A novel metrics to predict right heart failure after left ventricular assist device implantation

Right Heart Failure (RHF) is a severe complication that can occur after left ventricular assist device (LVAD) implantation, increasing early and late mortality. Although numerous RHF predictive scores have been developed, limited data exist on the external validation of these models. We therefore aimed at comparing existent risk score models and identifying predictors of severe RHF at our center. In this retrospective, single-center analysis, clinical, biological and functional data were collected in patients implanted with a LVAD between 2011 and 2020. Early severe RHF was defined as the use of inotropes for ≥ 14 days, nitric oxide use for ≥ 48 h or unplanned right-sided circulatory support. Risk models were evaluated for the primary outcome of RHF or RVAD implantation by means of logistic regression and receiver operating characteristic curves. Among 92 patients implanted, 24 (26%) developed early severe RHF. The EUROMACS-RHF risk score performed the best in predicting RHF (C = 0.82–95% CI: 0.68–0.90), compared with the other scores (Michigan, CRITT). In addition, we developed a new model, based on four variables selected for the best reduced logistic model: the INTERMACS level, the number of inotropes used, the ratio of right atrial/pulmonary capillary wedge pressure and the ratio of right ventricle/left ventricle diameters by echocardiography. This model demonstrated significant discrimination of RHF (C = 0.9–95% CI: 0.76–0.96). Amongst available risk scores, EUROMACS-RHF performs best to predict the occurrence of RHF after LVAD implantation. Our model’s performance compares well to the EUROMACS-RHF score, adding a more objective parameter to RV function evaluation.


Introduction
Heart transplantation (HT) is the gold-standard therapy for end-stage heart failure (HF) but represents a limited therapeutic option [1,2]. The increasing number of patients with refractory, advanced HF and the declining willingness for organ donation have resulted in expanded waiting lists, increased waiting times and mortality in HT waiting list patients [3][4][5].
At present, implantable left ventricular assist devices (LVADs) represent an available and effective alternative to HT, allowing a decrease in morbidity and mortality observed in advanced HF, particularly in patients on HT waiting lists [6,7].
Despite the widespread adoption and success of LVADs used for bridge-to-transplant (BTT) or destination-therapy (DT) indications, complications are numerous limiting its efficacy [8,9].
If the prevention and management of certain complications, such as major bleeding, pump thrombosis, Artificial Heart (Clinical) 1 3 neurological accidents and infections have improved, especially with new generation pumps, postoperative right heart failure (RHF) remains nevertheless a perennial issue to LVAD success.
Post-LVAD acute RHF is common and prevalence has been reported between 4 and 50% [10][11][12][13][14][15]. It is characterized by the Interagency Registry for Mechanically Assisted Circulatory Support (INTERMACS) [16] as documented elevations of central venous pressure associated with more than one clinical or biological manifestations [16]. Severe acute RHF is described by INTERMACS and most clinical investigators as the need for prolonged post-implant inotropes inhaled nitric oxide (NO) or intravenous vasodilators for more than 14 days after LVAD implant, or the requirement for RV mechanical support [16,17]. Severe RHF after LVAD implantation is associated with increased peri-operative mortality, prolonged length of stay, and decreased survival even after HT [13,15,18,19].
The pathophysiology of RHF after LVAD implant is complex and heterogeneous, making it difficult to establish a stratification or risk score [19]. Some risk scoring systems have been described to predict post-LVAD RHF [14,[20][21][22][23][24][25]. However, these studies are limited by the analysis of results in small cohorts, by their monocentric characteristics and by the heterogeneous nature of the LVADs.
The aim of this study was to perform a comparative analysis of commonly used RHF predictive risk models and to identify the best predictors of severe RHF in our cohort of patients undergoing primary LVAD implants.

Study design
This is a retrospective single centre study, who involved patients implanted with a continuous flow LVAD from February 2011 to February 2020. The devices used were all 2nd and 3rd generation continuous flow pumps: InCor (Berlin Heart, Berlin, Germany), Heartmate II and Heartmate 3 (Abbott Laboratories, USA) and Heartware HVAD (Medtronic, USA). The indication of LVAD implantation was bridge-to-transplantation or bridge-to-decision in all cases.
A retrospective analysis of pre-operative clinical, echocardiographic, laboratory, and hemodynamic data was performed to determine the risk for RH failure after LVAD implantation.
Early (< 30 days) severe RHF was defined as: receiving short-or long-term right-sided circulatory support (via an RV assist device or extracorporeal membrane oxygenator), continuous inotropic support for ≥ 14 days, or nitric oxide ventilation for ≥ 48 h [17].
All data were obtained from electronic patient records and the present study was approved by the institutional review board.

