Authors: Paul M. Heerdt (1Department of Anesthesiology, Applied Hemodynamics, Yale School of Medicine, New Haven, CT, USA.), Vitaly O. Kheyfets (2Department of Pediatrics-Critical Care Medicine, University of Colorado Denver - Anschutz Medical Campus, Denver, CO, USA), Hannah T. Oakland (3Division of Pulmonary, Critical Care, and Sleep Medicine, Department of Medicine, Yale New Haven Hospital and Yale School of Medicine, New Haven, CT, USA.), Phillip Joseph (3Division of Pulmonary, Critical Care, and Sleep Medicine, Department of Medicine, Yale New Haven Hospital and Yale School of Medicine, New Haven, CT, USA.), Inderjit Singh (3Division of Pulmonary, Critical Care, and Sleep Medicine, Department of Medicine, Yale New Haven Hospital and Yale School of Medicine, New Haven, CT, USA.)
Categories: Article, right ventricle, cardiac function, ejection fraction, contractility, afterload
Source: Journal of cardiothoracic and vascular anesthesia
Authors: Paul M. Heerdt, Vitaly O. Kheyfets, Hannah T. Oakland, Phillip Joseph, Inderjit Singh
Continuous measurement of pressure in the right atrium and pulmonary artery has commonly been used to monitor right ventricular function in critically ill and surgical patients. This approach is largely based upon the assumption that right atrial and pulmonary arterial pressures provide accurate surrogates for diastolic filling and peak right ventricular pressures, respectively. However, due to both technical and physiologic factors this assumption is not always true. Accordingly, recent studies have begun to emphasize the potential clinical value of also measuring right ventricular pressure at the bedside. This has highlighted both past and emerging research demonstrating utility of analyzing not only the amplitude of right ventricular pressure but also the shape of the pressure waveform. This brief review summarizes data demonstrating that combining conventional measurements of right ventricular pressure with variables derived from waveform shape allows for more comprehensive and ideally continuous bedside assessment of right ventricular function, particularly when combined with stroke volume measurement or 3-dimensional echocardiography, and discusses the potential use of right ventricular pressure analysis in computational models for evaluating cardiac function.
In the year 150 Galen proposed that blood moves from the right (RV) to left (LV) ventricle via invisible pores in the interventricular septum with only a small amount entering the pulmonary artery (PA) to “nourish” the lungs. This concept persisted for over a thousand years until the Arabic physician Ibn-al-Nafis challenged the notion of septal pores in 1242, stating that blood was heated in the RV and passed from the PA to pulmonary veins by direct connections^1^. It took European scholars another three hundred years to reach the same conclusion with Servetus writing in 1553, “… by a very ingenious arrangement the refined blood is urged forward from the right ventricle of the heart over a long course through the lungs…”^2^. Over the subsequent three centuries, increased understanding of cardiovascular physiology expanded knowledge of the processes by which the heart contracts, pressure is generated, and blood is “urged” from the RV to the pulmonary circulation.
From a clinical care perspective, a major advance occurred in 1929 when Forssmann described transvenous insertion of a catheter into his own right atrium^3^. This opened the door for right heart catheterization as a diagnostic tool that ultimately expanded into bedside monitoring following development of the balloon-directed PA catheter in 1970. While continuous PA catheter monitoring of right atrial and PA pressures subsequently became common, the ability to intermittently measure PA wedge pressure as a surrogate for left atrial pressure led many clinicians to regard the PA catheter as a device for primarily monitoring LV function in the critical care and perioperative environment^4, 5^. Alternatively, in the diagnostic environment, measurement of RV pressure by direct catheterization or during placement of a PA catheter has long been used to characterize both congenital and acquired right heart disease.
