=0.78, n=21, p<0.0001). Neither CO (B, r = 0.24, n = 81, p>0.05) or PAD (C, r = 0.64, n = 84, p>0.05) had as high of a correlation with PCWP. Abbreviations: NIVA= Non-Invasive Venous waveform Analysis; PCWP= pulmonary capillary wedge pressure; CO= cardiac output; PAD= pulmonary artery diastolic pressure.

Of the 84 subjects investigated for 30-day admission, 67 (80%) did not require admission within 30-days of their RHC and 17 (20%) were hospitalized for HF signs/symptoms. The median NIVA Score in subjects who required admission was 22 (IQR=18-30) compared to 13 (IQR=11-16, p<0.05; Figure 5A) in those not admitted. Median PCWP in subjects who were admitted was 22 mmHg (IQR=18-30 mmHg) and 13 mmHg in those that were not admitted (IQR=11-16 mmHg, p<0.05; Figure 5B). Median PAD in subjects who were admitted was 28 mmHg (IQR= 17-34 mmHg) and 16 mmHg in those not admitted (IQR= 13-20 mmHg, p<0.05; Figure 5C). The Receiver Operator Characteristic (ROC) Curve demonstrated an AUC of 0.84 for NIVA Score (Figure 5A), 0.87 for PCWP (Figure 5B), and 0.74 for PAD at predicting 30-day admission (Figure 5C). There were no significant differences between the three ROC curves based on their Z scores. NIVA Scores above 18 had a sensitivity 91% and specificity 56% at predicting the need for hospitalization within 30-days (p<0.0001; Figure 5A). A PCWP > 18mmHg had a sensitivity of 85% and a specificity of 63% of predicting admission (p<0.0001; Figure 5B) and PAD > 18mmHg had a sensitivity of 64% and a specificity of 75% for predicting admission (p=0.0034; Figure 5C). Cardiac output > 3.63 L/min, which was the value with the greatest likelihood ratio from the ROC sensitivity (88%) and specificity (56%) analysis, did not demonstrate significance for prediction of hospital admission (AUC = 0.68 p > 0.05; Figure 5D). When broken down into the three indications for RHC groupings, the ROC curve for HF diagnostic evaluation patients demonstrated an AUC of 0.88 for NIVA Score predicting 30-day admission, the ROC curve for HF maintenance care evaluation demonstrated an AUC of 0.80 for NIVA Score predicting 30-day admission, and the ROC curve for post-transplant graft evaluation demonstrated an AUC of 0.77 (Table 1).

Figure 5. NIVA Score and hemodynamic indices during RHC and admission to the hospital within 30 days of RHC.

Values of NIVA Score, PCWP, PAD and CO for subjects either admitted to the hospital within 30 days (n=17) or not admitted within 30 days (n=67). Median NIVA Score, PCWP, and PAD were significantly different (*, p<0.05) between each of their respective “admitted” and “not admitted” groups. For CO, there was no difference (p=0.15) between the two groups. Receiver Operator Characteristic (ROC) curves demonstrated an AUC = 0.84 for NIVA Score (A, p < 0.0001); AUC = 0.87 for PCWP (B, p < 0.0001); AUC = 0.74 for PAD (C, p < 0.0034): and 0.68 for CO (D, p = 0.08). NIVA Scores above 18 had a sensitivity 91% and specificity 56% at predicting the need for hospitalization within 30-days (A). A PCWP > 18mmHg had a sensitivity of 85% and a specificity of 63% of predicting admission (B) and PAD > 18mmHg had a sensitivity of 64% and a specificity of 75% for predicting admission (C). Cardiac output of 3.63 L/min did not demonstrate significance for prediction of hospital admission (D). There were no significant differences between the three ROC curves based on their Z scores. Abbreviations: NIVA= Non-Invasive Venous waveform Analysis; PCWP= pulmonary capillary wedge pressure; CO= cardiac output; PAD= pulmonary artery diastolic pressure; ROC= receiver operating characteristic; Std.= standard.

Residual analyses demonstrated that the accuracy of the NIVA Score to predict PCWP was not significantly affected by age, BMI, edema grade, Cr, eGFR, CKD stage, SBP, DBP, HR, RR, oxygen saturation, RAP, PAS pressure, PAD pressure, PAM pressure, PVR, SVR), CO, and EF (Supplementary Figures 1-3).