Variables collected
Clinical data included demographics, heart failure etiology, medical and surgical history, preoperative treatment, INTERMACS class), need for life support including mechanical ventilation, renal replacement therapy, temporary mechanical circulatory support (MCS), requirement for a continuous intravenous inotrope (milrinone, dobutamine, or levosimendan) or vasopressor (norepinephrine, vasopressin) therapies.
Pre-operative laboratory data were obtained < 24 h before LVAD implantation and included a complete blood count, liver enzymes, renal function, and coagulation parameters.
Echocardiographic measurements closest to the time of LVAD implantation were recorded and analyzed in accordance with published guidelines [27,28], including left ventricular ejection fraction (LVEF) and dimensions, right ventricular (RV) dysfunction on the visual score (based on the subjective visual kinetic function and graded as none, mildmoderate, severe), right ventricular enlargement (defined as basal RV diameter > 42 mm or mid cavitary RV diameter > 35 mm), tricuspid valvular regurgitation (none, mild, moderate, severe), values of tricuspid annular plane systolic excursion (TAPSE) and S' wave, estimation of systolic pulmonary arterial pressure (PAPs), inferior vena cava (IVC) enlargement (defined as IVC > 21 mm), and the ratio of RV/ LV diameter in four-chamber apical echocardiogram view.
Hemodynamic variables were collected during the right heart catheterization, performed closest to the time of LVAD implantation. We included measurements of heart rate, systolic and diastolic pressure, right atrial pressure (RAP), pulmonary artery pressure (PAP), and pulmonary capillary wedge pressure (PCWP). Cardiac output was assessed by thermodilution. Systemic vascular resistance, transpulmonary gradient, pulmonary vascular resistance (PVR), RV stroke work index (RVSWI), pulmonary artery pulsatility index (PAPi) and right atrial to pulmonary capillary wedge pressure ratio (RAP/PCWP) were calculated [29,30]. The transpulmonary gradient was calculated as the difference between the PA mean pressure and PCWP; pulmonary vascular resistance was calculated as the transpulmonary gradient divided by cardiac output. The RV systolic work index was calculated as: [RV stroke volume index × (mean PA pressure − central venous pressure) × 0.0136] expressed in grams per square meter per beat and the PAPi was calculated as: (systolic PA pressure − diastolic PA pressure)/ RAP). Samples of venous blood by the distal end of the pulmonary arterial catheter were collected to measure the mixed venous oxygen saturation (SvO 2 ).
Post-operative outcome, including the early mortality (defined as in-hospital death) and the adverse events as defined in INTERMACS adverse definition [16] were recorded. We included respiratory failure (defined as the need for reintubation, tracheostomy or the inability to discontinue ventilatory support within 6 days), acute renal dysfunction (defined as abnormal kidney function requiring dialysis or a rise in serum creatinine of greater than 3 times baseline or greater than 4 mg/dL or urine output < 0.3 ml/ kg/h sustained for over 24 h), early infection (clinical infection accompanied by pain, fever, drainage and/or leukocytosis that is treated by anti-microbial agents during the implantation stay), major bleeding (episode of suspected internal or external bleeding that results in death, re-operation, hospitalization or transfusions of > 4 U packed red blood cells) and neurological dysfunction (defined as any new neurologic dysfunction documented with appropriate diagnostic tests).

Right ventricular failure predictive risk scores
We applied all the predictive risk scores described in the literature and selected the EUROMACS-RHF risk score, the Michigan score and the CRITT score (Table 1) [14,21,23] to our population. We excluded the UTAH RVF risk score [20] as one of the score components was destination therapy (DT) not used in our population and the Pittsburgh decision tree, because of missing data. These three risk models (Michigan, EUROMACS-RHF, CRITT) were evaluated for the primary outcome of RHF or RVAD implantation by means of logistic regression and receiver operating characteristic curves from their respective scores in our population.