In 1960, Kittle reported post-cardiac surgery monitoring of RV pressure for up to 5 days using a catheter inserted via a saphenous vein and described marked RV hypertension as well as progressive RV outflow tract obstruction following repair of pulmonic valvular stenosis^6^. More recent reports indicate the potential utility of continuous RV pressure monitoring using modified PA catheters in perioperative and critical care settings or by implantable technology in ambulatory patients^7–11^. To date, perioperative and ICU analysis of RV pressure has largely been focused upon amplitude and subjective assessment of morphologic features linked to outcome^12^. However, similar to algorithms applied to the electrocardiogram signal that continuously define its amplitude, duration, and distortion, the RV pressure waveform can be analyzed to define specific events in the cardiac cycle and the pressures at these points, along with the time interval between them, incorporated into models to assess RV performance. The purpose of this brief review is to describe evolving approaches for using the RV pressure waveform to quantify RV function at the bedside. Frequently used abbreviations are shown in table 1.
In the intact mammalian heart, the relationship between chamber load and work is substantially different for the RV and LV reflecting differences in geometry, structure, contraction dynamics, and characteristics of the outflow circuit that are ultimately manifest in their respective pressure waveforms (Figure 1A). Anatomically, the RV is commonly regarded as having three free wall regions (inflow, apical, and outflow) wrapped around the interventricular septum^13^. Pressure generation is normally characterized by four mechanical a) free wall inward movement; b) long axis shortening; c) traction at LV attachment points; and d) longitudinal sequential contraction from apex to outflow tract^14–16^. Importantly, a substantial portion of RV systolic function is provided by LV contraction and septal movement^17–20^. Because of the highly compliant, low impedance pulmonary circulation, the normal RV can pump the same amount of blood as the LV by generating only ~20% of the systolic pressure with peak pressure reached early in systole (Fig 1A). However, even subtle changes in the pulmonary circulation such as those induced by mechanical ventilation can shift the timing of peak pressure from early to late in systole without a prominent increase in pressure (Fig 1B). This underscores that the factors determining RV afterload incorporate not only the steady-state pulmonary vascular resistance (PVR) - mean pressure/mean flow - but also pulsatile factors such as overall compliance and distensibility of the pulmonary circulation, stiffness of large vessels, and wave reflections^21^. In that 30–50% of RV work is to meet this pulsatile load^22,23^, a shift to a late systolic peak may indicate a change in the relative contributions of resistive and pulsatile factors to total afterload. For example, as shown is Figure 1C acute occlusion of the left PA in an experimental model produces only a small increase in RV pressure but a measurable decline in ejection fraction (EF) consistent with afterload dependence of RV systolic function. In this example there is little change in systemic blood pressure or cardiac output indictive of compensatory RV heterometric autoregulation^13–15^ and a modest rise in PVR. In contrast, PA elastance, an afterload metric that incorporates both resistive and pulsatile components^21^, shows a proportionally larger rise than PVR suggesting a greater relative contribution of pulsatile factors to the fall in EF. Panel 1D compares the RV pressure waveform before and after PA occlusion and highlights the fact that the largest difference in pressure occurs late in systole when pulsatile factors such as wave reflections are most prominent.