DISCUSSION

The NIVA Score is a numerical value generated from analysis of peripheral venous waveforms acquired using a non-invasive piezoelectric sensor in a housing applied over the volar aspect of the wrist and secured via a wristband. This study demonstrates the ability to train and test in-situ the NIVA Score to estimate PCWP with a high degree of accuracy (r=0.92). This method of analysis, training, validation, and testing was performed to produce the best possible correlative results. Compared to PCWP, NIVA Scores had a higher sensitivity and reasonably equivalent specificity for predicting heart failure hospitalizations within the specified 30-day period (Figure 5). These results demonstrate the ability of NIVA_HF to provide a value comparable to PCWP[30] and non-invasively identify patients with congestion who may be at increased risk for hospital admission. This technology also has potential value in the inpatient setting to assess the degree of congestion prior to hospital discharge, where there is growing interest in non-invasive devices.[9, 31] NIVA_HF may be particularly useful when evaluation of jugular venous pressure is difficult. The ease of use and non-invasive nature of NIVA_HF allows for use in the clinic and in outpatient settings. This non-invasive technology will likely offer significant benefit for assessment of congestion as serial measurements can be obtained in the inpatient, outpatient, and home settings conveniently with minimal risk. Finally, the ability to obtain repeat measures allows assessment of trends that may be useful for individual patients in designing and adjusting treatment approaches.

Other non-invasive modalities to directly estimate intracardiac filling pressures and congestion have begun to emerge in clinical practice, however, none have proven to have the reliability or accuracy to be incorporated into clinical practice. The earliest of these is inferior vena cava ultrasound for estimation of right atrial pressure; however, in heart failure patients these measurements are less reliable and subject to substantial user variability.[32, 33] The use of thoracic impedance to estimate hemodynamic volume status, either as an added functionality of implantable cardioverter-defibrillator and cardiac resynchronization therapy devices, or measured non-invasively (i.e. ZOE®, ReDS™) demonstrate moderate correlation with PCWP (ReDS™ correlation to PCWP: r=0.49, n=139, p<0.05),[34-36] but the specificity of impedance measurements is limited by patient positioning[37], body tissue (fat) composition[36], presence of cutaneous hair or sweat, and lead placement. Impedance technology relies on downstream consequences of congestion, such as pulmonary edema and intrathoracic fluid content, rather than direct estimation of intracardiac filling pressures, and it is unclear whether resolution of pulmonary edema correlates with complete intravascular decongestion.[38, 39] Thoracic impedance also reports the measure in Ohms in which is inverse to volume status (lower number = higher intravascular volume). This measure is difficult to understand and incorporate clinically as it does not parallel or mirror any existing clinically used measure of intravascular volume. NIVA_HF is simple to use and can provide a numeric output that estimates a widely understood clinical measure of intravascular volume, PCWP.

Venous waveform analysis is emerging as a novel and accurate approach for monitoring of intravascular volume status and hemodynamic congestion.[11, 12, 16-19] The NIVA_HF is a non-invasive, wearable piezoelectric sensor at the wrist to detect and analyze peripheral venous waveform. Current clinical evaluation of systemic congestion in heart failure relies largely on weight changes, intake and output balance, and exam findings such as jugular venous distention (JVD) and peripheral edema. It is well known that in most patients elevations in PCWP begin days or weeks prior to an admission for heart failure.[40, 41] RHC is the gold standard measure for volume status with PCWP used by clinicians to evaluate left-sided cardiac filling pressures, a surrogate for preload or systemic “volume status.”[11, 42] Elevated intracardiac filling pressures, as manifested by PCWP, have been associated with increased hospitalizations and mortality,[43, 44] and often precede the development of symptoms related to HF decompensation.[41] Data from the multicenter Evaluation Study of Congestive Heart Failure and Pulmonary Artery Catheterization Effectiveness (ESCAPE) trial demonstrated improvements in exercise and quality of life endpoints with PAC guided therapy in patients with acute decompensated heart failure, at the expense of an increased risk of adverse events, and no differences in mortality.[45] The 2013 ACCF/AHA Guidelines for the Management of Heart Failure continue to support the use of invasive hemodynamic monitoring in select patient populations with indeterminant volume status.[46] Pulmonary artery hemodynamic measurements have proven clinical value in patient management and quality of life. NIVA_HF can provide a comparable measure exhibiting promise for improving both inpatient and outpatient care of patients with HF without any procedural risk. NIVA_HF has the potential to detect hemodynamic congestion prior to onset of systemic symptoms, thus allowing earlier intervention similar to the CHAMPION trial.[40, 47] Presumably the NIVA_HF technology could non-invasively identify those at high risk of readmission.