Statistical analysis
The data are given as a number (%) or as a mean ± SD and median score with interquartile ranges. For the description of preoperative variables, categorical data were compared by Chi-square test and continuous nonlongitudinal variables were compared by a Mann-Whitney test.
A logistic regression model was fitted for right ventricular failure, defined according to EUROMACS as receiving short-or long-term right-sided circulatory support, or continuous inotropic support for ≥ 14 days, or NO administration for ≥ 48 h. For inclusion in the logistic regression model, analysis of individual variables was performed by a likelihood ratio chi-square test for the categorical data test and by fitting a univariate logistic regression model to obtain a likelihood ratio test for continuous variables. All variables with a p value < 0.25 were included to fit an initial full logistic regression model.
Given the small absolute numbers of occurrences (45 ventricular failures and 24 severe right ventricular failures), we searched for the best-reduced model that would not include more than four variables for right ventricular failure, and might not differ by a p value < 0.05 calculated from a chi-square distribution for the value of the difference between the log-likelihoods of the full and the reduced logistic models. The reduced logistic models were generated by a hierarchical backward selection with switching. Risk models (EUROMACS-RHF risk score, the Michigan score and the CRITT score) were evaluated for the primary outcome of RH failure or RVAD placement by means of logistic regression and receiver operating characteristic curves from their respective scores in our population. All analyses were performed using the NCSS 20.0.2 statistical package (NCSS, LLC; Kaysville, UT).
RHF was mostly observed in patients with a higher-acuity INTERMACS level (class 1 through 3), with a greater preoperative inotropes' requirement, as well as with a greater need for renal replacement therapy, mechanical ventilation and temporary MCS. They had significantly higher preoperative creatinine, bilirubin and aspartate transaminase, and lower haemoglobin. The level of N-terminal pro-brain natriuretic peptide (NTproBNP) was as well significantly higher in the group presenting a severe RHF. For right heart catheterization parameters, a higher RAP, higher RA/PWCP ratio and lower PAPi were predictive of RHF. A significantly lower cardiac index and SvO 2 were also present in the group of severe RHF. A RV enlargement with a greater ratio of RV/ LV diameters and inferior vena cava dilatation observed by echocardiographic analysis, were found in the RHF group.
Patients with severe RHF had higher early post-operatory mortality and more frequent complications (Table 3).
In-hospital death occurred in 18 patients (20%) after LVAD implantation, with mortality rates of 46% in the RHF group and 10% in patients without RHF (p = 0.0002). Three of the thirteen patients who survived in the RHF group needed a high urgent HT between 12 and 40 days after LVAD implantation.

Right heart failure and risk models
The variables included in the univariate logistic regression model are presented in Supplemental Table 1. The four parameters chosen as the best model provided by the calculation of logistic regression are presented in Table 4 the high INTERMACS level (class 1-3), the number of inotropes used, the ratio of right atrial pressure/pulmonary capillary wedge pressure (RAP/PWCP) at the right heart catheterization and the ratio of right ventricle/left ventricle (RV/LV) diameters in four-chamber apical echocardiogram view. This model demonstrated a significant improvement in the prediction of RHF. As shown in Fig. 1, the area under the ROC curve for severe RHF was 0.91 (CI: 0.76-0.96). We obtained the following fitted model: [ −11.48 + 2.98 × RV / LV + 6.17 × RAP / PCWP + 1.93 × inotropes + 1.73 × (INTERMACS_GROUP = "1-3")]. The plot of jittered outcome versus estimated probabilities from the fitted model is shown in Fig. 2.