Clinically, characteristic changes in RV and PA pressure waveforms are often evident in patients with advanced pulmonary hypertensive disease. Early work from Cournand et al in 1945 using simultaneous RV and PA waveforms demonstrated “ventricularization” of the PA pressure waveform characterized by a widened PA pulse pressure along with late systolic peaking of the RV pressure waveform in a patient with severe mitral stenosis^24^. More recently, the timing and magnitude of late systolic augmentation of RV pressure in patients with advanced pulmonary hypertension has been linked to stiffness of the proximal PA along with the speed and amplitude of reflected pressure waves and may have prognostic significance in terms of mortality^25^. This late systolic pressure augmentation is not clearly evident in proximal RV hypertensive disease such as pulmonic valve stenosis (Figure 1E) where PA systolic pressure is lower than RV peak pressure and the RV pressure waveform exhibits a smooth single peak without a shoulder followed by a secondary pressure rise.^26^
The observation that PA systolic pressure does not always reflect RV peak pressure highlights the role of factors other than the pulmonary circulation in determining afterload. In this context, even in the absence of pulmonic valve pathology some patients exhibit a measurable gradient between peak pressures in the RV and PA (RV > PA) that is present at rest and becomes further amplified during exercise^27,28^. These gradients result from the contraction pattern of the RV free wall and reflect in part embryological differences between the outflow tract and the rest of the RV^29^. Studies indicate that autonomic regulation/responsiveness of the RV outflow tract plays a role^30^ and with sympathetic nerve stimulation or catecholamine infusion these pressure gradients can be significant^31,32^ (figure 2A). At the far end of the spectrum, rapid afterload reduction of a hypertensive RV can result in profound dynamic RV outflow tract obstruction leading to cardiovascular collapse or what has been termed a “suicide RV”^33–35^, a situation in many ways analogous to the response of a hyperdynamic LV with hypertrophic cardiomyopathy to acute systemic vasodilation. Additionally, it is possible to generate measurable pressure gradients within the RV proper. In an experimental study, Pace and colleagues^31^ demonstrated that with isoproterenol infusion or stellate ganglion stimulation, gradients were induced between the inflow and outflow tracts (inflow > outflow), a response they speculated was the result of extensive circumferential shortening of myocardial fibers that encircle the RV outflow tract. Similar intra-chamber pressure gradients can be observed clinically (Fig 2B).
Algorithms for detecting RV pressure at peak, minimum, and end-diastole along with the first derivative of pressure to define the maximum rate of pressure rise (dP/dtmax) and fall (dP/dtmin) (figure 3A) are widely used for automated analysis during RHC. While similar RV pressure analysis is not commonly available for continuous bedside monitoring, the pressure rise from minimum to end-diastolic RV pressure has been proposed to have significance in cardiac surgical patients in terms of predicting the ease of separating from cardiopulmonary bypass^36^. In addition to amplitude, the relationship between dP/dt and associated events in the cardiac cycle allows for the extraction of further information. For example, dP/dtmax and dP/dtmin approximate the timing of pulmonic valve opening and closing, respectively, with the interval between them corresponding to ejection time. Within this construct, RV pressure at dP/dtmax can be used to approximate PA diastolic pressure^8^ since the valve opens when RV pressure > PA pressure thus the rate of RV pressure rise begins to decline.
Experimental recordings of RV pressure along with simultaneous RV volume and/or proximal PA flow over a broad range of physiological conditions have allowed for development of markers for approximating RV pressure at specific events in the cardiac cycle based upon pressure derivatives. Karamangolou et al described the use of second order and higher derivatives of RV pressure in an algorithm for estimating cardiac output from the RV pressure waveform^9^. Clinical application of this method to pressure waveforms obtained from micromanometers incorporated into pacing wires has been reported^37, 38^. More recently, other investigators^39, 40^ have focused on the second derivative of RV pressure in its native form or squared to produce only unidirectional upright peaks (Figure 3A) as a marker for events in the cardiac cycle. As shown in Figure 3B components of the second derivative squared correspond to events evident in RV volume and PA flow waveforms as well as RV pressure/RV volume and allow for reasonably high-fidelity definition of end-diastole, pulmonic valve opening and closing, peak PA flow, end-systole, and the onset and ending of RV filling. The RV pressure at each of these events, as well as the time interval between them, can then be used in analytic models^39–44^
When RV pressure is recorded with sufficient fidelity and heart rate is relatively slow, diastolic intervals (active relaxation, early filling, diastasis, and atrial systole) are also evident in the pressure waveform (Fig 3C). A characteristic of the diastolic RVP waveform frequently noted to be indicative of dysfunction is the “square root sign” or “dip and plateau” pattern of pressure change in diastole (Fig 3C). Caused by an abrupt halt of initial rapid filling due to a noncompliant RV secondary to adverse remodeling, marked dilation, or extrinsic compression, the square root sign is reported to have prognostic significance in the setting of progressive RV failure and tamponade^12^. However, in experimental studies the square root sign can be observed in the absence of RV dysfunction and be present when RVP is measured with a fluid-filled system but not by concomitant micromanometry (Fig 3C). Accordingly, while clinically predictive, technical factors may contribute to the square root sign and there does not appear to be absolute correlation with RV dysfunction, at least in experimental settings.