The ability of NIVA_HF to provide an accurate score was not affected within our cohort by any condition or vital sign tested: age, BMI, edema grade, creatinine, eGFR, CKD stage, systolic blood pressure, diastolic blood pressure, right atrial pressure, oxygen saturation, respiratory rate, heart rate, pulmonary artery systolic pressure, pulmonary artery diastolic pressure, pulmonary artery mean pressure, pulmonary vascular resistance, systolic vascular resistance, cardiac output and EF based on residual analysis of the data (Supplementary Figures 1-3). Previous publications of this technology reported an inability to analyze the venous waveforms of the pulse rate in the presence of atrial fibrillation/flutter, which, albeit a small sample size (N=3), may no longer be the case.[11] Thus, NIVA_HF has demonstrated the ability to capture data and report significant correlations to PCWP across a group of 106 subjects.

This study has several limitations, the first is that it was a single-center study performed at a quaternary referral center. While this observational study with a prototype NIVA_HF device suggests that venous waveform analysis has the potential to predict 30-day admission, further studies are needed to determine if treatment decisions based on NIVA_HF can effectively assist providers and patients in management of heart failure to improve quality of life, decrease readmission, and decrease mortality. Also, it must be noted that a PCWP > 18mmHg does not necessary mean that a patient is experiencing significant volume overload. The value of 18 was the point chosen as the cutoff for congestion[11, 25-27] and presented to highlight the results of PAD, PCWP and NIVA Score. This study used a conservative, frequently used data split for training, validating, and testing an algorithm—this observational study was not a device validation study. Correlations are observed on groups where there is a large amount of training data, potentially overfitting the PCWP to the patients’ NIVA Scores and larger validation studies are needed to confirm these results. The venous waveform is a low-amplitude signal and subject to noise secondary to motion artifact and requires that patients remain still for 2-5 minutes while the signals are acquired. If adequate signal is not obtained, the NIVA_HF device does not display a PCWP equivalent value. Further advancement in sensor apposition, waveform capture, and signal analysis may improve signal, analysis and reduce noise with NIVA_HF. An inherent limitation in comparison of NIVA Scores to PCWP lies in the reproducibility and accuracy of PCWP measurements.[48] To overcome this, PCWP measurements were analyzed in patients undergoing RHC, a controlled setting, and tracings reviewed by an independent, blinded cardiologist and/or cardiovascular intensivists (BA, JH, JS, JL). Finally, the admission data required a post hoc retrospective analysis. A prospective heart failure readmission study is needed to better investigate NIVA_HF’s ability in predicting readmission.

CONCLUSIONS

Results from this study suggest that NIVA_HF is a promising non-invasive technology that estimates PCWP over a wide range of values and in a medically diverse population. These results also demonstrate that NIVA_HF is as effective as the gold standard PCWP[49] at identifying patients with elevated filling pressures, who are at risk for hospital admission. The ease of use and non-invasive nature of NIVA_HF allows for use in multiple settings (hospital, clinic, home). This information also provides that there is adequate information in the venous waveform to establish a PCWP surrogate in estimating volume status of subjects. Future studies with this technology will continue to address the potential clinical utility for non-invasive assessment of congestion and optimization of therapy in patients with heart failure.

Highlights.

• NIVA_HF is a non-invasive, wristband, remote monitoring technology that is designed to provide a real time estimate of pulmonary capillary wedge pressure.

• NIVA Scores correlated with pulmonary capillary wedge pressures obtained during right heart catheterization (r=0.92, n=106, p<0.0001).

• A NIVA Score > 18 at time of discharge after right heart catheterization demonstrated a sensitivity for predicting 30-day hospital admission for heart failure exacerbation of 91% (AUC=0.84, p<0.0001).