Discussion
In this contemporary cohort of patients undergoing primary implantation of continuous-flow LVADs, we found that the EUROMACS-RHF risk score performed best in predicting right heart failure. Furthermore, we created a model with four preoperative metrics, which demonstrated good predictive ability in our population. RHF is an important and frequent complication in the early postoperative period after LVAD implantation. In prior studies, rates of post-LVAD RHF have ranged from 4 to 50% [10][11][12][13][14][15]. This wide range of reported RHF incidence is due to the heterogeneities in RHF definitions. Our definition was in line with the INTERMACS  definition of severe RHF [16] and was limited to the immediate postoperative period after LVAD implantation. In our study, the incidence of RHF was 26%, similar to the recently published EUROMACS study [23]. Furthermore, patients who develop RHF have also a greater risk to develop concomitant complications even death [13][14][15][16][17][18][19][20][21][22][23][24]. In our analysis, RHF was associated with a greater post-operatory mortality rate, as well as with additional comorbidities such as cerebrovascular accidents, respiratory failure, acute kidney injury and multiorgan failure. The challenge then for the medical team is to determine which patients will develop RHF after LVAD implantation. Numerous pre-operative risk scores have been proposed to quantify the risk of RHF in LVAD candidates [20][21][22][23][24][25]. With the exception of the EUROMACS RHF risk score, most predictors scores were typically developed in small single-center studies, used various definitions of RHF, the heterogeneous nature of LVADs leading to inconsistent predictors and no single model dependably forecasting RHF. Moreover, there is a paucity of external validation studies on these risk prediction models [24]. The well-known and most used risk scores include the Michigan RVF score [21], the Heartmate II bridge-to-transplantation RVF analysis [13], the Utah RVF risk score [20], the Pittsburgh decision tree [22], the CRITT score [14] and the EUROMACS RHF-risk score [23]. Results from these studies have reported varying predictive ability of RHF, but with no ideal C-statistics, mostly ranging from 0.6 to 0.7.
Peter et al. [24], have recently compared the performance of some of these risk scores, finding the Michigan score as the only risk model to demonstrate significant discrimination for RHF, even if modest (C = 0.74), compared with newer risk scores (Utah, Pitt, EUROMACS-RHF).
In a recent meta-analysis, Bellavia et al. [31] evaluated observational studies of risk factors associated with RHF after LVAD implantation. Variables found with the highest effect size in predicting RHF were: the need for mechanical ventilation and continuous renal replacement therapy, international normalized ratio and NT-proBNP, RV stroke work index and central venous pressure, pre-implant moderate to severe RV dysfunction and greater RV/LV diameter ratio.
In our study, we performed a comparative analysis between the risk scores mentioned above. We excluded the Utah risk score as it retains the presence of DT which was not used in our population, and the Pittsburgh score due the lacking of several parameters. Finally, we thus compared the Michigan score and the EUROMACS-RHF score for the severe RHF and the CRITT score for the need for RVAD implantation.
In our comparative analysis of these three scores, the EUROMACS-RHF risk score performed the best in predicting RHF. The Michigan score did not demonstrate a significant RHF predictor value in our population. In this score, four out of eleven points, are made by the vasopressors' administration, defined as phenylephrine, norepinephrine, or vasopressin. Or in our cohort, most of the patients were on inotropes and not on vasopressor therapy, making it harder to achieve a high-risk score. The other variables of Michigan score include just biological parameters (bilirubin, creatinine and AST) and not hemodynamic nor echocardiographic parameters.
EUROMACS-RHF [23] is a simple risk score comprising a range of variables including cardiopulmonary hemodynamic and echocardiographic metrics, patient characteristics, and preoperative medical management. It was derived from and validated by more than 2000 adults who underwent continuous-flow LVAD implantation across the European Union in the largest EU Registry of mechanical circulatory support devices. However, one of the limitations of this score is the semiqualitative assessment of RV function on echocardiography evaluated by the visual aspect of RV function.
In our analysis, the most significant predictors of RHF, were the advanced INTERMACS level, the use of a high number of inotropes, the RAP/PWCP ratio and the RV/LV diameters ratio.
The difference obtained with our analysis compared to EUROMACS-RHF was a quantitative RV assessment in echocardiography obtained by the ratio of the RV and LV diameters.
The present study validates the most used and recent RHF score in a continuous flow LVAD population and expands the comparative analysis to include newer predictive models and metrics. We highlight the severity of the population who develop RHF by the high INTERMACS level and the need for multiples inotropes, with a focus on the right ventricle itself by the dilatation and the elevation of right pressures. Judicious patient selection is vital to preventing RHF in patients undergoing LVAD implantation. A patient with a high-risk RHF score may require perioperative optimization of RV support [32] via reduction of preload [33,34], afterload [35], and RV contractility support [36], or early RV mechanical support or biventricular assist device [37,38]. In fact, despite aggressive risk stratification and medical management, some patients still develop RHF requiring RVAD support. The need for an RVAD is associated with more severe outcomes [37] and an elective RVAD correlates with better long-term survival than an emergency implantation [39], whilst also improving survival to transplant compared with delayed RVAD insertion [40]. The use of that predictive RHF score helps the medical team to better prepare high-risk patients and to be more quickly reactive to complications.

Limitations
This study is a single-center retrospective study. The sample size is fairly modest and the analyzed population was further reduced by missing hemodynamic variables in some of the patients. In addition, this study was unable to compare the UTAH and the Pittsburgh model.
Innovative parameters to evaluate the RV as the speckle tracking echocardiography (strain), the 3D-RV echocardiography, or the cardiac MRI were not studied to improve the evaluation of RV dysfunction.

Conclusion
Amongst the available risk scores, the EUROMACS performs best to predict the occurrence of RHF after LVAD implantation. Based on four simple metrics, our model's performance compares well to the EUROMACS score, adding a more objective quantitative evaluation of the preoperatory RV function.
Acknowledgements AR had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. FV and AR designed the study, interpreted data and drafted the manuscript. FV collected and analyzed data. EE performed the statistical analysis. CS, JLV, CD, EE, FVE have made substantial contributions to the conception and design and interpretation of data, Fig. 3 Receiver operating characteristic curves (ROC) derived from the logistic regression models for severe right ventricular failure (EUROMACS and Michigan risk score) or for right ventricular assist device placement (CRITT score) from their respective scores in our population, and the associated histograms of estimated probabilities from the fitted models