The reference standard for quantifying cardiac performance involves “multi-beat” definition of the end-systolic and end-diastolic PV relationships (ESPVR and EDPVR, respectively) from ventricular pressure-volume loops recorded over a range of preload (figure 4A)^45^. With this a) contractility can be quantified independent of load as end-systolic elastance (Ees); b) factors affecting diastolic filling can be expressed as a stiffness constant, capacitance, or end-diastolic elastance; c) a summation of forces opposing ejection and dictating afterload can be derived (arterial elastance or Ea); d) stroke work can be calculated from loop area; e) the potential energy stored in the ventricular walls can be estimated from the boundaries of the ESPVR and EDPVR outside the loop area; f) total pressure volume area, an index of oxygen consumption, can be calculated as stroke work + potential energy; and g) efficiency can be estimated as the ratio of stroke work to pressure volume area (figure 4B). The relationship between Ees in mmHg/mL and Ea, also in mmHg, expressed as Ea/Ees or more commonly Ees/Ea, has been widely used to summarize the contractility/afterload balance or mechanical coupling between the RV and pulmonary circulation.
Several equations for describing the multibeat RV EDPVR have been proposed (only one is shown in Figure 4B) that fundamentally include a factor for the pressure that would be remaining when RV volume is negligible (α) and an exponential function dictating the curvature, commonly referred to as the stiffness constant β^45^. With RV hypertrophic remodeling or dilation β in particular will change, allowing for quantification of changes in diastolic compliance.
Ultimately, while RV pressure volume analysis represents a powerful tool that has been applied clinically for assessment of RV function^45^, use of for bedside patient monitoring remains largely impractical due to the need for continuous and calibrated measurements of RV volume, the complexity of controlled preload alteration, and the lack of a simple, low cost platform to rapidly analyze multi-beat PV loops that can interface with common monitors in the operating room or ICU. In this context, alternative methods to estimate the functional metrics derived from multi-beat PV analysis are desirable.
Methods for quantifying RV:PA coupling as Ees/Ea using just the RV pressure waveform have been proposed^47^. This “single beat” approach is centered on (a) prediction of the maximum RV pressure that would have been achieved if no ejection occurred (Pmax)^39, 41, 43, 45–49^; and (b) definition of end-systolic pressure (ESP)^42^ (Figure 4C). From these data Ees/Ea can be calculated (Pmax/ESP)-1^47^. However, this approach yields unitless values and may be difficult for clinicians to interpret the clinical significance. Alternatively, these same data can be used to estimate RV EF as 1-(ESP/Pmax)^39^. This method yields a metric that clinicians are more familiar with and has been validated in both animal models and humans^39, 41, 43^. Recent data have suggested clinical utility of using implanted catheters for continuous monitoring of pressure-based RVEF.^11^ Finally, pressure/time intervals or the ratio of different areas under the systolic portion of the RV pressure curve have been proposed as potential approaches for approximating RV:PA coupling^50^. That said, accuracy of single beat estimation of RV EF and interpretation of absolute values for Ees/Ea among reported studies can be complicated by the fact that different methods have been used to predict Pmax and, more importantly, ESP^42^. (see below)
A common approach for assessing active relaxation is measurement of dP/dtmin (Fig 3A). However, just as dP/dtmax is a load-dependent index of contractility, dP/dtmin is a load-dependent index of diastolic function^51^. Alternatively, several approaches have been described for quantifying the very brief period of active relaxation as a time constant (tau), which is measured in msec by applying either an exponential or logistic model to define the rate of ventricular pressure decay over the period of isovolumic relaxation^52^. As with most metrics of cardiac function, tau measurements were initially developed for assessing LV pressure volume relationships (Figure 5A) and exhibit varying degrees of load-dependence. However, an intriguing caveat is the fact that under normal low RV pressure conditions, the RV can have little to no period of true isovolumic relaxation^53^, a feature evident in a triangular pressure volume loop with a prolonged ejection interval (Figure 5B). Thus, unlike the LV, calculation of RV tau may yield a value for an interval that technically does not exist. That said, with pulmonary hypertension and the transition to an RV pressure peak in late systole, the RV pressure volume loop becomes more rectangular like the LV with a well-defined isovolumic relaxation interval (Figure 5C). Measurements of RV tau have been widely reported and methods described for continuous measurement in critically ill patients^11^, but typically without specific knowledge of the pressure volume relationship. In terms of the passive filling phase of diastole, poor signal quality from low amplitude oscillations can complicate resolution of events. In contrast, active filling from atrial contraction is often evident with the amplitude of this component potentially related to diastolic compliance. Fundamentally, however, in the absence of concomitant volume measurement, assessment of RV diastolic filling is largely qualitative given the nonlinearity of diastolic compliance and the fact that factors such as remodeling, valve dysfunction, and changes in pericardial restraint (i.e., removal in cardiac surgical patients) can affect the EDPVR. A recent study attempted a more quantitative pressure-based approach by examining the relationship between the rise in RV pressure during diastole (end-diastolic pressure – minimum pressure) and difficulty in separating from cardiopulmonary bypass in cardiac surgical patients. Study results indicated that a pressure rise > 4 mmHg was associated with known preoperative risk factors but not an independent predictor of difficulty separating from bypass^36^.
A common single beat approach to clinical estimation of RV Ees and Ea as independent variables (i.e., not simply their ratio) involves the combination of Pmax and ESP with SV where Ees = (Pmax - ESP)/SV and Ea = ESP/SV^45–47^. With both contractility and afterload expressed in the same units (mmHg/mL), this allows for better interpretation of RV:PA coupling in terms of whether the impairment is the result of an abnormally high afterload relative to contractility, or abnormally low contractility relative to afterload. A recognized limitation of this single beat method is the assumption that the ESPVR volume intercept remains fixed and negligible.^48^ This is in contrast to the multibeat reference standard where the ESPVR volume intercept can vary widely, even within an individual subject as conditions change thus potentially affecting accuracy of the derived single beat Ees.^48^ Despite this limitation, the Pmax single beat model has been widely used and reported. In addition to proprietary data acquisition and analysis systems^8, 11^, Pmax and ESP can be estimated beat-to-beat during conventional right heart catheterization using commercial software when RV and PA pressures are measured^41,47^. Combining Pmax and ESP with SV also allows for deriving RVEF, end-diastolic volume (SV/EF), and Ees and Ea as continuous variables (Figure 6). As noted above, however, interpretation of reported absolute values for Ees and Ea among different studies can be complicated by differences in how Pmax, and particularly ESP, are derived^39, 42, 49, 55–57^. For example, mean PA pressure (mPAP) has been commonly used as a surrogate for ESP^46^ despite the fact accuracy declines as pressure rises^42^. This reflects in part the variable time delay between end-systole, defined as the point of maximal pressure/volume ratio^42^, and end-ejection or what has been termed the “hang out interval”^58^. Accordingly, Tello and colleagues reported a simple correction equation for mPAP based upon a regression of mPAP relative to ESP measured from pressure and volume data^56^. Wright and colleagues have also proposed an adjustment of mPAP that is more complex and involves calculating the time intervals between events in the cardiac cycle^57^. Ultimately, while applying different definitions for ESP may not decrease utility of tracking changes in Ees, Ea, and Ees/Ea in an individual patient over time, it does complicate defining normal values that can be used for universal reference and as a clinical trial end point.
Given curvilinearity of the EDPVR defined by multi-beat analysis during preload variation (figure 4A), single beat estimation of the EDPVR is not trivial as it requires simplifying assumptions and an end-diastolic volume prediction. Single beat methods have been described for approximating β using RV pressure in diastole along with a normalized fixed value for end-diastolic volume and also for deriving end-diastolic elastance expressed as mmHg/mL^40, 59^ with the latter metric having the advantage of incorporating both α and β components of the EDPVR.
The RV pressure waveform and SV can also be used to more fully define pulsatile components of RV afterload such as characteristic impedance (Zc), and wave reflections. Previous studies demonstrated that cardiac output can be estimated from the RV pressure waveform when corrected for Zc^37^. Within this model, Zc can be calculated from the ratio of estimated cardiac output to that measured by a standard technique such as thermodilution. When PA pressure is also measured, total pulmonary resistance (TPR) can be calculated (mPAP/cardiac output) and a wave reflection index estimated (TPR − Zc)/(TPR + Zc). This method has been validated in experimental models and used clinically to determine Zc and a wave reflection index in patients with pulmonary hypertension^21^.
While single beat methods based upon Pmax do not require RV volume measurement, they are affected by fidelity of the RV pressure signal, make assumptions about the ESPVR volume intercept that do not always hold true, and have primarily been used to quantify systolic function. Increasing access to simultaneous measurement of RV volume using 3D echo has made it possible to clinically apply other single beat PV models that do not require Pmax prediction, allow for calculation of the ESPVR volume intercept (see figure 4A) to improve accuracy, and also provide a more complete characterization of diastolic function^44, 60–62^. These models use only end-diastolic (EDV) and end-systolic (ESV) volumes, thus do not require construction of full pressure volume loops^44^. That said, when the 3D volume signal incorporates the full cardiac cycle (i.e., begins and ends with essentially the same end-diastolic volume) and can be optimally synchronized with RV pressure, PV loops can be derived for visual representation of the ESPVR and EDPVR (Figure 7)^10^.
Initially developed and validated for the assessment of LV function, single beat methods incorporating pressure analysis along with EDV and ESV make several assumptions about characteristics of the LV PV relationship^60–62^. Recent experimental studies on the RV, however, also demonstrated favorable comparison to the multibeat standard of PV loops (volume measured by conductance catheter) acquired during preload variation in terms of quantifying Ees based upon a more complete characterization of the ESPVR that does not make assumptions about the volume intercept^44^.
Knowledge of actual EDV and the ESPVR volume intercept allows for a more complete assessment of diastolic function expressed as the β stiffness constant or end-diastolic elastance since fewer assumptions are made^59^. Another index of diastolic function readily derived when EDV is known is capacitance quantified as the predicted EDV in mL at a specific EDP such as 15, 20, or 30 mmHg^62–64^. While this method requires assumptions about the shape of the EDPVR originally derived from studies of the LV, it has been validated for the RV^44^ and has a clinical context making it conceptually easier to grasp than the β stiffness constant.
Combining objectively measured EDV and ESV with the ESPVR volume intercept (Vo) and relevant time events from the RV pressure waveform allows for calculation of potential energy (PE) ESP(ESV-Vo)/2 -EDP(EDV-Vo)/4^65^. Summing PE and stroke work to calculate pressure volume area as a surrogate for oxygen consumption (Figure 4B) further allows for characterization of efficiency as stroke work/pressure volume area. Taken together, when directly measured variables are combined with derived metrics many of the aspects of multibeat PV analysis can generated. This may allow for more comprehensive bedside assessment of both baseline function and the response to a specific intervention. For example, recent reports indicate that in some patients RV contractility may decline during inhaled nitric oxide complicating interpretation of a fall in mPAP.^44,66^ While the mechanism behind this response remains unclear, consistent with these reports, integrating right heart catheterization and 3D echo data may allow for more a complete interpretation of the response to interventions such as inhaled nitric oxide (Figure 6).
As with any signal processing, results are influenced by fidelity of the source data. For RV pressure analysis, two main factors come into a) the limitations of fluid-filled systems for pressure measurement; and b) the dynamic nature of RV pressure morphology. The important considerations of catheter lumen size, frequency response, and damping for optimal RV pressure measurement using fluid-filled catheters have been well described^33^ and take on particular significance when portions of the waveform are used in predictive models such as estimation of Pmax. These limitations can be obviated by the use of larger fluid-filled lumens for RV pressure measurement or insertion of a high fidelity micromanometer through a fluid-filled lumen. That said, recent data indicate that under many circumstances fluid-filled systems can adequately replicate RV pressure measured by micromanometery^47^. As for dynamic features of the RV pressure waveform, analytic algorithms based on contour must be able to exclude distorted waveforms, accommodate respiratory variations and shifts in the timing of peak pressure, and adapt to rapid (even beat-to-beat) changes in morphology. Potential approaches include creation of ECG and respiratory-gated signal averages at regular intervals with exclusion of ectopic beats. Finally, the impact of valve dysfunction on analysis of the RV pressure waveform remains to be clearly defined particularly in regard to tricuspid insufficiency which can affect dP/dtmax and potentially event markers based upon pressure derivatives. A recent study, however, indicated that across a cohort of 25 patients with a mean tricuspid regurgitant fraction of 22% evident on magnetic resonance imaging, estimation of RV EF based entirely upon RV pressure waveform analysis was still able to predict clinically significant reductions in RV EF^41^.
Rapidly advancing technologies underscore the plausibility of combining RV pressure analysis with other clinical data to derive a more comprehensive assessment of overall cardiovascular function at the bedside. For example, incorporating echo assessment of not just RV volume but also valvular function and velocity waveforms with existing analytics of pressure waveforms makes possible formulation of a “hemodynamic digital twin”. More broadly is the potential for using the machine learning approaches of deep learning and feature-based predictive analysis^67^ to create models that leverage the strengths and dampen the weaknesses of individual components. In feature-based predictive analysis, humans draw on their experience in the field to extract features from the RV pressure waveform that they suspect can predict some established marker (e.g., EF, Ees) or be used to make a diagnosis. An empirical model (e.g., neural network, random forest, logistic regression) is then trained that can ideally be applied at the bedside to make predictions in real-time. In deep learning, the model is trained on the raw waveform data to make the same type of prediction, but there is no need to identify features. To our knowledge, there have been few attempts to apply machine learning for estimating established markers of ventricular function based upon pressure waveforms, but proof-of-concept studies have shown that this is theoretically possible^68, 69^. Deep learning and feature-based models have also been applied to echocardiography images to predict cardiac volume^70–72^ and provide a non-invasive diagnostic of pre/post-capillary pulmonary hypertension^73^. Current technology does not allow for continuous image-based measurement of RV volume at the bedside but in the future, deep learning algorithms could also be used to identify novel features of the pressure/volume time-series that might identify new markers of systolic/diastolic function or predict established markers with better accuracy. Furthermore, methods for more fully defining biventricular interaction are being explored^74, 75^.
While PA and RA pressures have long been used as surrogates for RV peak and end-diastolic pressures, respectively, these variables may not always fully represent actual events within the RV. Furthermore, both past and emerging research demonstrate that there is important information in not only the amplitude of RV pressure but also the shape of the pressure waveform. Recent data indicate that combining information derived from RV pressure amplitude and shape with measurements of SV, or even better maximum and minimum volume, may allow for more comprehensive – and potentially continuous - bedside assessment of RV function. This possibility is enhanced by the fact that the use of machine learning models for estimating cardiac function parameters at the bedside is quickly becoming a reality.